diff options
Diffstat (limited to 'third_party')
152 files changed, 12909 insertions, 6262 deletions
diff --git a/third_party/android/BUILD b/third_party/android/BUILD new file mode 100644 index 0000000000..e69de29bb2 --- /dev/null +++ b/third_party/android/BUILD diff --git a/third_party/android/android.bzl.tpl b/third_party/android/android.bzl.tpl new file mode 100644 index 0000000000..e6ed4994f3 --- /dev/null +++ b/third_party/android/android.bzl.tpl @@ -0,0 +1,9 @@ +"""Set up configurable Android SDK and NDK dependencies.""" + +def android_workspace(): + # String for replacement in Bazel template. + # These will either be replaced by android_sdk_repository if various ENV + # variables are set when `local_config_android` repo_rule is run, or they + # will be replaced by noops otherwise. + MAYBE_ANDROID_SDK_REPOSITORY + MAYBE_ANDROID_NDK_REPOSITORY diff --git a/third_party/android/android_configure.BUILD.tpl b/third_party/android/android_configure.BUILD.tpl new file mode 100644 index 0000000000..e69de29bb2 --- /dev/null +++ b/third_party/android/android_configure.BUILD.tpl diff --git a/third_party/android/android_configure.bzl b/third_party/android/android_configure.bzl new file mode 100644 index 0000000000..da09bdf39e --- /dev/null +++ b/third_party/android/android_configure.bzl @@ -0,0 +1,87 @@ +"""Repository rule for Android SDK and NDK autoconfiguration. + +`android_configure` depends on the following environment variables: + + * `ANDROID_NDK_HOME`: Location of Android NDK root. + * `ANDROID_SDK_HOME`: Location of Android SDK root. + * `ANDROID_SDK_API_LEVEL`: Desired Android SDK API version. + * `ANDROID_NDK_API_LEVEL`: Desired Android NDK API version. + * `ANDROID_BUILD_TOOLS_VERSION`: Desired Android build tools version. +""" + +# TODO(mikecase): Move logic for getting default values for the env variables +# from configure.py script into this rule. + +_ANDROID_NDK_HOME = "ANDROID_NDK_HOME" +_ANDROID_SDK_HOME = "ANDROID_SDK_HOME" +_ANDROID_NDK_API_VERSION = "ANDROID_NDK_API_LEVEL" +_ANDROID_SDK_API_VERSION = "ANDROID_SDK_API_LEVEL" +_ANDROID_BUILD_TOOLS_VERSION = "ANDROID_BUILD_TOOLS_VERSION" + +_ANDROID_SDK_REPO_TEMPLATE = """ + native.android_sdk_repository( + name="androidsdk", + path="%s", + api_level=%s, + build_tools_version="%s", + ) +""" + +_ANDROID_NDK_REPO_TEMPLATE = """ + native.android_ndk_repository( + name="androidndk", + path="%s", + api_level=%s, + ) +""" + +def _android_autoconf_impl(repository_ctx): + """Implementation of the android_autoconf repository rule.""" + sdk_home = repository_ctx.os.environ.get(_ANDROID_SDK_HOME) + sdk_api_level = repository_ctx.os.environ.get(_ANDROID_SDK_API_VERSION) + build_tools_version = repository_ctx.os.environ.get( + _ANDROID_BUILD_TOOLS_VERSION) + ndk_home = repository_ctx.os.environ.get(_ANDROID_NDK_HOME) + ndk_api_level = repository_ctx.os.environ.get(_ANDROID_NDK_API_VERSION) + + sdk_rule = "pass" + if all([sdk_home, sdk_api_level, build_tools_version]): + sdk_rule = _ANDROID_SDK_REPO_TEMPLATE % ( + sdk_home, sdk_api_level, build_tools_version) + + ndk_rule = "pass" + if all([ndk_home, ndk_api_level]): + ndk_rule = _ANDROID_NDK_REPO_TEMPLATE % (ndk_home, ndk_api_level) + + repository_ctx.template( + "BUILD", + Label("//third_party/android:android_configure.BUILD.tpl")) + repository_ctx.template( + "android.bzl", + Label("//third_party/android:android.bzl.tpl"), + substitutions={ + "MAYBE_ANDROID_SDK_REPOSITORY": sdk_rule, + "MAYBE_ANDROID_NDK_REPOSITORY": ndk_rule, + }) + +android_configure = repository_rule( + implementation = _android_autoconf_impl, + environ = [ + _ANDROID_SDK_API_VERSION, + _ANDROID_NDK_API_VERSION, + _ANDROID_BUILD_TOOLS_VERSION, + _ANDROID_NDK_HOME, + _ANDROID_SDK_HOME, + ], +) +"""Writes Android SDK and NDK rules. + +Add the following to your WORKSPACE FILE: + +```python +android_configure(name = "local_config_android") +``` + +Args: + name: A unique name for this workspace rule. +""" diff --git a/third_party/aws.BUILD b/third_party/aws.BUILD index 2dc921933c..5426f79e46 100644 --- a/third_party/aws.BUILD +++ b/third_party/aws.BUILD @@ -46,6 +46,8 @@ cc_library( "aws-cpp-sdk-core/source/utils/xml/**/*.cpp", "aws-cpp-sdk-core/source/utils/crypto/*.cpp", "aws-cpp-sdk-core/source/utils/crypto/factory/**/*.cpp", + "aws-cpp-sdk-kinesis/include/**/*.h", + "aws-cpp-sdk-kinesis/source/**/*.cpp", "aws-cpp-sdk-s3/include/**/*.h", "aws-cpp-sdk-s3/source/**/*.cpp", ]), @@ -72,6 +74,7 @@ cc_library( }), includes = [ "aws-cpp-sdk-core/include/", + "aws-cpp-sdk-kinesis/include/", "aws-cpp-sdk-s3/include/", ], deps = [ diff --git a/third_party/clang_toolchain/BUILD b/third_party/clang_toolchain/BUILD new file mode 100644 index 0000000000..e69de29bb2 --- /dev/null +++ b/third_party/clang_toolchain/BUILD diff --git a/third_party/clang_toolchain/cc_configure_clang.bzl b/third_party/clang_toolchain/cc_configure_clang.bzl new file mode 100644 index 0000000000..0778c43c53 --- /dev/null +++ b/third_party/clang_toolchain/cc_configure_clang.bzl @@ -0,0 +1,27 @@ +""" Downloads clang and configures the crosstool using bazel's autoconf.""" + +load("@bazel_tools//tools/cpp:cc_configure.bzl", "cc_autoconf_impl") +load(":download_clang.bzl", "download_clang") + +_TF_DOWNLOAD_CLANG = "TF_DOWNLOAD_CLANG" +_TF_NEED_CUDA = "TF_NEED_CUDA" + +def _cc_clang_autoconf(repo_ctx): + if repo_ctx.os.environ.get(_TF_DOWNLOAD_CLANG) != "1": + return + if repo_ctx.os.environ.get(_TF_NEED_CUDA) == "1": + # Clang is handled separately for CUDA configs. + # See cuda_configure.bzl for more details. + return + + download_clang(repo_ctx, out_folder = "extra_tools") + overriden_tools = {"gcc": "extra_tools/bin/clang"} + cc_autoconf_impl(repo_ctx, overriden_tools) + +cc_download_clang_toolchain = repository_rule( + environ = [ + _TF_DOWNLOAD_CLANG, + _TF_NEED_CUDA, + ], + implementation = _cc_clang_autoconf, +) diff --git a/third_party/clang_toolchain/download_clang.bzl b/third_party/clang_toolchain/download_clang.bzl new file mode 100644 index 0000000000..e782739661 --- /dev/null +++ b/third_party/clang_toolchain/download_clang.bzl @@ -0,0 +1,60 @@ +""" Helpers to download a recent clang release.""" + +def _get_platform_folder(os_name): + os_name = os_name.lower() + if os_name.startswith("windows"): + return "Win" + if os_name.startswith("mac os"): + return "Mac" + if not os_name.startswith("linux"): + fail("Unknown platform") + return "Linux_x64" + +def _download_chromium_clang( + repo_ctx, + platform_folder, + package_version, + sha256, + out_folder): + cds_url = "https://commondatastorage.googleapis.com/chromium-browser-clang" + cds_file = "clang-%s.tgz" % package_version + cds_full_url = "{0}/{1}/{2}".format(cds_url, platform_folder, cds_file) + repo_ctx.download_and_extract(cds_full_url, output = out_folder, sha256 = sha256) + +def download_clang(repo_ctx, out_folder): + """ Download a fresh clang release and put it into out_folder. + + Clang itself will be located in 'out_folder/bin/clang'. + We currently download one of the latest releases of clang by the + Chromium project (see + https://chromium.googlesource.com/chromium/src/+/master/docs/clang.md). + + Args: + repo_ctx: An instance of repository_context object. + out_folder: A folder to extract the compiler into. + """ + # TODO(ibiryukov): we currently download and extract some extra tools in the + # clang release (e.g., sanitizers). We should probably remove the ones + # we don't need and document the ones we want provide in addition to clang. + + # Latest CLANG_REVISION and CLANG_SUB_REVISION of the Chromiums's release + # can be found in https://chromium.googlesource.com/chromium/src/tools/clang/+/master/scripts/update.py + CLANG_REVISION = "340427" + CLANG_SUB_REVISION = 1 + + package_version = "%s-%s" % (CLANG_REVISION, CLANG_SUB_REVISION) + + checksums = { + "Linux_x64": "8a8f21fb624fc7be7e91e439a13114847185375bb932db51ba590174ecaf764b", + "Mac": "ba894536b7c8d37103a5ddba784f268d55e65bb2ea1200a2cf9f2ef1590eaacd", + "Win": "c3f5bd977266dfd011411c94a13e00974b643b70fb0225a5fb030f7f703fa474", + } + + platform_folder = _get_platform_folder(repo_ctx.os.name) + _download_chromium_clang( + repo_ctx, + platform_folder, + package_version, + checksums[platform_folder], + out_folder, + ) diff --git a/third_party/curl.BUILD b/third_party/curl.BUILD index 4def6f9489..c93fac6549 100644 --- a/third_party/curl.BUILD +++ b/third_party/curl.BUILD @@ -7,6 +7,7 @@ exports_files(["COPYING"]) CURL_WIN_COPTS = [ "/Iexternal/curl/lib", + "/DBUILDING_LIBCURL", "/DHAVE_CONFIG_H", "/DCURL_DISABLE_FTP", "/DCURL_DISABLE_NTLM", @@ -49,6 +50,8 @@ cc_library( "lib/curl_addrinfo.c", "lib/curl_addrinfo.h", "lib/curl_base64.h", + "lib/curl_ctype.c", + "lib/curl_ctype.h", "lib/curl_des.h", "lib/curl_endian.h", "lib/curl_fnmatch.c", @@ -75,6 +78,7 @@ cc_library( "lib/curl_sec.h", "lib/curl_setup.h", "lib/curl_setup_once.h", + "lib/curl_sha256.h", "lib/curl_sspi.c", "lib/curl_sspi.h", "lib/curl_threads.c", @@ -134,6 +138,8 @@ cc_library( "lib/md5.c", "lib/memdebug.c", "lib/memdebug.h", + "lib/mime.c", + "lib/mime.h", "lib/mprintf.c", "lib/multi.c", "lib/multihandle.h", @@ -153,8 +159,8 @@ cc_library( "lib/pop3.h", "lib/progress.c", "lib/progress.h", - "lib/rawstr.c", - "lib/rawstr.h", + "lib/rand.c", + "lib/rand.h", "lib/rtsp.c", "lib/rtsp.h", "lib/security.c", @@ -162,8 +168,11 @@ cc_library( "lib/select.h", "lib/sendf.c", "lib/sendf.h", + "lib/setopt.c", + "lib/setopt.h", "lib/setup-os400.h", "lib/setup-vms.h", + "lib/sha256.c", "lib/share.c", "lib/share.h", "lib/sigpipe.h", @@ -179,10 +188,10 @@ cc_library( "lib/splay.c", "lib/splay.h", "lib/ssh.h", + "lib/strcase.c", + "lib/strcase.h", "lib/strdup.c", "lib/strdup.h", - "lib/strequal.c", - "lib/strequal.h", "lib/strerror.c", "lib/strerror.h", "lib/strtok.c", @@ -234,28 +243,26 @@ cc_library( "lib/vtls/darwinssl.c", ], "@org_tensorflow//tensorflow:windows": CURL_WIN_SRCS, - "@org_tensorflow//tensorflow:windows_msvc": CURL_WIN_SRCS, "//conditions:default": [ "lib/vtls/openssl.c", ], }), hdrs = [ "include/curl/curl.h", - "include/curl/curlbuild.h", - "include/curl/curlrules.h", "include/curl/curlver.h", "include/curl/easy.h", "include/curl/mprintf.h", "include/curl/multi.h", "include/curl/stdcheaders.h", + "include/curl/system.h", "include/curl/typecheck-gcc.h", ], copts = select({ "@org_tensorflow//tensorflow:windows": CURL_WIN_COPTS, - "@org_tensorflow//tensorflow:windows_msvc": CURL_WIN_COPTS, "//conditions:default": [ "-Iexternal/curl/lib", "-D_GNU_SOURCE", + "-DBUILDING_LIBCURL", "-DHAVE_CONFIG_H", "-DCURL_DISABLE_FTP", "-DCURL_DISABLE_NTLM", # turning it off in configure is not enough @@ -271,10 +278,6 @@ cc_library( # See curl.h for discussion of write size and Windows "/DCURL_MAX_WRITE_SIZE=16384", ], - "@org_tensorflow//tensorflow:windows_msvc": [ - # See curl.h for discussion of write size and Windows - "/DCURL_MAX_WRITE_SIZE=16384", - ], "//conditions:default": [ "-DCURL_MAX_WRITE_SIZE=65536", ], @@ -298,12 +301,6 @@ cc_library( "-DEFAULTLIB:crypt32.lib", "-DEFAULTLIB:Normaliz.lib", ], - "@org_tensorflow//tensorflow:windows_msvc": [ - "-DEFAULTLIB:ws2_32.lib", - "-DEFAULTLIB:advapi32.lib", - "-DEFAULTLIB:crypt32.lib", - "-DEFAULTLIB:Normaliz.lib", - ], "//conditions:default": [ "-lrt", ], @@ -314,7 +311,6 @@ cc_library( ] + select({ "@org_tensorflow//tensorflow:ios": [], "@org_tensorflow//tensorflow:windows": [], - "@org_tensorflow//tensorflow:windows_msvc": [], "//conditions:default": [ "@boringssl//:ssl", ], @@ -417,7 +413,6 @@ cc_binary( ], copts = select({ "@org_tensorflow//tensorflow:windows": CURL_BIN_WIN_COPTS, - "@org_tensorflow//tensorflow:windows_msvc": CURL_BIN_WIN_COPTS, "//conditions:default": [ "-Iexternal/curl/lib", "-D_GNU_SOURCE", @@ -676,6 +671,7 @@ genrule( "# define SIZEOF_INT 4", "# define SIZEOF_LONG 8", "# define SIZEOF_OFF_T 8", + "# define SIZEOF_CURL_OFF_T 8", "# define SIZEOF_SHORT 2", "# define SIZEOF_SIZE_T 8", "# define SIZEOF_TIME_T 8", diff --git a/third_party/double_conversion.BUILD b/third_party/double_conversion.BUILD new file mode 100644 index 0000000000..d875a1a2b5 --- /dev/null +++ b/third_party/double_conversion.BUILD @@ -0,0 +1,42 @@ +# Bazel(http://bazel.io) BUILD file + +licenses(["notice"]) + +exports_files(["LICENSE"]) + +config_setting( + name = "windows", + values = {"cpu": "x64_windows"}, +) + +cc_library( + name = "double-conversion", + srcs = [ + "double-conversion/bignum.cc", + "double-conversion/bignum-dtoa.cc", + "double-conversion/cached-powers.cc", + "double-conversion/diy-fp.cc", + "double-conversion/double-conversion.cc", + "double-conversion/fast-dtoa.cc", + "double-conversion/fixed-dtoa.cc", + "double-conversion/strtod.cc", + "double-conversion/utils.h", + ], + hdrs = [ + "double-conversion/bignum.h", + "double-conversion/bignum-dtoa.h", + "double-conversion/cached-powers.h", + "double-conversion/diy-fp.h", + "double-conversion/double-conversion.h", + "double-conversion/fast-dtoa.h", + "double-conversion/fixed-dtoa.h", + "double-conversion/ieee.h", + "double-conversion/strtod.h", + ], + includes = ["."], + linkopts = select({ + ":windows": [], + "//conditions:default": ["-lm"], + }), + visibility = ["//visibility:public"], +) diff --git a/third_party/eigen.BUILD b/third_party/eigen.BUILD index 07bb6645eb..759f8a9be9 100644 --- a/third_party/eigen.BUILD +++ b/third_party/eigen.BUILD @@ -64,7 +64,14 @@ cc_library( # This define (mostly) guarantees we don't link any problematic # code. We use it, but we do not rely on it, as evidenced above. "EIGEN_MPL2_ONLY", + "EIGEN_MAX_ALIGN_BYTES=64", ], includes = ["."], visibility = ["//visibility:public"], ) + +filegroup( + name = "eigen_header_files", + srcs = EIGEN_MPL2_HEADER_FILES, + visibility = ["//visibility:public"], +) diff --git a/third_party/eigen3/BUILD b/third_party/eigen3/BUILD index f661093bc9..f072f2545a 100644 --- a/third_party/eigen3/BUILD +++ b/third_party/eigen3/BUILD @@ -17,21 +17,23 @@ load("//tensorflow:tensorflow.bzl", "if_mkl") # INTEL_MKL end load("//tensorflow:tensorflow.bzl", "if_mkl") +EIGEN3_THIRD_PARTY_HEADERS = [ + "Eigen/Core", + "Eigen/LU", + "Eigen/Cholesky", + "Eigen/Eigenvalues", + "Eigen/QR", + "Eigen/SVD", + "unsupported/Eigen/MatrixFunctions", + "unsupported/Eigen/SpecialFunctions", + "unsupported/Eigen/CXX11/ThreadPool", + "unsupported/Eigen/CXX11/Tensor", + "unsupported/Eigen/CXX11/FixedPoint", +] + glob(["unsupported/Eigen/CXX11/src/FixedPoint/*.h"]) + cc_library( name = "eigen3", - hdrs = glob(["unsupported/Eigen/CXX11/src/FixedPoint/*.h"]) + [ - "Eigen/Core", - "Eigen/LU", - "Eigen/Cholesky", - "Eigen/Eigenvalues", - "Eigen/QR", - "Eigen/SVD", - "unsupported/Eigen/MatrixFunctions", - "unsupported/Eigen/SpecialFunctions", - "unsupported/Eigen/CXX11/ThreadPool", - "unsupported/Eigen/CXX11/Tensor", - "unsupported/Eigen/CXX11/FixedPoint", - ], + hdrs = EIGEN3_THIRD_PARTY_HEADERS, includes = if_mkl(["./mkl_include"]), visibility = ["//visibility:public"], deps = [ @@ -48,3 +50,29 @@ filegroup( ), visibility = ["//tensorflow:__subpackages__"], ) + +filegroup( + name = "eigen_third_party_header_files", + srcs = EIGEN3_THIRD_PARTY_HEADERS, + visibility = ["//visibility:public"], +) + +genrule( + name = "install_eigen_headers", + srcs = [ + "@eigen_archive//:eigen_header_files", + ":eigen_third_party_header_files", + ], + outs = ["include"], + cmd = """ + mkdir $@ + for f in $(SRCS); do + d="$${f%/*}" + d="$${d#*external/eigen_archive/}" + + mkdir -p "$@/$${d}" + cp "$${f}" "$@/$${d}/" + done + """, + tags = ["manual"], +) diff --git a/third_party/eigen3/unsupported/Eigen/CXX11/Core b/third_party/eigen3/unsupported/Eigen/CXX11/Core deleted file mode 100644 index 1b3690716c..0000000000 --- a/third_party/eigen3/unsupported/Eigen/CXX11/Core +++ /dev/null @@ -1,46 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. -// -// Copyright (C) 2013 Christian Seiler <christian@iwakd.de> -// Copyright (C) 2014 Benoit Steiner <benoit.steiner.goog@gmail.com> -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#ifndef EIGEN_CXX11_CORE_MODULE -#define EIGEN_CXX11_CORE_MODULE - -#include <Eigen/Core> - -#include <Eigen/src/Core/util/DisableStupidWarnings.h> - -/** \defgroup CXX11_Core_Module C++11 Core Module - * - * This module provides common core features for all modules that - * explicitly depend on C++11. Currently, this is only the Tensor - * module. Note that at this stage, you should not need to include - * this module directly. - * - * It also provides a limited fallback for compilers that don't support - * CXX11 yet, such as nvcc. - * - * \code - * #include <Eigen/CXX11/Core> - * \endcode - */ - -// Only a subset of cxx11 is allowed at Google, so we default to emulate the -// cxx11 functionality that we need. -#include "src/Core/util/FixedSizeVector.h" -#if 1 -#include <vector> -#include "src/Core/util/EmulateCXX11Meta.h" -#else -#include "src/Core/util/CXX11Workarounds.h" -#include "src/Core/util/CXX11Meta.h" -#endif -#include <Eigen/src/Core/util/ReenableStupidWarnings.h> - -#endif // EIGEN_CXX11_CORE_MODULE - diff --git a/third_party/eigen3/unsupported/Eigen/CXX11/NeuralNetworks b/third_party/eigen3/unsupported/Eigen/CXX11/NeuralNetworks deleted file mode 100644 index 7741b68d8a..0000000000 --- a/third_party/eigen3/unsupported/Eigen/CXX11/NeuralNetworks +++ /dev/null @@ -1,35 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. -// -// Copyright (C) 2014 Benoit Steiner <benoit.steiner.goog@gmail.com> -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#ifndef EIGEN_CXX11_NEURAL_NETWORKS_MODULE -#define EIGEN_CXX11_NEURAL_NETWORKS_MODULE - -#include "unsupported/Eigen/CXX11/Tensor" - -/** \defgroup CXX11_NeuralNetworks_Module Neural Networks Module - * - * This module provides an efficient implementation of the common primitives - * used by neural networks. - * The primitives are built on top of the tensor library. - * - * \code - * #include <Eigen/CXX11/NeuralNetworks> - * \endcode - */ - -#include "unsupported/Eigen/CXX11/src/NeuralNetworks/Activations.h" -#include "unsupported/Eigen/CXX11/src/NeuralNetworks/Attention.h" -#include "unsupported/Eigen/CXX11/src/NeuralNetworks/Pooling.h" -#include "unsupported/Eigen/CXX11/src/NeuralNetworks/SoftMax.h" -#include "unsupported/Eigen/CXX11/src/NeuralNetworks/BackwardCuboidConvolutions.h" -#include "unsupported/Eigen/CXX11/src/NeuralNetworks/CuboidConvolution.h" -#include "unsupported/Eigen/CXX11/src/NeuralNetworks/BackwardSpatialConvolutions.h" -#include "unsupported/Eigen/CXX11/src/NeuralNetworks/SpatialConvolutions.h" - -#endif // EIGEN_CXX11_NEURAL_NETWORKS_MODULE diff --git a/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/FixedPointTypes.h b/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/FixedPointTypes.h index 6b625abc3e..5ab3664918 100644 --- a/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/FixedPointTypes.h +++ b/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/FixedPointTypes.h @@ -7,8 +7,8 @@ // Public License v. 2.0. If a copy of the MPL was not distributed // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. -#ifndef EIGEN_CXX11_FIXED_POINT_TYPES_H -#define EIGEN_CXX11_FIXED_POINT_TYPES_H +#ifndef CXX11_SRC_FIXEDPOINT_FIXEDPOINTTYPES_H_ +#define CXX11_SRC_FIXEDPOINT_FIXEDPOINTTYPES_H_ #include <cmath> #include <iostream> @@ -339,4 +339,4 @@ EIGEN_STRONG_INLINE std::ostream& operator<<(std::ostream& os, QInt32 a) { } // namespace Eigen -#endif // EIGEN_CXX11_FIXED_POINT_TYPES_H +#endif // CXX11_SRC_FIXEDPOINT_FIXEDPOINTTYPES_H_ diff --git a/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/MatMatProduct.h b/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/MatMatProduct.h index 4d0dca07df..e6f4080ae1 100644 --- a/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/MatMatProduct.h +++ b/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/MatMatProduct.h @@ -7,9 +7,8 @@ // Public License v. 2.0. If a copy of the MPL was not distributed // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. -#ifndef EIGEN_CXX11_FIXED_POINT_MAT_MAT_PRODUCT_H -#define EIGEN_CXX11_FIXED_POINT_MAT_MAT_PRODUCT_H - +#ifndef CXX11_SRC_FIXEDPOINT_MATMATPRODUCT_H_ +#define CXX11_SRC_FIXEDPOINT_MATMATPRODUCT_H_ namespace Eigen { namespace internal { @@ -24,6 +23,14 @@ template<> struct scalar_product_traits<QInt8, QInt8> typedef QInt32 ReturnType; }; +// Accumulate the product of 2 QInt16 inputs on 32 bits to prevent +// overflows +template <> +struct scalar_product_traits<QInt16, QInt16> { + enum { Defined = 1 }; + typedef QInt32 ReturnType; +}; + // Accumulate the product of QInt8 inputs with QUint8 inputs on 32 bits // to prevent overflows template<> struct scalar_product_traits<QInt8, QUInt8> @@ -247,9 +254,76 @@ void gebp_kernel<QUInt8, QInt8, Index, DataMapper, mr, nr, ConjugateLhs, Conjuga } #endif -} // namespace internal -} // namespace Eigen +#ifndef EIGEN_USE_OPTIMIZED_INT16_INT16_MAT_MAT_PRODUCT + +template <bool _ConjLhs, bool _ConjRhs> +class gebp_traits<QInt16, QInt16, _ConjLhs, _ConjRhs> { + public: + typedef QInt16 LhsScalar; + typedef QInt16 RhsScalar; + typedef QInt32 ResScalar; + + enum { + // register block size along the M and N directions + // One for the current implementation + nr = 1, + mr = 1, + // Progress made at each iteration of the product loop + // also 1 for the current implementation + LhsProgress = 1, + RhsProgress = 1 + }; +}; + +// The signed 16bit Mat-Mat product itself. +template <typename Index, typename DataMapper, int mr, int nr, + bool ConjugateLhs, bool ConjugateRhs> +struct gebp_kernel<QInt16, QInt16, Index, DataMapper, mr, nr, ConjugateLhs, + ConjugateRhs> { + EIGEN_DONT_INLINE + void operator()(const DataMapper& res, const QInt16* blockA, + const QInt16* blockB, Index rows, Index depth, Index cols, + QInt32 alpha, Index strideA = -1, Index strideB = -1, + Index offsetA = 0, Index offsetB = 0); +}; + +template <typename Index, typename DataMapper, int mr, int nr, + bool ConjugateLhs, bool ConjugateRhs> +EIGEN_DONT_INLINE void gebp_kernel<QInt16, QInt16, Index, DataMapper, mr, nr, + ConjugateLhs, ConjugateRhs>:: +operator()(const DataMapper& res, const QInt16* blockA, const QInt16* blockB, + Index rows, Index depth, Index cols, QInt32 alpha, Index strideA, + Index strideB, Index offsetA, Index offsetB) { + EIGEN_STATIC_ASSERT(!ConjugateLhs, YOU_MADE_A_PROGRAMMING_MISTAKE); + EIGEN_STATIC_ASSERT(!ConjugateRhs, YOU_MADE_A_PROGRAMMING_MISTAKE); + eigen_assert(alpha.value == 1); + eigen_assert(strideA == -1); + eigen_assert(strideB == -1); + eigen_assert(offsetA == 0); + eigen_assert(offsetB == 0); + + eigen_assert(rows > 0); + eigen_assert(cols > 0); + eigen_assert(depth > 0); + eigen_assert(blockA); + eigen_assert(blockB); + + for (Index j = 0; j < cols; ++j) { + Index startB = j * depth; + for (Index i = 0; i < rows; ++i) { + Index startA = i * depth; + + for (Index k = 0; k < depth; ++k) { + res(i, j) += blockA[startA + k] * blockB[startB + k]; + } + } + } +} +#endif + +} // namespace internal +} // namespace Eigen -#endif // EIGEN_CXX11_FIXED_POINT_MAT_MAT_PRODUCT_H +#endif // CXX11_SRC_FIXEDPOINT_MATMATPRODUCT_H_ diff --git a/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/MatMatProductAVX2.h b/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/MatMatProductAVX2.h index 6b4b0edcfb..66532fb600 100644 --- a/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/MatMatProductAVX2.h +++ b/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/MatMatProductAVX2.h @@ -3,18 +3,494 @@ // // Copyright (C) 2015 Benoit Steiner <benoit.steiner.goog@gmail.com> // Copyright (C) 2015 Matthew Sarett <msarett@google.com> +// Copyright (C) 2016 Nishant Patil <nishantpatil@google.com> // // This Source Code Form is subject to the terms of the Mozilla // Public License v. 2.0. If a copy of the MPL was not distributed // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. -#ifndef EIGEN_CXX11_FIXED_POINT_MAT_MAT_PRODUCT_AVX2_H -#define EIGEN_CXX11_FIXED_POINT_MAT_MAT_PRODUCT_AVX2_H +#ifndef CXX11_SRC_FIXEDPOINT_MATMATPRODUCTAVX2_H_ +#define CXX11_SRC_FIXEDPOINT_MATMATPRODUCTAVX2_H_ namespace Eigen { namespace internal { // AVX2 optimized implementation of Mat-Mat product. +// LHS is encoded using signed 16-bit integers. +// RHS is encoded using signed 16-bit integers. +#ifdef EIGEN_USE_OPTIMIZED_INT16_INT16_MAT_MAT_PRODUCT + +// Define quantized traits +template <bool _ConjLhs, bool _ConjRhs> +class gebp_traits<QInt16, QInt16, _ConjLhs, _ConjRhs> { + public: + typedef QInt16 LhsScalar; + typedef QInt16 RhsScalar; + typedef QInt32 ResScalar; + + enum { + // Define register blocking scheme. + nr = 16, + mr = 16, + kr = 4, + // Ignore progress tracking per loop iteration. + LhsProgress = -1, + RhsProgress = -1 + }; +}; + +// Specialized blocking for quantized implementations. +// Used by TensorContractionThreadPool, inputs must have dimensions that are +// multiples of 32. +template <typename Index, int ShardingType> +class TensorContractionBlocking<QInt16, QInt16, Index, ShardingType> { + public: + TensorContractionBlocking(Index k, Index m, Index n, Index num_threads = 1) + : kc_(((k + 15) / 16) * 16), + mc_(((m + 15) / 16) * 16), + nc_(((n + 15) / 16) * 16) { + eigen_assert(mc_ % 16 == 0); + eigen_assert(kc_ % 16 == 0); + if (!k || !m || !n) { + return; + } + + if (ShardingType == ShardByCol) { + eigen_assert(nc_ % 16 == 0); + nc_ = (((nc_ / num_threads) + 15) / 16) * 16; + } else { + eigen_assert(nc_ % 16 == 0); + mc_ = (((mc_ / num_threads) + 15) / 16) * 16; + } + } + + EIGEN_ALWAYS_INLINE Index kc() const { return kc_; } + EIGEN_ALWAYS_INLINE Index mc() const { return mc_; } + EIGEN_ALWAYS_INLINE Index nc() const { return nc_; } + + private: + Index kc_; + Index mc_; + Index nc_; +}; + +// Specialized blocking for quantized implementations. +// Used by TensorContraction and GeneralMatrixMatrix, inputs are padded to +// multiples of 32. +template <int MaxRows, int MaxCols, int MaxDepth, int KcFactor> +class gemm_blocking_space<ColMajor, QInt16, QInt16, MaxRows, MaxCols, MaxDepth, + KcFactor, false> + : public level3_blocking<QInt16, QInt16> { + DenseIndex m_sizeA; + DenseIndex m_sizeB; + + public: + gemm_blocking_space(DenseIndex rows, DenseIndex cols, DenseIndex depth, + DenseIndex /*num_threads*/, bool /*l3_blocking*/) { + this->m_mc = ((rows + 15) / 16) * 16; + this->m_nc = ((cols + 15) / 16) * 16; + this->m_kc = ((depth + 15) / 16) * 16; + m_sizeA = this->m_mc * this->m_kc; + m_sizeB = this->m_kc * this->m_nc; + } + void allocateA() { + if (this->m_blockA == 0) this->m_blockA = aligned_new<QInt16>(m_sizeA); + } + void allocateB() { + if (this->m_blockB == 0) this->m_blockB = aligned_new<QInt16>(m_sizeB); + } + void allocateAll() { + allocateA(); + allocateB(); + } + ~gemm_blocking_space() { + aligned_delete(this->m_blockA, m_sizeA); + aligned_delete(this->m_blockB, m_sizeB); + } +}; + +// Below are the fully optimized versions that are correct only for sizes that +// are multiple of 16. It is about a 10% performance benefit to keep these +// implementations separate. + +// Arrange a block of the left input matrix in contiguous memory. +// +// Given column major input (A0 beside A1 in memory): +// A0 B0 C0 D0 E0 F0 G0 H0 ... +// A1 B1 C1 D1 E1 F1 G1 H1 ... +// A2 B2 C2 D2 E2 F2 G2 H2 ... +// A3 B3 C3 D3 E3 F3 G3 H3 ... +// A4 B4 C4 D4 E4 F4 G4 H4 ... +// A5 B5 C5 D5 E5 F5 G5 H5 ... +// A6 B6 C6 D6 E6 F6 G6 H6 ... +// A7 B7 C7 D7 E7 F7 G7 H7 ... +// A8 ... +// ... +// +// Packing with m = 8 yields row major output (A0 beside B0 in memory): +// A0 B0 +// A1 B1 +// A2 B2 +// A3 B3 +// A4 B4 +// A5 B5 +// A6 B6 +// A7 B7 +// ... +// +// The purpose is to collect m rows of size k. Two elements of the same +// row are arranged contiguously because madd performs an adjacent addition +// in the kernel. + +template <typename Index, typename DataMapper, int Pack1, int Pack2, + bool Conjugate, bool PanelMode> +struct gemm_pack_lhs<QInt16, Index, DataMapper, Pack1, Pack2, ColMajor, + Conjugate, PanelMode> { + EIGEN_DONT_INLINE void operator()(QInt16* blockA, const DataMapper& lhs, + Index depth, Index rows, Index stride = 0, + Index offset = 0); +}; + +template <typename Index, typename DataMapper, int Pack1, int Pack2, + bool Conjugate, bool PanelMode> +EIGEN_DONT_INLINE void gemm_pack_lhs<QInt16, Index, DataMapper, Pack1, Pack2, + ColMajor, Conjugate, PanelMode>:: +operator()(QInt16* blockA, const DataMapper& lhs, Index depth, Index rows, + Index stride, Index offset) { + eigen_assert(stride == 0); + eigen_assert(offset == 0); + + // Use alternate function for weird sizes + if (rows % 16 != 0 || depth % 16 != 0) { + assert(false && + "only depths and rows that are a multiple of 16 are currently " + "supported"); + // gemm_pack_lhs_any<QInt16, Index, DataMapper, Pack1, Pack2, ColMajor, + // Conjugate, PanelMode> lhs_pack; + // return lhs_pack(blockA, lhs, depth, rows, stride, offset); + } + + // Get vector pointer + __m256i* blockA_256 = reinterpret_cast<__m256i*>(blockA); + + // Pack rows in sets of 16 + for (Index m = 0; m < rows; m += 16) { + // Pack depth in sets of 4 + for (Index k = 0; k < depth; k += 4) { + // Load vectors + __m256i L_A = lhs.loadPacket(m, k); + __m256i L_B = lhs.loadPacket(m, k + 1); + __m256i L_C = lhs.loadPacket(m, k + 2); + __m256i L_D = lhs.loadPacket(m, k + 3); + + // Rearrange the inputs as required by the kernel + __m256i L_AB0_AB7 = _mm256_unpacklo_epi16(L_A, L_B); + __m256i L_AB8_AB15 = _mm256_unpackhi_epi16(L_A, L_B); + __m256i L_CD0_CD7 = _mm256_unpacklo_epi16(L_C, L_D); + __m256i L_CD8_CD15 = _mm256_unpackhi_epi16(L_C, L_D); + + __m256i L_AD0 = _mm256_permute2x128_si256(L_AB0_AB7, L_AB8_AB15, 0x20); + _mm256_store_si256(blockA_256++, L_AD0); + __m256i L_AD8 = _mm256_permute2x128_si256(L_CD0_CD7, L_CD8_CD15, 0x20); + _mm256_store_si256(blockA_256++, L_AD8); + __m256i L_AD16 = _mm256_permute2x128_si256(L_AB0_AB7, L_AB8_AB15, 0x31); + _mm256_store_si256(blockA_256++, L_AD16); + __m256i L_AD24 = _mm256_permute2x128_si256(L_CD0_CD7, L_CD8_CD15, 0x31); + _mm256_store_si256(blockA_256++, L_AD24); + } + } +} + +// Arrange a block of the right input matrix in contiguous memory. +// +// Given column major input (A0 beside A1 in memory): +// A0 B0 C0 D0 E0 F0 G0 H0 ... +// A1 B1 C1 D1 E1 F1 G1 H1 ... +// A2 B2 C2 D2 E2 F2 G2 H2 ... +// A3 B3 C3 D3 E3 F3 G3 H3 ... +// A4 B4 C4 D4 E4 F4 G4 H4 ... +// A5 B5 C5 D5 E5 F5 G5 H5 ... +// A6 B6 C6 D6 E6 F6 G6 H6 ... +// A7 B7 C7 D7 E7 F7 G7 H7 ... +// A8 ... +// ... +// Packing yields row major output (A0 beside A1 in memory): +// A0 A1 A2 A3 A4 A5 A6 A7 +// B0 B1 B2 B3 B4 B5 B6 B7 +// ... +// +// At least two elements of the same col are arranged contiguously because +// maddubs and madd both perform an adjacent addition in the kernel. We can +// save work by leaving 4 adjacent elements because kr = 4. +// The purpose is to collect n cols of size k. Two elements of the same +// col are arranged contiguously because madd performs an adjacent addition +// in the kernel. +template <typename Index, typename DataMapper, int nr, bool Conjugate, + bool PanelMode> +struct gemm_pack_rhs<QInt16, Index, DataMapper, nr, ColMajor, Conjugate, + PanelMode> { + EIGEN_DONT_INLINE void operator()(QInt16* blockB, const DataMapper& rhs, + Index depth, Index cols, Index stride = 0, + Index offset = 0); +}; + +template <typename Index, typename DataMapper, int nr, bool Conjugate, + bool PanelMode> +EIGEN_DONT_INLINE void +gemm_pack_rhs<QInt16, Index, DataMapper, nr, ColMajor, Conjugate, PanelMode>:: +operator()(QInt16* blockB, const DataMapper& rhs, Index depth, Index cols, + Index stride, Index offset) { + eigen_assert(stride == 0); + eigen_assert(offset == 0); + + // Use alternate function for weird sizes + if (cols % 16 != 0 || depth % 16 != 0) { + assert(false && + "only depths and cols that are a multiple of 16 are currently " + "supported"); + // gemm_pack_rhs_any<QInt16, Index, DataMapper, nr, ColMajor, Conjugate, + // PanelMode> rhs_pack; + // return rhs_pack(blockB, rhs, depth, cols, stride, offset); + } + + // Get vector pointer + __m256i* blockB_256 = reinterpret_cast<__m256i*>(blockB); + + // Perform a step of the packing for 4 columns + __m256i R_AB_L, R_AB_H, R_CD_L, R_CD_H, R_AD_0, R_AD_4, R_AD_8, R_AD_12; +#define PACK_STEP \ + R_AB_L = _mm256_unpacklo_epi64(R_A, R_B); \ + R_CD_L = _mm256_unpacklo_epi64(R_C, R_D); \ + R_AB_H = _mm256_unpackhi_epi64(R_A, R_B); \ + R_CD_H = _mm256_unpackhi_epi64(R_C, R_D); \ + R_AD_0 = _mm256_permute2x128_si256(R_AB_L, R_CD_L, 0x20); \ + R_AD_8 = _mm256_permute2x128_si256(R_AB_L, R_CD_L, 0x31); \ + R_AD_4 = _mm256_permute2x128_si256(R_AB_H, R_CD_H, 0x20); \ + R_AD_12 = _mm256_permute2x128_si256(R_AB_H, R_CD_H, 0x31); \ + _mm256_store_si256(blockB_256, R_AD_0); \ + _mm256_store_si256(blockB_256 + 4, R_AD_4); \ + _mm256_store_si256(blockB_256 + 8, R_AD_8); \ + _mm256_store_si256(blockB_256 + 12, R_AD_12); \ + blockB_256++; + + // Pack cols in sets of 16 + for (Index n = 0; n < cols; n += 16) { + // Pack depth in sets of 16 + for (Index k = 0; k < depth; k += 16) { + __m256i R_A = rhs.loadPacket(k, n); + __m256i R_B = rhs.loadPacket(k, n + 1); + __m256i R_C = rhs.loadPacket(k, n + 2); + __m256i R_D = rhs.loadPacket(k, n + 3); + PACK_STEP; + + R_A = rhs.loadPacket(k, n + 4); + R_B = rhs.loadPacket(k, n + 5); + R_C = rhs.loadPacket(k, n + 6); + R_D = rhs.loadPacket(k, n + 7); + PACK_STEP; + + R_A = rhs.loadPacket(k, n + 8); + R_B = rhs.loadPacket(k, n + 9); + R_C = rhs.loadPacket(k, n + 10); + R_D = rhs.loadPacket(k, n + 11); + PACK_STEP; + + R_A = rhs.loadPacket(k, n + 12); + R_B = rhs.loadPacket(k, n + 13); + R_C = rhs.loadPacket(k, n + 14); + R_D = rhs.loadPacket(k, n + 15); + PACK_STEP; + + blockB_256 += 12; + } + } +#undef PACK_STEP +} + +// Perform the actual multiplication on packed inputs +template <typename Index, typename DataMapper, int mr, int nr, + bool ConjugateLhs, bool ConjugateRhs> +struct gebp_kernel<QInt16, QInt16, Index, DataMapper, mr, nr, ConjugateLhs, + ConjugateRhs> { + typedef typename DataMapper::LinearMapper LinearMapper; + + EIGEN_DONT_INLINE + void operator()(const DataMapper& res, const QInt16* blockA, + const QInt16* blockB, Index rows, Index depth, Index cols, + QInt32 alpha, Index strideA = -1, Index strideB = -1, + Index offsetA = 0, Index offsetB = 0); +}; + +template <typename Index, typename DataMapper, int mr, int nr, + bool ConjugateLhs, bool ConjugateRhs> +EIGEN_DONT_INLINE void gebp_kernel<QInt16, QInt16, Index, DataMapper, mr, nr, + ConjugateLhs, ConjugateRhs>:: +operator()(const DataMapper& res, const QInt16* blockA, const QInt16* blockB, + Index rows, Index depth, Index cols, QInt32 alpha, Index strideA, + Index strideB, Index offsetA, Index offsetB) { + EIGEN_STATIC_ASSERT(!ConjugateLhs, YOU_MADE_A_PROGRAMMING_MISTAKE); + EIGEN_STATIC_ASSERT(!ConjugateRhs, YOU_MADE_A_PROGRAMMING_MISTAKE); + eigen_assert(alpha.value == 1); + eigen_assert(strideA == -1); + eigen_assert(strideB == -1); + eigen_assert(offsetA == 0); + eigen_assert(offsetB == 0); + eigen_assert(rows > 0); + eigen_assert(cols > 0); + eigen_assert(depth > 0); + eigen_assert(blockA); + eigen_assert(blockB); + + // Use alternate function for weird sizes + if (rows % 16 != 0 || cols % 16 != 0 || depth % 16 != 0) { + assert(false && + "only depths, cols and rows that are a multiple of 16 are currently " + "supported"); + // gebp_kernel_any<QInt16, QInt16, Index, DataMapper, mr, nr, ConjugateLhs, + // ConjugateRhs> gebp; + // return gebp(res, blockA, blockB, rows, depth, cols, alpha, strideA, + // strideB, offsetA, offsetB); + } + + // Create result block + QInt32* blockO = aligned_new<QInt32>(16 * 16); + memset(blockO, 0, 16 * 16 * sizeof(QInt32)); + + // Get vectorized pointers + __m256i* blockO_256 = reinterpret_cast<__m256i*>(blockO); + const __m256i* blockA_256 = reinterpret_cast<const __m256i*>(blockA); + const __m256i* blockB_256 = reinterpret_cast<const __m256i*>(blockB); + + // Loop over blocks of 16 columns + for (Index n = 0; n < cols; n += 16) { + // Reset index into blockA + Index indexL = 0; + // Loop over blocks of 16 rows + for (Index m = 0; m < rows; m += 16) { + // Reset index into blockB + Index indexR = n / 16 * depth; + // Loop over blocks of 4 on depth + for (Index k = 0; k < depth; k += 4) { + // Load inputs + __m256i L_AD0 = blockA_256[indexL++]; + __m256i L_AD8 = blockA_256[indexL++]; + __m256i L_EH0 = blockA_256[indexL++]; + __m256i L_EH8 = blockA_256[indexL++]; + + __m256i R_AH0 = blockB_256[indexR++]; + __m256i R_AH4 = blockB_256[indexR++]; + __m256i R_AH8 = blockB_256[indexR++]; + __m256i R_AH12 = blockB_256[indexR++]; + + // Declare variables used in COMPUTE_STEP + __m256i P_32_A, P_32_B, P_32; + +#define COMPUTE_STEP(R_INPUT_A, R_INPUT_B, OFFSET) \ + P_32_A = _mm256_madd_epi16(R_INPUT_A, L_AD0); \ + P_32_B = _mm256_madd_epi16(R_INPUT_B, L_AD8); \ + P_32 = _mm256_add_epi32(P_32_A, P_32_B); \ + _mm256_store_si256( \ + blockO_256 + 2 * OFFSET, \ + _mm256_add_epi32(_mm256_load_si256(blockO_256 + 2 * OFFSET), P_32)); \ + \ + P_32_A = _mm256_madd_epi16(R_INPUT_A, L_EH0); \ + P_32_B = _mm256_madd_epi16(R_INPUT_B, L_EH8); \ + P_32 = _mm256_add_epi32(P_32_A, P_32_B); \ + _mm256_store_si256( \ + blockO_256 + 2 * OFFSET + 1, \ + _mm256_add_epi32(_mm256_load_si256(blockO_256 + 2 * OFFSET + 1), P_32)); + + // Permute and shuffle to copy a single value across the entire vector + // Then compute the multiplication + // Replicate lower 128-bits of R_AH0 across both lanes + __m256i R_AH0_ = _mm256_permute2x128_si256(R_AH0, R_AH0, 0x00); + // Copy first two elements of R_AH0 across entire vector + __m256i R_AD0 = _mm256_shuffle_epi32(R_AH0_, 0x00); + // Copy second two elements of R_AH0 across entire vector + __m256i R_EH0 = _mm256_shuffle_epi32(R_AH0_, 0x55); + + COMPUTE_STEP(R_AD0, R_EH0, 0); + __m256i R_AD1 = _mm256_shuffle_epi32(R_AH0_, 0xAA); + __m256i R_EH1 = _mm256_shuffle_epi32(R_AH0_, 0xFF); + COMPUTE_STEP(R_AD1, R_EH1, 1); + + // Replicate upper 128-bits of R_AH0 across both lanes + R_AH0_ = _mm256_permute2x128_si256(R_AH0, R_AH0, 0x11); + __m256i R_AD2 = _mm256_shuffle_epi32(R_AH0_, 0x00); + __m256i R_EH2 = _mm256_shuffle_epi32(R_AH0_, 0x55); + COMPUTE_STEP(R_AD2, R_EH2, 2); + __m256i R_AD3 = _mm256_shuffle_epi32(R_AH0_, 0xAA); + __m256i R_EH3 = _mm256_shuffle_epi32(R_AH0_, 0xFF); + COMPUTE_STEP(R_AD3, R_EH3, 3); + + R_AH0_ = _mm256_permute2x128_si256(R_AH4, R_AH4, 0x00); + R_AD0 = _mm256_shuffle_epi32(R_AH0_, 0x00); + R_EH0 = _mm256_shuffle_epi32(R_AH0_, 0x55); + COMPUTE_STEP(R_AD0, R_EH0, 4); + R_AD1 = _mm256_shuffle_epi32(R_AH0_, 0xAA); + R_EH1 = _mm256_shuffle_epi32(R_AH0_, 0xFF); + COMPUTE_STEP(R_AD1, R_EH1, 5); + R_AH0_ = _mm256_permute2x128_si256(R_AH4, R_AH4, 0x11); + R_AD2 = _mm256_shuffle_epi32(R_AH0_, 0x00); + R_EH2 = _mm256_shuffle_epi32(R_AH0_, 0x55); + COMPUTE_STEP(R_AD2, R_EH2, 6); + R_AD3 = _mm256_shuffle_epi32(R_AH0_, 0xAA); + R_EH3 = _mm256_shuffle_epi32(R_AH0_, 0xFF); + COMPUTE_STEP(R_AD3, R_EH3, 7); + + R_AH0_ = _mm256_permute2x128_si256(R_AH8, R_AH8, 0x00); + R_AD0 = _mm256_shuffle_epi32(R_AH0_, 0x00); + R_EH0 = _mm256_shuffle_epi32(R_AH0_, 0x55); + COMPUTE_STEP(R_AD0, R_EH0, 8); + R_AD1 = _mm256_shuffle_epi32(R_AH0_, 0xAA); + R_EH1 = _mm256_shuffle_epi32(R_AH0_, 0xFF); + COMPUTE_STEP(R_AD1, R_EH1, 9); + R_AH0_ = _mm256_permute2x128_si256(R_AH8, R_AH8, 0x11); + R_AD2 = _mm256_shuffle_epi32(R_AH0_, 0x00); + R_EH2 = _mm256_shuffle_epi32(R_AH0_, 0x55); + COMPUTE_STEP(R_AD2, R_EH2, 10); + R_AD3 = _mm256_shuffle_epi32(R_AH0_, 0xAA); + R_EH3 = _mm256_shuffle_epi32(R_AH0_, 0xFF); + COMPUTE_STEP(R_AD3, R_EH3, 11); + + R_AH0_ = _mm256_permute2x128_si256(R_AH12, R_AH12, 0x00); + R_AD0 = _mm256_shuffle_epi32(R_AH0_, 0x00); + R_EH0 = _mm256_shuffle_epi32(R_AH0_, 0x55); + COMPUTE_STEP(R_AD0, R_EH0, 12); + R_AD1 = _mm256_shuffle_epi32(R_AH0_, 0xAA); + R_EH1 = _mm256_shuffle_epi32(R_AH0_, 0xFF); + COMPUTE_STEP(R_AD1, R_EH1, 13); + R_AH0_ = _mm256_permute2x128_si256(R_AH12, R_AH12, 0x11); + R_AD2 = _mm256_shuffle_epi32(R_AH0_, 0x00); + R_EH2 = _mm256_shuffle_epi32(R_AH0_, 0x55); + COMPUTE_STEP(R_AD2, R_EH2, 14); + R_AD3 = _mm256_shuffle_epi32(R_AH0_, 0xAA); + R_EH3 = _mm256_shuffle_epi32(R_AH0_, 0xFF); + COMPUTE_STEP(R_AD3, R_EH3, 15); + +#undef COMPUTE_STEP + } + + // Transfer the results to the result matrix + Index i = 0; + for (Index j = n; j < n + 16; j++) { + LinearMapper r0 = res.getLinearMapper(m, j); + LinearMapper r1 = res.getLinearMapper(m + 8, j); + + r0.storePacket(0, _mm256_add_epi32(blockO_256[i++], r0.loadPacket(0))); + r1.storePacket(0, _mm256_add_epi32(blockO_256[i++], r1.loadPacket(0))); + } + + // Zero the result block so it can be reused + memset(blockO, 0, 16 * 16 * sizeof(QInt32)); + } + } + aligned_delete(blockO, 16 * 16); +} + +#endif + +// AVX2 optimized implementation of Mat-Mat product. // LHS is encoded using signed 8-bit integers. // RHS is encoded using unsigned 8-bit integers. #ifdef EIGEN_USE_OPTIMIZED_INT8_UINT8_MAT_MAT_PRODUCT @@ -1751,4 +2227,4 @@ void gebp_kernel<QInt8, QUInt8, Index, DataMapper, mr, nr, ConjugateLhs, Conjuga } // namespace internal } // namespace Eigen -#endif // EIGEN_CXX11_FIXED_POINT_MAT_MAT_PRODUCT_AVX2_H +#endif // CXX11_SRC_FIXEDPOINT_MATMATPRODUCTAVX2_H_ diff --git a/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/MatMatProductNEON.h b/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/MatMatProductNEON.h index 99894cafb5..9cd3157023 100644 --- a/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/MatMatProductNEON.h +++ b/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/MatMatProductNEON.h @@ -8,9 +8,8 @@ // Public License v. 2.0. If a copy of the MPL was not distributed // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. -#ifndef EIGEN_CXX11_FIXED_POINT_MAT_MAT_PRODUCT_NEON_H -#define EIGEN_CXX11_FIXED_POINT_MAT_MAT_PRODUCT_NEON_H - +#ifndef CXX11_SRC_FIXEDPOINT_MATMATPRODUCTNEON_H_ +#define CXX11_SRC_FIXEDPOINT_MATMATPRODUCTNEON_H_ namespace Eigen { namespace internal { @@ -90,6 +89,4 @@ void gebp_kernel<QInt8, QUInt8, Index, DataMapper, mr, nr, ConjugateLhs, Conjuga } // namespace internal } // namespace Eigen - - -#endif // EIGEN_CXX11_FIXED_POINT_MAT_MAT_PRODUCT_NEON_H +#endif // CXX11_SRC_FIXEDPOINT_MATMATPRODUCTNEON_H_ diff --git a/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/MatVecProduct.h b/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/MatVecProduct.h index 18b5085b89..ad11d3d44b 100644 --- a/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/MatVecProduct.h +++ b/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/MatVecProduct.h @@ -7,9 +7,8 @@ // Public License v. 2.0. If a copy of the MPL was not distributed // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. -#ifndef EIGEN_CXX11_FIXED_POINT_MAT_VEC_PRODUCT_H -#define EIGEN_CXX11_FIXED_POINT_MAT_VEC_PRODUCT_H - +#ifndef CXX11_SRC_FIXEDPOINT_MATVECPRODUCT_H_ +#define CXX11_SRC_FIXEDPOINT_MATVECPRODUCT_H_ namespace Eigen { namespace internal { @@ -47,6 +46,36 @@ EIGEN_DONT_INLINE void general_matrix_vector_product<Index,QInt8,LhsMapper,ColMa } } +// Mat-Vec product +// Both lhs and rhs are encoded as 16bit signed integers +template <typename Index, typename LhsMapper, bool ConjugateLhs, + typename RhsMapper, bool ConjugateRhs, int Version> +struct general_matrix_vector_product<Index, QInt16, LhsMapper, ColMajor, + ConjugateLhs, QInt16, RhsMapper, + ConjugateRhs, Version> { + EIGEN_DONT_INLINE static void run(Index rows, Index cols, + const LhsMapper& lhs, const RhsMapper& rhs, + QInt32* res, Index resIncr, QInt16 alpha); +}; + +template <typename Index, typename LhsMapper, bool ConjugateLhs, + typename RhsMapper, bool ConjugateRhs, int Version> +EIGEN_DONT_INLINE void general_matrix_vector_product< + Index, QInt16, LhsMapper, ColMajor, ConjugateLhs, QInt16, RhsMapper, + ConjugateRhs, Version>::run(Index rows, Index cols, const LhsMapper& lhs, + const RhsMapper& rhs, QInt32* res, + Index resIncr, QInt16 alpha) { + eigen_assert(alpha.value == 1); + eigen_assert(resIncr == 1); + eigen_assert(rows > 0); + eigen_assert(cols > 0); + + for (Index i = 0; i < rows; ++i) { + for (Index j = 0; j < cols; ++j) { + res[i] += lhs(i, j) * rhs(j, 0); + } + } +} // Mat-Vec product // The lhs is encoded using 8bit signed integers, the rhs using 8bit unsigned integers @@ -118,6 +147,4 @@ EIGEN_DONT_INLINE void general_matrix_vector_product<Index,QUInt8,LhsMapper,ColM } // namespace internal } // namespace Eigen - - -#endif // EIGEN_CXX11_FIXED_POINT_MAT_VEC_PRODUCT_H +#endif // CXX11_SRC_FIXEDPOINT_MATVECPRODUCT_H_ diff --git a/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/PacketMathAVX2.h b/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/PacketMathAVX2.h index cb1636256d..3abd4ee49c 100644 --- a/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/PacketMathAVX2.h +++ b/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/PacketMathAVX2.h @@ -1,6 +1,5 @@ -#ifndef EIGEN3_UNSUPPORTED_EIGEN_CXX11_SRC_FIXEDPOINT_PACKETMATHAVX2_H_ -#define EIGEN3_UNSUPPORTED_EIGEN_CXX11_SRC_FIXEDPOINT_PACKETMATHAVX2_H_ - +#ifndef CXX11_SRC_FIXEDPOINT_PACKETMATHAVX2_H_ +#define CXX11_SRC_FIXEDPOINT_PACKETMATHAVX2_H_ #ifdef _MSC_VER #include <immintrin.h> @@ -29,7 +28,6 @@ inline int _mm256_extract_epi8_N1(const __m256i X) return _mm_extract_epi8(_mm256_extractf128_si256((X), 1 >> 4), 1 % 16); } - namespace Eigen { namespace internal { @@ -502,4 +500,4 @@ struct functor_traits<scalar_product_op<QInt32, double>> { } // end namespace internal } // end namespace Eigen -#endif // EIGEN3_UNSUPPORTED_EIGEN_CXX11_SRC_FIXEDPOINT_PACKETMATHAVX2_H_ +#endif // CXX11_SRC_FIXEDPOINT_PACKETMATHAVX2_H_ diff --git a/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/PacketMathAVX512.h b/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/PacketMathAVX512.h index 8f9906dbf9..2092ce1d4c 100644 --- a/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/PacketMathAVX512.h +++ b/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/PacketMathAVX512.h @@ -1,5 +1,5 @@ -#ifndef EIGEN3_UNSUPPORTED_EIGEN_CXX11_SRC_FIXEDPOINT_PACKETMATHAVX512_H_ -#define EIGEN3_UNSUPPORTED_EIGEN_CXX11_SRC_FIXEDPOINT_PACKETMATHAVX512_H_ +#ifndef CXX11_SRC_FIXEDPOINT_PACKETMATHAVX512_H_ +#define CXX11_SRC_FIXEDPOINT_PACKETMATHAVX512_H_ #include "PacketMathAVX2.h" @@ -542,4 +542,4 @@ EIGEN_STRONG_INLINE QInt8 predux_max<Packet64q8i>(const Packet64q8i& a) { } // end namespace internal } // end namespace Eigen -#endif // EIGEN3_UNSUPPORTED_EIGEN_CXX11_SRC_FIXEDPOINT_PACKETMATHAVX512_H_ +#endif // CXX11_SRC_FIXEDPOINT_PACKETMATHAVX512_H_ diff --git a/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/TypeCastingAVX2.h b/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/TypeCastingAVX2.h index 7b4ecc752f..9561d6a338 100644 --- a/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/TypeCastingAVX2.h +++ b/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/TypeCastingAVX2.h @@ -1,5 +1,5 @@ -#ifndef EIGEN3_UNSUPPORTED_EIGEN_CXX11_SRC_FIXEDPOINT_TYPECASTINGAVX2_H_ -#define EIGEN3_UNSUPPORTED_EIGEN_CXX11_SRC_FIXEDPOINT_TYPECASTINGAVX2_H_ +#ifndef CXX11_SRC_FIXEDPOINT_TYPECASTINGAVX2_H_ +#define CXX11_SRC_FIXEDPOINT_TYPECASTINGAVX2_H_ namespace Eigen { namespace internal { @@ -52,8 +52,16 @@ template <> EIGEN_STRONG_INLINE Packet32q8u pcast<Packet8q32i, Packet32q8u>(const Packet8q32i& a, const Packet8q32i& b, const Packet8q32i& c, const Packet8q32i& d) { + // _mm256_packus_epi32 trims negative numbers to 0 but we can't allow numbers + // that are too large because _mm256_packus_epi16 expects signed input + // (example of problem input: 0x11111111, which saturates to 0xffff = -1, + // which saturates to 0). + const __m256i a_clip = _mm256_min_epi32(a, _mm256_set1_epi32(255)); + const __m256i b_clip = _mm256_min_epi32(b, _mm256_set1_epi32(255)); + const __m256i c_clip = _mm256_min_epi32(c, _mm256_set1_epi32(255)); + const __m256i d_clip = _mm256_min_epi32(d, _mm256_set1_epi32(255)); const __m256i converted = _mm256_packus_epi16( - _mm256_packs_epi32(a.val, b.val), _mm256_packs_epi32(c.val, d.val)); + _mm256_packus_epi32(a_clip, b_clip), _mm256_packus_epi32(c_clip, d_clip)); // Since packus does not cross 128 bit lane boundaries, // we have to permute to properly order the final result. const __m256i permute_mask = _mm256_set_epi32(7, 3, 6, 2, 5, 1, 4, 0); @@ -63,4 +71,4 @@ pcast<Packet8q32i, Packet32q8u>(const Packet8q32i& a, const Packet8q32i& b, } // end namespace internal } // end namespace Eigen -#endif // EIGEN3_UNSUPPORTED_EIGEN_CXX11_SRC_FIXEDPOINT_TYPECASTINGAVX2_H_ +#endif // CXX11_SRC_FIXEDPOINT_TYPECASTINGAVX2_H_ diff --git a/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/TypeCastingAVX512.h b/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/TypeCastingAVX512.h index 26735743d4..a09eac6707 100644 --- a/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/TypeCastingAVX512.h +++ b/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/TypeCastingAVX512.h @@ -1,5 +1,5 @@ -#ifndef EIGEN3_UNSUPPORTED_EIGEN_CXX11_SRC_FIXEDPOINT_TYPECASTINGAVX512_H_ -#define EIGEN3_UNSUPPORTED_EIGEN_CXX11_SRC_FIXEDPOINT_TYPECASTINGAVX512_H_ +#ifndef CXX11_SRC_FIXEDPOINT_TYPECASTINGAVX512_H_ +#define CXX11_SRC_FIXEDPOINT_TYPECASTINGAVX512_H_ namespace Eigen { namespace internal { @@ -132,8 +132,15 @@ pcast<Packet16q32i, Packet64q8i>(const Packet16q32i& a, const Packet16q32i& b, const Packet16q32i& c, const Packet16q32i& d) { - __m512i converted = _mm512_packs_epi16(_mm512_packs_epi32(a.val, b.val), - _mm512_packs_epi32(c.val, d.val)); + __m128i a_part = _mm512_cvtsepi32_epi8(a); + __m128i b_part = _mm512_cvtsepi32_epi8(b); + __m128i c_part = _mm512_cvtsepi32_epi8(c); + __m128i d_part = _mm512_cvtsepi32_epi8(d); + __m256i ab = + _mm256_inserti128_si256(_mm256_castsi128_si256(a_part), b_part, 1); + __m256i cd = + _mm256_inserti128_si256(_mm256_castsi128_si256(c_part), d_part, 1); + __m512i converted = _mm512_inserti64x4(_mm512_castsi256_si512(ab), cd, 1); return converted; } @@ -141,7 +148,10 @@ template <> EIGEN_STRONG_INLINE Packet32q16i pcast<Packet16q32i, Packet32q16i>(const Packet16q32i& a, const Packet16q32i& b) { - __m512i converted = _mm512_packs_epi32(a.val, b.val); + __m256i a_part = _mm512_cvtsepi32_epi16(a); + __m256i b_part = _mm512_cvtsepi32_epi16(b); + __m512i converted = + _mm512_inserti64x4(_mm512_castsi256_si512(a_part), b_part, 1); return converted; } @@ -154,22 +164,45 @@ template <> EIGEN_STRONG_INLINE Packet64q8u pcast<Packet16q32i, Packet64q8u>(const Packet16q32i& a, const Packet16q32i& b, const Packet16q32i& c, const Packet16q32i& d) { - const __m512i converted = _mm512_packus_epi16( - _mm512_packus_epi32(a.val, b.val), _mm512_packus_epi32(c.val, d.val)); + // Brute-force saturation since there isn't a pack operation for unsigned + // numbers that keeps the elements in order. + __m128i a_part = _mm512_cvtepi32_epi8(_mm512_max_epi32( + _mm512_min_epi32(a, _mm512_set1_epi32(255)), _mm512_setzero_si512())); + __m128i b_part = _mm512_cvtepi32_epi8(_mm512_max_epi32( + _mm512_min_epi32(b, _mm512_set1_epi32(255)), _mm512_setzero_si512())); + __m128i c_part = _mm512_cvtepi32_epi8(_mm512_max_epi32( + _mm512_min_epi32(c, _mm512_set1_epi32(255)), _mm512_setzero_si512())); + __m128i d_part = _mm512_cvtepi32_epi8(_mm512_max_epi32( + _mm512_min_epi32(d, _mm512_set1_epi32(255)), _mm512_setzero_si512())); + __m256i ab = + _mm256_inserti128_si256(_mm256_castsi128_si256(a_part), b_part, 1); + __m256i cd = + _mm256_inserti128_si256(_mm256_castsi128_si256(c_part), d_part, 1); + __m512i converted = _mm512_inserti64x4(_mm512_castsi256_si512(ab), cd, 1); return converted; } +#if 0 +// The type Packet32q16u does not exist for AVX-512 yet template <> struct type_casting_traits<QInt32, QUInt16> { enum { VectorizedCast = 1, SrcCoeffRatio = 2, TgtCoeffRatio = 1 }; }; -#if 0 template <> EIGEN_STRONG_INLINE Packet32q16u pcast<Packet16q32i, Packet32q16u>(const Packet16q32i& a, const Packet16q32i& b) { - const __m512i converted = _mm512_packus_epi32(a.val, b.val); + // Brute-force saturation since there isn't a pack operation for unsigned + // numbers that keeps the elements in order. + __m256i a_part = + _mm512_cvtepi32_epi16(_mm512_max_epi32( + _mm512_min_epi32(a, _mm512_set1_epi32(65535)), _mm512_setzero_si512())); + __m256i b_part = _mm512_cvtepi32_epi16( + _mm512_max_epi32(_mm512_min_epi32(b, _mm512_set1_epi32(65535)), + _mm512_setzero_si512())); + __m512i converted = + _mm512_inserti64x4(_mm512_castsi256_si512(a_part), b_part, 1); return converted; } #endif @@ -177,4 +210,4 @@ pcast<Packet16q32i, Packet32q16u>(const Packet16q32i& a, } // end namespace internal } // end namespace Eigen -#endif // EIGEN3_UNSUPPORTED_EIGEN_CXX11_SRC_FIXEDPOINT_TYPECASTINGAVX512_H_ +#endif // CXX11_SRC_FIXEDPOINT_TYPECASTINGAVX512_H_ diff --git a/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/Activations.h b/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/Activations.h deleted file mode 100644 index cbcce9e282..0000000000 --- a/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/Activations.h +++ /dev/null @@ -1,116 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. -// -// Copyright (C) 2015 Benoit Steiner <benoit.steiner.goog@gmail.com> -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. -#ifndef EIGEN_CXX11_NEURAL_NETWORKS_ACTIVATIONS_H -#define EIGEN_CXX11_NEURAL_NETWORKS_ACTIVATIONS_H - -namespace Eigen { - -/** scalar_sigmoid_fast_derivative_op - * \ingroup CXX11_NeuralNetworks_Module - * \brief Template functor to compute the fast derivative of a sigmoid - * - * Input should be the backpropagated gradient. - * - * \sa class CwiseUnaryOp, Cwise::sigmoid_fast_derivative() - */ -template <typename T> -struct scalar_sigmoid_fast_derivative_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_sigmoid_fast_derivative_op) - EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T operator()(const T& y) const { - const T one = T(1); - return (one - y) * y; - } - - template <typename Packet> - inline Packet packetOp(const Packet& y) const { - const Packet one = internal::pset1<Packet>(1); - return internal::pmul(internal::psub(one, y), y); - } -}; - -namespace internal { -template <typename T> -struct functor_traits<scalar_sigmoid_fast_derivative_op<T> > { - enum { - Cost = NumTraits<T>::AddCost * 2 + NumTraits<T>::MulCost, - PacketAccess = packet_traits<T>::HasAdd && packet_traits<T>::HasMul && - packet_traits<T>::HasNegate - }; -}; -} // namespace internal - -/** scalar_tanh_fast_derivative_op - * \ingroup CXX11_NeuralNetworks_Module - * \brief Template functor to compute the fast derivative of a tanh - * - * Input should be the backpropagated gradient. - * - * \sa class CwiseUnaryOp, Cwise::tanh_fast_derivative() - */ -template <typename T> -struct scalar_tanh_fast_derivative_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_tanh_fast_derivative_op) - EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T operator()(const T& y) const { - const T one = T(1); - return one - (y * y); - } - - template <typename Packet> - inline Packet packetOp(const Packet& y) const { - const Packet one = internal::pset1<Packet>(1); - return internal::psub(one, internal::pmul(y, y)); - } -}; - -namespace internal { -template <typename T> -struct functor_traits<scalar_tanh_fast_derivative_op<T> > { - enum { - Cost = NumTraits<T>::AddCost * 2 + NumTraits<T>::MulCost * 1, - PacketAccess = packet_traits<T>::HasAdd && packet_traits<T>::HasMul && - packet_traits<T>::HasNegate - }; -}; -} // namespace internal - -/** - * \ingroup CXX11_NeuralNetworks_Module - * \brief Template functor to clip the magnitude of the first scalar. - * - * \sa class CwiseBinaryOp, MatrixBase::Clip - */ -template <typename Scalar> -struct scalar_clip_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_clip_op) - EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar - operator()(const Scalar& a, const Scalar& b) const { - return numext::mini(numext::maxi(a, -b), b); - } - template <typename Packet> - EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet - packetOp(const Packet& a, const Packet& b) const { - return internal::pmin(internal::pmax(a, internal::pnegate(b)), b); - } -}; - -namespace internal { -template <typename Scalar> -struct functor_traits<scalar_clip_op<Scalar> > { - enum { - Cost = NumTraits<Scalar>::AddCost * 3, - PacketAccess = packet_traits<Scalar>::HasMax && - packet_traits<Scalar>::HasMin && - packet_traits<Scalar>::HasNegate - }; -}; -} // namespace internal - -} // end namespace Eigen - -#endif // EIGEN_CXX11_NEURAL_NETWORKS_ACTIVATIONS_H diff --git a/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/Attention.h b/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/Attention.h deleted file mode 100644 index d4bc7a3515..0000000000 --- a/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/Attention.h +++ /dev/null @@ -1,209 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. -// -// Copyright (C) 2015 Benoit Steiner <benoit.steiner.goog@gmail.com> -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. -#ifndef EIGEN_CXX11_NEURAL_NETWORKS_ATTENTION_H -#define EIGEN_CXX11_NEURAL_NETWORKS_ATTENTION_H - -namespace Eigen { - -/** ExtractGlimpses - * \ingroup CXX11_NeuralNetworks_Module - * - * \brief Extract glimpses from an input tensor. - * - * The input parameter is expected to be a col-major tensor with a rank of 4 (depth, x, y, and batch). - * The width and height parameters specify the extension of the returned glimpses. - * The offsets parameter specifies the x, y locations of the center of the glimpses relative to the center of the input image. The vector is expected to contain one IndexPair for each image in the batch dimension. - * The normalized boolean indicates if incoming coordinates are normalized so that 0.0 and 1.0 correspond to the minimum and maximum of each height and width dimension. - * The centered boolean indicates if incoming coordinates are centered relative to the image, in which case -1.0 and 1.0 correspond to minimum and maximum of each dimension while 0.0 corresponds to the center. - * - * The result can be assigned to a tensor of rank equal to that of the input. The result will be laid out in col-major order (depth, x, y, batch). - * The dimensions of the result will be equal to the dimensions of the input except for width and height which will be equal to the requested glimpse size. - */ -namespace { -template <typename Index> -struct GlimpseExtractionOp { - GlimpseExtractionOp(const Index width, const Index height, - const std::vector<IndexPair<float> >& offsets, - const bool normalized, - const bool centered, - const bool uniform_noise) : - width_(width), height_(height), offsets_(offsets), - normalized_(normalized), centered_(centered), uniform_noise_(uniform_noise) { } - - template <typename Input> - DSizes<Index, 4> dimensions(const Input& input) const { - typedef typename internal::traits<Input>::Index IndexType; - typedef TensorRef<Tensor<typename internal::traits<Input>::Scalar, 4, - internal::traits<Input>::Layout, IndexType> > Ref; - Ref in(input); - - DSizes<Index, 4> dims = in.dimensions(); - - dims[0] = in.dimension(0); - dims[1] = width_; - dims[2] = height_; - dims[3] = in.dimension(3); - return dims; - } - - template <typename Input, typename Output, typename Device> - EIGEN_DEVICE_FUNC - void eval(const Input& input, Output& output, const Device& device) const - { - typedef typename internal::traits<Input>::Index IndexType; - typedef TensorRef<Tensor<typename internal::traits<Input>::Scalar, 4, - internal::traits<Input>::Layout, IndexType> > Ref; - Ref in(input); - - const Index num_channels = in.dimension(0); - const Index input_width = in.dimension(1); - const Index input_height = in.dimension(2); - const Index batch_size = in.dimension(3); - eigen_assert(input_width > 0); - eigen_assert(input_height > 0); - - for (Index i = 0; i < batch_size; ++i) { - float x = offsets_[i].first, y = offsets_[i].second; - - // Un-normalize coordinates back to pixel space if normalized. - if (normalized_) { - x *= input_width; - y *= input_height; - } - // Un-center if coordinates are centered on the image center. - if (centered_) { - x /= 2.0f; - y /= 2.0f; - x += input_width / 2.0f; - y += input_height / 2.0f; - } - // Remove half of the glimpse window. - x -= width_ / 2.0f; - y -= height_ / 2.0f; - - const Index offset_x = (Index) x; - const Index offset_y = (Index) y; - Index glimpse_width = width_; - Index glimpse_height = height_; - bool partial_overlap = false; - DSizes<Index, 3> slice_offset(0, offset_x, offset_y); - DSizes<Index, 3> slice_extent(num_channels, width_, height_); - DSizes<Index, 3> base_offset(0, 0, 0); - - if (offset_x < 0) { - slice_offset[1] = 0; - glimpse_width = (std::max<Index>)(0, width_ + offset_x); - slice_extent[1] = glimpse_width; - base_offset[1] = width_ - glimpse_width; - partial_overlap = true; - } else if (offset_x + width_ >= input_width) { - glimpse_width = (std::max<Index>)(0, input_width - offset_x); - slice_extent[1] = glimpse_width; - partial_overlap = true; - } - if (offset_y < 0) { - slice_offset[2] = 0; - glimpse_height = (std::max<Index>)(0, height_ + offset_y); - slice_extent[2] = glimpse_height; - base_offset[2] = height_ - glimpse_height; - partial_overlap = true; - } else if (offset_y + height_ >= input_height) { - glimpse_height = (std::max<Index>)(0, input_height - offset_y); - slice_extent[2] = glimpse_height; - partial_overlap = true; - } - slice_extent[1] = std::min<Index>(input_width, slice_extent[1]); - slice_extent[2] = std::min<Index>(input_height, slice_extent[2]); - - if (partial_overlap) { - if (uniform_noise_) { - // Initialize the glimpse with uniform noise. - typedef typename internal::remove_const< - typename internal::traits<Input>::Scalar>::type Scalar; - TensorFixedSize<Scalar, Sizes<> > mini; - mini.device(device) = input.template chip<3>(i).minimum(); - TensorFixedSize<float, Sizes<> > range; - range.device(device) = - (input.template chip<3>(i).maximum() - mini).template cast<float>(); - - DSizes<Index, 3> glimpse_size(num_channels, width_, height_); - TensorMap<Tensor<float, 3> > tmp(NULL, glimpse_size); - output.template chip<3>(i).device(device) = - mini.reshape(Sizes<1,1,1>()).broadcast(glimpse_size) + - (tmp.random() * range.reshape(Sizes<1,1,1>()).broadcast(glimpse_size)).template cast<Scalar>(); - } else { - // Initialize the glimpse with white noise: compute the mean and sigma - // of each channel, and use them to shape the gaussian. - DSizes<Index, 2> glimpse_size(width_, height_); - DSizes<Index, 2> input_size(input_width, input_height); - typedef typename internal::remove_const< - typename internal::traits<Input>::Scalar>::type Scalar; - - for (int j = 0; j < num_channels; ++j) { - TensorFixedSize<Scalar, Sizes<> > mean; - mean.device(device) = input.template chip<3>(i).template chip<0>(j).template cast<float>().mean(); - TensorFixedSize<float, Sizes<> > sigma; - sigma.device(device) = - (input.template chip<3>(i).template chip<0>(j).template cast<float>() - mean.reshape(Sizes<1,1>()).broadcast(input_size)).square().mean().sqrt(); - TensorFixedSize<Scalar, Sizes<> > mini; - mini.device(device) = input.template chip<3>(i).template chip<0>(j).minimum(); - TensorFixedSize<float, Sizes<> > maxi; - maxi.device(device) = input.template chip<3>(i).template chip<0>(j).maximum(); - - TensorMap<Tensor<float, 2> > tmp(NULL, glimpse_size); - output.template chip<3>(i).template chip<0>(j).device(device) = - (mean.reshape(Sizes<1,1>()).broadcast(glimpse_size) + - (tmp.random(internal::NormalRandomGenerator<float>()) * sigma.reshape(Sizes<1,1>()).broadcast(glimpse_size)).template cast<Scalar>()).cwiseMin(maxi.reshape(Sizes<1,1>()).broadcast(glimpse_size)).cwiseMax(mini.reshape(Sizes<1,1>()).broadcast(glimpse_size)); - } - } - - // Copy the part of the glimpse that cover the input image if any. - if (glimpse_width == 0 || glimpse_height == 0) { - continue; - } - output.template chip<3>(i).slice(base_offset, slice_extent).device(device) = input.template chip<3>(i).slice(slice_offset, slice_extent); - } else { - output.template chip<3>(i).device(device) = input.template chip<3>(i).slice(slice_offset, slice_extent); - } - } - } - - private: - const Index width_; - const Index height_; - const std::vector<IndexPair<float> > offsets_; - const bool normalized_; - const bool centered_; - const bool uniform_noise_; -}; -} - - -template <typename Input> -EIGEN_ALWAYS_INLINE -static const TensorCustomUnaryOp<const GlimpseExtractionOp<typename internal::traits<Input>::Index>, const Input> -ExtractGlimpses(const Input& input, - const typename internal::traits<Input>::Index width, - const typename internal::traits<Input>::Index height, - const std::vector<IndexPair<float> >& offsets, - const bool normalized = true, const bool centered = true, - const bool uniform_noise = true) -{ - EIGEN_STATIC_ASSERT(internal::traits<Input>::Layout == ColMajor, YOU_MADE_A_PROGRAMMING_MISTAKE); - EIGEN_STATIC_ASSERT(internal::traits<Input>::NumDimensions == 4, YOU_MADE_A_PROGRAMMING_MISTAKE); - - typedef typename internal::traits<Input>::Index Index; - const GlimpseExtractionOp<Index> op(width, height, offsets, normalized, - centered, uniform_noise); - return input.customOp(op); -} - -} // end namespace Eigen - -#endif // EIGEN_CXX11_NEURAL_NETWORKS_ATTENTION_H diff --git a/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/BackwardCuboidConvolutions.h b/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/BackwardCuboidConvolutions.h deleted file mode 100644 index 12ce23444c..0000000000 --- a/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/BackwardCuboidConvolutions.h +++ /dev/null @@ -1,523 +0,0 @@ -#ifndef EIGEN_CXX11_NEURAL_NETWORKS_BACKWARD_CUBOID_CONVOLUTIONS_H -#define EIGEN_CXX11_NEURAL_NETWORKS_BACKWARD_CUBOID_CONVOLUTIONS_H - -#include "Patch3d.h" - -namespace Eigen { - -/** CuboidConvolutionBackwardInput - * \ingroup CXX11_NeuralNetworks_Module - * - * \brief Computes the backprop for the input of a 3D convolution. - * - * The output_backward parameter is expected to be a tensor with a rank of 4 or more (channels, depth, height, width, and optionally others) - * The kernel parameter is expected to be a 5D tensor (filters, channels, kernel_depth, kernel_height, kernel_width) - * output_backward and kernel have to be in the same layout. - * - * The dimensions of the result will be filters, depth, height, width (and others if applicable). - * - * It is possible to swap the order of the depth, width and height dimensions provided that the same order is used in the input, the kernel, and the output. - * - * All dimension orders above are given for col-major, and should be reversed for row-major. - */ - -template <typename OutputBackward, typename Kernel> -EIGEN_ALWAYS_INLINE static const typename internal::conditional< - internal::traits<OutputBackward>::Layout == ColMajor, - TensorReshapingOp< - const DSizes<typename internal::traits<OutputBackward>::Index, - internal::traits<OutputBackward>::NumDimensions>, - const TensorContractionOp< - const array< IndexPair<typename internal::traits<OutputBackward>::Index>, 2>, - const TensorReshapingOp< - const DSizes< typename internal::traits<OutputBackward>::Index, 3>, - const TensorReverseOp<const array<bool, 5>, const Kernel> - >, - const TensorReshapingOp< - const DSizes< typename internal::traits<OutputBackward>::Index, 3>, - const TensorVolumePatchOp<Dynamic, Dynamic, Dynamic, const OutputBackward> - > - > - >, - TensorReshapingOp< - const DSizes<typename internal::traits<OutputBackward>::Index, - internal::traits<OutputBackward>::NumDimensions>, - const TensorContractionOp< - const array< IndexPair<typename internal::traits<OutputBackward>::Index>, 2>, - const TensorReshapingOp< - const DSizes< typename internal::traits<OutputBackward>::Index, 3>, - const TensorVolumePatchOp<Dynamic, Dynamic, Dynamic, const OutputBackward> - >, - const TensorReshapingOp< - const DSizes<typename internal::traits<OutputBackward>::Index, 3>, - const TensorReverseOp<const array<bool, 5>, const Kernel> - > - > - > ->::type -CuboidConvolutionBackwardInput( - const Kernel& kernel, const OutputBackward& output_backward, - typename internal::traits<OutputBackward>::Index inputPlanes, - typename internal::traits<OutputBackward>::Index inputRows, - typename internal::traits<OutputBackward>::Index inputCols, - const DenseIndex stridePlanes = 1, const DenseIndex strideRows = 1, - const DenseIndex strideCols = 1) { - typedef typename internal::traits<OutputBackward>::Index TensorIndex; - const TensorRef<const Tensor<typename internal::traits<Kernel>::Scalar, internal::traits<Kernel>::NumDimensions, internal::traits<Kernel>::Layout, TensorIndex> > kern(kernel); - const TensorRef<const Tensor<typename internal::traits<OutputBackward>::Scalar, internal::traits<OutputBackward>::NumDimensions, internal::traits<OutputBackward>::Layout, TensorIndex> > out(output_backward); - - EIGEN_STATIC_ASSERT(internal::traits<Kernel>::Layout == internal::traits<OutputBackward>::Layout, YOU_MADE_A_PROGRAMMING_MISTAKE); - - static const bool isColMajor = (internal::traits<OutputBackward>::Layout == ColMajor); - - static const int NumDims = internal::traits<OutputBackward>::NumDimensions; - - // Number of filters to apply. This is the same as the output depth of the result - const TensorIndex kernelFilters = isColMajor ? kern.dimensions()[0] : kern.dimensions()[4]; - // Number of channels. This is the same as the input depth. - const TensorIndex kernelChannels = isColMajor ? kern.dimensions()[1] : kern.dimensions()[3]; - const TensorIndex kernelPlanes = isColMajor ? kern.dimensions()[2] : kern.dimensions()[2]; - const TensorIndex kernelRows = isColMajor ? kern.dimensions()[3] : kern.dimensions()[1]; - const TensorIndex kernelCols = isColMajor ? kern.dimensions()[4] : kern.dimensions()[0]; - - const TensorIndex outputPlanes = isColMajor ? out.dimensions()[1] : out.dimensions()[NumDims - 2]; - const TensorIndex outputRows = isColMajor ? out.dimensions()[2] : out.dimensions()[NumDims - 3]; - const TensorIndex outputCols = isColMajor ? out.dimensions()[3] : out.dimensions()[NumDims - 4]; - - TensorIndex forward_pad_z, forward_pad_y, forward_pad_x; - const TensorIndex size_z = ceil(inputPlanes / static_cast<float>(stridePlanes)); - const TensorIndex size_y = ceil(inputRows / static_cast<float>(strideRows)); - const TensorIndex size_x = ceil(inputCols / static_cast<float>(strideCols)); - - // Infer padding type. - if (size_z == outputPlanes && size_y == outputRows && size_x == outputCols) { - // SAME padding. - const TensorIndex dz = size_z * stridePlanes + kernelPlanes - 1 - inputPlanes; - const TensorIndex dy = size_y * strideRows + kernelRows - 1 - inputRows; - const TensorIndex dx = size_x * strideCols + kernelCols - 1 - inputCols; - - forward_pad_z = dz - dz / 2; - forward_pad_y = dy - dy / 2; - forward_pad_x = dx - dx / 2; - } else { - // VALID padding. - forward_pad_z = 0; - forward_pad_y = 0; - forward_pad_x = 0; - } - const TensorIndex padding_ztop = kernelPlanes - 1 - forward_pad_z; - const TensorIndex padding_top = kernelRows - 1 - forward_pad_y; - const TensorIndex padding_left = kernelCols - 1 - forward_pad_x; - - const TensorIndex padding_zbottom = inputPlanes + kernelPlanes - 1 - (outputPlanes - 1) * stridePlanes - 1 - padding_ztop; - const TensorIndex padding_bottom = inputRows + kernelRows - 1 - (outputRows - 1) * strideRows - 1 - padding_top; - const TensorIndex padding_right = inputCols + kernelCols - 1 - (outputCols - 1) * strideCols - 1 - padding_left; - - eigen_assert(padding_ztop >= 0); - eigen_assert(padding_zbottom >= 0); - eigen_assert(padding_top >= 0); - eigen_assert(padding_left >= 0); - eigen_assert(padding_bottom >= 0); - eigen_assert(padding_right >= 0); - - // The kernel has dimensions filters X channels X patch_planes X patch_rows X patch_cols. - // We need to reverse the kernel along the spatial dimensions. - array<bool, 5> kernel_reverse; - if (isColMajor) { - kernel_reverse[0] = false; - kernel_reverse[1] = false; - kernel_reverse[2] = true; - kernel_reverse[3] = true; - kernel_reverse[4] = true; - } else { - kernel_reverse[0] = true; - kernel_reverse[1] = true; - kernel_reverse[2] = true; - kernel_reverse[3] = false; - kernel_reverse[4] = false; - } - - DSizes<TensorIndex, 3> kernel_dims; - if (isColMajor) { - kernel_dims[0] = kernelFilters; - kernel_dims[1] = kernelChannels; - kernel_dims[2] = kernelRows * kernelCols * kernelPlanes; - } else { - kernel_dims[0] = kernelRows * kernelCols * kernelPlanes; - kernel_dims[1] = kernelChannels; - kernel_dims[2] = kernelFilters; - } - - // The output_backward has dimensions out_depth X out_planes X out_rows X out_cols X OTHERS - // When we extract the image patches from output_backward, it will have dimensions: - // out_depth X (patch_planes * patch_rows * patch_cols) X (input_planes * input_rows * input_cols * OTHERS) - DSizes<TensorIndex, 3> pre_contract_dims; - if (isColMajor) { - pre_contract_dims[0] = kernelFilters; - pre_contract_dims[1] = kernelRows * kernelCols * kernelPlanes; - pre_contract_dims[2] = inputRows * inputCols * inputPlanes; - for (int i = 4; i < NumDims; ++i) { - pre_contract_dims[2] *= out.dimension(i); - } - } else { - pre_contract_dims[2] = kernelFilters; - pre_contract_dims[1] = kernelRows * kernelCols * kernelPlanes; - pre_contract_dims[0] = inputRows * inputCols * inputPlanes; - for (int i = 0; i < NumDims - 4; ++i) { - pre_contract_dims[0] *= out.dimension(i); - } - } - - // We will contract along dimensions (0, 2) in kernel and (0, 1) in - // output_backward, if this is col-major, and - // dimensions (0, 2) in kernel and (1, 2) in output_backward, if this row-major. - array<IndexPair<TensorIndex>, 2> contract_dims; - if (isColMajor) { - // col-major: kernel.contract(output.patches) - contract_dims[0] = IndexPair<TensorIndex>(0, 0); - contract_dims[1] = IndexPair<TensorIndex>(2, 1); - } else { - // row-major: output.patches.contract(kernel) - contract_dims[0] = IndexPair<TensorIndex>(1, 0); - contract_dims[1] = IndexPair<TensorIndex>(2, 2); - } - - // Post contraction, the dimensions of the input_backprop is - // channels X input_planes X input_rows X input_cols X OTHERS - DSizes<TensorIndex, NumDims> post_contract_dims; - if (isColMajor) { - post_contract_dims[0] = kernelChannels; - post_contract_dims[1] = inputPlanes; - post_contract_dims[2] = inputRows; - post_contract_dims[3] = inputCols; - for (int i = 4; i < NumDims; ++i) { - post_contract_dims[i] = out.dimension(i); - } - } else { - post_contract_dims[NumDims - 1] = kernelChannels; - post_contract_dims[NumDims - 2] = inputPlanes; - post_contract_dims[NumDims - 3] = inputRows; - post_contract_dims[NumDims - 4] = inputCols; - for (int i = 0; i < NumDims - 4; ++i) { - post_contract_dims[i] = out.dimension(i); - } - } - - DSizes<TensorIndex, NumDims> strides; - for (int i = 0; i < NumDims; i++) { - strides[i] = 1; - } - if (isColMajor) { - strides[1] = stridePlanes; - strides[2] = strideRows; - strides[3] = strideCols; - } else { - strides[NumDims - 2] = stridePlanes; - strides[NumDims - 3] = strideRows; - strides[NumDims - 4] = strideCols; - } - - return choose( - Cond<internal::traits<OutputBackward>::Layout == ColMajor>(), - kernel.reverse(kernel_reverse) - .reshape(kernel_dims) - .contract( - output_backward.extract_volume_patches(kernelPlanes, kernelRows, kernelCols, - 1, 1, 1, stridePlanes, strideRows, strideCols, - padding_ztop, padding_zbottom, - padding_top, padding_bottom, - padding_left, padding_right) - .reshape(pre_contract_dims), - contract_dims) - .reshape(post_contract_dims), - output_backward.extract_volume_patches(kernelPlanes, kernelRows, kernelCols, - 1, 1, 1, stridePlanes, strideRows, strideCols, - padding_ztop, padding_zbottom, - padding_top, padding_bottom, - padding_left, padding_right) - .reshape(pre_contract_dims) - .contract(kernel.reverse(kernel_reverse).reshape(kernel_dims), - contract_dims) - .reshape(post_contract_dims)); -} - - -/** CuboidConvolutionBackwardKernel - * \ingroup CXX11_NeuralNetworks_Module - * - * \brief Computes the backprop for the filter of a 3D convolution. - * - * The output_backward parameter is expected to be a tensor with a rank of 4 or more (channels, depth, height, width, and optionally others) - * The kernel parameter is expected to be a 4D tensor (filters, channels, kernel_depth, kernel_height, kernel_width) - * output_backward and kernel have to be in the same layout. - * - * The dimensions of the result will be filters, depth, height, width (and others if applicable). - * - * It is possible to swap the order of the depth, width and height dimensions provided that the same order is used in the input, the kernel, and the output. - * - * All dimension orders above are given for col-major, and should be reversed for row-major. - */ -template <typename OutputBackward, typename Input> -EIGEN_ALWAYS_INLINE static const typename internal::conditional< - internal::traits<OutputBackward>::Layout == ColMajor, - const TensorShufflingOp< - const array<typename internal::traits<OutputBackward>::Index, 5>, - const TensorReverseOp< - const array<bool, 5>, - const TensorReshapingOp< - const DSizes<typename internal::traits<OutputBackward>::Index, 5>, - const TensorContractionOp< - const array< IndexPair<typename internal::traits<Input>::Index>, 2>, - const TensorReshapingOp< - const DSizes<typename internal::traits<Input>::Index, 3>, - const Input>, - const TensorReshapingOp< - const DSizes< typename internal::traits<OutputBackward>::Index, 4>, - const TensorVolumePatchOp<Dynamic, Dynamic, Dynamic, const OutputBackward> - > - > - > - > - >, - const TensorShufflingOp< - const array<typename internal::traits<OutputBackward>::Index, 5>, - const TensorReverseOp< - const array<bool, 5>, - const TensorReshapingOp< - const DSizes<typename internal::traits<OutputBackward>::Index, 5>, - const TensorContractionOp< - const array< IndexPair<typename internal::traits<Input>::Index>, 2>, - const TensorReshapingOp< - const DSizes< typename internal::traits<OutputBackward>::Index, 4>, - const TensorVolumePatchOp<Dynamic, Dynamic, Dynamic, const OutputBackward> - >, - const TensorReshapingOp< - const DSizes<typename internal::traits<Input>::Index, 3>, - const Input - > - > - > - > - > ->::type -CuboidConvolutionBackwardKernel( - const Input& input, const OutputBackward& output_backward, - typename internal::traits<Input>::Index kernelPlanes, - typename internal::traits<Input>::Index kernelRows, - typename internal::traits<Input>::Index kernelCols, - const DenseIndex stridePlanes = 1, - const DenseIndex strideRows = 1, - const DenseIndex strideCols = 1) { - typedef typename internal::traits<Input>::Index TensorIndex; - TensorRef<Tensor<typename internal::traits<Input>::Scalar, internal::traits<Input>::NumDimensions, internal::traits<Input>::Layout, TensorIndex> > in(input); - TensorRef<Tensor<typename internal::traits<OutputBackward>::Scalar, internal::traits<OutputBackward>::NumDimensions, internal::traits<OutputBackward>::Layout, TensorIndex> > out(output_backward); - - EIGEN_STATIC_ASSERT(internal::traits<Input>::Layout == internal::traits<OutputBackward>::Layout, YOU_MADE_A_PROGRAMMING_MISTAKE); - - static const bool isColMajor = (internal::traits<Input>::Layout == ColMajor); - - static const int NumDims = internal::traits<Input>::NumDimensions; - EIGEN_STATIC_ASSERT(internal::traits<Input>::NumDimensions == internal::traits<OutputBackward>::NumDimensions, YOU_MADE_A_PROGRAMMING_MISTAKE); - - const TensorIndex inputPlanes = isColMajor ? in.dimension(1) : in.dimension(NumDims - 2); - const TensorIndex inputRows = isColMajor ? in.dimension(2) : in.dimension(NumDims - 3); - const TensorIndex inputCols = isColMajor ? in.dimension(3) : in.dimension(NumDims - 4); - - const TensorIndex outputPlanes = isColMajor ? out.dimension(1) : out.dimension(NumDims - 2); - const TensorIndex outputRows = isColMajor ? out.dimension(2) : out.dimension(NumDims - 3); - const TensorIndex outputCols = isColMajor ? out.dimension(3) : out.dimension(NumDims - 4); - - const TensorIndex kernelFilters = isColMajor ? out.dimension(0) : out.dimension(NumDims - 1); - const TensorIndex kernelChannels = isColMajor ? in.dimension(0) : in.dimension(NumDims - 1); - - TensorIndex forward_pad_z, forward_pad_y, forward_pad_x; - const TensorIndex size_z = ceil(inputPlanes / static_cast<float>(stridePlanes)); - const TensorIndex size_y = ceil(inputRows / static_cast<float>(strideRows)); - const TensorIndex size_x = ceil(inputCols / static_cast<float>(strideCols)); - - // Infer padding type. - if (size_z == outputPlanes && size_y == outputRows && size_x == outputCols) { - // SAME padding. - const TensorIndex dz = size_z * stridePlanes + kernelPlanes - 1 - inputPlanes; - const TensorIndex dy = size_y * strideRows + kernelRows - 1 - inputRows; - const TensorIndex dx = size_x * strideCols + kernelCols - 1 - inputCols; - - forward_pad_z = dz - dz / 2; - forward_pad_y = dy - dy / 2; - forward_pad_x = dx - dx / 2; - } else { - // VALID padding. - forward_pad_z = 0; - forward_pad_y = 0; - forward_pad_x = 0; - } - - const TensorIndex padding_ztop = kernelPlanes - 1 - forward_pad_z; - const TensorIndex padding_top = kernelRows - 1 - forward_pad_y; - const TensorIndex padding_left = kernelCols - 1 - forward_pad_x; - - const TensorIndex padding_zbottom = inputPlanes + kernelPlanes - 1 - (outputPlanes - 1) * stridePlanes - 1 - padding_ztop; - const TensorIndex padding_bottom = inputRows + kernelRows - 1 - (outputRows - 1) * strideRows - 1 - padding_top; - const TensorIndex padding_right = inputCols + kernelCols - 1 - (outputCols - 1) * strideCols - 1 - padding_left; - - eigen_assert(padding_ztop >= 0); - eigen_assert(padding_zbottom >= 0); - eigen_assert(padding_top >= 0); - eigen_assert(padding_left >= 0); - eigen_assert(padding_bottom >= 0); - eigen_assert(padding_right >= 0); - - // The output_backward has dimensions out_depth X out_plaens X out_rows X out_cols X OTHERS - // When we extract the image patches from output_backward (with input as the - // kernel), it will have dimensions - // (out_depth) X (input_planes * input_rows * input_cols) X (kernel_planes * kernel_rows * kernel_cols) X OTHERS - DSizes<TensorIndex, 4> pre_contract_dims; - if (isColMajor) { - pre_contract_dims[0] = kernelFilters; - pre_contract_dims[1] = inputRows * inputCols * inputPlanes; - pre_contract_dims[2] = kernelRows * kernelCols * kernelPlanes; - pre_contract_dims[3] = 1; - for (int i = 4; i < NumDims; ++i) { - pre_contract_dims[3] *= out.dimension(i); - } - } else { - pre_contract_dims[3] = kernelFilters; - pre_contract_dims[2] = inputRows * inputCols * inputPlanes; - pre_contract_dims[1] = kernelRows * kernelCols * kernelPlanes; - pre_contract_dims[0] = 1; - for (int i = 0; i < NumDims - 4; ++i) { - pre_contract_dims[0] *= out.dimension(i); - } - } - - // The input has dimensions in_depth X (input_planes * input_rows * input_cols) X OTHERS - DSizes<TensorIndex, 3> input_dims; - if (isColMajor) { - input_dims[0] = kernelChannels; - input_dims[1] = inputRows * inputCols * inputPlanes; - input_dims[2] = 1; - for (int i = 4; i < NumDims; ++i) { - input_dims[2] *= in.dimension(i); - } - eigen_assert(input_dims[2] == pre_contract_dims[3]); - } else { - input_dims[2] = kernelChannels; - input_dims[1] = inputRows * inputCols * inputPlanes; - input_dims[0] = 1; - for (int i = 0; i < NumDims - 4; ++i) { - input_dims[0] *= in.dimension(i); - } - eigen_assert(input_dims[0] == pre_contract_dims[0]); - } - - // We will contract along dimensions (1, 2) in in and (1, 3) in out, if - // this is col-major. - // For row-major, it's dimensions (0, 1) in in and (0, 2) in out. - array<IndexPair<TensorIndex>, 2> contract_dims; - if (isColMajor) { - // col-major: in.contract(output.patches) - contract_dims[0] = IndexPair<TensorIndex>(1, 1); - contract_dims[1] = IndexPair<TensorIndex>(2, 3); - } else { - // row-major: output.patches.contract(in) - contract_dims[0] = IndexPair<TensorIndex>(0, 0); - contract_dims[1] = IndexPair<TensorIndex>(2, 1); - } - - // After the contraction, the kernel will have dimension - // in_depth X out_depth X kernel_patches X kernel_rows X kernel_cols - // We will need to shuffle the first two dimensions and reverse the spatial dimensions. - // The end shape is: - // out_depth X in_shape X kernel_planes X kernel_rows X kernel_cols - - // This is the shape of the kernel *before* the shuffling. - DSizes<TensorIndex, 5> kernel_dims; - if (isColMajor) { - kernel_dims[0] = kernelChannels; - kernel_dims[1] = kernelFilters; - kernel_dims[2] = kernelPlanes; - kernel_dims[3] = kernelRows; - kernel_dims[4] = kernelCols; - } else { - kernel_dims[0] = kernelCols; - kernel_dims[1] = kernelRows; - kernel_dims[2] = kernelPlanes; - kernel_dims[3] = kernelFilters; - kernel_dims[4] = kernelChannels; - } - - // Flip filters and channels. - array<TensorIndex, 5> kernel_shuffle; - if (isColMajor) { - kernel_shuffle[0] = 1; - kernel_shuffle[1] = 0; - kernel_shuffle[2] = 2; - kernel_shuffle[3] = 3; - kernel_shuffle[4] = 4; - } else { - kernel_shuffle[0] = 0; - kernel_shuffle[1] = 1; - kernel_shuffle[2] = 2; - kernel_shuffle[3] = 4; - kernel_shuffle[4] = 3; - } - - // Reverse the spatial dimensions. - array<bool, 5> kernel_reverse; - if (isColMajor) { - kernel_reverse[0] = false; - kernel_reverse[1] = false; - kernel_reverse[2] = true; - kernel_reverse[3] = true; - kernel_reverse[4] = true; - } else { - kernel_reverse[0] = true; - kernel_reverse[1] = true; - kernel_reverse[2] = true; - kernel_reverse[3] = false; - kernel_reverse[4] = false; - } - - DSizes<TensorIndex, NumDims> strides; - for (int i = 0; i < NumDims; i++) { - strides[i] = 1; - } - if (isColMajor) { - strides[1] = stridePlanes; - strides[2] = strideRows; - strides[3] = strideCols; - } else { - strides[NumDims - 2] = stridePlanes; - strides[NumDims - 3] = strideRows; - strides[NumDims - 4] = strideCols; - } - return choose( - Cond<internal::traits<Input>::Layout == ColMajor>(), - input.reshape(input_dims) - .contract( - output_backward.extract_volume_patches( - inputPlanes, inputRows, inputCols, 1, - 1, 1, stridePlanes, strideRows, strideCols, - - padding_ztop, padding_zbottom, padding_top, - padding_bottom, padding_left, padding_right) - .reshape(pre_contract_dims), - contract_dims) - .reshape(kernel_dims) - .reverse(kernel_reverse) - .shuffle(kernel_shuffle), - output_backward.extract_volume_patches( - inputPlanes, inputRows, inputCols, 1, 1, 1, - stridePlanes, strideRows, strideCols, padding_ztop, - padding_zbottom, padding_top, padding_bottom, - padding_left, padding_right) - .reshape(pre_contract_dims) - .contract(input.reshape(input_dims), contract_dims) - .reshape(kernel_dims) - .reverse(kernel_reverse) - .shuffle(kernel_shuffle)); -} - -} // end namespace Eigen - -#endif // EIGEN_CXX11_NEURAL_NETWORKS_BACKWARD_CUBOID_CONVOLUTIONS_H diff --git a/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/BackwardSpatialConvolutions.h b/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/BackwardSpatialConvolutions.h deleted file mode 100644 index 188dc75bf6..0000000000 --- a/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/BackwardSpatialConvolutions.h +++ /dev/null @@ -1,351 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. -// -// Copyright (C) 2015 Ke Yang <yangke@gmail.com> -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#ifndef EIGEN_CXX11_NEURAL_NETWORKS_BACKWARD_SPATIAL_CONVOLUTIONS_H -#define EIGEN_CXX11_NEURAL_NETWORKS_BACKWARD_SPATIAL_CONVOLUTIONS_H - -namespace Eigen { - -/** SpatialConvolutionBackwardInput - * \ingroup CXX11_NeuralNetworks_Module - * - * \brief Computes the backprop for the input of a 2D convolution. - * - * The output_backward parameter is expected to be a tensor with a rank of 3 or more (channels, height, width, and optionally others) - * The kernel parameter is expected to be a 4D tensor (filters, channels, kernel_height, kernel_width) - * The output_backward and the kernel must both be in col-major layout. The result will also be in col-major layout. - * - * If in_stride > 1, then applies convolution with holes (aka atrous convolution), sampling every in_stride input pixels. - * - * The result can be assigned to a tensor of rank equal to the rank of the output_backward. The dimensions of the result will be filters, height, width (and others if applicable). - * - * It is possible to swap the order of the width and height dimensions provided that the same order is used in the input, the kernel, and the output. - * - */ - -template <typename OutputBackward, typename Kernel> -EIGEN_ALWAYS_INLINE -static const typename internal::conditional< - internal::traits<OutputBackward>::Layout == ColMajor, - TensorReshapingOp<const DSizes<typename internal::traits<OutputBackward>::Index, internal::traits<OutputBackward>::NumDimensions>, const TensorContractionOp<const array<IndexPair<typename internal::traits<OutputBackward>::Index>, 2>, const TensorReshapingOp<const DSizes<typename internal::traits<OutputBackward>::Index, 3>, const TensorReverseOp<const array<bool, 4>, const Kernel> >, const TensorReshapingOp<const DSizes<typename internal::traits<OutputBackward>::Index, 3>, const TensorImagePatchOp<Dynamic, Dynamic, const OutputBackward> > > >, - TensorReshapingOp<const DSizes<typename internal::traits<OutputBackward>::Index, internal::traits<OutputBackward>::NumDimensions>, const TensorContractionOp<const array<IndexPair<typename internal::traits<OutputBackward>::Index>, 2>, const TensorReshapingOp<const DSizes<typename internal::traits<OutputBackward>::Index, 3>, const TensorImagePatchOp<Dynamic, Dynamic, const OutputBackward> >, const TensorReshapingOp<const DSizes<typename internal::traits<OutputBackward>::Index, 3>, const TensorReverseOp<const array<bool, 4>, const Kernel> > > > >::type -SpatialConvolutionBackwardInput(const Kernel& kernel, const OutputBackward& output_backward, typename internal::traits<OutputBackward>::Index inputRows, typename internal::traits<OutputBackward>::Index inputCols, const DenseIndex stride = 1, const DenseIndex in_stride = 1) { - - typedef typename internal::traits<OutputBackward>::Index TensorIndex; - TensorRef<Tensor<typename internal::traits<Kernel>::Scalar, internal::traits<Kernel>::NumDimensions, internal::traits<Kernel>::Layout, TensorIndex> > kern(kernel); - TensorRef<Tensor<typename internal::traits<OutputBackward>::Scalar, internal::traits<OutputBackward>::NumDimensions, internal::traits<OutputBackward>::Layout, TensorIndex> > out(output_backward); - - EIGEN_STATIC_ASSERT(internal::traits<Kernel>::Layout == internal::traits<OutputBackward>::Layout, YOU_MADE_A_PROGRAMMING_MISTAKE); - - static const bool isColMajor = (internal::traits<OutputBackward>::Layout == ColMajor); - - static const int NumDims = internal::traits<OutputBackward>::NumDimensions; - - // Number of filters to apply. This is the same as the output depth of the result - const TensorIndex kernelFilters = isColMajor ? kern.dimensions()[0] : kern.dimensions()[3]; - // Number of channels. This is the same as the input depth. - const TensorIndex kernelChannels = isColMajor ? kern.dimensions()[1] : kern.dimensions()[2]; - const TensorIndex kernelRows = isColMajor ? kern.dimensions()[2] : kern.dimensions()[1]; - const TensorIndex kernelCols = isColMajor ? kern.dimensions()[3] : kern.dimensions()[0]; - - // This is the effective kernel size, taking into account the (in_stride - 1) zero-values - // inserted between consecutive kernel elements in atrous convolution - const TensorIndex kernelRowsEff = kernelRows + (kernelRows - 1) * (in_stride - 1); - const TensorIndex kernelColsEff = kernelCols + (kernelCols - 1) * (in_stride - 1); - - const TensorIndex outputRows = isColMajor ? output_backward.dimension(1) : output_backward.dimension(NumDims - 2); - const TensorIndex outputCols = isColMajor ? output_backward.dimension(2) : output_backward.dimension(NumDims - 3); - - // Computing the forward padding - const TensorIndex forward_pad_top = ((outputRows - 1) * stride + kernelRowsEff - inputRows) / 2; - const TensorIndex forward_pad_left = ((outputCols - 1) * stride + kernelColsEff - inputCols) / 2; - - const TensorIndex padding_top = kernelRowsEff - 1 - forward_pad_top; - const TensorIndex padding_left = kernelColsEff - 1 - forward_pad_left; - const TensorIndex padding_bottom = inputRows + kernelRowsEff - 1 - (outputRows - 1) * stride - 1 - padding_top; - const TensorIndex padding_right = inputCols + kernelColsEff - 1 - (outputCols - 1) * stride - 1 - padding_left; - - eigen_assert(padding_top >= 0); - eigen_assert(padding_left >= 0); - eigen_assert(padding_bottom >= 0); - eigen_assert(padding_right >= 0); - - // The kernel has dimensions filters X channels X patch_rows X patch_cols - // We need to reverse the kernel along dimensions corresponding to rows and - // cols. - // TODO(yangke): we can make things slightly faster by collapsing the dimensions - // where we don't reverse. Try that once we have a faster compiler. - array<bool, 4> kernel_reverse; - if (isColMajor) { - kernel_reverse[0] = false; - kernel_reverse[1] = false; - kernel_reverse[2] = true; - kernel_reverse[3] = true; - } else { - kernel_reverse[0] = true; - kernel_reverse[1] = true; - kernel_reverse[2] = false; - kernel_reverse[3] = false; - } - - DSizes<TensorIndex, 3> kernel_dims; - if (isColMajor) { - kernel_dims[0] = kernelFilters; - kernel_dims[1] = kernelChannels; - kernel_dims[2] = kernelRows * kernelCols; - } else { - kernel_dims[0] = kernelRows * kernelCols; - kernel_dims[1] = kernelChannels; - kernel_dims[2] = kernelFilters; - } - - // The output_backward has dimensions out_depth X out_rows X out_cols X OTHERS - // When we extract the image patches from output_backward, it will have dimensions - // out_depth X (patch_rows * patch_cols) X (input_rows * input_cols * OTHERS) - DSizes<TensorIndex, 3> pre_contract_dims; - if (isColMajor) { - pre_contract_dims[0] = kernelFilters; - pre_contract_dims[1] = kernelRows * kernelCols; - pre_contract_dims[2] = inputRows * inputCols; - for (int i = 3; i < NumDims; ++i) { - pre_contract_dims[2] *= out.dimension(i); - } - } else { - pre_contract_dims[2] = kernelFilters; - pre_contract_dims[1] = kernelRows * kernelCols; - pre_contract_dims[0] = inputRows * inputCols; - for (int i = 0; i < NumDims - 3; ++i) { - pre_contract_dims[0] *= out.dimension(i); - } - } - - // We will contract along dimensions (0, 2) in kernel and (0, 1) in - // output_backward, if this is col-major, and - // dimensions (0, 2) in kernel and (1, 2) in output_backward, if this row-major. - array<IndexPair<TensorIndex>, 2> contract_dims; - if (isColMajor) { - // col-major: kernel.contract(output.patches) - contract_dims[0] = IndexPair<TensorIndex>(0, 0); - contract_dims[1] = IndexPair<TensorIndex>(2, 1); - } else { - // row-major: output.patches.contract(kernel) - contract_dims[0] = IndexPair<TensorIndex>(1, 0); - contract_dims[1] = IndexPair<TensorIndex>(2, 2); - } - - // Post contraction, the dimensions of the input_backprop is - // channels X input_rows X input_cols X OTHERS - DSizes<TensorIndex, NumDims> post_contract_dims; - if (isColMajor) { - post_contract_dims[0] = kernelChannels; - post_contract_dims[1] = inputRows; - post_contract_dims[2] = inputCols; - for (int i = 3; i < NumDims; ++i) { - post_contract_dims[i] = out.dimension(i); - } - } else { - post_contract_dims[NumDims - 1] = kernelChannels; - post_contract_dims[NumDims - 2] = inputRows; - post_contract_dims[NumDims - 3] = inputCols; - for (int i = 0; i < NumDims - 3; ++i) { - post_contract_dims[i] = out.dimension(i); - } - } - - return choose(Cond<internal::traits<OutputBackward>::Layout == ColMajor>(), - kernel.reverse(kernel_reverse).reshape(kernel_dims).contract(output_backward.extract_image_patches(kernelRows, kernelCols, 1, 1, in_stride, in_stride, stride, stride, padding_top, padding_bottom, padding_left, padding_right, 0).reshape(pre_contract_dims), contract_dims).reshape(post_contract_dims), - output_backward.extract_image_patches(kernelRows, kernelCols, 1, 1, in_stride, in_stride, stride, stride, padding_top, padding_bottom, padding_left, padding_right, 0).reshape(pre_contract_dims).contract(kernel.reverse(kernel_reverse).reshape(kernel_dims), contract_dims).reshape(post_contract_dims)); -} - - -/** SpatialConvolutionBackwardKernel - * \ingroup CXX11_NeuralNetworks_Module - * - * \brief Computes the backprop for the filter of a 2D convolution. - * - * The output_backward parameter is expected to be a tensor with a rank of 3 or more (channels, height, width, and optionally others) - * The kernel parameter is expected to be a 4D tensor (filters, channels, kernel_height, kernel_width) - * The output_backward and the kernel must both be in col-major layout. The result will also be in col-major layout. - * - * If in_stride > 1, then applies convolution with holes (aka atrous convolution), sampling every in_stride input pixels. - * - * The result can be assigned to a tensor of rank equal to the rank of the output_backward. The dimensions of the result will be filters, height, width (and others if applicable). - * - * It is possible to swap the order of the width and height dimensions provided that the same order is used in the input, the kernel, and the output. - * - */ -// TODO(gpapan): Resolve a bug in TensorContractionInputMapper at SpatialConvolutions.h that yangke circumvented by using .reshape().reshape(). -// This can significantly accelerate SpatialConvolutionBackwardKernel. - -template <typename OutputBackward, typename Input> -EIGEN_ALWAYS_INLINE -static const typename internal::conditional< - internal::traits<OutputBackward>::Layout == ColMajor, - const TensorShufflingOp<const array<typename internal::traits<OutputBackward>::Index, 4>, const TensorReverseOp<const array<bool, 4>, const TensorReshapingOp<const DSizes<typename internal::traits<OutputBackward>::Index, 4>, const TensorContractionOp<const array<IndexPair<typename internal::traits<Input>::Index>, 2>, const TensorReshapingOp<const DSizes<typename internal::traits<Input>::Index, 3>, const Input>, const TensorReshapingOp<const DSizes<typename internal::traits<OutputBackward>::Index, 4>, const TensorReshapingOp<const DSizes<typename internal::traits<OutputBackward>::Index, 4>, const TensorImagePatchOp<Dynamic, Dynamic, const OutputBackward> > > > > > >, - const TensorShufflingOp<const array<typename internal::traits<OutputBackward>::Index, 4>, const TensorReverseOp<const array<bool, 4>, const TensorReshapingOp<const DSizes<typename internal::traits<OutputBackward>::Index, 4>, const TensorContractionOp<const array<IndexPair<typename internal::traits<Input>::Index>, 2>, const TensorReshapingOp<const DSizes<typename internal::traits<OutputBackward>::Index, 4>, const TensorReshapingOp<const DSizes<typename internal::traits<OutputBackward>::Index, 4>, const TensorImagePatchOp<Dynamic, Dynamic, const OutputBackward> > >, const TensorReshapingOp<const DSizes<typename internal::traits<Input>::Index, 3>, const Input> > > > > >::type -SpatialConvolutionBackwardKernel(const Input& input, const OutputBackward& output_backward, typename internal::traits<Input>::Index kernelRows, typename internal::traits<Input>::Index kernelCols, const DenseIndex stride = 1, const DenseIndex in_stride = 1) { - - typedef typename internal::traits<Input>::Index TensorIndex; - TensorRef<Tensor<typename internal::traits<Input>::Scalar, internal::traits<Input>::NumDimensions, internal::traits<Input>::Layout, TensorIndex> > in(input); - TensorRef<Tensor<typename internal::traits<OutputBackward>::Scalar, internal::traits<OutputBackward>::NumDimensions, internal::traits<OutputBackward>::Layout, TensorIndex> > out(output_backward); - - EIGEN_STATIC_ASSERT(internal::traits<Input>::Layout == internal::traits<OutputBackward>::Layout, YOU_MADE_A_PROGRAMMING_MISTAKE); - - // stride and in_stride cannot both be larger than 1 - eigen_assert(!(stride > 1 && in_stride > 1)); - - static const bool isColMajor = (internal::traits<Input>::Layout == ColMajor); - - static const int NumDims = internal::traits<Input>::NumDimensions; - EIGEN_STATIC_ASSERT(internal::traits<Input>::NumDimensions == internal::traits<OutputBackward>::NumDimensions, YOU_MADE_A_PROGRAMMING_MISTAKE); - - const TensorIndex inputRows = isColMajor ? in.dimension(1) : in.dimension(NumDims - 2); - const TensorIndex inputCols = isColMajor ? in.dimension(2) : in.dimension(NumDims - 3); - - const TensorIndex outputRows = isColMajor ? output_backward.dimension(1) : output_backward.dimension(NumDims - 2); - const TensorIndex outputCols = isColMajor ? output_backward.dimension(2) : output_backward.dimension(NumDims - 3); - - // Number of filters to apply. This is the same as the output depth of the result - const TensorIndex kernelFilters = isColMajor ? out.dimensions()[0] : out.dimensions()[NumDims - 1]; - - // Number of channels. This is the same as the input depth. - const TensorIndex kernelChannels = isColMajor ? in.dimensions()[0] : in.dimensions()[NumDims - 1]; - - // This is the effective kernel size, taking into account the (in_stride - 1) zero-values - // inserted between consecutive kernel elements in atrous convolution - const TensorIndex kernelRowsEff = kernelRows + (kernelRows - 1) * (in_stride - 1); - const TensorIndex kernelColsEff = kernelCols + (kernelCols - 1) * (in_stride - 1); - - // Computing the forward padding - const TensorIndex forward_pad_top = ((outputRows - 1) * stride + kernelRowsEff - inputRows) / 2; - const TensorIndex forward_pad_left = ((outputCols - 1) * stride + kernelColsEff - inputCols) / 2; - - // TODO: factor out the padding computation. - const TensorIndex padding_top = kernelRowsEff - 1 - forward_pad_top; - const TensorIndex padding_left = kernelColsEff - 1 - forward_pad_left; - const TensorIndex padding_bottom = inputRows + kernelRowsEff - 1 - (outputRows - 1) * stride - 1 - padding_top; - const TensorIndex padding_right = inputCols + kernelColsEff - 1 - (outputCols - 1) * stride - 1 - padding_left; - - eigen_assert(padding_top >= 0); - eigen_assert(padding_left >= 0); - eigen_assert(padding_bottom >= 0); - eigen_assert(padding_right >= 0); - - // The output_backward has dimensions out_depth X out_rows X out_cols X OTHERS - // When we extract the image patches from output_backward (with input as the - // kernel), it will have dimensions - // (out_depth) X (input_rows * input_cols) X (kernel_rows * kernel_cols) X OTHERS - DSizes<TensorIndex, 4> pre_contract_dims; - if (isColMajor) { - pre_contract_dims[0] = kernelFilters; - pre_contract_dims[1] = inputRows * inputCols; - pre_contract_dims[2] = kernelRows * kernelCols; - pre_contract_dims[3] = 1; - for (int i = 3; i < NumDims; ++i) { - pre_contract_dims[3] *= out.dimension(i); - } - } else { - pre_contract_dims[3] = kernelFilters; - pre_contract_dims[2] = inputRows * inputCols; - pre_contract_dims[1] = kernelRows * kernelCols; - pre_contract_dims[0] = 1; - for (int i = 0; i < NumDims - 3; ++i) { - pre_contract_dims[0] *= out.dimension(i); - } - } - - // The input has dimensions in_depth X (input_rows * input_cols) X OTHERS - DSizes<TensorIndex, 3> input_dims; - if (isColMajor) { - input_dims[0] = kernelChannels; - input_dims[1] = inputRows * inputCols; - input_dims[2] = 1; - for (int i = 3; i < NumDims; ++i) { - input_dims[2] *= in.dimension(i); - } - eigen_assert(input_dims[2] == pre_contract_dims[3]); - } else { - input_dims[2] = kernelChannels; - input_dims[1] = inputRows * inputCols; - input_dims[0] = 1; - for (int i = 0; i < NumDims - 3; ++i) { - input_dims[0] *= in.dimension(i); - } - eigen_assert(input_dims[0] == pre_contract_dims[0]); - } - - // We will contract along dimensions (1, 2) in in and (1, 3) in out, if - // this is col-major. - // For row-major, it's dimensions (0, 1) in in and (0, 2) in out. - array<IndexPair<TensorIndex>, 2> contract_dims; - if (isColMajor) { - // col-major: in.contract(output.patches) - contract_dims[0] = IndexPair<TensorIndex>(1, 1); - contract_dims[1] = IndexPair<TensorIndex>(2, 3); - } else { - // row-major: output.patches.contract(in) - contract_dims[0] = IndexPair<TensorIndex>(0, 0); - contract_dims[1] = IndexPair<TensorIndex>(2, 1); - } - - // After the contraction, the kernel will have dimension - // in_depth X out_depth X kernel_rows X kernel_cols - // We will need to shuffle the first two dimensions and reverse the latter - // two dimensions. - // The end shape is - // out_depth X in_shape X kernel_rows X kernel_cols - - // This is the shape of the kernel *before* the shuffling. - DSizes<TensorIndex, 4> kernel_dims; - if (isColMajor) { - kernel_dims[0] = kernelChannels; - kernel_dims[1] = kernelFilters; - kernel_dims[2] = kernelRows; - kernel_dims[3] = kernelCols; - } else { - kernel_dims[0] = kernelCols; - kernel_dims[1] = kernelRows; - kernel_dims[2] = kernelFilters; - kernel_dims[3] = kernelChannels; - } - - array<TensorIndex, 4> kernel_shuffle; - if (isColMajor) { - kernel_shuffle[0] = 1; - kernel_shuffle[1] = 0; - kernel_shuffle[2] = 2; - kernel_shuffle[3] = 3; - } else { - kernel_shuffle[0] = 0; - kernel_shuffle[1] = 1; - kernel_shuffle[2] = 3; - kernel_shuffle[3] = 2; - } - - array<bool, 4> kernel_reverse; - if (isColMajor) { - kernel_reverse[0] = false; - kernel_reverse[1] = false; - kernel_reverse[2] = true; - kernel_reverse[3] = true; - } else { - kernel_reverse[0] = true; - kernel_reverse[1] = true; - kernel_reverse[2] = false; - kernel_reverse[3] = false; - } - - return choose(Cond<internal::traits<Input>::Layout == ColMajor>(), - input.reshape(input_dims).contract(output_backward.extract_image_patches(inputRows, inputCols, in_stride, in_stride, 1, 1, stride, stride, padding_top, padding_bottom, padding_left, padding_right, 0).reshape(pre_contract_dims).reshape(pre_contract_dims), contract_dims).reshape(kernel_dims).reverse(kernel_reverse).shuffle(kernel_shuffle), - output_backward.extract_image_patches(inputRows, inputCols, in_stride, in_stride, 1, 1, stride, stride, padding_top, padding_bottom, padding_left, padding_right, 0).reshape(pre_contract_dims).reshape(pre_contract_dims).contract(input.reshape(input_dims), contract_dims).reshape(kernel_dims).reverse(kernel_reverse).shuffle(kernel_shuffle)); -} - -} // end namespace Eigen - -#endif // EIGEN_CXX11_NEURAL_NETWORKS_BACKWARD_SPATIAL_CONVOLUTIONS_H diff --git a/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/CuboidConvolution.h b/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/CuboidConvolution.h deleted file mode 100644 index dfb9dcedba..0000000000 --- a/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/CuboidConvolution.h +++ /dev/null @@ -1,179 +0,0 @@ -#ifndef EIGEN_CXX11_SRC_NEURAL_NETWORKS_CUBOID_CONVOLUTION_H -#define EIGEN_CXX11_SRC_NEURAL_NETWORKS_CUBOID_CONVOLUTION_H - -#include "Patch3d.h" - -namespace Eigen { - -/** CuboidConvolution - * \ingroup CXX11_NeuralNetworks_Module - * - * \brief Applies a 3D convolution over a multichannel input voxel block. - * - * The input parameter is expected to be a tensor with a rank of 4 or more (channels, depth, height, width, and optionally others). - * The kernel parameter is expected to be a 5D tensor (filters, channels, kernel_depth, kernel_height, kernel_width). - * The result can be assigned to a tensor of rank equal to the rank of the input. The dimensions of the result will be filters, depth, height, width (and others if applicable). - * - * The input and kernel have to be in the same layout, and both row-major and - * col-major are supported. The shapes given above are for col-major layout. - * For row-major, all dimensions should be reversed. - * - * It is possible to swap the order of the depth, width, and height dimensions provided that the same order is used in the input, the kernel, and the output. - */ -template <typename Input, typename Kernel> -EIGEN_ALWAYS_INLINE -static const typename internal::conditional < - internal::traits<Input>::Layout == ColMajor, - TensorReshapingOp< - const DSizes<typename internal::traits<Input>::Index, - internal::traits<Input>::NumDimensions>, - const TensorContractionOp< - const array<IndexPair<typename internal::traits<Input>::Index>, 1>, - const TensorReshapingOp< - const DSizes<typename internal::traits<Input>::Index, 2>, - const Kernel>, - const TensorReshapingOp< - const DSizes<typename internal::traits<Input>::Index, 2>, - const TensorVolumePatchOp<Dynamic, Dynamic, Dynamic, - const Input> > > >, - TensorReshapingOp< - const DSizes<typename internal::traits<Input>::Index, - internal::traits<Input>::NumDimensions>, - const TensorContractionOp< - const array<IndexPair<typename internal::traits<Input>::Index>, 1>, - const TensorReshapingOp< - const DSizes<typename internal::traits<Input>::Index, 2>, - const TensorVolumePatchOp<Dynamic, Dynamic, Dynamic, - const Input> > , - const TensorReshapingOp< - const DSizes<typename internal::traits<Input>::Index, 2>, - const Kernel> > > >::type -CuboidConvolution(const Input& input, const Kernel& kernel, - const DenseIndex stridePlanes = 1, - const DenseIndex strideRows = 1, - const DenseIndex strideCols = 1, - const PaddingType padding_type = PADDING_SAME) { - typedef typename internal::traits<Input>::Index TensorIndex; - TensorRef<Tensor<typename internal::traits<Input>::Scalar, internal::traits<Input>::NumDimensions, internal::traits<Input>::Layout, TensorIndex> > in(input); - TensorRef<Tensor<typename internal::traits<Kernel>::Scalar, internal::traits<Kernel>::NumDimensions, internal::traits<Kernel>::Layout, TensorIndex> > kern(kernel); - - EIGEN_STATIC_ASSERT(internal::traits<Input>::Layout == internal::traits<Kernel>::Layout, YOU_MADE_A_PROGRAMMING_MISTAKE); - static const bool isColMajor = (internal::traits<Input>::Layout == ColMajor); - static const int NumDims = internal::traits<Input>::NumDimensions; - - // Number of filters to apply. This is the same as the output depth of the result. - const TensorIndex kernelFilters = isColMajor ? kern.dimensions()[0] : kern.dimensions()[4]; - const TensorIndex kernelChannels = isColMajor ? kern.dimensions()[1] : kern.dimensions()[3]; - - // Spatial size of the kernel. - const TensorIndex kernelDepth = isColMajor ? kern.dimensions()[2] : kern.dimensions()[2]; - const TensorIndex kernelRows = isColMajor ? kern.dimensions()[3] : kern.dimensions()[1]; - const TensorIndex kernelCols = isColMajor ? kern.dimensions()[4] : kern.dimensions()[0]; - - if (isColMajor) { - eigen_assert(kernelChannels == in.dimension(0)); - } else { - eigen_assert(kernelChannels == in.dimension(NumDims - 1)); - } - - const TensorIndex inputPlanes = isColMajor ? in.dimension(1) : in.dimension(NumDims - 2); - const TensorIndex inputRows = isColMajor ? in.dimension(2) : in.dimension(NumDims - 3); - const TensorIndex inputCols = isColMajor ? in.dimension(3) : in.dimension(NumDims - 4); - - const float stride_planes_f = static_cast<float>(stridePlanes); - const float stride_rows_f = static_cast<float>(strideRows); - const float stride_cols_f = static_cast<float>(strideCols); - TensorIndex out_depth; - TensorIndex out_height; - TensorIndex out_width; - switch (padding_type) { - case PADDING_VALID: - out_depth = ceil((inputPlanes - kernelDepth + 1.f) / stride_planes_f); - out_height = ceil((inputRows - kernelRows + 1.f) / stride_rows_f); - out_width = ceil((inputCols - kernelCols + 1.f) / stride_cols_f); - break; - case PADDING_SAME: - out_depth = ceil(inputPlanes / stride_planes_f); - out_height = ceil(inputRows / stride_rows_f); - out_width = ceil(inputCols / stride_cols_f); - break; - default: - eigen_assert(false && "unexpected padding"); - } - - DSizes<TensorIndex, 2> kernel_dims; - if (isColMajor) { - kernel_dims[0] = kernelFilters; - kernel_dims[1] = kernelChannels * kernelDepth * kernelRows * kernelCols; - } else { - kernel_dims[0] = kernelChannels * kernelDepth * kernelRows * kernelCols; - kernel_dims[1] = kernelFilters; - } - - // Molds the output of the patch extraction result into a 2D tensor: - // - the first dimension (dims[0]): the patch values to be multiplied with the kernels - // - the second dimension (dims[1]): everything else - DSizes<TensorIndex, 2> pre_contract_dims; - if (isColMajor) { - pre_contract_dims[0] = kernelChannels * kernelDepth * kernelRows * kernelCols; - pre_contract_dims[1] = out_depth * out_height * out_width; - for (int i = 4; i < NumDims; ++i) { - pre_contract_dims[1] *= in.dimension(i); - } - } else { - pre_contract_dims[1] = kernelChannels * kernelDepth * kernelRows * kernelCols; - pre_contract_dims[0] = out_depth * out_height * out_width; - for (int i = 0; i < NumDims - 4; ++i) { - pre_contract_dims[0] *= in.dimension(i); - } - } - - array<IndexPair<TensorIndex>, 1> contract_dims; - contract_dims[0] = IndexPair<TensorIndex>(1, 0); - - // Molds the output of the contraction into the shape expected by the user - // (assuming ColMajor): - // - 1st dim: kernel filters - // - 2nd dim: output depth - // - 3nd dim: output height - // - 4rd dim: output width - // - 5th dim and beyond: everything else including batch size - DSizes<TensorIndex, NumDims> post_contract_dims; - if (isColMajor) { - post_contract_dims[0] = kernelFilters; - post_contract_dims[1] = out_depth; - post_contract_dims[2] = out_height; - post_contract_dims[3] = out_width; - for (int i = 4; i < NumDims; ++i) { - post_contract_dims[i] = in.dimension(i); - } - } else { - post_contract_dims[NumDims - 1] = kernelFilters; - post_contract_dims[NumDims - 2] = out_depth; - post_contract_dims[NumDims - 3] = out_height; - post_contract_dims[NumDims - 4] = out_width; - for (int i = 0; i < NumDims - 4; ++i) { - post_contract_dims[i] = in.dimension(i); - } - } - - return choose( - Cond<internal::traits<Input>::Layout == ColMajor>(), - kernel.reshape(kernel_dims) - .contract(input.extract_volume_patches( - kernelDepth, kernelRows, kernelCols, stridePlanes, - strideRows, strideCols, padding_type) - .reshape(pre_contract_dims), - contract_dims) - .reshape(post_contract_dims), - input.extract_volume_patches(kernelDepth, kernelRows, kernelCols, - stridePlanes, strideRows, strideCols, - padding_type) - .reshape(pre_contract_dims) - .contract(kernel.reshape(kernel_dims), contract_dims) - .reshape(post_contract_dims)); -} - -} // end namespace Eigen - -#endif // EIGEN_CXX11_SRC_NEURAL_NETWORKS_CUBOID_CONVOLUTION_H diff --git a/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/Patch3d.h b/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/Patch3d.h deleted file mode 100644 index 2864f83299..0000000000 --- a/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/Patch3d.h +++ /dev/null @@ -1,240 +0,0 @@ -#ifndef EIGEN_CXX11_SRC_NEURAL_NETWORKS_PATCH3D_H -#define EIGEN_CXX11_SRC_NEURAL_NETWORKS_PATCH3D_H - -#if not defined(__CUDACC__) -#include <type_traits> -#endif - -namespace Eigen { -namespace internal { - -/** Extract3DPatches - * \ingroup CXX11_NeuralNetworksModule - * - * \brief Extracts 3D patches from a multichannel input volume. - * - * The input parameter is expected to be a tensor with a rank of 4 or more - * (channels, depth, height, width, optional others in col-major, and the - * reverse order in row-major). - - * The return value will be a tensor of 3 more dimension than the input tensor. - * In col-major, the first 4 dimensions of the result are: channels, patch_depth, - * patch_height, patch_width. The next dimensions will identify the patch - * position on the 3D grid of extracted patches: z, y, x. The remaining - * dimensions, if any, will be the same as the 'other' dimensions of the input - * tensor. - */ - -template <typename Input> -EIGEN_ALWAYS_INLINE static const TensorStridingOp< - const array<typename internal::traits<Input>::Index, - internal::traits<Input>::NumDimensions + 3>, - const TensorReshapingOp< - const DSizes<typename internal::traits<Input>::Index, - internal::traits<Input>::NumDimensions + 3>, - const TensorPatchOp< - const DSizes<typename internal::traits<Input>::Index, - internal::traits<Input>::NumDimensions>, - const TensorPaddingOp< - const array<IndexPair<typename internal::traits<Input>::Index>, - internal::traits<Input>::NumDimensions>, - const Input> > > > -Extract3DPatches( - const Input& input, const DenseIndex patchPlanes, - const DenseIndex patchRows, const DenseIndex patchCols, - const DenseIndex stridePlanes, const DenseIndex strideRows, - const DenseIndex strideCols, - const DenseIndex paddingZTop, const DenseIndex paddingZBottom, - const DenseIndex paddingTop, const DenseIndex paddingBottom, - const DenseIndex paddingLeft, const DenseIndex paddingRight, - const typename internal::traits<Input>::Scalar padding_value = 0) { - - typedef typename internal::traits<Input>::Index TensorIndex; - TensorRef<Tensor<typename internal::traits<Input>::Scalar, internal::traits<Input>::NumDimensions, internal::traits<Input>::Layout, TensorIndex> > in(input); - - EIGEN_STATIC_ASSERT(internal::traits<Input>::NumDimensions >= 4, YOU_MADE_A_PROGRAMMING_MISTAKE); - - static const bool isColMajor = (internal::traits<Input>::Layout == ColMajor); - static const int NumDims = internal::traits<Input>::NumDimensions; - static const int ExtDims = NumDims + 3; - - // Tensor size after patch extraction. We add three dimensions to unpack the - // linear patch index into a 3D grid over which stride() can work. - DSizes<TensorIndex, ExtDims> pre_stride_dims; - - if (isColMajor) { - pre_stride_dims[0] = in.dimension(0); - pre_stride_dims[1] = patchPlanes; - pre_stride_dims[2] = patchRows; - pre_stride_dims[3] = patchCols; - } else { - pre_stride_dims[ExtDims - 1] = in.dimension(NumDims - 1); - pre_stride_dims[ExtDims - 4] = patchCols; - pre_stride_dims[ExtDims - 3] = patchRows; - pre_stride_dims[ExtDims - 2] = patchPlanes; - } - - const TensorIndex inputPlanes = isColMajor ? in.dimension(1) : in.dimension(NumDims - 2); - const TensorIndex inputRows = isColMajor ? in.dimension(2) : in.dimension(NumDims - 3); - const TensorIndex inputCols = isColMajor ? in.dimension(3) : in.dimension(NumDims - 4); - - array<IndexPair<TensorIndex>, NumDims> paddings; - for (int i = 0; i < NumDims; ++i) { - paddings[i] = IndexPair<TensorIndex>(0, 0); - } - - paddings[isColMajor ? 1 : (NumDims - 2)] = IndexPair<TensorIndex>(paddingZTop, paddingZBottom); - paddings[isColMajor ? 2 : (NumDims - 3)] = IndexPair<TensorIndex>(paddingTop, paddingBottom); - paddings[isColMajor ? 3 : (NumDims - 4)] = IndexPair<TensorIndex>(paddingLeft, paddingRight); - - pre_stride_dims[isColMajor ? 4 : (ExtDims - 5)] = inputPlanes + paddingZBottom + paddingZTop - patchPlanes + 1; - pre_stride_dims[isColMajor ? 5 : (ExtDims - 6)] = inputRows + paddingTop + paddingBottom - patchRows + 1; - pre_stride_dims[isColMajor ? 6 : (ExtDims - 7)] = inputCols + paddingLeft + paddingRight - patchCols + 1; - - if (isColMajor) { - for (int i = 7; i < NumDims + 3; ++i) { - pre_stride_dims[i] = in.dimension(i - 3); - } - } else { - for (int i = 0; i < NumDims - 4; ++i) { - pre_stride_dims[i] = in.dimension(i); - } - } - - DSizes<TensorIndex, NumDims> patch_dims; - if (isColMajor) { - patch_dims[0] = in.dimension(0); - patch_dims[1] = patchPlanes; - patch_dims[2] = patchRows; - patch_dims[3] = patchCols; - for (int i = 4; i < NumDims; ++i) { - patch_dims[i] = 1; - } - } else { - patch_dims[NumDims - 1] = in.dimension(NumDims - 1); - patch_dims[NumDims - 4] = patchCols; - patch_dims[NumDims - 3] = patchRows; - patch_dims[NumDims - 2] = patchPlanes; - for (int i = 0; i < NumDims - 4; i++) { - patch_dims[i] = 1; - } - } - - array<TensorIndex, NumDims + 3> strides; - if (isColMajor) { - // No striding within the patches. - for (int i = 0; i < 4; ++i) { - strides[i] = 1; - } - // Apply striding in the spatial patch grid dimensions only. - strides[4] = stridePlanes; - strides[5] = strideRows; - strides[6] = strideCols; - // No striding in the remaining dimensions (batches, ...). - for (int i = 7; i < NumDims + 3; i++) { - strides[i] = 1; - } - } else { - // No striding within the patches. - for (int i = 1; i <= 4; ++i) { - strides[ExtDims - i] = 1; - } - // Apply striding in the spatial patch grid dimensions only. - strides[ExtDims - 7] = strideCols; - strides[ExtDims - 6] = strideRows; - strides[ExtDims - 5] = stridePlanes; - // No striding in the remaining dimensions (batches, ...). - for (int i = 0; i < NumDims - 4; i++) { - strides[i] = 1; - } - } - - // TODO(mjanusz): Consider getting rid of pad(), and stride() and extend - // extract_patches to take additional parameters for padding/striding, - // similarly to extract_image_patches. - return input.pad(paddings, padding_value).extract_patches(patch_dims).reshape(pre_stride_dims).stride(strides); -} - - -template <typename Input> -EIGEN_ALWAYS_INLINE static const TensorStridingOp< - const array<typename internal::traits<Input>::Index, - internal::traits<Input>::NumDimensions + 3>, - const TensorReshapingOp< - const DSizes<typename internal::traits<Input>::Index, - internal::traits<Input>::NumDimensions + 3>, - const TensorPatchOp< - const DSizes<typename internal::traits<Input>::Index, - internal::traits<Input>::NumDimensions>, - const TensorPaddingOp< - const array<IndexPair<typename internal::traits<Input>::Index>, - internal::traits<Input>::NumDimensions>, - const Input> > > > -Extract3DPatches( - const Input& input, const DenseIndex patchPlanes, - const DenseIndex patchRows, const DenseIndex patchCols, - const DenseIndex stridePlanes, const DenseIndex strideRows, - const DenseIndex strideCols, const PaddingType padding_type, - const typename internal::traits<Input>::Scalar padding_value = 0) { - typedef typename internal::traits<Input>::Index TensorIndex; - TensorRef<Tensor<typename internal::traits<Input>::Scalar, internal::traits<Input>::NumDimensions, internal::traits<Input>::Layout, TensorIndex> > in(input); - - EIGEN_STATIC_ASSERT(internal::traits<Input>::NumDimensions >= 4, YOU_MADE_A_PROGRAMMING_MISTAKE); - - static const bool isColMajor = (internal::traits<Input>::Layout == ColMajor); - static const int NumDims = internal::traits<Input>::NumDimensions; - - const TensorIndex inputPlanes = isColMajor ? in.dimension(1) : in.dimension(NumDims - 2); - const TensorIndex inputRows = isColMajor ? in.dimension(2) : in.dimension(NumDims - 3); - const TensorIndex inputCols = isColMajor ? in.dimension(3) : in.dimension(NumDims - 4); - - switch (padding_type) { - case PADDING_VALID: - // No padding in any dimension. - return Extract3DPatches(input, patchPlanes, patchRows, patchCols, - stridePlanes, strideRows, strideCols, - 0, 0, 0, 0, 0, 0, padding_value); - case PADDING_SAME: { - // The side of the tensor before striding should be just the expected - // output times the stride. - const TensorIndex size_z = ceil(inputPlanes / static_cast<float>(stridePlanes)) * stridePlanes; - const TensorIndex size_y = ceil(inputRows / static_cast<float>(strideRows)) * strideRows; - const TensorIndex size_x = ceil(inputCols / static_cast<float>(strideCols)) * strideCols; - - // The size of the patch space is going to be: padded_input_size - patch_size + 1. - // This has to match the expected size before striding (pre_stride_dims). - // The deltas below extend the input to the expected size. - const TensorIndex dz = size_z + patchPlanes - 1 - inputPlanes; - const TensorIndex dy = size_y + patchRows - 1 - inputRows; - const TensorIndex dx = size_x + patchCols - 1 - inputCols; - - return Extract3DPatches(input, patchPlanes, patchRows, patchCols, - stridePlanes, strideRows, strideCols, - dz - dz / 2, dz / 2, - dy - dy / 2, dy / 2, - dx - dx / 2, dx / 2, - padding_value); - } - default: - eigen_assert(false && "unexpected padding"); - // unreachable code to avoid missing return warning. - return Extract3DPatches(input, patchPlanes, patchRows, patchCols, - stridePlanes, strideRows, strideCols, - 0, 0, 0, 0, 0, 0, padding_value); - } -} - -// TODO(mjanusz): Switch this to a 'using' alias once CUDA supports C++11. -template <typename Input> -struct Extract3DPatchesType { - typedef const TensorStridingOp< const array<typename internal::traits<Input>::Index, internal::traits<Input>::NumDimensions + 3>, - const TensorReshapingOp< const DSizes<typename internal::traits<Input>::Index, internal::traits<Input>::NumDimensions + 3>, - const TensorPatchOp< const DSizes<typename internal::traits<Input>::Index, internal::traits<Input>::NumDimensions>, - const TensorPaddingOp< const array< IndexPair<typename internal::traits<Input>::Index>, internal::traits<Input>::NumDimensions>, - const Input> > > > type; -}; - -} // end namespace internal -} // end namespace Eigen - -#endif // EIGEN_CXX11_SRC_NEURAL_NETWORKS_PATCH3D_H diff --git a/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/Pooling.h b/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/Pooling.h deleted file mode 100644 index 942b060ba7..0000000000 --- a/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/Pooling.h +++ /dev/null @@ -1,433 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. -// -// Copyright (C) 2014 Benoit Steiner <benoit.steiner.goog@gmail.com> -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. -#ifndef EIGEN_CXX11_NEURAL_NETWORKS_POOLING_H -#define EIGEN_CXX11_NEURAL_NETWORKS_POOLING_H - -#include "Patch3d.h" - -namespace Eigen { - -/** SpatialMaxPooling - * \ingroup CXX11_NeuralNetworks_Module - * - * \brief Applies a max-pooling over a multichannel input image. - * - * The input parameter is expected to be a with a rank of 4 (channels, height, width, others in col-major, and the reverse of that in row-major). - * - * The result can be assigned to a tensor of rank equal to the rank of the input. The dimensions of the result will be channels, height, width, and others (in col-major, and the reverse of that if the input was row-major). - * - * The order of the width and height dimensions can be swapped if needed. - * -*/ -#if !defined(EIGEN_HAS_INDEX_LIST) -template <typename Input> -EIGEN_ALWAYS_INLINE -static const TensorReshapingOp<const Eigen::DSizes<typename internal::traits<Input>::Index, internal::traits<Input>::NumDimensions>, const TensorReductionOp<internal::MaxReducer<typename internal::remove_const<typename internal::traits<Input>::Scalar>::type>, const Eigen::array<int, 2>, const TensorImagePatchOp<Dynamic, Dynamic, const Input> > > -#else -template <typename Input> -EIGEN_ALWAYS_INLINE -static const TensorReshapingOp<const Eigen::DSizes<typename internal::traits<Input>::Index, internal::traits<Input>::NumDimensions>, const TensorReductionOp<internal::MaxReducer<typename internal::remove_const<typename internal::traits<Input>::Scalar>::type>, typename internal::conditional<internal::traits<Input>::Layout == ColMajor, const Eigen::IndexList<Eigen::type2index<1>, Eigen::type2index<2> >, const Eigen::IndexList<Eigen::type2index<2>, Eigen::type2index<3> > >::type, const TensorImagePatchOp<Dynamic, Dynamic, const Input> > > -#endif -SpatialMaxPooling(const Input& input, DenseIndex patchRows, DenseIndex patchCols, - DenseIndex strideRows, DenseIndex strideCols, const PaddingType padding_type, - DenseIndex in_strideRows = 1, DenseIndex in_strideCols = 1) -{ - EIGEN_STATIC_ASSERT(internal::traits<Input>::NumDimensions == 4, YOU_MADE_A_PROGRAMMING_MISTAKE); - - typedef typename internal::traits<Input>::Index TensorIndex; - TensorRef<Tensor<typename internal::traits<Input>::Scalar, internal::traits<Input>::NumDimensions, internal::traits<Input>::Layout, TensorIndex> > in(input); - - const DenseIndex patchRowsEff = patchRows + (patchRows - 1) * (in_strideRows - 1); - const DenseIndex patchColsEff = patchCols + (patchCols - 1) * (in_strideCols - 1); - - static const bool isColMajor = (internal::traits<Input>::Layout == ColMajor); - static const int idxRows = isColMajor ? 1 : 2; - static const int idxCols = isColMajor ? 2 : 1; - - // Molds the output of the reduction into the shape expected by the user. - // (assuming col-major): - // - 1st dim: channels - // - 2nd dim: output height - // - 3rd dim: output width - // - 4th dim and beyond: everything else including batch size - Eigen::DSizes<TensorIndex, internal::traits<Input>::NumDimensions> post_reduce_dims; - post_reduce_dims[0] = in.dimension(0); - if (padding_type == PADDING_VALID) { - post_reduce_dims[idxRows] = numext::ceil((in.dimension(idxRows) - patchRowsEff + 1.f) / static_cast<float>(strideRows)); - post_reduce_dims[idxCols] = numext::ceil((in.dimension(idxCols) - patchColsEff + 1.f) / static_cast<float>(strideCols)); - } else { - post_reduce_dims[idxRows] = numext::ceil(in.dimension(idxRows) / static_cast<float>(strideRows)); - post_reduce_dims[idxCols] = numext::ceil(in.dimension(idxCols) / static_cast<float>(strideCols)); - } - post_reduce_dims[3] = in.dimension(3); - -#if !defined(EIGEN_HAS_INDEX_LIST) - // nvcc doesn't support cxx11 - Eigen::array<int, 2> reduction_dims; - if (isColMajor) { - reduction_dims[0] = 1; - reduction_dims[1] = 2; - } else { - reduction_dims[0] = 2; - reduction_dims[1] = 3; - } -#else - // Take advantage of cxx11 to give the compiler information it can use to - // optimize the code. - typename internal::conditional<internal::traits<Input>::Layout == ColMajor, const Eigen::IndexList<Eigen::type2index<1>, Eigen::type2index<2> >, const Eigen::IndexList<Eigen::type2index<2>, Eigen::type2index<3> > >::type reduction_dims; -#endif - - return input.extract_image_patches(patchRows, patchCols, strideRows, strideCols, in_strideRows, in_strideCols, padding_type, -Eigen::NumTraits<typename internal::remove_const<typename internal::traits<Input>::Scalar>::type>::highest()).maximum(reduction_dims).reshape(post_reduce_dims); -} - -/** CuboidMaxPooling - * \ingroup CXX11_NeuralNetworks_Module - * - * \brief Applies a max-pooling over a multichannel input volume. - * - * The input parameter is expected to be a tensor with a rank of 5 (channels, depth, height, width, others in col-major, and the reverse of that in row-major). - * - * The result can be assigned to a tensor of rank equal to the rank of the input. The dimensions of the result will be channels, depth, height, width, and others (in col-major, and the reverse of that if the input was row-major). - * - * The order of the depth, width and height dimensions can be swapped if needed. - * -*/ -#if !defined(EIGEN_HAS_INDEX_LIST) -template <typename Input> -EIGEN_ALWAYS_INLINE static const TensorReshapingOp< - const Eigen::DSizes<DenseIndex, internal::traits<Input>::NumDimensions>, - const TensorReductionOp< - internal::MaxReducer<float>, const Eigen::array<int, 1>, - const TensorReshapingOp< - const Eigen::DSizes<DenseIndex, 3>, - const TensorVolumePatchOp<Dynamic, Dynamic, Dynamic, const Input> > > > -#else -template <typename Input> -EIGEN_ALWAYS_INLINE static const TensorReshapingOp< - const Eigen::DSizes<DenseIndex, internal::traits<Input>::NumDimensions>, - const TensorReductionOp< - internal::MaxReducer<float>, - const Eigen::IndexList<Eigen::type2index<1> >, - const TensorReshapingOp< - const Eigen::DSizes<DenseIndex, 3>, - const TensorVolumePatchOp<Dynamic, Dynamic, Dynamic, const Input> > > > -#endif -CuboidMaxPooling(const Input& input, DenseIndex patchPlanes, - DenseIndex patchRows, DenseIndex patchCols, - DenseIndex stridePlanes, DenseIndex strideRows, - DenseIndex strideCols, const PaddingType padding_type) { - EIGEN_STATIC_ASSERT(internal::traits<Input>::NumDimensions == 5, YOU_MADE_A_PROGRAMMING_MISTAKE); - static const bool isColMajor = (internal::traits<Input>::Layout == ColMajor); - - typedef typename internal::traits<Input>::Index TensorIndex; - TensorRef<Tensor<typename internal::traits<Input>::Scalar, internal::traits<Input>::NumDimensions, internal::traits<Input>::Layout, TensorIndex> > in(input); - - static const int idxPlanes = isColMajor ? 1 : 3; - static const int idxRows = 2; - static const int idxCols = isColMajor ? 3 : 1; - - // Molds the output of the reduction into the shape expected by the used - // (assuming col-major): - // - 1st dim: channels - // - 2nd dim: output depth - // - 3rd dim: output height - // - 4th dim: output width - // - 5th dim and beyond: everything else including batch size - Eigen::DSizes<DenseIndex, internal::traits<Input>::NumDimensions> post_reduce_dims; - post_reduce_dims[0] = in.dimension(0); - if (padding_type == PADDING_VALID) { - post_reduce_dims[idxPlanes] = numext::ceil((in.dimension(idxPlanes) - patchPlanes + 1.f) / static_cast<float>(stridePlanes)); - post_reduce_dims[idxRows] = numext::ceil((in.dimension(idxRows) - patchRows + 1.f) / static_cast<float>(strideRows)); - post_reduce_dims[idxCols] = numext::ceil((in.dimension(idxCols) - patchCols + 1.f) / static_cast<float>(strideCols)); - } else { - post_reduce_dims[idxPlanes] = numext::ceil(in.dimension(idxPlanes) / static_cast<float>(stridePlanes)); - post_reduce_dims[idxRows] = numext::ceil(in.dimension(idxRows) / static_cast<float>(strideRows)); - post_reduce_dims[idxCols] = numext::ceil(in.dimension(idxCols) / static_cast<float>(strideCols)); - } - post_reduce_dims[4] = in.dimension(4); - - Eigen::DSizes<DenseIndex, 3> pre_reduce_dims; - pre_reduce_dims[1] = patchRows * patchCols * patchPlanes; - if (isColMajor) { - pre_reduce_dims[0] = post_reduce_dims[0]; - pre_reduce_dims[2] = post_reduce_dims[1] * post_reduce_dims[2] * post_reduce_dims[3] * post_reduce_dims[4]; - } else { - pre_reduce_dims[0] = post_reduce_dims[0] * post_reduce_dims[1] * post_reduce_dims[2] * post_reduce_dims[3]; - pre_reduce_dims[2] = post_reduce_dims[4]; - } - -#if !defined(EIGEN_HAS_INDEX_LIST) - // nvcc doesn't support cxx11 - Eigen::array<int, 1> reduction_dims; - reduction_dims[0] = 1; -#else - // Take advantage of cxx11 to give the compiler information it can use to - // optimize the code. - Eigen::IndexList<Eigen::type2index<1> > reduction_dims; -#endif - return input.extract_volume_patches(patchPlanes, patchRows, patchCols, - stridePlanes, strideRows, strideCols, - padding_type, -Eigen::NumTraits<float>::highest()) - .reshape(pre_reduce_dims) - .maximum(reduction_dims) - .reshape(post_reduce_dims); -} - - -/** SpatialAvgPooling - * \ingroup CXX11_NeuralNetworks_Module - * - * \brief Applies an average pooling over a multichannel input image. - * - * The input parameter is expected to be a tensor with a rank of 4 (channels, height, width, others in col-major, and the reverse of that in row-major). - * - * The result can be assigned to a tensor of rank equal to the rank of the input. The dimensions of the result will be channels, height, width, and others (in col-major, and the reverse of that if the input was row-major). - * - * The order of the width and height dimensions can be swapped if needed. - * -*/ -namespace internal { - -template <typename T> struct AvgPoolMeanReducer -{ -#if (EIGEN_ARCH_i386 || EIGEN_ARCH_x86_64) && !defined(__CUDACC__) - // We only support packet access for floats. - static const bool PacketAccess = internal::is_same<T, float>::value; -#else - static const bool PacketAccess = false; -#endif - static const bool IsStateful = true; - - EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE AvgPoolMeanReducer() : scalarCount_(0) { - typedef typename packet_traits<T>::type Packet; - packetCount_ = pset1<Packet>(0.0); - } - - EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void reduce(const T t, T* accum) { - if (t != -Eigen::NumTraits<T>::highest()) { - (*accum) = (*accum) + t; - scalarCount_++; - } - } - - EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T initialize() const { - return static_cast<T>(0); - } - - EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T finalize(const T accum) const { - eigen_assert(scalarCount_ > 0); - return accum / scalarCount_; - } - -#if (EIGEN_ARCH_i386 || EIGEN_ARCH_x86_64) && !defined(__CUDACC__) -#ifdef EIGEN_VECTORIZE_AVX -#define pequal(a,b) _mm256_cmp_ps(a,b,_CMP_EQ_UQ) -#define psel(a,b,false_mask) _mm256_blendv_ps(a,b,false_mask) -#else -#define pequal(a,b) _mm_cmpeq_ps(a,b) -#define psel(a,b,false_mask) _mm_or_ps(_mm_andnot_ps(false_mask, a), _mm_and_ps(false_mask, b)) -#endif - - template <typename Packet> - EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void reducePacket(const Packet& p, Packet* accum) { - reducePacketWithType(static_cast<T>(0), p, accum); - } - - template <typename Packet> - void reducePacketWithType(T, const Packet& p, Packet* accum) { - Packet skip_mask = pequal(p, pset1<Packet>(-Eigen::NumTraits<T>::highest())); - (*accum) = padd<Packet>(*accum, psel(p, pset1<Packet>(0), skip_mask)); - packetCount_ = padd<Packet>(packetCount_, psel(pset1<Packet>(1), pset1<Packet>(0), skip_mask)); - } - - template <typename Packet> - EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet initializePacket() const { - return pset1<Packet>(0); - } - - template <typename Packet> - EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet finalizePacket(const Packet& vaccum) const { - return pdiv(vaccum, packetCount_); - } - template <typename Packet> - EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T finalizeBoth(const T saccum, const Packet& vaccum) const { - return (saccum + predux(vaccum)) / (scalarCount_ + predux(packetCount_)); - } -#endif - - protected: - typedef typename packet_traits<T>::type Packet; - int scalarCount_; - Packet packetCount_; -}; - -} // namespace internal - -#if !defined(EIGEN_HAS_INDEX_LIST) -template <typename Input> -EIGEN_ALWAYS_INLINE -static const TensorReshapingOp<const Eigen::DSizes<typename internal::traits<Input>::Index, internal::traits<Input>::NumDimensions>, const TensorReductionOp<internal::AvgPoolMeanReducer<typename internal::remove_const<typename internal::traits<Input>::Scalar>::type>, const Eigen::array<int, 2>, const TensorImagePatchOp<Dynamic, Dynamic, const Input> > > -#else -template <typename Input> -EIGEN_ALWAYS_INLINE -static const TensorReshapingOp<const Eigen::DSizes<typename internal::traits<Input>::Index, internal::traits<Input>::NumDimensions>, const TensorReductionOp<internal::AvgPoolMeanReducer<typename internal::remove_const<typename internal::traits<Input>::Scalar>::type>, typename internal::conditional<internal::traits<Input>::Layout == ColMajor, const Eigen::IndexList<Eigen::type2index<1>, Eigen::type2index<2> >, const Eigen::IndexList<Eigen::type2index<2>, Eigen::type2index<3> > >::type, const TensorImagePatchOp<Dynamic, Dynamic, const Input> > > -#endif -SpatialAvgPooling(const Input& input, DenseIndex patchRows, DenseIndex patchCols, - DenseIndex strideRows, DenseIndex strideCols, const PaddingType padding_type, - DenseIndex in_strideRows = 1, DenseIndex in_strideCols = 1) -{ - EIGEN_STATIC_ASSERT(internal::traits<Input>::NumDimensions == 4, YOU_MADE_A_PROGRAMMING_MISTAKE); - - typedef typename internal::traits<Input>::Index TensorIndex; - TensorRef<Tensor<typename internal::traits<Input>::Scalar, internal::traits<Input>::NumDimensions, internal::traits<Input>::Layout, TensorIndex> > in(input); - - const DenseIndex patchRowsEff = patchRows + (patchRows - 1) * (in_strideRows - 1); - const DenseIndex patchColsEff = patchCols + (patchCols - 1) * (in_strideCols - 1); - - static const bool isColMajor = (internal::traits<Input>::Layout == ColMajor); - static const int idxRows = isColMajor ? 1 : 2; - static const int idxCols = isColMajor ? 2 : 1; - - // Molds the output of the reduction into the shape expected by the user. - // (assuming col-major): - // - 1st dim: channels - // - 2nd dim: output height - // - 3rd dim: output width - // - 4th dim and beyond: everything else including batch size - Eigen::DSizes<TensorIndex, internal::traits<Input>::NumDimensions> post_reduce_dims; - post_reduce_dims[0] = in.dimension(0); - if (padding_type == PADDING_VALID) { - post_reduce_dims[idxRows] = numext::ceil((in.dimension(idxRows) - patchRowsEff + 1.f) / static_cast<float>(strideRows)); - post_reduce_dims[idxCols] = numext::ceil((in.dimension(idxCols) - patchColsEff + 1.f) / static_cast<float>(strideCols)); - } else { - post_reduce_dims[idxRows] = numext::ceil(in.dimension(idxRows) / static_cast<float>(strideRows)); - post_reduce_dims[idxCols] = numext::ceil(in.dimension(idxCols) / static_cast<float>(strideCols)); - } - post_reduce_dims[3] = in.dimension(3); - - typedef typename internal::remove_const<typename internal::traits<Input>::Scalar>::type CoeffReturnType; - internal::AvgPoolMeanReducer<CoeffReturnType> mean_with_nan; - -#if !defined(EIGEN_HAS_INDEX_LIST) - // nvcc doesn't support cxx11 - Eigen::array<int, 2> reduction_dims; - if (isColMajor) { - reduction_dims[0] = 1; - reduction_dims[1] = 2; - } else { - reduction_dims[0] = 2; - reduction_dims[1] = 3; - } -#else - // Take advantage of cxx11 to give the compiler information it can use to - // optimize the code. - typename internal::conditional<internal::traits<Input>::Layout == ColMajor, const Eigen::IndexList<Eigen::type2index<1>, Eigen::type2index<2> >, const Eigen::IndexList<Eigen::type2index<2>, Eigen::type2index<3> > >::type reduction_dims; -#endif - return input.extract_image_patches(patchRows, patchCols, strideRows, strideCols, in_strideRows, in_strideCols, padding_type, -Eigen::NumTraits<typename internal::remove_const<typename internal::traits<Input>::Scalar>::type>::highest()).reduce(reduction_dims, mean_with_nan).reshape(post_reduce_dims); -} - - -/** CuboidAvgPooling - * \ingroup CXX11_NeuralNetworks_Module - * - * \brief Applies an average pooling over a multichannel input volume. - * - * The input parameter is expected to be a tensor with a rank of 5 (channels, depth, height, width, others, and the reverse of that in row-major). - * - * The result can be assigned to a tensor of rank equal to the rank of the input. The dimensions of the result will be channels, depth, width, and others (in col-major, and the reverse of that if the input was row-major). - * - * The order of the depth, width and height dimensions can be swapped if needed. - * -*/ -#if !defined(EIGEN_HAS_INDEX_LIST) -template <typename Input> -EIGEN_ALWAYS_INLINE static const TensorReshapingOp< - const Eigen::DSizes<DenseIndex, internal::traits<Input>::NumDimensions>, - const TensorReductionOp< - internal::AvgPoolMeanReducer<float>, const Eigen::array<int, 1>, - const TensorReshapingOp< - const Eigen::DSizes<DenseIndex, 3>, - const TensorVolumePatchOp<Dynamic, Dynamic, Dynamic, const Input> > > > -#else -template <typename Input> -EIGEN_ALWAYS_INLINE static const TensorReshapingOp< - const Eigen::DSizes<DenseIndex, internal::traits<Input>::NumDimensions>, - const TensorReductionOp< - internal::AvgPoolMeanReducer<float>, - const Eigen::IndexList<Eigen::type2index<1> >, - const TensorReshapingOp< - const Eigen::DSizes<DenseIndex, 3>, - const TensorVolumePatchOp<Dynamic, Dynamic, Dynamic, const Input> > > > -#endif -CuboidAvgPooling(const Input& input, DenseIndex patchPlanes, - DenseIndex patchRows, DenseIndex patchCols, - DenseIndex stridePlanes, DenseIndex strideRows, - DenseIndex strideCols, const PaddingType padding_type) { - EIGEN_STATIC_ASSERT(internal::traits<Input>::NumDimensions == 5, YOU_MADE_A_PROGRAMMING_MISTAKE); - static const bool isColMajor = (internal::traits<Input>::Layout == ColMajor); - - typedef typename internal::traits<Input>::Index TensorIndex; - TensorRef<Tensor<typename internal::traits<Input>::Scalar, internal::traits<Input>::NumDimensions, internal::traits<Input>::Layout, TensorIndex> > in(input); - - static const int idxPlanes = isColMajor ? 1 : 3; - static const int idxRows = 2; - static const int idxCols = isColMajor ? 3 : 1; - // Molds the output of the reduction into the shape expected by the used - // (assuming col-major): - // - 1st dim: channels - // - 2nd dim: outupt depth - // - 3rd dim: output height - // - 4th dim: output width - // - 5th dim and beyond: everything else including batch size - Eigen::DSizes<DenseIndex, internal::traits<Input>::NumDimensions> post_reduce_dims; - post_reduce_dims[0] = in.dimension(0); - if (padding_type == PADDING_VALID) { - post_reduce_dims[idxPlanes] = numext::ceil((in.dimension(idxPlanes) - patchPlanes + 1.f) / static_cast<float>(stridePlanes)); - post_reduce_dims[idxRows] = numext::ceil((in.dimension(idxRows) - patchRows + 1.f) / static_cast<float>(strideRows)); - post_reduce_dims[idxCols] = numext::ceil((in.dimension(idxCols) - patchCols + 1.f) / static_cast<float>(strideCols)); - } else { - post_reduce_dims[idxPlanes] = numext::ceil(in.dimension(idxPlanes) / static_cast<float>(stridePlanes)); - post_reduce_dims[idxRows] = numext::ceil(in.dimension(idxRows) / static_cast<float>(strideRows)); - post_reduce_dims[idxCols] = numext::ceil(in.dimension(idxCols) / static_cast<float>(strideCols)); - } - post_reduce_dims[4] = in.dimension(4); - - Eigen::DSizes<DenseIndex, 3> pre_reduce_dims; - pre_reduce_dims[1] = patchRows * patchCols * patchPlanes; - if (isColMajor) { - pre_reduce_dims[0] = post_reduce_dims[0]; - pre_reduce_dims[2] = post_reduce_dims[1] * post_reduce_dims[2] * post_reduce_dims[3] * post_reduce_dims[4]; - } else { - pre_reduce_dims[0] = post_reduce_dims[0] * post_reduce_dims[1] * post_reduce_dims[2] * post_reduce_dims[3]; - pre_reduce_dims[2] = post_reduce_dims[4]; - } - - typedef typename internal::remove_const<typename internal::traits<Input>::Scalar>::type CoeffReturnType; - internal::AvgPoolMeanReducer<CoeffReturnType> mean_with_nan; - -#if !defined(EIGEN_HAS_INDEX_LIST) - // nvcc doesn't support cxx11 - Eigen::array<int, 1> reduction_dims; - reduction_dims[0] = 1; -#else - // Take advantage of cxx11 to give the compiler information it can use to - // optimize the code. - Eigen::IndexList<Eigen::type2index<1> > reduction_dims; -#endif - return input.extract_volume_patches(patchPlanes, patchRows, patchCols, - stridePlanes, strideRows, strideCols, - padding_type, -Eigen::NumTraits<float>::highest()) - .reshape(pre_reduce_dims) - .reduce(reduction_dims, mean_with_nan) - .reshape(post_reduce_dims); -} - -} // end namespace Eigen - -#endif // EIGEN_CXX11_NEURAL_NETWORKS_POOLING_H diff --git a/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/SoftMax.h b/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/SoftMax.h deleted file mode 100644 index f0e21ab9c2..0000000000 --- a/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/SoftMax.h +++ /dev/null @@ -1,83 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. -// -// Copyright (C) 2014 Benoit Steiner <benoit.steiner.goog@gmail.com> -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. -#ifndef EIGEN_CXX11_NEURAL_NETWORKS_SOFTMAX_H -#define EIGEN_CXX11_NEURAL_NETWORKS_SOFTMAX_H - -namespace Eigen { - -/** SoftMax - * \ingroup CXX11_NeuralNetworks_Module - * - * \brief Applies a softmax - * - * The input parameter is expected to be a col-major tensor with a rank of 2 (depth and other). - * - * The result can be assigned to a tensor of rank and dimensions equal to that of the input. The result will be laid out in col-major order. - * -*/ - -namespace { -class SoftmaxOp { - public: - EIGEN_ALWAYS_INLINE SoftmaxOp(const float beta) : beta_(beta) { } - - template <typename Input> EIGEN_ALWAYS_INLINE - typename Input::Dimensions dimensions(const Input& input) const { - return input.dimensions(); - } - - template <typename Input, typename Output, typename Device> - void eval(const Input& input, Output& output, const Device& device) const - { -#if !defined(EIGEN_HAS_INDEX_LIST) - // nvcc doesn't support cxx11 - Eigen::array<typename internal::traits<Input>::Index, 1> depth_dim; - depth_dim[0] = 0; - Eigen::array<typename internal::traits<Input>::Index, 2> bcast; - bcast[0] = dimensions(input)[0]; - bcast[1] = 1; - DSizes<typename internal::traits<Input>::Index, 2> dims2d; - dims2d[0] = 1; - dims2d[1] = dimensions(input)[1]; -#else - // Take advantage of cxx11 to give the compiler information it can use to - // optimize the code. - Eigen::IndexList<Eigen::type2index<0>> depth_dim; - Eigen::IndexList<int, Eigen::type2index<1>> bcast; - bcast.set(0, dimensions(input)[0]); - Eigen::IndexList<Eigen::type2index<1>, typename internal::traits<Input>::Index> dims2d; - dims2d.set(1, dimensions(input)[1]); -#endif - - output.device(device) = ((input - input.maximum(depth_dim).eval().reshape(dims2d).broadcast(bcast)) * beta_).exp(); - output.device(device) = output / (output.sum(depth_dim).eval().reshape(dims2d).broadcast(bcast)); - } - - private: - const float beta_; -}; -} - - -template <typename Input> -EIGEN_ALWAYS_INLINE -static const TensorCustomUnaryOp<const SoftmaxOp, const Input> -SoftMax(const Input& input, const float beta) -{ - EIGEN_STATIC_ASSERT(internal::traits<Input>::Layout == ColMajor, YOU_MADE_A_PROGRAMMING_MISTAKE); - EIGEN_STATIC_ASSERT(internal::traits<Input>::NumDimensions == 2, YOU_MADE_A_PROGRAMMING_MISTAKE); - - const SoftmaxOp op(beta); - return input.customOp(op); -} - - -} // end namespace Eigen - -#endif // EIGEN_CXX11_NEURAL_NETWORKS_SOFTMAX_H diff --git a/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/SpatialConvolutions.h b/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/SpatialConvolutions.h deleted file mode 100644 index 8e2ddca6b5..0000000000 --- a/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/SpatialConvolutions.h +++ /dev/null @@ -1,775 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. -// -// Copyright (C) 2014 Benoit Steiner <benoit.steiner.goog@gmail.com> -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. -#ifndef EIGEN_CXX11_NEURAL_NETWORKS_SPATIAL_CONVOLUTIONS_H -#define EIGEN_CXX11_NEURAL_NETWORKS_SPATIAL_CONVOLUTIONS_H - -namespace Eigen { - -namespace internal { - -// These optimizations require vector instructions -#ifdef EIGEN_VECTORIZE - -// TODO: Consolidate this part of the code with the image patch extraction code -// since they are both very similar. -template <typename NewDimension, DenseIndex Rows, DenseIndex Cols, typename ArgType, typename Device, - typename Scalar_, typename Index, - typename nocontract_t, typename contract_t, - int Side, size_t packet_size, - bool inner_dim_contiguous, bool inner_dim_reordered, int Alignment> -class TensorContractionInputMapper<Scalar_, Index, Side, TensorEvaluator<const TensorReshapingOp<NewDimension, const TensorImagePatchOp<Rows, Cols, ArgType> >, Device>, nocontract_t, contract_t, packet_size, inner_dim_contiguous, inner_dim_reordered, Alignment> -{ - public: - typedef TensorContractionInputMapper<Scalar_, Index, Side, TensorEvaluator<const TensorReshapingOp<NewDimension, const TensorImagePatchOp<Rows, Cols, ArgType> >, Device>, nocontract_t, contract_t, packet_size, inner_dim_contiguous, inner_dim_reordered, Alignment> Self; - typedef TensorContractionSubMapper<Scalar_, Index, Side, TensorEvaluator<const TensorReshapingOp<NewDimension, const TensorImagePatchOp<Rows, Cols, ArgType> >, Device>, nocontract_t, contract_t, packet_size, inner_dim_contiguous, inner_dim_reordered, Alignment> SubMapper; - typedef SubMapper VectorMapper; - typedef SubMapper LinearMapper; - typedef Scalar_ Scalar; - typedef typename packet_traits<Scalar>::type Packet; - - TensorContractionInputMapper(const TensorEvaluator<const TensorReshapingOp<NewDimension, const TensorImagePatchOp<Rows, Cols, ArgType> >, Device>& tensor, - const nocontract_t&, const nocontract_t&, - const contract_t&, const contract_t&) - : m_impl(tensor.impl().impl()) - { - Index patch_rows; - Index patch_depth; - if (internal::traits<ArgType>::Layout == ColMajor) { - patch_depth = tensor.impl().dimensions()[0]; - patch_rows = tensor.impl().dimensions()[1]; - m_patch_cols = tensor.impl().dimensions()[2]; - m_num_patches = tensor.impl().dimensions()[3]; - } else { - static const int NumDims = tensor.impl().dimensions().size(); - patch_depth = tensor.impl().dimensions()[NumDims - 1]; - patch_rows = tensor.impl().dimensions()[NumDims - 2]; - m_patch_cols = tensor.impl().dimensions()[NumDims - 3]; - m_num_patches = tensor.impl().dimensions()[NumDims - 4]; - } - m_patch_row_inflate_strides = tensor.impl().rowInflateStride(); - m_patch_col_inflate_strides = tensor.impl().colInflateStride(); - - m_colStride = patch_rows; - - m_outputRows = tensor.impl().outputRows(); - m_row_strides = tensor.impl().userRowStride(); - m_col_strides = tensor.impl().userColStride(); - - m_in_row_strides = tensor.impl().userInRowStride(); - m_in_col_strides = tensor.impl().userInColStride(); - - if (internal::traits<ArgType>::Layout == ColMajor) { - m_inputRows = tensor.impl().impl().dimensions()[1]; - m_inputCols = tensor.impl().impl().dimensions()[2]; - } else { - static const int NumDims = tensor.impl().impl().dimensions().size(); - m_inputRows = tensor.impl().impl().dimensions()[NumDims - 2]; - m_inputCols = tensor.impl().impl().dimensions()[NumDims - 3]; - } - - m_rowInputStride = patch_depth; - m_colInputStride = patch_depth * m_inputRows; - m_patchInputStride = patch_depth * m_inputRows * m_inputCols; - - m_rowPaddingTop = tensor.impl().rowPaddingTop(); - m_colPaddingLeft = tensor.impl().colPaddingLeft(); - - m_fastInputRowStride = internal::TensorIntDivisor<Index>(m_patch_row_inflate_strides); - m_fastInputColStride = internal::TensorIntDivisor<Index>(m_patch_col_inflate_strides); - m_fastNumPatches = internal::TensorIntDivisor<Index>(m_num_patches); - m_fastColStride = internal::TensorIntDivisor<Index>(m_colStride); - m_fastOutputRows = internal::TensorIntDivisor<Index>(m_outputRows); - m_fastDimZero = internal::TensorIntDivisor<Index>(patch_depth); - } - - TensorContractionInputMapper(const TensorContractionInputMapper& base_mapper) : - m_impl(base_mapper.m_impl) { - m_patch_cols = base_mapper.m_patch_cols; - m_num_patches = base_mapper.m_num_patches; - m_patch_row_inflate_strides = base_mapper.m_patch_row_inflate_strides; - m_patch_col_inflate_strides = base_mapper.m_patch_col_inflate_strides; - - m_colStride = base_mapper.m_colStride; - - m_rowInputStride = base_mapper.m_rowInputStride; - m_colInputStride = base_mapper.m_colInputStride; - m_patchInputStride = base_mapper.m_patchInputStride; - - m_inputRows = base_mapper.m_inputRows; - m_inputCols = base_mapper.m_inputCols; - - m_outputRows = base_mapper.m_outputRows; - m_row_strides = base_mapper.m_row_strides; - m_col_strides = base_mapper.m_col_strides; - - m_in_row_strides = base_mapper.m_in_row_strides; - m_in_col_strides = base_mapper.m_in_col_strides; - - m_rowPaddingTop = base_mapper.m_rowPaddingTop; - m_colPaddingLeft = base_mapper.m_colPaddingLeft; - - m_fastInputRowStride = base_mapper.m_fastInputRowStride; - m_fastInputColStride = base_mapper.m_fastInputColStride; - m_fastNumPatches = base_mapper.m_fastNumPatches; - m_fastColStride = base_mapper.m_fastColStride; - m_fastOutputRows = base_mapper.m_fastOutputRows; - m_fastDimZero = base_mapper.m_fastDimZero; - } - - // If true, turns off some optimizations for loading packets since the image - // patches are "non-standard" such as there are non-trivial strides or - // inflations in the input. - EIGEN_DEVICE_FUNC - EIGEN_ALWAYS_INLINE bool nonStandardPatches() const { - return m_in_row_strides != 1 || m_in_col_strides != 1 || m_patch_row_inflate_strides != 1 || m_patch_col_inflate_strides != 1; - } - - EIGEN_DEVICE_FUNC - EIGEN_STRONG_INLINE SubMapper getSubMapper(Index i, Index j) const { - return SubMapper(*this, i, j); - } - - EIGEN_DEVICE_FUNC - EIGEN_STRONG_INLINE LinearMapper getLinearMapper(Index i, Index j) const { - return LinearMapper(*this, i, j); - } - - EIGEN_DEVICE_FUNC - EIGEN_ALWAYS_INLINE Scalar operator()(Index row) const { - Index rowIndex, colIndex, otherIndex; - computeBaseIndices(0, rowIndex, colIndex, otherIndex); - return loadCoeff(row, rowIndex, colIndex, otherIndex); - } - - // Load the coefficient at the patchIndex location instead of the usual m_rowIndex, - // m_colIndex, m_otherIndex. This is currently only used by the gpu code. EIGEN_DEVICE_FUNC - EIGEN_DEVICE_FUNC - EIGEN_STRONG_INLINE Scalar operator()(Index row, Index patchIndex) const { - Index rowIndex, colIndex, otherIndex; - computeBaseIndices(patchIndex, rowIndex, colIndex, otherIndex); - return loadCoeff(row, rowIndex, colIndex, otherIndex); - } - - EIGEN_DEVICE_FUNC - EIGEN_ALWAYS_INLINE Packet loadPacket(Index row) const { - Index rowIndex, colIndex, otherIndex; - computeBaseIndices(0, rowIndex, colIndex, otherIndex); - return loadPacket(row, rowIndex, colIndex, otherIndex); - } - - // Load the packet at the patchIndex location instead of the usual m_rowIndex, - // m_colIndex, m_otherIndex. This is currently only used by the gpu code. - EIGEN_DEVICE_FUNC - EIGEN_ALWAYS_INLINE Packet loadPacket(Index row, Index patchIndex) const { - Index rowIndex, colIndex, otherIndex; - computeBaseIndices(patchIndex, rowIndex, colIndex, otherIndex); - return loadPacket(row, rowIndex, colIndex, otherIndex); - } - - EIGEN_DEVICE_FUNC - EIGEN_ALWAYS_INLINE const TensorEvaluator<ArgType, Device>& impl() const { return m_impl; } - - EIGEN_DEVICE_FUNC - EIGEN_ALWAYS_INLINE Index patchDepth() const { return m_rowInputStride; } - EIGEN_DEVICE_FUNC - EIGEN_ALWAYS_INLINE Index patchRows() const { return m_colStride; } - EIGEN_DEVICE_FUNC - EIGEN_ALWAYS_INLINE Index patchCols() const { return m_patch_cols; } - - EIGEN_DEVICE_FUNC - EIGEN_ALWAYS_INLINE Packet packetNoPadding(const Index depth, const Index baseIndex) const { - const Index inputIndex = depth + baseIndex; - return m_impl.template packet<Unaligned>(inputIndex); - } - - private: - friend class TensorContractionSubMapper<Scalar, Index, Side, TensorEvaluator<const TensorReshapingOp<NewDimension, const TensorImagePatchOp<Rows, Cols, ArgType> >, Device>, nocontract_t, contract_t, packet_size, inner_dim_contiguous, inner_dim_reordered, Alignment>; - - EIGEN_DEVICE_FUNC - EIGEN_STRONG_INLINE Scalar loadCoeff(Index patchId, Index rowIndex, Index colIndex, Index otherIndex) const { - // Find the offset of the element wrt the location of the first element. - const Index patchOffset = patchId / m_fastDimZero; - - const Index colOffset = patchOffset / m_fastColStride; - const Index inputCol = colIndex + colOffset * m_in_col_strides; - const Index origInputCol = (m_patch_col_inflate_strides == 1) ? inputCol : ((inputCol >= 0) ? (inputCol / m_fastInputColStride) : 0); - const Index rowOffset = patchOffset - colOffset * m_colStride; - const Index inputRow = rowIndex + rowOffset * m_in_row_strides; - const Index origInputRow = (m_patch_row_inflate_strides == 1) ? inputRow : ((inputRow >= 0) ? (inputRow / m_fastInputRowStride) : 0); - if (origInputCol < 0 | origInputRow < 0 | origInputCol >= m_inputCols | origInputRow >= m_inputRows | - (inputCol != origInputCol * m_patch_col_inflate_strides) | (inputRow != origInputRow * m_patch_row_inflate_strides)) { - return Scalar(0); - } - const Index depth = patchId - patchOffset * patchDepth(); - const Index inputIndex = depth + origInputRow * m_rowInputStride + origInputCol * m_colInputStride + otherIndex; - return m_impl.coeff(inputIndex); - } - - EIGEN_DEVICE_FUNC - EIGEN_STRONG_INLINE Scalar loadCoeffStandard(Index patchId, Index rowIndex, Index colIndex, Index otherIndex) const { - eigen_assert(!nonStandardPatches()); - - // Find the offset of the element wrt the location of the first element. - const Index patchOffset = patchId / m_fastDimZero; - - const Index colOffset = patchOffset / m_fastColStride; - const Index inputCol = colIndex + colOffset; - const Index rowOffset = patchOffset - colOffset * m_colStride; - const Index inputRow = rowIndex + rowOffset; - if (inputCol < 0 || inputCol >= m_inputCols || inputRow < 0 || inputRow >= m_inputRows) { - return Scalar(0); - } - const Index depth = patchId - patchOffset * patchDepth(); - const Index inputIndex = depth + inputRow * m_rowInputStride + inputCol * m_colInputStride + otherIndex; - return m_impl.coeff(inputIndex); - } - - EIGEN_DEVICE_FUNC - EIGEN_ALWAYS_INLINE Packet loadPacket(Index patchId, Index rowIndex, Index colIndex, Index otherIndex) const { - const Index packetSize = internal::unpacket_traits<Packet>::size; - EIGEN_STATIC_ASSERT(packetSize > 1, YOU_MADE_A_PROGRAMMING_MISTAKE) - eigen_assert(patchId < patchDepth()*patchRows()*m_patch_cols); - - if (nonStandardPatches()) { - return packetWithPossibleZero(patchId, rowIndex, colIndex, otherIndex); - } - return loadPacketStandard(patchId, rowIndex, colIndex, otherIndex); - } - - EIGEN_DEVICE_FUNC - EIGEN_ALWAYS_INLINE Packet loadPacketStandard(Index patchId, Index rowIndex, Index colIndex, Index otherIndex) const { - const Index packetSize = internal::unpacket_traits<Packet>::size; - EIGEN_STATIC_ASSERT(packetSize > 1, YOU_MADE_A_PROGRAMMING_MISTAKE) - eigen_assert(patchId < patchDepth()*patchRows()*m_patch_cols); - - eigen_assert(!nonStandardPatches()); - - if ((patchDepth() % packetSize) == 0) { - return loadPacketFast(patchId, rowIndex, colIndex, otherIndex); - } - else { - const Index patchOffsets[2] = {patchId / m_fastDimZero, (patchId + packetSize - 1) / m_fastDimZero}; - - const Index colOffsets[2] = {patchOffsets[0] / m_fastColStride, patchOffsets[1] / m_fastColStride}; - - const Index inputCols[2] = {colIndex + colOffsets[0], colIndex + colOffsets[1]}; - if (inputCols[0] >= m_inputCols | inputCols[1] < 0) { - // all zeros - return internal::pset1<Packet>(Scalar(0)); - } - - if (inputCols[0] == inputCols[1]) { - const Index rowOffsets[2] = {patchOffsets[0] - colOffsets[0]*m_colStride, patchOffsets[1] - colOffsets[1]*m_colStride}; - eigen_assert(rowOffsets[0] <= rowOffsets[1]); - const Index inputRows[2] = {rowIndex + rowOffsets[0], rowIndex + rowOffsets[1]}; - - if (inputRows[0] >= m_inputRows | inputRows[1] < 0) { - // all zeros - return internal::pset1<Packet>(Scalar(0)); - } - - if (inputRows[0] >= 0 & inputRows[1] < m_inputRows) { - // no padding - const Index depth = patchId - patchOffsets[0] * patchDepth(); - const Index inputIndex = depth + inputRows[0] * m_rowInputStride + inputCols[0] * m_colInputStride + otherIndex; - return m_impl.template packet<Unaligned>(inputIndex); - } - } - } - return packetWithPossibleZero(patchId, rowIndex, colIndex, otherIndex); - } - - EIGEN_DEVICE_FUNC - EIGEN_ALWAYS_INLINE Packet loadPacketFast(Index patchId, Index rowIndex, Index colIndex, Index otherIndex) const { - const Index packetSize = internal::unpacket_traits<Packet>::size; - EIGEN_STATIC_ASSERT(packetSize > 1, YOU_MADE_A_PROGRAMMING_MISTAKE) - eigen_assert(patchId < patchDepth()*patchRows()*m_patch_cols); - - eigen_assert(!nonStandardPatches()); - eigen_assert((patchDepth() % packetSize) == 0); - // Find the offset of the element wrt the location of the first element. - const Index patchOffset = patchId / m_fastDimZero; - eigen_assert((patchId + packetSize - 1) / m_fastDimZero == patchOffset); - - const Index colOffset = patchOffset / m_fastColStride; - const Index inputCol = colIndex + colOffset; - const Index rowOffset = patchOffset - colOffset*m_colStride; - const Index inputRow = rowIndex + rowOffset; - if (inputCol < 0 | inputRow < 0 | inputCol >= m_inputCols | inputRow >= m_inputRows) { - // all zeros - return internal::pset1<Packet>(Scalar(0)); - } - // no padding - const Index depth = patchId - patchOffset * patchDepth(); - const Index inputIndex = depth + inputRow * m_rowInputStride + inputCol * m_colInputStride + otherIndex; - return m_impl.template packet<Unaligned>(inputIndex); - } - - EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Packet packetWithPossibleZero(Index patchId, Index rowIndex, Index colIndex, Index otherIndex) const - { - const int packetSize = internal::unpacket_traits<Packet>::size; - EIGEN_ALIGN_MAX typename internal::remove_const<Scalar>::type values[packetSize]; - for (int i = 0; i < packetSize; ++i) { - values[i] = loadCoeff(patchId+i, rowIndex, colIndex, otherIndex); - } - Packet rslt = internal::pload<Packet>(values); - return rslt; - } - - EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void computeBaseIndices(Index patchIndex, Index& rowIndex, Index& colIndex, Index& otherIndex) const { - const int NumInputDims = array_size<typename TensorEvaluator<ArgType, Device>::Dimensions>::value; - otherIndex = (NumInputDims == 3) ? 0 : patchIndex / m_fastNumPatches; - const Index patch2DIndex = (NumInputDims == 3) ? patchIndex : (patchIndex - otherIndex * m_num_patches); - otherIndex *= m_patchInputStride; - colIndex = patch2DIndex / m_fastOutputRows; - rowIndex = patch2DIndex - colIndex * m_outputRows; - colIndex = colIndex * m_col_strides - m_colPaddingLeft; - rowIndex = rowIndex * m_row_strides - m_rowPaddingTop; - } - - Index m_patch_cols; // number of colums in the patch - Index m_num_patches; // number of patches to extract. - Index m_patch_row_inflate_strides; // the strides for row inflation in the image patch - Index m_patch_col_inflate_strides; // the strides for col inflation in the image patch - // Fast representation of inflation strides. - internal::TensorIntDivisor<Index> m_fastInputRowStride; - internal::TensorIntDivisor<Index> m_fastInputColStride; - - Index m_otherStride; - Index m_colStride; - internal::TensorIntDivisor<Index> m_fastNumPatches; - internal::TensorIntDivisor<Index> m_fastColStride; - - Index m_rowInputStride; // row stride in the input tensor - Index m_colInputStride; // col stride in the input tensor - Index m_patchInputStride; // patch stride in the input tensor - - Index m_inputRows; // Number of rows in the input tensor - Index m_inputCols; // Number of cols in the input tensor - - Index m_outputRows; // Number of patch rows - - Index m_row_strides; // User specified row stride - Index m_col_strides; // User specified col stride - - Index m_in_row_strides; // User specified input row stride - Index m_in_col_strides; // User specified input col stride - - Index m_rowPaddingTop; // Row padding - Index m_colPaddingLeft; // Column padding - - internal::TensorIntDivisor<Index> m_fastOutputRows; - internal::TensorIntDivisor<Index> m_fastDimZero; - - const TensorEvaluator<ArgType, Device> m_impl; -}; - - -template <typename NewDimension, DenseIndex Rows, DenseIndex Cols, typename ArgType, typename Device, - typename Scalar_, typename Index, - typename nocontract_t, typename contract_t, - int Side, size_t packet_size, - bool inner_dim_contiguous, bool inner_dim_reordered, int Alignment> -class TensorContractionSubMapper<Scalar_, Index, Side, TensorEvaluator<const TensorReshapingOp<NewDimension, const TensorImagePatchOp<Rows, Cols, ArgType> >, Device>, nocontract_t, contract_t, packet_size, inner_dim_contiguous, inner_dim_reordered, Alignment> -{ - public: - typedef Scalar_ Scalar; - typedef typename packet_traits<Scalar>::type Packet; - typedef typename packet_traits<Scalar>::half HalfPacket; - - typedef TensorContractionInputMapper<Scalar, Index, Side, TensorEvaluator<const TensorReshapingOp<NewDimension, const TensorImagePatchOp<Rows, Cols, ArgType> >, Device>, nocontract_t, contract_t, packet_size, inner_dim_contiguous, inner_dim_reordered, Alignment> ParentMapper; - typedef TensorContractionSubMapper<Scalar, Index, Side, TensorEvaluator<const TensorReshapingOp<NewDimension, const TensorImagePatchOp<Rows, Cols, ArgType> >, Device>, nocontract_t, contract_t, packet_size, inner_dim_contiguous, inner_dim_reordered, Alignment> Self; - typedef Self LinearMapper; - - EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorContractionSubMapper(const ParentMapper& base_mapper, Index vert_offset, Index horiz_offset) - : m_base_mapper(base_mapper), m_depth_offset(vert_offset), m_col_offset(horiz_offset) { - m_base_mapper.computeBaseIndices(m_col_offset, m_rowIndex, m_colIndex, m_otherIndex); - } - EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorContractionSubMapper(const Self& base_mapper, Index vert_offset, Index horiz_offset) - : m_base_mapper(base_mapper.m_base_mapper), m_depth_offset(vert_offset+base_mapper.m_depth_offset), m_col_offset(horiz_offset+base_mapper.m_col_offset) { - m_base_mapper.computeBaseIndices(m_col_offset, m_rowIndex, m_colIndex, m_otherIndex); - } - EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Scalar operator()(Index i) const { - return m_base_mapper.loadCoeff(i + m_depth_offset, m_rowIndex, m_colIndex, m_otherIndex); - } - EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Scalar operator()(Index i, Index j) const { - return m_base_mapper(i + m_depth_offset, j + m_col_offset); - } - - EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Packet loadPacket(Index i) const { - return m_base_mapper.loadPacket(i + m_depth_offset, m_rowIndex, m_colIndex, m_otherIndex); - } - EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Packet loadPacket(Index i, Index j) const { - return m_base_mapper.template loadPacket(i + m_depth_offset, j + m_col_offset); - } - - EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Scalar loadCoeffStandard(Index i) const { - return m_base_mapper.loadCoeffStandard(i + m_depth_offset, m_rowIndex, m_colIndex, m_otherIndex); - } - - EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Packet loadPacketFast(Index i) const { - return m_base_mapper.loadPacketFast(i + m_depth_offset, m_rowIndex, m_colIndex, m_otherIndex); - } - EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Packet loadPacketStandard(Index i) const { - return m_base_mapper.loadPacketStandard(i + m_depth_offset, m_rowIndex, m_colIndex, m_otherIndex); - } - template <typename Packet> - EIGEN_DEVICE_FUNC bool aligned(Index) const { - return false; - } - - EIGEN_DEVICE_FUNC - EIGEN_ALWAYS_INLINE bool nonStandardPatches() const { - return m_base_mapper.nonStandardPatches(); - } - - EIGEN_DEVICE_FUNC - EIGEN_ALWAYS_INLINE Index patchDepth() const { return m_base_mapper.m_rowInputStride; } - EIGEN_DEVICE_FUNC - EIGEN_ALWAYS_INLINE Index patchRows() const { return m_base_mapper.m_colStride; } - EIGEN_DEVICE_FUNC - EIGEN_ALWAYS_INLINE Index patchCols() const { return m_base_mapper.m_patch_cols; } - - EIGEN_DEVICE_FUNC - EIGEN_ALWAYS_INLINE Packet packetNoPadding(const Index depth, const Index baseIndex) const { - const Index inputIndex = depth + baseIndex; - return m_base_mapper.m_impl.template packet<Unaligned>(inputIndex); - } - - EIGEN_DEVICE_FUNC - EIGEN_ALWAYS_INLINE bool padRow(const Index row) const { - const Index r = m_rowIndex + row; - return r < 0 | r >= m_base_mapper.m_inputRows; - } - EIGEN_DEVICE_FUNC - EIGEN_ALWAYS_INLINE bool padCol(const Index col) const { - const Index c = m_colIndex + col; - return c < 0 | c >= m_base_mapper.m_inputCols; - } - EIGEN_DEVICE_FUNC - EIGEN_ALWAYS_INLINE Index baseIndex(const Index row, const Index col) const { - const Index r = m_rowIndex + row; - const Index c = m_colIndex + col; - return r * m_base_mapper.m_rowInputStride + c * m_base_mapper.m_colInputStride + m_otherIndex; - } - - EIGEN_DEVICE_FUNC - EIGEN_ALWAYS_INLINE Index rowOffset() const { - const Index patchOffset = m_depth_offset / m_base_mapper.m_fastDimZero; - const Index colOffset = patchOffset / m_base_mapper.m_fastColStride; - return patchOffset-colOffset*m_base_mapper.m_colStride; - } - - EIGEN_DEVICE_FUNC - EIGEN_ALWAYS_INLINE Index colOffset() const { - const Index patchOffset = m_depth_offset / m_base_mapper.m_fastDimZero; - const Index colOffset = patchOffset / m_base_mapper.m_fastColStride; - return colOffset; - } - - EIGEN_DEVICE_FUNC - EIGEN_ALWAYS_INLINE Index depthOffset() const { - const Index patchOffset = m_depth_offset % m_base_mapper.patchDepth(); - return patchOffset; - } - - EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE LinearMapper getLinearMapper(Index i, Index j) const { - return LinearMapper(m_base_mapper, i + m_depth_offset, j + m_col_offset); - } - - private: - const ParentMapper& m_base_mapper; // that was a reference before - Index m_depth_offset; // First row in the input matrix - Index m_col_offset; // First col in the input matrix - - Index m_rowIndex; // precomputed row index corresponding to the col offset - Index m_colIndex; // precomputed col index corresponding to the col offset - Index m_otherIndex; // precomputed other index corresponding to the col offset - -}; - - -template <typename NewDimension, DenseIndex Rows, DenseIndex Cols, typename ArgType, typename Device, - typename Scalar, typename Index, - typename nocontract_t, typename contract_t, - int Side, size_t packet_size, - bool inner_dim_contiguous, bool inner_dim_reordered, int Alignment, int nr> -struct gemm_pack_rhs<Scalar, Index, TensorContractionSubMapper<Scalar, Index, Side, TensorEvaluator<const TensorReshapingOp<NewDimension, const TensorImagePatchOp<Rows, Cols, ArgType> >, Device>, nocontract_t, contract_t, packet_size, inner_dim_contiguous, inner_dim_reordered, Alignment>, nr, ColMajor, false, false> { - - typedef TensorContractionSubMapper<Scalar, Index, Side, TensorEvaluator<const TensorReshapingOp<NewDimension, const TensorImagePatchOp<Rows, Cols, ArgType> >, Device>, nocontract_t, contract_t, packet_size, inner_dim_contiguous, inner_dim_reordered, Alignment> SubMapper; - typedef SubMapper DataMapper; - - static inline Index ceil_div(Index a, Index b) { - return (a + b - 1) / b; - } - - EIGEN_DONT_INLINE void operator()(Scalar* block, const DataMapper& rhs, Index depth, Index cols, Index stride=0, Index offset=0) const { - eigen_assert(stride == 0); - eigen_assert(offset == 0); - - EIGEN_STATIC_ASSERT((nr == 4), YOU_MADE_A_PROGRAMMING_MISTAKE); - typedef typename DataMapper::LinearMapper LinearMapper; - typedef typename packet_traits<Scalar>::type Packet; - - const Index packet_cols4 = (cols/4) * 4; - const Index peeled_k = (depth/packet_size) * packet_size; - const bool non_standard_patches = rhs.nonStandardPatches(); - - for(Index j2=0; j2<packet_cols4; j2+=4) - { - const SubMapper dm0 = rhs.getLinearMapper(0, j2 + 0); - const SubMapper dm1 = rhs.getLinearMapper(0, j2 + 1); - const SubMapper dm2 = rhs.getLinearMapper(0, j2 + 2); - const SubMapper dm3 = rhs.getLinearMapper(0, j2 + 3); - - Index k=0; - if((packet_size%4)==0 && !non_standard_patches) - { - const Index patch_depth = rhs.patchDepth(); - if ((patch_depth % packet_size) == 0) { - const Index patch_cols = rhs.patchCols(); - const Index patch_rows = rhs.patchRows(); - - const Index startCol = rhs.colOffset(); - const Index max_cols = std::min<Index>(ceil_div(peeled_k, patch_rows*patch_depth)+startCol, patch_cols); - - for (Index c = startCol; c < max_cols; ++c) { - eigen_assert(k < peeled_k); - const Index startRow = (c == startCol) ? rhs.rowOffset() : 0; - const Index max_rows = std::min<Index>(ceil_div(peeled_k-c*patch_rows*patch_depth, patch_depth)+startRow, patch_rows); - - const bool pad_col0 = dm0.padCol(c); - const bool pad_col1 = dm1.padCol(c); - const bool pad_col2 = dm2.padCol(c); - const bool pad_col3 = dm3.padCol(c); - for (Index r = startRow; r < max_rows; ++r) { - eigen_assert(k < peeled_k); - const bool pad0 = pad_col0 || dm0.padRow(r); - const bool pad1 = pad_col1 || dm1.padRow(r); - const bool pad2 = pad_col2 || dm2.padRow(r); - const bool pad3 = pad_col3 || dm3.padRow(r); - - const Index idx0 = dm0.baseIndex(r, c); - const Index idx1 = dm1.baseIndex(r, c); - const Index idx2 = dm2.baseIndex(r, c); - const Index idx3 = dm3.baseIndex(r, c); - - const Index startDepth = ((c == startCol) && (r == startRow)) ? rhs.depthOffset() : 0; - const Index max_depth = std::min<Index>(peeled_k-c*patch_rows*patch_depth-r*patch_depth+startDepth, patch_depth); - eigen_assert(max_depth % packet_size == 0); - for (Index d = startDepth; d < max_depth; d += packet_size) { - eigen_assert(k < peeled_k); - PacketBlock<Packet, 4> kernel; - kernel.packet[0] = pad0 ? pset1<Packet>(0) : rhs.packetNoPadding(d, idx0); - kernel.packet[1] = pad1 ? pset1<Packet>(0) : rhs.packetNoPadding(d, idx1); - kernel.packet[2] = pad2 ? pset1<Packet>(0) : rhs.packetNoPadding(d, idx2); - kernel.packet[3] = pad3 ? pset1<Packet>(0) : rhs.packetNoPadding(d, idx3); - ptranspose(kernel); - pstoreu(block+0*packet_size, kernel.packet[0]); - pstoreu(block+1*packet_size, kernel.packet[1]); - pstoreu(block+2*packet_size, kernel.packet[2]); - pstoreu(block+3*packet_size, kernel.packet[3]); - block+=4*packet_size; - k += packet_size; - } - } - } - - for(; k<peeled_k; k+=packet_size) { - PacketBlock<Packet, 4> kernel; - kernel.packet[0] = dm0.loadPacketFast(k); - kernel.packet[1] = dm1.loadPacketFast(k); - kernel.packet[2] = dm2.loadPacketFast(k); - kernel.packet[3] = dm3.loadPacketFast(k); - ptranspose(kernel); - pstoreu(block+0*packet_size, kernel.packet[0]); - pstoreu(block+1*packet_size, kernel.packet[1]); - pstoreu(block+2*packet_size, kernel.packet[2]); - pstoreu(block+3*packet_size, kernel.packet[3]); - block+=4*packet_size; - } - } - else { - for(; k<peeled_k; k+=packet_size) { - PacketBlock<Packet, 4> kernel; - kernel.packet[0] = dm0.loadPacketStandard(k); - kernel.packet[1] = dm1.loadPacketStandard(k); - kernel.packet[2] = dm2.loadPacketStandard(k); - kernel.packet[3] = dm3.loadPacketStandard(k); - ptranspose(kernel); - pstoreu(block+0*packet_size, kernel.packet[0]); - pstoreu(block+1*packet_size, kernel.packet[1]); - pstoreu(block+2*packet_size, kernel.packet[2]); - pstoreu(block+3*packet_size, kernel.packet[3]); - block+=4*packet_size; - } - } - } - if (!rhs.nonStandardPatches()) { - for(; k<depth; k++) - { - block[0] = dm0.loadCoeffStandard(k); - block[1] = dm1.loadCoeffStandard(k); - block[2] = dm2.loadCoeffStandard(k); - block[3] = dm3.loadCoeffStandard(k); - block += 4; - } - } - else { - for(; k<depth; k++) - { - block[0] = dm0(k); - block[1] = dm1(k); - block[2] = dm2(k); - block[3] = dm3(k); - block += 4; - } - } - } - - // copy the remaining columns one at a time (nr==1) - for(Index j2=packet_cols4; j2<cols; ++j2) - { - const SubMapper dm0 = rhs.getLinearMapper(0, j2); - for(Index k=0; k<depth; k++) - { - *block = dm0(k); - block += 1; - } - } - } -}; - -#endif // EIGEN_VECTORIZE -} // end namespace internal - - -/** SpatialConvolution - * \ingroup CXX11_NeuralNetworks_Module - * - * \brief Applies a 2D convolution over a multichannel input image. - * - * The input parameter is expected to be a tensor with a rank of 3 or more (channels, height, width, and optionally others) - * The kernel parameter is expected to be a 4D tensor (filters, channels, kernel_height, kernel_width) - * The input and the kernel must both be in col-major layout. The result will also be in col-major layout. - * - * If in_stride > 1, then applies convolution with holes (aka atrous convolution), sampling every in_stride input pixels. - * - * The result can be assigned to a tensor of rank equal to the rank of the input. The dimensions of the result will be filters, height, width (and others if applicable). - * - * It is possible to swap the order of the width and height dimensions provided that the same order is used in the input, the kernel, and the output. - * - */ -template <typename Input, typename Kernel> -EIGEN_ALWAYS_INLINE -static const typename internal::conditional< - internal::traits<Input>::Layout == ColMajor, - TensorReshapingOp<const DSizes<typename internal::traits<Input>::Index, internal::traits<Input>::NumDimensions>, const TensorContractionOp<const array<IndexPair<typename internal::traits<Input>::Index>, 1>, const TensorReshapingOp<const DSizes<typename internal::traits<Input>::Index, 2>, const Kernel>, const TensorReshapingOp<const DSizes<typename internal::traits<Input>::Index, 2>, const TensorImagePatchOp<Dynamic, Dynamic, const Input> > > >, - TensorReshapingOp<const DSizes<typename internal::traits<Input>::Index, internal::traits<Input>::NumDimensions>, const TensorContractionOp<const array<IndexPair<typename internal::traits<Input>::Index>, 1>, const TensorReshapingOp<const DSizes<typename internal::traits<Input>::Index, 2>, const TensorImagePatchOp<Dynamic, Dynamic, const Input> >, const TensorReshapingOp<const DSizes<typename internal::traits<Input>::Index, 2>, const Kernel> > > >::type -SpatialConvolution(const Input& input, const Kernel& kernel, const DenseIndex stride = 1, const PaddingType padding_type = PADDING_SAME, const DenseIndex in_stride = 1) { - - typedef typename internal::traits<Input>::Index TensorIndex; - TensorRef<Tensor<typename internal::traits<Input>::Scalar, internal::traits<Input>::NumDimensions, internal::traits<Input>::Layout, TensorIndex> > in(input); - TensorRef<Tensor<typename internal::traits<Kernel>::Scalar, internal::traits<Kernel>::NumDimensions, internal::traits<Kernel>::Layout, TensorIndex> > kern(kernel); - - EIGEN_STATIC_ASSERT(internal::traits<Input>::Layout == internal::traits<Kernel>::Layout, YOU_MADE_A_PROGRAMMING_MISTAKE); - static const bool isColMajor = (internal::traits<Input>::Layout == ColMajor); - - static const int NumDims = internal::traits<Input>::NumDimensions; - - // Number of filters to apply. This is the same as the output depth of the result - const TensorIndex kernelFilters = isColMajor ? kern.dimensions()[0] : kern.dimensions()[3]; - // Number of channels. This is the same as the input depth. - const TensorIndex kernelChannels = isColMajor ? kern.dimensions()[1] : kern.dimensions()[2]; - const TensorIndex kernelRows = isColMajor ? kern.dimensions()[2] : kern.dimensions()[1]; - const TensorIndex kernelCols = isColMajor ? kern.dimensions()[3] : kern.dimensions()[0]; - - const DenseIndex kernelRowsEff = kernelRows + (kernelRows - 1) * (in_stride - 1); - const DenseIndex kernelColsEff = kernelCols + (kernelCols - 1) * (in_stride - 1); - - array<IndexPair<TensorIndex>, 1> contract_dims; - contract_dims[0] = IndexPair<TensorIndex>(1, 0); - - const TensorIndex InputRows = isColMajor ? in.dimension(1) : in.dimension(NumDims - 2); - const TensorIndex InputCols = isColMajor ? in.dimension(2) : in.dimension(NumDims - 3); - - TensorIndex out_height; - TensorIndex out_width; - switch (padding_type) { - case PADDING_VALID: - out_height = numext::ceil((InputRows - kernelRowsEff + 1.f) / static_cast<float>(stride)); - out_width = numext::ceil((InputCols - kernelColsEff + 1.f) / static_cast<float>(stride)); - break; - case PADDING_SAME: - out_height = numext::ceil(InputRows / static_cast<float>(stride)); - out_width = numext::ceil(InputCols / static_cast<float>(stride)); - break; - default: - eigen_assert(false && "unexpected padding"); - } - - // Molds the output of the patch extraction code into a 2d tensor: - // - the first dimension (dims[0]): the patch values to be multiplied with the kernels - // - the second dimension (dims[1]): everything else - DSizes<TensorIndex, 2> pre_contract_dims; - if (isColMajor) { - pre_contract_dims[0] = kernelChannels * kernelRows * kernelCols; - pre_contract_dims[1] = out_height * out_width; - for (int i = 3; i < NumDims; ++i) { - pre_contract_dims[1] *= in.dimension(i); - } - } else { - pre_contract_dims[1] = kernelChannels * kernelRows * kernelCols; - pre_contract_dims[0] = out_height * out_width; - for (int i = 0; i < NumDims - 3; ++i) { - pre_contract_dims[0] *= in.dimension(i); - } - } - - // Molds the output of the contraction into the shape expected by the used - // (assuming this is ColMajor): - // - 1st dim: kernel filters - // - 2nd dim: output height - // - 3rd dim: output width - // - 4th dim and beyond: everything else including batch size - DSizes<TensorIndex, NumDims> post_contract_dims; - if (isColMajor) { - post_contract_dims[0] = kernelFilters; - post_contract_dims[1] = out_height; - post_contract_dims[2] = out_width; - for (int i = 3; i < NumDims; ++i) { - post_contract_dims[i] = in.dimension(i); - } - } else { - post_contract_dims[NumDims - 1] = kernelFilters; - post_contract_dims[NumDims - 2] = out_height; - post_contract_dims[NumDims - 3] = out_width; - for (int i = 0; i < NumDims - 3; ++i) { - post_contract_dims[i] = in.dimension(i); - } - } - - DSizes<TensorIndex, 2> kernel_dims; - if (isColMajor) { - kernel_dims[0] = kernelFilters; - kernel_dims[1] = kernelChannels * kernelRows * kernelCols; - } else { - kernel_dims[0] = kernelChannels * kernelRows * kernelCols; - kernel_dims[1] = kernelFilters; - } - // TODO(yangke): choose() is defined in TensorContraction.h -- consider - // moving it to somewhere more "common". - return choose(Cond<internal::traits<Input>::Layout == ColMajor>(), - kernel.reshape(kernel_dims).contract(input.extract_image_patches(kernelRows, kernelCols, stride, stride, in_stride, in_stride, padding_type).reshape(pre_contract_dims), contract_dims).reshape(post_contract_dims), - input.extract_image_patches(kernelRows, kernelCols, stride, stride, in_stride, in_stride, padding_type).reshape(pre_contract_dims).contract(kernel.reshape(kernel_dims), contract_dims).reshape(post_contract_dims)); -} - -} // end namespace Eigen - -#endif // EIGEN_CXX11_NEURAL_NETWORKS_SPATIAL_CONVOLUTIONS_H diff --git a/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/TensorConvolutionByFFT.h b/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/TensorConvolutionByFFT.h deleted file mode 100644 index 0e72173536..0000000000 --- a/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/TensorConvolutionByFFT.h +++ /dev/null @@ -1,289 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. -// -// Copyright (C) 2014 Benoit Steiner <benoit.steiner.goog@gmail.com> -// Copyright (C) 2015 Jianwei Cui <thucjw@gmail.com> -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#ifndef EIGEN_CXX11_TENSOR_TENSOR_CONVOLUTIONBYFFT_H -#define EIGEN_CXX11_TENSOR_TENSOR_CONVOLUTIONBYFFT_H - -namespace Eigen { - -/** \class TensorConvolutionByFFT - * \ingroup CXX11_Tensor_Module - * - * \brief Tensor convolution class. - * - * - */ -namespace internal { - - -template<typename Dimensions, typename InputXprType, typename KernelXprType> -struct traits<TensorConvolutionByFFTOp<Dimensions, InputXprType, KernelXprType> > -{ - // Type promotion to handle the case where the types of the lhs and the rhs are different. - typedef typename promote_storage_type<typename InputXprType::Scalar, - typename KernelXprType::Scalar>::ret Scalar; - typedef typename packet_traits<Scalar>::type Packet; - typedef typename promote_storage_type<typename traits<InputXprType>::StorageKind, - typename traits<KernelXprType>::StorageKind>::ret StorageKind; - typedef typename promote_index_type<typename traits<InputXprType>::Index, - typename traits<KernelXprType>::Index>::type Index; - typedef typename InputXprType::Nested LhsNested; - typedef typename KernelXprType::Nested RhsNested; - typedef typename remove_reference<LhsNested>::type _LhsNested; - typedef typename remove_reference<RhsNested>::type _RhsNested; - static const int NumDimensions = traits<InputXprType>::NumDimensions; - static const int Layout = traits<InputXprType>::Layout; - - enum { - Flags = 0, - }; -}; - -template<typename Dimensions, typename InputXprType, typename KernelXprType> -struct eval<TensorConvolutionByFFTOp<Dimensions, InputXprType, KernelXprType>, Eigen::Dense> -{ - typedef const TensorConvolutionByFFTOp<Dimensions, InputXprType, KernelXprType>& type; -}; - -template<typename Dimensions, typename InputXprType, typename KernelXprType> -struct nested<TensorConvolutionByFFTOp<Dimensions, InputXprType, KernelXprType>, 1, typename eval<TensorConvolutionByFFTOp<Dimensions, InputXprType, KernelXprType> >::type> -{ - typedef TensorConvolutionByFFTOp<Dimensions, InputXprType, KernelXprType> type; -}; - -} // end namespace internal - - - -template<typename Indices, typename InputXprType, typename KernelXprType> -class TensorConvolutionByFFTOp : public TensorBase<TensorConvolutionByFFTOp<Indices, InputXprType, KernelXprType> > -{ - public: - typedef typename Eigen::internal::traits<TensorConvolutionByFFTOp>::Scalar Scalar; - typedef typename Eigen::internal::traits<TensorConvolutionByFFTOp>::Packet Packet; - typedef typename Eigen::NumTraits<Scalar>::Real RealScalar; - typedef typename internal::promote_storage_type<typename InputXprType::CoeffReturnType, - typename KernelXprType::CoeffReturnType>::ret CoeffReturnType; - typedef typename internal::promote_storage_type<typename InputXprType::PacketReturnType, - typename KernelXprType::PacketReturnType>::ret PacketReturnType; - typedef typename Eigen::internal::nested<TensorConvolutionByFFTOp>::type Nested; - typedef typename Eigen::internal::traits<TensorConvolutionByFFTOp>::StorageKind StorageKind; - typedef typename Eigen::internal::traits<TensorConvolutionByFFTOp>::Index Index; - - EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorConvolutionByFFTOp(const InputXprType& input, const KernelXprType& kernel, const Indices& dims) - : m_input_xpr(input), m_kernel_xpr(kernel), m_indices(dims) {} - - EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE - const Indices& indices() const { return m_indices; } - - /** \returns the nested expressions */ - EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE - const typename internal::remove_all<typename InputXprType::Nested>::type& - inputExpression() const { return m_input_xpr; } - - EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE - const typename internal::remove_all<typename KernelXprType::Nested>::type& - kernelExpression() const { return m_kernel_xpr; } - - protected: - typename InputXprType::Nested m_input_xpr; - typename KernelXprType::Nested m_kernel_xpr; - const Indices m_indices; -}; - - -template<typename Indices, typename InputArgType, typename KernelArgType, typename Device> -struct TensorEvaluator<const TensorConvolutionByFFTOp<Indices, InputArgType, KernelArgType>, Device> -{ - typedef TensorConvolutionByFFTOp<Indices, InputArgType, KernelArgType> XprType; - - typedef typename XprType::Scalar Scalar; - typedef typename XprType::CoeffReturnType CoeffReturnType; - typedef typename XprType::PacketReturnType PacketReturnType; - - typedef typename Eigen::NumTraits<Scalar>::Real RealScalar; - - static const int NumDims = internal::array_size<typename TensorEvaluator<InputArgType, Device>::Dimensions>::value; - static const int NumKernelDims = internal::array_size<Indices>::value; - typedef typename XprType::Index Index; - typedef DSizes<Index, NumDims> Dimensions; - - enum { - IsAligned = TensorEvaluator<InputArgType, Device>::IsAligned & - TensorEvaluator<KernelArgType, Device>::IsAligned, - PacketAccess = false, - BlockAccess = false, - Layout = TensorEvaluator<InputArgType, Device>::Layout, - CoordAccess = false, // to be implemented - }; - - EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device) - : m_inputImpl(op.inputExpression(), device), m_kernelImpl(op.kernelExpression(), device), m_kernelArg(op.kernelExpression()), m_kernel(NULL), m_local_kernel(false), m_device(device) - { - EIGEN_STATIC_ASSERT((static_cast<int>(TensorEvaluator<InputArgType, Device>::Layout) == static_cast<int>(TensorEvaluator<KernelArgType, Device>::Layout)), YOU_MADE_A_PROGRAMMING_MISTAKE); - - const typename TensorEvaluator<InputArgType, Device>::Dimensions& input_dims = m_inputImpl.dimensions(); - const typename TensorEvaluator<KernelArgType, Device>::Dimensions& kernel_dims = m_kernelImpl.dimensions(); - - if (static_cast<int>(Layout) == static_cast<int>(ColMajor)) { - m_inputStride[0] = 1; - for (int i = 1; i < NumDims; ++i) { - m_inputStride[i] = m_inputStride[i - 1] * input_dims[i - 1]; - } - } else { - m_inputStride[NumDims - 1] = 1; - for (int i = NumDims - 2; i >= 0; --i) { - m_inputStride[i] = m_inputStride[i + 1] * input_dims[i + 1]; - } - } - - m_dimensions = m_inputImpl.dimensions(); - if (static_cast<int>(Layout) == static_cast<int>(ColMajor)) { - for (int i = 0; i < NumKernelDims; ++i) { - const Index index = op.indices()[i]; - const Index input_dim = input_dims[index]; - const Index kernel_dim = kernel_dims[i]; - const Index result_dim = input_dim - kernel_dim + 1; - m_dimensions[index] = result_dim; - if (i > 0) { - m_kernelStride[i] = m_kernelStride[i - 1] * kernel_dims[i - 1]; - } else { - m_kernelStride[0] = 1; - } - m_indexStride[i] = m_inputStride[index]; - } - - m_outputStride[0] = 1; - for (int i = 1; i < NumDims; ++i) { - m_outputStride[i] = m_outputStride[i - 1] * m_dimensions[i - 1]; - } - } else { - for (int i = NumKernelDims - 1; i >= 0; --i) { - const Index index = op.indices()[i]; - const Index input_dim = input_dims[index]; - const Index kernel_dim = kernel_dims[i]; - const Index result_dim = input_dim - kernel_dim + 1; - m_dimensions[index] = result_dim; - if (i < NumKernelDims - 1) { - m_kernelStride[i] = m_kernelStride[i + 1] * kernel_dims[i + 1]; - } else { - m_kernelStride[NumKernelDims - 1] = 1; - } - m_indexStride[i] = m_inputStride[index]; - } - - m_outputStride[NumDims - 1] = 1; - for (int i = NumDims - 2; i >= 0; --i) { - m_outputStride[i] = m_outputStride[i + 1] * m_dimensions[i + 1]; - } - } - } - - EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dimensions; } - - - EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(Scalar* data) { - m_inputImpl.evalSubExprsIfNeeded(NULL); - m_kernelImpl.evalSubExprsIfNeeded(NULL); - - typedef typename internal::traits<InputArgType>::Index TensorIndex; - - Tensor<Scalar, NumDims, Layout, TensorIndex> input(m_inputImpl.dimensions()); - for (int i = 0; i < m_inputImpl.dimensions().TotalSize(); ++i) { - input.data()[i] = m_inputImpl.coeff(i); - } - - Tensor<Scalar, NumDims, Layout, TensorIndex> kernel(m_kernelImpl.dimensions()); - for (int i = 0; i < m_kernelImpl.dimensions().TotalSize(); ++i) { - kernel.data()[i] = m_kernelImpl.coeff(i); - } - - array<std::pair<ptrdiff_t, ptrdiff_t>, NumDims> paddings; - for (int i = 0; i < NumDims; ++i) { - paddings[i] = std::make_pair(0, m_inputImpl.dimensions()[i] - m_kernelImpl.dimensions()[i]); - } - - Eigen::array<bool, NumKernelDims> reverse; - for (int i = 0; i < NumKernelDims; ++i) { - reverse[i] = true; - } - - Eigen::array<bool, NumDims> fft; - for (int i = 0; i < NumDims; ++i) { - fft[i] = i; - } - - Eigen::DSizes<TensorIndex, NumDims> slice_offsets; - for (int i = 0; i < NumDims; ++i) { - slice_offsets[i] = m_kernelImpl.dimensions()[i] - 1; - } - - Eigen::DSizes<TensorIndex, NumDims> slice_extents; - for (int i = 0; i < NumDims; ++i) { - slice_extents[i] = m_inputImpl.dimensions()[i] - m_kernelImpl.dimensions()[i] + 1; - } - - Tensor<Scalar, NumDims, Layout, TensorIndex> kernel_variant = kernel.reverse(reverse).pad(paddings); - Tensor<std::complex<Scalar>, NumDims, Layout, TensorIndex> kernel_fft = kernel_variant.template fft<Eigen::BothParts, FFT_FORWARD>(fft); - //Tensor<std::complex<Scalar>, NumDims, Layout|IndexType> kernel_fft = kernel.reverse(reverse).pad(paddings).template fft<2>(fft); - Tensor<std::complex<Scalar>, NumDims, Layout, TensorIndex> input_fft = input.template fft<Eigen::BothParts, FFT_FORWARD>(fft); - Tensor<std::complex<Scalar>, NumDims, Layout, TensorIndex> prod = (input_fft * kernel_fft).template fft<Eigen::BothParts, FFT_REVERSE>(fft); - Tensor<std::complex<Scalar>, NumDims, Layout, TensorIndex> tensor_result = prod.slice(slice_offsets, slice_extents); - - for (int i = 0; i < tensor_result.size(); ++i) { - data[i] = std::real(tensor_result.data()[i]); - } - return false; - } - - EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() { - m_inputImpl.cleanup(); - if (m_local_kernel) { - m_device.deallocate((void*)m_kernel); - m_local_kernel = false; - } - m_kernel = NULL; - } - - void evalTo(typename XprType::Scalar* buffer) { - evalSubExprsIfNeeded(NULL); - for (int i = 0; i < dimensions().TotalSize(); ++i) { - buffer[i] += coeff(i); - } - cleanup(); - } - - EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE CoeffReturnType coeff(Index index) const - { - CoeffReturnType result = CoeffReturnType(0); - return result; - } - - EIGEN_DEVICE_FUNC Scalar* data() const { return NULL; } - - private: - array<Index, NumDims> m_inputStride; - array<Index, NumDims> m_outputStride; - - array<Index, NumKernelDims> m_indexStride; - array<Index, NumKernelDims> m_kernelStride; - TensorEvaluator<InputArgType, Device> m_inputImpl; - TensorEvaluator<KernelArgType, Device> m_kernelImpl; - Dimensions m_dimensions; - - KernelArgType m_kernelArg; - const Scalar* m_kernel; - bool m_local_kernel; - const Device& m_device; -}; - -} // end namespace Eigen - -#endif // EIGEN_CXX11_TENSOR_TENSOR_CONVOLUTIONBYFFT_H diff --git a/third_party/eigen_fix_cuda_compilation.patch b/third_party/eigen_fix_cuda_compilation.patch deleted file mode 100644 index b921a7c31d..0000000000 --- a/third_party/eigen_fix_cuda_compilation.patch +++ /dev/null @@ -1,38 +0,0 @@ -diff --git a/Eigen/src/Core/ProductEvaluators.h b/Eigen/src/Core/ProductEvaluators.h ---- a/Eigen/src/Core/ProductEvaluators.h -+++ b/Eigen/src/Core/ProductEvaluators.h -@@ -137,7 +137,7 @@ struct Assignment<DstXprType, Product<Lh - typename enable_if<(Options==DefaultProduct || Options==AliasFreeProduct)>::type> - { - typedef Product<Lhs,Rhs,Options> SrcXprType; -- static EIGEN_STRONG_INLINE -+ static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE - void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,Scalar> &) - { - Index dstRows = src.rows(); -@@ -390,7 +390,7 @@ struct generic_product_impl<Lhs,Rhs,Dens - typedef typename Product<Lhs,Rhs>::Scalar Scalar; - - template<typename Dst> -- static EIGEN_STRONG_INLINE void evalTo(Dst& dst, const Lhs& lhs, const Rhs& rhs) -+ static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalTo(Dst& dst, const Lhs& lhs, const Rhs& rhs) - { - // Same as: dst.noalias() = lhs.lazyProduct(rhs); - // but easier on the compiler side -@@ -398,14 +398,14 @@ struct generic_product_impl<Lhs,Rhs,Dens - } - - template<typename Dst> -- static EIGEN_STRONG_INLINE void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs) -+ static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs) - { - // dst.noalias() += lhs.lazyProduct(rhs); - call_assignment_no_alias(dst, lhs.lazyProduct(rhs), internal::add_assign_op<typename Dst::Scalar,Scalar>()); - } - - template<typename Dst> -- static EIGEN_STRONG_INLINE void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs) -+ static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs) - { - // dst.noalias() -= lhs.lazyProduct(rhs); - call_assignment_no_alias(dst, lhs.lazyProduct(rhs), internal::sub_assign_op<typename Dst::Scalar,Scalar>()); diff --git a/third_party/examples/eager/spinn/README.md b/third_party/examples/eager/spinn/README.md index 7f477d1920..e2fd8009a0 100644 --- a/third_party/examples/eager/spinn/README.md +++ b/third_party/examples/eager/spinn/README.md @@ -22,7 +22,7 @@ Other eager execution examples can be found under [tensorflow/contrib/eager/pyth - [`data.py`](../../../../tensorflow/contrib/eager/python/examples/spinn/data.py): Pipeline for loading and preprocessing the [SNLI](https://nlp.stanford.edu/projects/snli/) data and [GloVe](https://nlp.stanford.edu/projects/glove/) word embedding, written - using the [`tf.data`](https://www.tensorflow.org/programmers_guide/datasets) + using the [`tf.data`](https://www.tensorflow.org/guide/datasets) API. - [`spinn.py`](./spinn.py): Model definition and training routines. This example illustrates how one might perform the following actions with @@ -70,7 +70,7 @@ Other eager execution examples can be found under [tensorflow/contrib/eager/pyth - After training, you may use the model to perform inference on input data in the SNLI data format. The premise and hypotheses sentences are specified with the command-line flags `--inference_premise` and `--inference_hypothesis`, - respecitvely. Each sentence should include the words, as well as parentheses + respectively. Each sentence should include the words, as well as parentheses representing a binary parsing of the sentence. The words and parentheses should all be separated by spaces. For instance, diff --git a/third_party/examples/eager/spinn/spinn.py b/third_party/examples/eager/spinn/spinn.py index 8a1c7db2ea..de63ebe9e6 100644 --- a/third_party/examples/eager/spinn/spinn.py +++ b/third_party/examples/eager/spinn/spinn.py @@ -51,6 +51,9 @@ import tensorflow.contrib.eager as tfe from tensorflow.contrib.eager.python.examples.spinn import data +layers = tf.keras.layers + + def _bundle(lstm_iter): """Concatenate a list of Tensors along 1st axis and split result into two. @@ -78,17 +81,16 @@ def _unbundle(state): return tf.split(tf.concat(state, 1), state[0].shape[0], axis=0) -class Reducer(tfe.Network): +# pylint: disable=not-callable +class Reducer(tf.keras.Model): """A module that applies reduce operation on left and right vectors.""" def __init__(self, size, tracker_size=None): super(Reducer, self).__init__() - self.left = self.track_layer(tf.layers.Dense(5 * size, activation=None)) - self.right = self.track_layer( - tf.layers.Dense(5 * size, activation=None, use_bias=False)) + self.left = layers.Dense(5 * size, activation=None) + self.right = layers.Dense(5 * size, activation=None, use_bias=False) if tracker_size is not None: - self.track = self.track_layer( - tf.layers.Dense(5 * size, activation=None, use_bias=False)) + self.track = layers.Dense(5 * size, activation=None, use_bias=False) else: self.track = None @@ -123,7 +125,7 @@ class Reducer(tfe.Network): return h, c -class Tracker(tfe.Network): +class Tracker(tf.keras.Model): """A module that tracks the history of the sentence with an LSTM.""" def __init__(self, tracker_size, predict): @@ -134,10 +136,10 @@ class Tracker(tfe.Network): predict: (`bool`) Whether prediction mode is enabled. """ super(Tracker, self).__init__() - self._rnn = self.track_layer(tf.nn.rnn_cell.LSTMCell(tracker_size)) + self._rnn = tf.nn.rnn_cell.LSTMCell(tracker_size) self._state_size = tracker_size if predict: - self._transition = self.track_layer(tf.layers.Dense(4)) + self._transition = layers.Dense(4) else: self._transition = None @@ -182,7 +184,7 @@ class Tracker(tfe.Network): return unbundled, None -class SPINN(tfe.Network): +class SPINN(tf.keras.Model): """Stack-augmented Parser-Interpreter Neural Network. See https://arxiv.org/abs/1603.06021 for more details. @@ -204,9 +206,9 @@ class SPINN(tfe.Network): """ super(SPINN, self).__init__() self.config = config - self.reducer = self.track_layer(Reducer(config.d_hidden, config.d_tracker)) + self.reducer = Reducer(config.d_hidden, config.d_tracker) if config.d_tracker is not None: - self.tracker = self.track_layer(Tracker(config.d_tracker, config.predict)) + self.tracker = Tracker(config.d_tracker, config.predict) else: self.tracker = None @@ -248,7 +250,7 @@ class SPINN(tfe.Network): trans = transitions[i] if self.tracker: # Invoke tracker to obtain the current tracker states for the sentences. - tracker_states, trans_hypothesis = self.tracker(buffers, stacks) + tracker_states, trans_hypothesis = self.tracker(buffers, stacks=stacks) if trans_hypothesis: trans = tf.argmax(trans_hypothesis, axis=-1) else: @@ -273,7 +275,27 @@ class SPINN(tfe.Network): return _bundle([stack.pop() for stack in stacks])[0] -class SNLIClassifier(tfe.Network): +class Perceptron(tf.keras.Model): + """One layer of the SNLIClassifier multi-layer perceptron.""" + + def __init__(self, dimension, dropout_rate, previous_layer): + """Configure the Perceptron.""" + super(Perceptron, self).__init__() + self.dense = tf.keras.layers.Dense(dimension, activation=tf.nn.elu) + self.batchnorm = layers.BatchNormalization() + self.dropout = layers.Dropout(rate=dropout_rate) + self.previous_layer = previous_layer + + def call(self, x, training): + """Run previous Perceptron layers, then this one.""" + x = self.previous_layer(x, training=training) + x = self.dense(x) + x = self.batchnorm(x, training=training) + x = self.dropout(x, training=training) + return x + + +class SNLIClassifier(tf.keras.Model): """SNLI Classifier Model. A model aimed at solving the SNLI (Standford Natural Language Inference) @@ -304,29 +326,24 @@ class SNLIClassifier(tfe.Network): self.config = config self.embed = tf.constant(embed) - self.projection = self.track_layer(tf.layers.Dense(config.d_proj)) - self.embed_bn = self.track_layer(tf.layers.BatchNormalization()) - self.embed_dropout = self.track_layer( - tf.layers.Dropout(rate=config.embed_dropout)) - self.encoder = self.track_layer(SPINN(config)) - - self.feature_bn = self.track_layer(tf.layers.BatchNormalization()) - self.feature_dropout = self.track_layer( - tf.layers.Dropout(rate=config.mlp_dropout)) - - self.mlp_dense = [] - self.mlp_bn = [] - self.mlp_dropout = [] - for _ in xrange(config.n_mlp_layers): - self.mlp_dense.append(self.track_layer(tf.layers.Dense(config.d_mlp))) - self.mlp_bn.append( - self.track_layer(tf.layers.BatchNormalization())) - self.mlp_dropout.append( - self.track_layer(tf.layers.Dropout(rate=config.mlp_dropout))) - self.mlp_output = self.track_layer(tf.layers.Dense( + self.projection = layers.Dense(config.d_proj) + self.embed_bn = layers.BatchNormalization() + self.embed_dropout = layers.Dropout(rate=config.embed_dropout) + self.encoder = SPINN(config) + + self.feature_bn = layers.BatchNormalization() + self.feature_dropout = layers.Dropout(rate=config.mlp_dropout) + + current_mlp = lambda result, training: result + for _ in range(config.n_mlp_layers): + current_mlp = Perceptron(dimension=config.d_mlp, + dropout_rate=config.mlp_dropout, + previous_layer=current_mlp) + self.mlp = current_mlp + self.mlp_output = layers.Dense( config.d_out, kernel_initializer=tf.random_uniform_initializer(minval=-5e-3, - maxval=5e-3))) + maxval=5e-3)) def call(self, premise, @@ -383,15 +400,12 @@ class SNLIClassifier(tfe.Network): self.feature_bn(logits, training=training), training=training) # Apply the multi-layer perceptron on the logits. - for dense, bn, dropout in zip( - self.mlp_dense, self.mlp_bn, self.mlp_dropout): - logits = tf.nn.elu(dense(logits)) - logits = dropout(bn(logits, training=training), training=training) + logits = self.mlp(logits, training=training) logits = self.mlp_output(logits) return logits -class SNLIClassifierTrainer(object): +class SNLIClassifierTrainer(tfe.Checkpointable): """A class that coordinates the training of an SNLIClassifier.""" def __init__(self, snli_classifier, lr): @@ -405,7 +419,7 @@ class SNLIClassifierTrainer(object): # Create a custom learning rate Variable for the RMSProp optimizer, because # the learning rate needs to be manually decayed later (see # decay_learning_rate()). - self._learning_rate = tfe.Variable(lr, name="learning_rate") + self._learning_rate = tf.Variable(lr, name="learning_rate") self._optimizer = tf.train.RMSPropOptimizer(self._learning_rate, epsilon=1e-6) @@ -448,7 +462,7 @@ class SNLIClassifierTrainer(object): 2. logits as a dense `Tensor` of shape (batch_size, d_out), where d_out is the output dimension size of the SNLIClassifier. """ - with tfe.GradientTape() as tape: + with tf.GradientTape() as tape: tape.watch(self._model.variables) logits = self._model(premise, premise_transition, @@ -609,29 +623,27 @@ def train_or_infer_spinn(embed, with tf.device(device), \ summary_writer.as_default(), \ tf.contrib.summary.always_record_summaries(): - with tfe.restore_variables_on_create( - tf.train.latest_checkpoint(config.logdir)): - model = SNLIClassifier(config, embed) - global_step = tf.train.get_or_create_global_step() - trainer = SNLIClassifierTrainer(model, config.lr) + model = SNLIClassifier(config, embed) + global_step = tf.train.get_or_create_global_step() + trainer = SNLIClassifierTrainer(model, config.lr) + checkpoint = tf.train.Checkpoint(trainer=trainer, global_step=global_step) + checkpoint.restore(tf.train.latest_checkpoint(config.logdir)) if inference_sentence_pair: # Inference mode. - with tfe.restore_variables_on_create( - tf.train.latest_checkpoint(config.logdir)): - prem, prem_trans = inference_sentence_pair[0] - hypo, hypo_trans = inference_sentence_pair[1] - hypo_trans = inference_sentence_pair[1][1] - inference_logits = model( # pylint: disable=not-callable - tf.constant(prem), tf.constant(prem_trans), - tf.constant(hypo), tf.constant(hypo_trans), training=False) - inference_logits = inference_logits[0][1:] - max_index = tf.argmax(inference_logits) - print("\nInference logits:") - for i, (label, logit) in enumerate( - zip(data.POSSIBLE_LABELS, inference_logits)): - winner_tag = " (winner)" if max_index == i else "" - print(" {0:<16}{1:.6f}{2}".format(label + ":", logit, winner_tag)) + prem, prem_trans = inference_sentence_pair[0] + hypo, hypo_trans = inference_sentence_pair[1] + hypo_trans = inference_sentence_pair[1][1] + inference_logits = model( + tf.constant(prem), tf.constant(prem_trans), + tf.constant(hypo), tf.constant(hypo_trans), training=False) + inference_logits = inference_logits[0][1:] + max_index = tf.argmax(inference_logits) + print("\nInference logits:") + for i, (label, logit) in enumerate( + zip(data.POSSIBLE_LABELS, inference_logits)): + winner_tag = " (winner)" if max_index == i else "" + print(" {0:<16}{1:.6f}{2}".format(label + ":", logit, winner_tag)) return inference_logits train_len = train_data.num_batches(config.batch_size) @@ -650,20 +662,15 @@ def train_or_infer_spinn(embed, # remain on CPU. Same in _evaluate_on_dataset(). iterations += 1 - with tfe.restore_variables_on_create( - tf.train.latest_checkpoint(config.logdir)): - batch_train_loss, batch_train_logits = trainer.train_batch( - label, prem, prem_trans, hypo, hypo_trans) + batch_train_loss, batch_train_logits = trainer.train_batch( + label, prem, prem_trans, hypo, hypo_trans) batch_size = tf.shape(label)[0] mean_loss(batch_train_loss.numpy(), weights=batch_size.gpu() if use_gpu else batch_size) accuracy(tf.argmax(batch_train_logits, axis=1), label) if iterations % config.save_every == 0: - all_variables = trainer.variables + [global_step] - saver = tfe.Saver(all_variables) - saver.save(os.path.join(config.logdir, "ckpt"), - global_step=global_step) + checkpoint.save(os.path.join(config.logdir, "ckpt")) if iterations % config.dev_every == 0: dev_loss, dev_frac_correct = _evaluate_on_dataset( diff --git a/third_party/farmhash.BUILD b/third_party/farmhash.BUILD index a51e1511c1..4b8464684a 100644 --- a/third_party/farmhash.BUILD +++ b/third_party/farmhash.BUILD @@ -3,13 +3,6 @@ licenses(["notice"]) # MIT exports_files(["COPYING"]) config_setting( - name = "windows_msvc", - values = { - "cpu": "x64_windows_msvc", - }, -) - -config_setting( name = "windows", values = { "cpu": "x64_windows", @@ -23,7 +16,6 @@ cc_library( # Disable __builtin_expect support on Windows copts = select({ ":windows": ["/DFARMHASH_OPTIONAL_BUILTIN_EXPECT"], - ":windows_msvc": ["/DFARMHASH_OPTIONAL_BUILTIN_EXPECT"], "//conditions:default": [], }), includes = ["src/."], diff --git a/third_party/fft2d/fft2d.BUILD b/third_party/fft2d/fft2d.BUILD index 3dbd36aec0..74dd3112fc 100644 --- a/third_party/fft2d/fft2d.BUILD +++ b/third_party/fft2d/fft2d.BUILD @@ -14,6 +14,11 @@ FFT2D_SRCS = [ "fft/fftsg.c", ] +config_setting( + name = "windows", + values = {"cpu": "x64_windows"}, +) + # This is the main 2D FFT library. The 2D FFTs in this library call # 1D FFTs. In addition, fast DCTs are provided for the special case # of 8x8 and 16x16. This code in this library is referred to as @@ -21,7 +26,10 @@ FFT2D_SRCS = [ cc_library( name = "fft2d", srcs = FFT2D_SRCS, - linkopts = ["-lm"], + linkopts = select({ + ":windows": [], + "//conditions:default": ["-lm"], + }), ) objc_library( diff --git a/third_party/flatbuffers/BUILD b/third_party/flatbuffers/BUILD index fbdf19f205..82bab3ffd9 100644 --- a/third_party/flatbuffers/BUILD +++ b/third_party/flatbuffers/BUILD @@ -1,15 +1 @@ -package(default_visibility = ["//visibility:public"]) - -licenses(["notice"]) # Apache 2.0 - -filegroup( - name = "all_files", - srcs = glob( - ["**/*"], - exclude = [ - "**/METADATA", - "**/OWNERS", - ], - ), - visibility = ["//tensorflow:__subpackages__"], -) +# This empty BUILD file is required to make Bazel treat this directory as a package. diff --git a/third_party/flatbuffers/flatbuffers.BUILD b/third_party/flatbuffers/BUILD.bazel index 824c97be60..934c0d9650 100644 --- a/third_party/flatbuffers/flatbuffers.BUILD +++ b/third_party/flatbuffers/BUILD.bazel @@ -12,12 +12,17 @@ config_setting( visibility = ["//visibility:public"], ) -FLATBUFFERS_COPTS = [ - "-fexceptions", -] + select({ - "@bazel_tools//src:windows": [], - "@bazel_tools//src:windows_msvc": [], - "//conditions:default": ["-Wno-implicit-fallthrough"], +config_setting( + name = "windows", + values = {"cpu": "x64_windows"}, +) + +FLATBUFFERS_COPTS = select({ + ":windows": [], + "//conditions:default": [ + "-Wno-implicit-fallthrough", + "-fexceptions", + ], }) # Public flatc library to compile flatbuffer files at runtime. @@ -98,6 +103,8 @@ cc_binary( "grpc/src/compiler/cpp_generator.h", "grpc/src/compiler/go_generator.cc", "grpc/src/compiler/go_generator.h", + "grpc/src/compiler/java_generator.cc", + "grpc/src/compiler/java_generator.h", "grpc/src/compiler/schema_interface.h", "src/flatc_main.cpp", "src/idl_gen_cpp.cpp", @@ -119,6 +126,7 @@ cc_binary( ":freebsd": [ "-lm", ], + ":windows": [], "//conditions:default": [ "-lm", "-ldl", @@ -134,6 +142,7 @@ filegroup( srcs = [ "include/flatbuffers/base.h", "include/flatbuffers/flatbuffers.h", + "include/flatbuffers/minireflect.h", "include/flatbuffers/stl_emulation.h", "include/flatbuffers/util.h", ], diff --git a/third_party/flatbuffers/BUILD.system b/third_party/flatbuffers/BUILD.system new file mode 100644 index 0000000000..14fceada82 --- /dev/null +++ b/third_party/flatbuffers/BUILD.system @@ -0,0 +1,38 @@ +licenses(["notice"]) # Apache 2.0 + +filegroup( + name = "LICENSE.txt", + visibility = ["//visibility:public"], +) + +# Public flatc library to compile flatbuffer files at runtime. +cc_library( + name = "flatbuffers", + linkopts = ["-lflatbuffers"], + visibility = ["//visibility:public"], +) + +# Public flatc compiler library. +cc_library( + name = "flatc_library", + linkopts = ["-lflatbuffers"], + visibility = ["//visibility:public"], +) + +genrule( + name = "lnflatc", + outs = ["flatc.bin"], + cmd = "ln -s $$(which flatc) $@", +) + +# Public flatc compiler. +sh_binary( + name = "flatc", + srcs = ["flatc.bin"], + visibility = ["//visibility:public"], +) + +cc_library( + name = "runtime_cc", + visibility = ["//visibility:public"], +) diff --git a/third_party/flatbuffers/build_defs.bzl b/third_party/flatbuffers/build_defs.bzl index ae8d7feebe..235b44f7cf 100644 --- a/third_party/flatbuffers/build_defs.bzl +++ b/third_party/flatbuffers/build_defs.bzl @@ -1,5 +1,4 @@ -# Description: -# BUILD rules for generating flatbuffer files. +"""BUILD rules for generating flatbuffer files.""" flatc_path = "@flatbuffers//:flatc" @@ -8,66 +7,50 @@ DEFAULT_FLATC_ARGS = [ "--gen-object-api", ] -def flatbuffer_library_public(name, - srcs, - outs, - language_flag, - out_prefix="", - includes=[], - include_paths=[], - flatc_args=DEFAULT_FLATC_ARGS, - reflection_name="", - reflection_visiblity=None, - output_to_bindir=False): - '''Generates code files for reading/writing the given flatbuffers in the requested language using the public compiler. - - Args: - name: Rule name. - srcs: Source .fbs files. Sent in order to the compiler. - outs: Output files from flatc. - language_flag: Target language flag. One of [-c, -j, -js]. - out_prefix: Prepend this path to the front of all generated files except on - single source targets. Usually is a directory name. - includes: Optional, list of filegroups of schemas that the srcs depend on. - include_paths: Optional, list of paths the includes files can be found in. - flatc_args: Optional, list of additional arguments to pass to flatc. - reflection_name: Optional, if set this will generate the flatbuffer - reflection binaries for the schemas. - reflection_visiblity: The visibility of the generated reflection Fileset. - output_to_bindir: Passed to genrule for output to bin directory. - Outs: - filegroup(name): all generated source files. - Fileset([reflection_name]): (Optional) all generated reflection binaries. - ''' - include_paths_cmd = ["-I %s" % (s) for s in include_paths] - # '$(@D)' when given a single source target will give the appropriate - # directory. Appending 'out_prefix' is only necessary when given a build - # target with multiple sources. - output_directory = ( - ("-o $(@D)/%s" % (out_prefix)) if len(srcs) > 1 else ("-o $(@D)")) - genrule_cmd = " ".join([ - "for f in $(SRCS); do", - "$(location %s)" % (flatc_path), - " ".join(flatc_args), - " ".join(include_paths_cmd), - language_flag, - output_directory, - "$$f;", - "done", - ]) - native.genrule( - name=name, - srcs=srcs, - outs=outs, - output_to_bindir=output_to_bindir, - tools=includes + [flatc_path,], - cmd=genrule_cmd, - message="Generating flatbuffer files for %s:" % (name),) - if reflection_name: - reflection_genrule_cmd = " ".join([ +def flatbuffer_library_public( + name, + srcs, + outs, + language_flag, + out_prefix = "", + includes = [], + include_paths = [], + flatc_args = DEFAULT_FLATC_ARGS, + reflection_name = "", + reflection_visiblity = None, + output_to_bindir = False): + """Generates code files for reading/writing the given flatbuffers in the requested language using the public compiler. + + Outs: + filegroup(name): all generated source files. + Fileset([reflection_name]): (Optional) all generated reflection binaries. + + Args: + name: Rule name. + srcs: Source .fbs files. Sent in order to the compiler. + outs: Output files from flatc. + language_flag: Target language flag. One of [-c, -j, -js]. + out_prefix: Prepend this path to the front of all generated files except on + single source targets. Usually is a directory name. + includes: Optional, list of filegroups of schemas that the srcs depend on. + include_paths: Optional, list of paths the includes files can be found in. + flatc_args: Optional, list of additional arguments to pass to flatc. + reflection_name: Optional, if set this will generate the flatbuffer + reflection binaries for the schemas. + reflection_visiblity: The visibility of the generated reflection Fileset. + output_to_bindir: Passed to genrule for output to bin directory. + """ + include_paths_cmd = ["-I %s" % (s) for s in include_paths] + + # '$(@D)' when given a single source target will give the appropriate + # directory. Appending 'out_prefix' is only necessary when given a build + # target with multiple sources. + output_directory = ( + ("-o $(@D)/%s" % (out_prefix)) if len(srcs) > 1 else ("-o $(@D)") + ) + genrule_cmd = " ".join([ "for f in $(SRCS); do", "$(location %s)" % (flatc_path), - "-b --schema", " ".join(flatc_args), " ".join(include_paths_cmd), language_flag, @@ -75,122 +58,160 @@ def flatbuffer_library_public(name, "$$f;", "done", ]) - reflection_outs = [ - (out_prefix + "%s.bfbs") % (s.replace(".fbs", "").split("/")[-1]) for s in srcs - ] native.genrule( - name= "%s_srcs" % reflection_name, - srcs=srcs, - outs=reflection_outs, - output_to_bindir=output_to_bindir, - tools=includes + [flatc_path,], - cmd=reflection_genrule_cmd, - message="Generating flatbuffer reflection binary for %s:" % (name),) - native.Fileset( - name=reflection_name, - out="%s_out" % reflection_name, - entries=[ - native.FilesetEntry(files=reflection_outs), - ], - visibility=reflection_visiblity + name = name, + srcs = srcs, + outs = outs, + output_to_bindir = output_to_bindir, + tools = includes + [flatc_path], + cmd = genrule_cmd, + message = "Generating flatbuffer files for %s:" % (name), ) + if reflection_name: + reflection_genrule_cmd = " ".join([ + "for f in $(SRCS); do", + "$(location %s)" % (flatc_path), + "-b --schema", + " ".join(flatc_args), + " ".join(include_paths_cmd), + language_flag, + output_directory, + "$$f;", + "done", + ]) + reflection_outs = [ + (out_prefix + "%s.bfbs") % (s.replace(".fbs", "").split("/")[-1]) + for s in srcs + ] + native.genrule( + name = "%s_srcs" % reflection_name, + srcs = srcs, + outs = reflection_outs, + output_to_bindir = output_to_bindir, + tools = includes + [flatc_path], + cmd = reflection_genrule_cmd, + message = "Generating flatbuffer reflection binary for %s:" % (name), + ) + # TODO(b/114456773): Make bazel rules proper and supported by flatbuffer + # Have to comment this since FilesetEntry is not supported in bazel + # skylark. + # native.Fileset( + # name = reflection_name, + # out = "%s_out" % reflection_name, + # entries = [ + # native.FilesetEntry(files = reflection_outs), + # ], + # visibility = reflection_visiblity, + # ) + +def flatbuffer_cc_library( + name, + srcs, + srcs_filegroup_name = "", + out_prefix = "", + includes = [], + include_paths = [], + flatc_args = DEFAULT_FLATC_ARGS, + visibility = None, + srcs_filegroup_visibility = None, + gen_reflections = False): + '''A cc_library with the generated reader/writers for the given flatbuffer definitions. + + Outs: + filegroup([name]_srcs): all generated .h files. + filegroup(srcs_filegroup_name if specified, or [name]_includes if not): + Other flatbuffer_cc_library's can pass this in for their `includes` + parameter, if they depend on the schemas in this library. + Fileset([name]_reflection): (Optional) all generated reflection binaries. + cc_library([name]): library with sources and flatbuffers deps. + + Remarks: + ** Because the genrule used to call flatc does not have any trivial way of + computing the output list of files transitively generated by includes and + --gen-includes (the default) being defined for flatc, the --gen-includes + flag will not work as expected. The way around this is to add a dependency + to the flatbuffer_cc_library defined alongside the flatc included Fileset. + For example you might define: + + flatbuffer_cc_library( + name = "my_fbs", + srcs = [ "schemas/foo.fbs" ], + includes = [ "//third_party/bazz:bazz_fbs_includes" ], + ) + In which foo.fbs includes a few files from the Fileset defined at + //third_party/bazz:bazz_fbs_includes. When compiling the library that + includes foo_generated.h, and therefore has my_fbs as a dependency, it + will fail to find any of the bazz *_generated.h files unless you also + add bazz's flatbuffer_cc_library to your own dependency list, e.g.: -def flatbuffer_cc_library(name, srcs, srcs_filegroup_name="", - out_prefix="", includes=[], include_paths=[], - flatc_args=DEFAULT_FLATC_ARGS, - visibility=None, srcs_filegroup_visibility=None, - gen_reflections=False): - '''A cc_library with the generated reader/writers for the given flatbuffer definitions. - - Args: - name: Rule name. - srcs: Source .fbs files. Sent in order to the compiler. - srcs_filegroup_name: Name of the output filegroup that holds srcs. Pass this - filegroup into the `includes` parameter of any other - flatbuffer_cc_library that depends on this one's schemas. - out_prefix: Prepend this path to the front of all generated files. Usually - is a directory name. - includes: Optional, list of filegroups of schemas that the srcs depend on. - ** SEE REMARKS BELOW ** - include_paths: Optional, list of paths the includes files can be found in. - flatc_args: Optional list of additional arguments to pass to flatc - (e.g. --gen-mutable). - visibility: The visibility of the generated cc_library. By default, use the - default visibility of the project. - srcs_filegroup_visibility: The visibility of the generated srcs filegroup. - By default, use the value of the visibility parameter above. - gen_reflections: Optional, if true this will generate the flatbuffer - reflection binaries for the schemas. - Outs: - filegroup([name]_srcs): all generated .h files. - filegroup(srcs_filegroup_name if specified, or [name]_includes if not): - Other flatbuffer_cc_library's can pass this in for their `includes` - parameter, if they depend on the schemas in this library. - Fileset([name]_reflection): (Optional) all generated reflection binaries. - cc_library([name]): library with sources and flatbuffers deps. - - Remarks: - ** Because the genrule used to call flatc does not have any trivial way of - computing the output list of files transitively generated by includes and - --gen-includes (the default) being defined for flatc, the --gen-includes - flag will not work as expected. The way around this is to add a dependency - to the flatbuffer_cc_library defined alongside the flatc included Fileset. - For example you might define: - - flatbuffer_cc_library( - name = "my_fbs", - srcs = [ "schemas/foo.fbs" ], - includes = [ "//third_party/bazz:bazz_fbs_includes" ], - ) - - In which foo.fbs includes a few files from the Fileset defined at - //third_party/bazz:bazz_fbs_includes. When compiling the library that - includes foo_generated.h, and therefore has my_fbs as a dependency, it - will fail to find any of the bazz *_generated.h files unless you also - add bazz's flatbuffer_cc_library to your own dependency list, e.g.: - - cc_library( - name = "my_lib", - deps = [ - ":my_fbs", - "//third_party/bazz:bazz_fbs" - ], - ) - - Happy dependent Flatbuffering! - ''' - output_headers = [ - (out_prefix + "%s_generated.h") % (s.replace(".fbs", "").split("/")[-1]) for s in srcs - ] - reflection_name = "%s_reflection" % name if gen_reflections else "" - - flatbuffer_library_public(name="%s_srcs" % (name), - srcs=srcs, - outs=output_headers, - language_flag="-c", - out_prefix=out_prefix, - includes=includes, - include_paths=include_paths, - flatc_args=flatc_args, - reflection_name=reflection_name, - reflection_visiblity=visibility,) - native.cc_library(name=name, - hdrs=output_headers, - srcs=output_headers, - features=[ - "-parse_headers", - ], - deps=[ - "@flatbuffers//:runtime_cc", - ], - includes=["."], - linkstatic=1, - visibility=visibility) - - # A filegroup for the `srcs`. That is, all the schema files for this - # Flatbuffer set. - native.filegroup( - name = srcs_filegroup_name if srcs_filegroup_name else "%s_includes" % (name), - srcs = srcs, - visibility=srcs_filegroup_visibility if srcs_filegroup_visibility != None else visibility) + cc_library( + name = "my_lib", + deps = [ + ":my_fbs", + "//third_party/bazz:bazz_fbs" + ], + ) + + Happy dependent Flatbuffering! + + Args: + name: Rule name. + srcs: Source .fbs files. Sent in order to the compiler. + srcs_filegroup_name: Name of the output filegroup that holds srcs. Pass this + filegroup into the `includes` parameter of any other + flatbuffer_cc_library that depends on this one's schemas. + out_prefix: Prepend this path to the front of all generated files. Usually + is a directory name. + includes: Optional, list of filegroups of schemas that the srcs depend on. + ** SEE REMARKS BELOW ** + include_paths: Optional, list of paths the includes files can be found in. + flatc_args: Optional list of additional arguments to pass to flatc + (e.g. --gen-mutable). + visibility: The visibility of the generated cc_library. By default, use the + default visibility of the project. + srcs_filegroup_visibility: The visibility of the generated srcs filegroup. + By default, use the value of the visibility parameter above. + gen_reflections: Optional, if true this will generate the flatbuffer + reflection binaries for the schemas. + ''' + output_headers = [ + (out_prefix + "%s_generated.h") % (s.replace(".fbs", "").split("/")[-1]) + for s in srcs + ] + reflection_name = "%s_reflection" % name if gen_reflections else "" + + flatbuffer_library_public( + name = "%s_srcs" % (name), + srcs = srcs, + outs = output_headers, + language_flag = "-c", + out_prefix = out_prefix, + includes = includes, + include_paths = include_paths, + flatc_args = flatc_args, + reflection_name = reflection_name, + reflection_visiblity = visibility, + ) + native.cc_library( + name = name, + hdrs = output_headers, + srcs = output_headers, + features = [ + "-parse_headers", + ], + deps = [ + "@flatbuffers//:runtime_cc", + ], + includes = ["."], + linkstatic = 1, + visibility = visibility, + ) + + # A filegroup for the `srcs`. That is, all the schema files for this + # Flatbuffer set. + native.filegroup( + name = srcs_filegroup_name if srcs_filegroup_name else "%s_includes" % (name), + srcs = srcs, + visibility = srcs_filegroup_visibility if srcs_filegroup_visibility != None else visibility, + ) diff --git a/third_party/flatbuffers/workspace.bzl b/third_party/flatbuffers/workspace.bzl new file mode 100644 index 0000000000..3aeef96a72 --- /dev/null +++ b/third_party/flatbuffers/workspace.bzl @@ -0,0 +1,19 @@ +"""Loads the Flatbuffers library, used by TF Lite.""" + +load("//third_party:repo.bzl", "third_party_http_archive") + +def repo(): + third_party_http_archive( + name = "flatbuffers", + strip_prefix = "flatbuffers-1.9.0", + sha256 = "5ca5491e4260cacae30f1a5786d109230db3f3a6e5a0eb45d0d0608293d247e3", + urls = [ + "https://mirror.bazel.build/github.com/google/flatbuffers/archive/v1.9.0.tar.gz", + "https://github.com/google/flatbuffers/archive/v1.9.0.tar.gz", + ], + build_file = "//third_party/flatbuffers:BUILD.bazel", + system_build_file = "//third_party/flatbuffers:BUILD.system", + link_files = { + "//third_party/flatbuffers:build_defs.bzl": "build_defs.bzl", + }, + ) diff --git a/third_party/gif.BUILD b/third_party/gif.BUILD index 78fbd6c0e0..cbe730fe10 100644 --- a/third_party/gif.BUILD +++ b/third_party/gif.BUILD @@ -21,7 +21,6 @@ cc_library( ], hdrs = ["lib/gif_lib.h"], defines = select({ - #"@org_tensorflow//tensorflow:android": [ ":android": [ "S_IREAD=S_IRUSR", "S_IWRITE=S_IWUSR", @@ -33,7 +32,6 @@ cc_library( visibility = ["//visibility:public"], deps = select({ ":windows": [":windows_polyfill"], - ":windows_msvc": [":windows_polyfill"], "//conditions:default": [], }), ) @@ -51,13 +49,6 @@ genrule( ) config_setting( - name = "windows_msvc", - values = { - "cpu": "x64_windows_msvc", - }, -) - -config_setting( name = "windows", values = { "cpu": "x64_windows", diff --git a/third_party/googleapis.BUILD b/third_party/googleapis.BUILD new file mode 100644 index 0000000000..95e999af18 --- /dev/null +++ b/third_party/googleapis.BUILD @@ -0,0 +1,45 @@ +# Copyright 2018 Google LLC +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +package(default_visibility = ["//visibility:public"]) +licenses(["notice"]) # Apache 2.0 +exports_files(["LICENSE"]) + +load("@protobuf_archive//:protobuf.bzl", "cc_proto_library") + +cc_proto_library( + name = "bigtable_protos", + srcs = [ + "google/bigtable/admin/v2/bigtable_instance_admin.proto", + "google/bigtable/admin/v2/bigtable_table_admin.proto", + "google/bigtable/admin/v2/common.proto", + "google/bigtable/admin/v2/instance.proto", + "google/bigtable/admin/v2/table.proto", + "google/bigtable/v2/bigtable.proto", + "google/bigtable/v2/data.proto", + "google/iam/v1/iam_policy.proto", + "google/iam/v1/policy.proto", + "google/longrunning/operations.proto", + "google/rpc/status.proto", + "google/rpc/error_details.proto", + "google/api/annotations.proto", + "google/api/auth.proto", + "google/api/http.proto", + ], + include = ".", + protoc = "@protobuf_archive//:protoc", + default_runtime = "@protobuf_archive//:protobuf", + deps = ["@protobuf_archive//:cc_wkt_protos"], + use_grpc_plugin = True, +) diff --git a/third_party/gpus/crosstool/BUILD.tpl b/third_party/gpus/crosstool/BUILD.tpl index 98cb326572..f638756d23 100644 --- a/third_party/gpus/crosstool/BUILD.tpl +++ b/third_party/gpus/crosstool/BUILD.tpl @@ -7,6 +7,7 @@ cc_toolchain_suite( toolchains = { "local|compiler": ":cc-compiler-local", "darwin|compiler": ":cc-compiler-darwin", + "x64_windows|msvc-cl": ":cc-compiler-windows", }, ) @@ -42,6 +43,20 @@ cc_toolchain( supports_param_files = 0, ) +cc_toolchain( + name = "cc-compiler-windows", + all_files = "%{win_linker_files}", + compiler_files = ":empty", + cpu = "x64_windows", + dwp_files = ":empty", + dynamic_runtime_libs = [":empty"], + linker_files = "%{win_linker_files}", + objcopy_files = ":empty", + static_runtime_libs = [":empty"], + strip_files = ":empty", + supports_param_files = 1, +) + filegroup( name = "empty", srcs = [], @@ -51,3 +66,8 @@ filegroup( name = "crosstool_wrapper_driver_is_not_gcc", srcs = ["clang/bin/crosstool_wrapper_driver_is_not_gcc"], ) + +filegroup( + name = "windows_msvc_wrapper_files", + srcs = glob(["windows/msvc_*"]), +) diff --git a/third_party/gpus/crosstool/CROSSTOOL.tpl b/third_party/gpus/crosstool/CROSSTOOL.tpl new file mode 100644 index 0000000000..3189cf8e31 --- /dev/null +++ b/third_party/gpus/crosstool/CROSSTOOL.tpl @@ -0,0 +1,1408 @@ +major_version: "local" +minor_version: "" +default_target_cpu: "same_as_host" + +default_toolchain { + cpu: "k8" + toolchain_identifier: "local_linux" +} +default_toolchain { + cpu: "piii" + toolchain_identifier: "local_linux" +} +default_toolchain { + cpu: "arm" + toolchain_identifier: "local_linux" +} +default_toolchain { + cpu: "darwin" + toolchain_identifier: "local_darwin" +} +default_toolchain { + cpu: "ppc" + toolchain_identifier: "local_linux" +} +default_toolchain { + cpu: "x64_windows" + toolchain_identifier: "local_windows" +} + +toolchain { + abi_version: "local" + abi_libc_version: "local" + compiler: "compiler" + host_system_name: "local" + needsPic: true + target_libc: "local" + target_cpu: "local" + target_system_name: "local" + toolchain_identifier: "local_linux" + + feature { + name: "c++11" + flag_set { + action: "c++-compile" + flag_group { + flag: "-std=c++11" + } + } + } + + feature { + name: "stdlib" + flag_set { + action: "c++-link-executable" + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "-lstdc++" + } + } + } + + feature { + name: "determinism" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + # Make C++ compilation deterministic. Use linkstamping instead of these + # compiler symbols. + flag: "-Wno-builtin-macro-redefined" + flag: "-D__DATE__=\"redacted\"" + flag: "-D__TIMESTAMP__=\"redacted\"" + flag: "-D__TIME__=\"redacted\"" + } + } + } + + feature { + name: "alwayslink" + flag_set { + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + action: "c++-link-executable" + flag_group { + flag: "-Wl,-no-as-needed" + } + } + } + + # This feature will be enabled for builds that support pic by bazel. + feature { + name: "pic" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + expand_if_all_available: "pic" + flag: "-fPIC" + } + flag_group { + expand_if_none_available: "pic" + flag: "-fPIE" + } + } + } + + # Security hardening on by default. + feature { + name: "hardening" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + # Conservative choice; -D_FORTIFY_SOURCE=2 may be unsafe in some cases. + # We need to undef it before redefining it as some distributions now + # have it enabled by default. + flag: "-U_FORTIFY_SOURCE" + flag: "-D_FORTIFY_SOURCE=1" + flag: "-fstack-protector" + } + } + flag_set { + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "-Wl,-z,relro,-z,now" + } + } + flag_set { + action: "c++-link-executable" + flag_group { + flag: "-pie" + flag: "-Wl,-z,relro,-z,now" + } + } + } + + feature { + name: "warnings" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + # All warnings are enabled. Maybe enable -Werror as well? + flag: "-Wall" + %{host_compiler_warnings} + } + } + } + + # Keep stack frames for debugging, even in opt mode. + feature { + name: "frame-pointer" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + flag: "-fno-omit-frame-pointer" + } + } + } + + feature { + name: "build-id" + flag_set { + action: "c++-link-executable" + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + flag_group { + # Stamp the binary with a unique identifier. + flag: "-Wl,--build-id=md5" + flag: "-Wl,--hash-style=gnu" + } + } + } + + feature { + name: "no-canonical-prefixes" + flag_set { + action: "c-compile" + action: "c++-compile" + action: "c++-link-executable" + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag:"-no-canonical-prefixes" + } + } + } + + feature { + name: "disable-assertions" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + flag: "-DNDEBUG" + } + } + } + + feature { + name: "linker-bin-path" + + flag_set { + action: "c++-link-executable" + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + flag_group { + %{linker_bin_path_flag} + } + } + } + + feature { + name: "common" + implies: "stdlib" + implies: "c++11" + implies: "determinism" + implies: "alwayslink" + implies: "hardening" + implies: "warnings" + implies: "frame-pointer" + implies: "build-id" + implies: "no-canonical-prefixes" + implies: "linker-bin-path" + } + + feature { + name: "opt" + implies: "common" + implies: "disable-assertions" + + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + # No debug symbols. + # Maybe we should enable https://gcc.gnu.org/wiki/DebugFission for opt + # or even generally? However, that can't happen here, as it requires + # special handling in Bazel. + flag: "-g0" + + # Conservative choice for -O + # -O3 can increase binary size and even slow down the resulting binaries. + # Profile first and / or use FDO if you need better performance than this. + flag: "-O2" + + # Removal of unused code and data at link time (can this increase binary size in some cases?). + flag: "-ffunction-sections" + flag: "-fdata-sections" + } + } + flag_set { + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + action: "c++-link-executable" + flag_group { + flag: "-Wl,--gc-sections" + } + } + } + + feature { + name: "fastbuild" + implies: "common" + } + + feature { + name: "dbg" + implies: "common" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + flag: "-g" + } + } + } + + # Set clang as a C/C++ compiler. + tool_path { name: "gcc" path: "%{host_compiler_path}" } + + # Use the default system toolchain for everything else. + tool_path { name: "ar" path: "/usr/bin/ar" } + tool_path { name: "compat-ld" path: "/usr/bin/ld" } + tool_path { name: "cpp" path: "/usr/bin/cpp" } + tool_path { name: "dwp" path: "/usr/bin/dwp" } + tool_path { name: "gcov" path: "/usr/bin/gcov" } + tool_path { name: "ld" path: "/usr/bin/ld" } + tool_path { name: "nm" path: "/usr/bin/nm" } + tool_path { name: "objcopy" path: "/usr/bin/objcopy" } + tool_path { name: "objdump" path: "/usr/bin/objdump" } + tool_path { name: "strip" path: "/usr/bin/strip" } + + # Enabled dynamic linking. + linking_mode_flags { mode: DYNAMIC } + +%{host_compiler_includes} +} + +toolchain { + abi_version: "local" + abi_libc_version: "local" + compiler: "compiler" + host_system_name: "local" + needsPic: true + target_libc: "macosx" + target_cpu: "darwin" + target_system_name: "local" + toolchain_identifier: "local_darwin" + feature { + name: "c++11" + flag_set { + action: "c++-compile" + flag_group { + flag: "-std=c++11" + } + } + } + + feature { + name: "stdlib" + flag_set { + action: "c++-link-executable" + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "-lc++" + } + } + } + + feature { + name: "determinism" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + # Make C++ compilation deterministic. Use linkstamping instead of these + # compiler symbols. + flag: "-Wno-builtin-macro-redefined" + flag: "-D__DATE__=\"redacted\"" + flag: "-D__TIMESTAMP__=\"redacted\"" + flag: "-D__TIME__=\"redacted\"" + } + } + } + + # This feature will be enabled for builds that support pic by bazel. + feature { + name: "pic" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + expand_if_all_available: "pic" + flag: "-fPIC" + } + flag_group { + expand_if_none_available: "pic" + flag: "-fPIE" + } + } + } + + # Security hardening on by default. + feature { + name: "hardening" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + # Conservative choice; -D_FORTIFY_SOURCE=2 may be unsafe in some cases. + # We need to undef it before redefining it as some distributions now + # have it enabled by default. + flag: "-U_FORTIFY_SOURCE" + flag: "-D_FORTIFY_SOURCE=1" + flag: "-fstack-protector" + } + } + flag_set { + action: "c++-link-executable" + flag_group { + flag: "-pie" + } + } + } + + feature { + name: "warnings" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + # All warnings are enabled. Maybe enable -Werror as well? + flag: "-Wall" + %{host_compiler_warnings} + } + } + } + + # Keep stack frames for debugging, even in opt mode. + feature { + name: "frame-pointer" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + flag: "-fno-omit-frame-pointer" + } + } + } + + feature { + name: "no-canonical-prefixes" + flag_set { + action: "c-compile" + action: "c++-compile" + action: "c++-link-executable" + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag:"-no-canonical-prefixes" + } + } + } + + feature { + name: "disable-assertions" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + flag: "-DNDEBUG" + } + } + } + + feature { + name: "linker-bin-path" + + flag_set { + action: "c++-link-executable" + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + flag_group { + %{linker_bin_path_flag} + } + } + } + + feature { + name: "undefined-dynamic" + flag_set { + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + action: "c++-link-executable" + flag_group { + flag: "-undefined" + flag: "dynamic_lookup" + } + } + } + + feature { + name: "common" + implies: "stdlib" + implies: "c++11" + implies: "determinism" + implies: "hardening" + implies: "warnings" + implies: "frame-pointer" + implies: "no-canonical-prefixes" + implies: "linker-bin-path" + implies: "undefined-dynamic" + } + + feature { + name: "opt" + implies: "common" + implies: "disable-assertions" + + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + # No debug symbols. + # Maybe we should enable https://gcc.gnu.org/wiki/DebugFission for opt + # or even generally? However, that can't happen here, as it requires + # special handling in Bazel. + flag: "-g0" + + # Conservative choice for -O + # -O3 can increase binary size and even slow down the resulting binaries. + # Profile first and / or use FDO if you need better performance than this. + flag: "-O2" + + # Removal of unused code and data at link time (can this increase binary size in some cases?). + flag: "-ffunction-sections" + flag: "-fdata-sections" + } + } + } + + feature { + name: "fastbuild" + implies: "common" + } + + feature { + name: "dbg" + implies: "common" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + flag: "-g" + } + } + } + + # Set clang as a C/C++ compiler. + tool_path { name: "gcc" path: "%{host_compiler_path}" } + + # Use the default system toolchain for everything else. + tool_path { name: "ar" path: "/usr/bin/libtool" } + tool_path { name: "compat-ld" path: "/usr/bin/ld" } + tool_path { name: "cpp" path: "/usr/bin/cpp" } + tool_path { name: "dwp" path: "/usr/bin/dwp" } + tool_path { name: "gcov" path: "/usr/bin/gcov" } + tool_path { name: "ld" path: "/usr/bin/ld" } + tool_path { name: "nm" path: "/usr/bin/nm" } + tool_path { name: "objcopy" path: "/usr/bin/objcopy" } + tool_path { name: "objdump" path: "/usr/bin/objdump" } + tool_path { name: "strip" path: "/usr/bin/strip" } + + # Enabled dynamic linking. + linking_mode_flags { mode: DYNAMIC } + +%{host_compiler_includes} +} + +toolchain { + toolchain_identifier: "local_windows" + host_system_name: "local" + target_system_name: "local" + + abi_version: "local" + abi_libc_version: "local" + target_cpu: "x64_windows" + compiler: "msvc-cl" + target_libc: "msvcrt" + +%{cxx_builtin_include_directory} + + tool_path { + name: "ar" + path: "%{msvc_lib_path}" + } + tool_path { + name: "ml" + path: "%{msvc_ml_path}" + } + tool_path { + name: "cpp" + path: "%{msvc_cl_path}" + } + tool_path { + name: "gcc" + path: "%{msvc_cl_path}" + } + tool_path { + name: "gcov" + path: "wrapper/bin/msvc_nop.bat" + } + tool_path { + name: "ld" + path: "%{msvc_link_path}" + } + tool_path { + name: "nm" + path: "wrapper/bin/msvc_nop.bat" + } + tool_path { + name: "objcopy" + path: "wrapper/bin/msvc_nop.bat" + } + tool_path { + name: "objdump" + path: "wrapper/bin/msvc_nop.bat" + } + tool_path { + name: "strip" + path: "wrapper/bin/msvc_nop.bat" + } + supports_interface_shared_objects: true + + # TODO(pcloudy): Review those flags below, they should be defined by cl.exe + compiler_flag: "/DCOMPILER_MSVC" + + # Don't define min/max macros in windows.h. + compiler_flag: "/DNOMINMAX" + + # Platform defines. + compiler_flag: "/D_WIN32_WINNT=0x0600" + # Turn off warning messages. + compiler_flag: "/D_CRT_SECURE_NO_DEPRECATE" + compiler_flag: "/D_CRT_SECURE_NO_WARNINGS" + compiler_flag: "/D_SILENCE_STDEXT_HASH_DEPRECATION_WARNINGS" + + # Useful options to have on for compilation. + # Increase the capacity of object files to 2^32 sections. + compiler_flag: "/bigobj" + # Allocate 500MB for precomputed headers. + compiler_flag: "/Zm500" + # Use unsigned char by default. + compiler_flag: "/J" + # Use function level linking. + compiler_flag: "/Gy" + # Use string pooling. + compiler_flag: "/GF" + # Catch C++ exceptions only and tell the compiler to assume that functions declared + # as extern "C" never throw a C++ exception. + compiler_flag: "/EHsc" + + # Globally disabled warnings. + # Don't warn about elements of array being be default initialized. + compiler_flag: "/wd4351" + # Don't warn about no matching delete found. + compiler_flag: "/wd4291" + # Don't warn about diamond inheritance patterns. + compiler_flag: "/wd4250" + # Don't warn about insecure functions (e.g. non _s functions). + compiler_flag: "/wd4996" + + linker_flag: "/MACHINE:X64" + + feature { + name: "no_legacy_features" + } + + # Suppress startup banner. + feature { + name: "nologo" + flag_set { + action: "c-compile" + action: "c++-compile" + action: "c++-module-compile" + action: "c++-module-codegen" + action: "c++-header-parsing" + action: "assemble" + action: "preprocess-assemble" + action: "c++-link-executable" + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + action: "c++-link-static-library" + flag_group { + flag: "/nologo" + } + } + } + + feature { + name: 'has_configured_linker_path' + } + + # This feature indicates strip is not supported, building stripped binary will just result a copy of orignial binary + feature { + name: 'no_stripping' + } + + # This feature indicates this is a toolchain targeting Windows. + feature { + name: 'targets_windows' + implies: 'copy_dynamic_libraries_to_binary' + enabled: true + } + + feature { + name: 'copy_dynamic_libraries_to_binary' + } + + action_config { + config_name: 'assemble' + action_name: 'assemble' + tool { + tool_path: '%{msvc_ml_path}' + } + implies: 'compiler_input_flags' + implies: 'compiler_output_flags' + implies: 'nologo' + implies: 'msvc_env' + implies: 'sysroot' + } + + action_config { + config_name: 'preprocess-assemble' + action_name: 'preprocess-assemble' + tool { + tool_path: '%{msvc_ml_path}' + } + implies: 'compiler_input_flags' + implies: 'compiler_output_flags' + implies: 'nologo' + implies: 'msvc_env' + implies: 'sysroot' + } + + action_config { + config_name: 'c-compile' + action_name: 'c-compile' + tool { + tool_path: '%{msvc_cl_path}' + } + implies: 'compiler_input_flags' + implies: 'compiler_output_flags' + implies: 'legacy_compile_flags' + implies: 'nologo' + implies: 'msvc_env' + implies: 'parse_showincludes' + implies: 'user_compile_flags' + implies: 'sysroot' + implies: 'unfiltered_compile_flags' + } + + action_config { + config_name: 'c++-compile' + action_name: 'c++-compile' + tool { + tool_path: '%{msvc_cl_path}' + } + implies: 'compiler_input_flags' + implies: 'compiler_output_flags' + implies: 'legacy_compile_flags' + implies: 'nologo' + implies: 'msvc_env' + implies: 'parse_showincludes' + implies: 'user_compile_flags' + implies: 'sysroot' + implies: 'unfiltered_compile_flags' + } + + action_config { + config_name: 'c++-link-executable' + action_name: 'c++-link-executable' + tool { + tool_path: '%{msvc_link_path}' + } + implies: 'nologo' + implies: 'linkstamps' + implies: 'output_execpath_flags' + implies: 'input_param_flags' + implies: 'user_link_flags' + implies: 'legacy_link_flags' + implies: 'linker_subsystem_flag' + implies: 'linker_param_file' + implies: 'msvc_env' + implies: 'no_stripping' + } + + action_config { + config_name: 'c++-link-dynamic-library' + action_name: 'c++-link-dynamic-library' + tool { + tool_path: '%{msvc_link_path}' + } + implies: 'nologo' + implies: 'shared_flag' + implies: 'linkstamps' + implies: 'output_execpath_flags' + implies: 'input_param_flags' + implies: 'user_link_flags' + implies: 'legacy_link_flags' + implies: 'linker_subsystem_flag' + implies: 'linker_param_file' + implies: 'msvc_env' + implies: 'no_stripping' + implies: 'has_configured_linker_path' + implies: 'def_file' + } + + action_config { + config_name: 'c++-link-nodeps-dynamic-library' + action_name: 'c++-link-nodeps-dynamic-library' + tool { + tool_path: '%{msvc_link_path}' + } + implies: 'nologo' + implies: 'shared_flag' + implies: 'linkstamps' + implies: 'output_execpath_flags' + implies: 'input_param_flags' + implies: 'user_link_flags' + implies: 'legacy_link_flags' + implies: 'linker_subsystem_flag' + implies: 'linker_param_file' + implies: 'msvc_env' + implies: 'no_stripping' + implies: 'has_configured_linker_path' + implies: 'def_file' + } + + action_config { + config_name: 'c++-link-static-library' + action_name: 'c++-link-static-library' + tool { + tool_path: '%{msvc_lib_path}' + } + implies: 'nologo' + implies: 'archiver_flags' + implies: 'input_param_flags' + implies: 'linker_param_file' + implies: 'msvc_env' + } + + # TODO(b/65151735): Remove legacy_compile_flags feature when legacy fields are + # not used in this crosstool + feature { + name: 'legacy_compile_flags' + flag_set { + expand_if_all_available: 'legacy_compile_flags' + action: 'preprocess-assemble' + action: 'c-compile' + action: 'c++-compile' + action: 'c++-header-parsing' + action: 'c++-module-compile' + action: 'c++-module-codegen' + flag_group { + iterate_over: 'legacy_compile_flags' + flag: '%{legacy_compile_flags}' + } + } + } + + feature { + name: "msvc_env" + env_set { + action: "c-compile" + action: "c++-compile" + action: "c++-module-compile" + action: "c++-module-codegen" + action: "c++-header-parsing" + action: "assemble" + action: "preprocess-assemble" + action: "c++-link-executable" + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + action: "c++-link-static-library" + env_entry { + key: "PATH" + value: "%{msvc_env_path}" + } + env_entry { + key: "INCLUDE" + value: "%{msvc_env_include}" + } + env_entry { + key: "LIB" + value: "%{msvc_env_lib}" + } + env_entry { + key: "TMP" + value: "%{msvc_env_tmp}" + } + env_entry { + key: "TEMP" + value: "%{msvc_env_tmp}" + } + } + } + + feature { + name: 'include_paths' + flag_set { + action: "assemble" + action: 'preprocess-assemble' + action: 'c-compile' + action: 'c++-compile' + action: 'c++-header-parsing' + action: 'c++-module-compile' + flag_group { + iterate_over: 'quote_include_paths' + flag: '/I%{quote_include_paths}' + } + flag_group { + iterate_over: 'include_paths' + flag: '/I%{include_paths}' + } + flag_group { + iterate_over: 'system_include_paths' + flag: '/I%{system_include_paths}' + } + } + } + + feature { + name: "preprocessor_defines" + flag_set { + action: "assemble" + action: "preprocess-assemble" + action: "c-compile" + action: "c++-compile" + action: "c++-header-parsing" + action: "c++-module-compile" + flag_group { + flag: "/D%{preprocessor_defines}" + iterate_over: "preprocessor_defines" + } + } + } + + # Tell Bazel to parse the output of /showIncludes + feature { + name: 'parse_showincludes' + flag_set { + action: 'preprocess-assemble' + action: 'c-compile' + action: 'c++-compile' + action: 'c++-module-compile' + action: 'c++-header-parsing' + flag_group { + flag: "/showIncludes" + } + } + } + + + feature { + name: 'generate_pdb_file' + requires: { + feature: 'dbg' + } + requires: { + feature: 'fastbuild' + } + } + + feature { + name: 'shared_flag' + flag_set { + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: '/DLL' + } + } + } + + feature { + name: 'linkstamps' + flag_set { + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + expand_if_all_available: 'linkstamp_paths' + flag_group { + iterate_over: 'linkstamp_paths' + flag: '%{linkstamp_paths}' + } + } + } + + feature { + name: 'output_execpath_flags' + flag_set { + expand_if_all_available: 'output_execpath' + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: '/OUT:%{output_execpath}' + } + } + } + + feature { + name: 'archiver_flags' + flag_set { + expand_if_all_available: 'output_execpath' + action: 'c++-link-static-library' + flag_group { + flag: '/OUT:%{output_execpath}' + } + } + } + + feature { + name: 'input_param_flags' + flag_set { + expand_if_all_available: 'interface_library_output_path' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "/IMPLIB:%{interface_library_output_path}" + } + } + flag_set { + expand_if_all_available: 'libopts' + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + iterate_over: 'libopts' + flag: '%{libopts}' + } + } + flag_set { + expand_if_all_available: 'libraries_to_link' + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + action: 'c++-link-static-library' + flag_group { + iterate_over: 'libraries_to_link' + flag_group { + expand_if_equal: { + variable: 'libraries_to_link.type' + value: 'object_file_group' + } + iterate_over: 'libraries_to_link.object_files' + flag_group { + flag: '%{libraries_to_link.object_files}' + } + } + flag_group { + expand_if_equal: { + variable: 'libraries_to_link.type' + value: 'object_file' + } + flag_group { + flag: '%{libraries_to_link.name}' + } + } + flag_group { + expand_if_equal: { + variable: 'libraries_to_link.type' + value: 'interface_library' + } + flag_group { + flag: '%{libraries_to_link.name}' + } + } + flag_group { + expand_if_equal: { + variable: 'libraries_to_link.type' + value: 'static_library' + } + flag_group { + expand_if_false: 'libraries_to_link.is_whole_archive' + flag: '%{libraries_to_link.name}' + } + flag_group { + expand_if_true: 'libraries_to_link.is_whole_archive' + flag: '/WHOLEARCHIVE:%{libraries_to_link.name}' + } + } + } + } + } + + # Since this feature is declared earlier in the CROSSTOOL than + # "user_link_flags", this feature will be applied prior to it anwyhere they + # are both implied. And since "user_link_flags" contains the linkopts from + # the build rule, this allows the user to override the /SUBSYSTEM in the BUILD + # file. + feature { + name: 'linker_subsystem_flag' + flag_set { + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: '/SUBSYSTEM:CONSOLE' + } + } + } + + # The "user_link_flags" contains user-defined linkopts (from build rules) + # so it should be defined after features that declare user-overridable flags. + # For example the "linker_subsystem_flag" defines a default "/SUBSYSTEM" flag + # but we want to let the user override it, therefore "link_flag_subsystem" is + # defined earlier in the CROSSTOOL file than "user_link_flags". + feature { + name: 'user_link_flags' + flag_set { + expand_if_all_available: 'user_link_flags' + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + iterate_over: 'user_link_flags' + flag: '%{user_link_flags}' + } + } + } + feature { + name: 'legacy_link_flags' + flag_set { + expand_if_all_available: 'legacy_link_flags' + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + iterate_over: 'legacy_link_flags' + flag: '%{legacy_link_flags}' + } + } + } + + feature { + name: 'linker_param_file' + flag_set { + expand_if_all_available: 'linker_param_file' + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + action: 'c++-link-static-library' + flag_group { + flag: '@%{linker_param_file}' + } + } + } + + feature { + name: 'static_link_msvcrt' + } + + feature { + name: 'static_link_msvcrt_no_debug' + flag_set { + action: 'c-compile' + action: 'c++-compile' + flag_group { + flag: "/MT" + } + } + flag_set { + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "/DEFAULTLIB:libcmt.lib" + } + } + requires: { feature: 'fastbuild'} + requires: { feature: 'opt'} + } + + feature { + name: 'dynamic_link_msvcrt_no_debug' + flag_set { + action: 'c-compile' + action: 'c++-compile' + flag_group { + flag: "/MD" + } + } + flag_set { + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "/DEFAULTLIB:msvcrt.lib" + } + } + requires: { feature: 'fastbuild'} + requires: { feature: 'opt'} + } + + feature { + name: 'static_link_msvcrt_debug' + flag_set { + action: 'c-compile' + action: 'c++-compile' + flag_group { + flag: "/MTd" + } + } + flag_set { + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "/DEFAULTLIB:libcmtd.lib" + } + } + requires: { feature: 'dbg'} + } + + feature { + name: 'dynamic_link_msvcrt_debug' + flag_set { + action: 'c-compile' + action: 'c++-compile' + flag_group { + flag: "/MDd" + } + } + flag_set { + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "/DEFAULTLIB:msvcrtd.lib" + } + } + requires: { feature: 'dbg'} + } + + feature { + name: 'dbg' + flag_set { + action: 'c-compile' + action: 'c++-compile' + flag_group { + flag: "/Od" + flag: "/Z7" + flag: "/DDEBUG" + } + } + flag_set { + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "/DEBUG:FULL" + flag: "/INCREMENTAL:NO" + } + } + implies: 'generate_pdb_file' + } + + feature { + name: 'fastbuild' + flag_set { + action: 'c-compile' + action: 'c++-compile' + flag_group { + flag: "/Od" + flag: "/Z7" + flag: "/DDEBUG" + } + } + flag_set { + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "/DEBUG:FASTLINK" + flag: "/INCREMENTAL:NO" + } + } + implies: 'generate_pdb_file' + } + + feature { + name: 'opt' + flag_set { + action: 'c-compile' + action: 'c++-compile' + flag_group { + flag: "/O2" + flag: "/DNDEBUG" + } + } + } + + feature { + name: 'user_compile_flags' + flag_set { + expand_if_all_available: 'user_compile_flags' + action: 'preprocess-assemble' + action: 'c-compile' + action: 'c++-compile' + action: 'c++-header-parsing' + action: 'c++-module-compile' + action: 'c++-module-codegen' + flag_group { + iterate_over: 'user_compile_flags' + flag: '%{user_compile_flags}' + } + } + } + + feature { + name: 'sysroot' + flag_set { + expand_if_all_available: 'sysroot' + action: 'assemble' + action: 'preprocess-assemble' + action: 'c-compile' + action: 'c++-compile' + action: 'c++-header-parsing' + action: 'c++-module-compile' + action: 'c++-module-codegen' + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + iterate_over: 'sysroot' + flag: '--sysroot=%{sysroot}' + } + } + } + + feature { + name: 'unfiltered_compile_flags' + flag_set { + expand_if_all_available: 'unfiltered_compile_flags' + action: 'preprocess-assemble' + action: 'c-compile' + action: 'c++-compile' + action: 'c++-header-parsing' + action: 'c++-module-compile' + action: 'c++-module-codegen' + flag_group { + iterate_over: 'unfiltered_compile_flags' + flag: '%{unfiltered_compile_flags}' + } + } + } + + feature { + name: 'compiler_output_flags' + flag_set { + action: 'assemble' + flag_group { + expand_if_all_available: 'output_file' + expand_if_none_available: 'output_assembly_file' + expand_if_none_available: 'output_preprocess_file' + flag: '/Fo%{output_file}' + flag: '/Zi' + } + } + flag_set { + action: 'preprocess-assemble' + action: 'c-compile' + action: 'c++-compile' + action: 'c++-header-parsing' + action: 'c++-module-compile' + action: 'c++-module-codegen' + flag_group { + expand_if_all_available: 'output_file' + expand_if_none_available: 'output_assembly_file' + expand_if_none_available: 'output_preprocess_file' + flag: '/Fo%{output_file}' + } + flag_group { + expand_if_all_available: 'output_file' + expand_if_all_available: 'output_assembly_file' + flag: '/Fa%{output_file}' + } + flag_group { + expand_if_all_available: 'output_file' + expand_if_all_available: 'output_preprocess_file' + flag: '/P' + flag: '/Fi%{output_file}' + } + } + } + + feature { + name: 'compiler_input_flags' + flag_set { + action: 'assemble' + action: 'preprocess-assemble' + action: 'c-compile' + action: 'c++-compile' + action: 'c++-header-parsing' + action: 'c++-module-compile' + action: 'c++-module-codegen' + flag_group { + expand_if_all_available: 'source_file' + flag: '/c' + flag: '%{source_file}' + } + } + } + + feature { + name : 'def_file', + flag_set { + expand_if_all_available: 'def_file_path' + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "/DEF:%{def_file_path}" + # We can specify a different DLL name in DEF file, /ignore:4070 suppresses + # the warning message about DLL name doesn't match the default one. + # See https://msdn.microsoft.com/en-us/library/sfkk2fz7.aspx + flag: "/ignore:4070" + } + } + } + + feature { + name: 'windows_export_all_symbols' + } + + feature { + name: 'no_windows_export_all_symbols' + } + + linking_mode_flags { mode: DYNAMIC } +} diff --git a/third_party/gpus/crosstool/CROSSTOOL_clang.tpl b/third_party/gpus/crosstool/CROSSTOOL_clang.tpl deleted file mode 100644 index 2f09473ee2..0000000000 --- a/third_party/gpus/crosstool/CROSSTOOL_clang.tpl +++ /dev/null @@ -1,299 +0,0 @@ -major_version: "local" -minor_version: "" -default_target_cpu: "same_as_host" - -default_toolchain { - cpu: "k8" - toolchain_identifier: "local_linux" -} -default_toolchain { - cpu: "piii" - toolchain_identifier: "local_linux" -} -default_toolchain { - cpu: "arm" - toolchain_identifier: "local_linux" -} -default_toolchain { - cpu: "darwin" - toolchain_identifier: "local_darwin" -} -default_toolchain { - cpu: "ppc" - toolchain_identifier: "local_linux" -} - -toolchain { - abi_version: "local" - abi_libc_version: "local" - compiler: "compiler" - host_system_name: "local" - needsPic: true - target_libc: "local" - target_cpu: "local" - target_system_name: "local" - toolchain_identifier: "local_linux" - - feature { - name: "c++11" - flag_set { - action: "c++-compile" - flag_group { - flag: "-std=c++11" - } - } - } - - feature { - name: "stdlib" - flag_set { - action: "c++-link-executable" - action: "c++-link-dynamic-library" - action: "c++-link-nodeps-dynamic-library" - flag_group { - flag: "-lstdc++" - } - } - } - - feature { - name: "determinism" - flag_set { - action: "c-compile" - action: "c++-compile" - flag_group { - # Make C++ compilation deterministic. Use linkstamping instead of these - # compiler symbols. - flag: "-Wno-builtin-macro-redefined" - flag: "-D__DATE__=\"redacted\"" - flag: "-D__TIMESTAMP__=\"redacted\"" - flag: "-D__TIME__=\"redacted\"" - } - } - } - - feature { - name: "alwayslink" - flag_set { - action: "c++-link-dynamic-library" - action: "c++-link-nodeps-dynamic-library" - action: "c++-link-executable" - flag_group { - flag: "-Wl,-no-as-needed" - } - } - } - - # This feature will be enabled for builds that support pic by bazel. - feature { - name: "pic" - flag_set { - action: "c-compile" - action: "c++-compile" - flag_group { - expand_if_all_available: "pic" - flag: "-fPIC" - } - flag_group { - expand_if_none_available: "pic" - flag: "-fPIE" - } - } - } - - # Security hardening on by default. - feature { - name: "hardening" - flag_set { - action: "c-compile" - action: "c++-compile" - flag_group { - # Conservative choice; -D_FORTIFY_SOURCE=2 may be unsafe in some cases. - # We need to undef it before redefining it as some distributions now - # have it enabled by default. - flag: "-U_FORTIFY_SOURCE" - flag: "-D_FORTIFY_SOURCE=1" - flag: "-fstack-protector" - } - } - flag_set { - action: "c++-link-dynamic-library" - action: "c++-link-nodeps-dynamic-library" - flag_group { - flag: "-Wl,-z,relro,-z,now" - } - } - flag_set { - action: "c++-link-executable" - flag_group { - flag: "-pie" - flag: "-Wl,-z,relro,-z,now" - } - } - } - - feature { - name: "warnings" - flag_set { - action: "c-compile" - action: "c++-compile" - flag_group { - # All warnings are enabled. Maybe enable -Werror as well? - flag: "-Wall" - # Some parts of the codebase set -Werror and hit this warning, so - # switch it off for now. - flag: "-Wno-invalid-partial-specialization" - } - } - } - - # Keep stack frames for debugging, even in opt mode. - feature { - name: "frame-pointer" - flag_set { - action: "c-compile" - action: "c++-compile" - flag_group { - flag: "-fno-omit-frame-pointer" - } - } - } - - feature { - name: "build-id" - flag_set { - action: "c++-link-executable" - action: "c++-link-dynamic-library" - action: "c++-link-nodeps-dynamic-library" - flag_group { - # Stamp the binary with a unique identifier. - flag: "-Wl,--build-id=md5" - flag: "-Wl,--hash-style=gnu" - } - } - } - - feature { - name: "no-canonical-prefixes" - flag_set { - action: "c-compile" - action: "c++-compile" - action: "c++-link-executable" - action: "c++-link-dynamic-library" - action: "c++-link-nodeps-dynamic-library" - flag_group { - flag:"-no-canonical-prefixes" - } - } - } - - feature { - name: "disable-assertions" - flag_set { - action: "c-compile" - action: "c++-compile" - flag_group { - flag: "-DNDEBUG" - } - } - } - - feature { - name: "linker-bin-path" - - flag_set { - action: "c++-link-executable" - action: "c++-link-dynamic-library" - action: "c++-link-nodeps-dynamic-library" - flag_group { - flag: "-B/usr/bin/" - } - } - } - - feature { - name: "common" - implies: "stdlib" - implies: "c++11" - implies: "determinism" - implies: "alwayslink" - implies: "hardening" - implies: "warnings" - implies: "frame-pointer" - implies: "build-id" - implies: "no-canonical-prefixes" - implies: "linker-bin-path" - } - - feature { - name: "opt" - implies: "common" - implies: "disable-assertions" - - flag_set { - action: "c-compile" - action: "c++-compile" - flag_group { - # No debug symbols. - # Maybe we should enable https://gcc.gnu.org/wiki/DebugFission for opt - # or even generally? However, that can't happen here, as it requires - # special handling in Bazel. - flag: "-g0" - - # Conservative choice for -O - # -O3 can increase binary size and even slow down the resulting binaries. - # Profile first and / or use FDO if you need better performance than this. - flag: "-O2" - - # Removal of unused code and data at link time (can this increase binary size in some cases?). - flag: "-ffunction-sections" - flag: "-fdata-sections" - } - } - flag_set { - action: "c++-link-dynamic-library" - action: "c++-link-nodeps-dynamic-library" - action: "c++-link-executable" - flag_group { - flag: "-Wl,--gc-sections" - } - } - } - - feature { - name: "fastbuild" - implies: "common" - } - - feature { - name: "dbg" - implies: "common" - flag_set { - action: "c-compile" - action: "c++-compile" - flag_group { - flag: "-g" - } - } - } - - # Set clang as a C/C++ compiler. - tool_path { name: "gcc" path: "%{clang_path}" } - - # Use the default system toolchain for everything else. - tool_path { name: "ar" path: "/usr/bin/ar" } - tool_path { name: "compat-ld" path: "/usr/bin/ld" } - tool_path { name: "cpp" path: "/usr/bin/cpp" } - tool_path { name: "dwp" path: "/usr/bin/dwp" } - tool_path { name: "gcov" path: "/usr/bin/gcov" } - tool_path { name: "ld" path: "/usr/bin/ld" } - tool_path { name: "nm" path: "/usr/bin/nm" } - tool_path { name: "objcopy" path: "/usr/bin/objcopy" } - tool_path { name: "objdump" path: "/usr/bin/objdump" } - tool_path { name: "strip" path: "/usr/bin/strip" } - - # Enabled dynamic linking. - linking_mode_flags { mode: DYNAMIC } - -%{host_compiler_includes} -} diff --git a/third_party/gpus/crosstool/CROSSTOOL_nvcc.tpl b/third_party/gpus/crosstool/CROSSTOOL_nvcc.tpl deleted file mode 100644 index 05290d647e..0000000000 --- a/third_party/gpus/crosstool/CROSSTOOL_nvcc.tpl +++ /dev/null @@ -1,249 +0,0 @@ -major_version: "local" -minor_version: "" -default_target_cpu: "same_as_host" - -default_toolchain { - cpu: "k8" - toolchain_identifier: "local_linux" -} -default_toolchain { - cpu: "piii" - toolchain_identifier: "local_linux" -} -default_toolchain { - cpu: "arm" - toolchain_identifier: "local_linux" -} -default_toolchain { - cpu: "darwin" - toolchain_identifier: "local_darwin" -} -default_toolchain { - cpu: "ppc" - toolchain_identifier: "local_linux" -} - -toolchain { - abi_version: "local" - abi_libc_version: "local" - builtin_sysroot: "" - compiler: "compiler" - host_system_name: "local" - needsPic: true - supports_gold_linker: false - supports_incremental_linker: false - supports_fission: false - supports_interface_shared_objects: false - supports_normalizing_ar: false - supports_start_end_lib: false - supports_thin_archives: false - target_libc: "local" - target_cpu: "local" - target_system_name: "local" - toolchain_identifier: "local_linux" - - tool_path { name: "ar" path: "/usr/bin/ar" } - tool_path { name: "compat-ld" path: "/usr/bin/ld" } - tool_path { name: "cpp" path: "/usr/bin/cpp" } - tool_path { name: "dwp" path: "/usr/bin/dwp" } - # As part of the TensorFlow release, we place some cuda-related compilation - # files in @local_config_cuda//crosstool/clang/bin, and this relative - # path, combined with the rest of our Bazel configuration causes our - # compilation to use those files. - tool_path { name: "gcc" path: "clang/bin/crosstool_wrapper_driver_is_not_gcc" } - # Use "-std=c++11" for nvcc. For consistency, force both the host compiler - # and the device compiler to use "-std=c++11". - cxx_flag: "-std=c++11" - linker_flag: "-Wl,-no-as-needed" - linker_flag: "-lstdc++" - linker_flag: "-B/usr/bin/" - -%{host_compiler_includes} - tool_path { name: "gcov" path: "/usr/bin/gcov" } - - # C(++) compiles invoke the compiler (as that is the one knowing where - # to find libraries), but we provide LD so other rules can invoke the linker. - tool_path { name: "ld" path: "/usr/bin/ld" } - - tool_path { name: "nm" path: "/usr/bin/nm" } - tool_path { name: "objcopy" path: "/usr/bin/objcopy" } - objcopy_embed_flag: "-I" - objcopy_embed_flag: "binary" - tool_path { name: "objdump" path: "/usr/bin/objdump" } - tool_path { name: "strip" path: "/usr/bin/strip" } - - # Anticipated future default. - unfiltered_cxx_flag: "-no-canonical-prefixes" - - # Make C++ compilation deterministic. Use linkstamping instead of these - # compiler symbols. - unfiltered_cxx_flag: "-Wno-builtin-macro-redefined" - unfiltered_cxx_flag: "-D__DATE__=\"redacted\"" - unfiltered_cxx_flag: "-D__TIMESTAMP__=\"redacted\"" - unfiltered_cxx_flag: "-D__TIME__=\"redacted\"" - - # Security hardening on by default. - # Conservative choice; -D_FORTIFY_SOURCE=2 may be unsafe in some cases. - # We need to undef it before redefining it as some distributions now have - # it enabled by default. - compiler_flag: "-U_FORTIFY_SOURCE" - compiler_flag: "-D_FORTIFY_SOURCE=1" - compiler_flag: "-fstack-protector" - compiler_flag: "-fPIE" - linker_flag: "-pie" - linker_flag: "-Wl,-z,relro,-z,now" - - # Enable coloring even if there's no attached terminal. Bazel removes the - # escape sequences if --nocolor is specified. This isn't supported by gcc - # on Ubuntu 14.04. - # compiler_flag: "-fcolor-diagnostics" - - # All warnings are enabled. Maybe enable -Werror as well? - compiler_flag: "-Wall" - # Enable a few more warnings that aren't part of -Wall. - compiler_flag: "-Wunused-but-set-parameter" - # But disable some that are problematic. - compiler_flag: "-Wno-free-nonheap-object" # has false positives - - # Keep stack frames for debugging, even in opt mode. - compiler_flag: "-fno-omit-frame-pointer" - - # Anticipated future default. - linker_flag: "-no-canonical-prefixes" - unfiltered_cxx_flag: "-fno-canonical-system-headers" - # Have gcc return the exit code from ld. - linker_flag: "-pass-exit-codes" - # Stamp the binary with a unique identifier. - linker_flag: "-Wl,--build-id=md5" - linker_flag: "-Wl,--hash-style=gnu" - # Gold linker only? Can we enable this by default? - # linker_flag: "-Wl,--warn-execstack" - # linker_flag: "-Wl,--detect-odr-violations" - - # Include directory for cuda headers. -%{cuda_include_path} - - compilation_mode_flags { - mode: DBG - # Enable debug symbols. - compiler_flag: "-g" - } - compilation_mode_flags { - mode: OPT - - # No debug symbols. - # Maybe we should enable https://gcc.gnu.org/wiki/DebugFission for opt or - # even generally? However, that can't happen here, as it requires special - # handling in Bazel. - compiler_flag: "-g0" - - # Conservative choice for -O - # -O3 can increase binary size and even slow down the resulting binaries. - # Profile first and / or use FDO if you need better performance than this. - compiler_flag: "-O2" - - # Disable assertions - compiler_flag: "-DNDEBUG" - - # Removal of unused code and data at link time (can this increase binary size in some cases?). - compiler_flag: "-ffunction-sections" - compiler_flag: "-fdata-sections" - linker_flag: "-Wl,--gc-sections" - } - linking_mode_flags { mode: DYNAMIC } -} - -toolchain { - abi_version: "local" - abi_libc_version: "local" - builtin_sysroot: "" - compiler: "compiler" - host_system_name: "local" - needsPic: true - target_libc: "macosx" - target_cpu: "darwin" - target_system_name: "local" - toolchain_identifier: "local_darwin" - - tool_path { name: "ar" path: "/usr/bin/libtool" } - tool_path { name: "compat-ld" path: "/usr/bin/ld" } - tool_path { name: "cpp" path: "/usr/bin/cpp" } - tool_path { name: "dwp" path: "/usr/bin/dwp" } - tool_path { name: "gcc" path: "clang/bin/crosstool_wrapper_driver_is_not_gcc" } - cxx_flag: "-std=c++11" - ar_flag: "-static" - ar_flag: "-s" - ar_flag: "-o" - linker_flag: "-lc++" - linker_flag: "-undefined" - linker_flag: "dynamic_lookup" - # TODO(ulfjack): This is wrong on so many levels. Figure out a way to auto-detect the proper - # setting from the local compiler, and also how to make incremental builds correct. - cxx_builtin_include_directory: "/" - tool_path { name: "gcov" path: "/usr/bin/gcov" } - tool_path { name: "ld" path: "/usr/bin/ld" } - tool_path { name: "nm" path: "/usr/bin/nm" } - tool_path { name: "objcopy" path: "/usr/bin/objcopy" } - objcopy_embed_flag: "-I" - objcopy_embed_flag: "binary" - tool_path { name: "objdump" path: "/usr/bin/objdump" } - tool_path { name: "strip" path: "/usr/bin/strip" } - - # Anticipated future default. - unfiltered_cxx_flag: "-no-canonical-prefixes" - # Make C++ compilation deterministic. Use linkstamping instead of these - # compiler symbols. - unfiltered_cxx_flag: "-Wno-builtin-macro-redefined" - unfiltered_cxx_flag: "-D__DATE__=\"redacted\"" - unfiltered_cxx_flag: "-D__TIMESTAMP__=\"redacted\"" - unfiltered_cxx_flag: "-D__TIME__=\"redacted\"" - - # Security hardening on by default. - # Conservative choice; -D_FORTIFY_SOURCE=2 may be unsafe in some cases. - compiler_flag: "-D_FORTIFY_SOURCE=1" - compiler_flag: "-fstack-protector" - - # Enable coloring even if there's no attached terminal. Bazel removes the - # escape sequences if --nocolor is specified. - compiler_flag: "-fcolor-diagnostics" - - # All warnings are enabled. Maybe enable -Werror as well? - compiler_flag: "-Wall" - # Enable a few more warnings that aren't part of -Wall. - compiler_flag: "-Wthread-safety" - compiler_flag: "-Wself-assign" - - # Keep stack frames for debugging, even in opt mode. - compiler_flag: "-fno-omit-frame-pointer" - - # Anticipated future default. - linker_flag: "-no-canonical-prefixes" - - # Include directory for cuda headers. -%{cuda_include_path} - - compilation_mode_flags { - mode: DBG - # Enable debug symbols. - compiler_flag: "-g" - } - compilation_mode_flags { - mode: OPT - # No debug symbols. - # Maybe we should enable https://gcc.gnu.org/wiki/DebugFission for opt or even generally? - # However, that can't happen here, as it requires special handling in Bazel. - compiler_flag: "-g0" - - # Conservative choice for -O - # -O3 can increase binary size and even slow down the resulting binaries. - # Profile first and / or use FDO if you need better performance than this. - compiler_flag: "-O2" - - # Disable assertions - compiler_flag: "-DNDEBUG" - - # Removal of unused code and data at link time (can this increase binary size in some cases?). - compiler_flag: "-ffunction-sections" - compiler_flag: "-fdata-sections" - } -} diff --git a/third_party/gpus/crosstool/clang/bin/crosstool_wrapper_driver_is_not_gcc.tpl b/third_party/gpus/crosstool/clang/bin/crosstool_wrapper_driver_is_not_gcc.tpl index 2558f46fd5..f4f4d0ee96 100755 --- a/third_party/gpus/crosstool/clang/bin/crosstool_wrapper_driver_is_not_gcc.tpl +++ b/third_party/gpus/crosstool/clang/bin/crosstool_wrapper_driver_is_not_gcc.tpl @@ -175,6 +175,11 @@ def InvokeNvcc(argv, log=False): # any other reliable way to just get the list of source files to be compiled. src_files = GetOptionValue(argv, 'c') + # Pass -w through from host to nvcc, but don't do anything fancier with + # warnings-related flags, since they're not necessarily the same across + # compilers. + warning_options = ' -w' if '-w' in argv else '' + if len(src_files) == 0: return 1 if len(out_file) != 1: @@ -205,6 +210,7 @@ def InvokeNvcc(argv, log=False): nvccopts += defines nvccopts += std_options nvccopts += m_options + nvccopts += warning_options if depfiles: # Generate the dependency file diff --git a/third_party/gpus/crosstool/windows/msvc_wrapper_for_nvcc.bat.tpl b/third_party/gpus/crosstool/windows/msvc_wrapper_for_nvcc.bat.tpl new file mode 100644 index 0000000000..8f8fb3e423 --- /dev/null +++ b/third_party/gpus/crosstool/windows/msvc_wrapper_for_nvcc.bat.tpl @@ -0,0 +1,20 @@ +:: Copyright 2015 The TensorFlow Authors. All Rights Reserved. +:: +:: Licensed under the Apache License, Version 2.0 (the "License"); +:: you may not use this file except in compliance with the License. +:: You may obtain a copy of the License at +:: +:: http://www.apache.org/licenses/LICENSE-2.0 +:: +:: Unless required by applicable law or agreed to in writing, software +:: distributed under the License is distributed on an "AS IS" BASIS, +:: WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +:: See the License for the specific language governing permissions and +:: limitations under the License. +:: ============================================================================= + +:: Invoke msvc_wrapper_for_nvcc.py, which is located in the same directory. +@echo OFF +set arg0=%~0 +for %%F in ("%arg0%") do set DRIVER_BIN=%%~dpF +"%{python_binary}" -B "%DRIVER_BIN%\msvc_wrapper_for_nvcc.py" %* diff --git a/third_party/gpus/crosstool/windows/msvc_wrapper_for_nvcc.py.tpl b/third_party/gpus/crosstool/windows/msvc_wrapper_for_nvcc.py.tpl new file mode 100644 index 0000000000..1a09756813 --- /dev/null +++ b/third_party/gpus/crosstool/windows/msvc_wrapper_for_nvcc.py.tpl @@ -0,0 +1,192 @@ +#!/usr/bin/env python +# Copyright 2015 The TensorFlow Authors. All Rights Reserved. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +# ============================================================================== + +"""Crosstool wrapper for compiling CUDA programs with nvcc on Windows. + +DESCRIPTION: + This script is the Windows version of //third_party/gpus/crosstool/crosstool_wrapper_is_not_gcc +""" + +from __future__ import print_function + +from argparse import ArgumentParser +import os +import subprocess +import re +import sys +import pipes + +# Template values set by cuda_autoconf. +CPU_COMPILER = ('%{cpu_compiler}') +GCC_HOST_COMPILER_PATH = ('%{gcc_host_compiler_path}') + +NVCC_PATH = '%{nvcc_path}' +NVCC_VERSION = '%{cuda_version}' +NVCC_TEMP_DIR = "%{nvcc_tmp_dir}" +supported_cuda_compute_capabilities = [ %{cuda_compute_capabilities} ] + +def Log(s): + print('gpus/crosstool: {0}'.format(s)) + + +def GetOptionValue(argv, option): + """Extract the list of values for option from options. + + Args: + option: The option whose value to extract, without the leading '/'. + + Returns: + 1. A list of values, either directly following the option, + (eg., /opt val1 val2) or values collected from multiple occurrences of + the option (eg., /opt val1 /opt val2). + 2. The leftover options. + """ + + parser = ArgumentParser(prefix_chars='/') + parser.add_argument('/' + option, nargs='*', action='append') + args, leftover = parser.parse_known_args(argv) + if args and vars(args)[option]: + return (sum(vars(args)[option], []), leftover) + return ([], leftover) + +def _update_options(nvcc_options): + if NVCC_VERSION in ("7.0",): + return nvcc_options + + update_options = { "relaxed-constexpr" : "expt-relaxed-constexpr" } + return [ update_options[opt] if opt in update_options else opt + for opt in nvcc_options ] + +def GetNvccOptions(argv): + """Collect the -nvcc_options values from argv. + + Args: + argv: A list of strings, possibly the argv passed to main(). + + Returns: + 1. The string that can be passed directly to nvcc. + 2. The leftover options. + """ + + parser = ArgumentParser() + parser.add_argument('-nvcc_options', nargs='*', action='append') + + args, leftover = parser.parse_known_args(argv) + + if args.nvcc_options: + options = _update_options(sum(args.nvcc_options, [])) + return (['--' + a for a in options], leftover) + return ([], leftover) + + +def InvokeNvcc(argv, log=False): + """Call nvcc with arguments assembled from argv. + + Args: + argv: A list of strings, possibly the argv passed to main(). + log: True if logging is requested. + + Returns: + The return value of calling os.system('nvcc ' + args) + """ + + src_files = [f for f in argv if + re.search('\.cpp$|\.cc$|\.c$|\.cxx$|\.C$', f)] + if len(src_files) == 0: + raise Error('No source files found for cuda compilation.') + + out_file = [ f for f in argv if f.startswith('/Fo') ] + if len(out_file) != 1: + raise Error('Please sepecify exactly one output file for cuda compilation.') + out = ['-o', out_file[0][len('/Fo'):]] + + nvcc_compiler_options, argv = GetNvccOptions(argv) + + opt_option, argv = GetOptionValue(argv, 'O') + opt = ['-g', '-G'] + if (len(opt_option) > 0 and opt_option[0] != 'd'): + opt = ['-O2'] + + include_options, argv = GetOptionValue(argv, 'I') + includes = ["-I " + include for include in include_options] + + defines, argv = GetOptionValue(argv, 'D') + defines = ['-D' + define for define in defines] + + undefines, argv = GetOptionValue(argv, 'U') + undefines = ['-U' + define for define in undefines] + + # The rest of the unrecongized options should be passed to host compiler + host_compiler_options = [option for option in argv if option not in (src_files + out_file)] + + m_options = ["-m64"] + + nvccopts = ['-D_FORCE_INLINES'] + for capability in supported_cuda_compute_capabilities: + capability = capability.replace('.', '') + nvccopts += [r'-gencode=arch=compute_%s,"code=sm_%s,compute_%s"' % ( + capability, capability, capability)] + nvccopts += nvcc_compiler_options + nvccopts += undefines + nvccopts += defines + nvccopts += m_options + nvccopts += ['--compiler-options="' + " ".join(host_compiler_options) + '"'] + nvccopts += ['-x', 'cu'] + opt + includes + out + ['-c'] + src_files + # If we don't specify --keep-dir, nvcc will generate intermediate files under TEMP + # Put them under NVCC_TEMP_DIR instead, then Bazel can ignore files under NVCC_TEMP_DIR during dependency check + # http://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html#options-for-guiding-compiler-driver + # Different actions are sharing NVCC_TEMP_DIR, so we cannot remove it if the directory already exists. + if os.path.isfile(NVCC_TEMP_DIR): + os.remove(NVCC_TEMP_DIR) + if not os.path.exists(NVCC_TEMP_DIR): + os.makedirs(NVCC_TEMP_DIR) + nvccopts += ['--keep', '--keep-dir', NVCC_TEMP_DIR] + cmd = [NVCC_PATH] + nvccopts + if log: + Log(cmd) + proc = subprocess.Popen(cmd, + stdout=sys.stdout, + stderr=sys.stderr, + env=os.environ.copy(), + shell=True) + proc.wait() + return proc.returncode + +def main(): + parser = ArgumentParser() + parser.add_argument('-x', nargs=1) + parser.add_argument('--cuda_log', action='store_true') + args, leftover = parser.parse_known_args(sys.argv[1:]) + + if args.x and args.x[0] == 'cuda': + if args.cuda_log: Log('-x cuda') + leftover = [pipes.quote(s) for s in leftover] + if args.cuda_log: Log('using nvcc') + return InvokeNvcc(leftover, log=args.cuda_log) + + # Strip our flags before passing through to the CPU compiler for files which + # are not -x cuda. We can't just pass 'leftover' because it also strips -x. + # We not only want to pass -x to the CPU compiler, but also keep it in its + # relative location in the argv list (the compiler is actually sensitive to + # this). + cpu_compiler_flags = [flag for flag in sys.argv[1:] + if not flag.startswith(('--cuda_log')) + and not flag.startswith(('-nvcc_options'))] + + return subprocess.call([CPU_COMPILER] + cpu_compiler_flags) + +if __name__ == '__main__': + sys.exit(main()) diff --git a/third_party/gpus/cuda/BUILD.tpl b/third_party/gpus/cuda/BUILD.tpl index 2a37c65bc7..f6b497f813 100644 --- a/third_party/gpus/cuda/BUILD.tpl +++ b/third_party/gpus/cuda/BUILD.tpl @@ -128,6 +128,15 @@ cc_library( ) cc_library( + name = "cudnn_header", + includes = [ + ".", + "cuda/include", + ], + visibility = ["//visibility:public"], +) + +cc_library( name = "cufft", srcs = ["cuda/lib/%{cufft_lib}"], data = ["cuda/lib/%{cufft_lib}"], diff --git a/third_party/gpus/cuda/BUILD.windows.tpl b/third_party/gpus/cuda/BUILD.windows.tpl new file mode 100644 index 0000000000..325d18b9cb --- /dev/null +++ b/third_party/gpus/cuda/BUILD.windows.tpl @@ -0,0 +1,164 @@ +licenses(["restricted"]) # MPL2, portions GPL v3, LGPL v3, BSD-like + +package(default_visibility = ["//visibility:public"]) + +config_setting( + name = "using_nvcc", + values = { + "define": "using_cuda_nvcc=true", + }, +) + +config_setting( + name = "using_clang", + values = { + "define": "using_cuda_clang=true", + }, +) + +# Equivalent to using_clang && -c opt. +config_setting( + name = "using_clang_opt", + values = { + "define": "using_cuda_clang=true", + "compilation_mode": "opt", + }, +) + +config_setting( + name = "darwin", + values = {"cpu": "darwin"}, + visibility = ["//visibility:public"], +) + +config_setting( + name = "freebsd", + values = {"cpu": "freebsd"}, + visibility = ["//visibility:public"], +) + +cc_library( + name = "cuda_headers", + hdrs = [ + "cuda/cuda_config.h", + %{cuda_headers} + ], + includes = [ + ".", + "cuda/include", + "cuda/include/crt", + ], + visibility = ["//visibility:public"], +) + +cc_import( + name = "cudart_static", + # /WHOLEARCHIVE:cudart_static.lib will cause a + # "Internal error during CImplib::EmitThunk" error. + # Treat this library as interface library to avoid being whole archived when + # linking a DLL that depends on this. + # TODO(pcloudy): Remove this rule after b/111278841 is resolved. + interface_library = "cuda/lib/%{cudart_static_lib}", + system_provided = 1, + visibility = ["//visibility:public"], +) + +cc_import( + name = "cuda_driver", + interface_library = "cuda/lib/%{cuda_driver_lib}", + system_provided = 1, + visibility = ["//visibility:public"], +) + +cc_import( + name = "cudart", + interface_library = "cuda/lib/%{cudart_lib}", + system_provided = 1, + visibility = ["//visibility:public"], +) + +cc_import( + name = "cublas", + interface_library = "cuda/lib/%{cublas_lib}", + system_provided = 1, + visibility = ["//visibility:public"], +) + +cc_import( + name = "cusolver", + interface_library = "cuda/lib/%{cusolver_lib}", + system_provided = 1, + visibility = ["//visibility:public"], +) + +cc_import( + name = "cudnn", + interface_library = "cuda/lib/%{cudnn_lib}", + system_provided = 1, + visibility = ["//visibility:public"], +) + +cc_library( + name = "cudnn_header", + includes = [ + ".", + "cuda/include", + ], + visibility = ["//visibility:public"], +) + +cc_import( + name = "cufft", + interface_library = "cuda/lib/%{cufft_lib}", + system_provided = 1, + visibility = ["//visibility:public"], +) + +cc_import( + name = "curand", + interface_library = "cuda/lib/%{curand_lib}", + system_provided = 1, + visibility = ["//visibility:public"], +) + +cc_library( + name = "cuda", + visibility = ["//visibility:public"], + deps = [ + ":cublas", + ":cuda_headers", + ":cudart", + ":cudnn", + ":cufft", + ":curand", + ], +) + +cc_library( + name = "cupti_headers", + hdrs = [ + "cuda/cuda_config.h", + ":cuda-extras", + ], + includes = [ + ".", + "cuda/", + "cuda/extras/CUPTI/include/", + ], + visibility = ["//visibility:public"], +) + +cc_import( + name = "cupti_dsos", + interface_library = "cuda/lib/%{cupti_lib}", + system_provided = 1, + visibility = ["//visibility:public"], +) + +cc_library( + name = "libdevice_root", + data = [":cuda-nvvm"], + visibility = ["//visibility:public"], +) + +%{cuda_include_genrules} diff --git a/third_party/gpus/cuda/remote.BUILD.tpl b/third_party/gpus/cuda/remote.BUILD.tpl index f774def5e6..100c7bb7c4 100644 --- a/third_party/gpus/cuda/remote.BUILD.tpl +++ b/third_party/gpus/cuda/remote.BUILD.tpl @@ -75,6 +75,11 @@ alias( ) alias( + name = "cudnn_header", + actual = "%{remote_cuda_repo}/cuda:cudnn_header", +) + +alias( name = "cufft", actual = "%{remote_cuda_repo}/cuda:cufft", ) diff --git a/third_party/gpus/cuda_configure.bzl b/third_party/gpus/cuda_configure.bzl index 6c9c128db6..5648b1525a 100644 --- a/third_party/gpus/cuda_configure.bzl +++ b/third_party/gpus/cuda_configure.bzl @@ -20,6 +20,7 @@ `/usr/local/cuda`. * `TF_CUDA_COMPUTE_CAPABILITIES`: The CUDA compute capabilities. Default is `3.5,5.2`. + * `PYTHON_BIN_PATH`: The python binary path """ _GCC_HOST_COMPILER_PATH = "GCC_HOST_COMPILER_PATH" @@ -31,6 +32,7 @@ _CUDNN_INSTALL_PATH = "CUDNN_INSTALL_PATH" _TF_CUDA_COMPUTE_CAPABILITIES = "TF_CUDA_COMPUTE_CAPABILITIES" _TF_CUDA_CONFIG_REPO = "TF_CUDA_CONFIG_REPO" _TF_DOWNLOAD_CLANG = "TF_DOWNLOAD_CLANG" +_PYTHON_BIN_PATH = "PYTHON_BIN_PATH" _DEFAULT_CUDA_VERSION = "" _DEFAULT_CUDNN_VERSION = "" @@ -44,12 +46,12 @@ _DEFAULT_CUDA_COMPUTE_CAPABILITIES = ["3.5", "5.2"] # will be used. For example, when looking for the cudart libraries, the first # attempt will be lib64/cudart inside the CUDA toolkit. CUDA_LIB_PATHS = [ - "lib64/", - "lib64/stubs/", - "lib/x86_64-linux-gnu/", - "lib/x64/", - "lib/", - "", + "lib64/", + "lib64/stubs/", + "lib/x86_64-linux-gnu/", + "lib/x64/", + "lib/", + "", ] # Lookup paths for cupti.h, relative to the CUDA toolkit directory. @@ -57,8 +59,9 @@ CUDA_LIB_PATHS = [ # On most systems, the cupti library is not installed in the same directory as # the other CUDA libraries but rather in a special extras/CUPTI directory. CUPTI_HEADER_PATHS = [ - "extras/CUPTI/include/", - "include/cuda/CUPTI/", + "extras/CUPTI/include/", + "include/cuda/CUPTI/", + "include/", ] # Lookup paths for the cupti library, relative to the @@ -66,25 +69,25 @@ CUPTI_HEADER_PATHS = [ # On most systems, the cupti library is not installed in the same directory as # the other CUDA libraries but rather in a special extras/CUPTI directory. CUPTI_LIB_PATHS = [ - "extras/CUPTI/lib64/", - "lib/x86_64-linux-gnu", - "lib64/", - "extras/CUPTI/libx64/", - "extras/CUPTI/lib/", - "lib/", + "extras/CUPTI/lib64/", + "lib/x86_64-linux-gnu/", + "lib64/", + "extras/CUPTI/libx64/", + "extras/CUPTI/lib/", + "lib/", ] # Lookup paths for CUDA headers (cuda.h) relative to the CUDA toolkit directory. CUDA_INCLUDE_PATHS = [ - "include/", - "include/cuda/" + "include/", + "include/cuda/", ] # Lookup paths for cudnn.h relative to the CUDNN install directory. CUDNN_INCLUDE_PATHS = [ - "", - "include/", - "include/cuda/", + "", + "include/", + "include/cuda/", ] # Lookup paths for NVVM libdevice relative to the CUDA directory toolkit. @@ -92,686 +95,842 @@ CUDNN_INCLUDE_PATHS = [ # libdevice implements mathematical functions for GPU kernels, and is provided # in NVVM bitcode (a subset of LLVM bitcode). NVVM_LIBDEVICE_PATHS = [ - "nvvm/libdevice/", - "share/cuda/", + "nvvm/libdevice/", + "share/cuda/", + "lib/nvidia-cuda-toolkit/libdevice/", ] -load(":download_clang.bzl", "download_clang") +# Files used to detect the NVVM libdevice path. +NVVM_LIBDEVICE_FILES = [ + # CUDA 9.0 has a single file. + "libdevice.10.bc", + + # CUDA 8.0 has separate files for compute versions 2.0, 3.0, 3.5 and 5.0. + # Probing for one of them is sufficient. + "libdevice.compute_20.10.bc", +] + +load("//third_party/clang_toolchain:download_clang.bzl", "download_clang") +load( + "@bazel_tools//tools/cpp:lib_cc_configure.bzl", + "escape_string", + "get_env_var", +) +load( + "@bazel_tools//tools/cpp:windows_cc_configure.bzl", + "find_msvc_tool", + "find_vc_path", + "setup_vc_env_vars", +) + +def _get_python_bin(repository_ctx): + """Gets the python bin path.""" + python_bin = repository_ctx.os.environ.get(_PYTHON_BIN_PATH) + if python_bin != None: + return python_bin + python_bin_name = "python.exe" if _is_windows(repository_ctx) else "python" + python_bin_path = repository_ctx.which(python_bin_name) + if python_bin_path != None: + return str(python_bin_path) + auto_configure_fail("Cannot find python in PATH, please make sure " + + "python is installed and add its directory in PATH, or --define " + + "%s='/something/else'.\nPATH=%s" % ( + _PYTHON_BIN_PATH, + repository_ctx.os.environ.get("PATH", ""), + )) + +def _get_nvcc_tmp_dir_for_windows(repository_ctx): + """Return the tmp directory for nvcc to generate intermediate source files.""" + escaped_tmp_dir = escape_string( + get_env_var(repository_ctx, "TMP", "C:\\Windows\\Temp").replace("\\", "\\\\"), + ) + return escaped_tmp_dir + "\\\\nvcc_inter_files_tmp_dir" + +def _get_msvc_compiler(repository_ctx): + vc_path = find_vc_path(repository_ctx) + return find_msvc_tool(repository_ctx, vc_path, "cl.exe").replace("\\", "/") + +def _get_win_cuda_defines(repository_ctx): + """Return CROSSTOOL defines for Windows""" + + # If we are not on Windows, return empty vaules for Windows specific fields. + # This ensures the CROSSTOOL file parser is happy. + if not _is_windows(repository_ctx): + return { + "%{msvc_env_tmp}": "", + "%{msvc_env_path}": "", + "%{msvc_env_include}": "", + "%{msvc_env_lib}": "", + "%{msvc_cl_path}": "", + "%{msvc_ml_path}": "", + "%{msvc_link_path}": "", + "%{msvc_lib_path}": "", + "%{cxx_builtin_include_directory}": "", + } + + vc_path = find_vc_path(repository_ctx) + if not vc_path: + auto_configure_fail("Visual C++ build tools not found on your machine." + + "Please check your installation following https://docs.bazel.build/versions/master/windows.html#using") + return {} + + env = setup_vc_env_vars(repository_ctx, vc_path) + escaped_paths = escape_string(env["PATH"]) + escaped_include_paths = escape_string(env["INCLUDE"]) + escaped_lib_paths = escape_string(env["LIB"]) + escaped_tmp_dir = escape_string( + get_env_var(repository_ctx, "TMP", "C:\\Windows\\Temp").replace("\\", "\\\\"), + ) + + msvc_cl_path = "windows/msvc_wrapper_for_nvcc.bat" + msvc_ml_path = find_msvc_tool(repository_ctx, vc_path, "ml64.exe").replace("\\", "/") + msvc_link_path = find_msvc_tool(repository_ctx, vc_path, "link.exe").replace("\\", "/") + msvc_lib_path = find_msvc_tool(repository_ctx, vc_path, "lib.exe").replace("\\", "/") + + # nvcc will generate some temporary source files under %{nvcc_tmp_dir} + # The generated files are guranteed to have unique name, so they can share the same tmp directory + escaped_cxx_include_directories = ["cxx_builtin_include_directory: \"%s\"" % _get_nvcc_tmp_dir_for_windows(repository_ctx)] + for path in escaped_include_paths.split(";"): + if path: + escaped_cxx_include_directories.append("cxx_builtin_include_directory: \"%s\"" % path) + + return { + "%{msvc_env_tmp}": escaped_tmp_dir, + "%{msvc_env_path}": escaped_paths, + "%{msvc_env_include}": escaped_include_paths, + "%{msvc_env_lib}": escaped_lib_paths, + "%{msvc_cl_path}": msvc_cl_path, + "%{msvc_ml_path}": msvc_ml_path, + "%{msvc_link_path}": msvc_link_path, + "%{msvc_lib_path}": msvc_lib_path, + "%{cxx_builtin_include_directory}": "\n".join(escaped_cxx_include_directories), + } # TODO(dzc): Once these functions have been factored out of Bazel's # cc_configure.bzl, load them from @bazel_tools instead. # BEGIN cc_configure common functions. def find_cc(repository_ctx): - """Find the C++ compiler.""" - # On Windows, we use Bazel's MSVC CROSSTOOL for GPU build - # Return a dummy value for GCC detection here to avoid error - if _is_windows(repository_ctx): - return "/use/--config=win-cuda --cpu=x64_windows_msvc/instead" - - if _use_cuda_clang(repository_ctx): - target_cc_name = "clang" - cc_path_envvar = _CLANG_CUDA_COMPILER_PATH - if _flag_enabled(repository_ctx, _TF_DOWNLOAD_CLANG): - return "extra_tools/bin/clang" - else: - target_cc_name = "gcc" - cc_path_envvar = _GCC_HOST_COMPILER_PATH - cc_name = target_cc_name - - if cc_path_envvar in repository_ctx.os.environ: - cc_name_from_env = repository_ctx.os.environ[cc_path_envvar].strip() - if cc_name_from_env: - cc_name = cc_name_from_env - if cc_name.startswith("/"): - # Absolute path, maybe we should make this supported by our which function. - return cc_name - cc = repository_ctx.which(cc_name) - if cc == None: - fail(("Cannot find {}, either correct your path or set the {}" + - " environment variable").format(target_cc_name, cc_path_envvar)) - return cc - + """Find the C++ compiler.""" + if _is_windows(repository_ctx): + return _get_msvc_compiler(repository_ctx) + + if _use_cuda_clang(repository_ctx): + target_cc_name = "clang" + cc_path_envvar = _CLANG_CUDA_COMPILER_PATH + if _flag_enabled(repository_ctx, _TF_DOWNLOAD_CLANG): + return "extra_tools/bin/clang" + else: + target_cc_name = "gcc" + cc_path_envvar = _GCC_HOST_COMPILER_PATH + cc_name = target_cc_name + + if cc_path_envvar in repository_ctx.os.environ: + cc_name_from_env = repository_ctx.os.environ[cc_path_envvar].strip() + if cc_name_from_env: + cc_name = cc_name_from_env + if cc_name.startswith("/"): + # Absolute path, maybe we should make this supported by our which function. + return cc_name + cc = repository_ctx.which(cc_name) + if cc == None: + fail(("Cannot find {}, either correct your path or set the {}" + + " environment variable").format(target_cc_name, cc_path_envvar)) + return cc _INC_DIR_MARKER_BEGIN = "#include <...>" - # OSX add " (framework directory)" at the end of line, strip it. _OSX_FRAMEWORK_SUFFIX = " (framework directory)" -_OSX_FRAMEWORK_SUFFIX_LEN = len(_OSX_FRAMEWORK_SUFFIX) -def _cxx_inc_convert(path): - """Convert path returned by cc -E xc++ in a complete path.""" - path = path.strip() - if path.endswith(_OSX_FRAMEWORK_SUFFIX): - path = path[:-_OSX_FRAMEWORK_SUFFIX_LEN].strip() - return path +_OSX_FRAMEWORK_SUFFIX_LEN = len(_OSX_FRAMEWORK_SUFFIX) +def _cxx_inc_convert(path): + """Convert path returned by cc -E xc++ in a complete path.""" + path = path.strip() + if path.endswith(_OSX_FRAMEWORK_SUFFIX): + path = path[:-_OSX_FRAMEWORK_SUFFIX_LEN].strip() + return path def _normalize_include_path(repository_ctx, path): - """Normalizes include paths before writing them to the crosstool. - - If path points inside the 'crosstool' folder of the repository, a relative - path is returned. - If path points outside the 'crosstool' folder, an absolute path is returned. - """ - path = str(repository_ctx.path(path)) - crosstool_folder = str(repository_ctx.path(".").get_child('crosstool')) + """Normalizes include paths before writing them to the crosstool. - if path.startswith(crosstool_folder): - # We drop the path to "$REPO/crosstool" and a trailing path separator. - return path[len(crosstool_folder)+1:] - return path + If path points inside the 'crosstool' folder of the repository, a relative + path is returned. + If path points outside the 'crosstool' folder, an absolute path is returned. + """ + path = str(repository_ctx.path(path)) + crosstool_folder = str(repository_ctx.path(".").get_child("crosstool")) + if path.startswith(crosstool_folder): + # We drop the path to "$REPO/crosstool" and a trailing path separator. + return path[len(crosstool_folder) + 1:] + return path def _get_cxx_inc_directories_impl(repository_ctx, cc, lang_is_cpp): - """Compute the list of default C or C++ include directories.""" - if lang_is_cpp: - lang = "c++" - else: - lang = "c" - result = repository_ctx.execute([cc, "-E", "-x" + lang, "-", "-v"]) - index1 = result.stderr.find(_INC_DIR_MARKER_BEGIN) - if index1 == -1: - return [] - index1 = result.stderr.find("\n", index1) - if index1 == -1: - return [] - index2 = result.stderr.rfind("\n ") - if index2 == -1 or index2 < index1: - return [] - index2 = result.stderr.find("\n", index2 + 1) - if index2 == -1: - inc_dirs = result.stderr[index1 + 1:] - else: - inc_dirs = result.stderr[index1 + 1:index2].strip() - - return [ - _normalize_include_path(repository_ctx, _cxx_inc_convert(p)) - for p in inc_dirs.split("\n") - ] + """Compute the list of default C or C++ include directories.""" + if lang_is_cpp: + lang = "c++" + else: + lang = "c" + result = repository_ctx.execute([cc, "-E", "-x" + lang, "-", "-v"]) + index1 = result.stderr.find(_INC_DIR_MARKER_BEGIN) + if index1 == -1: + return [] + index1 = result.stderr.find("\n", index1) + if index1 == -1: + return [] + index2 = result.stderr.rfind("\n ") + if index2 == -1 or index2 < index1: + return [] + index2 = result.stderr.find("\n", index2 + 1) + if index2 == -1: + inc_dirs = result.stderr[index1 + 1:] + else: + inc_dirs = result.stderr[index1 + 1:index2].strip() + return [ + _normalize_include_path(repository_ctx, _cxx_inc_convert(p)) + for p in inc_dirs.split("\n") + ] def get_cxx_inc_directories(repository_ctx, cc): - """Compute the list of default C and C++ include directories.""" - # For some reason `clang -xc` sometimes returns include paths that are - # different from the ones from `clang -xc++`. (Symlink and a dir) - # So we run the compiler with both `-xc` and `-xc++` and merge resulting lists - includes_cpp = _get_cxx_inc_directories_impl(repository_ctx, cc, True) - includes_c = _get_cxx_inc_directories_impl(repository_ctx, cc, False) + """Compute the list of default C and C++ include directories.""" - includes_cpp_set = depset(includes_cpp) - return includes_cpp + [inc for inc in includes_c - if inc not in includes_cpp_set] + # For some reason `clang -xc` sometimes returns include paths that are + # different from the ones from `clang -xc++`. (Symlink and a dir) + # So we run the compiler with both `-xc` and `-xc++` and merge resulting lists + includes_cpp = _get_cxx_inc_directories_impl(repository_ctx, cc, True) + includes_c = _get_cxx_inc_directories_impl(repository_ctx, cc, False) + includes_cpp_set = depset(includes_cpp) + return includes_cpp + [ + inc + for inc in includes_c + if inc not in includes_cpp_set + ] def auto_configure_fail(msg): - """Output failure message when cuda configuration fails.""" - red = "\033[0;31m" - no_color = "\033[0m" - fail("\n%sCuda Configuration Error:%s %s\n" % (red, no_color, msg)) -# END cc_configure common functions (see TODO above). + """Output failure message when cuda configuration fails.""" + red = "\033[0;31m" + no_color = "\033[0m" + fail("\n%sCuda Configuration Error:%s %s\n" % (red, no_color, msg)) +# END cc_configure common functions (see TODO above). def _host_compiler_includes(repository_ctx, cc): - """Generates the cxx_builtin_include_directory entries for gcc inc dirs. - - Args: - repository_ctx: The repository context. - cc: The path to the gcc host compiler. - - Returns: - A string containing the cxx_builtin_include_directory for each of the gcc - host compiler include directories, which can be added to the CROSSTOOL - file. - """ - inc_dirs = get_cxx_inc_directories(repository_ctx, cc) - inc_entries = [] - for inc_dir in inc_dirs: - inc_entries.append(" cxx_builtin_include_directory: \"%s\"" % inc_dir) - return "\n".join(inc_entries) + """Generates the cxx_builtin_include_directory entries for gcc inc dirs. + + Args: + repository_ctx: The repository context. + cc: The path to the gcc host compiler. + + Returns: + A string containing the cxx_builtin_include_directory for each of the gcc + host compiler include directories, which can be added to the CROSSTOOL + file. + """ + inc_dirs = get_cxx_inc_directories(repository_ctx, cc) + inc_entries = [] + for inc_dir in inc_dirs: + inc_entries.append(" cxx_builtin_include_directory: \"%s\"" % inc_dir) + return "\n".join(inc_entries) def _cuda_include_path(repository_ctx, cuda_config): - """Generates the cxx_builtin_include_directory entries for cuda inc dirs. - - Args: - repository_ctx: The repository context. - cc: The path to the gcc host compiler. - - Returns: - A string containing the cxx_builtin_include_directory for each of the gcc - host compiler include directories, which can be added to the CROSSTOOL - file. - """ - nvcc_path = repository_ctx.path("%s/bin/nvcc%s" % - (cuda_config.cuda_toolkit_path, - ".exe" if cuda_config.cpu_value == "Windows" else "")) - result = repository_ctx.execute([nvcc_path, '-v', - '/dev/null', '-o', '/dev/null']) - target_dir = "" - for one_line in result.stderr.splitlines(): - if one_line.startswith('#$ _TARGET_DIR_='): - target_dir = (cuda_config.cuda_toolkit_path + '/' + - one_line.replace('#$ _TARGET_DIR_=', '') + "/include") - inc_entries = [] - if target_dir != "": - inc_entries.append(" cxx_builtin_include_directory: \"%s\"" % target_dir) - default_include = cuda_config.cuda_toolkit_path + '/include' - inc_entries.append(" cxx_builtin_include_directory: \"%s\"" % - default_include) - return "\n".join(inc_entries) - + """Generates the cxx_builtin_include_directory entries for cuda inc dirs. + + Args: + repository_ctx: The repository context. + cc: The path to the gcc host compiler. + + Returns: + A string containing the cxx_builtin_include_directory for each of the gcc + host compiler include directories, which can be added to the CROSSTOOL + file. + """ + nvcc_path = repository_ctx.path("%s/bin/nvcc%s" % + ( + cuda_config.cuda_toolkit_path, + ".exe" if cuda_config.cpu_value == "Windows" else "", + )) + result = repository_ctx.execute([ + nvcc_path, + "-v", + "/dev/null", + "-o", + "/dev/null", + ]) + target_dir = "" + for one_line in result.stderr.splitlines(): + if one_line.startswith("#$ _TARGET_DIR_="): + target_dir = (cuda_config.cuda_toolkit_path + "/" + + one_line.replace("#$ _TARGET_DIR_=", "") + "/include") + inc_entries = [] + if target_dir != "": + inc_entries.append(" cxx_builtin_include_directory: \"%s\"" % target_dir) + default_include = cuda_config.cuda_toolkit_path + "/include" + inc_entries.append(" cxx_builtin_include_directory: \"%s\"" % + default_include) + return "\n".join(inc_entries) def _enable_cuda(repository_ctx): - if "TF_NEED_CUDA" in repository_ctx.os.environ: - enable_cuda = repository_ctx.os.environ["TF_NEED_CUDA"].strip() - return enable_cuda == "1" - return False - + if "TF_NEED_CUDA" in repository_ctx.os.environ: + enable_cuda = repository_ctx.os.environ["TF_NEED_CUDA"].strip() + return enable_cuda == "1" + return False def _cuda_toolkit_path(repository_ctx): - """Finds the cuda toolkit directory. - - Args: - repository_ctx: The repository context. + """Finds the cuda toolkit directory. - Returns: - A speculative real path of the cuda toolkit install directory. - """ - cuda_toolkit_path = _DEFAULT_CUDA_TOOLKIT_PATH - if _CUDA_TOOLKIT_PATH in repository_ctx.os.environ: - cuda_toolkit_path = repository_ctx.os.environ[_CUDA_TOOLKIT_PATH].strip() - if not repository_ctx.path(cuda_toolkit_path).exists: - auto_configure_fail("Cannot find cuda toolkit path.") - return str(repository_ctx.path(cuda_toolkit_path).realpath) + Args: + repository_ctx: The repository context. + Returns: + A speculative real path of the cuda toolkit install directory. + """ + cuda_toolkit_path = _DEFAULT_CUDA_TOOLKIT_PATH + if _CUDA_TOOLKIT_PATH in repository_ctx.os.environ: + cuda_toolkit_path = repository_ctx.os.environ[_CUDA_TOOLKIT_PATH].strip() + if not repository_ctx.path(cuda_toolkit_path).exists: + auto_configure_fail("Cannot find cuda toolkit path.") + return str(repository_ctx.path(cuda_toolkit_path).realpath) def _cudnn_install_basedir(repository_ctx): - """Finds the cudnn install directory.""" - cudnn_install_path = _DEFAULT_CUDNN_INSTALL_PATH - if _CUDNN_INSTALL_PATH in repository_ctx.os.environ: - cudnn_install_path = repository_ctx.os.environ[_CUDNN_INSTALL_PATH].strip() - if not repository_ctx.path(cudnn_install_path).exists: - auto_configure_fail("Cannot find cudnn install path.") - return cudnn_install_path - + """Finds the cudnn install directory.""" + cudnn_install_path = _DEFAULT_CUDNN_INSTALL_PATH + if _CUDNN_INSTALL_PATH in repository_ctx.os.environ: + cudnn_install_path = repository_ctx.os.environ[_CUDNN_INSTALL_PATH].strip() + if not repository_ctx.path(cudnn_install_path).exists: + auto_configure_fail("Cannot find cudnn install path.") + return cudnn_install_path def matches_version(environ_version, detected_version): - """Checks whether the user-specified version matches the detected version. - - This function performs a weak matching so that if the user specifies only the - major or major and minor versions, the versions are still considered matching - if the version parts match. To illustrate: - - environ_version detected_version result - ----------------------------------------- - 5.1.3 5.1.3 True - 5.1 5.1.3 True - 5 5.1 True - 5.1.3 5.1 False - 5.2.3 5.1.3 False - - Args: - environ_version: The version specified by the user via environment - variables. - detected_version: The version autodetected from the CUDA installation on - the system. - - Returns: True if user-specified version matches detected version and False - otherwise. - """ - environ_version_parts = environ_version.split(".") - detected_version_parts = detected_version.split(".") - if len(detected_version_parts) < len(environ_version_parts): - return False - for i, part in enumerate(detected_version_parts): - if i >= len(environ_version_parts): - break - if part != environ_version_parts[i]: - return False - return True - + """Checks whether the user-specified version matches the detected version. + + This function performs a weak matching so that if the user specifies only the + major or major and minor versions, the versions are still considered matching + if the version parts match. To illustrate: + + environ_version detected_version result + ----------------------------------------- + 5.1.3 5.1.3 True + 5.1 5.1.3 True + 5 5.1 True + 5.1.3 5.1 False + 5.2.3 5.1.3 False + + Args: + environ_version: The version specified by the user via environment + variables. + detected_version: The version autodetected from the CUDA installation on + the system. + + Returns: True if user-specified version matches detected version and False + otherwise. + """ + environ_version_parts = environ_version.split(".") + detected_version_parts = detected_version.split(".") + if len(detected_version_parts) < len(environ_version_parts): + return False + for i, part in enumerate(detected_version_parts): + if i >= len(environ_version_parts): + break + if part != environ_version_parts[i]: + return False + return True _NVCC_VERSION_PREFIX = "Cuda compilation tools, release " - def _cuda_version(repository_ctx, cuda_toolkit_path, cpu_value): - """Detects the version of CUDA installed on the system. - - Args: - repository_ctx: The repository context. - cuda_toolkit_path: The CUDA install directory. - - Returns: - String containing the version of CUDA. - """ - # Run nvcc --version and find the line containing the CUDA version. - nvcc_path = repository_ctx.path("%s/bin/nvcc%s" % - (cuda_toolkit_path, - ".exe" if cpu_value == "Windows" else "")) - if not nvcc_path.exists: - auto_configure_fail("Cannot find nvcc at %s" % str(nvcc_path)) - result = repository_ctx.execute([str(nvcc_path), '--version']) - if result.stderr: - auto_configure_fail("Error running nvcc --version: %s" % result.stderr) - lines = result.stdout.splitlines() - version_line = lines[len(lines) - 1] - if version_line.find(_NVCC_VERSION_PREFIX) == -1: - auto_configure_fail( - "Could not parse CUDA version from nvcc --version. Got: %s" % - result.stdout) - - # Parse the CUDA version from the line containing the CUDA version. - prefix_removed = version_line.replace(_NVCC_VERSION_PREFIX, '') - parts = prefix_removed.split(",") - if len(parts) != 2 or len(parts[0]) < 2: - auto_configure_fail( - "Could not parse CUDA version from nvcc --version. Got: %s" % - result.stdout) - full_version = parts[1].strip() - if full_version.startswith('V'): - full_version = full_version[1:] - - # Check whether TF_CUDA_VERSION was set by the user and fail if it does not - # match the detected version. - environ_version = "" - if _TF_CUDA_VERSION in repository_ctx.os.environ: - environ_version = repository_ctx.os.environ[_TF_CUDA_VERSION].strip() - if environ_version and not matches_version(environ_version, full_version): - auto_configure_fail( - ("CUDA version detected from nvcc (%s) does not match " + - "TF_CUDA_VERSION (%s)") % (full_version, environ_version)) - - # We only use the version consisting of the major and minor version numbers. - version_parts = full_version.split('.') - if len(version_parts) < 2: - auto_configure_fail("CUDA version detected from nvcc (%s) is incomplete.") - if cpu_value == "Windows": - version = "64_%s%s" % (version_parts[0], version_parts[1]) - else: - version = "%s.%s" % (version_parts[0], version_parts[1]) - return version + """Detects the version of CUDA installed on the system. + + Args: + repository_ctx: The repository context. + cuda_toolkit_path: The CUDA install directory. + + Returns: + String containing the version of CUDA. + """ + + # Run nvcc --version and find the line containing the CUDA version. + nvcc_path = repository_ctx.path("%s/bin/nvcc%s" % + ( + cuda_toolkit_path, + ".exe" if cpu_value == "Windows" else "", + )) + if not nvcc_path.exists: + auto_configure_fail("Cannot find nvcc at %s" % str(nvcc_path)) + result = repository_ctx.execute([str(nvcc_path), "--version"]) + if result.stderr: + auto_configure_fail("Error running nvcc --version: %s" % result.stderr) + lines = result.stdout.splitlines() + version_line = lines[len(lines) - 1] + if version_line.find(_NVCC_VERSION_PREFIX) == -1: + auto_configure_fail( + "Could not parse CUDA version from nvcc --version. Got: %s" % + result.stdout, + ) + # Parse the CUDA version from the line containing the CUDA version. + prefix_removed = version_line.replace(_NVCC_VERSION_PREFIX, "") + parts = prefix_removed.split(",") + if len(parts) != 2 or len(parts[0]) < 2: + auto_configure_fail( + "Could not parse CUDA version from nvcc --version. Got: %s" % + result.stdout, + ) + full_version = parts[1].strip() + if full_version.startswith("V"): + full_version = full_version[1:] + + # Check whether TF_CUDA_VERSION was set by the user and fail if it does not + # match the detected version. + environ_version = "" + if _TF_CUDA_VERSION in repository_ctx.os.environ: + environ_version = repository_ctx.os.environ[_TF_CUDA_VERSION].strip() + if environ_version and not matches_version(environ_version, full_version): + auto_configure_fail( + ("CUDA version detected from nvcc (%s) does not match " + + "TF_CUDA_VERSION (%s)") % (full_version, environ_version), + ) + + # We only use the version consisting of the major and minor version numbers. + version_parts = full_version.split(".") + if len(version_parts) < 2: + auto_configure_fail("CUDA version detected from nvcc (%s) is incomplete.") + if cpu_value == "Windows": + version = "64_%s%s" % (version_parts[0], version_parts[1]) + else: + version = "%s.%s" % (version_parts[0], version_parts[1]) + return version _DEFINE_CUDNN_MAJOR = "#define CUDNN_MAJOR" _DEFINE_CUDNN_MINOR = "#define CUDNN_MINOR" _DEFINE_CUDNN_PATCHLEVEL = "#define CUDNN_PATCHLEVEL" - def find_cuda_define(repository_ctx, header_dir, header_file, define): - """Returns the value of a #define in a header file. - - Greps through a header file and returns the value of the specified #define. - If the #define is not found, then raise an error. - - Args: - repository_ctx: The repository context. - header_dir: The directory containing the header file. - header_file: The header file name. - define: The #define to search for. - - Returns: - The value of the #define found in the header. - """ - # Confirm location of the header and grep for the line defining the macro. - h_path = repository_ctx.path("%s/%s" % (header_dir, header_file)) - if not h_path.exists: - auto_configure_fail("Cannot find %s at %s" % (header_file, str(h_path))) - result = repository_ctx.execute( - # Grep one more lines as some #defines are splitted into two lines. - ["grep", "--color=never", "-A1", "-E", define, str(h_path)]) - if result.stderr: - auto_configure_fail("Error reading %s: %s" % (str(h_path), result.stderr)) - - # Parse the version from the line defining the macro. - if result.stdout.find(define) == -1: - auto_configure_fail("Cannot find line containing '%s' in %s" % - (define, h_path)) - # Split results to lines - lines = result.stdout.split('\n') - num_lines = len(lines) - for l in range(num_lines): - line = lines[l] - if define in line: # Find the line with define - version = line - if l != num_lines-1 and line[-1] == '\\': # Add next line, if multiline - version = version[:-1] + lines[l+1] - break - # Remove any comments - version = version.split("//")[0] - # Remove define name - version = version.replace(define, "").strip() - # Remove the code after the version number. - version_end = version.find(" ") - if version_end != -1: - if version_end == 0: - auto_configure_fail( - "Cannot extract the version from line containing '%s' in %s" % - (define, str(h_path))) - version = version[:version_end].strip() - return version - + """Returns the value of a #define in a header file. + + Greps through a header file and returns the value of the specified #define. + If the #define is not found, then raise an error. + + Args: + repository_ctx: The repository context. + header_dir: The directory containing the header file. + header_file: The header file name. + define: The #define to search for. + + Returns: + The value of the #define found in the header. + """ + + # Confirm location of the header and grep for the line defining the macro. + h_path = repository_ctx.path("%s/%s" % (header_dir, header_file)) + if not h_path.exists: + auto_configure_fail("Cannot find %s at %s" % (header_file, str(h_path))) + result = repository_ctx.execute( + # Grep one more lines as some #defines are splitted into two lines. + ["grep", "--color=never", "-A1", "-E", define, str(h_path)], + ) + if result.stderr: + auto_configure_fail("Error reading %s: %s" % (str(h_path), result.stderr)) + + # Parse the version from the line defining the macro. + if result.stdout.find(define) == -1: + auto_configure_fail("Cannot find line containing '%s' in %s" % + (define, h_path)) + + # Split results to lines + lines = result.stdout.split("\n") + num_lines = len(lines) + for l in range(num_lines): + line = lines[l] + if define in line: # Find the line with define + version = line + if l != num_lines - 1 and line[-1] == "\\": # Add next line, if multiline + version = version[:-1] + lines[l + 1] + break + + # Remove any comments + version = version.split("//")[0] + + # Remove define name + version = version.replace(define, "").strip() + + # Remove the code after the version number. + version_end = version.find(" ") + if version_end != -1: + if version_end == 0: + auto_configure_fail( + "Cannot extract the version from line containing '%s' in %s" % + (define, str(h_path)), + ) + version = version[:version_end].strip() + return version def _cudnn_version(repository_ctx, cudnn_install_basedir, cpu_value): - """Detects the version of cuDNN installed on the system. - - Args: - repository_ctx: The repository context. - cpu_value: The name of the host operating system. - cudnn_install_basedir: The cuDNN install directory. - - Returns: - A string containing the version of cuDNN. - """ - cudnn_header_dir = _find_cudnn_header_dir(repository_ctx, - cudnn_install_basedir) - major_version = find_cuda_define( - repository_ctx, cudnn_header_dir, "cudnn.h", _DEFINE_CUDNN_MAJOR) - minor_version = find_cuda_define( - repository_ctx, cudnn_header_dir, "cudnn.h", _DEFINE_CUDNN_MINOR) - patch_version = find_cuda_define( - repository_ctx, cudnn_header_dir, "cudnn.h", _DEFINE_CUDNN_PATCHLEVEL) - full_version = "%s.%s.%s" % (major_version, minor_version, patch_version) - - # Check whether TF_CUDNN_VERSION was set by the user and fail if it does not - # match the detected version. - environ_version = "" - if _TF_CUDNN_VERSION in repository_ctx.os.environ: - environ_version = repository_ctx.os.environ[_TF_CUDNN_VERSION].strip() - if environ_version and not matches_version(environ_version, full_version): - cudnn_h_path = repository_ctx.path("%s/include/cudnn.h" % - cudnn_install_basedir) - auto_configure_fail( - ("cuDNN version detected from %s (%s) does not match " + - "TF_CUDNN_VERSION (%s)") % - (str(cudnn_h_path), full_version, environ_version)) - - # We only use the major version since we use the libcudnn libraries that are - # only versioned with the major version (e.g. libcudnn.so.5). - version = major_version - if cpu_value == "Windows": - version = "64_" + version - return version + """Detects the version of cuDNN installed on the system. + Args: + repository_ctx: The repository context. + cpu_value: The name of the host operating system. + cudnn_install_basedir: The cuDNN install directory. + + Returns: + A string containing the version of cuDNN. + """ + cudnn_header_dir = _find_cudnn_header_dir( + repository_ctx, + cudnn_install_basedir, + ) + major_version = find_cuda_define( + repository_ctx, + cudnn_header_dir, + "cudnn.h", + _DEFINE_CUDNN_MAJOR, + ) + minor_version = find_cuda_define( + repository_ctx, + cudnn_header_dir, + "cudnn.h", + _DEFINE_CUDNN_MINOR, + ) + patch_version = find_cuda_define( + repository_ctx, + cudnn_header_dir, + "cudnn.h", + _DEFINE_CUDNN_PATCHLEVEL, + ) + full_version = "%s.%s.%s" % (major_version, minor_version, patch_version) + + # Check whether TF_CUDNN_VERSION was set by the user and fail if it does not + # match the detected version. + environ_version = "" + if _TF_CUDNN_VERSION in repository_ctx.os.environ: + environ_version = repository_ctx.os.environ[_TF_CUDNN_VERSION].strip() + if environ_version and not matches_version(environ_version, full_version): + cudnn_h_path = repository_ctx.path("%s/include/cudnn.h" % + cudnn_install_basedir) + auto_configure_fail( + ("cuDNN version detected from %s (%s) does not match " + + "TF_CUDNN_VERSION (%s)") % + (str(cudnn_h_path), full_version, environ_version), + ) -def _compute_capabilities(repository_ctx): - """Returns a list of strings representing cuda compute capabilities.""" - if _TF_CUDA_COMPUTE_CAPABILITIES not in repository_ctx.os.environ: - return _DEFAULT_CUDA_COMPUTE_CAPABILITIES - capabilities_str = repository_ctx.os.environ[_TF_CUDA_COMPUTE_CAPABILITIES] - capabilities = capabilities_str.split(",") - for capability in capabilities: - # Workaround for Skylark's lack of support for regex. This check should - # be equivalent to checking: - # if re.match("[0-9]+.[0-9]+", capability) == None: - parts = capability.split(".") - if len(parts) != 2 or not parts[0].isdigit() or not parts[1].isdigit(): - auto_configure_fail("Invalid compute capability: %s" % capability) - return capabilities + # We only use the major version since we use the libcudnn libraries that are + # only versioned with the major version (e.g. libcudnn.so.5). + version = major_version + if cpu_value == "Windows": + version = "64_" + version + return version +def _compute_capabilities(repository_ctx): + """Returns a list of strings representing cuda compute capabilities.""" + if _TF_CUDA_COMPUTE_CAPABILITIES not in repository_ctx.os.environ: + return _DEFAULT_CUDA_COMPUTE_CAPABILITIES + capabilities_str = repository_ctx.os.environ[_TF_CUDA_COMPUTE_CAPABILITIES] + capabilities = capabilities_str.split(",") + for capability in capabilities: + # Workaround for Skylark's lack of support for regex. This check should + # be equivalent to checking: + # if re.match("[0-9]+.[0-9]+", capability) == None: + parts = capability.split(".") + if len(parts) != 2 or not parts[0].isdigit() or not parts[1].isdigit(): + auto_configure_fail("Invalid compute capability: %s" % capability) + return capabilities def get_cpu_value(repository_ctx): - """Returns the name of the host operating system. + """Returns the name of the host operating system. + + Args: + repository_ctx: The repository context. + + Returns: + A string containing the name of the host operating system. + """ + os_name = repository_ctx.os.name.lower() + if os_name.startswith("mac os"): + return "Darwin" + if os_name.find("windows") != -1: + return "Windows" + result = repository_ctx.execute(["uname", "-s"]) + return result.stdout.strip() - Args: - repository_ctx: The repository context. - - Returns: - A string containing the name of the host operating system. - """ - os_name = repository_ctx.os.name.lower() - if os_name.startswith("mac os"): - return "Darwin" - if os_name.find("windows") != -1: - return "Windows" - result = repository_ctx.execute(["uname", "-s"]) - return result.stdout.strip() +def _is_windows(repository_ctx): + """Returns true if the host operating system is windows.""" + return get_cpu_value(repository_ctx) == "Windows" +def _lib_name(lib, cpu_value, version = "", static = False): + """Constructs the platform-specific name of a library. -def _is_windows(repository_ctx): - """Returns true if the host operating system is windows.""" - return get_cpu_value(repository_ctx) == "Windows" - -def _lib_name(lib, cpu_value, version="", static=False): - """Constructs the platform-specific name of a library. - - Args: - lib: The name of the library, such as "cudart" - cpu_value: The name of the host operating system. - version: The version of the library. - static: True the library is static or False if it is a shared object. - - Returns: - The platform-specific name of the library. - """ - if cpu_value in ("Linux", "FreeBSD"): - if static: - return "lib%s.a" % lib - else: - if version: - version = ".%s" % version - return "lib%s.so%s" % (lib, version) - elif cpu_value == "Windows": - return "%s.lib" % lib - elif cpu_value == "Darwin": - if static: - return "lib%s.a" % lib + Args: + lib: The name of the library, such as "cudart" + cpu_value: The name of the host operating system. + version: The version of the library. + static: True the library is static or False if it is a shared object. + + Returns: + The platform-specific name of the library. + """ + if cpu_value in ("Linux", "FreeBSD"): + if static: + return "lib%s.a" % lib + else: + if version: + version = ".%s" % version + return "lib%s.so%s" % (lib, version) + elif cpu_value == "Windows": + return "%s.lib" % lib + elif cpu_value == "Darwin": + if static: + return "lib%s.a" % lib + elif version: + version = ".%s" % version + return "lib%s%s.dylib" % (lib, version) else: - if version: - version = ".%s" % version - return "lib%s%s.dylib" % (lib, version) - else: - auto_configure_fail("Invalid cpu_value: %s" % cpu_value) - - -def _find_cuda_lib(lib, repository_ctx, cpu_value, basedir, version="", - static=False): - """Finds the given CUDA or cuDNN library on the system. - - Args: - lib: The name of the library, such as "cudart" - repository_ctx: The repository context. - cpu_value: The name of the host operating system. - basedir: The install directory of CUDA or cuDNN. - version: The version of the library. - static: True if static library, False if shared object. - - Returns: - Returns a struct with the following fields: - file_name: The basename of the library found on the system. - path: The full path to the library. - """ - file_name = _lib_name(lib, cpu_value, version, static) - for relative_path in CUDA_LIB_PATHS: - path = repository_ctx.path("%s/%s%s" % (basedir, relative_path, file_name)) - if path.exists: - return struct(file_name=file_name, path=str(path.realpath)) - auto_configure_fail("Cannot find cuda library %s" % file_name) - + auto_configure_fail("Invalid cpu_value: %s" % cpu_value) + +def _find_cuda_lib( + lib, + repository_ctx, + cpu_value, + basedir, + version = "", + static = False): + """Finds the given CUDA or cuDNN library on the system. + + Args: + lib: The name of the library, such as "cudart" + repository_ctx: The repository context. + cpu_value: The name of the host operating system. + basedir: The install directory of CUDA or cuDNN. + version: The version of the library. + static: True if static library, False if shared object. + + Returns: + Returns a struct with the following fields: + file_name: The basename of the library found on the system. + path: The full path to the library. + """ + file_name = _lib_name(lib, cpu_value, version, static) + for relative_path in CUDA_LIB_PATHS: + path = repository_ctx.path("%s/%s%s" % (basedir, relative_path, file_name)) + if path.exists: + return struct(file_name = file_name, path = str(path.realpath)) + auto_configure_fail("Cannot find cuda library %s" % file_name) def _find_cupti_header_dir(repository_ctx, cuda_config): - """Returns the path to the directory containing cupti.h - - On most systems, the cupti library is not installed in the same directory as - the other CUDA libraries but rather in a special extras/CUPTI directory. + """Returns the path to the directory containing cupti.h - Args: - repository_ctx: The repository context. - cuda_config: The CUDA config as returned by _get_cuda_config + On most systems, the cupti library is not installed in the same directory as + the other CUDA libraries but rather in a special extras/CUPTI directory. - Returns: - The path of the directory containing the cupti header. - """ - cuda_toolkit_path = cuda_config.cuda_toolkit_path - for relative_path in CUPTI_HEADER_PATHS: - if repository_ctx.path("%s/%scupti.h" % (cuda_toolkit_path, relative_path)).exists: - return ("%s/%s" % (cuda_toolkit_path, relative_path))[:-1] - auto_configure_fail("Cannot find cupti.h under %s" % cuda_toolkit_path) + Args: + repository_ctx: The repository context. + cuda_config: The CUDA config as returned by _get_cuda_config + Returns: + The path of the directory containing the cupti header. + """ + cuda_toolkit_path = cuda_config.cuda_toolkit_path + for relative_path in CUPTI_HEADER_PATHS: + if repository_ctx.path("%s/%scupti.h" % (cuda_toolkit_path, relative_path)).exists: + return ("%s/%s" % (cuda_toolkit_path, relative_path))[:-1] + auto_configure_fail("Cannot find cupti.h under %s" % ", ".join([cuda_toolkit_path + "/" + s for s in CUPTI_HEADER_PATHS])) def _find_cupti_lib(repository_ctx, cuda_config): - """Finds the cupti library on the system. - - On most systems, the cupti library is not installed in the same directory as - the other CUDA libraries but rather in a special extras/CUPTI directory. - - Args: - repository_ctx: The repository context. - cuda_config: The cuda configuration as returned by _get_cuda_config. - - Returns: - Returns a struct with the following fields: - file_name: The basename of the library found on the system. - path: The full path to the library. - """ - file_name = _lib_name("cupti", cuda_config.cpu_value, - cuda_config.cuda_version) - cuda_toolkit_path = cuda_config.cuda_toolkit_path - for relative_path in CUPTI_LIB_PATHS: - path = repository_ctx.path( - "%s/%s%s" % (cuda_toolkit_path, relative_path, file_name)) - if path.exists: - return struct(file_name=file_name, path=str(path.realpath)) - - auto_configure_fail("Cannot find cupti library %s" % file_name) + """Finds the cupti library on the system. + + On most systems, the cupti library is not installed in the same directory as + the other CUDA libraries but rather in a special extras/CUPTI directory. + + Args: + repository_ctx: The repository context. + cuda_config: The cuda configuration as returned by _get_cuda_config. + + Returns: + Returns a struct with the following fields: + file_name: The basename of the library found on the system. + path: The full path to the library. + """ + file_name = _lib_name( + "cupti", + cuda_config.cpu_value, + cuda_config.cuda_version, + ) + cuda_toolkit_path = cuda_config.cuda_toolkit_path + for relative_path in CUPTI_LIB_PATHS: + path = repository_ctx.path( + "%s/%s%s" % (cuda_toolkit_path, relative_path, file_name), + ) + if path.exists: + return struct(file_name = file_name, path = str(path.realpath)) -def _find_libs(repository_ctx, cuda_config): - """Returns the CUDA and cuDNN libraries on the system. - - Args: - repository_ctx: The repository context. - cuda_config: The CUDA config as returned by _get_cuda_config - - Returns: - Map of library names to structs of filename and path. - """ - cpu_value = cuda_config.cpu_value - return { - "cuda": _find_cuda_lib("cuda", repository_ctx, cpu_value, cuda_config.cuda_toolkit_path), - "cudart": _find_cuda_lib( - "cudart", repository_ctx, cpu_value, cuda_config.cuda_toolkit_path, - cuda_config.cuda_version), - "cudart_static": _find_cuda_lib( - "cudart_static", repository_ctx, cpu_value, - cuda_config.cuda_toolkit_path, cuda_config.cuda_version, static=True), - "cublas": _find_cuda_lib( - "cublas", repository_ctx, cpu_value, cuda_config.cuda_toolkit_path, - cuda_config.cuda_version), - "cusolver": _find_cuda_lib( - "cusolver", repository_ctx, cpu_value, cuda_config.cuda_toolkit_path, - cuda_config.cuda_version), - "curand": _find_cuda_lib( - "curand", repository_ctx, cpu_value, cuda_config.cuda_toolkit_path, - cuda_config.cuda_version), - "cufft": _find_cuda_lib( - "cufft", repository_ctx, cpu_value, cuda_config.cuda_toolkit_path, - cuda_config.cuda_version), - "cudnn": _find_cuda_lib( - "cudnn", repository_ctx, cpu_value, cuda_config.cudnn_install_basedir, - cuda_config.cudnn_version), - "cupti": _find_cupti_lib(repository_ctx, cuda_config) - } + auto_configure_fail("Cannot find cupti library %s" % file_name) +def _find_libs(repository_ctx, cuda_config): + """Returns the CUDA and cuDNN libraries on the system. + + Args: + repository_ctx: The repository context. + cuda_config: The CUDA config as returned by _get_cuda_config + + Returns: + Map of library names to structs of filename and path. + """ + cpu_value = cuda_config.cpu_value + return { + "cuda": _find_cuda_lib("cuda", repository_ctx, cpu_value, cuda_config.cuda_toolkit_path), + "cudart": _find_cuda_lib( + "cudart", + repository_ctx, + cpu_value, + cuda_config.cuda_toolkit_path, + cuda_config.cuda_version, + ), + "cudart_static": _find_cuda_lib( + "cudart_static", + repository_ctx, + cpu_value, + cuda_config.cuda_toolkit_path, + cuda_config.cuda_version, + static = True, + ), + "cublas": _find_cuda_lib( + "cublas", + repository_ctx, + cpu_value, + cuda_config.cuda_toolkit_path, + cuda_config.cuda_version, + ), + "cusolver": _find_cuda_lib( + "cusolver", + repository_ctx, + cpu_value, + cuda_config.cuda_toolkit_path, + cuda_config.cuda_version, + ), + "curand": _find_cuda_lib( + "curand", + repository_ctx, + cpu_value, + cuda_config.cuda_toolkit_path, + cuda_config.cuda_version, + ), + "cufft": _find_cuda_lib( + "cufft", + repository_ctx, + cpu_value, + cuda_config.cuda_toolkit_path, + cuda_config.cuda_version, + ), + "cudnn": _find_cuda_lib( + "cudnn", + repository_ctx, + cpu_value, + cuda_config.cudnn_install_basedir, + cuda_config.cudnn_version, + ), + "cupti": _find_cupti_lib(repository_ctx, cuda_config), + } def _find_cuda_include_path(repository_ctx, cuda_config): - """Returns the path to the directory containing cuda.h - - Args: - repository_ctx: The repository context. - cuda_config: The CUDA config as returned by _get_cuda_config + """Returns the path to the directory containing cuda.h - Returns: - The path of the directory containing the CUDA headers. - """ - cuda_toolkit_path = cuda_config.cuda_toolkit_path - for relative_path in CUDA_INCLUDE_PATHS: - if repository_ctx.path("%s/%scuda.h" % (cuda_toolkit_path, relative_path)).exists: - return ("%s/%s" % (cuda_toolkit_path, relative_path))[:-1] - auto_configure_fail("Cannot find cuda.h under %s" % cuda_toolkit_path) + Args: + repository_ctx: The repository context. + cuda_config: The CUDA config as returned by _get_cuda_config + Returns: + The path of the directory containing the CUDA headers. + """ + cuda_toolkit_path = cuda_config.cuda_toolkit_path + for relative_path in CUDA_INCLUDE_PATHS: + if repository_ctx.path("%s/%scuda.h" % (cuda_toolkit_path, relative_path)).exists: + return ("%s/%s" % (cuda_toolkit_path, relative_path))[:-1] + auto_configure_fail("Cannot find cuda.h under %s" % cuda_toolkit_path) def _find_cudnn_header_dir(repository_ctx, cudnn_install_basedir): - """Returns the path to the directory containing cudnn.h - - Args: - repository_ctx: The repository context. - cudnn_install_basedir: The cudnn install directory as returned by - _cudnn_install_basedir. - - Returns: - The path of the directory containing the cudnn header. - """ - for relative_path in CUDA_INCLUDE_PATHS: - if repository_ctx.path("%s/%scudnn.h" % (cudnn_install_basedir, relative_path)).exists: - return ("%s/%s" % (cudnn_install_basedir, relative_path))[:-1] - if repository_ctx.path("/usr/include/cudnn.h").exists: - return "/usr/include" - auto_configure_fail("Cannot find cudnn.h under %s" % cudnn_install_basedir) - + """Returns the path to the directory containing cudnn.h + + Args: + repository_ctx: The repository context. + cudnn_install_basedir: The cudnn install directory as returned by + _cudnn_install_basedir. + + Returns: + The path of the directory containing the cudnn header. + """ + for relative_path in CUDA_INCLUDE_PATHS: + if repository_ctx.path("%s/%scudnn.h" % (cudnn_install_basedir, relative_path)).exists: + return ("%s/%s" % (cudnn_install_basedir, relative_path))[:-1] + if repository_ctx.path("/usr/include/cudnn.h").exists: + return "/usr/include" + auto_configure_fail("Cannot find cudnn.h under %s" % cudnn_install_basedir) def _find_nvvm_libdevice_dir(repository_ctx, cuda_config): - """Returns the path to the directory containing libdevice in bitcode format. - - Args: - repository_ctx: The repository context. - cuda_config: The CUDA config as returned by _get_cuda_config - - Returns: - The path of the directory containing the CUDA headers. - """ - cuda_toolkit_path = cuda_config.cuda_toolkit_path - for relative_path in NVVM_LIBDEVICE_PATHS: - if repository_ctx.path("%s/%slibdevice.10.bc" % (cuda_toolkit_path, relative_path)).exists: - return ("%s/%s" % (cuda_toolkit_path, relative_path))[:-1] - auto_configure_fail("Cannot find libdevice.10.bc under %s" % cuda_toolkit_path) - + """Returns the path to the directory containing libdevice in bitcode format. + + Args: + repository_ctx: The repository context. + cuda_config: The CUDA config as returned by _get_cuda_config + + Returns: + The path of the directory containing the CUDA headers. + """ + cuda_toolkit_path = cuda_config.cuda_toolkit_path + for libdevice_file in NVVM_LIBDEVICE_FILES: + for relative_path in NVVM_LIBDEVICE_PATHS: + if repository_ctx.path("%s/%s%s" % (cuda_toolkit_path, relative_path, libdevice_file)).exists: + return ("%s/%s" % (cuda_toolkit_path, relative_path))[:-1] + auto_configure_fail("Cannot find libdevice*.bc files under %s" % cuda_toolkit_path) def _cudart_static_linkopt(cpu_value): - """Returns additional platform-specific linkopts for cudart.""" - return "" if cpu_value == "Darwin" else "\"-lrt\"," + """Returns additional platform-specific linkopts for cudart.""" + return "" if cpu_value == "Darwin" else "\"-lrt\"," def _get_cuda_config(repository_ctx): - """Detects and returns information about the CUDA installation on the system. - - Args: - repository_ctx: The repository context. - - Returns: - A struct containing the following fields: - cuda_toolkit_path: The CUDA toolkit installation directory. - cudnn_install_basedir: The cuDNN installation directory. - cuda_version: The version of CUDA on the system. - cudnn_version: The version of cuDNN on the system. - compute_capabilities: A list of the system's CUDA compute capabilities. - cpu_value: The name of the host operating system. - """ - cpu_value = get_cpu_value(repository_ctx) - cuda_toolkit_path = _cuda_toolkit_path(repository_ctx) - cuda_version = _cuda_version(repository_ctx, cuda_toolkit_path, cpu_value) - cudnn_install_basedir = _cudnn_install_basedir(repository_ctx) - cudnn_version = _cudnn_version(repository_ctx, cudnn_install_basedir, cpu_value) - return struct( - cuda_toolkit_path = cuda_toolkit_path, - cudnn_install_basedir = cudnn_install_basedir, - cuda_version = cuda_version, - cudnn_version = cudnn_version, - compute_capabilities = _compute_capabilities(repository_ctx), - cpu_value = cpu_value) - - -def _tpl(repository_ctx, tpl, substitutions={}, out=None): - if not out: - out = tpl.replace(":", "/") - repository_ctx.template( - out, - Label("//third_party/gpus/%s.tpl" % tpl), - substitutions) - + """Detects and returns information about the CUDA installation on the system. + + Args: + repository_ctx: The repository context. + + Returns: + A struct containing the following fields: + cuda_toolkit_path: The CUDA toolkit installation directory. + cudnn_install_basedir: The cuDNN installation directory. + cuda_version: The version of CUDA on the system. + cudnn_version: The version of cuDNN on the system. + compute_capabilities: A list of the system's CUDA compute capabilities. + cpu_value: The name of the host operating system. + """ + cpu_value = get_cpu_value(repository_ctx) + cuda_toolkit_path = _cuda_toolkit_path(repository_ctx) + cuda_version = _cuda_version(repository_ctx, cuda_toolkit_path, cpu_value) + cudnn_install_basedir = _cudnn_install_basedir(repository_ctx) + cudnn_version = _cudnn_version(repository_ctx, cudnn_install_basedir, cpu_value) + return struct( + cuda_toolkit_path = cuda_toolkit_path, + cudnn_install_basedir = cudnn_install_basedir, + cuda_version = cuda_version, + cudnn_version = cudnn_version, + compute_capabilities = _compute_capabilities(repository_ctx), + cpu_value = cpu_value, + ) + +def _tpl(repository_ctx, tpl, substitutions = {}, out = None): + if not out: + out = tpl.replace(":", "/") + repository_ctx.template( + out, + Label("//third_party/gpus/%s.tpl" % tpl), + substitutions, + ) def _file(repository_ctx, label): - repository_ctx.template( - label.replace(":", "/"), - Label("//third_party/gpus/%s.tpl" % label), - {}) - + repository_ctx.template( + label.replace(":", "/"), + Label("//third_party/gpus/%s.tpl" % label), + {}, + ) _DUMMY_CROSSTOOL_BZL_FILE = """ def error_gpu_disabled(): @@ -792,355 +951,511 @@ def error_gpu_disabled(): ) """ - _DUMMY_CROSSTOOL_BUILD_FILE = """ load("//crosstool:error_gpu_disabled.bzl", "error_gpu_disabled") error_gpu_disabled() """ - def _create_dummy_repository(repository_ctx): - cpu_value = get_cpu_value(repository_ctx) - - # Set up BUILD file for cuda/. - _tpl(repository_ctx, "cuda:build_defs.bzl", - { - "%{cuda_is_configured}": "False", - "%{cuda_extra_copts}": "[]", - }) - _tpl(repository_ctx, "cuda:BUILD", - { - "%{cuda_driver_lib}": _lib_name("cuda", cpu_value), - "%{cudart_static_lib}": _lib_name("cudart_static", cpu_value, - static=True), - "%{cudart_static_linkopt}": _cudart_static_linkopt(cpu_value), - "%{cudart_lib}": _lib_name("cudart", cpu_value), - "%{cublas_lib}": _lib_name("cublas", cpu_value), - "%{cusolver_lib}": _lib_name("cusolver", cpu_value), - "%{cudnn_lib}": _lib_name("cudnn", cpu_value), - "%{cufft_lib}": _lib_name("cufft", cpu_value), - "%{curand_lib}": _lib_name("curand", cpu_value), - "%{cupti_lib}": _lib_name("cupti", cpu_value), - "%{cuda_include_genrules}": '', - "%{cuda_headers}": '', - }) - - # Create dummy files for the CUDA toolkit since they are still required by - # tensorflow/core/platform/default/build_config:cuda. - repository_ctx.file("cuda/cuda/include/cuda.h", "") - repository_ctx.file("cuda/cuda/include/cublas.h", "") - repository_ctx.file("cuda/cuda/include/cudnn.h", "") - repository_ctx.file("cuda/cuda/extras/CUPTI/include/cupti.h", "") - repository_ctx.file("cuda/cuda/lib/%s" % _lib_name("cuda", cpu_value)) - repository_ctx.file("cuda/cuda/lib/%s" % _lib_name("cudart", cpu_value)) - repository_ctx.file("cuda/cuda/lib/%s" % _lib_name("cudart_static", cpu_value)) - repository_ctx.file("cuda/cuda/lib/%s" % _lib_name("cublas", cpu_value)) - repository_ctx.file("cuda/cuda/lib/%s" % _lib_name("cusolver", cpu_value)) - repository_ctx.file("cuda/cuda/lib/%s" % _lib_name("cudnn", cpu_value)) - repository_ctx.file("cuda/cuda/lib/%s" % _lib_name("curand", cpu_value)) - repository_ctx.file("cuda/cuda/lib/%s" % _lib_name("cufft", cpu_value)) - repository_ctx.file("cuda/cuda/lib/%s" % _lib_name("cupti", cpu_value)) - - # Set up cuda_config.h, which is used by - # tensorflow/stream_executor/dso_loader.cc. - _tpl(repository_ctx, "cuda:cuda_config.h", - { - "%{cuda_version}": _DEFAULT_CUDA_VERSION, - "%{cudnn_version}": _DEFAULT_CUDNN_VERSION, - "%{cuda_compute_capabilities}": ",".join([ - "CudaVersion(\"%s\")" % c - for c in _DEFAULT_CUDA_COMPUTE_CAPABILITIES]), - "%{cuda_toolkit_path}": _DEFAULT_CUDA_TOOLKIT_PATH, - }, "cuda/cuda/cuda_config.h") - - # If cuda_configure is not configured to build with GPU support, and the user - # attempts to build with --config=cuda, add a dummy build rule to intercept - # this and fail with an actionable error message. - repository_ctx.file("crosstool/error_gpu_disabled.bzl", - _DUMMY_CROSSTOOL_BZL_FILE) - repository_ctx.file("crosstool/BUILD", _DUMMY_CROSSTOOL_BUILD_FILE) - - -def _execute(repository_ctx, cmdline, error_msg=None, error_details=None, - empty_stdout_fine=False): - """Executes an arbitrary shell command. - - Args: - repository_ctx: the repository_ctx object - cmdline: list of strings, the command to execute - error_msg: string, a summary of the error if the command fails - error_details: string, details about the error or steps to fix it - empty_stdout_fine: bool, if True, an empty stdout result is fine, otherwise - it's an error - Return: - the result of repository_ctx.execute(cmdline) - """ - result = repository_ctx.execute(cmdline) - if result.stderr or not (empty_stdout_fine or result.stdout): - auto_configure_fail( - "\n".join([ - error_msg.strip() if error_msg else "Repository command failed", - result.stderr.strip(), - error_details if error_details else ""])) - return result - + cpu_value = get_cpu_value(repository_ctx) + + # Set up BUILD file for cuda/. + _tpl( + repository_ctx, + "cuda:build_defs.bzl", + { + "%{cuda_is_configured}": "False", + "%{cuda_extra_copts}": "[]", + }, + ) + _tpl( + repository_ctx, + "cuda:BUILD", + { + "%{cuda_driver_lib}": _lib_name("cuda", cpu_value), + "%{cudart_static_lib}": _lib_name( + "cudart_static", + cpu_value, + static = True, + ), + "%{cudart_static_linkopt}": _cudart_static_linkopt(cpu_value), + "%{cudart_lib}": _lib_name("cudart", cpu_value), + "%{cublas_lib}": _lib_name("cublas", cpu_value), + "%{cusolver_lib}": _lib_name("cusolver", cpu_value), + "%{cudnn_lib}": _lib_name("cudnn", cpu_value), + "%{cufft_lib}": _lib_name("cufft", cpu_value), + "%{curand_lib}": _lib_name("curand", cpu_value), + "%{cupti_lib}": _lib_name("cupti", cpu_value), + "%{cuda_include_genrules}": "", + "%{cuda_headers}": "", + }, + ) + + # Create dummy files for the CUDA toolkit since they are still required by + # tensorflow/core/platform/default/build_config:cuda. + repository_ctx.file("cuda/cuda/include/cuda.h", "") + repository_ctx.file("cuda/cuda/include/cublas.h", "") + repository_ctx.file("cuda/cuda/include/cudnn.h", "") + repository_ctx.file("cuda/cuda/extras/CUPTI/include/cupti.h", "") + repository_ctx.file("cuda/cuda/lib/%s" % _lib_name("cuda", cpu_value)) + repository_ctx.file("cuda/cuda/lib/%s" % _lib_name("cudart", cpu_value)) + repository_ctx.file("cuda/cuda/lib/%s" % _lib_name("cudart_static", cpu_value)) + repository_ctx.file("cuda/cuda/lib/%s" % _lib_name("cublas", cpu_value)) + repository_ctx.file("cuda/cuda/lib/%s" % _lib_name("cusolver", cpu_value)) + repository_ctx.file("cuda/cuda/lib/%s" % _lib_name("cudnn", cpu_value)) + repository_ctx.file("cuda/cuda/lib/%s" % _lib_name("curand", cpu_value)) + repository_ctx.file("cuda/cuda/lib/%s" % _lib_name("cufft", cpu_value)) + repository_ctx.file("cuda/cuda/lib/%s" % _lib_name("cupti", cpu_value)) + + # Set up cuda_config.h, which is used by + # tensorflow/stream_executor/dso_loader.cc. + _tpl( + repository_ctx, + "cuda:cuda_config.h", + { + "%{cuda_version}": _DEFAULT_CUDA_VERSION, + "%{cudnn_version}": _DEFAULT_CUDNN_VERSION, + "%{cuda_compute_capabilities}": ",".join([ + "CudaVersion(\"%s\")" % c + for c in _DEFAULT_CUDA_COMPUTE_CAPABILITIES + ]), + "%{cuda_toolkit_path}": _DEFAULT_CUDA_TOOLKIT_PATH, + }, + "cuda/cuda/cuda_config.h", + ) + + # If cuda_configure is not configured to build with GPU support, and the user + # attempts to build with --config=cuda, add a dummy build rule to intercept + # this and fail with an actionable error message. + repository_ctx.file( + "crosstool/error_gpu_disabled.bzl", + _DUMMY_CROSSTOOL_BZL_FILE, + ) + repository_ctx.file("crosstool/BUILD", _DUMMY_CROSSTOOL_BUILD_FILE) + +def _execute( + repository_ctx, + cmdline, + error_msg = None, + error_details = None, + empty_stdout_fine = False): + """Executes an arbitrary shell command. + + Args: + repository_ctx: the repository_ctx object + cmdline: list of strings, the command to execute + error_msg: string, a summary of the error if the command fails + error_details: string, details about the error or steps to fix it + empty_stdout_fine: bool, if True, an empty stdout result is fine, otherwise + it's an error + Return: + the result of repository_ctx.execute(cmdline) + """ + result = repository_ctx.execute(cmdline) + if result.stderr or not (empty_stdout_fine or result.stdout): + auto_configure_fail( + "\n".join([ + error_msg.strip() if error_msg else "Repository command failed", + result.stderr.strip(), + error_details if error_details else "", + ]), + ) + return result def _norm_path(path): - """Returns a path with '/' and remove the trailing slash.""" - path = path.replace("\\", "/") - if path[-1] == "/": - path = path[:-1] - return path - - -def symlink_genrule_for_dir(repository_ctx, src_dir, dest_dir, genrule_name, - src_files = [], dest_files = []): - """Returns a genrule to symlink(or copy if on Windows) a set of files. - - If src_dir is passed, files will be read from the given directory; otherwise - we assume files are in src_files and dest_files - """ - if src_dir != None: - src_dir = _norm_path(src_dir) - dest_dir = _norm_path(dest_dir) - files = '\n'.join(sorted(_read_dir(repository_ctx, src_dir).splitlines())) - # Create a list with the src_dir stripped to use for outputs. - dest_files = files.replace(src_dir, '').splitlines() - src_files = files.splitlines() - command = [] - if not _is_windows(repository_ctx): - # We clear folders that might have been generated previously to avoid - # undesired inclusions - command.append('if [ -d "$(@D)/extras" ]; then rm $(@D)/extras -drf; fi') - command.append('if [ -d "$(@D)/include" ]; then rm $(@D)/include -drf; fi') - command.append('if [ -d "$(@D)/lib" ]; then rm $(@D)/lib -drf; fi') - command.append('if [ -d "$(@D)/nvvm" ]; then rm $(@D)/nvvm -drf; fi') - outs = [] - for i in range(len(dest_files)): - if dest_files[i] != "": - # If we have only one file to link we do not want to use the dest_dir, as - # $(@D) will include the full path to the file. - dest = '$(@D)/' + dest_dir + dest_files[i] if len(dest_files) != 1 else '$(@D)/' + dest_files[i] - # On Windows, symlink is not supported, so we just copy all the files. - cmd = 'cp -f' if _is_windows(repository_ctx) else 'ln -s' - command.append(cmd + ' "%s" "%s"' % (src_files[i] , dest)) - outs.append(' "' + dest_dir + dest_files[i] + '",') - genrule = _genrule(src_dir, genrule_name, " && ".join(command), - "\n".join(outs)) - return genrule - + """Returns a path with '/' and remove the trailing slash.""" + path = path.replace("\\", "/") + if path[-1] == "/": + path = path[:-1] + return path + +def symlink_genrule_for_dir( + repository_ctx, + src_dir, + dest_dir, + genrule_name, + src_files = [], + dest_files = []): + """Returns a genrule to symlink(or copy if on Windows) a set of files. + + If src_dir is passed, files will be read from the given directory; otherwise + we assume files are in src_files and dest_files + """ + if src_dir != None: + src_dir = _norm_path(src_dir) + dest_dir = _norm_path(dest_dir) + files = "\n".join(sorted(_read_dir(repository_ctx, src_dir).splitlines())) + + # Create a list with the src_dir stripped to use for outputs. + dest_files = files.replace(src_dir, "").splitlines() + src_files = files.splitlines() + command = [] + if not _is_windows(repository_ctx): + # We clear folders that might have been generated previously to avoid + # undesired inclusions + command.append('if [ -d "$(@D)/extras" ]; then rm $(@D)/extras -drf; fi') + command.append('if [ -d "$(@D)/include" ]; then rm $(@D)/include -drf; fi') + command.append('if [ -d "$(@D)/lib" ]; then rm $(@D)/lib -drf; fi') + command.append('if [ -d "$(@D)/nvvm" ]; then rm $(@D)/nvvm -drf; fi') + outs = [] + for i in range(len(dest_files)): + if dest_files[i] != "": + # If we have only one file to link we do not want to use the dest_dir, as + # $(@D) will include the full path to the file. + dest = "$(@D)/" + dest_dir + dest_files[i] if len(dest_files) != 1 else "$(@D)/" + dest_files[i] + + # On Windows, symlink is not supported, so we just copy all the files. + cmd = "cp -f" if _is_windows(repository_ctx) else "ln -s" + command.append(cmd + ' "%s" "%s"' % (src_files[i], dest)) + outs.append(' "' + dest_dir + dest_files[i] + '",') + genrule = _genrule( + src_dir, + genrule_name, + " && ".join(command), + "\n".join(outs), + ) + return genrule def _genrule(src_dir, genrule_name, command, outs): - """Returns a string with a genrule. - - Genrule executes the given command and produces the given outputs. - """ - return ( - 'genrule(\n' + - ' name = "' + - genrule_name + '",\n' + - ' outs = [\n' + - outs + - '\n ],\n' + - ' cmd = """\n' + - command + - '\n """,\n' + - ')\n' - ) - + """Returns a string with a genrule. + + Genrule executes the given command and produces the given outputs. + """ + return ( + "genrule(\n" + + ' name = "' + + genrule_name + '",\n' + + " outs = [\n" + + outs + + "\n ],\n" + + ' cmd = """\n' + + command + + '\n """,\n' + + ")\n" + ) def _read_dir(repository_ctx, src_dir): - """Returns a string with all files in a directory. - - Finds all files inside a directory, traversing subfolders and following - symlinks. The returned string contains the full path of all files - separated by line breaks. - """ - if _is_windows(repository_ctx): - src_dir = src_dir.replace("/", "\\") - find_result = _execute( - repository_ctx, ["cmd.exe", "/c", "dir", src_dir, "/b", "/s", "/a-d"], - empty_stdout_fine=True) - # src_files will be used in genrule.outs where the paths must - # use forward slashes. - result = find_result.stdout.replace("\\", "/") - else: - find_result = _execute( - repository_ctx, ["find", src_dir, "-follow", "-type", "f"], - empty_stdout_fine=True) - result = find_result.stdout - return result + """Returns a string with all files in a directory. + + Finds all files inside a directory, traversing subfolders and following + symlinks. The returned string contains the full path of all files + separated by line breaks. + """ + if _is_windows(repository_ctx): + src_dir = src_dir.replace("/", "\\") + find_result = _execute( + repository_ctx, + ["cmd.exe", "/c", "dir", src_dir, "/b", "/s", "/a-d"], + empty_stdout_fine = True, + ) + + # src_files will be used in genrule.outs where the paths must + # use forward slashes. + result = find_result.stdout.replace("\\", "/") + else: + find_result = _execute( + repository_ctx, + ["find", src_dir, "-follow", "-type", "f"], + empty_stdout_fine = True, + ) + result = find_result.stdout + return result def _flag_enabled(repository_ctx, flag_name): - if flag_name in repository_ctx.os.environ: - value = repository_ctx.os.environ[flag_name].strip() - return value == "1" - return False + if flag_name in repository_ctx.os.environ: + value = repository_ctx.os.environ[flag_name].strip() + return value == "1" + return False def _use_cuda_clang(repository_ctx): - return _flag_enabled(repository_ctx, "TF_CUDA_CLANG") + return _flag_enabled(repository_ctx, "TF_CUDA_CLANG") def _compute_cuda_extra_copts(repository_ctx, compute_capabilities): - if _use_cuda_clang(repository_ctx): - capability_flags = ["--cuda-gpu-arch=sm_" + - cap.replace(".", "") for cap in compute_capabilities] - else: - # Capabilities are handled in the "crosstool_wrapper_driver_is_not_gcc" for nvcc - capability_flags = [] - return str(capability_flags) + if _use_cuda_clang(repository_ctx): + capability_flags = ["--cuda-gpu-arch=sm_" + + cap.replace(".", "") for cap in compute_capabilities] + else: + # Capabilities are handled in the "crosstool_wrapper_driver_is_not_gcc" for nvcc + capability_flags = [] + return str(capability_flags) def _create_local_cuda_repository(repository_ctx): - """Creates the repository containing files set up to build with CUDA.""" - cuda_config = _get_cuda_config(repository_ctx) - - cuda_include_path = _find_cuda_include_path(repository_ctx, cuda_config) - cudnn_header_dir = _find_cudnn_header_dir(repository_ctx, - cuda_config.cudnn_install_basedir) - cupti_header_dir = _find_cupti_header_dir(repository_ctx, cuda_config) - nvvm_libdevice_dir = _find_nvvm_libdevice_dir(repository_ctx, cuda_config) - - # Set up symbolic links for the cuda toolkit by creating genrules to do - # symlinking. We create one genrule for each directory we want to track under - # cuda_toolkit_path - cuda_toolkit_path = cuda_config.cuda_toolkit_path - genrules = [symlink_genrule_for_dir(repository_ctx, - cuda_include_path, "cuda/include", "cuda-include")] - genrules.append(symlink_genrule_for_dir(repository_ctx, - nvvm_libdevice_dir, "cuda/nvvm/libdevice", "cuda-nvvm")) - genrules.append(symlink_genrule_for_dir(repository_ctx, - cupti_header_dir, "cuda/extras/CUPTI/include", "cuda-extras")) - - cuda_libs = _find_libs(repository_ctx, cuda_config) - cuda_lib_src = [] - cuda_lib_dest = [] - for lib in cuda_libs.values(): - cuda_lib_src.append(lib.path) - cuda_lib_dest.append("cuda/lib/" + lib.file_name) - genrules.append(symlink_genrule_for_dir(repository_ctx, None, "", "cuda-lib", - cuda_lib_src, cuda_lib_dest)) - - # Set up the symbolic links for cudnn if cndnn was not installed to - # CUDA_TOOLKIT_PATH. - included_files = _read_dir(repository_ctx, cuda_include_path).replace( - cuda_include_path, '').splitlines() - if '/cudnn.h' not in included_files: - genrules.append(symlink_genrule_for_dir(repository_ctx, None, - "cuda/include/", "cudnn-include", [cudnn_header_dir + "/cudnn.h"], - ["cudnn.h"])) - else: - genrules.append( - 'filegroup(\n' + + """Creates the repository containing files set up to build with CUDA.""" + cuda_config = _get_cuda_config(repository_ctx) + + cuda_include_path = _find_cuda_include_path(repository_ctx, cuda_config) + cudnn_header_dir = _find_cudnn_header_dir( + repository_ctx, + cuda_config.cudnn_install_basedir, + ) + cupti_header_dir = _find_cupti_header_dir(repository_ctx, cuda_config) + nvvm_libdevice_dir = _find_nvvm_libdevice_dir(repository_ctx, cuda_config) + + # Set up symbolic links for the cuda toolkit by creating genrules to do + # symlinking. We create one genrule for each directory we want to track under + # cuda_toolkit_path + cuda_toolkit_path = cuda_config.cuda_toolkit_path + genrules = [symlink_genrule_for_dir( + repository_ctx, + cuda_include_path, + "cuda/include", + "cuda-include", + )] + genrules.append(symlink_genrule_for_dir( + repository_ctx, + nvvm_libdevice_dir, + "cuda/nvvm/libdevice", + "cuda-nvvm", + )) + genrules.append(symlink_genrule_for_dir( + repository_ctx, + cupti_header_dir, + "cuda/extras/CUPTI/include", + "cuda-extras", + )) + + cuda_libs = _find_libs(repository_ctx, cuda_config) + cuda_lib_src = [] + cuda_lib_dest = [] + for lib in cuda_libs.values(): + cuda_lib_src.append(lib.path) + cuda_lib_dest.append("cuda/lib/" + lib.file_name) + genrules.append(symlink_genrule_for_dir( + repository_ctx, + None, + "", + "cuda-lib", + cuda_lib_src, + cuda_lib_dest, + )) + + # Set up the symbolic links for cudnn if cndnn was not installed to + # CUDA_TOOLKIT_PATH. + included_files = _read_dir(repository_ctx, cuda_include_path).replace( + cuda_include_path, + "", + ).splitlines() + if "/cudnn.h" not in included_files: + genrules.append(symlink_genrule_for_dir( + repository_ctx, + None, + "cuda/include/", + "cudnn-include", + [cudnn_header_dir + "/cudnn.h"], + ["cudnn.h"], + )) + else: + genrules.append( + "filegroup(\n" + ' name = "cudnn-include",\n' + - ' srcs = [],\n' + - ')\n' + " srcs = [],\n" + + ")\n", ) - # Set up BUILD file for cuda/ - _tpl(repository_ctx, "cuda:build_defs.bzl", - { - "%{cuda_is_configured}": "True", - "%{cuda_extra_copts}": _compute_cuda_extra_copts( - repository_ctx, cuda_config.compute_capabilities), - }) - _tpl(repository_ctx, "cuda:BUILD", - { - "%{cuda_driver_lib}": cuda_libs["cuda"].file_name, - "%{cudart_static_lib}": cuda_libs["cudart_static"].file_name, - "%{cudart_static_linkopt}": _cudart_static_linkopt( - cuda_config.cpu_value), - "%{cudart_lib}": cuda_libs["cudart"].file_name, - "%{cublas_lib}": cuda_libs["cublas"].file_name, - "%{cusolver_lib}": cuda_libs["cusolver"].file_name, - "%{cudnn_lib}": cuda_libs["cudnn"].file_name, - "%{cufft_lib}": cuda_libs["cufft"].file_name, - "%{curand_lib}": cuda_libs["curand"].file_name, - "%{cupti_lib}": cuda_libs["cupti"].file_name, - "%{cuda_include_genrules}": "\n".join(genrules), - "%{cuda_headers}": ('":cuda-include",\n' + - ' ":cudnn-include",') - }) - - is_cuda_clang = _use_cuda_clang(repository_ctx) - - should_download_clang = is_cuda_clang and _flag_enabled( - repository_ctx, _TF_DOWNLOAD_CLANG) - if should_download_clang: - download_clang(repository_ctx, "crosstool/extra_tools") - - # Set up crosstool/ - cc = find_cc(repository_ctx) - cc_fullpath = cc if not should_download_clang else "crosstool/" + cc - - host_compiler_includes = _host_compiler_includes(repository_ctx, cc_fullpath) - cuda_defines = { - "%{cuda_include_path}": _cuda_include_path(repository_ctx, - cuda_config), - "%{host_compiler_includes}": host_compiler_includes, - } - if is_cuda_clang: - cuda_defines["%{clang_path}"] = cc - _tpl(repository_ctx, "crosstool:BUILD", {"%{linker_files}": ":empty"}) - _tpl(repository_ctx, "crosstool:CROSSTOOL_clang", cuda_defines, out="crosstool/CROSSTOOL") - repository_ctx.file("crosstool/clang/bin/crosstool_wrapper_driver_is_not_gcc", "") - else: - nvcc_path = str(repository_ctx.path("%s/bin/nvcc%s" % - (cuda_config.cuda_toolkit_path, - ".exe" if cuda_config.cpu_value == "Windows" else ""))) - _tpl(repository_ctx, "crosstool:BUILD", - {"%{linker_files}": ":crosstool_wrapper_driver_is_not_gcc"}) - _tpl(repository_ctx, "crosstool:CROSSTOOL_nvcc", cuda_defines, out="crosstool/CROSSTOOL") - _tpl(repository_ctx, - "crosstool:clang/bin/crosstool_wrapper_driver_is_not_gcc", - { - "%{cpu_compiler}": str(cc), - "%{cuda_version}": cuda_config.cuda_version, - "%{nvcc_path}": nvcc_path, - "%{gcc_host_compiler_path}": str(cc), - "%{cuda_compute_capabilities}": ", ".join( - ["\"%s\"" % c for c in cuda_config.compute_capabilities]), - }) - - # Set up cuda_config.h, which is used by - # tensorflow/stream_executor/dso_loader.cc. - _tpl(repository_ctx, "cuda:cuda_config.h", - { - "%{cuda_version}": cuda_config.cuda_version, - "%{cudnn_version}": cuda_config.cudnn_version, - "%{cuda_compute_capabilities}": ",".join( - ["CudaVersion(\"%s\")" % c - for c in cuda_config.compute_capabilities]), - "%{cuda_toolkit_path}": cuda_config.cuda_toolkit_path, - }, "cuda/cuda/cuda_config.h") + # Set up BUILD file for cuda/ + _tpl( + repository_ctx, + "cuda:build_defs.bzl", + { + "%{cuda_is_configured}": "True", + "%{cuda_extra_copts}": _compute_cuda_extra_copts( + repository_ctx, + cuda_config.compute_capabilities, + ), + }, + ) + _tpl( + repository_ctx, + "cuda:BUILD.windows" if _is_windows(repository_ctx) else "cuda:BUILD", + { + "%{cuda_driver_lib}": cuda_libs["cuda"].file_name, + "%{cudart_static_lib}": cuda_libs["cudart_static"].file_name, + "%{cudart_static_linkopt}": _cudart_static_linkopt( + cuda_config.cpu_value, + ), + "%{cudart_lib}": cuda_libs["cudart"].file_name, + "%{cublas_lib}": cuda_libs["cublas"].file_name, + "%{cusolver_lib}": cuda_libs["cusolver"].file_name, + "%{cudnn_lib}": cuda_libs["cudnn"].file_name, + "%{cufft_lib}": cuda_libs["cufft"].file_name, + "%{curand_lib}": cuda_libs["curand"].file_name, + "%{cupti_lib}": cuda_libs["cupti"].file_name, + "%{cuda_include_genrules}": "\n".join(genrules), + "%{cuda_headers}": ('":cuda-include",\n' + + ' ":cudnn-include",'), + }, + "cuda/BUILD", + ) + + is_cuda_clang = _use_cuda_clang(repository_ctx) + + should_download_clang = is_cuda_clang and _flag_enabled( + repository_ctx, + _TF_DOWNLOAD_CLANG, + ) + if should_download_clang: + download_clang(repository_ctx, "crosstool/extra_tools") + + # Set up crosstool/ + cc = find_cc(repository_ctx) + cc_fullpath = cc if not should_download_clang else "crosstool/" + cc + + host_compiler_includes = _host_compiler_includes(repository_ctx, cc_fullpath) + cuda_defines = {} + # Bazel sets '-B/usr/bin' flag to workaround build errors on RHEL (see + # https://github.com/bazelbuild/bazel/issues/760). + # However, this stops our custom clang toolchain from picking the provided + # LLD linker, so we're only adding '-B/usr/bin' when using non-downloaded + # toolchain. + # TODO: when bazel stops adding '-B/usr/bin' by default, remove this + # flag from the CROSSTOOL completely (see + # https://github.com/bazelbuild/bazel/issues/5634) + if should_download_clang: + cuda_defines["%{linker_bin_path_flag}"] = "" + else: + cuda_defines["%{linker_bin_path_flag}"] = 'flag: "-B/usr/bin"' + + if is_cuda_clang: + cuda_defines["%{host_compiler_path}"] = str(cc) + cuda_defines["%{host_compiler_warnings}"] = """ + # Some parts of the codebase set -Werror and hit this warning, so + # switch it off for now. + flag: "-Wno-invalid-partial-specialization" + """ + cuda_defines["%{host_compiler_includes}"] = host_compiler_includes + _tpl(repository_ctx, "crosstool:BUILD", {"%{linker_files}": ":empty", "%{win_linker_files}": ":empty"}) + repository_ctx.file("crosstool/clang/bin/crosstool_wrapper_driver_is_not_gcc", "") + repository_ctx.file("crosstool/windows/msvc_wrapper_for_nvcc.py", "") + repository_ctx.file("crosstool/windows/msvc_wrapper_for_nvcc.bat", "") + else: + cuda_defines["%{host_compiler_path}"] = "clang/bin/crosstool_wrapper_driver_is_not_gcc" + cuda_defines["%{host_compiler_warnings}"] = "" + + # TODO(klimek): We currently need to inject "/" as builtin directory path + # to disable bazel's dependency checks. + # The problem is that: + # - the python rules symlink the python headers into the bazel root + # - the rules use 'includes' in the BUILD file to redirect includes of the + # python headers through those paths + # - bazel currently uses -isystem for include paths specified via 'includes' + # - gcc follows symlinks when resolving files via -isystem paths, and puts + # the resolved paths into the .d file, which makes the dependency check + # fail for bazel + # There are multiple possible ways to solve this: + # 1. make bazel not use -isystem for paths specified via 'includes' + # 2. cp the headers instead of symlinking them + # + # Once this is fixed, the right builtin directory path is: + # (host_compiler_includes + + # "\n cxx_builtin_include_directory: \"%s\"" % cuda_include_path) + # The cuda directory needs to be passed, as there is currently no rule + # providing the cuda headers in the same way the python headers are + # provided. + cuda_defines["%{host_compiler_includes}"] = "\n cxx_builtin_include_directory: \"/\"" + nvcc_path = str(repository_ctx.path("%s/bin/nvcc%s" % + ( + cuda_config.cuda_toolkit_path, + ".exe" if _is_windows(repository_ctx) else "", + ))) + _tpl( + repository_ctx, + "crosstool:BUILD", + { + "%{linker_files}": ":crosstool_wrapper_driver_is_not_gcc", + "%{win_linker_files}": ":windows_msvc_wrapper_files", + }, + ) + wrapper_defines = { + "%{cpu_compiler}": str(cc), + "%{cuda_version}": cuda_config.cuda_version, + "%{nvcc_path}": nvcc_path, + "%{gcc_host_compiler_path}": str(cc), + "%{cuda_compute_capabilities}": ", ".join( + ["\"%s\"" % c for c in cuda_config.compute_capabilities], + ), + "%{nvcc_tmp_dir}": _get_nvcc_tmp_dir_for_windows(repository_ctx), + } + _tpl( + repository_ctx, + "crosstool:clang/bin/crosstool_wrapper_driver_is_not_gcc", + wrapper_defines, + ) + _tpl( + repository_ctx, + "crosstool:windows/msvc_wrapper_for_nvcc.py", + wrapper_defines, + ) + _tpl( + repository_ctx, + "crosstool:windows/msvc_wrapper_for_nvcc.bat", + { + "%{python_binary}": _get_python_bin(repository_ctx), + }, + ) + + _tpl( + repository_ctx, + "crosstool:CROSSTOOL", + cuda_defines + _get_win_cuda_defines(repository_ctx), + out = "crosstool/CROSSTOOL", + ) + + # Set up cuda_config.h, which is used by + # tensorflow/stream_executor/dso_loader.cc. + _tpl( + repository_ctx, + "cuda:cuda_config.h", + { + "%{cuda_version}": cuda_config.cuda_version, + "%{cudnn_version}": cuda_config.cudnn_version, + "%{cuda_compute_capabilities}": ",".join( + [ + "CudaVersion(\"%s\")" % c + for c in cuda_config.compute_capabilities + ], + ), + "%{cuda_toolkit_path}": cuda_config.cuda_toolkit_path, + }, + "cuda/cuda/cuda_config.h", + ) def _create_remote_cuda_repository(repository_ctx, remote_config_repo): - """Creates pointers to a remotely configured repo set up to build with CUDA.""" - _tpl(repository_ctx, "cuda:build_defs.bzl", - { - "%{cuda_is_configured}": "True", - "%{cuda_extra_copts}": _compute_cuda_extra_copts( - repository_ctx, _compute_capabilities(repository_ctx)), - - }) - _tpl(repository_ctx, "cuda:remote.BUILD", - { - "%{remote_cuda_repo}": remote_config_repo, - }, "cuda/BUILD") - _tpl(repository_ctx, "crosstool:remote.BUILD", { - "%{remote_cuda_repo}": remote_config_repo, - }, "crosstool/BUILD") + """Creates pointers to a remotely configured repo set up to build with CUDA.""" + _tpl( + repository_ctx, + "cuda:build_defs.bzl", + { + "%{cuda_is_configured}": "True", + "%{cuda_extra_copts}": _compute_cuda_extra_copts( + repository_ctx, + _compute_capabilities(repository_ctx), + ), + }, + ) + _tpl( + repository_ctx, + "cuda:remote.BUILD", + { + "%{remote_cuda_repo}": remote_config_repo, + }, + "cuda/BUILD", + ) + _tpl(repository_ctx, "crosstool:remote.BUILD", { + "%{remote_cuda_repo}": remote_config_repo, + }, "crosstool/BUILD") def _cuda_autoconf_impl(repository_ctx): - """Implementation of the cuda_autoconf repository rule.""" - if not _enable_cuda(repository_ctx): - _create_dummy_repository(repository_ctx) - else: - if _TF_CUDA_CONFIG_REPO in repository_ctx.os.environ: - _create_remote_cuda_repository(repository_ctx, - repository_ctx.os.environ[_TF_CUDA_CONFIG_REPO]) + """Implementation of the cuda_autoconf repository rule.""" + if not _enable_cuda(repository_ctx): + _create_dummy_repository(repository_ctx) + elif _TF_CUDA_CONFIG_REPO in repository_ctx.os.environ: + _create_remote_cuda_repository( + repository_ctx, + repository_ctx.os.environ[_TF_CUDA_CONFIG_REPO], + ) else: - _create_local_cuda_repository(repository_ctx) - + _create_local_cuda_repository(repository_ctx) cuda_configure = repository_rule( implementation = _cuda_autoconf_impl, @@ -1157,6 +1472,7 @@ cuda_configure = repository_rule( _TF_CUDA_COMPUTE_CAPABILITIES, _TF_CUDA_CONFIG_REPO, "NVVMIR_LIBRARY_DIR", + _PYTHON_BIN_PATH, ], ) diff --git a/third_party/gpus/download_clang.bzl b/third_party/gpus/download_clang.bzl deleted file mode 100644 index 54d383d7d7..0000000000 --- a/third_party/gpus/download_clang.bzl +++ /dev/null @@ -1,54 +0,0 @@ -""" Helpers to download a recent clang release.""" - -def _get_platform_folder(os_name): - os_name = os_name.lower() - if os_name.startswith('windows'): - return 'Win' - if os_name.startswith('mac os'): - return 'Mac' - if not os_name.startswith('linux'): - fail('Unknown platform') - return 'Linux_x64' - -def _download_chromium_clang(repo_ctx, platform_folder, package_version, sha256, - out_folder): - cds_url = 'https://commondatastorage.googleapis.com/chromium-browser-clang' - cds_file = 'clang-%s.tgz' % package_version - cds_full_url = '{0}/{1}/{2}'.format(cds_url, platform_folder, cds_file) - repo_ctx.download_and_extract(cds_full_url, output=out_folder, sha256=sha256) - -def download_clang(repo_ctx, out_folder): - """ Download a fresh clang release and put it into out_folder. - - Clang itself will be located in 'out_folder/bin/clang'. - We currently download one of the latest releases of clang by the - Chromium project (see - https://chromium.googlesource.com/chromium/src/+/master/docs/clang.md). - - Args: - repo_ctx: An instance of repository_context object. - out_folder: A folder to extract the compiler into. - """ - # TODO(ibiryukov): we currently download and extract some extra tools in the - # clang release (e.g., sanitizers). We should probably remove the ones - # we don't need and document the ones we want provide in addition to clang. - - # Latest CLANG_REVISION and CLANG_SUB_REVISION of the Chromiums's release - # can be found in https://chromium.googlesource.com/chromium/src/tools/clang/+/master/scripts/update.py - CLANG_REVISION = '321529' - CLANG_SUB_REVISION = 2 - - package_version = '%s-%s' % (CLANG_REVISION, CLANG_SUB_REVISION) - - checksums = { - 'Linux_x64': - '76d4eb1ad011e3127c4a9de9b9f5d4ac624b5a9395c4d7395c9e0a487b13daf6', - 'Mac': - '4b2a7a65ac1ee892b318c723eec8771f514bb306f346aa8216bb0006f19d87b7', - 'Win': - 'eba51bb8f84af41a85903113666bd21c22709010c39c4cb19dc20cf1ed14581b', - } - - platform_folder = _get_platform_folder(repo_ctx.os.name) - _download_chromium_clang(repo_ctx, platform_folder, package_version, - checksums[platform_folder], out_folder) diff --git a/third_party/hadoop/BUILD b/third_party/hadoop/BUILD index 9e98154400..c3c5e428be 100644 --- a/third_party/hadoop/BUILD +++ b/third_party/hadoop/BUILD @@ -4,18 +4,6 @@ licenses(["notice"]) # Apache 2.0 exports_files(["LICENSE.txt"]) -filegroup( - name = "all_files", - srcs = glob( - ["**/*"], - exclude = [ - "**/METADATA", - "**/OWNERS", - ], - ), - visibility = ["//tensorflow:__subpackages__"], -) - cc_library( name = "hdfs", hdrs = ["hdfs.h"], diff --git a/third_party/hadoop/hdfs.h b/third_party/hadoop/hdfs.h index a664f3b50c..30c277a450 100644 --- a/third_party/hadoop/hdfs.h +++ b/third_party/hadoop/hdfs.h @@ -16,8 +16,8 @@ * limitations under the License. */ -#ifndef LIBHDFS_HDFS_H -#define LIBHDFS_HDFS_H +#ifndef TENSORFLOW_THIRD_PARTY_HADOOP_HDFS_H_ +#define TENSORFLOW_THIRD_PARTY_HADOOP_HDFS_H_ #include <errno.h> /* for EINTERNAL, etc. */ #include <fcntl.h> /* for O_RDONLY, O_WRONLY */ @@ -904,7 +904,7 @@ void hadoopRzBufferFree(hdfsFile file, struct hadoopRzBuffer *buffer); #endif #undef LIBHDFS_EXTERNAL -#endif /*LIBHDFS_HDFS_H*/ +#endif // TENSORFLOW_THIRD_PARTY_HADOOP_HDFS_H_ /** * vim: ts=4: sw=4: et diff --git a/third_party/highwayhash.BUILD b/third_party/highwayhash.BUILD index 1b8e40765e..08cb84ea2c 100644 --- a/third_party/highwayhash.BUILD +++ b/third_party/highwayhash.BUILD @@ -10,6 +10,7 @@ cc_library( srcs = ["highwayhash/sip_hash.cc"], hdrs = [ "highwayhash/sip_hash.h", + "highwayhash/endianess.h", "highwayhash/state_helpers.h", ], visibility = ["//visibility:public"], diff --git a/third_party/jpeg/jpeg.BUILD b/third_party/jpeg/jpeg.BUILD index 87a23925c4..1b9b9bf2f5 100644 --- a/third_party/jpeg/jpeg.BUILD +++ b/third_party/jpeg/jpeg.BUILD @@ -11,8 +11,8 @@ libjpegturbo_nocopts = "-[W]error" WIN_COPTS = [ "/Ox", - "/w14711", # function 'function' selected for inline expansion - "/w14710", # 'function' : function not inlined + "-DWITH_SIMD", + "-wd4996", ] libjpegturbo_copts = select({ @@ -22,7 +22,6 @@ libjpegturbo_copts = select({ "-w", ], ":windows": WIN_COPTS, - ":windows_msvc": WIN_COPTS, "//conditions:default": [ "-O3", "-w", @@ -128,6 +127,7 @@ cc_library( ":armeabi-v7a": [":simd_armv7a"], ":arm64-v8a": [":simd_armv8a"], ":linux_ppc64le": [":simd_altivec"], + ":windows": [":simd_win_x86_64"], "//conditions:default": [":simd_none"], }), ) @@ -145,27 +145,27 @@ cc_library( "jpeglib.h", "jsimd.h", "jsimddct.h", - "simd/jccolor-altivec.c", - "simd/jcgray-altivec.c", - "simd/jcsample.h", - "simd/jcsample-altivec.c", - "simd/jdcolor-altivec.c", - "simd/jdmerge-altivec.c", - "simd/jdsample-altivec.c", - "simd/jfdctfst-altivec.c", - "simd/jfdctint-altivec.c", - "simd/jidctfst-altivec.c", - "simd/jidctint-altivec.c", - "simd/jquanti-altivec.c", "simd/jsimd.h", - "simd/jsimd_altivec.h", - "simd/jsimd_powerpc.c", + "simd/powerpc/jccolor-altivec.c", + "simd/powerpc/jcgray-altivec.c", + "simd/powerpc/jcsample-altivec.c", + "simd/powerpc/jdcolor-altivec.c", + "simd/powerpc/jdmerge-altivec.c", + "simd/powerpc/jdsample-altivec.c", + "simd/powerpc/jfdctfst-altivec.c", + "simd/powerpc/jfdctint-altivec.c", + "simd/powerpc/jidctfst-altivec.c", + "simd/powerpc/jidctint-altivec.c", + "simd/powerpc/jquanti-altivec.c", + "simd/powerpc/jsimd.c", ], hdrs = [ - "simd/jccolext-altivec.c", # should have been named .inc - "simd/jcgryext-altivec.c", # should have been named .inc - "simd/jdcolext-altivec.c", # should have been named .inc - "simd/jdmrgext-altivec.c", # should have been named .inc + "simd/powerpc/jccolext-altivec.c", + "simd/powerpc/jcgryext-altivec.c", + "simd/powerpc/jdcolext-altivec.c", + "simd/powerpc/jdmrgext-altivec.c", + "simd/powerpc/jcsample.h", + "simd/powerpc/jsimd_altivec.h", ], copts = libjpegturbo_copts, nocopts = libjpegturbo_nocopts, @@ -176,6 +176,7 @@ cc_library( srcs = [ "jchuff.h", "jconfig.h", + "jconfigint.h", "jdct.h", "jerror.h", "jinclude.h", @@ -184,24 +185,35 @@ cc_library( "jpeglib.h", "jsimd.h", "jsimddct.h", - "simd/jccolor-sse2-64.o", - "simd/jcgray-sse2-64.o", - "simd/jchuff-sse2-64.o", - "simd/jcsample-sse2-64.o", - "simd/jdcolor-sse2-64.o", - "simd/jdmerge-sse2-64.o", - "simd/jdsample-sse2-64.o", - "simd/jfdctflt-sse-64.o", - "simd/jfdctfst-sse2-64.o", - "simd/jfdctint-sse2-64.o", - "simd/jidctflt-sse2-64.o", - "simd/jidctfst-sse2-64.o", - "simd/jidctint-sse2-64.o", - "simd/jidctred-sse2-64.o", - "simd/jquantf-sse2-64.o", - "simd/jquanti-sse2-64.o", "simd/jsimd.h", - "simd/jsimd_x86_64.c", + "simd/x86_64/jsimd.c", + "simd/x86_64/jccolor-avx2.o", + "simd/x86_64/jccolor-sse2.o", + "simd/x86_64/jcgray-avx2.o", + "simd/x86_64/jcgray-sse2.o", + "simd/x86_64/jchuff-sse2.o", + "simd/x86_64/jcphuff-sse2.o", + "simd/x86_64/jcsample-avx2.o", + "simd/x86_64/jcsample-sse2.o", + "simd/x86_64/jdcolor-avx2.o", + "simd/x86_64/jdcolor-sse2.o", + "simd/x86_64/jdmerge-avx2.o", + "simd/x86_64/jdmerge-sse2.o", + "simd/x86_64/jdsample-avx2.o", + "simd/x86_64/jdsample-sse2.o", + "simd/x86_64/jfdctflt-sse.o", + "simd/x86_64/jfdctfst-sse2.o", + "simd/x86_64/jfdctint-avx2.o", + "simd/x86_64/jfdctint-sse2.o", + "simd/x86_64/jidctflt-sse2.o", + "simd/x86_64/jidctfst-sse2.o", + "simd/x86_64/jidctint-avx2.o", + "simd/x86_64/jidctint-sse2.o", + "simd/x86_64/jidctred-sse2.o", + "simd/x86_64/jquantf-sse2.o", + "simd/x86_64/jquanti-avx2.o", + "simd/x86_64/jquanti-sse2.o", + "simd/x86_64/jsimdcpu.o", ], copts = libjpegturbo_copts, linkstatic = 1, @@ -211,57 +223,88 @@ cc_library( genrule( name = "simd_x86_64_assemblage23", srcs = [ - "simd/jccolext-sse2-64.asm", - "simd/jccolor-sse2-64.asm", - "simd/jcgray-sse2-64.asm", - "simd/jcgryext-sse2-64.asm", - "simd/jchuff-sse2-64.asm", - "simd/jcolsamp.inc", - "simd/jcsample-sse2-64.asm", - "simd/jdcolext-sse2-64.asm", - "simd/jdcolor-sse2-64.asm", - "simd/jdct.inc", - "simd/jdmerge-sse2-64.asm", - "simd/jdmrgext-sse2-64.asm", - "simd/jdsample-sse2-64.asm", - "simd/jfdctflt-sse-64.asm", - "simd/jfdctfst-sse2-64.asm", - "simd/jfdctint-sse2-64.asm", - "simd/jidctflt-sse2-64.asm", - "simd/jidctfst-sse2-64.asm", - "simd/jidctint-sse2-64.asm", - "simd/jidctred-sse2-64.asm", - "simd/jpeg_nbits_table.inc", - "simd/jquantf-sse2-64.asm", - "simd/jquanti-sse2-64.asm", - "simd/jsimdcfg.inc", - "simd/jsimdext.inc", + "jconfig.h", + "jconfigint.h", + "simd/x86_64/jccolext-avx2.asm", + "simd/x86_64/jccolext-sse2.asm", + "simd/x86_64/jccolor-avx2.asm", + "simd/x86_64/jccolor-sse2.asm", + "simd/x86_64/jcgray-avx2.asm", + "simd/x86_64/jcgray-sse2.asm", + "simd/x86_64/jcgryext-avx2.asm", + "simd/x86_64/jcgryext-sse2.asm", + "simd/x86_64/jchuff-sse2.asm", + "simd/x86_64/jcphuff-sse2.asm", + "simd/x86_64/jcsample-avx2.asm", + "simd/x86_64/jcsample-sse2.asm", + "simd/x86_64/jdcolext-avx2.asm", + "simd/x86_64/jdcolext-sse2.asm", + "simd/x86_64/jdcolor-avx2.asm", + "simd/x86_64/jdcolor-sse2.asm", + "simd/x86_64/jdmerge-avx2.asm", + "simd/x86_64/jdmerge-sse2.asm", + "simd/x86_64/jdmrgext-avx2.asm", + "simd/x86_64/jdmrgext-sse2.asm", + "simd/x86_64/jdsample-avx2.asm", + "simd/x86_64/jdsample-sse2.asm", + "simd/x86_64/jfdctflt-sse.asm", + "simd/x86_64/jfdctfst-sse2.asm", + "simd/x86_64/jfdctint-avx2.asm", + "simd/x86_64/jfdctint-sse2.asm", + "simd/x86_64/jidctflt-sse2.asm", + "simd/x86_64/jidctfst-sse2.asm", + "simd/x86_64/jidctint-avx2.asm", + "simd/x86_64/jidctint-sse2.asm", + "simd/x86_64/jidctred-sse2.asm", + "simd/x86_64/jquantf-sse2.asm", + "simd/x86_64/jquanti-avx2.asm", + "simd/x86_64/jquanti-sse2.asm", + "simd/x86_64/jsimdcpu.asm", + "simd/nasm/jcolsamp.inc", + "simd/nasm/jdct.inc", + "simd/nasm/jpeg_nbits_table.inc", + "simd/nasm/jsimdcfg.inc", + "simd/nasm/jsimdcfg.inc.h", + "simd/nasm/jsimdext.inc", ], outs = [ - "simd/jccolor-sse2-64.o", - "simd/jcgray-sse2-64.o", - "simd/jchuff-sse2-64.o", - "simd/jcsample-sse2-64.o", - "simd/jdcolor-sse2-64.o", - "simd/jdmerge-sse2-64.o", - "simd/jdsample-sse2-64.o", - "simd/jfdctflt-sse-64.o", - "simd/jfdctfst-sse2-64.o", - "simd/jfdctint-sse2-64.o", - "simd/jidctflt-sse2-64.o", - "simd/jidctfst-sse2-64.o", - "simd/jidctint-sse2-64.o", - "simd/jidctred-sse2-64.o", - "simd/jquantf-sse2-64.o", - "simd/jquanti-sse2-64.o", + "simd/x86_64/jccolor-avx2.o", + "simd/x86_64/jccolor-sse2.o", + "simd/x86_64/jcgray-avx2.o", + "simd/x86_64/jcgray-sse2.o", + "simd/x86_64/jchuff-sse2.o", + "simd/x86_64/jcphuff-sse2.o", + "simd/x86_64/jcsample-avx2.o", + "simd/x86_64/jcsample-sse2.o", + "simd/x86_64/jdcolor-avx2.o", + "simd/x86_64/jdcolor-sse2.o", + "simd/x86_64/jdmerge-avx2.o", + "simd/x86_64/jdmerge-sse2.o", + "simd/x86_64/jdsample-avx2.o", + "simd/x86_64/jdsample-sse2.o", + "simd/x86_64/jfdctflt-sse.o", + "simd/x86_64/jfdctfst-sse2.o", + "simd/x86_64/jfdctint-avx2.o", + "simd/x86_64/jfdctint-sse2.o", + "simd/x86_64/jidctflt-sse2.o", + "simd/x86_64/jidctfst-sse2.o", + "simd/x86_64/jidctint-avx2.o", + "simd/x86_64/jidctint-sse2.o", + "simd/x86_64/jidctred-sse2.o", + "simd/x86_64/jquantf-sse2.o", + "simd/x86_64/jquanti-avx2.o", + "simd/x86_64/jquanti-sse2.o", + "simd/x86_64/jsimdcpu.o", ], cmd = "for out in $(OUTS); do\n" + " $(location @nasm//:nasm) -f elf64" + - " -DELF -DPIC -DRGBX_FILLER_0XFF -D__x86_64__ -DARCH_X86_64" + - " -I $$(dirname $(location simd/jdct.inc))/" + - " -I $$(dirname $(location simd/jsimdcfg.inc))/" + + " -DELF -DPIC -D__x86_64__" + + " -I $$(dirname $(location jconfig.h))/" + + " -I $$(dirname $(location jconfigint.h))/" + + " -I $$(dirname $(location simd/nasm/jsimdcfg.inc.h))/" + + " -I $$(dirname $(location simd/x86_64/jccolext-sse2.asm))/" + " -o $$out" + - " $$(dirname $(location simd/jdct.inc))/$$(basename $${out%.o}.asm)\n" + + " $$(dirname $(location simd/x86_64/jccolext-sse2.asm))/$$(basename $${out%.o}.asm)\n" + "done", tools = ["@nasm"], ) @@ -272,14 +315,16 @@ cc_library( "jchuff.h", "jconfig.h", "jdct.h", + "jerror.h", "jinclude.h", "jmorecfg.h", + "jpegint.h", "jpeglib.h", "jsimd.h", "jsimddct.h", "simd/jsimd.h", - "simd/jsimd_arm.c", - "simd/jsimd_arm_neon.S", + "simd/arm/jsimd.c", + "simd/arm/jsimd_neon.S", ], copts = libjpegturbo_copts, nocopts = libjpegturbo_nocopts, @@ -291,20 +336,156 @@ cc_library( "jchuff.h", "jconfig.h", "jdct.h", + "jerror.h", "jinclude.h", "jmorecfg.h", + "jpegint.h", "jpeglib.h", "jsimd.h", "jsimddct.h", "simd/jsimd.h", - "simd/jsimd_arm64.c", - "simd/jsimd_arm64_neon.S", + "simd/arm64/jsimd.c", + "simd/arm64/jsimd_neon.S", ], copts = libjpegturbo_copts, nocopts = libjpegturbo_nocopts, ) cc_library( + name = "simd_win_x86_64", + srcs = [ + "jchuff.h", + "jconfig.h", + "jconfigint.h", + "jdct.h", + "jerror.h", + "jinclude.h", + "jmorecfg.h", + "jpegint.h", + "jpeglib.h", + "jsimd.h", + "jsimddct.h", + "simd/jsimd.h", + "simd/x86_64/jsimd.c", + "simd/x86_64/jccolor-avx2.obj", + "simd/x86_64/jccolor-sse2.obj", + "simd/x86_64/jcgray-avx2.obj", + "simd/x86_64/jcgray-sse2.obj", + "simd/x86_64/jchuff-sse2.obj", + "simd/x86_64/jcphuff-sse2.obj", + "simd/x86_64/jcsample-avx2.obj", + "simd/x86_64/jcsample-sse2.obj", + "simd/x86_64/jdcolor-avx2.obj", + "simd/x86_64/jdcolor-sse2.obj", + "simd/x86_64/jdmerge-avx2.obj", + "simd/x86_64/jdmerge-sse2.obj", + "simd/x86_64/jdsample-avx2.obj", + "simd/x86_64/jdsample-sse2.obj", + "simd/x86_64/jfdctflt-sse.obj", + "simd/x86_64/jfdctfst-sse2.obj", + "simd/x86_64/jfdctint-avx2.obj", + "simd/x86_64/jfdctint-sse2.obj", + "simd/x86_64/jidctflt-sse2.obj", + "simd/x86_64/jidctfst-sse2.obj", + "simd/x86_64/jidctint-avx2.obj", + "simd/x86_64/jidctint-sse2.obj", + "simd/x86_64/jidctred-sse2.obj", + "simd/x86_64/jquantf-sse2.obj", + "simd/x86_64/jquanti-avx2.obj", + "simd/x86_64/jquanti-sse2.obj", + "simd/x86_64/jsimdcpu.obj", + ], + copts = libjpegturbo_copts, +) + +genrule( + name = "simd_win_x86_64_assemble", + srcs = [ + "jconfig.h", + "jconfigint.h", + "simd/x86_64/jccolext-avx2.asm", + "simd/x86_64/jccolext-sse2.asm", + "simd/x86_64/jccolor-avx2.asm", + "simd/x86_64/jccolor-sse2.asm", + "simd/x86_64/jcgray-avx2.asm", + "simd/x86_64/jcgray-sse2.asm", + "simd/x86_64/jcgryext-avx2.asm", + "simd/x86_64/jcgryext-sse2.asm", + "simd/x86_64/jchuff-sse2.asm", + "simd/x86_64/jcphuff-sse2.asm", + "simd/x86_64/jcsample-avx2.asm", + "simd/x86_64/jcsample-sse2.asm", + "simd/x86_64/jdcolext-avx2.asm", + "simd/x86_64/jdcolext-sse2.asm", + "simd/x86_64/jdcolor-avx2.asm", + "simd/x86_64/jdcolor-sse2.asm", + "simd/x86_64/jdmerge-avx2.asm", + "simd/x86_64/jdmerge-sse2.asm", + "simd/x86_64/jdmrgext-avx2.asm", + "simd/x86_64/jdmrgext-sse2.asm", + "simd/x86_64/jdsample-avx2.asm", + "simd/x86_64/jdsample-sse2.asm", + "simd/x86_64/jfdctflt-sse.asm", + "simd/x86_64/jfdctfst-sse2.asm", + "simd/x86_64/jfdctint-avx2.asm", + "simd/x86_64/jfdctint-sse2.asm", + "simd/x86_64/jidctflt-sse2.asm", + "simd/x86_64/jidctfst-sse2.asm", + "simd/x86_64/jidctint-avx2.asm", + "simd/x86_64/jidctint-sse2.asm", + "simd/x86_64/jidctred-sse2.asm", + "simd/x86_64/jquantf-sse2.asm", + "simd/x86_64/jquanti-avx2.asm", + "simd/x86_64/jquanti-sse2.asm", + "simd/x86_64/jsimdcpu.asm", + "simd/nasm/jcolsamp.inc", + "simd/nasm/jdct.inc", + "simd/nasm/jpeg_nbits_table.inc", + "simd/nasm/jsimdcfg.inc", + "simd/nasm/jsimdcfg.inc.h", + "simd/nasm/jsimdext.inc", + ], + outs = [ + "simd/x86_64/jccolor-avx2.obj", + "simd/x86_64/jccolor-sse2.obj", + "simd/x86_64/jcgray-avx2.obj", + "simd/x86_64/jcgray-sse2.obj", + "simd/x86_64/jchuff-sse2.obj", + "simd/x86_64/jcphuff-sse2.obj", + "simd/x86_64/jcsample-avx2.obj", + "simd/x86_64/jcsample-sse2.obj", + "simd/x86_64/jdcolor-avx2.obj", + "simd/x86_64/jdcolor-sse2.obj", + "simd/x86_64/jdmerge-avx2.obj", + "simd/x86_64/jdmerge-sse2.obj", + "simd/x86_64/jdsample-avx2.obj", + "simd/x86_64/jdsample-sse2.obj", + "simd/x86_64/jfdctflt-sse.obj", + "simd/x86_64/jfdctfst-sse2.obj", + "simd/x86_64/jfdctint-avx2.obj", + "simd/x86_64/jfdctint-sse2.obj", + "simd/x86_64/jidctflt-sse2.obj", + "simd/x86_64/jidctfst-sse2.obj", + "simd/x86_64/jidctint-avx2.obj", + "simd/x86_64/jidctint-sse2.obj", + "simd/x86_64/jidctred-sse2.obj", + "simd/x86_64/jquantf-sse2.obj", + "simd/x86_64/jquanti-avx2.obj", + "simd/x86_64/jquanti-sse2.obj", + "simd/x86_64/jsimdcpu.obj", + ], + cmd = "for out in $(OUTS); do\n" + + " $(location @nasm//:nasm) -fwin64 -DWIN64 -D__x86_64__" + + " -I $$(dirname $(location simd/x86_64/jccolext-sse2.asm))/" + + " -I $$(dirname $(location simd/nasm/jdct.inc))/" + + " -I $$(dirname $(location simd/nasm/jdct.inc))/../../win/" + + " -o $$out" + + " $$(dirname $(location simd/x86_64/jccolext-sse2.asm))/$$(basename $${out%.obj}.asm)\n" + + "done", + tools = ["@nasm"], +) + +cc_library( name = "simd_none", srcs = [ "jchuff.h", @@ -329,50 +510,44 @@ template_rule( out = "jconfig_win.h", substitutions = { "@JPEG_LIB_VERSION@": "62", - "@VERSION@": "1.5.1", - "@LIBJPEG_TURBO_VERSION_NUMBER@": "1005001", - "cmakedefine": "define", + "@VERSION@": "2.0.0", + "@LIBJPEG_TURBO_VERSION_NUMBER@": "2000000", "@BITS_IN_JSAMPLE@": "8", - }, -) - -template_rule( - name = "jconfigint_win", - src = "win/jconfigint.h.in", - out = "jconfigint_win.h", - substitutions = { - "@VERSION@": "1.5.1", - "@BUILD@": "20161115", - "@CMAKE_PROJECT_NAME@": "libjpeg-turbo", + "#cmakedefine C_ARITH_CODING_SUPPORTED": "#define C_ARITH_CODING_SUPPORTED", + "#cmakedefine D_ARITH_CODING_SUPPORTED": "#define D_ARITH_CODING_SUPPORTED", + "#cmakedefine MEM_SRCDST_SUPPORTED": "#define MEM_SRCDST_SUPPORTED", + "#cmakedefine WITH_SIMD": "", }, ) JCONFIG_NOWIN_COMMON_SUBSTITUTIONS = { - "LIBJPEG_TURBO_VERSION 0": "LIBJPEG_TURBO_VERSION 1.5.1", - "LIBJPEG_TURBO_VERSION_NUMBER 0": "LIBJPEG_TURBO_VERSION_NUMBER 1005001", - "#undef C_ARITH_CODING_SUPPORTED": "#define C_ARITH_CODING_SUPPORTED 1", - "#undef D_ARITH_CODING_SUPPORTED": "#define D_ARITH_CODING_SUPPORTED 1", - "#undef HAVE_LOCALE_H": "#define HAVE_LOCALE_H 1", - "#undef HAVE_STDDEF_H": "#define HAVE_STDDEF_H 1", - "#undef HAVE_STDLIB_H": "#define HAVE_STDLIB_H 1", - "#undef HAVE_UNSIGNED_CHAR": "#define HAVE_UNSIGNED_CHAR 1", - "#undef HAVE_UNSIGNED_SHORT": "#define HAVE_UNSIGNED_SHORT 1", - "#undef INCOMPLETE_TYPES_BROKEN": "", - "#undef MEM_SRCDST_SUPPORTED": "#define MEM_SRCDST_SUPPORTED 1", - "#undef NEED_BSD_STRINGS": "", - "#undef NEED_SYS_TYPES_H": "#define NEED_SYS_TYPES_H 1", - "#undef __CHAR_UNSIGNED__": "", + "@JPEG_LIB_VERSION@": "62", + "@VERSION@": "2.0.0", + "@LIBJPEG_TURBO_VERSION_NUMBER@": "2000000", + "#cmakedefine C_ARITH_CODING_SUPPORTED": "#define C_ARITH_CODING_SUPPORTED", + "#cmakedefine D_ARITH_CODING_SUPPORTED": "#define D_ARITH_CODING_SUPPORTED", + "#cmakedefine MEM_SRCDST_SUPPORTED": "#define MEM_SRCDST_SUPPORTED", + "@BITS_IN_JSAMPLE@": "8", + "#cmakedefine HAVE_LOCALE_H": "#define HAVE_LOCALE_H 1", + "#cmakedefine HAVE_STDDEF_H": "#define HAVE_STDDEF_H 1", + "#cmakedefine HAVE_STDLIB_H": "#define HAVE_STDLIB_H 1", + "#cmakedefine NEED_SYS_TYPES_H": "#define NEED_SYS_TYPES_H", + "#cmakedefine NEED_BSD_STRINGS": "", + "#cmakedefine HAVE_UNSIGNED_CHAR": "#define HAVE_UNSIGNED_CHAR 1", + "#cmakedefine HAVE_UNSIGNED_SHORT": "#define HAVE_UNSIGNED_SHORT 1", + "#cmakedefine INCOMPLETE_TYPES_BROKEN": "", + "#cmakedefine RIGHT_SHIFT_IS_UNSIGNED": "", + "#cmakedefine __CHAR_UNSIGNED__": "", "#undef const": "", "#undef size_t": "", - "#undef RIGHT_SHIFT_IS_UNSIGNED": "", } JCONFIG_NOWIN_SIMD_SUBSTITUTIONS = { - "#undef WITH_SIMD": "#define WITH_SIMD 1", + "#cmakedefine WITH_SIMD": "#define WITH_SIMD", } JCONFIG_NOWIN_NOSIMD_SUBSTITUTIONS = { - "#undef WITH_SIMD": "", + "#cmakedefine WITH_SIMD": "", } JCONFIG_NOWIN_SIMD_SUBSTITUTIONS.update(JCONFIG_NOWIN_COMMON_SUBSTITUTIONS) @@ -393,22 +568,55 @@ template_rule( substitutions = JCONFIG_NOWIN_SIMD_SUBSTITUTIONS, ) +JCONFIGINT_COMMON_SUBSTITUTIONS = { + "@BUILD@": "20180831", + "@VERSION@": "2.0.0", + "@CMAKE_PROJECT_NAME@": "libjpeg-turbo", + "#undef inline": "", + "#cmakedefine HAVE_INTRIN_H": "", +} + +JCONFIGINT_NOWIN_SUBSTITUTIONS = { + "#cmakedefine HAVE_BUILTIN_CTZL": "#define HAVE_BUILTIN_CTZL", + "@INLINE@": "inline __attribute__((always_inline))", + "#define SIZEOF_SIZE_T @SIZE_T@": "#if (__WORDSIZE==64 && !defined(__native_client__))\n" + + "#define SIZEOF_SIZE_T 8\n" + + "#else\n" + + "#define SIZEOF_SIZE_T 4\n" + + "#endif\n", +} + +JCONFIGINT_WIN_SUBSTITUTIONS = { + "#cmakedefine HAVE_BUILTIN_CTZL": "", + "#define INLINE @INLINE@": "#if defined(__GNUC__)\n" + + "#define INLINE inline __attribute__((always_inline))\n" + + "#elif defined(_MSC_VER)\n" + + "#define INLINE __forceinline\n" + + "#else\n" + + "#define INLINE\n" + + "#endif\n", + "#define SIZEOF_SIZE_T @SIZE_T@": "#if (__WORDSIZE==64)\n" + + "#define SIZEOF_SIZE_T 8\n" + + "#else\n" + + "#define SIZEOF_SIZE_T 4\n" + + "#endif\n", +} + +JCONFIGINT_NOWIN_SUBSTITUTIONS.update(JCONFIGINT_COMMON_SUBSTITUTIONS) +JCONFIGINT_WIN_SUBSTITUTIONS.update(JCONFIGINT_COMMON_SUBSTITUTIONS) + template_rule( name = "jconfigint_nowin", src = "jconfigint.h.in", out = "jconfigint_nowin.h", - substitutions = { - "#undef BUILD": "#define BUILD \"20161115\"", - "#undef inline": "", - "#undef INLINE": "#define INLINE inline __attribute__((always_inline))", - "#undef PACKAGE_NAME": "#define PACKAGE_NAME \"libjpeg-turbo\"", - "#undef VERSION": "#define VERSION \"1.5.1\"", - "#undef SIZEOF_SIZE_T": "#if (__WORDSIZE==64 && !defined(__native_client__))\n" + - "#define SIZEOF_SIZE_T 8\n" + - "#else\n" + - "#define SIZEOF_SIZE_T 4\n" + - "#endif\n", - }, + substitutions = JCONFIGINT_NOWIN_SUBSTITUTIONS, +) + +template_rule( + name = "jconfigint_win", + src = "jconfigint.h.in", + out = "jconfigint_win.h", + substitutions = JCONFIGINT_WIN_SUBSTITUTIONS, ) genrule( @@ -421,7 +629,6 @@ genrule( outs = ["jconfig.h"], cmd = select({ ":windows": "cp $(location jconfig_win.h) $@", - ":windows_msvc": "cp $(location jconfig_win.h) $@", ":k8": "cp $(location jconfig_nowin_simd.h) $@", ":armeabi-v7a": "cp $(location jconfig_nowin_simd.h) $@", ":arm64-v8a": "cp $(location jconfig_nowin_simd.h) $@", @@ -439,7 +646,6 @@ genrule( outs = ["jconfigint.h"], cmd = select({ ":windows": "cp $(location jconfigint_win.h) $@", - ":windows_msvc": "cp $(location jconfigint_win.h) $@", "//conditions:default": "cp $(location jconfigint_nowin.h) $@", }), ) @@ -526,12 +732,12 @@ config_setting( config_setting( name = "armeabi-v7a", - values = {"android_cpu": "armeabi-v7a"}, + values = {"cpu": "armeabi-v7a"}, ) config_setting( name = "arm64-v8a", - values = {"android_cpu": "arm64-v8a"}, + values = {"cpu": "arm64-v8a"}, ) config_setting( @@ -540,11 +746,6 @@ config_setting( ) config_setting( - name = "windows_msvc", - values = {"cpu": "x64_windows_msvc"}, -) - -config_setting( name = "linux_ppc64le", values = {"cpu": "ppc"}, ) diff --git a/third_party/jsoncpp.BUILD b/third_party/jsoncpp.BUILD index 65f98410b2..cf3cba0555 100644 --- a/third_party/jsoncpp.BUILD +++ b/third_party/jsoncpp.BUILD @@ -6,7 +6,6 @@ cc_library( name = "jsoncpp", srcs = [ "include/json/assertions.h", - "src/lib_json/json_batchallocator.h", "src/lib_json/json_reader.cpp", "src/lib_json/json_tool.h", "src/lib_json/json_value.cpp", @@ -20,9 +19,13 @@ cc_library( "include/json/json.h", "include/json/reader.h", "include/json/value.h", + "include/json/version.h", "include/json/writer.h", ], - copts = ["-DJSON_USE_EXCEPTION=0"], + copts = [ + "-DJSON_USE_EXCEPTION=0", + "-DJSON_HAS_INT64", + ], includes = ["include"], visibility = ["//visibility:public"], deps = [":private"], diff --git a/third_party/kafka/BUILD b/third_party/kafka/BUILD index a839ca717e..11ec50069a 100644 --- a/third_party/kafka/BUILD +++ b/third_party/kafka/BUILD @@ -15,6 +15,7 @@ cc_library( "src-cpp/KafkaConsumerImpl.cpp", "src-cpp/MessageImpl.cpp", "src-cpp/MetadataImpl.cpp", + "src-cpp/ProducerImpl.cpp", "src-cpp/QueueImpl.cpp", "src-cpp/RdKafka.cpp", "src-cpp/TopicImpl.cpp", @@ -47,8 +48,13 @@ cc_library( "src/rdinterval.h", "src/rdkafka.c", "src/rdkafka.h", + "src/rdkafka_admin.c", + "src/rdkafka_admin.h", "src/rdkafka_assignor.c", "src/rdkafka_assignor.h", + "src/rdkafka_aux.c", + "src/rdkafka_aux.h", + "src/rdkafka_background.c", "src/rdkafka_broker.c", "src/rdkafka_broker.h", "src/rdkafka_buf.c", @@ -57,9 +63,12 @@ cc_library( "src/rdkafka_cgrp.h", "src/rdkafka_conf.c", "src/rdkafka_conf.h", + "src/rdkafka_confval.h", "src/rdkafka_event.h", "src/rdkafka_feature.c", "src/rdkafka_feature.h", + "src/rdkafka_header.c", + "src/rdkafka_header.h", "src/rdkafka_int.h", "src/rdkafka_interceptor.c", "src/rdkafka_interceptor.h", @@ -93,7 +102,6 @@ cc_library( "src/rdkafka_sasl_int.h", "src/rdkafka_sasl_plain.c", "src/rdkafka_subscription.c", - "src/rdkafka_subscription.h", "src/rdkafka_timer.c", "src/rdkafka_timer.h", "src/rdkafka_topic.c", @@ -105,6 +113,8 @@ cc_library( "src/rdlist.h", "src/rdlog.c", "src/rdlog.h", + "src/rdmurmur2.c", + "src/rdmurmur2.h", "src/rdports.c", "src/rdports.h", "src/rdposix.h", @@ -127,7 +137,15 @@ cc_library( "src/tinycthread.h", "src/xxhash.c", "src/xxhash.h", - ], + ] + select({ + "@org_tensorflow//tensorflow:windows": [ + "src/rdkafka_sasl_win32.c", + "src/rdwin32.h", + "src/regexp.c", + "src/regexp.h", + ], + "//conditions:default": [], + }), hdrs = [ "config.h", "src-cpp/rdkafkacpp.h", @@ -135,15 +153,25 @@ cc_library( "src/lz4.c", "src/snappy_compat.h", ], - copts = [ - "-Iexternal/kafka/src", - "-Iexternal/kafka/src-cpp", - ], - defines = [ - ], - linkopts = [ - "-lpthread", + copts = select({ + "@org_tensorflow//tensorflow:windows": [ + "-DWIN32_LEAN_AND_MEAN", + "-DWITHOUT_WIN32_CONFIG", + "-DWITH_ZLIB=1", + "-DWITH_SSL=1", + "-DWITH_SNAPPY=1", + ], + "//conditions:default": [], + }), + defines = ["LIBRDKAFKA_STATICLIB"], + includes = [ + "src", + "src-cpp", ], + linkopts = select({ + "@org_tensorflow//tensorflow:windows": ["-defaultlib:crypt32.lib"], + "//conditions:default": ["-lpthread"], + }), visibility = ["//visibility:public"], deps = [ "@boringssl//:ssl", diff --git a/third_party/libxsmm.BUILD b/third_party/libxsmm.BUILD index 4124f2db63..ee49d281ab 100644 --- a/third_party/libxsmm.BUILD +++ b/third_party/libxsmm.BUILD @@ -3,7 +3,7 @@ licenses(["notice"]) # BSD 3-clause -exports_files(["LICENSE"]) +exports_files(["LICENSE.md"]) # Arguments to ./scripts/libxsmm_interface.py, see that file for detailed description. # precision: SP & DP @@ -38,8 +38,8 @@ genrule( ":libxsmm_interface", ], visibility = [ - "//tensorflow/core/kernels:__pkg__", "//third_party/eigen3:__pkg__", + "//tensorflow/core/kernels:__pkg__", ], ) diff --git a/third_party/llvm/llvm.BUILD b/third_party/llvm/llvm.autogenerated.BUILD index 28293a3659..776935739a 100644 --- a/third_party/llvm/llvm.BUILD +++ b/third_party/llvm/llvm.autogenerated.BUILD @@ -8,10 +8,14 @@ exports_files(["LICENSE.TXT"]) load( "@org_tensorflow//third_party/llvm:llvm.bzl", - "gentbl", - "expand_cmake_vars", - "llvm_target_cmake_vars", "cmake_var_string", + "expand_cmake_vars", + "gentbl", + "llvm_all_cmake_vars", + "llvm_copts", + "llvm_defines", + "llvm_linkopts", + "llvm_support_platform_specific_srcs_glob", ) load( "@org_tensorflow//third_party:common.bzl", @@ -24,9 +28,7 @@ llvm_host_triple = "x86_64-unknown-linux_gnu" llvm_targets = [ "AArch64", - # Uncomment to enable the AMDGPU backend. - # TODO(phawkins): use a configure-time test. - # "AMDGPU", + "AMDGPU", "ARM", "NVPTX", "PowerPC", @@ -39,154 +41,25 @@ llvm_target_asm_printers = llvm_targets llvm_target_disassemblers = llvm_targets -# TODO(phawkins): the set of CMake variables was hardcoded for expediency. -# However, we should really detect many of these via configure-time tests. - -# The set of CMake variables common to all targets. -cmake_vars = { - # Headers - "HAVE_DIRENT_H": 1, - "HAVE_DLFCN_H": 1, - "HAVE_ERRNO_H": 1, - "HAVE_EXECINFO_H": 1, - "HAVE_FCNTL_H": 1, - "HAVE_INTTYPES_H": 1, - "HAVE_PTHREAD_H": 1, - "HAVE_SIGNAL_H": 1, - "HAVE_STDINT_H": 1, - "HAVE_SYS_IOCTL_H": 1, - "HAVE_SYS_MMAN_H": 1, - "HAVE_SYS_PARAM_H": 1, - "HAVE_SYS_RESOURCE_H": 1, - "HAVE_SYS_STAT_H": 1, - "HAVE_SYS_TIME_H": 1, - "HAVE_SYS_TYPES_H": 1, - "HAVE_TERMIOS_H": 1, - "HAVE_UNISTD_H": 1, - "HAVE_ZLIB_H": 1, - - # Features - "HAVE_BACKTRACE": 1, - "BACKTRACE_HEADER": "execinfo.h", - "HAVE_DLOPEN": 1, - "HAVE_FUTIMES": 1, - "HAVE_GETCWD": 1, - "HAVE_GETPAGESIZE": 1, - "HAVE_GETRLIMIT": 1, - "HAVE_GETRUSAGE": 1, - "HAVE_GETTIMEOFDAY": 1, - "HAVE_INT64_T": 1, - "HAVE_ISATTY": 1, - "HAVE_LIBEDIT": 1, - "HAVE_LIBPTHREAD": 1, - "HAVE_LIBZ": 1, - "HAVE_MKDTEMP": 1, - "HAVE_MKSTEMP": 1, - "HAVE_MKTEMP": 1, - "HAVE_PREAD": 1, - "HAVE_PTHREAD_GETSPECIFIC": 1, - "HAVE_PTHREAD_MUTEX_LOCK": 1, - "HAVE_PTHREAD_RWLOCK_INIT": 1, - "HAVE_REALPATH": 1, - "HAVE_SBRK": 1, - "HAVE_SETENV": 1, - "HAVE_SETRLIMIT": 1, - "HAVE_SIGALTSTACK": 1, - "HAVE_STRERROR": 1, - "HAVE_STRERROR_R": 1, - "HAVE_STRTOLL": 1, - "HAVE_SYSCONF": 1, - "HAVE_UINT64_T": 1, - "HAVE__UNWIND_BACKTRACE": 1, - - # LLVM features - "ENABLE_BACKTRACES": 1, - "LLVM_BINDIR": "/dev/null", - "LLVM_DISABLE_ABI_BREAKING_CHECKS_ENFORCING": 0, - "LLVM_ENABLE_ABI_BREAKING_CHECKS": 0, - "LLVM_ENABLE_THREADS": 1, - "LLVM_ENABLE_ZLIB": 1, - "LLVM_HAS_ATOMICS": 1, - "LLVM_INCLUDEDIR": "/dev/null", - "LLVM_INFODIR": "/dev/null", - "LLVM_MANDIR": "/dev/null", - "LLVM_NATIVE_TARGET": 1, - "LLVM_NATIVE_TARGETINFO": 1, - "LLVM_NATIVE_TARGETMC": 1, - "LLVM_NATIVE_ASMPRINTER": 1, - "LLVM_NATIVE_ASMPARSER": 1, - "LLVM_NATIVE_DISASSEMBLER": 1, - "LLVM_ON_UNIX": 1, - "LLVM_PREFIX": "/dev/null", - "LLVM_VERSION_MAJOR": 0, - "LLVM_VERSION_MINOR": 0, - "LLVM_VERSION_PATCH": 0, - "LTDL_SHLIB_EXT": ".so", - "PACKAGE_NAME": "llvm", - "PACKAGE_STRING": "llvm tensorflow-trunk", - "PACKAGE_VERSION": "tensorflow-trunk", - "RETSIGTYPE": "void", -} - -# CMake variables specific to the Linux platform -linux_cmake_vars = { - "HAVE_MALLOC_H": 1, - "HAVE_LINK_H": 1, - "HAVE_MALLINFO": 1, - "HAVE_FUTIMENS": 1, -} - -# CMake variables specific to the Darwin (Mac OS X) platform. -darwin_cmake_vars = { - "HAVE_MALLOC_MALLOC_H": 1, -} - -# Select a set of CMake variables based on the platform. -# TODO(phawkins): use a better method to select the right host triple, rather -# than hardcoding x86_64. -all_cmake_vars = select({ - "@org_tensorflow//tensorflow:darwin": cmake_var_string( - cmake_vars + llvm_target_cmake_vars("X86", "x86_64-apple-darwin") + - darwin_cmake_vars, - ), - "@org_tensorflow//tensorflow:linux_ppc64le": cmake_var_string( - cmake_vars + - llvm_target_cmake_vars("PowerPC", "powerpc64le-unknown-linux_gnu") + - linux_cmake_vars, - ), - "//conditions:default": cmake_var_string( - cmake_vars + - llvm_target_cmake_vars("X86", "x86_64-unknown-linux_gnu") + - linux_cmake_vars, - ), -}) - # Performs CMake variable substitutions on configuration header files. expand_cmake_vars( - name = "datatypes_gen", - src = "include/llvm/Support/DataTypes.h.cmake", - cmake_vars = all_cmake_vars, - dst = "include/llvm/Support/DataTypes.h", -) - -expand_cmake_vars( name = "config_gen", src = "include/llvm/Config/config.h.cmake", - cmake_vars = all_cmake_vars, + cmake_vars = llvm_all_cmake_vars, dst = "include/llvm/Config/config.h", ) expand_cmake_vars( name = "llvm_config_gen", src = "include/llvm/Config/llvm-config.h.cmake", - cmake_vars = all_cmake_vars, + cmake_vars = llvm_all_cmake_vars, dst = "include/llvm/Config/llvm-config.h", ) expand_cmake_vars( name = "abi_breaking_gen", src = "include/llvm/Config/abi-breaking.h.cmake", - cmake_vars = all_cmake_vars, + cmake_vars = llvm_all_cmake_vars, dst = "include/llvm/Config/abi-breaking.h", ) @@ -236,25 +109,25 @@ template_rule( ) # A common library that all LLVM targets depend on. +# TODO(b/113996071): We need to glob all potentially #included files and stage +# them here because LLVM's build files are not strict headers clean, and remote +# build execution requires all inputs to be depended upon. cc_library( name = "config", - hdrs = [ + hdrs = glob([ + "**/*.h", + "**/*.def", + "**/*.inc.cpp", + ]) + [ "include/llvm/Config/AsmParsers.def", "include/llvm/Config/AsmPrinters.def", "include/llvm/Config/Disassemblers.def", "include/llvm/Config/Targets.def", - "include/llvm/Config/abi-breaking.h", "include/llvm/Config/config.h", "include/llvm/Config/llvm-config.h", + "include/llvm/Config/abi-breaking.h", ], - defines = [ - "LLVM_ENABLE_STATS", - "__STDC_LIMIT_MACROS", - "__STDC_CONSTANT_MACROS", - "__STDC_FORMAT_MACROS", - "_DEBUG", - "LLVM_BUILD_GLOBAL_ISEL", - ], + defines = llvm_defines, includes = ["include"], ) @@ -270,19 +143,8 @@ genrule( # Rules that apply the LLVM tblgen tool. gentbl( - name = "intrinsics_gen", - tbl_outs = [("-gen-intrinsic", "include/llvm/IR/Intrinsics.gen")], - tblgen = ":llvm-tblgen", - td_file = "include/llvm/IR/Intrinsics.td", - td_srcs = glob([ - "include/llvm/CodeGen/*.td", - "include/llvm/IR/Intrinsics*.td", - ]), -) - -gentbl( name = "attributes_gen", - tbl_outs = [("-gen-attrs", "include/llvm/IR/Attributes.gen")], + tbl_outs = [("-gen-attrs", "include/llvm/IR/Attributes.inc")], tblgen = ":llvm-tblgen", td_file = "include/llvm/IR/Attributes.td", td_srcs = ["include/llvm/IR/Attributes.td"], @@ -299,20 +161,51 @@ gentbl( ], ) +gentbl( + name = "instcombine_transforms_gen", + tbl_outs = [( + "-gen-searchable-tables", + "lib/Transforms/InstCombine/InstCombineTables.inc", + )], + tblgen = ":llvm-tblgen", + td_file = "lib/Transforms/InstCombine/InstCombineTables.td", + td_srcs = glob([ + "include/llvm/CodeGen/*.td", + "include/llvm/IR/Intrinsics*.td", + ]) + ["include/llvm/TableGen/SearchableTable.td"], +) + +gentbl( + name = "intrinsic_enums_gen", + tbl_outs = [("-gen-intrinsic-enums", "include/llvm/IR/IntrinsicEnums.inc")], + tblgen = ":llvm-tblgen", + td_file = "include/llvm/IR/Intrinsics.td", + td_srcs = glob([ + "include/llvm/CodeGen/*.td", + "include/llvm/IR/Intrinsics*.td", + ]), +) + +gentbl( + name = "intrinsics_impl_gen", + tbl_outs = [("-gen-intrinsic-impl", "include/llvm/IR/IntrinsicImpl.inc")], + tblgen = ":llvm-tblgen", + td_file = "include/llvm/IR/Intrinsics.td", + td_srcs = glob([ + "include/llvm/CodeGen/*.td", + "include/llvm/IR/Intrinsics*.td", + ]), +) + # Binary targets used by Tensorflow. cc_binary( name = "llvm-tblgen", srcs = glob([ "utils/TableGen/*.cpp", "utils/TableGen/*.h", - ]) + [ - "lib/Target/X86/Disassembler/X86DisassemblerDecoderCommon.h", - ], - linkopts = [ - "-lm", - "-ldl", - "-lpthread", - ], + ]), + copts = llvm_copts, + linkopts = llvm_linkopts, stamp = 0, deps = [ ":config", @@ -328,11 +221,8 @@ cc_binary( "utils/FileCheck/*.cpp", "utils/FileCheck/*.h", ]), - linkopts = [ - "-ldl", - "-lm", - "-lpthread", - ], + copts = llvm_copts, + linkopts = llvm_linkopts, stamp = 0, deps = [":support"], ) @@ -371,13 +261,31 @@ llvm_target_list = [ ("-gen-dag-isel", "lib/Target/AMDGPU/AMDGPUGenDAGISel.inc"), ("-gen-callingconv", "lib/Target/AMDGPU/AMDGPUGenCallingConv.inc"), ("-gen-subtarget", "lib/Target/AMDGPU/AMDGPUGenSubtargetInfo.inc"), - ("-gen-tgt-intrinsic", "lib/Target/AMDGPU/AMDGPUGenIntrinsics.inc"), + ("-gen-tgt-intrinsic-impl", "lib/Target/AMDGPU/AMDGPUGenIntrinsicImpl.inc"), + ("-gen-tgt-intrinsic-enums", "lib/Target/AMDGPU/AMDGPUGenIntrinsicEnums.inc"), ("-gen-emitter", "lib/Target/AMDGPU/AMDGPUGenMCCodeEmitter.inc"), ("-gen-dfa-packetizer", "lib/Target/AMDGPU/AMDGPUGenDFAPacketizer.inc"), ("-gen-asm-writer", "lib/Target/AMDGPU/AMDGPUGenAsmWriter.inc"), ("-gen-asm-matcher", "lib/Target/AMDGPU/AMDGPUGenAsmMatcher.inc"), ("-gen-disassembler", "lib/Target/AMDGPU/AMDGPUGenDisassemblerTables.inc"), ("-gen-pseudo-lowering", "lib/Target/AMDGPU/AMDGPUGenMCPseudoLowering.inc"), + ("-gen-searchable-tables", "lib/Target/AMDGPU/AMDGPUGenSearchableTables.inc"), + ("-gen-global-isel", "lib/Target/AMDGPU/AMDGPUGenGlobalISel.inc"), + ], + }, + { + "name": "AMDGPU", + "lower_name": "amdgpu_r600", + "short_name": "R600", + "tbl_outs": [ + ("-gen-asm-writer", "lib/Target/AMDGPU/R600GenAsmWriter.inc"), + ("-gen-callingconv", "lib/Target/AMDGPU/R600GenCallingConv.inc"), + ("-gen-dag-isel", "lib/Target/AMDGPU/R600GenDAGISel.inc"), + ("-gen-dfa-packetizer", "lib/Target/AMDGPU/R600GenDFAPacketizer.inc"), + ("-gen-instr-info", "lib/Target/AMDGPU/R600GenInstrInfo.inc"), + ("-gen-emitter", "lib/Target/AMDGPU/R600GenMCCodeEmitter.inc"), + ("-gen-register-info", "lib/Target/AMDGPU/R600GenRegisterInfo.inc"), + ("-gen-subtarget", "lib/Target/AMDGPU/R600GenSubtargetInfo.inc"), ], }, { @@ -503,7 +411,7 @@ cc_library( "include/llvm/Target/AArch64/AsmParser/*.inc", "lib/Target/AArch64/AsmParser/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/AArch64"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/AArch64"], deps = [ ":aarch64_desc", ":aarch64_info", @@ -528,7 +436,7 @@ cc_library( "include/llvm/Target/AArch64/InstPrinter/*.inc", "lib/Target/AArch64/InstPrinter/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/AArch64"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/AArch64"], deps = [ ":aarch64_target_gen", ":aarch64_utils", @@ -551,7 +459,7 @@ cc_library( "include/llvm/Target/AArch64/*.inc", "lib/Target/AArch64/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/AArch64"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/AArch64"], deps = [ ":aarch64_asm_printer", ":aarch64_desc", @@ -584,14 +492,15 @@ cc_library( "include/llvm/Target/AArch64/MCTargetDesc/*.inc", "lib/Target/AArch64/MCTargetDesc/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/AArch64"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/AArch64"], deps = [ ":aarch64_asm_printer", ":aarch64_info", ":aarch64_target_gen", ":attributes_gen", ":config", - ":intrinsics_gen", + ":intrinsic_enums_gen", + ":intrinsics_impl_gen", ":mc", ":support", ], @@ -610,7 +519,7 @@ cc_library( "include/llvm/Target/AArch64/Disassembler/*.inc", "lib/Target/AArch64/Disassembler/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/AArch64"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/AArch64"], deps = [ ":aarch64_desc", ":aarch64_info", @@ -638,7 +547,7 @@ cc_library( "lib/Target/AArch64/AArch64*.h", "lib/Target/AArch64/TargetInfo/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/AArch64"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/AArch64"], deps = [ ":code_gen", ":config", @@ -661,7 +570,7 @@ cc_library( "include/llvm/Target/AArch64/Utils/*.inc", "lib/Target/AArch64/Utils/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/AArch64"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/AArch64"], deps = [ ":aarch64_target_gen", ":config", @@ -683,6 +592,7 @@ cc_library( "include/llvm/Transforms/AggressiveInstCombine/*.def", "include/llvm/Transforms/AggressiveInstCombine/*.inc", ]), + copts = llvm_copts, deps = [ ":analysis", ":config", @@ -707,6 +617,7 @@ cc_library( "include/llvm/Analysis/*.def", "include/llvm/Analysis/*.inc", ]), + copts = llvm_copts, deps = [ ":binary_format", ":config", @@ -730,7 +641,7 @@ cc_library( "include/llvm/Target/AMDGPU/MCTargetDesc/*.inc", "lib/Target/AMDGPU/MCTargetDesc/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/AMDGPU"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/AMDGPU"], deps = [ ":amdgpu_asm_printer", ":amdgpu_info", @@ -755,7 +666,7 @@ cc_library( "include/llvm/Target/AMDGPU/Disassembler/*.inc", "lib/Target/AMDGPU/Disassembler/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/AMDGPU"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/AMDGPU"], deps = [ ":amdgpu_desc", ":amdgpu_info", @@ -780,8 +691,9 @@ cc_library( "include/llvm/Target/AMDGPU/TargetInfo/*.inc", "lib/Target/AMDGPU/TargetInfo/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/AMDGPU"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/AMDGPU"], deps = [ + ":amdgpu_r600_target_gen", ":amdgpu_target_gen", ":config", ":core", @@ -802,8 +714,9 @@ cc_library( "include/llvm/Target/AMDGPU/Utils/*.inc", "lib/Target/AMDGPU/Utils/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/AMDGPU"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/AMDGPU"], deps = [ + ":amdgpu_r600_target_gen", ":amdgpu_target_gen", ":config", ":core", @@ -825,7 +738,7 @@ cc_library( "include/llvm/Target/AMDGPU/AsmParser/*.inc", "lib/Target/AMDGPU/AsmParser/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/AMDGPU"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/AMDGPU"], deps = [ ":amdgpu_desc", ":amdgpu_info", @@ -850,7 +763,7 @@ cc_library( "include/llvm/Target/AMDGPU/InstPrinter/*.inc", "lib/Target/AMDGPU/InstPrinter/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/AMDGPU"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/AMDGPU"], deps = [ ":amdgpu_utils", ":config", @@ -872,7 +785,7 @@ cc_library( "include/llvm/Target/AMDGPU/*.inc", "lib/Target/AMDGPU/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/AMDGPU"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/AMDGPU"], deps = [ ":amdgpu_asm_printer", ":amdgpu_desc", @@ -908,7 +821,7 @@ cc_library( "include/llvm/Target/ARM/AsmParser/*.inc", "lib/Target/ARM/AsmParser/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/ARM"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/ARM"], deps = [ ":arm_desc", ":arm_info", @@ -934,7 +847,7 @@ cc_library( "lib/Target/ARM/*.h", "lib/Target/ARM/InstPrinter/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/ARM"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/ARM"], deps = [ ":arm_info", ":arm_target_gen", @@ -958,7 +871,7 @@ cc_library( "include/llvm/Target/ARM/*.inc", "lib/Target/ARM/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/ARM"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/ARM"], deps = [ ":analysis", ":arm_asm_printer", @@ -975,6 +888,7 @@ cc_library( ":selection_dag", ":support", ":target", + ":transform_utils", ], ) @@ -993,14 +907,15 @@ cc_library( "include/llvm/Target/ARM/MCTargetDesc/*.inc", "lib/Target/ARM/MCTargetDesc/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/ARM"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/ARM"], deps = [ ":arm_asm_printer", ":arm_info", ":arm_target_gen", ":attributes_gen", ":config", - ":intrinsics_gen", + ":intrinsic_enums_gen", + ":intrinsics_impl_gen", ":mc", ":mc_disassembler", ":support", @@ -1020,7 +935,7 @@ cc_library( "include/llvm/Target/ARM/Disassembler/*.inc", "lib/Target/ARM/Disassembler/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/ARM"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/ARM"], deps = [ ":arm_desc", ":arm_info", @@ -1045,7 +960,7 @@ cc_library( "include/llvm/Target/ARM/TargetInfo/*.inc", "lib/Target/ARM/TargetInfo/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/ARM"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/ARM"], deps = [ ":arm_target_gen", ":config", @@ -1068,7 +983,7 @@ cc_library( "include/llvm/Target/ARM/Utils/*.inc", "lib/Target/ARM/Utils/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/ARM"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/ARM"], deps = [ ":arm_target_gen", ":config", @@ -1090,6 +1005,7 @@ cc_library( "include/llvm/AsmParser/*.def", "include/llvm/AsmParser/*.inc", ]), + copts = llvm_copts, deps = [ ":binary_format", ":config", @@ -1112,6 +1028,7 @@ cc_library( "include/llvm/CodeGen/AsmPrinter/*.inc", "lib/CodeGen/AsmPrinter/*.def", ]), + copts = llvm_copts, deps = [ ":analysis", ":binary_format", @@ -1142,6 +1059,7 @@ cc_library( "include/llvm/BinaryFormat/ELFRelocs/*.def", "include/llvm/BinaryFormat/WasmRelocs/*.def", ]), + copts = llvm_copts, deps = [ ":config", ":support", @@ -1162,6 +1080,7 @@ cc_library( "include/llvm/Bitcode/Reader/*.inc", "include/llvm/Bitcode/BitstreamReader.h", ]), + copts = llvm_copts, deps = [ ":config", ":core", @@ -1185,6 +1104,7 @@ cc_library( "include/llvm/Bitcode/BitcodeWriterPass.h", "include/llvm/Bitcode/BitstreamWriter.h", ]), + copts = llvm_copts, deps = [ ":analysis", ":config", @@ -1209,6 +1129,7 @@ cc_library( "include/llvm/CodeGen/*.inc", "include/llvm/CodeGen/**/*.h", ]), + copts = llvm_copts, deps = [ ":analysis", ":bit_reader", @@ -1246,12 +1167,14 @@ cc_library( "include/llvm/*.h", "include/llvm/Analysis/*.def", ]), + copts = llvm_copts, deps = [ ":attributes_compat_gen", ":attributes_gen", ":binary_format", ":config", - ":intrinsics_gen", + ":intrinsic_enums_gen", + ":intrinsics_impl_gen", ":support", ], ) @@ -1269,6 +1192,7 @@ cc_library( "include/llvm/DebugInfo/CodeView/*.def", "include/llvm/DebugInfo/CodeView/*.inc", ]), + copts = llvm_copts, deps = [ ":binary_format", ":config", @@ -1290,6 +1214,7 @@ cc_library( "include/llvm/DebugInfo/MSF/*.def", "include/llvm/DebugInfo/MSF/*.inc", ]), + copts = llvm_copts, deps = [ ":config", ":support", @@ -1309,6 +1234,7 @@ cc_library( "include/llvm/Demangle/*.def", "include/llvm/Demangle/*.inc", ]), + copts = llvm_copts, deps = [":config"], ) @@ -1325,6 +1251,7 @@ cc_library( "include/llvm/ExecutionEngine/*.def", "include/llvm/ExecutionEngine/*.inc", ]), + copts = llvm_copts, deps = [ ":config", ":core", @@ -1349,6 +1276,7 @@ cc_library( "include/llvm/CodeGen/GlobalISel/*.def", "include/llvm/CodeGen/GlobalISel/*.inc", ]), + copts = llvm_copts, deps = [ ":analysis", ":code_gen", @@ -1378,6 +1306,7 @@ cc_library( "include/llvm/Transforms/InstrProfiling.h", "include/llvm/Transforms/PGOInstrumentation.h", ]), + copts = llvm_copts, deps = [ ":analysis", ":config", @@ -1402,10 +1331,12 @@ cc_library( "include/llvm/Transforms/InstCombine/*.def", "include/llvm/Transforms/InstCombine/*.inc", ]), + copts = llvm_copts, deps = [ ":analysis", ":config", ":core", + ":instcombine_transforms_gen", ":support", ":transform_utils", ], @@ -1427,6 +1358,7 @@ cc_library( "include/llvm/Transforms/IPO/*.def", "include/llvm/Transforms/IPO/*.inc", ]), + copts = llvm_copts, deps = [ ":aggressive_inst_combine", ":analysis", @@ -1460,6 +1392,7 @@ cc_library( "include/llvm/IRReader/*.def", "include/llvm/IRReader/*.inc", ]), + copts = llvm_copts, deps = [ ":asm_parser", ":bit_reader", @@ -1482,6 +1415,7 @@ cc_library( "include/llvm/Linker/*.def", "include/llvm/Linker/*.inc", ]), + copts = llvm_copts, deps = [ ":config", ":core", @@ -1503,6 +1437,7 @@ cc_library( "include/llvm/MC/*.def", "include/llvm/MC/*.inc", ]), + copts = llvm_copts, deps = [ ":binary_format", ":config", @@ -1524,6 +1459,7 @@ cc_library( "include/llvm/MC/MCDisassembler/*.def", "include/llvm/MC/MCDisassembler/*.inc", ]), + copts = llvm_copts, deps = [ ":config", ":mc", @@ -1544,6 +1480,7 @@ cc_library( "include/llvm/MC/MCParser/*.def", "include/llvm/MC/MCParser/*.inc", ]), + copts = llvm_copts, deps = [ ":config", ":mc", @@ -1564,7 +1501,7 @@ cc_library( "include/llvm/Target/NVPTX/InstPrinter/*.inc", "lib/Target/NVPTX/InstPrinter/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/NVPTX"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/NVPTX"], deps = [ "nvptx_target_gen", ":attributes_gen", @@ -1588,7 +1525,7 @@ cc_library( "include/llvm/Target/NVPTX/*.inc", "lib/Target/NVPTX/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/NVPTX"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/NVPTX"], deps = [ ":analysis", ":asm_printer", @@ -1622,7 +1559,7 @@ cc_library( "include/llvm/Target/NVPTX/MCTargetDesc/*.inc", "lib/Target/NVPTX/MCTargetDesc/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/NVPTX"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/NVPTX"], deps = [ "nvptx_target_gen", ":config", @@ -1648,7 +1585,7 @@ cc_library( "lib/Target/NVPTX/NVPTX.h", "lib/Target/NVPTX/TargetInfo/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/NVPTX"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/NVPTX"], deps = [ "nvptx_target_gen", ":attributes_gen", @@ -1672,6 +1609,7 @@ cc_library( "include/llvm/Object/*.def", "include/llvm/Object/*.inc", ]), + copts = llvm_copts, deps = [ ":binary_format", ":bit_reader", @@ -1697,6 +1635,7 @@ cc_library( "include/llvm/Transforms/ObjCARC/*.def", "include/llvm/Transforms/ObjCARC/*.inc", ]), + copts = llvm_copts, deps = [ ":analysis", ":config", @@ -1719,13 +1658,16 @@ cc_library( "include/llvm/ExecutionEngine/Orc/*.def", "include/llvm/ExecutionEngine/Orc/*.inc", ]), + copts = llvm_copts, deps = [ ":config", ":core", ":execution_engine", + ":mc", ":object", ":runtime_dyld", ":support", + ":target", ":transform_utils", ], ) @@ -1743,7 +1685,7 @@ cc_library( "include/llvm/Target/PowerPC/AsmParser/*.inc", "lib/Target/PowerPC/AsmParser/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/PowerPC"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/PowerPC"], deps = [ ":config", ":mc", @@ -1767,11 +1709,12 @@ cc_library( "include/llvm/Target/PowerPC/InstPrinter/*.inc", "lib/Target/PowerPC/InstPrinter/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/PowerPC"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/PowerPC"], deps = [ ":attributes_gen", ":config", - ":intrinsics_gen", + ":intrinsic_enums_gen", + ":intrinsics_impl_gen", ":mc", ":powerpc_info", ":powerpc_target_gen", @@ -1792,7 +1735,7 @@ cc_library( "include/llvm/Target/PowerPC/*.inc", "lib/Target/PowerPC/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/PowerPC"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/PowerPC"], deps = [ ":analysis", ":asm_printer", @@ -1824,11 +1767,12 @@ cc_library( "include/llvm/Target/PowerPC/MCTargetDesc/*.inc", "lib/Target/PowerPC/MCTargetDesc/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/PowerPC"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/PowerPC"], deps = [ ":attributes_gen", ":config", - ":intrinsics_gen", + ":intrinsic_enums_gen", + ":intrinsics_impl_gen", ":mc", ":powerpc_asm_printer", ":powerpc_info", @@ -1850,7 +1794,7 @@ cc_library( "include/llvm/Target/PowerPC/Disassembler/*.inc", "lib/Target/PowerPC/Disassembler/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/PowerPC"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/PowerPC"], deps = [ ":config", ":mc_disassembler", @@ -1874,12 +1818,11 @@ cc_library( "lib/Target/PowerPC/PPC*.h", "lib/Target/PowerPC/TargetInfo/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/PowerPC"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/PowerPC"], deps = [ ":attributes_gen", ":config", ":core", - ":intrinsics_gen", ":powerpc_target_gen", ":support", ":target", @@ -1899,6 +1842,7 @@ cc_library( "include/llvm/ProfileData/*.def", "include/llvm/ProfileData/*.inc", ]), + copts = llvm_copts, deps = [ ":config", ":core", @@ -1927,6 +1871,7 @@ cc_library( "include/llvm/ExecutionEngine/RTDyldMemoryManager.h", "include/llvm/ExecutionEngine/RuntimeDyld*.h", ]), + copts = llvm_copts, deps = [ ":config", ":mc", @@ -1954,6 +1899,7 @@ cc_library( "include/llvm/Transforms/IPO.h", "include/llvm/Transforms/IPO/SCCP.h", ]), + copts = llvm_copts, deps = [ ":aggressive_inst_combine", ":analysis", @@ -1979,6 +1925,7 @@ cc_library( "include/llvm/CodeGen/SelectionDAG/*.def", "include/llvm/CodeGen/SelectionDAG/*.inc", ]), + copts = llvm_copts, deps = [ ":analysis", ":code_gen", @@ -1997,14 +1944,12 @@ cc_library( "lib/Support/*.c", "lib/Support/*.cpp", "lib/Support/*.inc", - "lib/Support/Unix/*.inc", - "lib/Support/Unix/*.h", "include/llvm-c/*.h", "include/llvm/CodeGen/MachineValueType.h", "include/llvm/BinaryFormat/COFF.h", "include/llvm/BinaryFormat/MachO.h", "lib/Support/*.h", - ]), + ]) + llvm_support_platform_specific_srcs_glob(), hdrs = glob([ "include/llvm/Support/*.h", "include/llvm/Support/*.def", @@ -2014,10 +1959,9 @@ cc_library( "include/llvm/Support/WasmRelocs/*.def", ]) + [ "include/llvm/BinaryFormat/MachO.def", - "include/llvm/Support/DataTypes.h", "include/llvm/Support/VCSRevision.h", - "include/llvm/ExecutionEngine/ObjectMemoryBuffer.h", ], + copts = llvm_copts, deps = [ ":config", ":demangle", @@ -2040,6 +1984,7 @@ cc_library( "include/llvm/TableGen/*.inc", "include/llvm/Target/*.def", ]), + copts = llvm_copts, deps = [ ":config", ":mc", @@ -2063,7 +2008,9 @@ cc_library( "include/llvm/Target/*.def", "include/llvm/Target/*.inc", "include/llvm/CodeGen/*.def", + "include/llvm/CodeGen/*.inc", ]), + copts = llvm_copts, deps = [ ":analysis", ":config", @@ -2088,6 +2035,7 @@ cc_library( "include/llvm/Transforms/Utils/*.def", "include/llvm/Transforms/Utils/*.inc", ]), + copts = llvm_copts, deps = [ ":analysis", ":config", @@ -2111,6 +2059,7 @@ cc_library( "include/llvm/Transforms/Vectorize/*.inc", "include/llvm/Transforms/Vectorize.h", ]), + copts = llvm_copts, deps = [ ":analysis", ":config", @@ -2134,7 +2083,7 @@ cc_library( "include/llvm/Target/X86/AsmParser/*.inc", "lib/Target/X86/AsmParser/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/X86"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/X86"], deps = [ ":config", ":mc", @@ -2159,7 +2108,7 @@ cc_library( "include/llvm/Target/X86/InstPrinter/*.inc", "lib/Target/X86/InstPrinter/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/X86"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/X86"], deps = [ ":config", ":mc", @@ -2183,7 +2132,7 @@ cc_library( "include/llvm/Target/X86/*.inc", "lib/Target/X86/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/X86"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/X86"], deps = [ ":analysis", ":asm_printer", @@ -2216,7 +2165,7 @@ cc_library( "include/llvm/Target/X86/MCTargetDesc/*.inc", "lib/Target/X86/MCTargetDesc/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/X86"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/X86"], deps = [ ":config", ":mc", @@ -2241,7 +2190,7 @@ cc_library( "include/llvm/Target/X86/Disassembler/*.inc", "lib/Target/X86/Disassembler/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/X86"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/X86"], deps = [ ":config", ":mc_disassembler", @@ -2264,7 +2213,7 @@ cc_library( "include/llvm/Target/X86/TargetInfo/*.inc", "lib/Target/X86/TargetInfo/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/X86"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/X86"], deps = [ ":config", ":mc", @@ -2286,7 +2235,7 @@ cc_library( "include/llvm/Target/X86/Utils/*.inc", "lib/Target/X86/Utils/*.h", ]), - copts = ["-Iexternal/llvm/lib/Target/X86"], + copts = llvm_copts + ["-Iexternal/llvm/lib/Target/X86"], deps = [ ":code_gen", ":config", diff --git a/third_party/llvm/llvm.bzl b/third_party/llvm/llvm.bzl index 0efcf319bd..54ca86f327 100644 --- a/third_party/llvm/llvm.bzl +++ b/third_party/llvm/llvm.bzl @@ -7,101 +7,408 @@ TODO(chandlerc): Currently this expresses include-based dependencies as correctly understood by the build system. """ +def _dict_add(*dictionaries): + """Returns a new `dict` that has all the entries of the given dictionaries. + + If the same key is present in more than one of the input dictionaries, the + last of them in the argument list overrides any earlier ones. + + This function is designed to take zero or one arguments as well as multiple + dictionaries, so that it follows arithmetic identities and callers can avoid + special cases for their inputs: the sum of zero dictionaries is the empty + dictionary, and the sum of a single dictionary is a copy of itself. + + Re-implemented here to avoid adding a dependency on skylib. + + Args: + *dictionaries: Zero or more dictionaries to be added. + + Returns: + A new `dict` that has all the entries of the given dictionaries. + """ + result = {} + for d in dictionaries: + result.update(d) + return result + def gentbl(name, tblgen, td_file, td_srcs, tbl_outs, library = True, **kwargs): - """gentbl() generates tabular code from a table definition file. - - Args: - name: The name of the build rule for use in dependencies. - tblgen: The binary used to produce the output. - td_file: The primary table definitions file. - td_srcs: A list of table definition files included transitively. - tbl_outs: A list of tuples (opts, out), where each opts is a string of - options passed to tblgen, and the out is the corresponding output file - produced. - library: Whether to bundle the generated files into a library. - **kwargs: Keyword arguments to pass to subsidiary cc_library() rule. - """ - if td_file not in td_srcs: - td_srcs += [td_file] - includes = [] - for (opts, out) in tbl_outs: - outdir = out[:out.rindex("/")] - if outdir not in includes: - includes.append(outdir) - rule_suffix = "_".join(opts.replace("-", "_").replace("=", "_").split(" ")) - native.genrule( - name="%s_%s_genrule" % (name, rule_suffix), - srcs=td_srcs, - outs=[out], - tools=[tblgen], - message="Generating code from table: %s" % td_file, - cmd=(("$(location %s) " + "-I external/llvm/include " + - "-I external/llvm/tools/clang/include " + - "-I $$(dirname $(location %s)) " + "%s $(location %s) -o $@") % ( - tblgen, td_file, opts, td_file))) - # For now, all generated files can be assumed to comprise public interfaces. - # If this is not true, you should specify library = False - # and list the generated '.inc' files in "srcs". - if library: - native.cc_library(name=name, textual_hdrs=[f for (_, f) in tbl_outs], - includes=includes, **kwargs) + """gentbl() generates tabular code from a table definition file. + + Args: + name: The name of the build rule for use in dependencies. + tblgen: The binary used to produce the output. + td_file: The primary table definitions file. + td_srcs: A list of table definition files included transitively. + tbl_outs: A list of tuples (opts, out), where each opts is a string of + options passed to tblgen, and the out is the corresponding output file + produced. + library: Whether to bundle the generated files into a library. + **kwargs: Keyword arguments to pass to subsidiary cc_library() rule. + """ + if td_file not in td_srcs: + td_srcs += [td_file] + includes = [] + for (opts, out) in tbl_outs: + outdir = out[:out.rindex("/")] + if outdir not in includes: + includes.append(outdir) + rule_suffix = "_".join(opts.replace("-", "_").replace("=", "_").split(" ")) + native.genrule( + name = "%s_%s_genrule" % (name, rule_suffix), + srcs = td_srcs, + outs = [out], + tools = [tblgen], + message = "Generating code from table: %s" % td_file, + cmd = (("$(location %s) " + "-I external/llvm/include " + + "-I external/llvm/tools/clang/include " + + "-I $$(dirname $(location %s)) " + "%s $(location %s) -o $@") % ( + tblgen, + td_file, + opts, + td_file, + )), + ) + + # For now, all generated files can be assumed to comprise public interfaces. + # If this is not true, you should specify library = False + # and list the generated '.inc' files in "srcs". + if library: + native.cc_library( + name = name, + textual_hdrs = [f for (_, f) in tbl_outs], + includes = includes, + **kwargs + ) def llvm_target_cmake_vars(native_arch, target_triple): - return { - "LLVM_HOST_TRIPLE": target_triple, - "LLVM_DEFAULT_TARGET_TRIPLE": target_triple, - "LLVM_NATIVE_ARCH": native_arch, - } + return { + "LLVM_HOST_TRIPLE": target_triple, + "LLVM_DEFAULT_TARGET_TRIPLE": target_triple, + "LLVM_NATIVE_ARCH": native_arch, + } def _quote(s): - """Quotes the given string for use in a shell command. - - This function double-quotes the given string (in case it contains spaces or - other special characters) and escapes any special characters (dollar signs, - double-quotes, and backslashes) that may be present. - - Args: - s: The string to quote. - Returns: - An escaped and quoted version of the string that can be passed to a shell - command. - """ - return ('"' + - s.replace("\\", "\\\\").replace("$", "\\$").replace('"', '\\"') + - '"') + """Quotes the given string for use in a shell command. + + This function double-quotes the given string (in case it contains spaces or + other special characters) and escapes any special characters (dollar signs, + double-quotes, and backslashes) that may be present. + + Args: + s: The string to quote. + + Returns: + An escaped and quoted version of the string that can be passed to a shell + command. + """ + return ('"' + + s.replace("\\", "\\\\").replace("$", "\\$").replace('"', '\\"') + + '"') def cmake_var_string(cmake_vars): - """Converts a dictionary to an input suitable for expand_cmake_vars. + """Converts a dictionary to an input suitable for expand_cmake_vars. - Ideally we would jist stringify in the expand_cmake_vars() rule, but select() - interacts badly with genrules. + Ideally we would jist stringify in the expand_cmake_vars() rule, but select() + interacts badly with genrules. - TODO(phawkins): replace the genrule() with native rule and delete this rule. + TODO(phawkins): replace the genrule() with native rule and delete this rule. - Args: - cmake_vars: a dictionary with string keys and values that are convertable to - strings. - """ - return " ".join([_quote("{}={}".format(k, str(v))) - for (k, v) in cmake_vars.items()]) + Args: + cmake_vars: a dictionary with string keys and values that are convertable to + strings. + + Returns: + cmake_vars in a form suitable for passing to expand_cmake_vars. + """ + return " ".join([ + _quote("{}={}".format(k, str(v))) + for (k, v) in cmake_vars.items() + ]) def expand_cmake_vars(name, src, dst, cmake_vars): - """Expands #cmakedefine, #cmakedefine01, and CMake variables in a text file. - - Args: - name: the name of the rule - src: the input of the rule - dst: the output of the rule - cmake_vars: a string containing the CMake variables, as generated by - cmake_var_string. - """ - expand_cmake_vars_tool = Label("@org_tensorflow//third_party/llvm:expand_cmake_vars") - native.genrule( - name = name, - srcs = [src], - tools = [expand_cmake_vars_tool], - outs = [dst], - cmd = ("$(location {}) ".format(expand_cmake_vars_tool) + cmake_vars + - "< $< > $@") - ) + """Expands #cmakedefine, #cmakedefine01, and CMake variables in a text file. + + Args: + name: the name of the rule + src: the input of the rule + dst: the output of the rule + cmake_vars: a string containing the CMake variables, as generated by + cmake_var_string. + """ + expand_cmake_vars_tool = Label("@org_tensorflow//third_party/llvm:expand_cmake_vars") + native.genrule( + name = name, + srcs = [src], + tools = [expand_cmake_vars_tool], + outs = [dst], + cmd = ("$(location {}) ".format(expand_cmake_vars_tool) + cmake_vars + + "< $< > $@"), + ) + +# TODO(phawkins): the set of CMake variables was hardcoded for expediency. +# However, we should really detect many of these via configure-time tests. + +# The set of CMake variables common to all targets. +cmake_vars = { + # LLVM features + "ENABLE_BACKTRACES": 1, + "LLVM_BINDIR": "/dev/null", + "LLVM_DISABLE_ABI_BREAKING_CHECKS_ENFORCING": 0, + "LLVM_ENABLE_ABI_BREAKING_CHECKS": 0, + "LLVM_ENABLE_THREADS": 1, + "LLVM_ENABLE_ZLIB": 1, + "LLVM_HAS_ATOMICS": 1, + "LLVM_INCLUDEDIR": "/dev/null", + "LLVM_INFODIR": "/dev/null", + "LLVM_MANDIR": "/dev/null", + "LLVM_NATIVE_TARGET": 1, + "LLVM_NATIVE_TARGETINFO": 1, + "LLVM_NATIVE_TARGETMC": 1, + "LLVM_NATIVE_ASMPRINTER": 1, + "LLVM_NATIVE_ASMPARSER": 1, + "LLVM_NATIVE_DISASSEMBLER": 1, + "LLVM_PREFIX": "/dev/null", + "LLVM_VERSION_MAJOR": 0, + "LLVM_VERSION_MINOR": 0, + "LLVM_VERSION_PATCH": 0, + "PACKAGE_NAME": "llvm", + "PACKAGE_STRING": "llvm tensorflow-trunk", + "PACKAGE_VERSION": "tensorflow-trunk", + "RETSIGTYPE": "void", +} + +# The set of CMake variables common to POSIX targets. +posix_cmake_vars = { + # Headers + "HAVE_DIRENT_H": 1, + "HAVE_DLFCN_H": 1, + "HAVE_ERRNO_H": 1, + "HAVE_EXECINFO_H": 1, + "HAVE_FCNTL_H": 1, + "HAVE_INTTYPES_H": 1, + "HAVE_PTHREAD_H": 1, + "HAVE_SIGNAL_H": 1, + "HAVE_STDINT_H": 1, + "HAVE_SYS_IOCTL_H": 1, + "HAVE_SYS_MMAN_H": 1, + "HAVE_SYS_PARAM_H": 1, + "HAVE_SYS_RESOURCE_H": 1, + "HAVE_SYS_STAT_H": 1, + "HAVE_SYS_TIME_H": 1, + "HAVE_SYS_TYPES_H": 1, + "HAVE_TERMIOS_H": 1, + "HAVE_UNISTD_H": 1, + "HAVE_ZLIB_H": 1, + + # Features + "HAVE_BACKTRACE": 1, + "BACKTRACE_HEADER": "execinfo.h", + "HAVE_DLOPEN": 1, + "HAVE_FUTIMES": 1, + "HAVE_GETCWD": 1, + "HAVE_GETPAGESIZE": 1, + "HAVE_GETRLIMIT": 1, + "HAVE_GETRUSAGE": 1, + "HAVE_GETTIMEOFDAY": 1, + "HAVE_INT64_T": 1, + "HAVE_ISATTY": 1, + "HAVE_LIBEDIT": 1, + "HAVE_LIBPTHREAD": 1, + "HAVE_LIBZ": 1, + "HAVE_MKDTEMP": 1, + "HAVE_MKSTEMP": 1, + "HAVE_MKTEMP": 1, + "HAVE_PREAD": 1, + "HAVE_PTHREAD_GETSPECIFIC": 1, + "HAVE_PTHREAD_MUTEX_LOCK": 1, + "HAVE_PTHREAD_RWLOCK_INIT": 1, + "HAVE_REALPATH": 1, + "HAVE_SBRK": 1, + "HAVE_SETENV": 1, + "HAVE_SETRLIMIT": 1, + "HAVE_SIGALTSTACK": 1, + "HAVE_STRERROR": 1, + "HAVE_STRERROR_R": 1, + "HAVE_STRTOLL": 1, + "HAVE_SYSCONF": 1, + "HAVE_UINT64_T": 1, + "HAVE__UNWIND_BACKTRACE": 1, + + # LLVM features + "LLVM_ON_UNIX": 1, + "LTDL_SHLIB_EXT": ".so", +} + +# CMake variables specific to the Linux platform +linux_cmake_vars = { + "HAVE_MALLOC_H": 1, + "HAVE_LINK_H": 1, + "HAVE_MALLINFO": 1, + "HAVE_FUTIMENS": 1, +} + +# CMake variables specific to the Darwin (Mac OS X) platform. +darwin_cmake_vars = { + "HAVE_MALLOC_MALLOC_H": 1, +} + +# CMake variables specific to the Windows platform. +win32_cmake_vars = { + # Headers + "HAVE_ERRNO_H": 1, + "HAVE_EXECINFO_H": 1, + "HAVE_FCNTL_H": 1, + "HAVE_FENV_H": 1, + "HAVE_INTTYPES_H": 1, + "HAVE_MALLOC_H": 1, + "HAVE_SIGNAL_H": 1, + "HAVE_STDINT_H": 1, + "HAVE_SYS_STAT_H": 1, + "HAVE_SYS_TYPES_H": 1, + "HAVE_ZLIB_H": 1, + + # Features + "BACKTRACE_HEADER": "execinfo.h", + "HAVE_GETCWD": 1, + "HAVE_INT64_T": 1, + "HAVE_STRERROR": 1, + "HAVE_STRTOLL": 1, + "HAVE_SYSCONF": 1, + "HAVE_UINT64_T": 1, + "HAVE__CHSIZE_S": 1, + "HAVE___CHKSTK": 1, + + # MSVC specific + "stricmp": "_stricmp", + "strdup": "_strdup", + + # LLVM features + "LTDL_SHLIB_EXT": ".dll", +} + +# Select a set of CMake variables based on the platform. +# TODO(phawkins): use a better method to select the right host triple, rather +# than hardcoding x86_64. +llvm_all_cmake_vars = select({ + "@org_tensorflow//tensorflow:darwin": cmake_var_string( + _dict_add( + cmake_vars, + llvm_target_cmake_vars("X86", "x86_64-apple-darwin"), + posix_cmake_vars, + darwin_cmake_vars, + ), + ), + "@org_tensorflow//tensorflow:linux_ppc64le": cmake_var_string( + _dict_add( + cmake_vars, + llvm_target_cmake_vars("PowerPC", "powerpc64le-unknown-linux_gnu"), + posix_cmake_vars, + linux_cmake_vars, + ), + ), + "@org_tensorflow//tensorflow:windows": cmake_var_string( + _dict_add( + cmake_vars, + llvm_target_cmake_vars("X86", "x86_64-pc-win32"), + win32_cmake_vars, + ), + ), + "//conditions:default": cmake_var_string( + _dict_add( + cmake_vars, + llvm_target_cmake_vars("X86", "x86_64-unknown-linux_gnu"), + posix_cmake_vars, + linux_cmake_vars, + ), + ), +}) + +llvm_linkopts = select({ + "@org_tensorflow//tensorflow:windows": [], + "//conditions:default": ["-ldl", "-lm", "-lpthread"], +}) + +llvm_defines = select({ + "@org_tensorflow//tensorflow:windows": [ + "_CRT_SECURE_NO_DEPRECATE", + "_CRT_SECURE_NO_WARNINGS", + "_CRT_NONSTDC_NO_DEPRECATE", + "_CRT_NONSTDC_NO_WARNINGS", + "_SCL_SECURE_NO_DEPRECATE", + "_SCL_SECURE_NO_WARNINGS", + "UNICODE", + "_UNICODE", + ], + "//conditions:default": ["_DEBUG"], +}) + [ + "LLVM_ENABLE_STATS", + "__STDC_LIMIT_MACROS", + "__STDC_CONSTANT_MACROS", + "__STDC_FORMAT_MACROS", + "LLVM_BUILD_GLOBAL_ISEL", +] + +llvm_copts = select({ + "@org_tensorflow//tensorflow:windows": [ + "-Zc:inline", + "-Zc:strictStrings", + "-Zc:rvalueCast", + "-Oi", + "-wd4141", + "-wd4146", + "-wd4180", + "-wd4244", + "-wd4258", + "-wd4267", + "-wd4291", + "-wd4345", + "-wd4351", + "-wd4355", + "-wd4456", + "-wd4457", + "-wd4458", + "-wd4459", + "-wd4503", + "-wd4624", + "-wd4722", + "-wd4800", + "-wd4100", + "-wd4127", + "-wd4512", + "-wd4505", + "-wd4610", + "-wd4510", + "-wd4702", + "-wd4245", + "-wd4706", + "-wd4310", + "-wd4701", + "-wd4703", + "-wd4389", + "-wd4611", + "-wd4805", + "-wd4204", + "-wd4577", + "-wd4091", + "-wd4592", + "-wd4319", + "-wd4324", + "-w14062", + "-we4238", + ], + "//conditions:default": [], +}) + +# Platform specific sources for libSupport. +def llvm_support_platform_specific_srcs_glob(): + return select({ + "@org_tensorflow//tensorflow:windows": native.glob([ + "lib/Support/Windows/*.inc", + "lib/Support/Windows/*.h", + ]), + "//conditions:default": native.glob([ + "lib/Support/Unix/*.inc", + "lib/Support/Unix/*.h", + ]), + }) diff --git a/third_party/lmdb.BUILD b/third_party/lmdb.BUILD index 9b3e1d97c8..f36a698ee3 100644 --- a/third_party/lmdb.BUILD +++ b/third_party/lmdb.BUILD @@ -20,7 +20,6 @@ cc_library( ], linkopts = select({ ":windows": ["-DEFAULTLIB:advapi32.lib"], # InitializeSecurityDescriptor, SetSecurityDescriptorDacl - ":windows_msvc": ["-DEFAULTLIB:advapi32.lib"], "//conditions:default": ["-lpthread"], }), visibility = ["//visibility:public"], @@ -30,8 +29,3 @@ config_setting( name = "windows", values = {"cpu": "x64_windows"}, ) - -config_setting( - name = "windows_msvc", - values = {"cpu": "x64_windows_msvc"}, -) diff --git a/third_party/mkl/BUILD b/third_party/mkl/BUILD index b27d341404..efff7fd51b 100644 --- a/third_party/mkl/BUILD +++ b/third_party/mkl/BUILD @@ -1,11 +1,29 @@ licenses(["notice"]) # 3-Clause BSD -exports_files(["LICENSE"]) - config_setting( name = "using_mkl", + define_values = { + "using_mkl": "true", + }, + visibility = ["//visibility:public"], +) + +config_setting( + name = "using_mkl_ml_only", + define_values = { + "using_mkl": "true", + "using_mkl_ml_only": "true", + }, + visibility = ["//visibility:public"], +) + +config_setting( + name = "using_mkl_lnx_x64", + define_values = { + "using_mkl": "true", + }, values = { - "define": "using_mkl=true", + "cpu": "k8", }, visibility = ["//visibility:public"], ) @@ -15,12 +33,39 @@ load( "if_mkl", ) +filegroup( + name = "LICENSE", + srcs = ["MKL_LICENSE"] + select({ + "@org_tensorflow//tensorflow:linux_x86_64": [ + "@mkl_linux//:LICENSE", + ], + "@org_tensorflow//tensorflow:darwin": [ + "@mkl_darwin//:LICENSE", + ], + "@org_tensorflow//tensorflow:windows": [ + "@mkl_windows//:LICENSE", + ], + "//conditions:default": [], + }), + visibility = ["//visibility:public"], +) + cc_library( name = "intel_binary_blob", - srcs = if_mkl([ - "@mkl//:libmklml_intel.so", - "@mkl//:libiomp5.so", - ]), visibility = ["//visibility:public"], - deps = ["@mkl//:mkl_headers"], + deps = select({ + "@org_tensorflow//tensorflow:linux_x86_64": [ + "@mkl_linux//:mkl_headers", + "@mkl_linux//:mkl_libs_linux", + ], + "@org_tensorflow//tensorflow:darwin": [ + "@mkl_darwin//:mkl_headers", + "@mkl_darwin//:mkl_libs_darwin", + ], + "@org_tensorflow//tensorflow:windows": [ + "@mkl_windows//:mkl_headers", + "@mkl_windows//:mkl_libs_windows", + ], + "//conditions:default": [], + }), ) diff --git a/third_party/mkl/MKL_LICENSE b/third_party/mkl/MKL_LICENSE new file mode 100644 index 0000000000..9c8f3ea087 --- /dev/null +++ b/third_party/mkl/MKL_LICENSE @@ -0,0 +1,201 @@ + Apache License + Version 2.0, January 2004 + http://www.apache.org/licenses/ + + TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION + + 1. Definitions. + + "License" shall mean the terms and conditions for use, reproduction, + and distribution as defined by Sections 1 through 9 of this document. + + "Licensor" shall mean the copyright owner or entity authorized by + the copyright owner that is granting the License. + + "Legal Entity" shall mean the union of the acting entity and all + other entities that control, are controlled by, or are under common + control with that entity. For the purposes of this definition, + "control" means (i) the power, direct or indirect, to cause the + direction or management of such entity, whether by contract or + otherwise, or (ii) ownership of fifty percent (50%) or more of the + outstanding shares, or (iii) beneficial ownership of such entity. + + "You" (or "Your") shall mean an individual or Legal Entity + exercising permissions granted by this License. + + "Source" form shall mean the preferred form for making modifications, + including but not limited to software source code, documentation + source, and configuration files. + + "Object" form shall mean any form resulting from mechanical + transformation or translation of a Source form, including but + not limited to compiled object code, generated documentation, + and conversions to other media types. + + "Work" shall mean the work of authorship, whether in Source or + Object form, made available under the License, as indicated by a + copyright notice that is included in or attached to the work + (an example is provided in the Appendix below). + + "Derivative Works" shall mean any work, whether in Source or Object + form, that is based on (or derived from) the Work and for which the + editorial revisions, annotations, elaborations, or other modifications + represent, as a whole, an original work of authorship. For the purposes + of this License, Derivative Works shall not include works that remain + separable from, or merely link (or bind by name) to the interfaces of, + the Work and Derivative Works thereof. + + "Contribution" shall mean any work of authorship, including + the original version of the Work and any modifications or additions + to that Work or Derivative Works thereof, that is intentionally + submitted to Licensor for inclusion in the Work by the copyright owner + or by an individual or Legal Entity authorized to submit on behalf of + the copyright owner. For the purposes of this definition, "submitted" + means any form of electronic, verbal, or written communication sent + to the Licensor or its representatives, including but not limited to + communication on electronic mailing lists, source code control systems, + and issue tracking systems that are managed by, or on behalf of, the + Licensor for the purpose of discussing and improving the Work, but + excluding communication that is conspicuously marked or otherwise + designated in writing by the copyright owner as "Not a Contribution." + + "Contributor" shall mean Licensor and any individual or Legal Entity + on behalf of whom a Contribution has been received by Licensor and + subsequently incorporated within the Work. + + 2. Grant of Copyright License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + copyright license to reproduce, prepare Derivative Works of, + publicly display, publicly perform, sublicense, and distribute the + Work and such Derivative Works in Source or Object form. + + 3. Grant of Patent License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + (except as stated in this section) patent license to make, have made, + use, offer to sell, sell, import, and otherwise transfer the Work, + where such license applies only to those patent claims licensable + by such Contributor that are necessarily infringed by their + Contribution(s) alone or by combination of their Contribution(s) + with the Work to which such Contribution(s) was submitted. If You + institute patent litigation against any entity (including a + cross-claim or counterclaim in a lawsuit) alleging that the Work + or a Contribution incorporated within the Work constitutes direct + or contributory patent infringement, then any patent licenses + granted to You under this License for that Work shall terminate + as of the date such litigation is filed. + + 4. Redistribution. You may reproduce and distribute copies of the + Work or Derivative Works thereof in any medium, with or without + modifications, and in Source or Object form, provided that You + meet the following conditions: + + (a) You must give any other recipients of the Work or + Derivative Works a copy of this License; and + + (b) You must cause any modified files to carry prominent notices + stating that You changed the files; and + + (c) You must retain, in the Source form of any Derivative Works + that You distribute, all copyright, patent, trademark, and + attribution notices from the Source form of the Work, + excluding those notices that do not pertain to any part of + the Derivative Works; and + + (d) If the Work includes a "NOTICE" text file as part of its + distribution, then any Derivative Works that You distribute must + include a readable copy of the attribution notices contained + within such NOTICE file, excluding those notices that do not + pertain to any part of the Derivative Works, in at least one + of the following places: within a NOTICE text file distributed + as part of the Derivative Works; within the Source form or + documentation, if provided along with the Derivative Works; or, + within a display generated by the Derivative Works, if and + wherever such third-party notices normally appear. The contents + of the NOTICE file are for informational purposes only and + do not modify the License. You may add Your own attribution + notices within Derivative Works that You distribute, alongside + or as an addendum to the NOTICE text from the Work, provided + that such additional attribution notices cannot be construed + as modifying the License. + + You may add Your own copyright statement to Your modifications and + may provide additional or different license terms and conditions + for use, reproduction, or distribution of Your modifications, or + for any such Derivative Works as a whole, provided Your use, + reproduction, and distribution of the Work otherwise complies with + the conditions stated in this License. + + 5. Submission of Contributions. Unless You explicitly state otherwise, + any Contribution intentionally submitted for inclusion in the Work + by You to the Licensor shall be under the terms and conditions of + this License, without any additional terms or conditions. + Notwithstanding the above, nothing herein shall supersede or modify + the terms of any separate license agreement you may have executed + with Licensor regarding such Contributions. + + 6. Trademarks. This License does not grant permission to use the trade + names, trademarks, service marks, or product names of the Licensor, + except as required for reasonable and customary use in describing the + origin of the Work and reproducing the content of the NOTICE file. + + 7. Disclaimer of Warranty. Unless required by applicable law or + agreed to in writing, Licensor provides the Work (and each + Contributor provides its Contributions) on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or + implied, including, without limitation, any warranties or conditions + of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A + PARTICULAR PURPOSE. You are solely responsible for determining the + appropriateness of using or redistributing the Work and assume any + risks associated with Your exercise of permissions under this License. + + 8. Limitation of Liability. In no event and under no legal theory, + whether in tort (including negligence), contract, or otherwise, + unless required by applicable law (such as deliberate and grossly + negligent acts) or agreed to in writing, shall any Contributor be + liable to You for damages, including any direct, indirect, special, + incidental, or consequential damages of any character arising as a + result of this License or out of the use or inability to use the + Work (including but not limited to damages for loss of goodwill, + work stoppage, computer failure or malfunction, or any and all + other commercial damages or losses), even if such Contributor + has been advised of the possibility of such damages. + + 9. Accepting Warranty or Additional Liability. While redistributing + the Work or Derivative Works thereof, You may choose to offer, + and charge a fee for, acceptance of support, warranty, indemnity, + or other liability obligations and/or rights consistent with this + License. However, in accepting such obligations, You may act only + on Your own behalf and on Your sole responsibility, not on behalf + of any other Contributor, and only if You agree to indemnify, + defend, and hold each Contributor harmless for any liability + incurred by, or claims asserted against, such Contributor by reason + of your accepting any such warranty or additional liability. + + END OF TERMS AND CONDITIONS + + APPENDIX: How to apply the Apache License to your work. + + To apply the Apache License to your work, attach the following + boilerplate notice, with the fields enclosed by brackets "{}" + replaced with your own identifying information. (Don't include + the brackets!) The text should be enclosed in the appropriate + comment syntax for the file format. We also recommend that a + file or class name and description of purpose be included on the + same "printed page" as the copyright notice for easier + identification within third-party archives. + + Copyright {yyyy} {name of copyright owner} + + Licensed under the Apache License, Version 2.0 (the "License"); + you may not use this file except in compliance with the License. + You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + + Unless required by applicable law or agreed to in writing, software + distributed under the License is distributed on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + See the License for the specific language governing permissions and + limitations under the License.
\ No newline at end of file diff --git a/third_party/mkl/build_defs.bzl b/third_party/mkl/build_defs.bzl index 8b73ddabdd..b645c0fc5c 100644 --- a/third_party/mkl/build_defs.bzl +++ b/third_party/mkl/build_defs.bzl @@ -1,6 +1,9 @@ # -*- Python -*- """Skylark macros for MKL. if_mkl is a conditional to check if MKL is enabled or not. +if_mkl_ml is a conditional to check if MKL-ML is enabled. +if_mkl_ml_only is a conditional to check for MKL-ML-only (no MKL-DNN) mode. +if_mkl_lnx_x64 is a conditional to check for MKL mkl_repository is a repository rule for creating MKL repository rule that can be pointed to either a local folder, or download it from the internet. @@ -8,50 +11,116 @@ mkl_repository depends on the following environment variables: * `TF_MKL_ROOT`: The root folder where a copy of libmkl is located. """ - _TF_MKL_ROOT = "TF_MKL_ROOT" - def if_mkl(if_true, if_false = []): """Shorthand for select()'ing on whether we're building with MKL. - Returns a select statement which evaluates to if_true if we're building - with MKL enabled. Otherwise, the select statement evaluates to if_false. + Args: + if_true: expression to evaluate if building with MKL. + if_false: expression to evaluate if building without MKL. + Returns: + a select evaluating to either if_true or if_false as appropriate. """ return select({ str(Label("//third_party/mkl:using_mkl")): if_true, - "//conditions:default": if_false + "//conditions:default": if_false, }) +def if_mkl_ml(if_true, if_false = []): + """Shorthand for select()'ing on whether we're building with MKL-ML. -def _enable_local_mkl(repository_ctx): - return _TF_MKL_ROOT in repository_ctx.os.environ + Args: + if_true: expression to evaluate if building with MKL-ML. + if_false: expression to evaluate if building without MKL-ML + (i.e. without MKL at all, or with MKL-DNN only). + + Returns: + a select evaluating to either if_true or if_false as appropriate. + """ + return select({ + str(Label("//third_party/mkl_dnn:using_mkl_dnn_only")): if_false, + str(Label("//third_party/mkl:using_mkl")): if_true, + "//conditions:default": if_false, + }) + +def if_mkl_ml_only(if_true, if_false = []): + """Shorthand for select()'ing on whether we're building with MKL-ML only. + + Args: + if_true: expression to evaluate if building with MKL-ML only. + if_false: expression to evaluate if building without MKL, or with MKL-DNN. + + Returns: + a select evaluating to either if_true or if_false as appropriate. + """ + return select({ + str(Label("//third_party/mkl:using_mkl_ml_only")): if_true, + "//conditions:default": if_false, + }) +def if_mkl_lnx_x64(if_true, if_false = []): + """Shorthand to select() on if MKL is on and the target is Linux x86-64. + + Args: + if_true: expression to evaluate if building with MKL is enabled and the + target platform is Linux x86-64. + if_false: expression to evaluate if building without MKL or for a + different platform. + + Returns: + a select evaluating to either if_true or if_false as appropriate. + """ + return select({ + str(Label("//third_party/mkl:using_mkl_lnx_x64")): if_true, + "//conditions:default": if_false, + }) + +def mkl_deps(): + """Shorthand for select() to pull in the correct set of MKL library deps. + + Can pull in MKL-ML, MKL-DNN, both, or neither depending on config settings. + + Returns: + a select evaluating to a list of library dependencies, suitable for + inclusion in the deps attribute of rules. + """ + return select({ + str(Label("//third_party/mkl_dnn:using_mkl_dnn_only")): ["@mkl_dnn"], + str(Label("//third_party/mkl:using_mkl_ml_only")): ["//third_party/mkl:intel_binary_blob"], + str(Label("//third_party/mkl:using_mkl")): [ + "//third_party/mkl:intel_binary_blob", + "@mkl_dnn", + ], + "//conditions:default": [], + }) + +def _enable_local_mkl(repository_ctx): + return _TF_MKL_ROOT in repository_ctx.os.environ def _mkl_autoconf_impl(repository_ctx): - """Implementation of the local_mkl_autoconf repository rule.""" - - if _enable_local_mkl(repository_ctx): - # Symlink lib and include local folders. - mkl_root = repository_ctx.os.environ[_TF_MKL_ROOT] - mkl_lib_path = "%s/lib" % mkl_root - repository_ctx.symlink(mkl_lib_path, "lib") - mkl_include_path = "%s/include" % mkl_root - repository_ctx.symlink(mkl_include_path, "include") - mkl_license_path = "%s/license.txt" % mkl_root - repository_ctx.symlink(mkl_license_path, "license.txt") - else: - # setup remote mkl repository. - repository_ctx.download_and_extract( - repository_ctx.attr.urls, - sha256=repository_ctx.attr.sha256, - stripPrefix=repository_ctx.attr.strip_prefix, - ) - - # Also setup BUILD file. - repository_ctx.symlink(repository_ctx.attr.build_file, "BUILD") + """Implementation of the local_mkl_autoconf repository rule.""" + + if _enable_local_mkl(repository_ctx): + # Symlink lib and include local folders. + mkl_root = repository_ctx.os.environ[_TF_MKL_ROOT] + mkl_lib_path = "%s/lib" % mkl_root + repository_ctx.symlink(mkl_lib_path, "lib") + mkl_include_path = "%s/include" % mkl_root + repository_ctx.symlink(mkl_include_path, "include") + mkl_license_path = "%s/license.txt" % mkl_root + repository_ctx.symlink(mkl_license_path, "license.txt") + else: + # setup remote mkl repository. + repository_ctx.download_and_extract( + repository_ctx.attr.urls, + sha256 = repository_ctx.attr.sha256, + stripPrefix = repository_ctx.attr.strip_prefix, + ) + # Also setup BUILD file. + repository_ctx.symlink(repository_ctx.attr.build_file, "BUILD") mkl_repository = repository_rule( implementation = _mkl_autoconf_impl, diff --git a/third_party/mkl/mkl.BUILD b/third_party/mkl/mkl.BUILD index 8db97232e1..c3a71e4ff9 100644 --- a/third_party/mkl/mkl.BUILD +++ b/third_party/mkl/mkl.BUILD @@ -17,14 +17,29 @@ cc_library( visibility = ["//visibility:public"], ) -filegroup( - name = "libmklml_intel.so", - srcs = ["lib/libmklml_intel.so"], +cc_library( + name = "mkl_libs_linux", + srcs = [ + "lib/libiomp5.so", + "lib/libmklml_intel.so", + ], visibility = ["//visibility:public"], ) -filegroup( - name = "libiomp5.so", - srcs = ["lib/libiomp5.so"], +cc_library( + name = "mkl_libs_darwin", + srcs = [ + "lib/libiomp5.dylib", + "lib/libmklml.dylib", + ], + visibility = ["//visibility:public"], +) + +cc_library( + name = "mkl_libs_windows", + srcs = [ + "lib/libiomp5md.lib", + "lib/mklml.lib", + ], visibility = ["//visibility:public"], ) diff --git a/third_party/mkl_dnn/BUILD b/third_party/mkl_dnn/BUILD index 5b01f6e3e4..3e567fa9fc 100644 --- a/third_party/mkl_dnn/BUILD +++ b/third_party/mkl_dnn/BUILD @@ -1 +1,12 @@ licenses(["notice"]) + +exports_files(["LICENSE"]) + +config_setting( + name = "using_mkl_dnn_only", + define_values = { + "using_mkl": "true", + "using_mkl_dnn_only": "true", + }, + visibility = ["//visibility:public"], +) diff --git a/third_party/mkl_dnn/LICENSE b/third_party/mkl_dnn/LICENSE new file mode 100644 index 0000000000..8dada3edaf --- /dev/null +++ b/third_party/mkl_dnn/LICENSE @@ -0,0 +1,201 @@ + Apache License + Version 2.0, January 2004 + http://www.apache.org/licenses/ + + TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION + + 1. Definitions. + + "License" shall mean the terms and conditions for use, reproduction, + and distribution as defined by Sections 1 through 9 of this document. + + "Licensor" shall mean the copyright owner or entity authorized by + the copyright owner that is granting the License. + + "Legal Entity" shall mean the union of the acting entity and all + other entities that control, are controlled by, or are under common + control with that entity. For the purposes of this definition, + "control" means (i) the power, direct or indirect, to cause the + direction or management of such entity, whether by contract or + otherwise, or (ii) ownership of fifty percent (50%) or more of the + outstanding shares, or (iii) beneficial ownership of such entity. + + "You" (or "Your") shall mean an individual or Legal Entity + exercising permissions granted by this License. + + "Source" form shall mean the preferred form for making modifications, + including but not limited to software source code, documentation + source, and configuration files. + + "Object" form shall mean any form resulting from mechanical + transformation or translation of a Source form, including but + not limited to compiled object code, generated documentation, + and conversions to other media types. + + "Work" shall mean the work of authorship, whether in Source or + Object form, made available under the License, as indicated by a + copyright notice that is included in or attached to the work + (an example is provided in the Appendix below). + + "Derivative Works" shall mean any work, whether in Source or Object + form, that is based on (or derived from) the Work and for which the + editorial revisions, annotations, elaborations, or other modifications + represent, as a whole, an original work of authorship. For the purposes + of this License, Derivative Works shall not include works that remain + separable from, or merely link (or bind by name) to the interfaces of, + the Work and Derivative Works thereof. + + "Contribution" shall mean any work of authorship, including + the original version of the Work and any modifications or additions + to that Work or Derivative Works thereof, that is intentionally + submitted to Licensor for inclusion in the Work by the copyright owner + or by an individual or Legal Entity authorized to submit on behalf of + the copyright owner. For the purposes of this definition, "submitted" + means any form of electronic, verbal, or written communication sent + to the Licensor or its representatives, including but not limited to + communication on electronic mailing lists, source code control systems, + and issue tracking systems that are managed by, or on behalf of, the + Licensor for the purpose of discussing and improving the Work, but + excluding communication that is conspicuously marked or otherwise + designated in writing by the copyright owner as "Not a Contribution." + + "Contributor" shall mean Licensor and any individual or Legal Entity + on behalf of whom a Contribution has been received by Licensor and + subsequently incorporated within the Work. + + 2. Grant of Copyright License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + copyright license to reproduce, prepare Derivative Works of, + publicly display, publicly perform, sublicense, and distribute the + Work and such Derivative Works in Source or Object form. + + 3. Grant of Patent License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + (except as stated in this section) patent license to make, have made, + use, offer to sell, sell, import, and otherwise transfer the Work, + where such license applies only to those patent claims licensable + by such Contributor that are necessarily infringed by their + Contribution(s) alone or by combination of their Contribution(s) + with the Work to which such Contribution(s) was submitted. If You + institute patent litigation against any entity (including a + cross-claim or counterclaim in a lawsuit) alleging that the Work + or a Contribution incorporated within the Work constitutes direct + or contributory patent infringement, then any patent licenses + granted to You under this License for that Work shall terminate + as of the date such litigation is filed. + + 4. Redistribution. You may reproduce and distribute copies of the + Work or Derivative Works thereof in any medium, with or without + modifications, and in Source or Object form, provided that You + meet the following conditions: + + (a) You must give any other recipients of the Work or + Derivative Works a copy of this License; and + + (b) You must cause any modified files to carry prominent notices + stating that You changed the files; and + + (c) You must retain, in the Source form of any Derivative Works + that You distribute, all copyright, patent, trademark, and + attribution notices from the Source form of the Work, + excluding those notices that do not pertain to any part of + the Derivative Works; and + + (d) If the Work includes a "NOTICE" text file as part of its + distribution, then any Derivative Works that You distribute must + include a readable copy of the attribution notices contained + within such NOTICE file, excluding those notices that do not + pertain to any part of the Derivative Works, in at least one + of the following places: within a NOTICE text file distributed + as part of the Derivative Works; within the Source form or + documentation, if provided along with the Derivative Works; or, + within a display generated by the Derivative Works, if and + wherever such third-party notices normally appear. The contents + of the NOTICE file are for informational purposes only and + do not modify the License. You may add Your own attribution + notices within Derivative Works that You distribute, alongside + or as an addendum to the NOTICE text from the Work, provided + that such additional attribution notices cannot be construed + as modifying the License. + + You may add Your own copyright statement to Your modifications and + may provide additional or different license terms and conditions + for use, reproduction, or distribution of Your modifications, or + for any such Derivative Works as a whole, provided Your use, + reproduction, and distribution of the Work otherwise complies with + the conditions stated in this License. + + 5. Submission of Contributions. Unless You explicitly state otherwise, + any Contribution intentionally submitted for inclusion in the Work + by You to the Licensor shall be under the terms and conditions of + this License, without any additional terms or conditions. + Notwithstanding the above, nothing herein shall supersede or modify + the terms of any separate license agreement you may have executed + with Licensor regarding such Contributions. + + 6. Trademarks. This License does not grant permission to use the trade + names, trademarks, service marks, or product names of the Licensor, + except as required for reasonable and customary use in describing the + origin of the Work and reproducing the content of the NOTICE file. + + 7. Disclaimer of Warranty. Unless required by applicable law or + agreed to in writing, Licensor provides the Work (and each + Contributor provides its Contributions) on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or + implied, including, without limitation, any warranties or conditions + of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A + PARTICULAR PURPOSE. You are solely responsible for determining the + appropriateness of using or redistributing the Work and assume any + risks associated with Your exercise of permissions under this License. + + 8. Limitation of Liability. In no event and under no legal theory, + whether in tort (including negligence), contract, or otherwise, + unless required by applicable law (such as deliberate and grossly + negligent acts) or agreed to in writing, shall any Contributor be + liable to You for damages, including any direct, indirect, special, + incidental, or consequential damages of any character arising as a + result of this License or out of the use or inability to use the + Work (including but not limited to damages for loss of goodwill, + work stoppage, computer failure or malfunction, or any and all + other commercial damages or losses), even if such Contributor + has been advised of the possibility of such damages. + + 9. Accepting Warranty or Additional Liability. While redistributing + the Work or Derivative Works thereof, You may choose to offer, + and charge a fee for, acceptance of support, warranty, indemnity, + or other liability obligations and/or rights consistent with this + License. However, in accepting such obligations, You may act only + on Your own behalf and on Your sole responsibility, not on behalf + of any other Contributor, and only if You agree to indemnify, + defend, and hold each Contributor harmless for any liability + incurred by, or claims asserted against, such Contributor by reason + of your accepting any such warranty or additional liability. + + END OF TERMS AND CONDITIONS + + APPENDIX: How to apply the Apache License to your work. + + To apply the Apache License to your work, attach the following + boilerplate notice, with the fields enclosed by brackets "{}" + replaced with your own identifying information. (Don't include + the brackets!) The text should be enclosed in the appropriate + comment syntax for the file format. We also recommend that a + file or class name and description of purpose be included on the + same "printed page" as the copyright notice for easier + identification within third-party archives. + + Copyright {yyyy} {name of copyright owner} + + Licensed under the Apache License, Version 2.0 (the "License"); + you may not use this file except in compliance with the License. + You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + + Unless required by applicable law or agreed to in writing, software + distributed under the License is distributed on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + See the License for the specific language governing permissions and + limitations under the License. diff --git a/third_party/mkl_dnn/build_defs.bzl b/third_party/mkl_dnn/build_defs.bzl new file mode 100644 index 0000000000..7ce2a7d9b0 --- /dev/null +++ b/third_party/mkl_dnn/build_defs.bzl @@ -0,0 +1,13 @@ +def if_mkl_open_source_only(if_true, if_false = []): + """Shorthand for select()'ing on whether we're building with + MKL-DNN open source lib only, without depending on MKL binary form. + + Returns a select statement which evaluates to if_true if we're building + with MKL-DNN open source lib only. Otherwise, + the select statement evaluates to if_false. + + """ + return select({ + str(Label("//third_party/mkl_dnn:using_mkl_dnn_only")): if_true, + "//conditions:default": if_false, + }) diff --git a/third_party/mkl_dnn/mkldnn.BUILD b/third_party/mkl_dnn/mkldnn.BUILD index 58bb7a6a5d..597ac69e2f 100644 --- a/third_party/mkl_dnn/mkldnn.BUILD +++ b/third_party/mkl_dnn/mkldnn.BUILD @@ -1,16 +1,40 @@ exports_files(["LICENSE"]) +load( + "@org_tensorflow//third_party/mkl_dnn:build_defs.bzl", + "if_mkl_open_source_only", +) + +config_setting( + name = "clang_linux_x86_64", + values = { + "cpu": "k8", + "define": "using_clang=true", + }, +) + cc_library( name = "mkl_dnn", srcs = glob([ "src/common/*.cpp", "src/cpu/*.cpp", + "src/cpu/gemm/*.cpp", ]), hdrs = glob(["include/*"]), - copts = ["-fexceptions"] + select({ + copts = [ + "-fexceptions", + "-DUSE_MKL", + "-DUSE_CBLAS", + ] + if_mkl_open_source_only([ + "-UUSE_MKL", + "-UUSE_CBLAS", + ]) + select({ "@org_tensorflow//tensorflow:linux_x86_64": [ - "-fopenmp", + "-fopenmp", # only works with gcc ], + # TODO(ibiryukov): enable openmp with clang by including libomp as a + # dependency. + ":clang_linux_x86_64": [], "//conditions:default": [], }), includes = [ @@ -19,7 +43,23 @@ cc_library( "src/common", "src/cpu", "src/cpu/xbyak", + "src/cpu/gemm", ], nocopts = "-fno-exceptions", visibility = ["//visibility:public"], + deps = select({ + "@org_tensorflow//tensorflow:linux_x86_64": [ + "@mkl_linux//:mkl_headers", + "@mkl_linux//:mkl_libs_linux", + ], + "@org_tensorflow//tensorflow:darwin": [ + "@mkl_darwin//:mkl_headers", + "@mkl_darwin//:mkl_libs_darwin", + ], + "@org_tensorflow//tensorflow:windows": [ + "@mkl_windows//:mkl_headers", + "@mkl_windows//:mkl_libs_windows", + ], + "//conditions:default": [], + }), ) diff --git a/third_party/mpi/BUILD b/third_party/mpi/BUILD index ff3f437e92..1d6ac2fceb 100644 --- a/third_party/mpi/BUILD +++ b/third_party/mpi/BUILD @@ -1,17 +1,5 @@ licenses(["restricted"]) -filegroup( - name = "all_files", - srcs = glob( - ["**/*"], - exclude = [ - "**/METADATA", - "**/OWNERS", - ], - ), - visibility = ["//tensorflow:__subpackages__"], -) - load("//third_party/mpi:mpi.bzl", "mpi_hdr") load("//third_party/mpi:mpi.bzl", "if_mpi") diff --git a/third_party/nanopb.BUILD b/third_party/nanopb.BUILD new file mode 100644 index 0000000000..d21866911b --- /dev/null +++ b/third_party/nanopb.BUILD @@ -0,0 +1,23 @@ +# Description: +# Nanopb, a tiny ANSI C protobuf implementation for use on embedded devices. + +licenses(["notice"]) # zlib license + +exports_files(["LICENSE.txt"]) + +cc_library( + name = "nanopb", + srcs = [ + "pb_common.c", + "pb_decode.c", + "pb_encode.c", + ], + hdrs = [ + "pb.h", + "pb_common.h", + "pb_decode.h", + "pb_encode.h", + ], + includes = ["."], + visibility = ["//visibility:public"], +) diff --git a/third_party/nasm.BUILD b/third_party/nasm.BUILD index 341d58068b..d746a65e7e 100644 --- a/third_party/nasm.BUILD +++ b/third_party/nasm.BUILD @@ -8,45 +8,93 @@ exports_files(["LICENSE"]) cc_binary( name = "nasm", srcs = [ - "assemble.c", - "assemble.h", - "compiler.h", - "crc64.c", - "directiv.c", - "directiv.h", - "disp8.c", - "disp8.h", - "eval.c", - "eval.h", - "exprlib.c", - "float.c", - "float.h", - "hashtbl.c", - "hashtbl.h", - "iflag.c", - "iflag.h", - "iflaggen.h", - "ilog2.c", - "insns.h", - "insnsa.c", - "insnsb.c", - "insnsi.h", - "labels.c", - "labels.h", - "lib/strlcpy.c", - "listing.c", - "listing.h", - "macros.c", - "md5.h", - "md5c.c", - "nasm.c", - "nasm.h", - "nasmlib.c", - "nasmlib.h", - "opflags.h", + "asm/assemble.c", + "asm/assemble.h", + "asm/directbl.c", + "asm/directiv.c", + "asm/directiv.h", + "asm/error.c", + "asm/eval.c", + "asm/eval.h", + "asm/exprdump.c", + "asm/exprlib.c", + "asm/float.c", + "asm/float.h", + "asm/labels.c", + "asm/listing.c", + "asm/listing.h", + "asm/nasm.c", + "asm/parser.c", + "asm/parser.h", + "asm/pptok.c", + "asm/pptok.h", + "asm/pragma.c", + "asm/preproc.c", + "asm/preproc.h", + "asm/preproc-nop.c", + "asm/quote.c", + "asm/quote.h", + "asm/rdstrnum.c", + "asm/segalloc.c", + "asm/stdscan.c", + "asm/stdscan.h", + "asm/strfunc.c", + "asm/tokens.h", + "asm/tokhash.c", + "common/common.c", + "config/unknown.h", + "disasm/disasm.c", + "disasm/disasm.h", + "disasm/sync.c", + "disasm/sync.h", + "include/compiler.h", + "include/disp8.h", + "include/error.h", + "include/hashtbl.h", + "include/iflag.h", + "include/insns.h", + "include/labels.h", + "include/md5.h", + "include/nasm.h", + "include/nasmint.h", + "include/nasmlib.h", + "include/opflags.h", + "include/perfhash.h", + "include/raa.h", + "include/rbtree.h", + "include/rdoff.h", + "include/saa.h", + "include/strlist.h", + "include/tables.h", + "include/ver.h", + "macros/macros.c", + "nasmlib/badenum.c", + "nasmlib/bsi.c", + "nasmlib/crc64.c", + "nasmlib/file.c", + "nasmlib/file.h", + "nasmlib/filename.c", + "nasmlib/hashtbl.c", + "nasmlib/ilog2.c", + "nasmlib/malloc.c", + "nasmlib/md5c.c", + "nasmlib/mmap.c", + "nasmlib/path.c", + "nasmlib/perfhash.c", + "nasmlib/raa.c", + "nasmlib/rbtree.c", + "nasmlib/readnum.c", + "nasmlib/realpath.c", + "nasmlib/saa.c", + "nasmlib/srcfile.c", + "nasmlib/string.c", + "nasmlib/strlist.c", + "nasmlib/ver.c", + "nasmlib/zerobuf.c", "output/codeview.c", "output/dwarf.h", "output/elf.h", + "output/legacy.c", "output/nulldbg.c", "output/nullout.c", "output/outaout.c", @@ -56,9 +104,6 @@ cc_binary( "output/outdbg.c", "output/outelf.c", "output/outelf.h", - "output/outelf32.c", - "output/outelf64.c", - "output/outelfx32.c", "output/outform.c", "output/outform.h", "output/outieee.c", @@ -69,39 +114,37 @@ cc_binary( "output/outrdf2.c", "output/pecoff.h", "output/stabs.h", - "parser.c", - "parser.h", - "pptok.c", - "pptok.h", - "preproc.c", - "preproc.h", - "preproc-nop.c", - "quote.c", - "quote.h", - "raa.c", - "raa.h", - "rbtree.c", - "rbtree.h", - "rdoff/rdoff.h", - "realpath.c", - "regflags.c", - "regs.h", - "regvals.c", - "saa.c", - "saa.h", - "srcfile.c", - "stdscan.c", - "stdscan.h", - "strfunc.c", - "tables.h", - "tokens.h", - "tokhash.c", - "ver.c", + "stdlib/snprintf.c", + "stdlib/strlcpy.c", + "stdlib/strnlen.c", + "stdlib/vsnprintf.c", "version.h", + "x86/disp8.c", + "x86/iflag.c", + "x86/iflaggen.h", + "x86/insnsa.c", + "x86/insnsb.c", + "x86/insnsd.c", + "x86/insnsi.h", + "x86/insnsn.c", + "x86/regdis.c", + "x86/regdis.h", + "x86/regflags.c", + "x86/regs.c", + "x86/regs.h", + "x86/regvals.c", + ] + select({ + ":windows": ["config/msvc.h"], + "//conditions:default": [], + }), + includes = [ + "asm", + "include", + "output", + "x86", ], copts = select({ ":windows": [], - ":windows_msvc": [], "//conditions:default": [ "-w", "-std=c99", @@ -109,20 +152,15 @@ cc_binary( }), defines = select({ ":windows": [], - ":windows_msvc": [], - "//conditions:default": ["HAVE_SNPRINTF"], + "//conditions:default": [ + "HAVE_SNPRINTF", + "HAVE_SYS_TYPES_H", + ], }), visibility = ["@jpeg//:__pkg__"], ) config_setting( - name = "windows_msvc", - values = { - "cpu": "x64_windows_msvc", - }, -) - -config_setting( name = "windows", values = { "cpu": "x64_windows", diff --git a/third_party/nccl/BUILD b/third_party/nccl/BUILD new file mode 100644 index 0000000000..e69de29bb2 --- /dev/null +++ b/third_party/nccl/BUILD diff --git a/third_party/nccl/LICENSE b/third_party/nccl/LICENSE new file mode 100644 index 0000000000..146d9b765c --- /dev/null +++ b/third_party/nccl/LICENSE @@ -0,0 +1,203 @@ +Copyright 2018 The TensorFlow Authors. All rights reserved. + + Apache License + Version 2.0, January 2004 + http://www.apache.org/licenses/ + + TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION + + 1. Definitions. + + "License" shall mean the terms and conditions for use, reproduction, + and distribution as defined by Sections 1 through 9 of this document. + + "Licensor" shall mean the copyright owner or entity authorized by + the copyright owner that is granting the License. + + "Legal Entity" shall mean the union of the acting entity and all + other entities that control, are controlled by, or are under common + control with that entity. For the purposes of this definition, + "control" means (i) the power, direct or indirect, to cause the + direction or management of such entity, whether by contract or + otherwise, or (ii) ownership of fifty percent (50%) or more of the + outstanding shares, or (iii) beneficial ownership of such entity. + + "You" (or "Your") shall mean an individual or Legal Entity + exercising permissions granted by this License. + + "Source" form shall mean the preferred form for making modifications, + including but not limited to software source code, documentation + source, and configuration files. + + "Object" form shall mean any form resulting from mechanical + transformation or translation of a Source form, including but + not limited to compiled object code, generated documentation, + and conversions to other media types. + + "Work" shall mean the work of authorship, whether in Source or + Object form, made available under the License, as indicated by a + copyright notice that is included in or attached to the work + (an example is provided in the Appendix below). + + "Derivative Works" shall mean any work, whether in Source or Object + form, that is based on (or derived from) the Work and for which the + editorial revisions, annotations, elaborations, or other modifications + represent, as a whole, an original work of authorship. For the purposes + of this License, Derivative Works shall not include works that remain + separable from, or merely link (or bind by name) to the interfaces of, + the Work and Derivative Works thereof. + + "Contribution" shall mean any work of authorship, including + the original version of the Work and any modifications or additions + to that Work or Derivative Works thereof, that is intentionally + submitted to Licensor for inclusion in the Work by the copyright owner + or by an individual or Legal Entity authorized to submit on behalf of + the copyright owner. For the purposes of this definition, "submitted" + means any form of electronic, verbal, or written communication sent + to the Licensor or its representatives, including but not limited to + communication on electronic mailing lists, source code control systems, + and issue tracking systems that are managed by, or on behalf of, the + Licensor for the purpose of discussing and improving the Work, but + excluding communication that is conspicuously marked or otherwise + designated in writing by the copyright owner as "Not a Contribution." + + "Contributor" shall mean Licensor and any individual or Legal Entity + on behalf of whom a Contribution has been received by Licensor and + subsequently incorporated within the Work. + + 2. Grant of Copyright License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + copyright license to reproduce, prepare Derivative Works of, + publicly display, publicly perform, sublicense, and distribute the + Work and such Derivative Works in Source or Object form. + + 3. Grant of Patent License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + (except as stated in this section) patent license to make, have made, + use, offer to sell, sell, import, and otherwise transfer the Work, + where such license applies only to those patent claims licensable + by such Contributor that are necessarily infringed by their + Contribution(s) alone or by combination of their Contribution(s) + with the Work to which such Contribution(s) was submitted. If You + institute patent litigation against any entity (including a + cross-claim or counterclaim in a lawsuit) alleging that the Work + or a Contribution incorporated within the Work constitutes direct + or contributory patent infringement, then any patent licenses + granted to You under this License for that Work shall terminate + as of the date such litigation is filed. + + 4. Redistribution. You may reproduce and distribute copies of the + Work or Derivative Works thereof in any medium, with or without + modifications, and in Source or Object form, provided that You + meet the following conditions: + + (a) You must give any other recipients of the Work or + Derivative Works a copy of this License; and + + (b) You must cause any modified files to carry prominent notices + stating that You changed the files; and + + (c) You must retain, in the Source form of any Derivative Works + that You distribute, all copyright, patent, trademark, and + attribution notices from the Source form of the Work, + excluding those notices that do not pertain to any part of + the Derivative Works; and + + (d) If the Work includes a "NOTICE" text file as part of its + distribution, then any Derivative Works that You distribute must + include a readable copy of the attribution notices contained + within such NOTICE file, excluding those notices that do not + pertain to any part of the Derivative Works, in at least one + of the following places: within a NOTICE text file distributed + as part of the Derivative Works; within the Source form or + documentation, if provided along with the Derivative Works; or, + within a display generated by the Derivative Works, if and + wherever such third-party notices normally appear. The contents + of the NOTICE file are for informational purposes only and + do not modify the License. You may add Your own attribution + notices within Derivative Works that You distribute, alongside + or as an addendum to the NOTICE text from the Work, provided + that such additional attribution notices cannot be construed + as modifying the License. + + You may add Your own copyright statement to Your modifications and + may provide additional or different license terms and conditions + for use, reproduction, or distribution of Your modifications, or + for any such Derivative Works as a whole, provided Your use, + reproduction, and distribution of the Work otherwise complies with + the conditions stated in this License. + + 5. Submission of Contributions. Unless You explicitly state otherwise, + any Contribution intentionally submitted for inclusion in the Work + by You to the Licensor shall be under the terms and conditions of + this License, without any additional terms or conditions. + Notwithstanding the above, nothing herein shall supersede or modify + the terms of any separate license agreement you may have executed + with Licensor regarding such Contributions. + + 6. Trademarks. This License does not grant permission to use the trade + names, trademarks, service marks, or product names of the Licensor, + except as required for reasonable and customary use in describing the + origin of the Work and reproducing the content of the NOTICE file. + + 7. Disclaimer of Warranty. Unless required by applicable law or + agreed to in writing, Licensor provides the Work (and each + Contributor provides its Contributions) on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or + implied, including, without limitation, any warranties or conditions + of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A + PARTICULAR PURPOSE. You are solely responsible for determining the + appropriateness of using or redistributing the Work and assume any + risks associated with Your exercise of permissions under this License. + + 8. Limitation of Liability. In no event and under no legal theory, + whether in tort (including negligence), contract, or otherwise, + unless required by applicable law (such as deliberate and grossly + negligent acts) or agreed to in writing, shall any Contributor be + liable to You for damages, including any direct, indirect, special, + incidental, or consequential damages of any character arising as a + result of this License or out of the use or inability to use the + Work (including but not limited to damages for loss of goodwill, + work stoppage, computer failure or malfunction, or any and all + other commercial damages or losses), even if such Contributor + has been advised of the possibility of such damages. + + 9. Accepting Warranty or Additional Liability. While redistributing + the Work or Derivative Works thereof, You may choose to offer, + and charge a fee for, acceptance of support, warranty, indemnity, + or other liability obligations and/or rights consistent with this + License. However, in accepting such obligations, You may act only + on Your own behalf and on Your sole responsibility, not on behalf + of any other Contributor, and only if You agree to indemnify, + defend, and hold each Contributor harmless for any liability + incurred by, or claims asserted against, such Contributor by reason + of your accepting any such warranty or additional liability. + + END OF TERMS AND CONDITIONS + + APPENDIX: How to apply the Apache License to your work. + + To apply the Apache License to your work, attach the following + boilerplate notice, with the fields enclosed by brackets "[]" + replaced with your own identifying information. (Don't include + the brackets!) The text should be enclosed in the appropriate + comment syntax for the file format. We also recommend that a + file or class name and description of purpose be included on the + same "printed page" as the copyright notice for easier + identification within third-party archives. + + Copyright 2018, The TensorFlow Authors. + + Licensed under the Apache License, Version 2.0 (the "License"); + you may not use this file except in compliance with the License. + You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + + Unless required by applicable law or agreed to in writing, software + distributed under the License is distributed on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + See the License for the specific language governing permissions and + limitations under the License. diff --git a/third_party/nccl.BUILD b/third_party/nccl/nccl_archive.BUILD index b2b8e18824..a05899e38d 100644 --- a/third_party/nccl.BUILD +++ b/third_party/nccl/nccl_archive.BUILD @@ -43,6 +43,7 @@ cc_library( "-Iexternal/nccl_archive/src", "-O3", ] + cuda_default_copts(), + include_prefix = "third_party/nccl", linkopts = select({ "@org_tensorflow//tensorflow:android": [ "-pie", @@ -61,6 +62,7 @@ cc_library( "-lrt", ], }), + strip_include_prefix = "src", visibility = ["//visibility:public"], deps = ["@local_config_cuda//cuda:cuda_headers"], ) diff --git a/third_party/nccl/nccl_configure.bzl b/third_party/nccl/nccl_configure.bzl new file mode 100644 index 0000000000..ce9447096e --- /dev/null +++ b/third_party/nccl/nccl_configure.bzl @@ -0,0 +1,165 @@ +# -*- Python -*- +"""Repository rule for NCCL configuration. + +`nccl_configure` depends on the following environment variables: + + * `TF_NCCL_VERSION`: The NCCL version. + * `NCCL_INSTALL_PATH`: The installation path of the NCCL library. +""" + +load( + "//third_party/gpus:cuda_configure.bzl", + "auto_configure_fail", + "find_cuda_define", + "matches_version", +) + +_NCCL_INSTALL_PATH = "NCCL_INSTALL_PATH" +_TF_NCCL_VERSION = "TF_NCCL_VERSION" +_TF_NCCL_CONFIG_REPO = "TF_NCCL_CONFIG_REPO" + +_DEFINE_NCCL_MAJOR = "#define NCCL_MAJOR" +_DEFINE_NCCL_MINOR = "#define NCCL_MINOR" +_DEFINE_NCCL_PATCH = "#define NCCL_PATCH" + +_NCCL_DUMMY_BUILD_CONTENT = """ +filegroup( + name = "LICENSE", + visibility = ["//visibility:public"], +) + +cc_library( + name = "nccl", + visibility = ["//visibility:public"], +) +""" + +_NCCL_ARCHIVE_BUILD_CONTENT = """ +filegroup( + name = "LICENSE", + data = ["@nccl_archive//:LICENSE.txt"], + visibility = ["//visibility:public"], +) + +alias( + name = "nccl", + actual = "@nccl_archive//:nccl", + visibility = ["//visibility:public"], +) +""" + +# Local build results in dynamic link and the license should not be included. +_NCCL_REMOTE_BUILD_TEMPLATE = Label("//third_party/nccl:remote.BUILD.tpl") +_NCCL_LOCAL_BUILD_TEMPLATE = Label("//third_party/nccl:system.BUILD.tpl") + +def _find_nccl_header(repository_ctx, nccl_install_path): + """Finds the NCCL header on the system. + + Args: + repository_ctx: The repository context. + nccl_install_path: The NCCL library install directory. + + Returns: + The path to the NCCL header. + """ + header_path = repository_ctx.path("%s/include/nccl.h" % nccl_install_path) + if not header_path.exists: + auto_configure_fail("Cannot find %s" % str(header_path)) + return header_path + + +def _check_nccl_version(repository_ctx, nccl_install_path, nccl_version): + """Checks whether the header file matches the specified version of NCCL. + + Args: + repository_ctx: The repository context. + nccl_install_path: The NCCL library install directory. + nccl_version: The expected NCCL version. + + Returns: + A string containing the library version of NCCL. + """ + header_path = _find_nccl_header(repository_ctx, nccl_install_path) + header_dir = str(header_path.realpath.dirname) + major_version = find_cuda_define(repository_ctx, header_dir, "nccl.h", + _DEFINE_NCCL_MAJOR) + minor_version = find_cuda_define(repository_ctx, header_dir, "nccl.h", + _DEFINE_NCCL_MINOR) + patch_version = find_cuda_define(repository_ctx, header_dir, "nccl.h", + _DEFINE_NCCL_PATCH) + header_version = "%s.%s.%s" % (major_version, minor_version, patch_version) + if not matches_version(nccl_version, header_version): + auto_configure_fail( + ("NCCL library version detected from %s/nccl.h (%s) does not match " + + "TF_NCCL_VERSION (%s). To fix this rerun configure again.") % + (header_dir, header_version, nccl_version)) + + +def _find_nccl_lib(repository_ctx, nccl_install_path, nccl_version): + """Finds the given NCCL library on the system. + + Args: + repository_ctx: The repository context. + nccl_install_path: The NCCL library installation directory. + nccl_version: The version of NCCL library files as returned + by _nccl_version. + + Returns: + The path to the NCCL library. + """ + lib_path = repository_ctx.path("%s/lib/libnccl.so.%s" % (nccl_install_path, + nccl_version)) + if not lib_path.exists: + auto_configure_fail("Cannot find NCCL library %s" % str(lib_path)) + return lib_path + + +def _nccl_configure_impl(repository_ctx): + """Implementation of the nccl_configure repository rule.""" + if _TF_NCCL_VERSION not in repository_ctx.os.environ: + # Add a dummy build file to make bazel query happy. + repository_ctx.file("BUILD", _NCCL_DUMMY_BUILD_CONTENT) + return + + if _TF_NCCL_CONFIG_REPO in repository_ctx.os.environ: + # Forward to the pre-configured remote repository. + repository_ctx.template("BUILD", _NCCL_REMOTE_BUILD_TEMPLATE, { + "%{target}": repository_ctx.os.environ[_TF_NCCL_CONFIG_REPO], + }) + return + + nccl_version = repository_ctx.os.environ[_TF_NCCL_VERSION].strip() + if matches_version("1", nccl_version): + # Alias to GitHub target from @nccl_archive. + if not matches_version(nccl_version, "1.3"): + auto_configure_fail( + "NCCL from GitHub must use version 1.3 (got %s)" % nccl_version) + repository_ctx.file("BUILD", _NCCL_ARCHIVE_BUILD_CONTENT) + else: + # Create target for locally installed NCCL. + nccl_install_path = repository_ctx.os.environ[_NCCL_INSTALL_PATH].strip() + _check_nccl_version(repository_ctx, nccl_install_path, nccl_version) + repository_ctx.template("BUILD", _NCCL_LOCAL_BUILD_TEMPLATE, { + "%{version}": nccl_version, + "%{install_path}": nccl_install_path, + }) + + +nccl_configure = repository_rule( + implementation=_nccl_configure_impl, + environ=[ + _NCCL_INSTALL_PATH, + _TF_NCCL_VERSION, + ], +) +"""Detects and configures the NCCL configuration. + +Add the following to your WORKSPACE FILE: + +```python +nccl_configure(name = "local_config_nccl") +``` + +Args: + name: A unique name for this workspace rule. +""" diff --git a/third_party/nccl/remote.BUILD.tpl b/third_party/nccl/remote.BUILD.tpl new file mode 100644 index 0000000000..d66fc5563d --- /dev/null +++ b/third_party/nccl/remote.BUILD.tpl @@ -0,0 +1,6 @@ +licenses(["restricted"]) + +package(default_visibility = ["//visibility:public"]) + +alias(name="LICENSE", actual = "%{target}:LICENSE") +alias(name = "nccl", actual = "%{target}:nccl") diff --git a/third_party/nccl/system.BUILD.tpl b/third_party/nccl/system.BUILD.tpl new file mode 100644 index 0000000000..7ca835dedf --- /dev/null +++ b/third_party/nccl/system.BUILD.tpl @@ -0,0 +1,26 @@ +filegroup( + name = "LICENSE", + visibility = ["//visibility:public"], +) + +cc_library( + name = "nccl", + srcs = ["libnccl.so.%{version}"], + hdrs = ["nccl.h"], + include_prefix = "third_party/nccl", + deps = [ + "@local_config_cuda//cuda:cuda_headers", + ], + visibility = ["//visibility:public"], +) + +genrule( + name = "nccl-files", + outs = [ + "libnccl.so.%{version}", + "nccl.h", + ], + cmd = """cp "%{install_path}/include/nccl.h" "$(@D)/nccl.h" && + cp "%{install_path}/lib/libnccl.so.%{version}" "$(@D)/libnccl.so.%{version}" """, +) + diff --git a/third_party/ngraph/BUILD b/third_party/ngraph/BUILD new file mode 100644 index 0000000000..067771b43f --- /dev/null +++ b/third_party/ngraph/BUILD @@ -0,0 +1 @@ +licenses(["notice"]) # 3-Clause BSD diff --git a/third_party/ngraph/LICENSE b/third_party/ngraph/LICENSE new file mode 100644 index 0000000000..9c8f3ea087 --- /dev/null +++ b/third_party/ngraph/LICENSE @@ -0,0 +1,201 @@ + Apache License + Version 2.0, January 2004 + http://www.apache.org/licenses/ + + TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION + + 1. Definitions. + + "License" shall mean the terms and conditions for use, reproduction, + and distribution as defined by Sections 1 through 9 of this document. + + "Licensor" shall mean the copyright owner or entity authorized by + the copyright owner that is granting the License. + + "Legal Entity" shall mean the union of the acting entity and all + other entities that control, are controlled by, or are under common + control with that entity. For the purposes of this definition, + "control" means (i) the power, direct or indirect, to cause the + direction or management of such entity, whether by contract or + otherwise, or (ii) ownership of fifty percent (50%) or more of the + outstanding shares, or (iii) beneficial ownership of such entity. + + "You" (or "Your") shall mean an individual or Legal Entity + exercising permissions granted by this License. + + "Source" form shall mean the preferred form for making modifications, + including but not limited to software source code, documentation + source, and configuration files. + + "Object" form shall mean any form resulting from mechanical + transformation or translation of a Source form, including but + not limited to compiled object code, generated documentation, + and conversions to other media types. + + "Work" shall mean the work of authorship, whether in Source or + Object form, made available under the License, as indicated by a + copyright notice that is included in or attached to the work + (an example is provided in the Appendix below). + + "Derivative Works" shall mean any work, whether in Source or Object + form, that is based on (or derived from) the Work and for which the + editorial revisions, annotations, elaborations, or other modifications + represent, as a whole, an original work of authorship. For the purposes + of this License, Derivative Works shall not include works that remain + separable from, or merely link (or bind by name) to the interfaces of, + the Work and Derivative Works thereof. + + "Contribution" shall mean any work of authorship, including + the original version of the Work and any modifications or additions + to that Work or Derivative Works thereof, that is intentionally + submitted to Licensor for inclusion in the Work by the copyright owner + or by an individual or Legal Entity authorized to submit on behalf of + the copyright owner. For the purposes of this definition, "submitted" + means any form of electronic, verbal, or written communication sent + to the Licensor or its representatives, including but not limited to + communication on electronic mailing lists, source code control systems, + and issue tracking systems that are managed by, or on behalf of, the + Licensor for the purpose of discussing and improving the Work, but + excluding communication that is conspicuously marked or otherwise + designated in writing by the copyright owner as "Not a Contribution." + + "Contributor" shall mean Licensor and any individual or Legal Entity + on behalf of whom a Contribution has been received by Licensor and + subsequently incorporated within the Work. + + 2. Grant of Copyright License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + copyright license to reproduce, prepare Derivative Works of, + publicly display, publicly perform, sublicense, and distribute the + Work and such Derivative Works in Source or Object form. + + 3. Grant of Patent License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + (except as stated in this section) patent license to make, have made, + use, offer to sell, sell, import, and otherwise transfer the Work, + where such license applies only to those patent claims licensable + by such Contributor that are necessarily infringed by their + Contribution(s) alone or by combination of their Contribution(s) + with the Work to which such Contribution(s) was submitted. If You + institute patent litigation against any entity (including a + cross-claim or counterclaim in a lawsuit) alleging that the Work + or a Contribution incorporated within the Work constitutes direct + or contributory patent infringement, then any patent licenses + granted to You under this License for that Work shall terminate + as of the date such litigation is filed. + + 4. Redistribution. You may reproduce and distribute copies of the + Work or Derivative Works thereof in any medium, with or without + modifications, and in Source or Object form, provided that You + meet the following conditions: + + (a) You must give any other recipients of the Work or + Derivative Works a copy of this License; and + + (b) You must cause any modified files to carry prominent notices + stating that You changed the files; and + + (c) You must retain, in the Source form of any Derivative Works + that You distribute, all copyright, patent, trademark, and + attribution notices from the Source form of the Work, + excluding those notices that do not pertain to any part of + the Derivative Works; and + + (d) If the Work includes a "NOTICE" text file as part of its + distribution, then any Derivative Works that You distribute must + include a readable copy of the attribution notices contained + within such NOTICE file, excluding those notices that do not + pertain to any part of the Derivative Works, in at least one + of the following places: within a NOTICE text file distributed + as part of the Derivative Works; within the Source form or + documentation, if provided along with the Derivative Works; or, + within a display generated by the Derivative Works, if and + wherever such third-party notices normally appear. The contents + of the NOTICE file are for informational purposes only and + do not modify the License. You may add Your own attribution + notices within Derivative Works that You distribute, alongside + or as an addendum to the NOTICE text from the Work, provided + that such additional attribution notices cannot be construed + as modifying the License. + + You may add Your own copyright statement to Your modifications and + may provide additional or different license terms and conditions + for use, reproduction, or distribution of Your modifications, or + for any such Derivative Works as a whole, provided Your use, + reproduction, and distribution of the Work otherwise complies with + the conditions stated in this License. + + 5. Submission of Contributions. Unless You explicitly state otherwise, + any Contribution intentionally submitted for inclusion in the Work + by You to the Licensor shall be under the terms and conditions of + this License, without any additional terms or conditions. + Notwithstanding the above, nothing herein shall supersede or modify + the terms of any separate license agreement you may have executed + with Licensor regarding such Contributions. + + 6. Trademarks. This License does not grant permission to use the trade + names, trademarks, service marks, or product names of the Licensor, + except as required for reasonable and customary use in describing the + origin of the Work and reproducing the content of the NOTICE file. + + 7. Disclaimer of Warranty. Unless required by applicable law or + agreed to in writing, Licensor provides the Work (and each + Contributor provides its Contributions) on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or + implied, including, without limitation, any warranties or conditions + of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A + PARTICULAR PURPOSE. You are solely responsible for determining the + appropriateness of using or redistributing the Work and assume any + risks associated with Your exercise of permissions under this License. + + 8. Limitation of Liability. In no event and under no legal theory, + whether in tort (including negligence), contract, or otherwise, + unless required by applicable law (such as deliberate and grossly + negligent acts) or agreed to in writing, shall any Contributor be + liable to You for damages, including any direct, indirect, special, + incidental, or consequential damages of any character arising as a + result of this License or out of the use or inability to use the + Work (including but not limited to damages for loss of goodwill, + work stoppage, computer failure or malfunction, or any and all + other commercial damages or losses), even if such Contributor + has been advised of the possibility of such damages. + + 9. Accepting Warranty or Additional Liability. While redistributing + the Work or Derivative Works thereof, You may choose to offer, + and charge a fee for, acceptance of support, warranty, indemnity, + or other liability obligations and/or rights consistent with this + License. However, in accepting such obligations, You may act only + on Your own behalf and on Your sole responsibility, not on behalf + of any other Contributor, and only if You agree to indemnify, + defend, and hold each Contributor harmless for any liability + incurred by, or claims asserted against, such Contributor by reason + of your accepting any such warranty or additional liability. + + END OF TERMS AND CONDITIONS + + APPENDIX: How to apply the Apache License to your work. + + To apply the Apache License to your work, attach the following + boilerplate notice, with the fields enclosed by brackets "{}" + replaced with your own identifying information. (Don't include + the brackets!) The text should be enclosed in the appropriate + comment syntax for the file format. We also recommend that a + file or class name and description of purpose be included on the + same "printed page" as the copyright notice for easier + identification within third-party archives. + + Copyright {yyyy} {name of copyright owner} + + Licensed under the Apache License, Version 2.0 (the "License"); + you may not use this file except in compliance with the License. + You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + + Unless required by applicable law or agreed to in writing, software + distributed under the License is distributed on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + See the License for the specific language governing permissions and + limitations under the License.
\ No newline at end of file diff --git a/third_party/ngraph/NGRAPH_LICENSE b/third_party/ngraph/NGRAPH_LICENSE new file mode 100644 index 0000000000..9c8f3ea087 --- /dev/null +++ b/third_party/ngraph/NGRAPH_LICENSE @@ -0,0 +1,201 @@ + Apache License + Version 2.0, January 2004 + http://www.apache.org/licenses/ + + TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION + + 1. Definitions. + + "License" shall mean the terms and conditions for use, reproduction, + and distribution as defined by Sections 1 through 9 of this document. + + "Licensor" shall mean the copyright owner or entity authorized by + the copyright owner that is granting the License. + + "Legal Entity" shall mean the union of the acting entity and all + other entities that control, are controlled by, or are under common + control with that entity. For the purposes of this definition, + "control" means (i) the power, direct or indirect, to cause the + direction or management of such entity, whether by contract or + otherwise, or (ii) ownership of fifty percent (50%) or more of the + outstanding shares, or (iii) beneficial ownership of such entity. + + "You" (or "Your") shall mean an individual or Legal Entity + exercising permissions granted by this License. + + "Source" form shall mean the preferred form for making modifications, + including but not limited to software source code, documentation + source, and configuration files. + + "Object" form shall mean any form resulting from mechanical + transformation or translation of a Source form, including but + not limited to compiled object code, generated documentation, + and conversions to other media types. + + "Work" shall mean the work of authorship, whether in Source or + Object form, made available under the License, as indicated by a + copyright notice that is included in or attached to the work + (an example is provided in the Appendix below). + + "Derivative Works" shall mean any work, whether in Source or Object + form, that is based on (or derived from) the Work and for which the + editorial revisions, annotations, elaborations, or other modifications + represent, as a whole, an original work of authorship. For the purposes + of this License, Derivative Works shall not include works that remain + separable from, or merely link (or bind by name) to the interfaces of, + the Work and Derivative Works thereof. + + "Contribution" shall mean any work of authorship, including + the original version of the Work and any modifications or additions + to that Work or Derivative Works thereof, that is intentionally + submitted to Licensor for inclusion in the Work by the copyright owner + or by an individual or Legal Entity authorized to submit on behalf of + the copyright owner. For the purposes of this definition, "submitted" + means any form of electronic, verbal, or written communication sent + to the Licensor or its representatives, including but not limited to + communication on electronic mailing lists, source code control systems, + and issue tracking systems that are managed by, or on behalf of, the + Licensor for the purpose of discussing and improving the Work, but + excluding communication that is conspicuously marked or otherwise + designated in writing by the copyright owner as "Not a Contribution." + + "Contributor" shall mean Licensor and any individual or Legal Entity + on behalf of whom a Contribution has been received by Licensor and + subsequently incorporated within the Work. + + 2. Grant of Copyright License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + copyright license to reproduce, prepare Derivative Works of, + publicly display, publicly perform, sublicense, and distribute the + Work and such Derivative Works in Source or Object form. + + 3. Grant of Patent License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + (except as stated in this section) patent license to make, have made, + use, offer to sell, sell, import, and otherwise transfer the Work, + where such license applies only to those patent claims licensable + by such Contributor that are necessarily infringed by their + Contribution(s) alone or by combination of their Contribution(s) + with the Work to which such Contribution(s) was submitted. If You + institute patent litigation against any entity (including a + cross-claim or counterclaim in a lawsuit) alleging that the Work + or a Contribution incorporated within the Work constitutes direct + or contributory patent infringement, then any patent licenses + granted to You under this License for that Work shall terminate + as of the date such litigation is filed. + + 4. Redistribution. You may reproduce and distribute copies of the + Work or Derivative Works thereof in any medium, with or without + modifications, and in Source or Object form, provided that You + meet the following conditions: + + (a) You must give any other recipients of the Work or + Derivative Works a copy of this License; and + + (b) You must cause any modified files to carry prominent notices + stating that You changed the files; and + + (c) You must retain, in the Source form of any Derivative Works + that You distribute, all copyright, patent, trademark, and + attribution notices from the Source form of the Work, + excluding those notices that do not pertain to any part of + the Derivative Works; and + + (d) If the Work includes a "NOTICE" text file as part of its + distribution, then any Derivative Works that You distribute must + include a readable copy of the attribution notices contained + within such NOTICE file, excluding those notices that do not + pertain to any part of the Derivative Works, in at least one + of the following places: within a NOTICE text file distributed + as part of the Derivative Works; within the Source form or + documentation, if provided along with the Derivative Works; or, + within a display generated by the Derivative Works, if and + wherever such third-party notices normally appear. The contents + of the NOTICE file are for informational purposes only and + do not modify the License. You may add Your own attribution + notices within Derivative Works that You distribute, alongside + or as an addendum to the NOTICE text from the Work, provided + that such additional attribution notices cannot be construed + as modifying the License. + + You may add Your own copyright statement to Your modifications and + may provide additional or different license terms and conditions + for use, reproduction, or distribution of Your modifications, or + for any such Derivative Works as a whole, provided Your use, + reproduction, and distribution of the Work otherwise complies with + the conditions stated in this License. + + 5. Submission of Contributions. Unless You explicitly state otherwise, + any Contribution intentionally submitted for inclusion in the Work + by You to the Licensor shall be under the terms and conditions of + this License, without any additional terms or conditions. + Notwithstanding the above, nothing herein shall supersede or modify + the terms of any separate license agreement you may have executed + with Licensor regarding such Contributions. + + 6. Trademarks. This License does not grant permission to use the trade + names, trademarks, service marks, or product names of the Licensor, + except as required for reasonable and customary use in describing the + origin of the Work and reproducing the content of the NOTICE file. + + 7. Disclaimer of Warranty. Unless required by applicable law or + agreed to in writing, Licensor provides the Work (and each + Contributor provides its Contributions) on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or + implied, including, without limitation, any warranties or conditions + of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A + PARTICULAR PURPOSE. You are solely responsible for determining the + appropriateness of using or redistributing the Work and assume any + risks associated with Your exercise of permissions under this License. + + 8. Limitation of Liability. In no event and under no legal theory, + whether in tort (including negligence), contract, or otherwise, + unless required by applicable law (such as deliberate and grossly + negligent acts) or agreed to in writing, shall any Contributor be + liable to You for damages, including any direct, indirect, special, + incidental, or consequential damages of any character arising as a + result of this License or out of the use or inability to use the + Work (including but not limited to damages for loss of goodwill, + work stoppage, computer failure or malfunction, or any and all + other commercial damages or losses), even if such Contributor + has been advised of the possibility of such damages. + + 9. Accepting Warranty or Additional Liability. While redistributing + the Work or Derivative Works thereof, You may choose to offer, + and charge a fee for, acceptance of support, warranty, indemnity, + or other liability obligations and/or rights consistent with this + License. However, in accepting such obligations, You may act only + on Your own behalf and on Your sole responsibility, not on behalf + of any other Contributor, and only if You agree to indemnify, + defend, and hold each Contributor harmless for any liability + incurred by, or claims asserted against, such Contributor by reason + of your accepting any such warranty or additional liability. + + END OF TERMS AND CONDITIONS + + APPENDIX: How to apply the Apache License to your work. + + To apply the Apache License to your work, attach the following + boilerplate notice, with the fields enclosed by brackets "{}" + replaced with your own identifying information. (Don't include + the brackets!) The text should be enclosed in the appropriate + comment syntax for the file format. We also recommend that a + file or class name and description of purpose be included on the + same "printed page" as the copyright notice for easier + identification within third-party archives. + + Copyright {yyyy} {name of copyright owner} + + Licensed under the Apache License, Version 2.0 (the "License"); + you may not use this file except in compliance with the License. + You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + + Unless required by applicable law or agreed to in writing, software + distributed under the License is distributed on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + See the License for the specific language governing permissions and + limitations under the License.
\ No newline at end of file diff --git a/third_party/ngraph/build_defs.bzl b/third_party/ngraph/build_defs.bzl new file mode 100644 index 0000000000..3c34be524b --- /dev/null +++ b/third_party/ngraph/build_defs.bzl @@ -0,0 +1,11 @@ +"""Build configurations for nGraph.""" + +def clean_dep(dep): + return str(Label(dep)) + +def if_ngraph(if_true, if_false = []): + """select()'ing on whether we're building with nGraph support.""" + return select({ + clean_dep("//tensorflow:with_ngraph_support"): if_true, + "//conditions:default": if_false, + }) diff --git a/third_party/ngraph/ngraph.BUILD b/third_party/ngraph/ngraph.BUILD new file mode 100644 index 0000000000..31aa3cee51 --- /dev/null +++ b/third_party/ngraph/ngraph.BUILD @@ -0,0 +1,37 @@ +licenses(["notice"]) # 3-Clause BSD + +exports_files(["LICENSE"]) + +cc_library( + name = "ngraph_core", + srcs = glob([ + "src/ngraph/*.cpp", + "src/ngraph/autodiff/*.cpp", + "src/ngraph/builder/*.cpp", + "src/ngraph/descriptor/*.cpp", + "src/ngraph/descriptor/layout/*.cpp", + "src/ngraph/op/*.cpp", + "src/ngraph/op/util/*.cpp", + "src/ngraph/pattern/*.cpp", + "src/ngraph/pattern/*.hpp", + "src/ngraph/pass/*.cpp", + "src/ngraph/pass/*.hpp", + "src/ngraph/runtime/*.cpp", + "src/ngraph/type/*.cpp", + "src/ngraph/runtime/interpreter/*.cpp", + "src/ngraph/runtime/interpreter/*.hpp", + ]), + hdrs = glob(["src/ngraph/**/*.hpp"]), + deps = [ + "@eigen_archive//:eigen", + "@nlohmann_json_lib", + ], + copts = [ + "-I external/ngraph/src", + "-I external/nlohmann_json_lib/include/", + '-D SHARED_LIB_EXT=\\".so\\"', + '-D NGRAPH_VERSION=\\"0.5.0\\"', + ], + visibility = ["//visibility:public"], + alwayslink = 1, +) diff --git a/third_party/ngraph/ngraph_tf.BUILD b/third_party/ngraph/ngraph_tf.BUILD new file mode 100644 index 0000000000..4d96ccf2f2 --- /dev/null +++ b/third_party/ngraph/ngraph_tf.BUILD @@ -0,0 +1,88 @@ +licenses(["notice"]) # 3-Clause BSD + +exports_files(["LICENSE"]) + +load( + "@org_tensorflow//tensorflow:tensorflow.bzl", + "tf_cc_test", +) + +cc_library( + name = "ngraph_libs_linux", + srcs = [ + "lib/libiomp5.so", + "lib/libmklml_intel.so", + ], + visibility = ["//visibility:public"], +) + +cc_library( + name = "ngraph_tf", + srcs = [ + "src/ngraph_builder.h", + "src/ngraph_builder.cc", + "src/ngraph_cluster.h", + "src/ngraph_cluster.cc", + "src/ngraph_cluster_manager.h", + "src/ngraph_cluster_manager.cc", + "src/ngraph_confirm_pass.cc", + "src/ngraph_device.cc", + "src/ngraph_encapsulate_op.cc", + "src/ngraph_encapsulate_pass.cc", + "src/ngraph_freshness_tracker.h", + "src/ngraph_freshness_tracker.cc", + "src/ngraph_graph_rewrite_passes.cc", + "src/ngraph_liberate_pass.cc", + "src/ngraph_op_kernels.cc", + "src/ngraph_stub_ops.cc", + "src/ngraph_utils.h", + "src/ngraph_utils.cc", + "src/ngraph_send_recv_ops.cc", + "src/ngraph_variable_ops.cc", + "src/tf_graphcycles.cc", + "logging/ngraph_log.h", + "logging/ngraph_log.cc", + "logging/tf_graph_writer.h", + "logging/tf_graph_writer.cc", + ], + hdrs = [ + "src/tf_graphcycles.h", + ], + deps = [ + "@org_tensorflow//tensorflow/core:protos_all_proto_text", + "@org_tensorflow//tensorflow/core:framework_headers_lib", + "@org_tensorflow//tensorflow/core:core_cpu_headers_lib", + "@ngraph//:ngraph_core", + ], + copts = [ + "-I external/ngraph_tf/src", + "-I external/ngraph_tf/logging", + "-I external/ngraph/src", + "-D NGRAPH_EMBEDDED_IN_TENSORFLOW=1", + ], + alwayslink = 1, + visibility = ["//visibility:public"], +) + +tf_cc_test( + name = "ngraph_tf_tests", + size = "small", + srcs = [ + "test/tf_exec.cpp", + "test/main.cpp", + ], + deps = [ + ":ngraph_tf", + "@com_google_googletest//:gtest", + "@org_tensorflow//tensorflow/cc:cc_ops", + "@org_tensorflow//tensorflow/cc:client_session", + "@org_tensorflow//tensorflow/core:tensorflow", + ], + extra_copts = [ + "-fexceptions ", + "-D NGRAPH_EMBEDDED_IN_TENSORFLOW=1", + "-I external/ngraph_tf/src", + "-I external/ngraph_tf/logging", + "-I external/ngraph/src", + ], +) diff --git a/third_party/ngraph/nlohmann_json.BUILD b/third_party/ngraph/nlohmann_json.BUILD new file mode 100644 index 0000000000..04c8db6a96 --- /dev/null +++ b/third_party/ngraph/nlohmann_json.BUILD @@ -0,0 +1,15 @@ +licenses(["notice"]) # 3-Clause BSD + +exports_files(["LICENSE.MIT"]) + +cc_library( + name = "nlohmann_json_lib", + hdrs = glob([ + "include/nlohmann/**/*.hpp", + ]), + copts = [ + "-I external/nlohmann_json_lib", + ], + visibility = ["//visibility:public"], + alwayslink = 1, +) diff --git a/third_party/png.BUILD b/third_party/png.BUILD index 6a7ad719aa..c26a289717 100644 --- a/third_party/png.BUILD +++ b/third_party/png.BUILD @@ -9,27 +9,62 @@ cc_library( name = "png", srcs = [ "png.c", + "pngdebug.h", "pngerror.c", "pngget.c", + "pnginfo.h", + "pnglibconf.h", "pngmem.c", "pngpread.c", + "pngpriv.h", "pngread.c", "pngrio.c", "pngrtran.c", "pngrutil.c", "pngset.c", + "pngstruct.h", "pngtrans.c", "pngwio.c", "pngwrite.c", "pngwtran.c", "pngwutil.c", - ], + ] + select({ + ":windows": [ + "intel/intel_init.c", + "intel/filter_sse2_intrinsics.c", + ], + "@org_tensorflow//tensorflow:linux_ppc64le": [ + "powerpc/powerpc_init.c", + "powerpc/filter_vsx_intrinsics.c", + ], + "//conditions:default": [ + ], + }), hdrs = [ "png.h", "pngconf.h", ], includes = ["."], - linkopts = ["-lm"], + copts = select({ + ":windows": ["-DPNG_INTEL_SSE_OPT=1"], + "//conditions:default": [], + }), + linkopts = select({ + ":windows": [], + "//conditions:default": ["-lm"], + }), visibility = ["//visibility:public"], deps = ["@zlib_archive//:zlib"], ) + +genrule( + name = "snappy_stubs_public_h", + srcs = ["scripts/pnglibconf.h.prebuilt"], + outs = ["pnglibconf.h"], + cmd = "sed -e 's/PNG_ZLIB_VERNUM 0/PNG_ZLIB_VERNUM 0x12b0/' $< >$@", +) + +config_setting( + name = "windows", + values = {"cpu": "x64_windows"}, +) diff --git a/third_party/png_fix_rpi.patch b/third_party/png_fix_rpi.patch new file mode 100644 index 0000000000..80da7b3c06 --- /dev/null +++ b/third_party/png_fix_rpi.patch @@ -0,0 +1,16 @@ +diff -r -u /tmp/libpng-1.6.34/scripts/pnglibconf.h.prebuilt ./scripts/pnglibconf.h.prebuilt +--- /tmp/libpng-1.6.34/scripts/pnglibconf.h.prebuilt 2017-09-29 01:42:33.000000000 -0700 ++++ ./scripts/pnglibconf.h.prebuilt 2018-05-01 09:51:24.719318242 -0700 +@@ -20,6 +20,12 @@ + #define PNG_ALIGNED_MEMORY_SUPPORTED + /*#undef PNG_ARM_NEON_API_SUPPORTED*/ + /*#undef PNG_ARM_NEON_CHECK_SUPPORTED*/ ++ ++/* Workaround not having a great build file by forcing ++ * png filter optimization to be disabled on arm */ ++#define PNG_ARM_NEON_OPT 0 ++ ++ + /*#undef PNG_POWERPC_VSX_API_SUPPORTED*/ + /*#undef PNG_POWERPC_VSX_CHECK_SUPPORTED*/ + #define PNG_BENIGN_ERRORS_SUPPORTED diff --git a/third_party/py/python_configure.bzl b/third_party/py/python_configure.bzl index 954f21f5f8..3c7e5c8469 100644 --- a/third_party/py/python_configure.bzl +++ b/third_party/py/python_configure.bzl @@ -6,6 +6,7 @@ * `PYTHON_LIB_PATH`: Location of python libraries. """ +_BAZEL_SH = "BAZEL_SH" _PYTHON_BIN_PATH = "PYTHON_BIN_PATH" _PYTHON_LIB_PATH = "PYTHON_LIB_PATH" _TF_PYTHON_CONFIG_REPO = "TF_PYTHON_CONFIG_REPO" @@ -152,6 +153,22 @@ def _get_python_bin(repository_ctx): _PYTHON_BIN_PATH, repository_ctx.os.environ.get("PATH", ""))) +def _get_bash_bin(repository_ctx): + """Gets the bash bin path.""" + bash_bin = repository_ctx.os.environ.get(_BAZEL_SH) + if bash_bin != None: + return bash_bin + else: + bash_bin_path = repository_ctx.which("bash") + if bash_bin_path != None: + return str(bash_bin_path) + else: + _fail("Cannot find bash in PATH, please make sure " + + "bash is installed and add its directory in PATH, or --define " + + "%s='/path/to/bash'.\nPATH=%s" % ( + _BAZEL_SH, repository_ctx.os.environ.get("PATH", ""))) + + def _get_python_lib(repository_ctx, python_bin): """Gets the python lib path.""" python_lib = repository_ctx.os.environ.get(_PYTHON_LIB_PATH) @@ -184,14 +201,14 @@ def _get_python_lib(repository_ctx, python_bin): " print(paths[0])\n" + "END") cmd = '%s - %s' % (python_bin, print_lib) - result = repository_ctx.execute(["bash", "-c", cmd]) + result = repository_ctx.execute([_get_bash_bin(repository_ctx), "-c", cmd]) return result.stdout.strip('\n') def _check_python_lib(repository_ctx, python_lib): """Checks the python lib path.""" cmd = 'test -d "%s" -a -x "%s"' % (python_lib, python_lib) - result = repository_ctx.execute(["bash", "-c", cmd]) + result = repository_ctx.execute([_get_bash_bin(repository_ctx), "-c", cmd]) if result.return_code == 1: _fail("Invalid python library path: %s" % python_lib) @@ -199,7 +216,7 @@ def _check_python_lib(repository_ctx, python_lib): def _check_python_bin(repository_ctx, python_bin): """Checks the python bin path.""" cmd = '[[ -x "%s" ]] && [[ ! -d "%s" ]]' % (python_bin, python_bin) - result = repository_ctx.execute(["bash", "-c", cmd]) + result = repository_ctx.execute([_get_bash_bin(repository_ctx), "-c", cmd]) if result.return_code == 1: _fail("--define %s='%s' is not executable. Is it the python binary?" % ( _PYTHON_BIN_PATH, python_bin)) @@ -294,6 +311,7 @@ def _python_autoconf_impl(repository_ctx): python_configure = repository_rule( implementation = _python_autoconf_impl, environ = [ + _BAZEL_SH, _PYTHON_BIN_PATH, _PYTHON_LIB_PATH, _TF_PYTHON_CONFIG_REPO, diff --git a/third_party/python_runtime/BUILD b/third_party/python_runtime/BUILD new file mode 100644 index 0000000000..2a1609191f --- /dev/null +++ b/third_party/python_runtime/BUILD @@ -0,0 +1,8 @@ +licenses(["notice"]) # New BSD, Python Software Foundation + +package(default_visibility = ["//visibility:public"]) + +alias( + name = "headers", + actual = "@local_config_python//:python_headers", +) diff --git a/third_party/repo.bzl b/third_party/repo.bzl index aa178fa8ca..7d1aa5dce9 100644 --- a/third_party/repo.bzl +++ b/third_party/repo.bzl @@ -16,95 +16,193 @@ _SINGLE_URL_WHITELIST = depset([ "arm_compiler", - "ortools_archive", ]) def _is_windows(ctx): - return ctx.os.name.lower().find("windows") != -1 + return ctx.os.name.lower().find("windows") != -1 def _wrap_bash_cmd(ctx, cmd): - if _is_windows(ctx): - bazel_sh = _get_env_var(ctx, "BAZEL_SH") - if not bazel_sh: - fail("BAZEL_SH environment variable is not set") - cmd = [bazel_sh, "-l", "-c", " ".join(cmd)] - return cmd + if _is_windows(ctx): + bazel_sh = _get_env_var(ctx, "BAZEL_SH") + if not bazel_sh: + fail("BAZEL_SH environment variable is not set") + cmd = [bazel_sh, "-l", "-c", " ".join(cmd)] + return cmd def _get_env_var(ctx, name): - if name in ctx.os.environ: - return ctx.os.environ[name] - else: - return None + if name in ctx.os.environ: + return ctx.os.environ[name] + else: + return None + +# Checks if we should use the system lib instead of the bundled one +def _use_system_lib(ctx, name): + syslibenv = _get_env_var(ctx, "TF_SYSTEM_LIBS") + if syslibenv: + for n in syslibenv.strip().split(","): + if n.strip() == name: + return True + return False # Executes specified command with arguments and calls 'fail' if it exited with # non-zero code def _execute_and_check_ret_code(repo_ctx, cmd_and_args): - result = repo_ctx.execute(cmd_and_args, timeout=10) - if result.return_code != 0: - fail(("Non-zero return code({1}) when executing '{0}':\n" + "Stdout: {2}\n" - + "Stderr: {3}").format(" ".join(cmd_and_args), result.return_code, - result.stdout, result.stderr)) + result = repo_ctx.execute(cmd_and_args, timeout = 10) + if result.return_code != 0: + fail(("Non-zero return code({1}) when executing '{0}':\n" + "Stdout: {2}\n" + + "Stderr: {3}").format( + " ".join(cmd_and_args), + result.return_code, + result.stdout, + result.stderr, + )) def _repos_are_siblings(): - return Label("@foo//bar").workspace_root.startswith("../") + return Label("@foo//bar").workspace_root.startswith("../") # Apply a patch_file to the repository root directory # Runs 'patch -p1' def _apply_patch(ctx, patch_file): - # Don't check patch on Windows, because patch is only available under bash. - if not _is_windows(ctx) and not ctx.which("patch"): - fail("patch command is not found, please install it") - cmd = _wrap_bash_cmd( - ctx, ["patch", "-p1", "-d", ctx.path("."), "-i", ctx.path(patch_file)]) - _execute_and_check_ret_code(ctx, cmd) + # Don't check patch on Windows, because patch is only available under bash. + if not _is_windows(ctx) and not ctx.which("patch"): + fail("patch command is not found, please install it") + cmd = _wrap_bash_cmd( + ctx, + ["patch", "-p1", "-d", ctx.path("."), "-i", ctx.path(patch_file)], + ) + _execute_and_check_ret_code(ctx, cmd) def _apply_delete(ctx, paths): - for path in paths: - if path.startswith("/"): - fail("refusing to rm -rf path starting with '/': " + path) - if ".." in path: - fail("refusing to rm -rf path containing '..': " + path) - cmd = _wrap_bash_cmd(ctx, ["rm", "-rf"] + [ctx.path(path) for path in paths]) - _execute_and_check_ret_code(ctx, cmd) + for path in paths: + if path.startswith("/"): + fail("refusing to rm -rf path starting with '/': " + path) + if ".." in path: + fail("refusing to rm -rf path containing '..': " + path) + cmd = _wrap_bash_cmd(ctx, ["rm", "-rf"] + [ctx.path(path) for path in paths]) + _execute_and_check_ret_code(ctx, cmd) def _tf_http_archive(ctx): - if ("mirror.bazel.build" not in ctx.attr.urls[0] or - (len(ctx.attr.urls) < 2 and - ctx.attr.name not in _SINGLE_URL_WHITELIST)): - fail("tf_http_archive(urls) must have redundant URLs. The " + - "mirror.bazel.build URL must be present and it must come first. " + - "Even if you don't have permission to mirror the file, please " + - "put the correctly formatted mirror URL there anyway, because " + - "someone will come along shortly thereafter and mirror the file.") - ctx.download_and_extract( - ctx.attr.urls, - "", - ctx.attr.sha256, - ctx.attr.type, - ctx.attr.strip_prefix) - if ctx.attr.delete: - _apply_delete(ctx, ctx.attr.delete) - if ctx.attr.patch_file != None: - _apply_patch(ctx, ctx.attr.patch_file) - if ctx.attr.build_file != None: - ctx.template("BUILD", ctx.attr.build_file, { - "%prefix%": ".." if _repos_are_siblings() else "external", - }, False) + if ("mirror.bazel.build" not in ctx.attr.urls[0] and + (len(ctx.attr.urls) < 2 and + ctx.attr.name not in _SINGLE_URL_WHITELIST)): + fail("tf_http_archive(urls) must have redundant URLs. The " + + "mirror.bazel.build URL must be present and it must come first. " + + "Even if you don't have permission to mirror the file, please " + + "put the correctly formatted mirror URL there anyway, because " + + "someone will come along shortly thereafter and mirror the file.") + + use_syslib = _use_system_lib(ctx, ctx.attr.name) + if not use_syslib: + ctx.download_and_extract( + ctx.attr.urls, + "", + ctx.attr.sha256, + ctx.attr.type, + ctx.attr.strip_prefix, + ) + if ctx.attr.delete: + _apply_delete(ctx, ctx.attr.delete) + if ctx.attr.patch_file != None: + _apply_patch(ctx, ctx.attr.patch_file) + + if use_syslib and ctx.attr.system_build_file != None: + # Use BUILD.bazel to avoid conflict with third party projects with + # BUILD or build (directory) underneath. + ctx.template("BUILD.bazel", ctx.attr.system_build_file, { + "%prefix%": ".." if _repos_are_siblings() else "external", + }, False) + + elif ctx.attr.build_file != None: + # Use BUILD.bazel to avoid conflict with third party projects with + # BUILD or build (directory) underneath. + ctx.template("BUILD.bazel", ctx.attr.build_file, { + "%prefix%": ".." if _repos_are_siblings() else "external", + }, False) tf_http_archive = repository_rule( - implementation=_tf_http_archive, - attrs={ - "sha256": attr.string(mandatory=True), - "urls": attr.string_list(mandatory=True, allow_empty=False), + implementation = _tf_http_archive, + attrs = { + "sha256": attr.string(mandatory = True), + "urls": attr.string_list(mandatory = True, allow_empty = False), "strip_prefix": attr.string(), "type": attr.string(), "delete": attr.string_list(), "patch_file": attr.label(), "build_file": attr.label(), - }) + "system_build_file": attr.label(), + }, + environ = [ + "TF_SYSTEM_LIBS", + ], +) """Downloads and creates Bazel repos for dependencies. This is a swappable replacement for both http_archive() and new_http_archive() that offers some additional features. It also helps ensure best practices are followed. """ + +def _third_party_http_archive(ctx): + if ("mirror.bazel.build" not in ctx.attr.urls[0] and + (len(ctx.attr.urls) < 2 and + ctx.attr.name not in _SINGLE_URL_WHITELIST)): + fail("tf_http_archive(urls) must have redundant URLs. The " + + "mirror.bazel.build URL must be present and it must come first. " + + "Even if you don't have permission to mirror the file, please " + + "put the correctly formatted mirror URL there anyway, because " + + "someone will come along shortly thereafter and mirror the file.") + + use_syslib = _use_system_lib(ctx, ctx.attr.name) + + # Use "BUILD.bazel" to avoid conflict with third party projects that contain a + # file or directory called "BUILD" + buildfile_path = ctx.path("BUILD.bazel") + + if use_syslib: + if ctx.attr.system_build_file == None: + fail("Bazel was configured with TF_SYSTEM_LIBS to use a system " + + "library for %s, but no system build file for %s was configured. " + + "Please add a system_build_file attribute to the repository rule" + + "for %s." % (ctx.attr.name, ctx.attr.name, ctx.attr.name)) + ctx.symlink(Label(ctx.attr.system_build_file), buildfile_path) + + else: + ctx.download_and_extract( + ctx.attr.urls, + "", + ctx.attr.sha256, + ctx.attr.type, + ctx.attr.strip_prefix, + ) + if ctx.attr.delete: + _apply_delete(ctx, ctx.attr.delete) + if ctx.attr.patch_file != None: + _apply_patch(ctx, ctx.attr.patch_file) + ctx.symlink(Label(ctx.attr.build_file), buildfile_path) + + for internal_src, external_dest in ctx.attr.link_files.items(): + ctx.symlink(Label(internal_src), ctx.path(external_dest)) + +# Downloads and creates Bazel repos for dependencies. +# +# This is an upgrade for tf_http_archive that works with go/tfbr-thirdparty. +# +# For link_files, specify each dict entry as: +# "//path/to/source:file": "localfile" +third_party_http_archive = repository_rule( + implementation = _third_party_http_archive, + attrs = { + "sha256": attr.string(mandatory = True), + "urls": attr.string_list(mandatory = True, allow_empty = False), + "strip_prefix": attr.string(), + "type": attr.string(), + "delete": attr.string_list(), + "build_file": attr.string(mandatory = True), + "system_build_file": attr.string(mandatory = False), + "patch_file": attr.label(), + "link_files": attr.string_dict(), + }, + environ = [ + "TF_SYSTEM_LIBS", + ], +) diff --git a/third_party/snappy.BUILD b/third_party/snappy.BUILD index fd48ed8941..d93f030769 100644 --- a/third_party/snappy.BUILD +++ b/third_party/snappy.BUILD @@ -4,25 +4,12 @@ licenses(["notice"]) # BSD 3-Clause exports_files(["COPYING"]) -config_setting( - name = "windows", - values = {"cpu": "x64_windows"}, - visibility = ["//visibility:public"], -) - -config_setting( - name = "windows_msvc", - values = {"cpu": "x64_windows_msvc"}, - visibility = ["//visibility:public"], -) - cc_library( name = "snappy", srcs = [ + "config.h", "snappy.cc", "snappy.h", - "snappy-c.cc", - "snappy-c.h", "snappy-internal.h", "snappy-sinksource.cc", "snappy-sinksource.h", @@ -31,10 +18,11 @@ cc_library( "snappy-stubs-public.h", ], hdrs = ["snappy.h"], - copts = select({ - ":windows": [], - ":windows_msvc": [], + copts = ["-DHAVE_CONFIG_H"] + select({ + "@org_tensorflow//tensorflow:windows": [], "//conditions:default": [ + "-fno-exceptions", + "-Wno-sign-compare", "-Wno-shift-negative-value", "-Wno-implicit-function-declaration", ], @@ -42,20 +30,66 @@ cc_library( ) genrule( + name = "config_h", + outs = ["config.h"], + cmd = "\n".join([ + "cat <<'EOF' >$@", + "#define HAVE_STDDEF_H 1", + "#define HAVE_STDINT_H 1", + "", + "#ifdef __has_builtin", + "# if !defined(HAVE_BUILTIN_EXPECT) && __has_builtin(__builtin_expect)", + "# define HAVE_BUILTIN_EXPECT 1", + "# endif", + "# if !defined(HAVE_BUILTIN_CTZ) && __has_builtin(__builtin_ctzll)", + "# define HAVE_BUILTIN_CTZ 1", + "# endif", + "#elif defined(__GNUC__) && (__GNUC__ > 3 || __GNUC__ == 3 && __GNUC_MINOR__ >= 4)", + "# ifndef HAVE_BUILTIN_EXPECT", + "# define HAVE_BUILTIN_EXPECT 1", + "# endif", + "# ifndef HAVE_BUILTIN_CTZ", + "# define HAVE_BUILTIN_CTZ 1", + "# endif", + "#endif", + "", + "#ifdef __has_include", + "# if !defined(HAVE_BYTESWAP_H) && __has_include(<byteswap.h>)", + "# define HAVE_BYTESWAP_H 1", + "# endif", + "# if !defined(HAVE_UNISTD_H) && __has_include(<unistd.h>)", + "# define HAVE_UNISTD_H 1", + "# endif", + "# if !defined(HAVE_SYS_ENDIAN_H) && __has_include(<sys/endian.h>)", + "# define HAVE_SYS_ENDIAN_H 1", + "# endif", + "# if !defined(HAVE_SYS_MMAN_H) && __has_include(<sys/mman.h>)", + "# define HAVE_SYS_MMAN_H 1", + "# endif", + "# if !defined(HAVE_SYS_UIO_H) && __has_include(<sys/uio.h>)", + "# define HAVE_SYS_UIO_H 1", + "# endif", + "#endif", + "", + "#ifndef SNAPPY_IS_BIG_ENDIAN", + "# ifdef __s390x__", + "# define SNAPPY_IS_BIG_ENDIAN 1", + "# elif defined(__BYTE_ORDER__) && defined(__ORDER_BIG_ENDIAN__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__", + "# define SNAPPY_IS_BIG_ENDIAN 1", + "# endif", + "#endif", + "EOF", + ]), +) + +genrule( name = "snappy_stubs_public_h", srcs = ["snappy-stubs-public.h.in"], outs = ["snappy-stubs-public.h"], cmd = ("sed " + - "-e 's/@ac_cv_have_stdint_h@/1/g' " + - "-e 's/@ac_cv_have_stddef_h@/1/g' " + - "-e 's/@ac_cv_have_stdint_h@/1/g' " + - select({ - "@org_tensorflow//tensorflow:windows": "-e 's/@ac_cv_have_sys_uio_h@/0/g' ", - "@org_tensorflow//tensorflow:windows_msvc": "-e 's/@ac_cv_have_sys_uio_h@/0/g' ", - "//conditions:default": "-e 's/@ac_cv_have_sys_uio_h@/1/g' ", - }) + - "-e 's/@SNAPPY_MAJOR@/1/g' " + - "-e 's/@SNAPPY_MINOR@/1/g' " + - "-e 's/@SNAPPY_PATCHLEVEL@/4/g' " + + "-e 's/$${\\(.*\\)_01}/\\1/g' " + + "-e 's/$${SNAPPY_MAJOR}/1/g' " + + "-e 's/$${SNAPPY_MINOR}/1/g' " + + "-e 's/$${SNAPPY_PATCHLEVEL}/4/g' " + "$< >$@"), ) diff --git a/third_party/sqlite.BUILD b/third_party/sqlite.BUILD index 6da7953589..8b876fb56f 100644 --- a/third_party/sqlite.BUILD +++ b/third_party/sqlite.BUILD @@ -4,7 +4,7 @@ licenses(["unencumbered"]) # Public Domain SQLITE_COPTS = [ - "-Os", + "-DSQLITE_ENABLE_JSON1", "-DHAVE_DECL_STRERROR_R=1", "-DHAVE_STDINT_H=1", "-DHAVE_INTTYPES_H=1", @@ -14,15 +14,14 @@ SQLITE_COPTS = [ "@org_tensorflow//tensorflow:windows": [ "-DSQLITE_MAX_TRIGGER_DEPTH=100", ], - "@org_tensorflow//tensorflow:windows_msvc": [ - "-DSQLITE_MAX_TRIGGER_DEPTH=100", - ], "@org_tensorflow//tensorflow:darwin": [ + "-Os", "-DHAVE_GMTIME_R=1", "-DHAVE_LOCALTIME_R=1", "-DHAVE_USLEEP=1", ], "//conditions:default": [ + "-Os", "-DHAVE_FDATASYNC=1", "-DHAVE_GMTIME_R=1", "-DHAVE_LOCALTIME_R=1", @@ -47,7 +46,7 @@ cc_library( "SQLITE_OMIT_DEPRECATED", ], linkopts = select({ - "@org_tensorflow//tensorflow:windows_msvc": [], + "@org_tensorflow//tensorflow:windows": [], "//conditions:default": [ "-ldl", "-lpthread", diff --git a/third_party/swig.BUILD b/third_party/swig.BUILD index f2f647401b..59a3d9e671 100644 --- a/third_party/swig.BUILD +++ b/third_party/swig.BUILD @@ -71,7 +71,6 @@ cc_binary( ], copts = ["$(STACK_FRAME_UNLIMITED)"] + select({ ":windows": [], - ":windows_msvc": [], "//conditions:default": [ "-Wno-parentheses", "-Wno-unused-variable", @@ -332,11 +331,6 @@ genrule( ) config_setting( - name = "windows_msvc", - values = {"cpu": "x64_windows_msvc"}, -) - -config_setting( name = "windows", values = {"cpu": "x64_windows"}, ) diff --git a/third_party/sycl/BUILD b/third_party/sycl/BUILD index fbdf19f205..f631b6df06 100644 --- a/third_party/sycl/BUILD +++ b/third_party/sycl/BUILD @@ -1,15 +1,3 @@ package(default_visibility = ["//visibility:public"]) licenses(["notice"]) # Apache 2.0 - -filegroup( - name = "all_files", - srcs = glob( - ["**/*"], - exclude = [ - "**/METADATA", - "**/OWNERS", - ], - ), - visibility = ["//tensorflow:__subpackages__"], -) diff --git a/third_party/sycl/sycl/BUILD b/third_party/sycl/sycl/BUILD index bc1d18b7b5..b045609954 100644 --- a/third_party/sycl/sycl/BUILD +++ b/third_party/sycl/sycl/BUILD @@ -5,15 +5,3 @@ package(default_visibility = ["//visibility:public"]) licenses(["notice"]) # Apache 2.0 - -filegroup( - name = "all_files", - srcs = glob( - ["**/*"], - exclude = [ - "**/METADATA", - "**/OWNERS", - ], - ), - visibility = ["//tensorflow:__subpackages__"], -) diff --git a/third_party/sycl/sycl/BUILD.tpl b/third_party/sycl/sycl/BUILD.tpl index 21b1a2bbf7..b7e9aa8edb 100755 --- a/third_party/sycl/sycl/BUILD.tpl +++ b/third_party/sycl/sycl/BUILD.tpl @@ -21,7 +21,7 @@ config_setting( name = "using_sycl_trisycl", define_values = { "using_sycl": "true", - "using_trisycl": "false", + "using_trisycl": "true", }, ) diff --git a/third_party/systemlibs/BUILD b/third_party/systemlibs/BUILD new file mode 100644 index 0000000000..e69de29bb2 --- /dev/null +++ b/third_party/systemlibs/BUILD diff --git a/third_party/systemlibs/BUILD.tpl b/third_party/systemlibs/BUILD.tpl new file mode 100644 index 0000000000..e69de29bb2 --- /dev/null +++ b/third_party/systemlibs/BUILD.tpl diff --git a/third_party/systemlibs/astor.BUILD b/third_party/systemlibs/astor.BUILD new file mode 100644 index 0000000000..497ec4bcea --- /dev/null +++ b/third_party/systemlibs/astor.BUILD @@ -0,0 +1,12 @@ +licenses(["notice"]) # New BSD + +filegroup( + name = "LICENSE", + visibility = ["//visibility:public"], +) + +py_library( + name = "astor", + srcs_version = "PY2AND3", + visibility = ["//visibility:public"], +) diff --git a/third_party/systemlibs/build_defs.bzl.tpl b/third_party/systemlibs/build_defs.bzl.tpl new file mode 100644 index 0000000000..3faa46c581 --- /dev/null +++ b/third_party/systemlibs/build_defs.bzl.tpl @@ -0,0 +1,32 @@ +# -*- Python -*- +"""Skylark macros for system libraries. +""" + +SYSTEM_LIBS_ENABLED = %{syslibs_enabled} + +SYSTEM_LIBS_LIST = [ +%{syslibs_list} +] + + +def if_any_system_libs(a, b=[]): + """Conditional which evaluates to 'a' if any system libraries are configured.""" + if SYSTEM_LIBS_ENABLED: + return a + else: + return b + + +def if_system_lib(lib, a, b=[]): + """Conditional which evaluates to 'a' if we're using the system version of lib""" + + if SYSTEM_LIBS_ENABLED and lib in SYSTEM_LIBS_LIST: + return a + else: + return b + + +def if_not_system_lib(lib, a, b=[]): + """Conditional which evaluates to 'a' if we're using the system version of lib""" + + return if_system_lib(lib, b, a) diff --git a/third_party/systemlibs/curl.BUILD b/third_party/systemlibs/curl.BUILD new file mode 100644 index 0000000000..c5f125caa9 --- /dev/null +++ b/third_party/systemlibs/curl.BUILD @@ -0,0 +1,12 @@ +licenses(["notice"]) # MIT/X derivative license + +filegroup( + name = "COPYING", + visibility = ["//visibility:public"], +) + +cc_library( + name = "curl", + linkopts = ["-lcurl"], + visibility = ["//visibility:public"], +) diff --git a/third_party/systemlibs/cython.BUILD b/third_party/systemlibs/cython.BUILD new file mode 100644 index 0000000000..1d52587676 --- /dev/null +++ b/third_party/systemlibs/cython.BUILD @@ -0,0 +1,13 @@ +licenses(["notice"]) # Apache-2.0 + +genrule( + name = "lncython", + outs = ["cython"], + cmd = "ln -s $$(which cython) $@", +) + +sh_binary( + name = "cython_binary", + srcs = ["cython"], + visibility = ["//visibility:public"], +) diff --git a/third_party/systemlibs/gif.BUILD b/third_party/systemlibs/gif.BUILD new file mode 100644 index 0000000000..5eb2c918ba --- /dev/null +++ b/third_party/systemlibs/gif.BUILD @@ -0,0 +1,12 @@ +licenses(["notice"]) # MIT + +filegroup( + name = "COPYING", + visibility = ["//visibility:public"], +) + +cc_library( + name = "gif", + linkopts = ["-lgif"], + visibility = ["//visibility:public"], +) diff --git a/third_party/systemlibs/grpc.BUILD b/third_party/systemlibs/grpc.BUILD new file mode 100644 index 0000000000..fd90eb0dd3 --- /dev/null +++ b/third_party/systemlibs/grpc.BUILD @@ -0,0 +1,54 @@ +licenses(["notice"]) # Apache v2 + +filegroup( + name = "LICENSE", + visibility = ["//visibility:public"], +) + +cc_library( + name = "grpc", + linkopts = ["-lgrpc"], + visibility = ["//visibility:public"], +) + +cc_library( + name = "grpc++", + linkopts = ["-lgrpc++"], + visibility = ["//visibility:public"], +) + +cc_library( + name = "grpc_unsecure", + linkopts = ["-lgrpc_unsecure"], + visibility = ["//visibility:public"], +) + +cc_library( + name = "grpc++_unsecure", + linkopts = ["-lgrpc++_unsecure"], + visibility = ["//visibility:public"], +) + +genrule( + name = "ln_grpc_cpp_plugin", + outs = ["grpc_cpp_plugin.bin"], + cmd = "ln -s $$(which grpc_cpp_plugin) $@", +) + +sh_binary( + name = "grpc_cpp_plugin", + srcs = ["grpc_cpp_plugin.bin"], + visibility = ["//visibility:public"], +) + +genrule( + name = "ln_grpc_python_plugin", + outs = ["grpc_python_plugin.bin"], + cmd = "ln -s $$(which grpc_python_plugin) $@", +) + +sh_binary( + name = "grpc_python_plugin", + srcs = ["grpc_python_plugin.bin"], + visibility = ["//visibility:public"], +) diff --git a/third_party/systemlibs/jemalloc.BUILD b/third_party/systemlibs/jemalloc.BUILD new file mode 100644 index 0000000000..6a48d582ba --- /dev/null +++ b/third_party/systemlibs/jemalloc.BUILD @@ -0,0 +1,30 @@ +licenses(["notice"]) # BSD + +filegroup( + name = "COPYING", + visibility = ["//visibility:public"], +) + +cc_library( + name = "jemalloc_headers", + defines = [ + "jemalloc_posix_memalign=posix_memalign", + "jemalloc_malloc=malloc", + "jemalloc_realloc=realloc", + "jemalloc_free=free", + ], + visibility = ["//visibility:public"], +) + +cc_library( + name = "jemalloc_impl", + linkopts = ["-ljemalloc"], + defines = [ + "jemalloc_posix_memalign=posix_memalign", + "jemalloc_malloc=malloc", + "jemalloc_realloc=realloc", + "jemalloc_free=free", + ], + visibility = ["//visibility:public"], + deps = [":jemalloc_headers"], +) diff --git a/third_party/systemlibs/jpeg.BUILD b/third_party/systemlibs/jpeg.BUILD new file mode 100644 index 0000000000..f4f52da9bd --- /dev/null +++ b/third_party/systemlibs/jpeg.BUILD @@ -0,0 +1,12 @@ +licenses(["notice"]) # custom notice-style license, see LICENSE.md + +filegroup( + name = "LICENSE.md", + visibility = ["//visibility:public"], +) + +cc_library( + name = "jpeg", + linkopts = ["-ljpeg"], + visibility = ["//visibility:public"], +) diff --git a/third_party/systemlibs/jsoncpp.BUILD b/third_party/systemlibs/jsoncpp.BUILD new file mode 100644 index 0000000000..cf91917cfb --- /dev/null +++ b/third_party/systemlibs/jsoncpp.BUILD @@ -0,0 +1,37 @@ +licenses(["unencumbered"]) # Public Domain or MIT + +filegroup( + name = "LICENSE", + visibility = ["//visibility:public"], +) + +HEADERS = [ + "include/json/autolink.h", + "include/json/config.h", + "include/json/features.h", + "include/json/forwards.h", + "include/json/json.h", + "include/json/reader.h", + "include/json/value.h", + "include/json/version.h", + "include/json/writer.h", +] + +genrule( + name = "link_headers", + outs = HEADERS, + cmd = """ + for i in $(OUTS); do + i=$${i##*/} + ln -vsf /usr/include/jsoncpp/json/$$i $(@D)/include/json/$$i + done + """, +) + +cc_library( + name = "jsoncpp", + hdrs = HEADERS, + includes = ["."], + linkopts = ["-ljsoncpp"], + visibility = ["//visibility:public"], +) diff --git a/third_party/systemlibs/lmdb.BUILD b/third_party/systemlibs/lmdb.BUILD new file mode 100644 index 0000000000..6177b095ec --- /dev/null +++ b/third_party/systemlibs/lmdb.BUILD @@ -0,0 +1,12 @@ +licenses(["notice"]) # OpenLDAP Public License + +filegroup( + name = "LICENSE", + visibility = ["//visibility:public"], +) + +cc_library( + name = "lmdb", + linkopts = ["-llmdb"], + visibility = ["//visibility:public"], +) diff --git a/third_party/systemlibs/nasm.BUILD b/third_party/systemlibs/nasm.BUILD new file mode 100644 index 0000000000..10ef8d8832 --- /dev/null +++ b/third_party/systemlibs/nasm.BUILD @@ -0,0 +1,12 @@ +licenses(["notice"]) # BSD 2-clause + +filegroup( + name = "LICENSE", + visibility = ["//visibility:public"], +) + +sh_binary( + name = "nasm", + srcs = ["nasm"], + visibility = ["@jpeg//:__pkg__"], +) diff --git a/third_party/systemlibs/nsync.BUILD b/third_party/systemlibs/nsync.BUILD new file mode 100644 index 0000000000..c5d4ad0a76 --- /dev/null +++ b/third_party/systemlibs/nsync.BUILD @@ -0,0 +1,23 @@ +licenses(["notice"]) # BSD 3-Clause + +filegroup( + name = "LICENSE", + visibility = ["//visibility:public"], +) + +cc_library( + name = "nsync_headers", + visibility = ["//visibility:public"], +) + +cc_library( + name = "nsync", + linkopts = ["-lnsync"], + visibility = ["//visibility:public"], +) + +cc_library( + name = "nsync_cpp", + linkopts = ["-lnsync_cpp"], + visibility = ["//visibility:public"], +) diff --git a/third_party/systemlibs/pcre.BUILD b/third_party/systemlibs/pcre.BUILD new file mode 100644 index 0000000000..df74238847 --- /dev/null +++ b/third_party/systemlibs/pcre.BUILD @@ -0,0 +1,12 @@ +licenses(["notice"]) # BSD + +filegroup( + name = "LICENCE", + visibility = ["//visibility:public"], +) + +cc_library( + name = "pcre", + linkopts = ["-lpcre"], + visibility = ["//visibility:public"], +) diff --git a/third_party/systemlibs/png.BUILD b/third_party/systemlibs/png.BUILD new file mode 100644 index 0000000000..fc6b6f2d8b --- /dev/null +++ b/third_party/systemlibs/png.BUILD @@ -0,0 +1,12 @@ +licenses(["notice"]) # BSD/MIT-like license + +filegroup( + name = "LICENSE", + visibility = ["//visibility:public"], +) + +cc_library( + name = "png", + linkopts = ["-lpng"], + visibility = ["//visibility:public"], +) diff --git a/third_party/systemlibs/re2.BUILD b/third_party/systemlibs/re2.BUILD new file mode 100644 index 0000000000..c18e252dbc --- /dev/null +++ b/third_party/systemlibs/re2.BUILD @@ -0,0 +1,12 @@ +licenses(["notice"]) # BSD/MIT-like license + +filegroup( + name = "LICENSE", + visibility = ["//visibility:public"], +) + +cc_library( + name = "re2", + linkopts = ["-lre2"], + visibility = ["//visibility:public"], +) diff --git a/third_party/systemlibs/six.BUILD b/third_party/systemlibs/six.BUILD new file mode 100644 index 0000000000..ff9b1a540b --- /dev/null +++ b/third_party/systemlibs/six.BUILD @@ -0,0 +1,11 @@ +licenses(["notice"]) # MIT + +filegroup( + name = "LICENSE", + visibility = ["//visibility:public"], +) + +py_library( + name = "six", + visibility = ["//visibility:public"], +) diff --git a/third_party/systemlibs/snappy.BUILD b/third_party/systemlibs/snappy.BUILD new file mode 100644 index 0000000000..fd2db9e2df --- /dev/null +++ b/third_party/systemlibs/snappy.BUILD @@ -0,0 +1,12 @@ +licenses(["notice"]) # BSD 3-Clause + +filegroup( + name = "COPYING", + visibility = ["//visibility:public"], +) + +cc_library( + name = "snappy", + linkopts = ["-lsnappy"], + visibility = ["//visibility:public"], +) diff --git a/third_party/systemlibs/sqlite.BUILD b/third_party/systemlibs/sqlite.BUILD new file mode 100644 index 0000000000..20ee1ebbef --- /dev/null +++ b/third_party/systemlibs/sqlite.BUILD @@ -0,0 +1,15 @@ +licenses(["unencumbered"]) # Public Domain + +# Production build of SQLite library that's baked into TensorFlow. +cc_library( + name = "org_sqlite", + linkopts = ["-lsqlite3"], + visibility = ["//visibility:public"], +) + +# This is a Copybara sync helper for Google. +py_library( + name = "python", + srcs_version = "PY2AND3", + visibility = ["//visibility:public"], +) diff --git a/third_party/systemlibs/swig.BUILD b/third_party/systemlibs/swig.BUILD new file mode 100644 index 0000000000..4c9b74dadb --- /dev/null +++ b/third_party/systemlibs/swig.BUILD @@ -0,0 +1,23 @@ +licenses(["restricted"]) # GPLv3 + +filegroup( + name = "LICENSE", + visibility = ["//visibility:public"], +) + +filegroup( + name = "templates", + visibility = ["//visibility:public"], +) + +genrule( + name = "lnswiglink", + outs = ["swiglink"], + cmd = "ln -s $$(which swig) $@", +) + +sh_binary( + name = "swig", + srcs = ["swiglink"], + visibility = ["//visibility:public"], +) diff --git a/third_party/systemlibs/syslibs_configure.bzl b/third_party/systemlibs/syslibs_configure.bzl new file mode 100644 index 0000000000..8b09c9ac1f --- /dev/null +++ b/third_party/systemlibs/syslibs_configure.bzl @@ -0,0 +1,158 @@ +# -*- Python -*- +"""Repository rule for system library autoconfiguration. + +`syslibs_configure` depends on the following environment variables: + + * `TF_SYSTEM_LIBS`: list of third party dependencies that should use + the system version instead +""" + +_TF_SYSTEM_LIBS = "TF_SYSTEM_LIBS" + +VALID_LIBS = [ + "astor_archive", + "com_googlesource_code_re2", + "curl", + "cython", + "flatbuffers", + "gif_archive", + "grpc", + "jemalloc", + "jpeg", + "jsoncpp_git", + "lmdb", + "nasm", + "nsync", + "org_sqlite", + "pcre", + "png_archive", + "six_archive", + "snappy", + "swig", + "termcolor_archive", + "zlib_archive", +] + +def auto_configure_fail(msg): + """Output failure message when syslibs configuration fails.""" + red = "\033[0;31m" + no_color = "\033[0m" + fail("\n%sSystem Library Configuration Error:%s %s\n" % (red, no_color, msg)) + +def _is_windows(repository_ctx): + """Returns true if the host operating system is windows.""" + os_name = repository_ctx.os.name.lower() + if os_name.find("windows") != -1: + return True + return False + +def _enable_syslibs(repository_ctx): + s = repository_ctx.os.environ.get(_TF_SYSTEM_LIBS, "").strip() + if not _is_windows(repository_ctx) and s != None and s != "": + return True + return False + +def _get_system_lib_list(repository_ctx): + """Gets the list of deps that should use the system lib. + + Args: + repository_ctx: The repository context. + + Returns: + A string version of a python list + """ + if _TF_SYSTEM_LIBS not in repository_ctx.os.environ: + return [] + + libenv = repository_ctx.os.environ[_TF_SYSTEM_LIBS].strip() + libs = [] + + for lib in list(libenv.split(",")): + lib = lib.strip() + if lib == "": + continue + if lib not in VALID_LIBS: + auto_configure_fail("Invalid system lib set: %s" % lib) + return [] + libs.append(lib) + + return libs + +def _format_system_lib_list(repository_ctx): + """Formats the list of deps that should use the system lib. + + Args: + repository_ctx: The repository context. + + Returns: + A list of the names of deps that should use the system lib. + """ + libs = _get_system_lib_list(repository_ctx) + ret = "" + for lib in libs: + ret += "'%s',\n" % lib + + return ret + +def _tpl(repository_ctx, tpl, substitutions = {}, out = None): + if not out: + out = tpl.replace(":", "") + repository_ctx.template( + out, + Label("//third_party/systemlibs%s.tpl" % tpl), + substitutions, + False, + ) + +def _create_dummy_repository(repository_ctx): + """Creates the dummy repository to build with all bundled libraries.""" + + _tpl(repository_ctx, ":BUILD") + _tpl( + repository_ctx, + ":build_defs.bzl", + { + "%{syslibs_enabled}": "False", + "%{syslibs_list}": "", + }, + ) + +def _create_local_repository(repository_ctx): + """Creates the repository to build with system libraries.""" + + _tpl(repository_ctx, ":BUILD") + _tpl( + repository_ctx, + ":build_defs.bzl", + { + "%{syslibs_enabled}": "True", + "%{syslibs_list}": _format_system_lib_list(repository_ctx), + }, + ) + +def _syslibs_autoconf_impl(repository_ctx): + """Implementation of the syslibs_configure repository rule.""" + if not _enable_syslibs(repository_ctx): + _create_dummy_repository(repository_ctx) + else: + _create_local_repository(repository_ctx) + +syslibs_configure = repository_rule( + implementation = _syslibs_autoconf_impl, + environ = [ + _TF_SYSTEM_LIBS, + ], +) + +"""Configures the build to link to system libraries +instead of using bundled versions. + +Add the following to your WORKSPACE FILE: + +```python +syslibs_configure(name = "local_config_syslibs") +``` + +Args: + name: A unique name for this workspace rule. +""" diff --git a/third_party/systemlibs/termcolor.BUILD b/third_party/systemlibs/termcolor.BUILD new file mode 100644 index 0000000000..915eb621d5 --- /dev/null +++ b/third_party/systemlibs/termcolor.BUILD @@ -0,0 +1,12 @@ +licenses(["notice"]) # MIT + +filegroup( + name = "COPYING.txt", + visibility = ["//visibility:public"], +) + +py_library( + name = "termcolor", + srcs_version = "PY2AND3", + visibility = ["//visibility:public"], +) diff --git a/third_party/systemlibs/zlib.BUILD b/third_party/systemlibs/zlib.BUILD new file mode 100644 index 0000000000..69462ae6cb --- /dev/null +++ b/third_party/systemlibs/zlib.BUILD @@ -0,0 +1,12 @@ +licenses(["notice"]) # BSD/MIT-like license (for zlib) + +filegroup( + name = "zlib.h", + visibility = ["//visibility:public"], +) + +cc_library( + name = "zlib", + linkopts = ["-lz"], + visibility = ["//visibility:public"], +) diff --git a/third_party/tensorrt/tensorrt_configure.bzl b/third_party/tensorrt/tensorrt_configure.bzl index 8e76e5d02a..9b946505a6 100644 --- a/third_party/tensorrt/tensorrt_configure.bzl +++ b/third_party/tensorrt/tensorrt_configure.bzl @@ -57,6 +57,10 @@ def _find_trt_header_dir(repository_ctx, trt_install_path): path = "/usr/include/x86_64-linux-gnu" if _headers_exist(repository_ctx, path): return path + if trt_install_path == "/usr/lib/aarch64-linux-gnu": + path = "/usr/include/aarch64-linux-gnu" + if _headers_exist(repository_ctx, path): + return path path = str(repository_ctx.path("%s/../include" % trt_install_path).realpath) if _headers_exist(repository_ctx, path): return path diff --git a/third_party/tflite_ovic_testdata.BUILD b/third_party/tflite_ovic_testdata.BUILD new file mode 100644 index 0000000000..de47ed61f9 --- /dev/null +++ b/third_party/tflite_ovic_testdata.BUILD @@ -0,0 +1,12 @@ +package(default_visibility = ["//visibility:public"]) + +licenses(["notice"]) # Apache 2.0 + +exports_files( + glob( + ["**/*"], + exclude = [ + "BUILD", + ], + ), +) diff --git a/third_party/toolchains/BUILD b/third_party/toolchains/BUILD new file mode 100644 index 0000000000..4303751452 --- /dev/null +++ b/third_party/toolchains/BUILD @@ -0,0 +1,37 @@ +licenses(["restricted"]) + +package(default_visibility = ["//visibility:public"]) + +# Platform for use with remote execution with +# custom container based off RBE Ubuntu16_04 +# http://gcr.io/cloud-marketplace/google/rbe-ubuntu16-04 +# Built with //tensorflow/tools/ci_build/Dockerfile.rbe.cpu +platform( + name = "rbe_ubuntu16_04-tf", + constraint_values = [ + "@bazel_tools//platforms:x86_64", + "@bazel_tools//platforms:linux", + "@bazel_tools//tools/cpp:clang", + "@bazel_toolchains//constraints:xenial", + ], + remote_execution_properties = """ + properties: { + name: "container-image" + value:"docker://gcr.io/asci-toolchain/nosla-ubuntu16_04-tf@sha256:495a025ed5e273cfa5d53357ef93ac20500c008994e0be106c509f51555fb93c" + }""", +) + +platform( + name = "rbe_cuda9.0-cudnn7-ubuntu14.04", + constraint_values = [ + "@bazel_tools//platforms:x86_64", + "@bazel_tools//platforms:linux", + "@bazel_tools//tools/cpp:clang", + "@bazel_toolchains//constraints:xenial", + ], + remote_execution_properties = """ + properties: { + name: "container-image" + value:"docker://gcr.io/asci-toolchain/nosla-cuda9.0-cudnn7-ubuntu14.04@sha256:ae58329b961e7c17d89725bf8fd72dfbd5850f4f3313de58e0cafbf5b0343735" + }""", +) diff --git a/third_party/toolchains/clang6/CROSSTOOL.tpl b/third_party/toolchains/clang6/CROSSTOOL.tpl index 6b7e5a8808..ffba9850bb 100644 --- a/third_party/toolchains/clang6/CROSSTOOL.tpl +++ b/third_party/toolchains/clang6/CROSSTOOL.tpl @@ -76,9 +76,6 @@ toolchain { # This adds a little bit more durability to our Clang build. # - # At the moment, this only only be needed for: - # - add_boringssl_s390x.patch: --Wa,--noexecstack - # # Folks who do maintenance work on TF Bazel Clang should consider # commenting out these lines, while doing that work, to gain a better # understanding of what the intersection of support looks like between GCC diff --git a/third_party/toolchains/cpus/py/BUILD b/third_party/toolchains/cpus/py/BUILD index c175742cbf..1235988abb 100644 --- a/third_party/toolchains/cpus/py/BUILD +++ b/third_party/toolchains/cpus/py/BUILD @@ -6,18 +6,24 @@ licenses(["restricted"]) package(default_visibility = ["//visibility:public"]) +# To build Python C/C++ extension on Windows, we need to link to python import library pythonXY.lib +# See https://docs.python.org/3/extending/windows.html +cc_import( + name = "python_lib", + interface_library = select({ + ":windows": ":python_import_lib", + # A placeholder for Unix platforms which makes --no_build happy. + "//conditions:default": "not-existing.lib", + }), + system_provided = 1, +) + cc_library( name = "python_headers", hdrs = [":python_include"], - data = select({ - ":windows": [":python_import_lib"], - "//conditions:default": [], - }), includes = ["python_include"], - linkopts = select({ - # TODO(pcloudy): Ideally, this should just go into deps after resolving - # https://github.com/bazelbuild/bazel/issues/3237, - ":windows": ["$(locations :python_import_lib)"], + deps = select({ + ":windows": [":python_lib"], "//conditions:default": [], }), ) @@ -37,161 +43,135 @@ config_setting( genrule( name = "python_include", outs = [ + "python_include/Python-ast.h", + "python_include/Python.h", + "python_include/abstract.h", + "python_include/asdl.h", + "python_include/ast.h", + "python_include/bitset.h", + "python_include/boolobject.h", + "python_include/bufferobject.h", + "python_include/bytearrayobject.h", + "python_include/bytes_methods.h", + "python_include/bytesobject.h", + "python_include/cStringIO.h", + "python_include/cellobject.h", + "python_include/ceval.h", + "python_include/classobject.h", + "python_include/cobject.h", "python_include/code.h", + "python_include/codecs.h", + "python_include/compile.h", + "python_include/complexobject.h", + "python_include/datetime.h", + "python_include/descrobject.h", + "python_include/dictobject.h", "python_include/dtoa.h", - "python_include/tupleobject.h", - "python_include/object.h", - "python_include/ast.h", - "python_include/pymacconfig.h", + "python_include/enumobject.h", "python_include/errcode.h", + "python_include/eval.h", + "python_include/fileobject.h", + "python_include/floatobject.h", "python_include/frameobject.h", - "python_include/pgenheaders.h", - "python_include/cellobject.h", + "python_include/funcobject.h", + "python_include/genobject.h", + "python_include/graminit.h", + "python_include/grammar.h", + "python_include/import.h", "python_include/intobject.h", - "python_include/pythread.h", - "python_include/cStringIO.h", - "python_include/boolobject.h", + "python_include/intrcheck.h", + "python_include/iterobject.h", + "python_include/listobject.h", + "python_include/longintrepr.h", + "python_include/longobject.h", + "python_include/marshal.h", + "python_include/memoryobject.h", + "python_include/metagrammar.h", + "python_include/methodobject.h", "python_include/modsupport.h", - "python_include/import.h", - "python_include/pymath.h", + "python_include/moduleobject.h", "python_include/node.h", - "python_include/funcobject.h", - "python_include/eval.h", - "python_include/longintrepr.h", - "python_include/floatobject.h", - "python_include/rangeobject.h", - "python_include/pyfpe.h", - "python_include/pystrcmp.h", - "python_include/dictobject.h", - "python_include/pyarena.h", + "python_include/object.h", "python_include/objimpl.h", - "python_include/bitset.h", - "python_include/memoryobject.h", - "python_include/bytearrayobject.h", + "python_include/opcode.h", + "python_include/osdefs.h", + "python_include/parsetok.h", + "python_include/patchlevel.h", + "python_include/pgen.h", + "python_include/pgenheaders.h", + "python_include/py_curses.h", + "python_include/pyarena.h", + "python_include/pycapsule.h", + "python_include/pyconfig.h", + "python_include/pyctype.h", "python_include/pydebug.h", "python_include/pyerrors.h", - "python_include/weakrefobject.h", - "python_include/grammar.h", - "python_include/symtable.h", - "python_include/longobject.h", - "python_include/structmember.h", - "python_include/enumobject.h", - "python_include/classobject.h", - "python_include/unicodeobject.h", - "python_include/sliceobject.h", - "python_include/pystrtod.h", - "python_include/genobject.h", - "python_include/pymactoolbox.h", - "python_include/compile.h", "python_include/pyexpat.h", - "python_include/asdl.h", - "python_include/codecs.h", - "python_include/pyctype.h", - "python_include/sysmodule.h", - "python_include/methodobject.h", - "python_include/graminit.h", - "python_include/cobject.h", - "python_include/intrcheck.h", - "python_include/pyport.h", - "python_include/warnings.h", - "python_include/osdefs.h", - "python_include/fileobject.h", - "python_include/stringobject.h", - "python_include/timefuncs.h", - "python_include/traceback.h", - "python_include/ceval.h", - "python_include/bytes_methods.h", - "python_include/pyconfig.h", - "python_include/Python.h", - "python_include/moduleobject.h", - "python_include/pystate.h", - "python_include/descrobject.h", - "python_include/ucnhash.h", + "python_include/pyfpe.h", "python_include/pygetopt.h", + "python_include/pymacconfig.h", + "python_include/pymactoolbox.h", + "python_include/pymath.h", "python_include/pymem.h", - "python_include/complexobject.h", - "python_include/structseq.h", - "python_include/datetime.h", + "python_include/pyport.h", + "python_include/pystate.h", + "python_include/pystrcmp.h", + "python_include/pystrtod.h", "python_include/pythonrun.h", - "python_include/numpy/oldnumeric.h", - "python_include/numpy/npy_1_7_deprecated_api.h", - "python_include/numpy/ufunc_api.txt", - "python_include/numpy/multiarray_api.txt", - "python_include/numpy/halffloat.h", - "python_include/numpy/npy_common.h", - "python_include/numpy/utils.h", - "python_include/numpy/npy_interrupt.h", - "python_include/numpy/npy_endian.h", - "python_include/numpy/__ufunc_api.h", - "python_include/numpy/_neighborhood_iterator_imp.h", - "python_include/numpy/ufuncobject.h", - "python_include/numpy/ndarraytypes.h", - "python_include/numpy/npy_math.h", - "python_include/numpy/noprefix.h", - "python_include/numpy/npy_3kcompat.h", - "python_include/numpy/arrayscalars.h", - "python_include/numpy/npy_os.h", - "python_include/numpy/ndarrayobject.h", - "python_include/numpy/npy_no_deprecated_api.h", - "python_include/numpy/arrayobject.h", - "python_include/numpy/_numpyconfig.h", - "python_include/numpy/__multiarray_api.h", - "python_include/numpy/npy_cpu.h", - "python_include/numpy/old_defines.h", - "python_include/numpy/numpyconfig.h", - "python_include/pycapsule.h", + "python_include/pythread.h", + "python_include/rangeobject.h", "python_include/setobject.h", - "python_include/listobject.h", - "python_include/bytesobject.h", - "python_include/pgen.h", - "python_include/patchlevel.h", - "python_include/opcode.h", - "python_include/parsetok.h", - "python_include/marshal.h", + "python_include/sliceobject.h", + "python_include/stringobject.h", + "python_include/structmember.h", + "python_include/structseq.h", + "python_include/symtable.h", + "python_include/sysmodule.h", + "python_include/timefuncs.h", "python_include/token.h", - "python_include/iterobject.h", - "python_include/abstract.h", - "python_include/py_curses.h", - "python_include/metagrammar.h", - "python_include/bufferobject.h", - "python_include/Python-ast.h", + "python_include/traceback.h", + "python_include/tupleobject.h", + "python_include/ucnhash.h", + "python_include/unicodeobject.h", + "python_include/warnings.h", + "python_include/weakrefobject.h", ], cmd = """ -cp "/usr/include/python2.7/code.h" "$(@D)/python_include/code.h" && cp "/usr/include/python2.7/dtoa.h" "$(@D)/python_include/dtoa.h" && cp "/usr/include/python2.7/tupleobject.h" "$(@D)/python_include/tupleobject.h" && cp "/usr/include/python2.7/object.h" "$(@D)/python_include/object.h" && cp "/usr/include/python2.7/ast.h" "$(@D)/python_include/ast.h" && cp "/usr/include/python2.7/pymacconfig.h" "$(@D)/python_include/pymacconfig.h" && cp "/usr/include/python2.7/errcode.h" "$(@D)/python_include/errcode.h" && cp "/usr/include/python2.7/frameobject.h" "$(@D)/python_include/frameobject.h" && cp "/usr/include/python2.7/pgenheaders.h" "$(@D)/python_include/pgenheaders.h" && cp "/usr/include/python2.7/cellobject.h" "$(@D)/python_include/cellobject.h" && cp "/usr/include/python2.7/intobject.h" "$(@D)/python_include/intobject.h" && cp "/usr/include/python2.7/pythread.h" "$(@D)/python_include/pythread.h" && cp "/usr/include/python2.7/cStringIO.h" "$(@D)/python_include/cStringIO.h" && cp "/usr/include/python2.7/boolobject.h" "$(@D)/python_include/boolobject.h" && cp "/usr/include/python2.7/modsupport.h" "$(@D)/python_include/modsupport.h" && cp "/usr/include/python2.7/import.h" "$(@D)/python_include/import.h" && cp "/usr/include/python2.7/pymath.h" "$(@D)/python_include/pymath.h" && cp "/usr/include/python2.7/node.h" "$(@D)/python_include/node.h" && cp "/usr/include/python2.7/funcobject.h" "$(@D)/python_include/funcobject.h" && cp "/usr/include/python2.7/eval.h" "$(@D)/python_include/eval.h" && cp "/usr/include/python2.7/longintrepr.h" "$(@D)/python_include/longintrepr.h" && cp "/usr/include/python2.7/floatobject.h" "$(@D)/python_include/floatobject.h" && cp "/usr/include/python2.7/rangeobject.h" "$(@D)/python_include/rangeobject.h" && cp "/usr/include/python2.7/pyfpe.h" "$(@D)/python_include/pyfpe.h" && cp "/usr/include/python2.7/pystrcmp.h" "$(@D)/python_include/pystrcmp.h" && cp "/usr/include/python2.7/dictobject.h" "$(@D)/python_include/dictobject.h" && cp "/usr/include/python2.7/pyarena.h" "$(@D)/python_include/pyarena.h" && cp "/usr/include/python2.7/objimpl.h" "$(@D)/python_include/objimpl.h" && cp "/usr/include/python2.7/bitset.h" "$(@D)/python_include/bitset.h" && cp "/usr/include/python2.7/memoryobject.h" "$(@D)/python_include/memoryobject.h" && cp "/usr/include/python2.7/bytearrayobject.h" "$(@D)/python_include/bytearrayobject.h" && cp "/usr/include/python2.7/pydebug.h" "$(@D)/python_include/pydebug.h" && cp "/usr/include/python2.7/pyerrors.h" "$(@D)/python_include/pyerrors.h" && cp "/usr/include/python2.7/weakrefobject.h" "$(@D)/python_include/weakrefobject.h" && cp "/usr/include/python2.7/grammar.h" "$(@D)/python_include/grammar.h" && cp "/usr/include/python2.7/symtable.h" "$(@D)/python_include/symtable.h" && cp "/usr/include/python2.7/longobject.h" "$(@D)/python_include/longobject.h" && cp "/usr/include/python2.7/structmember.h" "$(@D)/python_include/structmember.h" && cp "/usr/include/python2.7/enumobject.h" "$(@D)/python_include/enumobject.h" && cp "/usr/include/python2.7/classobject.h" "$(@D)/python_include/classobject.h" && cp "/usr/include/python2.7/unicodeobject.h" "$(@D)/python_include/unicodeobject.h" && cp "/usr/include/python2.7/sliceobject.h" "$(@D)/python_include/sliceobject.h" && cp "/usr/include/python2.7/pystrtod.h" "$(@D)/python_include/pystrtod.h" && cp "/usr/include/python2.7/genobject.h" "$(@D)/python_include/genobject.h" && cp "/usr/include/python2.7/pymactoolbox.h" "$(@D)/python_include/pymactoolbox.h" && cp "/usr/include/python2.7/compile.h" "$(@D)/python_include/compile.h" && cp "/usr/include/python2.7/pyexpat.h" "$(@D)/python_include/pyexpat.h" && cp "/usr/include/python2.7/asdl.h" "$(@D)/python_include/asdl.h" && cp "/usr/include/python2.7/codecs.h" "$(@D)/python_include/codecs.h" && cp "/usr/include/python2.7/pyctype.h" "$(@D)/python_include/pyctype.h" && cp "/usr/include/python2.7/sysmodule.h" "$(@D)/python_include/sysmodule.h" && cp "/usr/include/python2.7/methodobject.h" "$(@D)/python_include/methodobject.h" && cp "/usr/include/python2.7/graminit.h" "$(@D)/python_include/graminit.h" && cp "/usr/include/python2.7/cobject.h" "$(@D)/python_include/cobject.h" && cp "/usr/include/python2.7/intrcheck.h" "$(@D)/python_include/intrcheck.h" && cp "/usr/include/python2.7/pyport.h" "$(@D)/python_include/pyport.h" && cp "/usr/include/python2.7/warnings.h" "$(@D)/python_include/warnings.h" && cp "/usr/include/python2.7/osdefs.h" "$(@D)/python_include/osdefs.h" && cp "/usr/include/python2.7/fileobject.h" "$(@D)/python_include/fileobject.h" && cp "/usr/include/python2.7/stringobject.h" "$(@D)/python_include/stringobject.h" && cp "/usr/include/python2.7/timefuncs.h" "$(@D)/python_include/timefuncs.h" && cp "/usr/include/python2.7/traceback.h" "$(@D)/python_include/traceback.h" && cp "/usr/include/python2.7/ceval.h" "$(@D)/python_include/ceval.h" && cp "/usr/include/python2.7/bytes_methods.h" "$(@D)/python_include/bytes_methods.h" && cp "/usr/include/python2.7/pyconfig.h" "$(@D)/python_include/pyconfig.h" && cp "/usr/include/python2.7/Python.h" "$(@D)/python_include/Python.h" && cp "/usr/include/python2.7/moduleobject.h" "$(@D)/python_include/moduleobject.h" && cp "/usr/include/python2.7/pystate.h" "$(@D)/python_include/pystate.h" && cp "/usr/include/python2.7/descrobject.h" "$(@D)/python_include/descrobject.h" && cp "/usr/include/python2.7/ucnhash.h" "$(@D)/python_include/ucnhash.h" && cp "/usr/include/python2.7/pygetopt.h" "$(@D)/python_include/pygetopt.h" && cp "/usr/include/python2.7/pymem.h" "$(@D)/python_include/pymem.h" && cp "/usr/include/python2.7/complexobject.h" "$(@D)/python_include/complexobject.h" && cp "/usr/include/python2.7/structseq.h" "$(@D)/python_include/structseq.h" && cp "/usr/include/python2.7/datetime.h" "$(@D)/python_include/datetime.h" && cp "/usr/include/python2.7/pythonrun.h" "$(@D)/python_include/pythonrun.h" && cp "/usr/include/python2.7/numpy/oldnumeric.h" "$(@D)/python_include/numpy/oldnumeric.h" && cp "/usr/include/python2.7/numpy/npy_1_7_deprecated_api.h" "$(@D)/python_include/numpy/npy_1_7_deprecated_api.h" && cp "/usr/include/python2.7/numpy/ufunc_api.txt" "$(@D)/python_include/numpy/ufunc_api.txt" && cp "/usr/include/python2.7/numpy/multiarray_api.txt" "$(@D)/python_include/numpy/multiarray_api.txt" && cp "/usr/include/python2.7/numpy/halffloat.h" "$(@D)/python_include/numpy/halffloat.h" && cp "/usr/include/python2.7/numpy/npy_common.h" "$(@D)/python_include/numpy/npy_common.h" && cp "/usr/include/python2.7/numpy/utils.h" "$(@D)/python_include/numpy/utils.h" && cp "/usr/include/python2.7/numpy/npy_interrupt.h" "$(@D)/python_include/numpy/npy_interrupt.h" && cp "/usr/include/python2.7/numpy/npy_endian.h" "$(@D)/python_include/numpy/npy_endian.h" && cp "/usr/include/python2.7/numpy/__ufunc_api.h" "$(@D)/python_include/numpy/__ufunc_api.h" && cp "/usr/include/python2.7/numpy/_neighborhood_iterator_imp.h" "$(@D)/python_include/numpy/_neighborhood_iterator_imp.h" && cp "/usr/include/python2.7/numpy/ufuncobject.h" "$(@D)/python_include/numpy/ufuncobject.h" && cp "/usr/include/python2.7/numpy/ndarraytypes.h" "$(@D)/python_include/numpy/ndarraytypes.h" && cp "/usr/include/python2.7/numpy/npy_math.h" "$(@D)/python_include/numpy/npy_math.h" && cp "/usr/include/python2.7/numpy/noprefix.h" "$(@D)/python_include/numpy/noprefix.h" && cp "/usr/include/python2.7/numpy/npy_3kcompat.h" "$(@D)/python_include/numpy/npy_3kcompat.h" && cp "/usr/include/python2.7/numpy/arrayscalars.h" "$(@D)/python_include/numpy/arrayscalars.h" && cp "/usr/include/python2.7/numpy/npy_os.h" "$(@D)/python_include/numpy/npy_os.h" && cp "/usr/include/python2.7/numpy/ndarrayobject.h" "$(@D)/python_include/numpy/ndarrayobject.h" && cp "/usr/include/python2.7/numpy/npy_no_deprecated_api.h" "$(@D)/python_include/numpy/npy_no_deprecated_api.h" && cp "/usr/include/python2.7/numpy/arrayobject.h" "$(@D)/python_include/numpy/arrayobject.h" && cp "/usr/include/python2.7/numpy/_numpyconfig.h" "$(@D)/python_include/numpy/_numpyconfig.h" && cp "/usr/include/python2.7/numpy/__multiarray_api.h" "$(@D)/python_include/numpy/__multiarray_api.h" && cp "/usr/include/python2.7/numpy/npy_cpu.h" "$(@D)/python_include/numpy/npy_cpu.h" && cp "/usr/include/python2.7/numpy/old_defines.h" "$(@D)/python_include/numpy/old_defines.h" && cp "/usr/include/python2.7/numpy/numpyconfig.h" "$(@D)/python_include/numpy/numpyconfig.h" && cp "/usr/include/python2.7/pycapsule.h" "$(@D)/python_include/pycapsule.h" && cp "/usr/include/python2.7/setobject.h" "$(@D)/python_include/setobject.h" && cp "/usr/include/python2.7/listobject.h" "$(@D)/python_include/listobject.h" && cp "/usr/include/python2.7/bytesobject.h" "$(@D)/python_include/bytesobject.h" && cp "/usr/include/python2.7/pgen.h" "$(@D)/python_include/pgen.h" && cp "/usr/include/python2.7/patchlevel.h" "$(@D)/python_include/patchlevel.h" && cp "/usr/include/python2.7/opcode.h" "$(@D)/python_include/opcode.h" && cp "/usr/include/python2.7/parsetok.h" "$(@D)/python_include/parsetok.h" && cp "/usr/include/python2.7/marshal.h" "$(@D)/python_include/marshal.h" && cp "/usr/include/python2.7/token.h" "$(@D)/python_include/token.h" && cp "/usr/include/python2.7/iterobject.h" "$(@D)/python_include/iterobject.h" && cp "/usr/include/python2.7/abstract.h" "$(@D)/python_include/abstract.h" && cp "/usr/include/python2.7/py_curses.h" "$(@D)/python_include/py_curses.h" && cp "/usr/include/python2.7/metagrammar.h" "$(@D)/python_include/metagrammar.h" && cp "/usr/include/python2.7/bufferobject.h" "$(@D)/python_include/bufferobject.h" && cp "/usr/include/python2.7/Python-ast.h" "$(@D)/python_include/Python-ast.h" +cp "/usr/include/python2.7/Python-ast.h" "$(@D)/python_include/Python-ast.h" && cp "/usr/include/python2.7/Python.h" "$(@D)/python_include/Python.h" && cp "/usr/include/python2.7/abstract.h" "$(@D)/python_include/abstract.h" && cp "/usr/include/python2.7/asdl.h" "$(@D)/python_include/asdl.h" && cp "/usr/include/python2.7/ast.h" "$(@D)/python_include/ast.h" && cp "/usr/include/python2.7/bitset.h" "$(@D)/python_include/bitset.h" && cp "/usr/include/python2.7/boolobject.h" "$(@D)/python_include/boolobject.h" && cp "/usr/include/python2.7/bufferobject.h" "$(@D)/python_include/bufferobject.h" && cp "/usr/include/python2.7/bytearrayobject.h" "$(@D)/python_include/bytearrayobject.h" && cp "/usr/include/python2.7/bytes_methods.h" "$(@D)/python_include/bytes_methods.h" && cp "/usr/include/python2.7/bytesobject.h" "$(@D)/python_include/bytesobject.h" && cp "/usr/include/python2.7/cStringIO.h" "$(@D)/python_include/cStringIO.h" && cp "/usr/include/python2.7/cellobject.h" "$(@D)/python_include/cellobject.h" && cp "/usr/include/python2.7/ceval.h" "$(@D)/python_include/ceval.h" && cp "/usr/include/python2.7/classobject.h" "$(@D)/python_include/classobject.h" && cp "/usr/include/python2.7/cobject.h" "$(@D)/python_include/cobject.h" && cp "/usr/include/python2.7/code.h" "$(@D)/python_include/code.h" && cp "/usr/include/python2.7/codecs.h" "$(@D)/python_include/codecs.h" && cp "/usr/include/python2.7/compile.h" "$(@D)/python_include/compile.h" && cp "/usr/include/python2.7/complexobject.h" "$(@D)/python_include/complexobject.h" && cp "/usr/include/python2.7/datetime.h" "$(@D)/python_include/datetime.h" && cp "/usr/include/python2.7/descrobject.h" "$(@D)/python_include/descrobject.h" && cp "/usr/include/python2.7/dictobject.h" "$(@D)/python_include/dictobject.h" && cp "/usr/include/python2.7/dtoa.h" "$(@D)/python_include/dtoa.h" && cp "/usr/include/python2.7/enumobject.h" "$(@D)/python_include/enumobject.h" && cp "/usr/include/python2.7/errcode.h" "$(@D)/python_include/errcode.h" && cp "/usr/include/python2.7/eval.h" "$(@D)/python_include/eval.h" && cp "/usr/include/python2.7/fileobject.h" "$(@D)/python_include/fileobject.h" && cp "/usr/include/python2.7/floatobject.h" "$(@D)/python_include/floatobject.h" && cp "/usr/include/python2.7/frameobject.h" "$(@D)/python_include/frameobject.h" && cp "/usr/include/python2.7/funcobject.h" "$(@D)/python_include/funcobject.h" && cp "/usr/include/python2.7/genobject.h" "$(@D)/python_include/genobject.h" && cp "/usr/include/python2.7/graminit.h" "$(@D)/python_include/graminit.h" && cp "/usr/include/python2.7/grammar.h" "$(@D)/python_include/grammar.h" && cp "/usr/include/python2.7/import.h" "$(@D)/python_include/import.h" && cp "/usr/include/python2.7/intobject.h" "$(@D)/python_include/intobject.h" && cp "/usr/include/python2.7/intrcheck.h" "$(@D)/python_include/intrcheck.h" && cp "/usr/include/python2.7/iterobject.h" "$(@D)/python_include/iterobject.h" && cp "/usr/include/python2.7/listobject.h" "$(@D)/python_include/listobject.h" && cp "/usr/include/python2.7/longintrepr.h" "$(@D)/python_include/longintrepr.h" && cp "/usr/include/python2.7/longobject.h" "$(@D)/python_include/longobject.h" && cp "/usr/include/python2.7/marshal.h" "$(@D)/python_include/marshal.h" && cp "/usr/include/python2.7/memoryobject.h" "$(@D)/python_include/memoryobject.h" && cp "/usr/include/python2.7/metagrammar.h" "$(@D)/python_include/metagrammar.h" && cp "/usr/include/python2.7/methodobject.h" "$(@D)/python_include/methodobject.h" && cp "/usr/include/python2.7/modsupport.h" "$(@D)/python_include/modsupport.h" && cp "/usr/include/python2.7/moduleobject.h" "$(@D)/python_include/moduleobject.h" && cp "/usr/include/python2.7/node.h" "$(@D)/python_include/node.h" && cp "/usr/include/python2.7/object.h" "$(@D)/python_include/object.h" && cp "/usr/include/python2.7/objimpl.h" "$(@D)/python_include/objimpl.h" && cp "/usr/include/python2.7/opcode.h" "$(@D)/python_include/opcode.h" && cp "/usr/include/python2.7/osdefs.h" "$(@D)/python_include/osdefs.h" && cp "/usr/include/python2.7/parsetok.h" "$(@D)/python_include/parsetok.h" && cp "/usr/include/python2.7/patchlevel.h" "$(@D)/python_include/patchlevel.h" && cp "/usr/include/python2.7/pgen.h" "$(@D)/python_include/pgen.h" && cp "/usr/include/python2.7/pgenheaders.h" "$(@D)/python_include/pgenheaders.h" && cp "/usr/include/python2.7/py_curses.h" "$(@D)/python_include/py_curses.h" && cp "/usr/include/python2.7/pyarena.h" "$(@D)/python_include/pyarena.h" && cp "/usr/include/python2.7/pycapsule.h" "$(@D)/python_include/pycapsule.h" && cp "/usr/include/python2.7/pyconfig.h" "$(@D)/python_include/pyconfig.h" && cp "/usr/include/python2.7/pyctype.h" "$(@D)/python_include/pyctype.h" && cp "/usr/include/python2.7/pydebug.h" "$(@D)/python_include/pydebug.h" && cp "/usr/include/python2.7/pyerrors.h" "$(@D)/python_include/pyerrors.h" && cp "/usr/include/python2.7/pyexpat.h" "$(@D)/python_include/pyexpat.h" && cp "/usr/include/python2.7/pyfpe.h" "$(@D)/python_include/pyfpe.h" && cp "/usr/include/python2.7/pygetopt.h" "$(@D)/python_include/pygetopt.h" && cp "/usr/include/python2.7/pymacconfig.h" "$(@D)/python_include/pymacconfig.h" && cp "/usr/include/python2.7/pymactoolbox.h" "$(@D)/python_include/pymactoolbox.h" && cp "/usr/include/python2.7/pymath.h" "$(@D)/python_include/pymath.h" && cp "/usr/include/python2.7/pymem.h" "$(@D)/python_include/pymem.h" && cp "/usr/include/python2.7/pyport.h" "$(@D)/python_include/pyport.h" && cp "/usr/include/python2.7/pystate.h" "$(@D)/python_include/pystate.h" && cp "/usr/include/python2.7/pystrcmp.h" "$(@D)/python_include/pystrcmp.h" && cp "/usr/include/python2.7/pystrtod.h" "$(@D)/python_include/pystrtod.h" && cp "/usr/include/python2.7/pythonrun.h" "$(@D)/python_include/pythonrun.h" && cp "/usr/include/python2.7/pythread.h" "$(@D)/python_include/pythread.h" && cp "/usr/include/python2.7/rangeobject.h" "$(@D)/python_include/rangeobject.h" && cp "/usr/include/python2.7/setobject.h" "$(@D)/python_include/setobject.h" && cp "/usr/include/python2.7/sliceobject.h" "$(@D)/python_include/sliceobject.h" && cp "/usr/include/python2.7/stringobject.h" "$(@D)/python_include/stringobject.h" && cp "/usr/include/python2.7/structmember.h" "$(@D)/python_include/structmember.h" && cp "/usr/include/python2.7/structseq.h" "$(@D)/python_include/structseq.h" && cp "/usr/include/python2.7/symtable.h" "$(@D)/python_include/symtable.h" && cp "/usr/include/python2.7/sysmodule.h" "$(@D)/python_include/sysmodule.h" && cp "/usr/include/python2.7/timefuncs.h" "$(@D)/python_include/timefuncs.h" && cp "/usr/include/python2.7/token.h" "$(@D)/python_include/token.h" && cp "/usr/include/python2.7/traceback.h" "$(@D)/python_include/traceback.h" && cp "/usr/include/python2.7/tupleobject.h" "$(@D)/python_include/tupleobject.h" && cp "/usr/include/python2.7/ucnhash.h" "$(@D)/python_include/ucnhash.h" && cp "/usr/include/python2.7/unicodeobject.h" "$(@D)/python_include/unicodeobject.h" && cp "/usr/include/python2.7/warnings.h" "$(@D)/python_include/warnings.h" && cp "/usr/include/python2.7/weakrefobject.h" "$(@D)/python_include/weakrefobject.h" """, ) genrule( name = "numpy_include", outs = [ - "numpy_include/numpy/oldnumeric.h", - "numpy_include/numpy/npy_1_7_deprecated_api.h", - "numpy_include/numpy/ufunc_api.txt", - "numpy_include/numpy/multiarray_api.txt", - "numpy_include/numpy/halffloat.h", - "numpy_include/numpy/npy_common.h", - "numpy_include/numpy/utils.h", - "numpy_include/numpy/npy_interrupt.h", - "numpy_include/numpy/npy_endian.h", + "numpy_include/numpy/__multiarray_api.h", "numpy_include/numpy/__ufunc_api.h", "numpy_include/numpy/_neighborhood_iterator_imp.h", - "numpy_include/numpy/ufuncobject.h", + "numpy_include/numpy/_numpyconfig.h", + "numpy_include/numpy/arrayobject.h", + "numpy_include/numpy/arrayscalars.h", + "numpy_include/numpy/halffloat.h", + "numpy_include/numpy/multiarray_api.txt", + "numpy_include/numpy/ndarrayobject.h", "numpy_include/numpy/ndarraytypes.h", - "numpy_include/numpy/npy_math.h", "numpy_include/numpy/noprefix.h", + "numpy_include/numpy/npy_1_7_deprecated_api.h", "numpy_include/numpy/npy_3kcompat.h", - "numpy_include/numpy/arrayscalars.h", - "numpy_include/numpy/npy_os.h", - "numpy_include/numpy/ndarrayobject.h", - "numpy_include/numpy/npy_no_deprecated_api.h", - "numpy_include/numpy/arrayobject.h", - "numpy_include/numpy/_numpyconfig.h", - "numpy_include/numpy/__multiarray_api.h", + "numpy_include/numpy/npy_common.h", "numpy_include/numpy/npy_cpu.h", - "numpy_include/numpy/old_defines.h", + "numpy_include/numpy/npy_endian.h", + "numpy_include/numpy/npy_interrupt.h", + "numpy_include/numpy/npy_math.h", + "numpy_include/numpy/npy_no_deprecated_api.h", + "numpy_include/numpy/npy_os.h", "numpy_include/numpy/numpyconfig.h", + "numpy_include/numpy/old_defines.h", + "numpy_include/numpy/oldnumeric.h", + "numpy_include/numpy/ufunc_api.txt", + "numpy_include/numpy/ufuncobject.h", + "numpy_include/numpy/utils.h", ], cmd = """ -cp "/usr/lib/python2.7/dist-packages/numpy/core/include/numpy/oldnumeric.h" "$(@D)/numpy_include/numpy/oldnumeric.h" && cp "/usr/lib/python2.7/dist-packages/numpy/core/include/numpy/npy_1_7_deprecated_api.h" "$(@D)/numpy_include/numpy/npy_1_7_deprecated_api.h" && cp "/usr/lib/python2.7/dist-packages/numpy/core/include/numpy/ufunc_api.txt" "$(@D)/numpy_include/numpy/ufunc_api.txt" && cp "/usr/lib/python2.7/dist-packages/numpy/core/include/numpy/multiarray_api.txt" "$(@D)/numpy_include/numpy/multiarray_api.txt" && cp "/usr/lib/python2.7/dist-packages/numpy/core/include/numpy/halffloat.h" "$(@D)/numpy_include/numpy/halffloat.h" && cp "/usr/lib/python2.7/dist-packages/numpy/core/include/numpy/npy_common.h" "$(@D)/numpy_include/numpy/npy_common.h" && cp "/usr/lib/python2.7/dist-packages/numpy/core/include/numpy/utils.h" "$(@D)/numpy_include/numpy/utils.h" && cp "/usr/lib/python2.7/dist-packages/numpy/core/include/numpy/npy_interrupt.h" "$(@D)/numpy_include/numpy/npy_interrupt.h" && cp "/usr/lib/python2.7/dist-packages/numpy/core/include/numpy/npy_endian.h" "$(@D)/numpy_include/numpy/npy_endian.h" && cp "/usr/lib/python2.7/dist-packages/numpy/core/include/numpy/__ufunc_api.h" "$(@D)/numpy_include/numpy/__ufunc_api.h" && cp "/usr/lib/python2.7/dist-packages/numpy/core/include/numpy/_neighborhood_iterator_imp.h" "$(@D)/numpy_include/numpy/_neighborhood_iterator_imp.h" && cp "/usr/lib/python2.7/dist-packages/numpy/core/include/numpy/ufuncobject.h" "$(@D)/numpy_include/numpy/ufuncobject.h" && cp "/usr/lib/python2.7/dist-packages/numpy/core/include/numpy/ndarraytypes.h" "$(@D)/numpy_include/numpy/ndarraytypes.h" && cp "/usr/lib/python2.7/dist-packages/numpy/core/include/numpy/npy_math.h" "$(@D)/numpy_include/numpy/npy_math.h" && cp "/usr/lib/python2.7/dist-packages/numpy/core/include/numpy/noprefix.h" "$(@D)/numpy_include/numpy/noprefix.h" && cp "/usr/lib/python2.7/dist-packages/numpy/core/include/numpy/npy_3kcompat.h" "$(@D)/numpy_include/numpy/npy_3kcompat.h" && cp "/usr/lib/python2.7/dist-packages/numpy/core/include/numpy/arrayscalars.h" "$(@D)/numpy_include/numpy/arrayscalars.h" && cp "/usr/lib/python2.7/dist-packages/numpy/core/include/numpy/npy_os.h" "$(@D)/numpy_include/numpy/npy_os.h" && cp "/usr/lib/python2.7/dist-packages/numpy/core/include/numpy/ndarrayobject.h" "$(@D)/numpy_include/numpy/ndarrayobject.h" && cp "/usr/lib/python2.7/dist-packages/numpy/core/include/numpy/npy_no_deprecated_api.h" "$(@D)/numpy_include/numpy/npy_no_deprecated_api.h" && cp "/usr/lib/python2.7/dist-packages/numpy/core/include/numpy/arrayobject.h" "$(@D)/numpy_include/numpy/arrayobject.h" && cp "/usr/lib/python2.7/dist-packages/numpy/core/include/numpy/_numpyconfig.h" "$(@D)/numpy_include/numpy/_numpyconfig.h" && cp "/usr/lib/python2.7/dist-packages/numpy/core/include/numpy/__multiarray_api.h" "$(@D)/numpy_include/numpy/__multiarray_api.h" && cp "/usr/lib/python2.7/dist-packages/numpy/core/include/numpy/npy_cpu.h" "$(@D)/numpy_include/numpy/npy_cpu.h" && cp "/usr/lib/python2.7/dist-packages/numpy/core/include/numpy/old_defines.h" "$(@D)/numpy_include/numpy/old_defines.h" && cp "/usr/lib/python2.7/dist-packages/numpy/core/include/numpy/numpyconfig.h" "$(@D)/numpy_include/numpy/numpyconfig.h" +cp "/usr/local/lib/python2.7/dist-packages/numpy/core/include/numpy/__multiarray_api.h" "$(@D)/numpy_include/numpy/__multiarray_api.h" && cp "/usr/local/lib/python2.7/dist-packages/numpy/core/include/numpy/__ufunc_api.h" "$(@D)/numpy_include/numpy/__ufunc_api.h" && cp "/usr/local/lib/python2.7/dist-packages/numpy/core/include/numpy/_neighborhood_iterator_imp.h" "$(@D)/numpy_include/numpy/_neighborhood_iterator_imp.h" && cp "/usr/local/lib/python2.7/dist-packages/numpy/core/include/numpy/_numpyconfig.h" "$(@D)/numpy_include/numpy/_numpyconfig.h" && cp "/usr/local/lib/python2.7/dist-packages/numpy/core/include/numpy/arrayobject.h" "$(@D)/numpy_include/numpy/arrayobject.h" && cp "/usr/local/lib/python2.7/dist-packages/numpy/core/include/numpy/arrayscalars.h" "$(@D)/numpy_include/numpy/arrayscalars.h" && cp "/usr/local/lib/python2.7/dist-packages/numpy/core/include/numpy/halffloat.h" "$(@D)/numpy_include/numpy/halffloat.h" && cp "/usr/local/lib/python2.7/dist-packages/numpy/core/include/numpy/multiarray_api.txt" "$(@D)/numpy_include/numpy/multiarray_api.txt" && cp "/usr/local/lib/python2.7/dist-packages/numpy/core/include/numpy/ndarrayobject.h" "$(@D)/numpy_include/numpy/ndarrayobject.h" && cp "/usr/local/lib/python2.7/dist-packages/numpy/core/include/numpy/ndarraytypes.h" "$(@D)/numpy_include/numpy/ndarraytypes.h" && cp "/usr/local/lib/python2.7/dist-packages/numpy/core/include/numpy/noprefix.h" "$(@D)/numpy_include/numpy/noprefix.h" && cp "/usr/local/lib/python2.7/dist-packages/numpy/core/include/numpy/npy_1_7_deprecated_api.h" "$(@D)/numpy_include/numpy/npy_1_7_deprecated_api.h" && cp "/usr/local/lib/python2.7/dist-packages/numpy/core/include/numpy/npy_3kcompat.h" "$(@D)/numpy_include/numpy/npy_3kcompat.h" && cp "/usr/local/lib/python2.7/dist-packages/numpy/core/include/numpy/npy_common.h" "$(@D)/numpy_include/numpy/npy_common.h" && cp "/usr/local/lib/python2.7/dist-packages/numpy/core/include/numpy/npy_cpu.h" "$(@D)/numpy_include/numpy/npy_cpu.h" && cp "/usr/local/lib/python2.7/dist-packages/numpy/core/include/numpy/npy_endian.h" "$(@D)/numpy_include/numpy/npy_endian.h" && cp "/usr/local/lib/python2.7/dist-packages/numpy/core/include/numpy/npy_interrupt.h" "$(@D)/numpy_include/numpy/npy_interrupt.h" && cp "/usr/local/lib/python2.7/dist-packages/numpy/core/include/numpy/npy_math.h" "$(@D)/numpy_include/numpy/npy_math.h" && cp "/usr/local/lib/python2.7/dist-packages/numpy/core/include/numpy/npy_no_deprecated_api.h" "$(@D)/numpy_include/numpy/npy_no_deprecated_api.h" && cp "/usr/local/lib/python2.7/dist-packages/numpy/core/include/numpy/npy_os.h" "$(@D)/numpy_include/numpy/npy_os.h" && cp "/usr/local/lib/python2.7/dist-packages/numpy/core/include/numpy/numpyconfig.h" "$(@D)/numpy_include/numpy/numpyconfig.h" && cp "/usr/local/lib/python2.7/dist-packages/numpy/core/include/numpy/old_defines.h" "$(@D)/numpy_include/numpy/old_defines.h" && cp "/usr/local/lib/python2.7/dist-packages/numpy/core/include/numpy/oldnumeric.h" "$(@D)/numpy_include/numpy/oldnumeric.h" && cp "/usr/local/lib/python2.7/dist-packages/numpy/core/include/numpy/ufunc_api.txt" "$(@D)/numpy_include/numpy/ufunc_api.txt" && cp "/usr/local/lib/python2.7/dist-packages/numpy/core/include/numpy/ufuncobject.h" "$(@D)/numpy_include/numpy/ufuncobject.h" && cp "/usr/local/lib/python2.7/dist-packages/numpy/core/include/numpy/utils.h" "$(@D)/numpy_include/numpy/utils.h" """, ) diff --git a/third_party/toolchains/cpus/py3/BUILD b/third_party/toolchains/cpus/py3/BUILD index 932a25239f..d47256ebef 100644 --- a/third_party/toolchains/cpus/py3/BUILD +++ b/third_party/toolchains/cpus/py3/BUILD @@ -6,18 +6,24 @@ licenses(["restricted"]) package(default_visibility = ["//visibility:public"]) +# To build Python C/C++ extension on Windows, we need to link to python import library pythonXY.lib +# See https://docs.python.org/3/extending/windows.html +cc_import( + name = "python_lib", + interface_library = select({ + ":windows": ":python_import_lib", + # A placeholder for Unix platforms which makes --no_build happy. + "//conditions:default": "not-existing.lib", + }), + system_provided = 1, +) + cc_library( name = "python_headers", hdrs = [":python_include"], - data = select({ - ":windows": [":python_import_lib"], - "//conditions:default": [], - }), includes = ["python_include"], - linkopts = select({ - # TODO(pcloudy): Ideally, this should just go into deps after resolving - # https://github.com/bazelbuild/bazel/issues/3237, - ":windows": ["$(locations :python_import_lib)"], + deps = select({ + ":windows": [":python_lib"], "//conditions:default": [], }), ) @@ -37,143 +43,143 @@ config_setting( genrule( name = "python_include", outs = [ - "python_include/code.h", - "python_include/dtoa.h", - "python_include/tupleobject.h", - "python_include/object.h", - "python_include/ast.h", - "python_include/pymacconfig.h", - "python_include/errcode.h", - "python_include/frameobject.h", - "python_include/typeslots.h", - "python_include/pgenheaders.h", - "python_include/cellobject.h", - "python_include/pythread.h", - "python_include/boolobject.h", + "python_include/Python-ast.h", + "python_include/Python.h", + "python_include/abstract.h", "python_include/accu.h", - "python_include/modsupport.h", - "python_include/import.h", - "python_include/pymath.h", - "python_include/node.h", - "python_include/funcobject.h", - "python_include/eval.h", - "python_include/pyatomic.h", - "python_include/longintrepr.h", - "python_include/floatobject.h", - "python_include/rangeobject.h", - "python_include/pyfpe.h", - "python_include/pystrcmp.h", - "python_include/fileutils.h", - "python_include/dictobject.h", - "python_include/pyarena.h", - "python_include/osmodule.h", - "python_include/objimpl.h", + "python_include/asdl.h", + "python_include/ast.h", "python_include/bitset.h", - "python_include/memoryobject.h", + "python_include/bltinmodule.h", + "python_include/boolobject.h", "python_include/bytearrayobject.h", - "python_include/pydebug.h", - "python_include/pyerrors.h", - "python_include/weakrefobject.h", - "python_include/grammar.h", - "python_include/symtable.h", - "python_include/longobject.h", - "python_include/structmember.h", - "python_include/enumobject.h", - "python_include/pymacro.h", + "python_include/bytes_methods.h", + "python_include/bytesobject.h", + "python_include/cellobject.h", + "python_include/ceval.h", "python_include/classobject.h", - "python_include/unicodeobject.h", - "python_include/sliceobject.h", - "python_include/pystrtod.h", - "python_include/genobject.h", - "python_include/compile.h", - "python_include/pyexpat.h", - "python_include/asdl.h", + "python_include/code.h", "python_include/codecs.h", + "python_include/compile.h", + "python_include/complexobject.h", + "python_include/datetime.h", + "python_include/descrobject.h", + "python_include/dictobject.h", + "python_include/dtoa.h", "python_include/dynamic_annotations.h", - "python_include/pyctype.h", - "python_include/sysmodule.h", - "python_include/methodobject.h", + "python_include/enumobject.h", + "python_include/errcode.h", + "python_include/eval.h", + "python_include/fileobject.h", + "python_include/fileutils.h", + "python_include/floatobject.h", + "python_include/frameobject.h", + "python_include/funcobject.h", + "python_include/genobject.h", "python_include/graminit.h", - "python_include/bltinmodule.h", + "python_include/grammar.h", + "python_include/import.h", "python_include/intrcheck.h", - "python_include/pyport.h", - "python_include/warnings.h", - "python_include/osdefs.h", - "python_include/pydtrace.h", - "python_include/pylifecycle.h", - "python_include/fileobject.h", - "python_include/pytime.h", - "python_include/traceback.h", - "python_include/ceval.h", - "python_include/bytes_methods.h", - "python_include/namespaceobject.h", - "python_include/pyconfig.h", - "python_include/Python.h", + "python_include/iterobject.h", + "python_include/listobject.h", + "python_include/longintrepr.h", + "python_include/longobject.h", + "python_include/marshal.h", + "python_include/memoryobject.h", + "python_include/metagrammar.h", + "python_include/methodobject.h", + "python_include/modsupport.h", "python_include/moduleobject.h", - "python_include/pystate.h", - "python_include/descrobject.h", + "python_include/namespaceobject.h", + "python_include/node.h", + "python_include/object.h", + "python_include/objimpl.h", "python_include/odictobject.h", - "python_include/ucnhash.h", + "python_include/opcode.h", + "python_include/osdefs.h", + "python_include/osmodule.h", + "python_include/parsetok.h", + "python_include/patchlevel.h", + "python_include/pgen.h", + "python_include/pgenheaders.h", + "python_include/py_curses.h", + "python_include/pyarena.h", + "python_include/pyatomic.h", + "python_include/pycapsule.h", + "python_include/pyconfig.h", + "python_include/pyctype.h", + "python_include/pydebug.h", + "python_include/pydtrace.h", + "python_include/pyerrors.h", + "python_include/pyexpat.h", + "python_include/pyfpe.h", "python_include/pygetopt.h", + "python_include/pyhash.h", + "python_include/pylifecycle.h", + "python_include/pymacconfig.h", + "python_include/pymacro.h", + "python_include/pymath.h", "python_include/pymem.h", - "python_include/complexobject.h", - "python_include/structseq.h", - "python_include/datetime.h", + "python_include/pyport.h", + "python_include/pystate.h", + "python_include/pystrcmp.h", + "python_include/pystrhex.h", + "python_include/pystrtod.h", "python_include/pythonrun.h", - "python_include/pyhash.h", - "python_include/pycapsule.h", + "python_include/pythread.h", + "python_include/pytime.h", + "python_include/rangeobject.h", "python_include/setobject.h", - "python_include/listobject.h", - "python_include/bytesobject.h", - "python_include/pgen.h", - "python_include/patchlevel.h", - "python_include/opcode.h", - "python_include/parsetok.h", - "python_include/pystrhex.h", - "python_include/marshal.h", + "python_include/sliceobject.h", + "python_include/structmember.h", + "python_include/structseq.h", + "python_include/symtable.h", + "python_include/sysmodule.h", "python_include/token.h", - "python_include/iterobject.h", - "python_include/abstract.h", - "python_include/py_curses.h", - "python_include/metagrammar.h", - "python_include/Python-ast.h", + "python_include/traceback.h", + "python_include/tupleobject.h", + "python_include/typeslots.h", + "python_include/ucnhash.h", + "python_include/unicodeobject.h", + "python_include/warnings.h", + "python_include/weakrefobject.h", ], cmd = """ -cp "/opt/python3.6/include/python3.6m/code.h" "$(@D)/python_include/code.h" && cp "/opt/python3.6/include/python3.6m/dtoa.h" "$(@D)/python_include/dtoa.h" && cp "/opt/python3.6/include/python3.6m/tupleobject.h" "$(@D)/python_include/tupleobject.h" && cp "/opt/python3.6/include/python3.6m/object.h" "$(@D)/python_include/object.h" && cp "/opt/python3.6/include/python3.6m/ast.h" "$(@D)/python_include/ast.h" && cp "/opt/python3.6/include/python3.6m/pymacconfig.h" "$(@D)/python_include/pymacconfig.h" && cp "/opt/python3.6/include/python3.6m/errcode.h" "$(@D)/python_include/errcode.h" && cp "/opt/python3.6/include/python3.6m/frameobject.h" "$(@D)/python_include/frameobject.h" && cp "/opt/python3.6/include/python3.6m/typeslots.h" "$(@D)/python_include/typeslots.h" && cp "/opt/python3.6/include/python3.6m/pgenheaders.h" "$(@D)/python_include/pgenheaders.h" && cp "/opt/python3.6/include/python3.6m/cellobject.h" "$(@D)/python_include/cellobject.h" && cp "/opt/python3.6/include/python3.6m/pythread.h" "$(@D)/python_include/pythread.h" && cp "/opt/python3.6/include/python3.6m/boolobject.h" "$(@D)/python_include/boolobject.h" && cp "/opt/python3.6/include/python3.6m/accu.h" "$(@D)/python_include/accu.h" && cp "/opt/python3.6/include/python3.6m/modsupport.h" "$(@D)/python_include/modsupport.h" && cp "/opt/python3.6/include/python3.6m/import.h" "$(@D)/python_include/import.h" && cp "/opt/python3.6/include/python3.6m/pymath.h" "$(@D)/python_include/pymath.h" && cp "/opt/python3.6/include/python3.6m/node.h" "$(@D)/python_include/node.h" && cp "/opt/python3.6/include/python3.6m/funcobject.h" "$(@D)/python_include/funcobject.h" && cp "/opt/python3.6/include/python3.6m/eval.h" "$(@D)/python_include/eval.h" && cp "/opt/python3.6/include/python3.6m/pyatomic.h" "$(@D)/python_include/pyatomic.h" && cp "/opt/python3.6/include/python3.6m/longintrepr.h" "$(@D)/python_include/longintrepr.h" && cp "/opt/python3.6/include/python3.6m/floatobject.h" "$(@D)/python_include/floatobject.h" && cp "/opt/python3.6/include/python3.6m/rangeobject.h" "$(@D)/python_include/rangeobject.h" && cp "/opt/python3.6/include/python3.6m/pyfpe.h" "$(@D)/python_include/pyfpe.h" && cp "/opt/python3.6/include/python3.6m/pystrcmp.h" "$(@D)/python_include/pystrcmp.h" && cp "/opt/python3.6/include/python3.6m/fileutils.h" "$(@D)/python_include/fileutils.h" && cp "/opt/python3.6/include/python3.6m/dictobject.h" "$(@D)/python_include/dictobject.h" && cp "/opt/python3.6/include/python3.6m/pyarena.h" "$(@D)/python_include/pyarena.h" && cp "/opt/python3.6/include/python3.6m/osmodule.h" "$(@D)/python_include/osmodule.h" && cp "/opt/python3.6/include/python3.6m/objimpl.h" "$(@D)/python_include/objimpl.h" && cp "/opt/python3.6/include/python3.6m/bitset.h" "$(@D)/python_include/bitset.h" && cp "/opt/python3.6/include/python3.6m/memoryobject.h" "$(@D)/python_include/memoryobject.h" && cp "/opt/python3.6/include/python3.6m/bytearrayobject.h" "$(@D)/python_include/bytearrayobject.h" && cp "/opt/python3.6/include/python3.6m/pydebug.h" "$(@D)/python_include/pydebug.h" && cp "/opt/python3.6/include/python3.6m/pyerrors.h" "$(@D)/python_include/pyerrors.h" && cp "/opt/python3.6/include/python3.6m/weakrefobject.h" "$(@D)/python_include/weakrefobject.h" && cp "/opt/python3.6/include/python3.6m/grammar.h" "$(@D)/python_include/grammar.h" && cp "/opt/python3.6/include/python3.6m/symtable.h" "$(@D)/python_include/symtable.h" && cp "/opt/python3.6/include/python3.6m/longobject.h" "$(@D)/python_include/longobject.h" && cp "/opt/python3.6/include/python3.6m/structmember.h" "$(@D)/python_include/structmember.h" && cp "/opt/python3.6/include/python3.6m/enumobject.h" "$(@D)/python_include/enumobject.h" && cp "/opt/python3.6/include/python3.6m/pymacro.h" "$(@D)/python_include/pymacro.h" && cp "/opt/python3.6/include/python3.6m/classobject.h" "$(@D)/python_include/classobject.h" && cp "/opt/python3.6/include/python3.6m/unicodeobject.h" "$(@D)/python_include/unicodeobject.h" && cp "/opt/python3.6/include/python3.6m/sliceobject.h" "$(@D)/python_include/sliceobject.h" && cp "/opt/python3.6/include/python3.6m/pystrtod.h" "$(@D)/python_include/pystrtod.h" && cp "/opt/python3.6/include/python3.6m/genobject.h" "$(@D)/python_include/genobject.h" && cp "/opt/python3.6/include/python3.6m/compile.h" "$(@D)/python_include/compile.h" && cp "/opt/python3.6/include/python3.6m/pyexpat.h" "$(@D)/python_include/pyexpat.h" && cp "/opt/python3.6/include/python3.6m/asdl.h" "$(@D)/python_include/asdl.h" && cp "/opt/python3.6/include/python3.6m/codecs.h" "$(@D)/python_include/codecs.h" && cp "/opt/python3.6/include/python3.6m/dynamic_annotations.h" "$(@D)/python_include/dynamic_annotations.h" && cp "/opt/python3.6/include/python3.6m/pyctype.h" "$(@D)/python_include/pyctype.h" && cp "/opt/python3.6/include/python3.6m/sysmodule.h" "$(@D)/python_include/sysmodule.h" && cp "/opt/python3.6/include/python3.6m/methodobject.h" "$(@D)/python_include/methodobject.h" && cp "/opt/python3.6/include/python3.6m/graminit.h" "$(@D)/python_include/graminit.h" && cp "/opt/python3.6/include/python3.6m/bltinmodule.h" "$(@D)/python_include/bltinmodule.h" && cp "/opt/python3.6/include/python3.6m/intrcheck.h" "$(@D)/python_include/intrcheck.h" && cp "/opt/python3.6/include/python3.6m/pyport.h" "$(@D)/python_include/pyport.h" && cp "/opt/python3.6/include/python3.6m/warnings.h" "$(@D)/python_include/warnings.h" && cp "/opt/python3.6/include/python3.6m/osdefs.h" "$(@D)/python_include/osdefs.h" && cp "/opt/python3.6/include/python3.6m/pydtrace.h" "$(@D)/python_include/pydtrace.h" && cp "/opt/python3.6/include/python3.6m/pylifecycle.h" "$(@D)/python_include/pylifecycle.h" && cp "/opt/python3.6/include/python3.6m/fileobject.h" "$(@D)/python_include/fileobject.h" && cp "/opt/python3.6/include/python3.6m/pytime.h" "$(@D)/python_include/pytime.h" && cp "/opt/python3.6/include/python3.6m/traceback.h" "$(@D)/python_include/traceback.h" && cp "/opt/python3.6/include/python3.6m/ceval.h" "$(@D)/python_include/ceval.h" && cp "/opt/python3.6/include/python3.6m/bytes_methods.h" "$(@D)/python_include/bytes_methods.h" && cp "/opt/python3.6/include/python3.6m/namespaceobject.h" "$(@D)/python_include/namespaceobject.h" && cp "/opt/python3.6/include/python3.6m/pyconfig.h" "$(@D)/python_include/pyconfig.h" && cp "/opt/python3.6/include/python3.6m/Python.h" "$(@D)/python_include/Python.h" && cp "/opt/python3.6/include/python3.6m/moduleobject.h" "$(@D)/python_include/moduleobject.h" && cp "/opt/python3.6/include/python3.6m/pystate.h" "$(@D)/python_include/pystate.h" && cp "/opt/python3.6/include/python3.6m/descrobject.h" "$(@D)/python_include/descrobject.h" && cp "/opt/python3.6/include/python3.6m/odictobject.h" "$(@D)/python_include/odictobject.h" && cp "/opt/python3.6/include/python3.6m/ucnhash.h" "$(@D)/python_include/ucnhash.h" && cp "/opt/python3.6/include/python3.6m/pygetopt.h" "$(@D)/python_include/pygetopt.h" && cp "/opt/python3.6/include/python3.6m/pymem.h" "$(@D)/python_include/pymem.h" && cp "/opt/python3.6/include/python3.6m/complexobject.h" "$(@D)/python_include/complexobject.h" && cp "/opt/python3.6/include/python3.6m/structseq.h" "$(@D)/python_include/structseq.h" && cp "/opt/python3.6/include/python3.6m/datetime.h" "$(@D)/python_include/datetime.h" && cp "/opt/python3.6/include/python3.6m/pythonrun.h" "$(@D)/python_include/pythonrun.h" && cp "/opt/python3.6/include/python3.6m/pyhash.h" "$(@D)/python_include/pyhash.h" && cp "/opt/python3.6/include/python3.6m/pycapsule.h" "$(@D)/python_include/pycapsule.h" && cp "/opt/python3.6/include/python3.6m/setobject.h" "$(@D)/python_include/setobject.h" && cp "/opt/python3.6/include/python3.6m/listobject.h" "$(@D)/python_include/listobject.h" && cp "/opt/python3.6/include/python3.6m/bytesobject.h" "$(@D)/python_include/bytesobject.h" && cp "/opt/python3.6/include/python3.6m/pgen.h" "$(@D)/python_include/pgen.h" && cp "/opt/python3.6/include/python3.6m/patchlevel.h" "$(@D)/python_include/patchlevel.h" && cp "/opt/python3.6/include/python3.6m/opcode.h" "$(@D)/python_include/opcode.h" && cp "/opt/python3.6/include/python3.6m/parsetok.h" "$(@D)/python_include/parsetok.h" && cp "/opt/python3.6/include/python3.6m/pystrhex.h" "$(@D)/python_include/pystrhex.h" && cp "/opt/python3.6/include/python3.6m/marshal.h" "$(@D)/python_include/marshal.h" && cp "/opt/python3.6/include/python3.6m/token.h" "$(@D)/python_include/token.h" && cp "/opt/python3.6/include/python3.6m/iterobject.h" "$(@D)/python_include/iterobject.h" && cp "/opt/python3.6/include/python3.6m/abstract.h" "$(@D)/python_include/abstract.h" && cp "/opt/python3.6/include/python3.6m/py_curses.h" "$(@D)/python_include/py_curses.h" && cp "/opt/python3.6/include/python3.6m/metagrammar.h" "$(@D)/python_include/metagrammar.h" && cp "/opt/python3.6/include/python3.6m/Python-ast.h" "$(@D)/python_include/Python-ast.h" +cp "/opt/python3.6/include/python3.6m/Python-ast.h" "$(@D)/python_include/Python-ast.h" && cp "/opt/python3.6/include/python3.6m/Python.h" "$(@D)/python_include/Python.h" && cp "/opt/python3.6/include/python3.6m/abstract.h" "$(@D)/python_include/abstract.h" && cp "/opt/python3.6/include/python3.6m/accu.h" "$(@D)/python_include/accu.h" && cp "/opt/python3.6/include/python3.6m/asdl.h" "$(@D)/python_include/asdl.h" && cp "/opt/python3.6/include/python3.6m/ast.h" "$(@D)/python_include/ast.h" && cp "/opt/python3.6/include/python3.6m/bitset.h" "$(@D)/python_include/bitset.h" && cp "/opt/python3.6/include/python3.6m/bltinmodule.h" "$(@D)/python_include/bltinmodule.h" && cp "/opt/python3.6/include/python3.6m/boolobject.h" "$(@D)/python_include/boolobject.h" && cp "/opt/python3.6/include/python3.6m/bytearrayobject.h" "$(@D)/python_include/bytearrayobject.h" && cp "/opt/python3.6/include/python3.6m/bytes_methods.h" "$(@D)/python_include/bytes_methods.h" && cp "/opt/python3.6/include/python3.6m/bytesobject.h" "$(@D)/python_include/bytesobject.h" && cp "/opt/python3.6/include/python3.6m/cellobject.h" "$(@D)/python_include/cellobject.h" && cp "/opt/python3.6/include/python3.6m/ceval.h" "$(@D)/python_include/ceval.h" && cp "/opt/python3.6/include/python3.6m/classobject.h" "$(@D)/python_include/classobject.h" && cp "/opt/python3.6/include/python3.6m/code.h" "$(@D)/python_include/code.h" && cp "/opt/python3.6/include/python3.6m/codecs.h" "$(@D)/python_include/codecs.h" && cp "/opt/python3.6/include/python3.6m/compile.h" "$(@D)/python_include/compile.h" && cp "/opt/python3.6/include/python3.6m/complexobject.h" "$(@D)/python_include/complexobject.h" && cp "/opt/python3.6/include/python3.6m/datetime.h" "$(@D)/python_include/datetime.h" && cp "/opt/python3.6/include/python3.6m/descrobject.h" "$(@D)/python_include/descrobject.h" && cp "/opt/python3.6/include/python3.6m/dictobject.h" "$(@D)/python_include/dictobject.h" && cp "/opt/python3.6/include/python3.6m/dtoa.h" "$(@D)/python_include/dtoa.h" && cp "/opt/python3.6/include/python3.6m/dynamic_annotations.h" "$(@D)/python_include/dynamic_annotations.h" && cp "/opt/python3.6/include/python3.6m/enumobject.h" "$(@D)/python_include/enumobject.h" && cp "/opt/python3.6/include/python3.6m/errcode.h" "$(@D)/python_include/errcode.h" && cp "/opt/python3.6/include/python3.6m/eval.h" "$(@D)/python_include/eval.h" && cp "/opt/python3.6/include/python3.6m/fileobject.h" "$(@D)/python_include/fileobject.h" && cp "/opt/python3.6/include/python3.6m/fileutils.h" "$(@D)/python_include/fileutils.h" && cp "/opt/python3.6/include/python3.6m/floatobject.h" "$(@D)/python_include/floatobject.h" && cp "/opt/python3.6/include/python3.6m/frameobject.h" "$(@D)/python_include/frameobject.h" && cp "/opt/python3.6/include/python3.6m/funcobject.h" "$(@D)/python_include/funcobject.h" && cp "/opt/python3.6/include/python3.6m/genobject.h" "$(@D)/python_include/genobject.h" && cp "/opt/python3.6/include/python3.6m/graminit.h" "$(@D)/python_include/graminit.h" && cp "/opt/python3.6/include/python3.6m/grammar.h" "$(@D)/python_include/grammar.h" && cp "/opt/python3.6/include/python3.6m/import.h" "$(@D)/python_include/import.h" && cp "/opt/python3.6/include/python3.6m/intrcheck.h" "$(@D)/python_include/intrcheck.h" && cp "/opt/python3.6/include/python3.6m/iterobject.h" "$(@D)/python_include/iterobject.h" && cp "/opt/python3.6/include/python3.6m/listobject.h" "$(@D)/python_include/listobject.h" && cp "/opt/python3.6/include/python3.6m/longintrepr.h" "$(@D)/python_include/longintrepr.h" && cp "/opt/python3.6/include/python3.6m/longobject.h" "$(@D)/python_include/longobject.h" && cp "/opt/python3.6/include/python3.6m/marshal.h" "$(@D)/python_include/marshal.h" && cp "/opt/python3.6/include/python3.6m/memoryobject.h" "$(@D)/python_include/memoryobject.h" && cp "/opt/python3.6/include/python3.6m/metagrammar.h" "$(@D)/python_include/metagrammar.h" && cp "/opt/python3.6/include/python3.6m/methodobject.h" "$(@D)/python_include/methodobject.h" && cp "/opt/python3.6/include/python3.6m/modsupport.h" "$(@D)/python_include/modsupport.h" && cp "/opt/python3.6/include/python3.6m/moduleobject.h" "$(@D)/python_include/moduleobject.h" && cp "/opt/python3.6/include/python3.6m/namespaceobject.h" "$(@D)/python_include/namespaceobject.h" && cp "/opt/python3.6/include/python3.6m/node.h" "$(@D)/python_include/node.h" && cp "/opt/python3.6/include/python3.6m/object.h" "$(@D)/python_include/object.h" && cp "/opt/python3.6/include/python3.6m/objimpl.h" "$(@D)/python_include/objimpl.h" && cp "/opt/python3.6/include/python3.6m/odictobject.h" "$(@D)/python_include/odictobject.h" && cp "/opt/python3.6/include/python3.6m/opcode.h" "$(@D)/python_include/opcode.h" && cp "/opt/python3.6/include/python3.6m/osdefs.h" "$(@D)/python_include/osdefs.h" && cp "/opt/python3.6/include/python3.6m/osmodule.h" "$(@D)/python_include/osmodule.h" && cp "/opt/python3.6/include/python3.6m/parsetok.h" "$(@D)/python_include/parsetok.h" && cp "/opt/python3.6/include/python3.6m/patchlevel.h" "$(@D)/python_include/patchlevel.h" && cp "/opt/python3.6/include/python3.6m/pgen.h" "$(@D)/python_include/pgen.h" && cp "/opt/python3.6/include/python3.6m/pgenheaders.h" "$(@D)/python_include/pgenheaders.h" && cp "/opt/python3.6/include/python3.6m/py_curses.h" "$(@D)/python_include/py_curses.h" && cp "/opt/python3.6/include/python3.6m/pyarena.h" "$(@D)/python_include/pyarena.h" && cp "/opt/python3.6/include/python3.6m/pyatomic.h" "$(@D)/python_include/pyatomic.h" && cp "/opt/python3.6/include/python3.6m/pycapsule.h" "$(@D)/python_include/pycapsule.h" && cp "/opt/python3.6/include/python3.6m/pyconfig.h" "$(@D)/python_include/pyconfig.h" && cp "/opt/python3.6/include/python3.6m/pyctype.h" "$(@D)/python_include/pyctype.h" && cp "/opt/python3.6/include/python3.6m/pydebug.h" "$(@D)/python_include/pydebug.h" && cp "/opt/python3.6/include/python3.6m/pydtrace.h" "$(@D)/python_include/pydtrace.h" && cp "/opt/python3.6/include/python3.6m/pyerrors.h" "$(@D)/python_include/pyerrors.h" && cp "/opt/python3.6/include/python3.6m/pyexpat.h" "$(@D)/python_include/pyexpat.h" && cp "/opt/python3.6/include/python3.6m/pyfpe.h" "$(@D)/python_include/pyfpe.h" && cp "/opt/python3.6/include/python3.6m/pygetopt.h" "$(@D)/python_include/pygetopt.h" && cp "/opt/python3.6/include/python3.6m/pyhash.h" "$(@D)/python_include/pyhash.h" && cp "/opt/python3.6/include/python3.6m/pylifecycle.h" "$(@D)/python_include/pylifecycle.h" && cp "/opt/python3.6/include/python3.6m/pymacconfig.h" "$(@D)/python_include/pymacconfig.h" && cp "/opt/python3.6/include/python3.6m/pymacro.h" "$(@D)/python_include/pymacro.h" && cp "/opt/python3.6/include/python3.6m/pymath.h" "$(@D)/python_include/pymath.h" && cp "/opt/python3.6/include/python3.6m/pymem.h" "$(@D)/python_include/pymem.h" && cp "/opt/python3.6/include/python3.6m/pyport.h" "$(@D)/python_include/pyport.h" && cp "/opt/python3.6/include/python3.6m/pystate.h" "$(@D)/python_include/pystate.h" && cp "/opt/python3.6/include/python3.6m/pystrcmp.h" "$(@D)/python_include/pystrcmp.h" && cp "/opt/python3.6/include/python3.6m/pystrhex.h" "$(@D)/python_include/pystrhex.h" && cp "/opt/python3.6/include/python3.6m/pystrtod.h" "$(@D)/python_include/pystrtod.h" && cp "/opt/python3.6/include/python3.6m/pythonrun.h" "$(@D)/python_include/pythonrun.h" && cp "/opt/python3.6/include/python3.6m/pythread.h" "$(@D)/python_include/pythread.h" && cp "/opt/python3.6/include/python3.6m/pytime.h" "$(@D)/python_include/pytime.h" && cp "/opt/python3.6/include/python3.6m/rangeobject.h" "$(@D)/python_include/rangeobject.h" && cp "/opt/python3.6/include/python3.6m/setobject.h" "$(@D)/python_include/setobject.h" && cp "/opt/python3.6/include/python3.6m/sliceobject.h" "$(@D)/python_include/sliceobject.h" && cp "/opt/python3.6/include/python3.6m/structmember.h" "$(@D)/python_include/structmember.h" && cp "/opt/python3.6/include/python3.6m/structseq.h" "$(@D)/python_include/structseq.h" && cp "/opt/python3.6/include/python3.6m/symtable.h" "$(@D)/python_include/symtable.h" && cp "/opt/python3.6/include/python3.6m/sysmodule.h" "$(@D)/python_include/sysmodule.h" && cp "/opt/python3.6/include/python3.6m/token.h" "$(@D)/python_include/token.h" && cp "/opt/python3.6/include/python3.6m/traceback.h" "$(@D)/python_include/traceback.h" && cp "/opt/python3.6/include/python3.6m/tupleobject.h" "$(@D)/python_include/tupleobject.h" && cp "/opt/python3.6/include/python3.6m/typeslots.h" "$(@D)/python_include/typeslots.h" && cp "/opt/python3.6/include/python3.6m/ucnhash.h" "$(@D)/python_include/ucnhash.h" && cp "/opt/python3.6/include/python3.6m/unicodeobject.h" "$(@D)/python_include/unicodeobject.h" && cp "/opt/python3.6/include/python3.6m/warnings.h" "$(@D)/python_include/warnings.h" && cp "/opt/python3.6/include/python3.6m/weakrefobject.h" "$(@D)/python_include/weakrefobject.h" """, ) genrule( name = "numpy_include", outs = [ - "numpy_include/numpy/oldnumeric.h", - "numpy_include/numpy/npy_1_7_deprecated_api.h", - "numpy_include/numpy/ufunc_api.txt", - "numpy_include/numpy/multiarray_api.txt", - "numpy_include/numpy/halffloat.h", - "numpy_include/numpy/npy_common.h", - "numpy_include/numpy/utils.h", - "numpy_include/numpy/npy_interrupt.h", - "numpy_include/numpy/npy_endian.h", + "numpy_include/numpy/__multiarray_api.h", "numpy_include/numpy/__ufunc_api.h", "numpy_include/numpy/_neighborhood_iterator_imp.h", - "numpy_include/numpy/ufuncobject.h", + "numpy_include/numpy/_numpyconfig.h", + "numpy_include/numpy/arrayobject.h", + "numpy_include/numpy/arrayscalars.h", + "numpy_include/numpy/halffloat.h", + "numpy_include/numpy/multiarray_api.txt", + "numpy_include/numpy/ndarrayobject.h", "numpy_include/numpy/ndarraytypes.h", - "numpy_include/numpy/npy_math.h", "numpy_include/numpy/noprefix.h", + "numpy_include/numpy/npy_1_7_deprecated_api.h", "numpy_include/numpy/npy_3kcompat.h", - "numpy_include/numpy/arrayscalars.h", - "numpy_include/numpy/npy_os.h", - "numpy_include/numpy/ndarrayobject.h", - "numpy_include/numpy/npy_no_deprecated_api.h", - "numpy_include/numpy/arrayobject.h", - "numpy_include/numpy/_numpyconfig.h", - "numpy_include/numpy/__multiarray_api.h", + "numpy_include/numpy/npy_common.h", "numpy_include/numpy/npy_cpu.h", - "numpy_include/numpy/old_defines.h", + "numpy_include/numpy/npy_endian.h", + "numpy_include/numpy/npy_interrupt.h", + "numpy_include/numpy/npy_math.h", + "numpy_include/numpy/npy_no_deprecated_api.h", + "numpy_include/numpy/npy_os.h", "numpy_include/numpy/numpyconfig.h", + "numpy_include/numpy/old_defines.h", + "numpy_include/numpy/oldnumeric.h", + "numpy_include/numpy/ufunc_api.txt", + "numpy_include/numpy/ufuncobject.h", + "numpy_include/numpy/utils.h", ], cmd = """ -cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/oldnumeric.h" "$(@D)/numpy_include/numpy/oldnumeric.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/npy_1_7_deprecated_api.h" "$(@D)/numpy_include/numpy/npy_1_7_deprecated_api.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/ufunc_api.txt" "$(@D)/numpy_include/numpy/ufunc_api.txt" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/multiarray_api.txt" "$(@D)/numpy_include/numpy/multiarray_api.txt" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/halffloat.h" "$(@D)/numpy_include/numpy/halffloat.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/npy_common.h" "$(@D)/numpy_include/numpy/npy_common.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/utils.h" "$(@D)/numpy_include/numpy/utils.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/npy_interrupt.h" "$(@D)/numpy_include/numpy/npy_interrupt.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/npy_endian.h" "$(@D)/numpy_include/numpy/npy_endian.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/__ufunc_api.h" "$(@D)/numpy_include/numpy/__ufunc_api.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/_neighborhood_iterator_imp.h" "$(@D)/numpy_include/numpy/_neighborhood_iterator_imp.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/ufuncobject.h" "$(@D)/numpy_include/numpy/ufuncobject.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/ndarraytypes.h" "$(@D)/numpy_include/numpy/ndarraytypes.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/npy_math.h" "$(@D)/numpy_include/numpy/npy_math.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/noprefix.h" "$(@D)/numpy_include/numpy/noprefix.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/npy_3kcompat.h" "$(@D)/numpy_include/numpy/npy_3kcompat.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/arrayscalars.h" "$(@D)/numpy_include/numpy/arrayscalars.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/npy_os.h" "$(@D)/numpy_include/numpy/npy_os.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/ndarrayobject.h" "$(@D)/numpy_include/numpy/ndarrayobject.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/npy_no_deprecated_api.h" "$(@D)/numpy_include/numpy/npy_no_deprecated_api.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/arrayobject.h" "$(@D)/numpy_include/numpy/arrayobject.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/_numpyconfig.h" "$(@D)/numpy_include/numpy/_numpyconfig.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/__multiarray_api.h" "$(@D)/numpy_include/numpy/__multiarray_api.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/npy_cpu.h" "$(@D)/numpy_include/numpy/npy_cpu.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/old_defines.h" "$(@D)/numpy_include/numpy/old_defines.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/numpyconfig.h" "$(@D)/numpy_include/numpy/numpyconfig.h" +cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/__multiarray_api.h" "$(@D)/numpy_include/numpy/__multiarray_api.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/__ufunc_api.h" "$(@D)/numpy_include/numpy/__ufunc_api.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/_neighborhood_iterator_imp.h" "$(@D)/numpy_include/numpy/_neighborhood_iterator_imp.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/_numpyconfig.h" "$(@D)/numpy_include/numpy/_numpyconfig.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/arrayobject.h" "$(@D)/numpy_include/numpy/arrayobject.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/arrayscalars.h" "$(@D)/numpy_include/numpy/arrayscalars.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/halffloat.h" "$(@D)/numpy_include/numpy/halffloat.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/multiarray_api.txt" "$(@D)/numpy_include/numpy/multiarray_api.txt" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/ndarrayobject.h" "$(@D)/numpy_include/numpy/ndarrayobject.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/ndarraytypes.h" "$(@D)/numpy_include/numpy/ndarraytypes.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/noprefix.h" "$(@D)/numpy_include/numpy/noprefix.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/npy_1_7_deprecated_api.h" "$(@D)/numpy_include/numpy/npy_1_7_deprecated_api.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/npy_3kcompat.h" "$(@D)/numpy_include/numpy/npy_3kcompat.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/npy_common.h" "$(@D)/numpy_include/numpy/npy_common.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/npy_cpu.h" "$(@D)/numpy_include/numpy/npy_cpu.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/npy_endian.h" "$(@D)/numpy_include/numpy/npy_endian.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/npy_interrupt.h" "$(@D)/numpy_include/numpy/npy_interrupt.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/npy_math.h" "$(@D)/numpy_include/numpy/npy_math.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/npy_no_deprecated_api.h" "$(@D)/numpy_include/numpy/npy_no_deprecated_api.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/npy_os.h" "$(@D)/numpy_include/numpy/npy_os.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/numpyconfig.h" "$(@D)/numpy_include/numpy/numpyconfig.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/old_defines.h" "$(@D)/numpy_include/numpy/old_defines.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/oldnumeric.h" "$(@D)/numpy_include/numpy/oldnumeric.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/ufunc_api.txt" "$(@D)/numpy_include/numpy/ufunc_api.txt" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/ufuncobject.h" "$(@D)/numpy_include/numpy/ufuncobject.h" && cp "/opt/python3.6/lib/python3.6/site-packages/numpy/core/include/numpy/utils.h" "$(@D)/numpy_include/numpy/utils.h" """, ) diff --git a/third_party/toolchains/gpus/crosstool/BUILD b/third_party/toolchains/gpus/crosstool/BUILD index 1f9065007c..bb0b6b3bbb 100644 --- a/third_party/toolchains/gpus/crosstool/BUILD +++ b/third_party/toolchains/gpus/crosstool/BUILD @@ -11,6 +11,7 @@ cc_toolchain_suite( toolchains = { "local|compiler": ":cc-compiler-local", "darwin|compiler": ":cc-compiler-darwin", + "x64_windows|msvc-cl": ":cc-compiler-windows", }, ) @@ -46,6 +47,20 @@ cc_toolchain( supports_param_files = 0, ) +cc_toolchain( + name = "cc-compiler-windows", + all_files = ":empty", + compiler_files = ":empty", + cpu = "x64_windows", + dwp_files = ":empty", + dynamic_runtime_libs = [":empty"], + linker_files = ":empty", + objcopy_files = ":empty", + static_runtime_libs = [":empty"], + strip_files = ":empty", + supports_param_files = 1, +) + filegroup( name = "empty", srcs = [], @@ -55,3 +70,8 @@ filegroup( name = "crosstool_wrapper_driver_is_not_gcc", srcs = ["clang/bin/crosstool_wrapper_driver_is_not_gcc"], ) + +filegroup( + name = "windows_msvc_wrapper_files", + srcs = glob(["windows/msvc_*"]), +) diff --git a/third_party/toolchains/gpus/crosstool/CROSSTOOL b/third_party/toolchains/gpus/crosstool/CROSSTOOL index d6ee7e38c4..b8eeb31ecb 100644 --- a/third_party/toolchains/gpus/crosstool/CROSSTOOL +++ b/third_party/toolchains/gpus/crosstool/CROSSTOOL @@ -26,6 +26,10 @@ default_toolchain { cpu: "ppc" toolchain_identifier: "local_linux" } +default_toolchain { + cpu: "x64_windows" + toolchain_identifier: "local_windows" +} toolchain { abi_version: "local" @@ -144,9 +148,11 @@ toolchain { flag_group { # All warnings are enabled. Maybe enable -Werror as well? flag: "-Wall" + # Some parts of the codebase set -Werror and hit this warning, so # switch it off for now. flag: "-Wno-invalid-partial-specialization" + } } } @@ -307,3 +313,1120 @@ toolchain { cxx_builtin_include_directory: "/usr/include/x86_64-linux-gnu" cxx_builtin_include_directory: "/usr/include" } + +toolchain { + abi_version: "local" + abi_libc_version: "local" + compiler: "compiler" + host_system_name: "local" + needsPic: true + target_libc: "macosx" + target_cpu: "darwin" + target_system_name: "local" + toolchain_identifier: "local_darwin" + feature { + name: "c++11" + flag_set { + action: "c++-compile" + flag_group { + flag: "-std=c++11" + } + } + } + + feature { + name: "stdlib" + flag_set { + action: "c++-link-executable" + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "-lc++" + } + } + } + + feature { + name: "determinism" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + # Make C++ compilation deterministic. Use linkstamping instead of these + # compiler symbols. + flag: "-Wno-builtin-macro-redefined" + flag: "-D__DATE__=\"redacted\"" + flag: "-D__TIMESTAMP__=\"redacted\"" + flag: "-D__TIME__=\"redacted\"" + } + } + } + + # This feature will be enabled for builds that support pic by bazel. + feature { + name: "pic" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + expand_if_all_available: "pic" + flag: "-fPIC" + } + flag_group { + expand_if_none_available: "pic" + flag: "-fPIE" + } + } + } + + # Security hardening on by default. + feature { + name: "hardening" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + # Conservative choice; -D_FORTIFY_SOURCE=2 may be unsafe in some cases. + # We need to undef it before redefining it as some distributions now + # have it enabled by default. + flag: "-U_FORTIFY_SOURCE" + flag: "-D_FORTIFY_SOURCE=1" + flag: "-fstack-protector" + } + } + flag_set { + action: "c++-link-executable" + flag_group { + flag: "-pie" + } + } + } + + feature { + name: "warnings" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + # All warnings are enabled. Maybe enable -Werror as well? + flag: "-Wall" + + # Some parts of the codebase set -Werror and hit this warning, so + # switch it off for now. + flag: "-Wno-invalid-partial-specialization" + + } + } + } + + # Keep stack frames for debugging, even in opt mode. + feature { + name: "frame-pointer" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + flag: "-fno-omit-frame-pointer" + } + } + } + + feature { + name: "no-canonical-prefixes" + flag_set { + action: "c-compile" + action: "c++-compile" + action: "c++-link-executable" + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag:"-no-canonical-prefixes" + } + } + } + + feature { + name: "disable-assertions" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + flag: "-DNDEBUG" + } + } + } + + feature { + name: "linker-bin-path" + + flag_set { + action: "c++-link-executable" + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "-B/usr/bin/" + } + } + } + + feature { + name: "undefined-dynamic" + flag_set { + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + action: "c++-link-executable" + flag_group { + flag: "-undefined" + flag: "dynamic_lookup" + } + } + } + + feature { + name: "common" + implies: "stdlib" + implies: "c++11" + implies: "determinism" + implies: "hardening" + implies: "warnings" + implies: "frame-pointer" + implies: "no-canonical-prefixes" + implies: "linker-bin-path" + implies: "undefined-dynamic" + } + + feature { + name: "opt" + implies: "common" + implies: "disable-assertions" + + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + # No debug symbols. + # Maybe we should enable https://gcc.gnu.org/wiki/DebugFission for opt + # or even generally? However, that can't happen here, as it requires + # special handling in Bazel. + flag: "-g0" + + # Conservative choice for -O + # -O3 can increase binary size and even slow down the resulting binaries. + # Profile first and / or use FDO if you need better performance than this. + flag: "-O2" + + # Removal of unused code and data at link time (can this increase binary size in some cases?). + flag: "-ffunction-sections" + flag: "-fdata-sections" + } + } + } + + feature { + name: "fastbuild" + implies: "common" + } + + feature { + name: "dbg" + implies: "common" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + flag: "-g" + } + } + } + + # Set clang as a C/C++ compiler. + tool_path { name: "gcc" path: "/usr/local/bin/clang" } + + # Use the default system toolchain for everything else. + tool_path { name: "ar" path: "/usr/bin/libtool" } + tool_path { name: "compat-ld" path: "/usr/bin/ld" } + tool_path { name: "cpp" path: "/usr/bin/cpp" } + tool_path { name: "dwp" path: "/usr/bin/dwp" } + tool_path { name: "gcov" path: "/usr/bin/gcov" } + tool_path { name: "ld" path: "/usr/bin/ld" } + tool_path { name: "nm" path: "/usr/bin/nm" } + tool_path { name: "objcopy" path: "/usr/bin/objcopy" } + tool_path { name: "objdump" path: "/usr/bin/objdump" } + tool_path { name: "strip" path: "/usr/bin/strip" } + + # Enabled dynamic linking. + linking_mode_flags { mode: DYNAMIC } + + cxx_builtin_include_directory: "/usr/include/c++/5.4.0" + cxx_builtin_include_directory: "/usr/include/x86_64-linux-gnu/c++/5.4.0" + cxx_builtin_include_directory: "/usr/include/c++/5.4.0/backward" + cxx_builtin_include_directory: "/usr/local/include" + cxx_builtin_include_directory: "/usr/local/lib/clang/7.0.0/include" + cxx_builtin_include_directory: "/usr/include/x86_64-linux-gnu" + cxx_builtin_include_directory: "/usr/include" +} + +toolchain { + toolchain_identifier: "local_windows" + host_system_name: "local" + target_system_name: "local" + + abi_version: "local" + abi_libc_version: "local" + target_cpu: "x64_windows" + compiler: "msvc-cl" + target_libc: "msvcrt" + + + + tool_path { + name: "ar" + path: "" + } + tool_path { + name: "ml" + path: "" + } + tool_path { + name: "cpp" + path: "" + } + tool_path { + name: "gcc" + path: "" + } + tool_path { + name: "gcov" + path: "wrapper/bin/msvc_nop.bat" + } + tool_path { + name: "ld" + path: "" + } + tool_path { + name: "nm" + path: "wrapper/bin/msvc_nop.bat" + } + tool_path { + name: "objcopy" + path: "wrapper/bin/msvc_nop.bat" + } + tool_path { + name: "objdump" + path: "wrapper/bin/msvc_nop.bat" + } + tool_path { + name: "strip" + path: "wrapper/bin/msvc_nop.bat" + } + supports_interface_shared_objects: true + + # TODO(pcloudy): Review those flags below, they should be defined by cl.exe + compiler_flag: "/DCOMPILER_MSVC" + + # Don't define min/max macros in windows.h. + compiler_flag: "/DNOMINMAX" + + # Platform defines. + compiler_flag: "/D_WIN32_WINNT=0x0600" + # Turn off warning messages. + compiler_flag: "/D_CRT_SECURE_NO_DEPRECATE" + compiler_flag: "/D_CRT_SECURE_NO_WARNINGS" + compiler_flag: "/D_SILENCE_STDEXT_HASH_DEPRECATION_WARNINGS" + + # Useful options to have on for compilation. + # Increase the capacity of object files to 2^32 sections. + compiler_flag: "/bigobj" + # Allocate 500MB for precomputed headers. + compiler_flag: "/Zm500" + # Use unsigned char by default. + compiler_flag: "/J" + # Use function level linking. + compiler_flag: "/Gy" + # Use string pooling. + compiler_flag: "/GF" + # Catch C++ exceptions only and tell the compiler to assume that functions declared + # as extern "C" never throw a C++ exception. + compiler_flag: "/EHsc" + + # Globally disabled warnings. + # Don't warn about elements of array being be default initialized. + compiler_flag: "/wd4351" + # Don't warn about no matching delete found. + compiler_flag: "/wd4291" + # Don't warn about diamond inheritance patterns. + compiler_flag: "/wd4250" + # Don't warn about insecure functions (e.g. non _s functions). + compiler_flag: "/wd4996" + + linker_flag: "/MACHINE:X64" + + feature { + name: "no_legacy_features" + } + + # Suppress startup banner. + feature { + name: "nologo" + flag_set { + action: "c-compile" + action: "c++-compile" + action: "c++-module-compile" + action: "c++-module-codegen" + action: "c++-header-parsing" + action: "assemble" + action: "preprocess-assemble" + action: "c++-link-executable" + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + action: "c++-link-static-library" + flag_group { + flag: "/nologo" + } + } + } + + feature { + name: 'has_configured_linker_path' + } + + # This feature indicates strip is not supported, building stripped binary will just result a copy of orignial binary + feature { + name: 'no_stripping' + } + + # This feature indicates this is a toolchain targeting Windows. + feature { + name: 'targets_windows' + implies: 'copy_dynamic_libraries_to_binary' + enabled: true + } + + feature { + name: 'copy_dynamic_libraries_to_binary' + } + + action_config { + config_name: 'assemble' + action_name: 'assemble' + tool { + tool_path: '' + } + implies: 'compiler_input_flags' + implies: 'compiler_output_flags' + implies: 'nologo' + implies: 'msvc_env' + implies: 'sysroot' + } + + action_config { + config_name: 'preprocess-assemble' + action_name: 'preprocess-assemble' + tool { + tool_path: '' + } + implies: 'compiler_input_flags' + implies: 'compiler_output_flags' + implies: 'nologo' + implies: 'msvc_env' + implies: 'sysroot' + } + + action_config { + config_name: 'c-compile' + action_name: 'c-compile' + tool { + tool_path: '' + } + implies: 'compiler_input_flags' + implies: 'compiler_output_flags' + implies: 'legacy_compile_flags' + implies: 'nologo' + implies: 'msvc_env' + implies: 'parse_showincludes' + implies: 'user_compile_flags' + implies: 'sysroot' + implies: 'unfiltered_compile_flags' + } + + action_config { + config_name: 'c++-compile' + action_name: 'c++-compile' + tool { + tool_path: '' + } + implies: 'compiler_input_flags' + implies: 'compiler_output_flags' + implies: 'legacy_compile_flags' + implies: 'nologo' + implies: 'msvc_env' + implies: 'parse_showincludes' + implies: 'user_compile_flags' + implies: 'sysroot' + implies: 'unfiltered_compile_flags' + } + + action_config { + config_name: 'c++-link-executable' + action_name: 'c++-link-executable' + tool { + tool_path: '' + } + implies: 'nologo' + implies: 'linkstamps' + implies: 'output_execpath_flags' + implies: 'input_param_flags' + implies: 'user_link_flags' + implies: 'legacy_link_flags' + implies: 'linker_subsystem_flag' + implies: 'linker_param_file' + implies: 'msvc_env' + implies: 'no_stripping' + } + + action_config { + config_name: 'c++-link-dynamic-library' + action_name: 'c++-link-dynamic-library' + tool { + tool_path: '' + } + implies: 'nologo' + implies: 'shared_flag' + implies: 'linkstamps' + implies: 'output_execpath_flags' + implies: 'input_param_flags' + implies: 'user_link_flags' + implies: 'legacy_link_flags' + implies: 'linker_subsystem_flag' + implies: 'linker_param_file' + implies: 'msvc_env' + implies: 'no_stripping' + implies: 'has_configured_linker_path' + implies: 'def_file' + } + + action_config { + config_name: 'c++-link-nodeps-dynamic-library' + action_name: 'c++-link-nodeps-dynamic-library' + tool { + tool_path: '' + } + implies: 'nologo' + implies: 'shared_flag' + implies: 'linkstamps' + implies: 'output_execpath_flags' + implies: 'input_param_flags' + implies: 'user_link_flags' + implies: 'legacy_link_flags' + implies: 'linker_subsystem_flag' + implies: 'linker_param_file' + implies: 'msvc_env' + implies: 'no_stripping' + implies: 'has_configured_linker_path' + implies: 'def_file' + } + + action_config { + config_name: 'c++-link-static-library' + action_name: 'c++-link-static-library' + tool { + tool_path: '' + } + implies: 'nologo' + implies: 'archiver_flags' + implies: 'input_param_flags' + implies: 'linker_param_file' + implies: 'msvc_env' + } + + # TODO(b/65151735): Remove legacy_compile_flags feature when legacy fields are + # not used in this crosstool + feature { + name: 'legacy_compile_flags' + flag_set { + expand_if_all_available: 'legacy_compile_flags' + action: 'preprocess-assemble' + action: 'c-compile' + action: 'c++-compile' + action: 'c++-header-parsing' + action: 'c++-module-compile' + action: 'c++-module-codegen' + flag_group { + iterate_over: 'legacy_compile_flags' + flag: '%{legacy_compile_flags}' + } + } + } + + feature { + name: "msvc_env" + env_set { + action: "c-compile" + action: "c++-compile" + action: "c++-module-compile" + action: "c++-module-codegen" + action: "c++-header-parsing" + action: "assemble" + action: "preprocess-assemble" + action: "c++-link-executable" + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + action: "c++-link-static-library" + env_entry { + key: "PATH" + value: "" + } + env_entry { + key: "INCLUDE" + value: "" + } + env_entry { + key: "LIB" + value: "" + } + env_entry { + key: "TMP" + value: "" + } + env_entry { + key: "TEMP" + value: "" + } + } + } + + feature { + name: 'include_paths' + flag_set { + action: "assemble" + action: 'preprocess-assemble' + action: 'c-compile' + action: 'c++-compile' + action: 'c++-header-parsing' + action: 'c++-module-compile' + flag_group { + iterate_over: 'quote_include_paths' + flag: '/I%{quote_include_paths}' + } + flag_group { + iterate_over: 'include_paths' + flag: '/I%{include_paths}' + } + flag_group { + iterate_over: 'system_include_paths' + flag: '/I%{system_include_paths}' + } + } + } + + feature { + name: "preprocessor_defines" + flag_set { + action: "assemble" + action: "preprocess-assemble" + action: "c-compile" + action: "c++-compile" + action: "c++-header-parsing" + action: "c++-module-compile" + flag_group { + flag: "/D%{preprocessor_defines}" + iterate_over: "preprocessor_defines" + } + } + } + + # Tell Bazel to parse the output of /showIncludes + feature { + name: 'parse_showincludes' + flag_set { + action: 'preprocess-assemble' + action: 'c-compile' + action: 'c++-compile' + action: 'c++-module-compile' + action: 'c++-header-parsing' + flag_group { + flag: "/showIncludes" + } + } + } + + + feature { + name: 'generate_pdb_file' + requires: { + feature: 'dbg' + } + requires: { + feature: 'fastbuild' + } + } + + feature { + name: 'shared_flag' + flag_set { + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: '/DLL' + } + } + } + + feature { + name: 'linkstamps' + flag_set { + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + expand_if_all_available: 'linkstamp_paths' + flag_group { + iterate_over: 'linkstamp_paths' + flag: '%{linkstamp_paths}' + } + } + } + + feature { + name: 'output_execpath_flags' + flag_set { + expand_if_all_available: 'output_execpath' + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: '/OUT:%{output_execpath}' + } + } + } + + feature { + name: 'archiver_flags' + flag_set { + expand_if_all_available: 'output_execpath' + action: 'c++-link-static-library' + flag_group { + flag: '/OUT:%{output_execpath}' + } + } + } + + feature { + name: 'input_param_flags' + flag_set { + expand_if_all_available: 'interface_library_output_path' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "/IMPLIB:%{interface_library_output_path}" + } + } + flag_set { + expand_if_all_available: 'libopts' + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + iterate_over: 'libopts' + flag: '%{libopts}' + } + } + flag_set { + expand_if_all_available: 'libraries_to_link' + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + action: 'c++-link-static-library' + flag_group { + iterate_over: 'libraries_to_link' + flag_group { + expand_if_equal: { + variable: 'libraries_to_link.type' + value: 'object_file_group' + } + iterate_over: 'libraries_to_link.object_files' + flag_group { + flag: '%{libraries_to_link.object_files}' + } + } + flag_group { + expand_if_equal: { + variable: 'libraries_to_link.type' + value: 'object_file' + } + flag_group { + flag: '%{libraries_to_link.name}' + } + } + flag_group { + expand_if_equal: { + variable: 'libraries_to_link.type' + value: 'interface_library' + } + flag_group { + flag: '%{libraries_to_link.name}' + } + } + flag_group { + expand_if_equal: { + variable: 'libraries_to_link.type' + value: 'static_library' + } + flag_group { + expand_if_false: 'libraries_to_link.is_whole_archive' + flag: '%{libraries_to_link.name}' + } + flag_group { + expand_if_true: 'libraries_to_link.is_whole_archive' + flag: '/WHOLEARCHIVE:%{libraries_to_link.name}' + } + } + } + } + } + + # Since this feature is declared earlier in the CROSSTOOL than + # "user_link_flags", this feature will be applied prior to it anwyhere they + # are both implied. And since "user_link_flags" contains the linkopts from + # the build rule, this allows the user to override the /SUBSYSTEM in the BUILD + # file. + feature { + name: 'linker_subsystem_flag' + flag_set { + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: '/SUBSYSTEM:CONSOLE' + } + } + } + + # The "user_link_flags" contains user-defined linkopts (from build rules) + # so it should be defined after features that declare user-overridable flags. + # For example the "linker_subsystem_flag" defines a default "/SUBSYSTEM" flag + # but we want to let the user override it, therefore "link_flag_subsystem" is + # defined earlier in the CROSSTOOL file than "user_link_flags". + feature { + name: 'user_link_flags' + flag_set { + expand_if_all_available: 'user_link_flags' + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + iterate_over: 'user_link_flags' + flag: '%{user_link_flags}' + } + } + } + feature { + name: 'legacy_link_flags' + flag_set { + expand_if_all_available: 'legacy_link_flags' + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + iterate_over: 'legacy_link_flags' + flag: '%{legacy_link_flags}' + } + } + } + + feature { + name: 'linker_param_file' + flag_set { + expand_if_all_available: 'linker_param_file' + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + action: 'c++-link-static-library' + flag_group { + flag: '@%{linker_param_file}' + } + } + } + + feature { + name: 'static_link_msvcrt' + } + + feature { + name: 'static_link_msvcrt_no_debug' + flag_set { + action: 'c-compile' + action: 'c++-compile' + flag_group { + flag: "/MT" + } + } + flag_set { + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "/DEFAULTLIB:libcmt.lib" + } + } + requires: { feature: 'fastbuild'} + requires: { feature: 'opt'} + } + + feature { + name: 'dynamic_link_msvcrt_no_debug' + flag_set { + action: 'c-compile' + action: 'c++-compile' + flag_group { + flag: "/MD" + } + } + flag_set { + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "/DEFAULTLIB:msvcrt.lib" + } + } + requires: { feature: 'fastbuild'} + requires: { feature: 'opt'} + } + + feature { + name: 'static_link_msvcrt_debug' + flag_set { + action: 'c-compile' + action: 'c++-compile' + flag_group { + flag: "/MTd" + } + } + flag_set { + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "/DEFAULTLIB:libcmtd.lib" + } + } + requires: { feature: 'dbg'} + } + + feature { + name: 'dynamic_link_msvcrt_debug' + flag_set { + action: 'c-compile' + action: 'c++-compile' + flag_group { + flag: "/MDd" + } + } + flag_set { + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "/DEFAULTLIB:msvcrtd.lib" + } + } + requires: { feature: 'dbg'} + } + + feature { + name: 'dbg' + flag_set { + action: 'c-compile' + action: 'c++-compile' + flag_group { + flag: "/Od" + flag: "/Z7" + flag: "/DDEBUG" + } + } + flag_set { + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "/DEBUG:FULL" + flag: "/INCREMENTAL:NO" + } + } + implies: 'generate_pdb_file' + } + + feature { + name: 'fastbuild' + flag_set { + action: 'c-compile' + action: 'c++-compile' + flag_group { + flag: "/Od" + flag: "/Z7" + flag: "/DDEBUG" + } + } + flag_set { + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "/DEBUG:FASTLINK" + flag: "/INCREMENTAL:NO" + } + } + implies: 'generate_pdb_file' + } + + feature { + name: 'opt' + flag_set { + action: 'c-compile' + action: 'c++-compile' + flag_group { + flag: "/O2" + flag: "/DNDEBUG" + } + } + } + + feature { + name: 'user_compile_flags' + flag_set { + expand_if_all_available: 'user_compile_flags' + action: 'preprocess-assemble' + action: 'c-compile' + action: 'c++-compile' + action: 'c++-header-parsing' + action: 'c++-module-compile' + action: 'c++-module-codegen' + flag_group { + iterate_over: 'user_compile_flags' + flag: '%{user_compile_flags}' + } + } + } + + feature { + name: 'sysroot' + flag_set { + expand_if_all_available: 'sysroot' + action: 'assemble' + action: 'preprocess-assemble' + action: 'c-compile' + action: 'c++-compile' + action: 'c++-header-parsing' + action: 'c++-module-compile' + action: 'c++-module-codegen' + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + iterate_over: 'sysroot' + flag: '--sysroot=%{sysroot}' + } + } + } + + feature { + name: 'unfiltered_compile_flags' + flag_set { + expand_if_all_available: 'unfiltered_compile_flags' + action: 'preprocess-assemble' + action: 'c-compile' + action: 'c++-compile' + action: 'c++-header-parsing' + action: 'c++-module-compile' + action: 'c++-module-codegen' + flag_group { + iterate_over: 'unfiltered_compile_flags' + flag: '%{unfiltered_compile_flags}' + } + } + } + + feature { + name: 'compiler_output_flags' + flag_set { + action: 'assemble' + flag_group { + expand_if_all_available: 'output_file' + expand_if_none_available: 'output_assembly_file' + expand_if_none_available: 'output_preprocess_file' + flag: '/Fo%{output_file}' + flag: '/Zi' + } + } + flag_set { + action: 'preprocess-assemble' + action: 'c-compile' + action: 'c++-compile' + action: 'c++-header-parsing' + action: 'c++-module-compile' + action: 'c++-module-codegen' + flag_group { + expand_if_all_available: 'output_file' + expand_if_none_available: 'output_assembly_file' + expand_if_none_available: 'output_preprocess_file' + flag: '/Fo%{output_file}' + } + flag_group { + expand_if_all_available: 'output_file' + expand_if_all_available: 'output_assembly_file' + flag: '/Fa%{output_file}' + } + flag_group { + expand_if_all_available: 'output_file' + expand_if_all_available: 'output_preprocess_file' + flag: '/P' + flag: '/Fi%{output_file}' + } + } + } + + feature { + name: 'compiler_input_flags' + flag_set { + action: 'assemble' + action: 'preprocess-assemble' + action: 'c-compile' + action: 'c++-compile' + action: 'c++-header-parsing' + action: 'c++-module-compile' + action: 'c++-module-codegen' + flag_group { + expand_if_all_available: 'source_file' + flag: '/c' + flag: '%{source_file}' + } + } + } + + feature { + name : 'def_file', + flag_set { + expand_if_all_available: 'def_file_path' + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "/DEF:%{def_file_path}" + # We can specify a different DLL name in DEF file, /ignore:4070 suppresses + # the warning message about DLL name doesn't match the default one. + # See https://msdn.microsoft.com/en-us/library/sfkk2fz7.aspx + flag: "/ignore:4070" + } + } + } + + feature { + name: 'windows_export_all_symbols' + } + + feature { + name: 'no_windows_export_all_symbols' + } + + linking_mode_flags { mode: DYNAMIC } +}
\ No newline at end of file diff --git a/third_party/toolchains/gpus/cuda/BUILD b/third_party/toolchains/gpus/cuda/BUILD index cfc6930851..f59e025019 100644 --- a/third_party/toolchains/gpus/cuda/BUILD +++ b/third_party/toolchains/gpus/cuda/BUILD @@ -133,6 +133,15 @@ cc_library( ) cc_library( + name = "cudnn_header", + includes = [ + ".", + "cuda/include", + ], + visibility = ["//visibility:public"], +) + +cc_library( name = "cufft", srcs = ["cuda/lib/libcufft.so.9.0"], data = ["cuda/lib/libcufft.so.9.0"], @@ -1191,33 +1200,10 @@ if [ -d "$(@D)/extras" ]; then rm $(@D)/extras -drf; fi && if [ -d "$(@D)/includ genrule( name = "cuda-nvvm", outs = [ - "cuda/nvvm/bin/cicc", - "cuda/nvvm/include/nvvm.h", - "cuda/nvvm/lib64/libnvvm.so", - "cuda/nvvm/lib64/libnvvm.so.3", - "cuda/nvvm/lib64/libnvvm.so.3.2.0", "cuda/nvvm/libdevice/libdevice.10.bc", - "cuda/nvvm/libnvvm-samples/CMakeLists.txt", - "cuda/nvvm/libnvvm-samples/README.txt", - "cuda/nvvm/libnvvm-samples/build.bat", - "cuda/nvvm/libnvvm-samples/build.sh", - "cuda/nvvm/libnvvm-samples/common/include/DDSWriter.h", - "cuda/nvvm/libnvvm-samples/common/include/drvapi_error_string.h", - "cuda/nvvm/libnvvm-samples/cuda-c-linking/CMakeLists.txt", - "cuda/nvvm/libnvvm-samples/cuda-c-linking/README.txt", - "cuda/nvvm/libnvvm-samples/cuda-c-linking/cuda-c-linking.cpp", - "cuda/nvvm/libnvvm-samples/cuda-c-linking/math-funcs.cu", - "cuda/nvvm/libnvvm-samples/ptxgen/CMakeLists.txt", - "cuda/nvvm/libnvvm-samples/ptxgen/README.txt", - "cuda/nvvm/libnvvm-samples/ptxgen/ptxgen.c", - "cuda/nvvm/libnvvm-samples/simple/CMakeLists.txt", - "cuda/nvvm/libnvvm-samples/simple/README.txt", - "cuda/nvvm/libnvvm-samples/simple/simple-gpu.ll", - "cuda/nvvm/libnvvm-samples/simple/simple-gpu64.ll", - "cuda/nvvm/libnvvm-samples/simple/simple.c", ], cmd = """ -if [ -d "$(@D)/extras" ]; then rm $(@D)/extras -drf; fi && if [ -d "$(@D)/include" ]; then rm $(@D)/include -drf; fi && if [ -d "$(@D)/lib" ]; then rm $(@D)/lib -drf; fi && if [ -d "$(@D)/nvvm" ]; then rm $(@D)/nvvm -drf; fi && cp "/usr/local/cuda-9.0/nvvm/bin/cicc" "$(@D)/cuda/nvvm/bin/cicc" && cp "/usr/local/cuda-9.0/nvvm/include/nvvm.h" "$(@D)/cuda/nvvm/include/nvvm.h" && cp "/usr/local/cuda-9.0/nvvm/lib64/libnvvm.so" "$(@D)/cuda/nvvm/lib64/libnvvm.so" && cp "/usr/local/cuda-9.0/nvvm/lib64/libnvvm.so.3" "$(@D)/cuda/nvvm/lib64/libnvvm.so.3" && cp "/usr/local/cuda-9.0/nvvm/lib64/libnvvm.so.3.2.0" "$(@D)/cuda/nvvm/lib64/libnvvm.so.3.2.0" && cp "/usr/local/cuda-9.0/nvvm/libdevice/libdevice.10.bc" "$(@D)/cuda/nvvm/libdevice/libdevice.10.bc" && cp "/usr/local/cuda-9.0/nvvm/libnvvm-samples/CMakeLists.txt" "$(@D)/cuda/nvvm/libnvvm-samples/CMakeLists.txt" && cp "/usr/local/cuda-9.0/nvvm/libnvvm-samples/README.txt" "$(@D)/cuda/nvvm/libnvvm-samples/README.txt" && cp "/usr/local/cuda-9.0/nvvm/libnvvm-samples/build.bat" "$(@D)/cuda/nvvm/libnvvm-samples/build.bat" && cp "/usr/local/cuda-9.0/nvvm/libnvvm-samples/build.sh" "$(@D)/cuda/nvvm/libnvvm-samples/build.sh" && cp "/usr/local/cuda-9.0/nvvm/libnvvm-samples/common/include/DDSWriter.h" "$(@D)/cuda/nvvm/libnvvm-samples/common/include/DDSWriter.h" && cp "/usr/local/cuda-9.0/nvvm/libnvvm-samples/common/include/drvapi_error_string.h" "$(@D)/cuda/nvvm/libnvvm-samples/common/include/drvapi_error_string.h" && cp "/usr/local/cuda-9.0/nvvm/libnvvm-samples/cuda-c-linking/CMakeLists.txt" "$(@D)/cuda/nvvm/libnvvm-samples/cuda-c-linking/CMakeLists.txt" && cp "/usr/local/cuda-9.0/nvvm/libnvvm-samples/cuda-c-linking/README.txt" "$(@D)/cuda/nvvm/libnvvm-samples/cuda-c-linking/README.txt" && cp "/usr/local/cuda-9.0/nvvm/libnvvm-samples/cuda-c-linking/cuda-c-linking.cpp" "$(@D)/cuda/nvvm/libnvvm-samples/cuda-c-linking/cuda-c-linking.cpp" && cp "/usr/local/cuda-9.0/nvvm/libnvvm-samples/cuda-c-linking/math-funcs.cu" "$(@D)/cuda/nvvm/libnvvm-samples/cuda-c-linking/math-funcs.cu" && cp "/usr/local/cuda-9.0/nvvm/libnvvm-samples/ptxgen/CMakeLists.txt" "$(@D)/cuda/nvvm/libnvvm-samples/ptxgen/CMakeLists.txt" && cp "/usr/local/cuda-9.0/nvvm/libnvvm-samples/ptxgen/README.txt" "$(@D)/cuda/nvvm/libnvvm-samples/ptxgen/README.txt" && cp "/usr/local/cuda-9.0/nvvm/libnvvm-samples/ptxgen/ptxgen.c" "$(@D)/cuda/nvvm/libnvvm-samples/ptxgen/ptxgen.c" && cp "/usr/local/cuda-9.0/nvvm/libnvvm-samples/simple/CMakeLists.txt" "$(@D)/cuda/nvvm/libnvvm-samples/simple/CMakeLists.txt" && cp "/usr/local/cuda-9.0/nvvm/libnvvm-samples/simple/README.txt" "$(@D)/cuda/nvvm/libnvvm-samples/simple/README.txt" && cp "/usr/local/cuda-9.0/nvvm/libnvvm-samples/simple/simple-gpu.ll" "$(@D)/cuda/nvvm/libnvvm-samples/simple/simple-gpu.ll" && cp "/usr/local/cuda-9.0/nvvm/libnvvm-samples/simple/simple-gpu64.ll" "$(@D)/cuda/nvvm/libnvvm-samples/simple/simple-gpu64.ll" && cp "/usr/local/cuda-9.0/nvvm/libnvvm-samples/simple/simple.c" "$(@D)/cuda/nvvm/libnvvm-samples/simple/simple.c" +if [ -d "$(@D)/extras" ]; then rm $(@D)/extras -drf; fi && if [ -d "$(@D)/include" ]; then rm $(@D)/include -drf; fi && if [ -d "$(@D)/lib" ]; then rm $(@D)/lib -drf; fi && if [ -d "$(@D)/nvvm" ]; then rm $(@D)/nvvm -drf; fi && cp "/usr/local/cuda-9.0/nvvm/libdevice/libdevice.10.bc" "$(@D)//libdevice.10.bc" """, ) @@ -1272,7 +1258,7 @@ genrule( "cuda/lib/libcupti.so.9.0", ], cmd = """ -if [ -d "$(@D)/extras" ]; then rm $(@D)/extras -drf; fi && if [ -d "$(@D)/include" ]; then rm $(@D)/include -drf; fi && if [ -d "$(@D)/lib" ]; then rm $(@D)/lib -drf; fi && if [ -d "$(@D)/nvvm" ]; then rm $(@D)/nvvm -drf; fi && cp "/usr/local/cuda-9.0/targets/x86_64-linux/lib/stubs/libcuda.so" "$(@D)/cuda/lib/libcuda.so" && cp "/usr/local/cuda-9.0/targets/x86_64-linux/lib/libcudart.so.9.0.176" "$(@D)/cuda/lib/libcudart.so.9.0" && cp "/usr/local/cuda-9.0/targets/x86_64-linux/lib/libcudart_static.a" "$(@D)/cuda/lib/libcudart_static.a" && cp "/usr/local/cuda-9.0/targets/x86_64-linux/lib/libcublas.so.9.0.282" "$(@D)/cuda/lib/libcublas.so.9.0" && cp "/usr/local/cuda-9.0/targets/x86_64-linux/lib/libcusolver.so.9.0.176" "$(@D)/cuda/lib/libcusolver.so.9.0" && cp "/usr/local/cuda-9.0/targets/x86_64-linux/lib/libcurand.so.9.0.176" "$(@D)/cuda/lib/libcurand.so.9.0" && cp "/usr/local/cuda-9.0/targets/x86_64-linux/lib/libcufft.so.9.0.176" "$(@D)/cuda/lib/libcufft.so.9.0" && cp "/usr/lib/x86_64-linux-gnu/libcudnn.so.7.0.5" "$(@D)/cuda/lib/libcudnn.so.7" && cp "/usr/local/cuda-9.0/extras/CUPTI/lib64/libcupti.so.9.0.176" "$(@D)/cuda/lib/libcupti.so.9.0" +if [ -d "$(@D)/extras" ]; then rm $(@D)/extras -drf; fi && if [ -d "$(@D)/include" ]; then rm $(@D)/include -drf; fi && if [ -d "$(@D)/lib" ]; then rm $(@D)/lib -drf; fi && if [ -d "$(@D)/nvvm" ]; then rm $(@D)/nvvm -drf; fi && cp "/usr/local/cuda-9.0/targets/x86_64-linux/lib/stubs/libcuda.so" "$(@D)/cuda/lib/libcuda.so" && cp "/usr/local/cuda-9.0/targets/x86_64-linux/lib/libcudart.so.9.0.176" "$(@D)/cuda/lib/libcudart.so.9.0" && cp "/usr/local/cuda-9.0/targets/x86_64-linux/lib/libcudart_static.a" "$(@D)/cuda/lib/libcudart_static.a" && cp "/usr/local/cuda-9.0/targets/x86_64-linux/lib/libcublas.so.9.0.480" "$(@D)/cuda/lib/libcublas.so.9.0" && cp "/usr/local/cuda-9.0/targets/x86_64-linux/lib/libcusolver.so.9.0.176" "$(@D)/cuda/lib/libcusolver.so.9.0" && cp "/usr/local/cuda-9.0/targets/x86_64-linux/lib/libcurand.so.9.0.176" "$(@D)/cuda/lib/libcurand.so.9.0" && cp "/usr/local/cuda-9.0/targets/x86_64-linux/lib/libcufft.so.9.0.176" "$(@D)/cuda/lib/libcufft.so.9.0" && cp "/usr/lib/x86_64-linux-gnu/libcudnn.so.7.2.1" "$(@D)/cuda/lib/libcudnn.so.7" && cp "/usr/local/cuda-9.0/extras/CUPTI/lib64/libcupti.so.9.0.176" "$(@D)/cuda/lib/libcupti.so.9.0" """, ) diff --git a/third_party/toolchains/gpus/cuda/build_defs.bzl b/third_party/toolchains/gpus/cuda/build_defs.bzl index badaf43019..9210bfe016 100644 --- a/third_party/toolchains/gpus/cuda/build_defs.bzl +++ b/third_party/toolchains/gpus/cuda/build_defs.bzl @@ -2,6 +2,7 @@ # execution service. # DO NOT EDIT: automatically generated file +# Macros for building CUDA code. def if_cuda(if_true, if_false = []): """Shorthand for select()'ing on whether we're building with CUDA. @@ -12,15 +13,13 @@ def if_cuda(if_true, if_false = []): return select({ "@local_config_cuda//cuda:using_nvcc": if_true, "@local_config_cuda//cuda:using_clang": if_true, - "//conditions:default": if_false + "//conditions:default": if_false, }) - def cuda_default_copts(): """Default options for all CUDA compilations.""" return if_cuda(["-x", "cuda", "-DGOOGLE_CUDA=1"] + ["--cuda-gpu-arch=sm_30"]) - def cuda_is_configured(): """Returns true if CUDA was enabled during the configure process.""" return True @@ -32,6 +31,5 @@ def if_cuda_is_configured(x): --config=cuda. Used to allow non-CUDA code to depend on CUDA libraries. """ if cuda_is_configured(): - return x + return x return [] - diff --git a/third_party/toolchains/gpus/cuda/cuda/cuda_config.h b/third_party/toolchains/gpus/cuda/cuda/cuda_config.h index f6662274cc..7cdaf144ad 100644 --- a/third_party/toolchains/gpus/cuda/cuda/cuda_config.h +++ b/third_party/toolchains/gpus/cuda/cuda/cuda_config.h @@ -19,9 +19,9 @@ limitations under the License. #define TF_CUDA_CAPABILITIES CudaVersion("3.0") -#define TF_CUDA_VERSION "8.0" -#define TF_CUDNN_VERSION "5" +#define TF_CUDA_VERSION "9.0" +#define TF_CUDNN_VERSION "7" -#define TF_CUDA_TOOLKIT_PATH "/usr/local/cuda-8.0" +#define TF_CUDA_TOOLKIT_PATH "/usr/local/cuda-9.0" #endif // CUDA_CUDA_CONFIG_H_ diff --git a/third_party/toolchains/gpus/py/BUILD b/third_party/toolchains/gpus/py/BUILD index 2d5ace93ff..1235988abb 100644 --- a/third_party/toolchains/gpus/py/BUILD +++ b/third_party/toolchains/gpus/py/BUILD @@ -6,18 +6,24 @@ licenses(["restricted"]) package(default_visibility = ["//visibility:public"]) +# To build Python C/C++ extension on Windows, we need to link to python import library pythonXY.lib +# See https://docs.python.org/3/extending/windows.html +cc_import( + name = "python_lib", + interface_library = select({ + ":windows": ":python_import_lib", + # A placeholder for Unix platforms which makes --no_build happy. + "//conditions:default": "not-existing.lib", + }), + system_provided = 1, +) + cc_library( name = "python_headers", hdrs = [":python_include"], - data = select({ - ":windows": [":python_import_lib"], - "//conditions:default": [], - }), includes = ["python_include"], - linkopts = select({ - # TODO(pcloudy): Ideally, this should just go into deps after resolving - # https://github.com/bazelbuild/bazel/issues/3237, - ":windows": ["$(locations :python_import_lib)"], + deps = select({ + ":windows": [":python_lib"], "//conditions:default": [], }), ) diff --git a/third_party/toolchains/preconfig/ubuntu14.04/cuda9.0-cudnn7/WORKSPACE b/third_party/toolchains/preconfig/ubuntu14.04/cuda9.0-cudnn7/WORKSPACE new file mode 100644 index 0000000000..b61f572d6d --- /dev/null +++ b/third_party/toolchains/preconfig/ubuntu14.04/cuda9.0-cudnn7/WORKSPACE @@ -0,0 +1,2 @@ +# DO NOT EDIT: automatically generated WORKSPACE file for cuda_configure rule +workspace(name = "local_config_cuda") diff --git a/third_party/toolchains/preconfig/ubuntu14.04/cuda9.0-cudnn7/cuda/BUILD b/third_party/toolchains/preconfig/ubuntu14.04/cuda9.0-cudnn7/cuda/BUILD new file mode 100755 index 0000000000..2d3e41127d --- /dev/null +++ b/third_party/toolchains/preconfig/ubuntu14.04/cuda9.0-cudnn7/cuda/BUILD @@ -0,0 +1,1268 @@ +licenses(["restricted"]) # MPL2, portions GPL v3, LGPL v3, BSD-like + +package(default_visibility = ["//visibility:public"]) + +config_setting( + name = "using_nvcc", + values = { + "define": "using_cuda_nvcc=true", + }, +) + +config_setting( + name = "using_clang", + values = { + "define": "using_cuda_clang=true", + }, +) + +# Equivalent to using_clang && -c opt. +config_setting( + name = "using_clang_opt", + values = { + "define": "using_cuda_clang=true", + "compilation_mode": "opt", + }, +) + +config_setting( + name = "darwin", + values = {"cpu": "darwin"}, + visibility = ["//visibility:public"], +) + +config_setting( + name = "freebsd", + values = {"cpu": "freebsd"}, + visibility = ["//visibility:public"], +) + +cc_library( + name = "cuda_headers", + hdrs = [ + "cuda/cuda_config.h", + ":cuda-include", + ":cudnn-include", + ], + includes = [ + ".", + "cuda/include", + "cuda/include/crt", + ], + visibility = ["//visibility:public"], +) + +cc_library( + name = "cudart_static", + srcs = ["cuda/lib/libcudart_static.a"], + includes = [ + ".", + "cuda/include", + ], + linkopts = select({ + ":freebsd": [], + "//conditions:default": ["-ldl"], + }) + [ + "-lpthread", + "-lrt", + ], + visibility = ["//visibility:public"], +) + +cc_library( + name = "cuda_driver", + srcs = ["cuda/lib/libcuda.so"], + includes = [ + ".", + "cuda/include", + ], + visibility = ["//visibility:public"], +) + +cc_library( + name = "cudart", + srcs = ["cuda/lib/libcudart.so.9.0"], + data = ["cuda/lib/libcudart.so.9.0"], + includes = [ + ".", + "cuda/include", + ], + linkstatic = 1, + visibility = ["//visibility:public"], +) + +cc_library( + name = "cublas", + srcs = ["cuda/lib/libcublas.so.9.0"], + data = ["cuda/lib/libcublas.so.9.0"], + includes = [ + ".", + "cuda/include", + ], + linkstatic = 1, + visibility = ["//visibility:public"], +) + +cc_library( + name = "cusolver", + srcs = ["cuda/lib/libcusolver.so.9.0"], + data = ["cuda/lib/libcusolver.so.9.0"], + includes = [ + ".", + "cuda/include", + ], + linkopts = ["-lgomp"], + linkstatic = 1, + visibility = ["//visibility:public"], +) + +cc_library( + name = "cudnn", + srcs = ["cuda/lib/libcudnn.so.7"], + data = ["cuda/lib/libcudnn.so.7"], + includes = [ + ".", + "cuda/include", + ], + linkstatic = 1, + visibility = ["//visibility:public"], +) + +cc_library( + name = "cudnn_header", + includes = [ + ".", + "cuda/include", + ], + visibility = ["//visibility:public"], +) + +cc_library( + name = "cufft", + srcs = ["cuda/lib/libcufft.so.9.0"], + data = ["cuda/lib/libcufft.so.9.0"], + includes = [ + ".", + "cuda/include", + ], + linkstatic = 1, + visibility = ["//visibility:public"], +) + +cc_library( + name = "curand", + srcs = ["cuda/lib/libcurand.so.9.0"], + data = ["cuda/lib/libcurand.so.9.0"], + includes = [ + ".", + "cuda/include", + ], + linkstatic = 1, + visibility = ["//visibility:public"], +) + +cc_library( + name = "cuda", + visibility = ["//visibility:public"], + deps = [ + ":cublas", + ":cuda_headers", + ":cudart", + ":cudnn", + ":cufft", + ":curand", + ], +) + +cc_library( + name = "cupti_headers", + hdrs = [ + "cuda/cuda_config.h", + ":cuda-extras", + ], + includes = [ + ".", + "cuda/extras/CUPTI/include/", + ], + visibility = ["//visibility:public"], +) + +cc_library( + name = "cupti_dsos", + data = ["cuda/lib/libcupti.so.9.0"], + includes = [ + ".", + "cuda/include", + ], + visibility = ["//visibility:public"], +) + +cc_library( + name = "libdevice_root", + data = [":cuda-nvvm"], + visibility = ["//visibility:public"], +) + +genrule( + name = "cuda-include", + outs = [ + "cuda/include/CL/cl.h", + "cuda/include/CL/cl.hpp", + "cuda/include/CL/cl_egl.h", + "cuda/include/CL/cl_ext.h", + "cuda/include/CL/cl_gl.h", + "cuda/include/CL/cl_gl_ext.h", + "cuda/include/CL/cl_platform.h", + "cuda/include/CL/opencl.h", + "cuda/include/builtin_types.h", + "cuda/include/channel_descriptor.h", + "cuda/include/common_functions.h", + "cuda/include/cooperative_groups.h", + "cuda/include/cooperative_groups_helpers.h", + "cuda/include/crt/common_functions.h", + "cuda/include/crt/device_double_functions.h", + "cuda/include/crt/device_double_functions.hpp", + "cuda/include/crt/device_functions.h", + "cuda/include/crt/device_functions.hpp", + "cuda/include/crt/func_macro.h", + "cuda/include/crt/host_config.h", + "cuda/include/crt/host_defines.h", + "cuda/include/crt/host_runtime.h", + "cuda/include/crt/math_functions.h", + "cuda/include/crt/math_functions.hpp", + "cuda/include/crt/mma.h", + "cuda/include/crt/mma.hpp", + "cuda/include/crt/nvfunctional", + "cuda/include/crt/sm_70_rt.h", + "cuda/include/crt/sm_70_rt.hpp", + "cuda/include/crt/storage_class.h", + "cuda/include/cuComplex.h", + "cuda/include/cublas.h", + "cuda/include/cublasXt.h", + "cuda/include/cublas_api.h", + "cuda/include/cublas_v2.h", + "cuda/include/cuda.h", + "cuda/include/cudaEGL.h", + "cuda/include/cudaGL.h", + "cuda/include/cudaProfiler.h", + "cuda/include/cudaVDPAU.h", + "cuda/include/cuda_device_runtime_api.h", + "cuda/include/cuda_fp16.h", + "cuda/include/cuda_fp16.hpp", + "cuda/include/cuda_gl_interop.h", + "cuda/include/cuda_occupancy.h", + "cuda/include/cuda_profiler_api.h", + "cuda/include/cuda_runtime.h", + "cuda/include/cuda_runtime_api.h", + "cuda/include/cuda_surface_types.h", + "cuda/include/cuda_texture_types.h", + "cuda/include/cuda_vdpau_interop.h", + "cuda/include/cudalibxt.h", + "cuda/include/cufft.h", + "cuda/include/cufftXt.h", + "cuda/include/cufftw.h", + "cuda/include/curand.h", + "cuda/include/curand_discrete.h", + "cuda/include/curand_discrete2.h", + "cuda/include/curand_globals.h", + "cuda/include/curand_kernel.h", + "cuda/include/curand_lognormal.h", + "cuda/include/curand_mrg32k3a.h", + "cuda/include/curand_mtgp32.h", + "cuda/include/curand_mtgp32_host.h", + "cuda/include/curand_mtgp32_kernel.h", + "cuda/include/curand_mtgp32dc_p_11213.h", + "cuda/include/curand_normal.h", + "cuda/include/curand_normal_static.h", + "cuda/include/curand_philox4x32_x.h", + "cuda/include/curand_poisson.h", + "cuda/include/curand_precalc.h", + "cuda/include/curand_uniform.h", + "cuda/include/cusolverDn.h", + "cuda/include/cusolverRf.h", + "cuda/include/cusolverSp.h", + "cuda/include/cusolverSp_LOWLEVEL_PREVIEW.h", + "cuda/include/cusolver_common.h", + "cuda/include/cusparse.h", + "cuda/include/cusparse_v2.h", + "cuda/include/device_atomic_functions.h", + "cuda/include/device_atomic_functions.hpp", + "cuda/include/device_double_functions.h", + "cuda/include/device_double_functions.hpp", + "cuda/include/device_functions.h", + "cuda/include/device_functions.hpp", + "cuda/include/device_functions_decls.h", + "cuda/include/device_launch_parameters.h", + "cuda/include/device_types.h", + "cuda/include/driver_functions.h", + "cuda/include/driver_types.h", + "cuda/include/dynlink_cuda.h", + "cuda/include/dynlink_cuda_cuda.h", + "cuda/include/dynlink_cuviddec.h", + "cuda/include/dynlink_nvcuvid.h", + "cuda/include/fatBinaryCtl.h", + "cuda/include/fatbinary.h", + "cuda/include/host_config.h", + "cuda/include/host_defines.h", + "cuda/include/library_types.h", + "cuda/include/math_constants.h", + "cuda/include/math_functions.h", + "cuda/include/math_functions.hpp", + "cuda/include/math_functions_dbl_ptx3.h", + "cuda/include/math_functions_dbl_ptx3.hpp", + "cuda/include/mma.h", + "cuda/include/npp.h", + "cuda/include/nppcore.h", + "cuda/include/nppdefs.h", + "cuda/include/nppi.h", + "cuda/include/nppi_arithmetic_and_logical_operations.h", + "cuda/include/nppi_color_conversion.h", + "cuda/include/nppi_compression_functions.h", + "cuda/include/nppi_computer_vision.h", + "cuda/include/nppi_data_exchange_and_initialization.h", + "cuda/include/nppi_filtering_functions.h", + "cuda/include/nppi_geometry_transforms.h", + "cuda/include/nppi_linear_transforms.h", + "cuda/include/nppi_morphological_operations.h", + "cuda/include/nppi_statistics_functions.h", + "cuda/include/nppi_support_functions.h", + "cuda/include/nppi_threshold_and_compare_operations.h", + "cuda/include/npps.h", + "cuda/include/npps_arithmetic_and_logical_operations.h", + "cuda/include/npps_conversion_functions.h", + "cuda/include/npps_filtering_functions.h", + "cuda/include/npps_initialization.h", + "cuda/include/npps_statistics_functions.h", + "cuda/include/npps_support_functions.h", + "cuda/include/nppversion.h", + "cuda/include/nvToolsExt.h", + "cuda/include/nvToolsExtCuda.h", + "cuda/include/nvToolsExtCudaRt.h", + "cuda/include/nvToolsExtMeta.h", + "cuda/include/nvToolsExtSync.h", + "cuda/include/nvblas.h", + "cuda/include/nvfunctional", + "cuda/include/nvgraph.h", + "cuda/include/nvml.h", + "cuda/include/nvrtc.h", + "cuda/include/sm_20_atomic_functions.h", + "cuda/include/sm_20_atomic_functions.hpp", + "cuda/include/sm_20_intrinsics.h", + "cuda/include/sm_20_intrinsics.hpp", + "cuda/include/sm_30_intrinsics.h", + "cuda/include/sm_30_intrinsics.hpp", + "cuda/include/sm_32_atomic_functions.h", + "cuda/include/sm_32_atomic_functions.hpp", + "cuda/include/sm_32_intrinsics.h", + "cuda/include/sm_32_intrinsics.hpp", + "cuda/include/sm_35_atomic_functions.h", + "cuda/include/sm_35_intrinsics.h", + "cuda/include/sm_60_atomic_functions.h", + "cuda/include/sm_60_atomic_functions.hpp", + "cuda/include/sm_61_intrinsics.h", + "cuda/include/sm_61_intrinsics.hpp", + "cuda/include/sobol_direction_vectors.h", + "cuda/include/surface_functions.h", + "cuda/include/surface_functions.hpp", + "cuda/include/surface_indirect_functions.h", + "cuda/include/surface_indirect_functions.hpp", + "cuda/include/surface_types.h", + "cuda/include/texture_fetch_functions.h", + "cuda/include/texture_fetch_functions.hpp", + "cuda/include/texture_indirect_functions.h", + "cuda/include/texture_indirect_functions.hpp", + "cuda/include/texture_types.h", + "cuda/include/thrust/adjacent_difference.h", + "cuda/include/thrust/advance.h", + "cuda/include/thrust/binary_search.h", + "cuda/include/thrust/complex.h", + "cuda/include/thrust/copy.h", + "cuda/include/thrust/count.h", + "cuda/include/thrust/detail/adjacent_difference.inl", + "cuda/include/thrust/detail/advance.inl", + "cuda/include/thrust/detail/allocator/allocator_traits.h", + "cuda/include/thrust/detail/allocator/allocator_traits.inl", + "cuda/include/thrust/detail/allocator/copy_construct_range.h", + "cuda/include/thrust/detail/allocator/copy_construct_range.inl", + "cuda/include/thrust/detail/allocator/default_construct_range.h", + "cuda/include/thrust/detail/allocator/default_construct_range.inl", + "cuda/include/thrust/detail/allocator/destroy_range.h", + "cuda/include/thrust/detail/allocator/destroy_range.inl", + "cuda/include/thrust/detail/allocator/fill_construct_range.h", + "cuda/include/thrust/detail/allocator/fill_construct_range.inl", + "cuda/include/thrust/detail/allocator/malloc_allocator.h", + "cuda/include/thrust/detail/allocator/malloc_allocator.inl", + "cuda/include/thrust/detail/allocator/no_throw_allocator.h", + "cuda/include/thrust/detail/allocator/tagged_allocator.h", + "cuda/include/thrust/detail/allocator/tagged_allocator.inl", + "cuda/include/thrust/detail/allocator/temporary_allocator.h", + "cuda/include/thrust/detail/allocator/temporary_allocator.inl", + "cuda/include/thrust/detail/binary_search.inl", + "cuda/include/thrust/detail/complex/arithmetic.h", + "cuda/include/thrust/detail/complex/c99math.h", + "cuda/include/thrust/detail/complex/catrig.h", + "cuda/include/thrust/detail/complex/catrigf.h", + "cuda/include/thrust/detail/complex/ccosh.h", + "cuda/include/thrust/detail/complex/ccoshf.h", + "cuda/include/thrust/detail/complex/cexp.h", + "cuda/include/thrust/detail/complex/cexpf.h", + "cuda/include/thrust/detail/complex/clog.h", + "cuda/include/thrust/detail/complex/clogf.h", + "cuda/include/thrust/detail/complex/complex.inl", + "cuda/include/thrust/detail/complex/cpow.h", + "cuda/include/thrust/detail/complex/cpowf.h", + "cuda/include/thrust/detail/complex/cproj.h", + "cuda/include/thrust/detail/complex/csinh.h", + "cuda/include/thrust/detail/complex/csinhf.h", + "cuda/include/thrust/detail/complex/csqrt.h", + "cuda/include/thrust/detail/complex/csqrtf.h", + "cuda/include/thrust/detail/complex/ctanh.h", + "cuda/include/thrust/detail/complex/ctanhf.h", + "cuda/include/thrust/detail/complex/math_private.h", + "cuda/include/thrust/detail/complex/stream.h", + "cuda/include/thrust/detail/config.h", + "cuda/include/thrust/detail/config/compiler.h", + "cuda/include/thrust/detail/config/compiler_fence.h", + "cuda/include/thrust/detail/config/config.h", + "cuda/include/thrust/detail/config/debug.h", + "cuda/include/thrust/detail/config/device_system.h", + "cuda/include/thrust/detail/config/exec_check_disable.h", + "cuda/include/thrust/detail/config/forceinline.h", + "cuda/include/thrust/detail/config/global_workarounds.h", + "cuda/include/thrust/detail/config/host_device.h", + "cuda/include/thrust/detail/config/host_system.h", + "cuda/include/thrust/detail/config/simple_defines.h", + "cuda/include/thrust/detail/contiguous_storage.h", + "cuda/include/thrust/detail/contiguous_storage.inl", + "cuda/include/thrust/detail/copy.h", + "cuda/include/thrust/detail/copy.inl", + "cuda/include/thrust/detail/copy_if.h", + "cuda/include/thrust/detail/copy_if.inl", + "cuda/include/thrust/detail/count.inl", + "cuda/include/thrust/detail/cstdint.h", + "cuda/include/thrust/detail/device_delete.inl", + "cuda/include/thrust/detail/device_free.inl", + "cuda/include/thrust/detail/device_malloc.inl", + "cuda/include/thrust/detail/device_new.inl", + "cuda/include/thrust/detail/device_ptr.inl", + "cuda/include/thrust/detail/device_reference.inl", + "cuda/include/thrust/detail/device_vector.inl", + "cuda/include/thrust/detail/dispatch/is_trivial_copy.h", + "cuda/include/thrust/detail/distance.inl", + "cuda/include/thrust/detail/equal.inl", + "cuda/include/thrust/detail/execute_with_allocator.h", + "cuda/include/thrust/detail/execution_policy.h", + "cuda/include/thrust/detail/extrema.inl", + "cuda/include/thrust/detail/fill.inl", + "cuda/include/thrust/detail/find.inl", + "cuda/include/thrust/detail/for_each.inl", + "cuda/include/thrust/detail/function.h", + "cuda/include/thrust/detail/functional.inl", + "cuda/include/thrust/detail/functional/actor.h", + "cuda/include/thrust/detail/functional/actor.inl", + "cuda/include/thrust/detail/functional/argument.h", + "cuda/include/thrust/detail/functional/composite.h", + "cuda/include/thrust/detail/functional/operators.h", + "cuda/include/thrust/detail/functional/operators/arithmetic_operators.h", + "cuda/include/thrust/detail/functional/operators/assignment_operator.h", + "cuda/include/thrust/detail/functional/operators/bitwise_operators.h", + "cuda/include/thrust/detail/functional/operators/compound_assignment_operators.h", + "cuda/include/thrust/detail/functional/operators/logical_operators.h", + "cuda/include/thrust/detail/functional/operators/operator_adaptors.h", + "cuda/include/thrust/detail/functional/operators/relational_operators.h", + "cuda/include/thrust/detail/functional/placeholder.h", + "cuda/include/thrust/detail/functional/value.h", + "cuda/include/thrust/detail/gather.inl", + "cuda/include/thrust/detail/generate.inl", + "cuda/include/thrust/detail/get_iterator_value.h", + "cuda/include/thrust/detail/host_vector.inl", + "cuda/include/thrust/detail/inner_product.inl", + "cuda/include/thrust/detail/integer_math.h", + "cuda/include/thrust/detail/integer_traits.h", + "cuda/include/thrust/detail/internal_functional.h", + "cuda/include/thrust/detail/logical.inl", + "cuda/include/thrust/detail/malloc_and_free.h", + "cuda/include/thrust/detail/merge.inl", + "cuda/include/thrust/detail/minmax.h", + "cuda/include/thrust/detail/mismatch.inl", + "cuda/include/thrust/detail/mpl/math.h", + "cuda/include/thrust/detail/numeric_traits.h", + "cuda/include/thrust/detail/overlapped_copy.h", + "cuda/include/thrust/detail/pair.inl", + "cuda/include/thrust/detail/partition.inl", + "cuda/include/thrust/detail/pointer.h", + "cuda/include/thrust/detail/pointer.inl", + "cuda/include/thrust/detail/range/head_flags.h", + "cuda/include/thrust/detail/range/tail_flags.h", + "cuda/include/thrust/detail/raw_pointer_cast.h", + "cuda/include/thrust/detail/raw_reference_cast.h", + "cuda/include/thrust/detail/reduce.inl", + "cuda/include/thrust/detail/reference.h", + "cuda/include/thrust/detail/reference.inl", + "cuda/include/thrust/detail/reference_forward_declaration.h", + "cuda/include/thrust/detail/remove.inl", + "cuda/include/thrust/detail/replace.inl", + "cuda/include/thrust/detail/reverse.inl", + "cuda/include/thrust/detail/scan.inl", + "cuda/include/thrust/detail/scatter.inl", + "cuda/include/thrust/detail/seq.h", + "cuda/include/thrust/detail/sequence.inl", + "cuda/include/thrust/detail/set_operations.inl", + "cuda/include/thrust/detail/sort.inl", + "cuda/include/thrust/detail/static_assert.h", + "cuda/include/thrust/detail/static_map.h", + "cuda/include/thrust/detail/swap.h", + "cuda/include/thrust/detail/swap.inl", + "cuda/include/thrust/detail/swap_ranges.inl", + "cuda/include/thrust/detail/tabulate.inl", + "cuda/include/thrust/detail/temporary_array.h", + "cuda/include/thrust/detail/temporary_array.inl", + "cuda/include/thrust/detail/temporary_buffer.h", + "cuda/include/thrust/detail/transform.inl", + "cuda/include/thrust/detail/transform_reduce.inl", + "cuda/include/thrust/detail/transform_scan.inl", + "cuda/include/thrust/detail/trivial_sequence.h", + "cuda/include/thrust/detail/tuple.inl", + "cuda/include/thrust/detail/tuple_meta_transform.h", + "cuda/include/thrust/detail/tuple_transform.h", + "cuda/include/thrust/detail/type_traits.h", + "cuda/include/thrust/detail/type_traits/algorithm/intermediate_type_from_function_and_iterators.h", + "cuda/include/thrust/detail/type_traits/function_traits.h", + "cuda/include/thrust/detail/type_traits/has_member_function.h", + "cuda/include/thrust/detail/type_traits/has_nested_type.h", + "cuda/include/thrust/detail/type_traits/has_trivial_assign.h", + "cuda/include/thrust/detail/type_traits/is_call_possible.h", + "cuda/include/thrust/detail/type_traits/is_metafunction_defined.h", + "cuda/include/thrust/detail/type_traits/iterator/is_discard_iterator.h", + "cuda/include/thrust/detail/type_traits/iterator/is_output_iterator.h", + "cuda/include/thrust/detail/type_traits/minimum_type.h", + "cuda/include/thrust/detail/type_traits/pointer_traits.h", + "cuda/include/thrust/detail/type_traits/result_of_adaptable_function.h", + "cuda/include/thrust/detail/uninitialized_copy.inl", + "cuda/include/thrust/detail/uninitialized_fill.inl", + "cuda/include/thrust/detail/unique.inl", + "cuda/include/thrust/detail/use_default.h", + "cuda/include/thrust/detail/util/align.h", + "cuda/include/thrust/detail/util/blocking.h", + "cuda/include/thrust/detail/vector_base.h", + "cuda/include/thrust/detail/vector_base.inl", + "cuda/include/thrust/device_allocator.h", + "cuda/include/thrust/device_delete.h", + "cuda/include/thrust/device_free.h", + "cuda/include/thrust/device_malloc.h", + "cuda/include/thrust/device_malloc_allocator.h", + "cuda/include/thrust/device_new.h", + "cuda/include/thrust/device_new_allocator.h", + "cuda/include/thrust/device_ptr.h", + "cuda/include/thrust/device_reference.h", + "cuda/include/thrust/device_vector.h", + "cuda/include/thrust/distance.h", + "cuda/include/thrust/equal.h", + "cuda/include/thrust/execution_policy.h", + "cuda/include/thrust/extrema.h", + "cuda/include/thrust/fill.h", + "cuda/include/thrust/find.h", + "cuda/include/thrust/for_each.h", + "cuda/include/thrust/functional.h", + "cuda/include/thrust/gather.h", + "cuda/include/thrust/generate.h", + "cuda/include/thrust/host_vector.h", + "cuda/include/thrust/inner_product.h", + "cuda/include/thrust/iterator/constant_iterator.h", + "cuda/include/thrust/iterator/counting_iterator.h", + "cuda/include/thrust/iterator/detail/any_assign.h", + "cuda/include/thrust/iterator/detail/any_system_tag.h", + "cuda/include/thrust/iterator/detail/constant_iterator_base.h", + "cuda/include/thrust/iterator/detail/counting_iterator.inl", + "cuda/include/thrust/iterator/detail/device_system_tag.h", + "cuda/include/thrust/iterator/detail/discard_iterator_base.h", + "cuda/include/thrust/iterator/detail/distance_from_result.h", + "cuda/include/thrust/iterator/detail/host_system_tag.h", + "cuda/include/thrust/iterator/detail/is_iterator_category.h", + "cuda/include/thrust/iterator/detail/is_trivial_iterator.h", + "cuda/include/thrust/iterator/detail/iterator_adaptor_base.h", + "cuda/include/thrust/iterator/detail/iterator_category_to_system.h", + "cuda/include/thrust/iterator/detail/iterator_category_to_traversal.h", + "cuda/include/thrust/iterator/detail/iterator_category_with_system_and_traversal.h", + "cuda/include/thrust/iterator/detail/iterator_facade_category.h", + "cuda/include/thrust/iterator/detail/iterator_traits.inl", + "cuda/include/thrust/iterator/detail/iterator_traversal_tags.h", + "cuda/include/thrust/iterator/detail/join_iterator.h", + "cuda/include/thrust/iterator/detail/minimum_category.h", + "cuda/include/thrust/iterator/detail/minimum_system.h", + "cuda/include/thrust/iterator/detail/normal_iterator.h", + "cuda/include/thrust/iterator/detail/permutation_iterator_base.h", + "cuda/include/thrust/iterator/detail/retag.h", + "cuda/include/thrust/iterator/detail/reverse_iterator.inl", + "cuda/include/thrust/iterator/detail/reverse_iterator_base.h", + "cuda/include/thrust/iterator/detail/tagged_iterator.h", + "cuda/include/thrust/iterator/detail/transform_iterator.inl", + "cuda/include/thrust/iterator/detail/transform_output_iterator.inl", + "cuda/include/thrust/iterator/detail/tuple_of_iterator_references.h", + "cuda/include/thrust/iterator/detail/universal_categories.h", + "cuda/include/thrust/iterator/detail/zip_iterator.inl", + "cuda/include/thrust/iterator/detail/zip_iterator_base.h", + "cuda/include/thrust/iterator/discard_iterator.h", + "cuda/include/thrust/iterator/iterator_adaptor.h", + "cuda/include/thrust/iterator/iterator_categories.h", + "cuda/include/thrust/iterator/iterator_facade.h", + "cuda/include/thrust/iterator/iterator_traits.h", + "cuda/include/thrust/iterator/permutation_iterator.h", + "cuda/include/thrust/iterator/retag.h", + "cuda/include/thrust/iterator/reverse_iterator.h", + "cuda/include/thrust/iterator/transform_iterator.h", + "cuda/include/thrust/iterator/transform_output_iterator.h", + "cuda/include/thrust/iterator/zip_iterator.h", + "cuda/include/thrust/logical.h", + "cuda/include/thrust/memory.h", + "cuda/include/thrust/merge.h", + "cuda/include/thrust/mismatch.h", + "cuda/include/thrust/pair.h", + "cuda/include/thrust/partition.h", + "cuda/include/thrust/random.h", + "cuda/include/thrust/random/detail/discard_block_engine.inl", + "cuda/include/thrust/random/detail/linear_congruential_engine.inl", + "cuda/include/thrust/random/detail/linear_congruential_engine_discard.h", + "cuda/include/thrust/random/detail/linear_feedback_shift_engine.inl", + "cuda/include/thrust/random/detail/linear_feedback_shift_engine_wordmask.h", + "cuda/include/thrust/random/detail/mod.h", + "cuda/include/thrust/random/detail/normal_distribution.inl", + "cuda/include/thrust/random/detail/normal_distribution_base.h", + "cuda/include/thrust/random/detail/random_core_access.h", + "cuda/include/thrust/random/detail/subtract_with_carry_engine.inl", + "cuda/include/thrust/random/detail/uniform_int_distribution.inl", + "cuda/include/thrust/random/detail/uniform_real_distribution.inl", + "cuda/include/thrust/random/detail/xor_combine_engine.inl", + "cuda/include/thrust/random/detail/xor_combine_engine_max.h", + "cuda/include/thrust/random/discard_block_engine.h", + "cuda/include/thrust/random/linear_congruential_engine.h", + "cuda/include/thrust/random/linear_feedback_shift_engine.h", + "cuda/include/thrust/random/normal_distribution.h", + "cuda/include/thrust/random/subtract_with_carry_engine.h", + "cuda/include/thrust/random/uniform_int_distribution.h", + "cuda/include/thrust/random/uniform_real_distribution.h", + "cuda/include/thrust/random/xor_combine_engine.h", + "cuda/include/thrust/reduce.h", + "cuda/include/thrust/remove.h", + "cuda/include/thrust/replace.h", + "cuda/include/thrust/reverse.h", + "cuda/include/thrust/scan.h", + "cuda/include/thrust/scatter.h", + "cuda/include/thrust/sequence.h", + "cuda/include/thrust/set_operations.h", + "cuda/include/thrust/sort.h", + "cuda/include/thrust/swap.h", + "cuda/include/thrust/system/cpp/detail/adjacent_difference.h", + "cuda/include/thrust/system/cpp/detail/assign_value.h", + "cuda/include/thrust/system/cpp/detail/binary_search.h", + "cuda/include/thrust/system/cpp/detail/copy.h", + "cuda/include/thrust/system/cpp/detail/copy_if.h", + "cuda/include/thrust/system/cpp/detail/count.h", + "cuda/include/thrust/system/cpp/detail/equal.h", + "cuda/include/thrust/system/cpp/detail/execution_policy.h", + "cuda/include/thrust/system/cpp/detail/extrema.h", + "cuda/include/thrust/system/cpp/detail/fill.h", + "cuda/include/thrust/system/cpp/detail/find.h", + "cuda/include/thrust/system/cpp/detail/for_each.h", + "cuda/include/thrust/system/cpp/detail/gather.h", + "cuda/include/thrust/system/cpp/detail/generate.h", + "cuda/include/thrust/system/cpp/detail/get_value.h", + "cuda/include/thrust/system/cpp/detail/inner_product.h", + "cuda/include/thrust/system/cpp/detail/iter_swap.h", + "cuda/include/thrust/system/cpp/detail/logical.h", + "cuda/include/thrust/system/cpp/detail/malloc_and_free.h", + "cuda/include/thrust/system/cpp/detail/memory.inl", + "cuda/include/thrust/system/cpp/detail/merge.h", + "cuda/include/thrust/system/cpp/detail/mismatch.h", + "cuda/include/thrust/system/cpp/detail/par.h", + "cuda/include/thrust/system/cpp/detail/partition.h", + "cuda/include/thrust/system/cpp/detail/reduce.h", + "cuda/include/thrust/system/cpp/detail/reduce_by_key.h", + "cuda/include/thrust/system/cpp/detail/remove.h", + "cuda/include/thrust/system/cpp/detail/replace.h", + "cuda/include/thrust/system/cpp/detail/reverse.h", + "cuda/include/thrust/system/cpp/detail/scan.h", + "cuda/include/thrust/system/cpp/detail/scan_by_key.h", + "cuda/include/thrust/system/cpp/detail/scatter.h", + "cuda/include/thrust/system/cpp/detail/sequence.h", + "cuda/include/thrust/system/cpp/detail/set_operations.h", + "cuda/include/thrust/system/cpp/detail/sort.h", + "cuda/include/thrust/system/cpp/detail/swap_ranges.h", + "cuda/include/thrust/system/cpp/detail/tabulate.h", + "cuda/include/thrust/system/cpp/detail/temporary_buffer.h", + "cuda/include/thrust/system/cpp/detail/transform.h", + "cuda/include/thrust/system/cpp/detail/transform_reduce.h", + "cuda/include/thrust/system/cpp/detail/transform_scan.h", + "cuda/include/thrust/system/cpp/detail/uninitialized_copy.h", + "cuda/include/thrust/system/cpp/detail/uninitialized_fill.h", + "cuda/include/thrust/system/cpp/detail/unique.h", + "cuda/include/thrust/system/cpp/detail/unique_by_key.h", + "cuda/include/thrust/system/cpp/detail/vector.inl", + "cuda/include/thrust/system/cpp/execution_policy.h", + "cuda/include/thrust/system/cpp/memory.h", + "cuda/include/thrust/system/cpp/vector.h", + "cuda/include/thrust/system/cuda/config.h", + "cuda/include/thrust/system/cuda/detail/adjacent_difference.h", + "cuda/include/thrust/system/cuda/detail/assign_value.h", + "cuda/include/thrust/system/cuda/detail/binary_search.h", + "cuda/include/thrust/system/cuda/detail/copy.h", + "cuda/include/thrust/system/cuda/detail/copy_if.h", + "cuda/include/thrust/system/cuda/detail/core/agent_launcher.h", + "cuda/include/thrust/system/cuda/detail/core/alignment.h", + "cuda/include/thrust/system/cuda/detail/core/triple_chevron_launch.h", + "cuda/include/thrust/system/cuda/detail/core/util.h", + "cuda/include/thrust/system/cuda/detail/count.h", + "cuda/include/thrust/system/cuda/detail/cross_system.h", + "cuda/include/thrust/system/cuda/detail/cub/agent/agent_histogram.cuh", + "cuda/include/thrust/system/cuda/detail/cub/agent/agent_radix_sort_downsweep.cuh", + "cuda/include/thrust/system/cuda/detail/cub/agent/agent_radix_sort_upsweep.cuh", + "cuda/include/thrust/system/cuda/detail/cub/agent/agent_reduce.cuh", + "cuda/include/thrust/system/cuda/detail/cub/agent/agent_reduce_by_key.cuh", + "cuda/include/thrust/system/cuda/detail/cub/agent/agent_rle.cuh", + "cuda/include/thrust/system/cuda/detail/cub/agent/agent_scan.cuh", + "cuda/include/thrust/system/cuda/detail/cub/agent/agent_segment_fixup.cuh", + "cuda/include/thrust/system/cuda/detail/cub/agent/agent_select_if.cuh", + "cuda/include/thrust/system/cuda/detail/cub/agent/agent_spmv_csrt.cuh", + "cuda/include/thrust/system/cuda/detail/cub/agent/agent_spmv_orig.cuh", + "cuda/include/thrust/system/cuda/detail/cub/agent/agent_spmv_row_based.cuh", + "cuda/include/thrust/system/cuda/detail/cub/agent/single_pass_scan_operators.cuh", + "cuda/include/thrust/system/cuda/detail/cub/block/block_adjacent_difference.cuh", + "cuda/include/thrust/system/cuda/detail/cub/block/block_discontinuity.cuh", + "cuda/include/thrust/system/cuda/detail/cub/block/block_exchange.cuh", + "cuda/include/thrust/system/cuda/detail/cub/block/block_histogram.cuh", + "cuda/include/thrust/system/cuda/detail/cub/block/block_load.cuh", + "cuda/include/thrust/system/cuda/detail/cub/block/block_radix_rank.cuh", + "cuda/include/thrust/system/cuda/detail/cub/block/block_radix_sort.cuh", + "cuda/include/thrust/system/cuda/detail/cub/block/block_raking_layout.cuh", + "cuda/include/thrust/system/cuda/detail/cub/block/block_reduce.cuh", + "cuda/include/thrust/system/cuda/detail/cub/block/block_scan.cuh", + "cuda/include/thrust/system/cuda/detail/cub/block/block_shuffle.cuh", + "cuda/include/thrust/system/cuda/detail/cub/block/block_store.cuh", + "cuda/include/thrust/system/cuda/detail/cub/block/specializations/block_histogram_atomic.cuh", + "cuda/include/thrust/system/cuda/detail/cub/block/specializations/block_histogram_sort.cuh", + "cuda/include/thrust/system/cuda/detail/cub/block/specializations/block_reduce_raking.cuh", + "cuda/include/thrust/system/cuda/detail/cub/block/specializations/block_reduce_raking_commutative_only.cuh", + "cuda/include/thrust/system/cuda/detail/cub/block/specializations/block_reduce_warp_reductions.cuh", + "cuda/include/thrust/system/cuda/detail/cub/block/specializations/block_scan_raking.cuh", + "cuda/include/thrust/system/cuda/detail/cub/block/specializations/block_scan_warp_scans.cuh", + "cuda/include/thrust/system/cuda/detail/cub/block/specializations/block_scan_warp_scans2.cuh", + "cuda/include/thrust/system/cuda/detail/cub/block/specializations/block_scan_warp_scans3.cuh", + "cuda/include/thrust/system/cuda/detail/cub/cub.cuh", + "cuda/include/thrust/system/cuda/detail/cub/device/device_histogram.cuh", + "cuda/include/thrust/system/cuda/detail/cub/device/device_partition.cuh", + "cuda/include/thrust/system/cuda/detail/cub/device/device_radix_sort.cuh", + "cuda/include/thrust/system/cuda/detail/cub/device/device_reduce.cuh", + "cuda/include/thrust/system/cuda/detail/cub/device/device_run_length_encode.cuh", + "cuda/include/thrust/system/cuda/detail/cub/device/device_scan.cuh", + "cuda/include/thrust/system/cuda/detail/cub/device/device_segmented_radix_sort.cuh", + "cuda/include/thrust/system/cuda/detail/cub/device/device_segmented_reduce.cuh", + "cuda/include/thrust/system/cuda/detail/cub/device/device_select.cuh", + "cuda/include/thrust/system/cuda/detail/cub/device/device_spmv.cuh", + "cuda/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_histogram.cuh", + "cuda/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_radix_sort.cuh", + "cuda/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_reduce.cuh", + "cuda/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_reduce_by_key.cuh", + "cuda/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_rle.cuh", + "cuda/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_scan.cuh", + "cuda/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_select_if.cuh", + "cuda/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_spmv_csrt.cuh", + "cuda/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_spmv_orig.cuh", + "cuda/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_spmv_row_based.cuh", + "cuda/include/thrust/system/cuda/detail/cub/grid/grid_barrier.cuh", + "cuda/include/thrust/system/cuda/detail/cub/grid/grid_even_share.cuh", + "cuda/include/thrust/system/cuda/detail/cub/grid/grid_mapping.cuh", + "cuda/include/thrust/system/cuda/detail/cub/grid/grid_queue.cuh", + "cuda/include/thrust/system/cuda/detail/cub/host/mutex.cuh", + "cuda/include/thrust/system/cuda/detail/cub/iterator/arg_index_input_iterator.cuh", + "cuda/include/thrust/system/cuda/detail/cub/iterator/cache_modified_input_iterator.cuh", + "cuda/include/thrust/system/cuda/detail/cub/iterator/cache_modified_output_iterator.cuh", + "cuda/include/thrust/system/cuda/detail/cub/iterator/constant_input_iterator.cuh", + "cuda/include/thrust/system/cuda/detail/cub/iterator/counting_input_iterator.cuh", + "cuda/include/thrust/system/cuda/detail/cub/iterator/discard_output_iterator.cuh", + "cuda/include/thrust/system/cuda/detail/cub/iterator/tex_obj_input_iterator.cuh", + "cuda/include/thrust/system/cuda/detail/cub/iterator/tex_ref_input_iterator.cuh", + "cuda/include/thrust/system/cuda/detail/cub/iterator/transform_input_iterator.cuh", + "cuda/include/thrust/system/cuda/detail/cub/thread/thread_load.cuh", + "cuda/include/thrust/system/cuda/detail/cub/thread/thread_operators.cuh", + "cuda/include/thrust/system/cuda/detail/cub/thread/thread_reduce.cuh", + "cuda/include/thrust/system/cuda/detail/cub/thread/thread_scan.cuh", + "cuda/include/thrust/system/cuda/detail/cub/thread/thread_search.cuh", + "cuda/include/thrust/system/cuda/detail/cub/thread/thread_store.cuh", + "cuda/include/thrust/system/cuda/detail/cub/util_allocator.cuh", + "cuda/include/thrust/system/cuda/detail/cub/util_arch.cuh", + "cuda/include/thrust/system/cuda/detail/cub/util_debug.cuh", + "cuda/include/thrust/system/cuda/detail/cub/util_device.cuh", + "cuda/include/thrust/system/cuda/detail/cub/util_macro.cuh", + "cuda/include/thrust/system/cuda/detail/cub/util_namespace.cuh", + "cuda/include/thrust/system/cuda/detail/cub/util_ptx.cuh", + "cuda/include/thrust/system/cuda/detail/cub/util_type.cuh", + "cuda/include/thrust/system/cuda/detail/cub/warp/specializations/warp_reduce_shfl.cuh", + "cuda/include/thrust/system/cuda/detail/cub/warp/specializations/warp_reduce_smem.cuh", + "cuda/include/thrust/system/cuda/detail/cub/warp/specializations/warp_scan_shfl.cuh", + "cuda/include/thrust/system/cuda/detail/cub/warp/specializations/warp_scan_smem.cuh", + "cuda/include/thrust/system/cuda/detail/cub/warp/warp_reduce.cuh", + "cuda/include/thrust/system/cuda/detail/cub/warp/warp_scan.cuh", + "cuda/include/thrust/system/cuda/detail/equal.h", + "cuda/include/thrust/system/cuda/detail/error.inl", + "cuda/include/thrust/system/cuda/detail/execution_policy.h", + "cuda/include/thrust/system/cuda/detail/extrema.h", + "cuda/include/thrust/system/cuda/detail/fill.h", + "cuda/include/thrust/system/cuda/detail/find.h", + "cuda/include/thrust/system/cuda/detail/for_each.h", + "cuda/include/thrust/system/cuda/detail/gather.h", + "cuda/include/thrust/system/cuda/detail/generate.h", + "cuda/include/thrust/system/cuda/detail/get_value.h", + "cuda/include/thrust/system/cuda/detail/guarded_cuda_runtime_api.h", + "cuda/include/thrust/system/cuda/detail/guarded_driver_types.h", + "cuda/include/thrust/system/cuda/detail/inner_product.h", + "cuda/include/thrust/system/cuda/detail/internal/copy_cross_system.h", + "cuda/include/thrust/system/cuda/detail/internal/copy_device_to_device.h", + "cuda/include/thrust/system/cuda/detail/iter_swap.h", + "cuda/include/thrust/system/cuda/detail/logical.h", + "cuda/include/thrust/system/cuda/detail/malloc_and_free.h", + "cuda/include/thrust/system/cuda/detail/memory.inl", + "cuda/include/thrust/system/cuda/detail/memory_buffer.h", + "cuda/include/thrust/system/cuda/detail/merge.h", + "cuda/include/thrust/system/cuda/detail/mismatch.h", + "cuda/include/thrust/system/cuda/detail/par.h", + "cuda/include/thrust/system/cuda/detail/par_to_seq.h", + "cuda/include/thrust/system/cuda/detail/parallel_for.h", + "cuda/include/thrust/system/cuda/detail/partition.h", + "cuda/include/thrust/system/cuda/detail/reduce.h", + "cuda/include/thrust/system/cuda/detail/reduce_by_key.h", + "cuda/include/thrust/system/cuda/detail/remove.h", + "cuda/include/thrust/system/cuda/detail/replace.h", + "cuda/include/thrust/system/cuda/detail/reverse.h", + "cuda/include/thrust/system/cuda/detail/scan.h", + "cuda/include/thrust/system/cuda/detail/scan_by_key.h", + "cuda/include/thrust/system/cuda/detail/scatter.h", + "cuda/include/thrust/system/cuda/detail/sequence.h", + "cuda/include/thrust/system/cuda/detail/set_operations.h", + "cuda/include/thrust/system/cuda/detail/sort.h", + "cuda/include/thrust/system/cuda/detail/swap_ranges.h", + "cuda/include/thrust/system/cuda/detail/tabulate.h", + "cuda/include/thrust/system/cuda/detail/temporary_buffer.h", + "cuda/include/thrust/system/cuda/detail/terminate.h", + "cuda/include/thrust/system/cuda/detail/transform.h", + "cuda/include/thrust/system/cuda/detail/transform_reduce.h", + "cuda/include/thrust/system/cuda/detail/transform_scan.h", + "cuda/include/thrust/system/cuda/detail/uninitialized_copy.h", + "cuda/include/thrust/system/cuda/detail/uninitialized_fill.h", + "cuda/include/thrust/system/cuda/detail/unique.h", + "cuda/include/thrust/system/cuda/detail/unique_by_key.h", + "cuda/include/thrust/system/cuda/detail/util.h", + "cuda/include/thrust/system/cuda/detail/vector.inl", + "cuda/include/thrust/system/cuda/error.h", + "cuda/include/thrust/system/cuda/execution_policy.h", + "cuda/include/thrust/system/cuda/experimental/pinned_allocator.h", + "cuda/include/thrust/system/cuda/memory.h", + "cuda/include/thrust/system/cuda/vector.h", + "cuda/include/thrust/system/detail/adl/adjacent_difference.h", + "cuda/include/thrust/system/detail/adl/assign_value.h", + "cuda/include/thrust/system/detail/adl/binary_search.h", + "cuda/include/thrust/system/detail/adl/copy.h", + "cuda/include/thrust/system/detail/adl/copy_if.h", + "cuda/include/thrust/system/detail/adl/count.h", + "cuda/include/thrust/system/detail/adl/equal.h", + "cuda/include/thrust/system/detail/adl/extrema.h", + "cuda/include/thrust/system/detail/adl/fill.h", + "cuda/include/thrust/system/detail/adl/find.h", + "cuda/include/thrust/system/detail/adl/for_each.h", + "cuda/include/thrust/system/detail/adl/gather.h", + "cuda/include/thrust/system/detail/adl/generate.h", + "cuda/include/thrust/system/detail/adl/get_value.h", + "cuda/include/thrust/system/detail/adl/inner_product.h", + "cuda/include/thrust/system/detail/adl/iter_swap.h", + "cuda/include/thrust/system/detail/adl/logical.h", + "cuda/include/thrust/system/detail/adl/malloc_and_free.h", + "cuda/include/thrust/system/detail/adl/merge.h", + "cuda/include/thrust/system/detail/adl/mismatch.h", + "cuda/include/thrust/system/detail/adl/partition.h", + "cuda/include/thrust/system/detail/adl/reduce.h", + "cuda/include/thrust/system/detail/adl/reduce_by_key.h", + "cuda/include/thrust/system/detail/adl/remove.h", + "cuda/include/thrust/system/detail/adl/replace.h", + "cuda/include/thrust/system/detail/adl/reverse.h", + "cuda/include/thrust/system/detail/adl/scan.h", + "cuda/include/thrust/system/detail/adl/scan_by_key.h", + "cuda/include/thrust/system/detail/adl/scatter.h", + "cuda/include/thrust/system/detail/adl/sequence.h", + "cuda/include/thrust/system/detail/adl/set_operations.h", + "cuda/include/thrust/system/detail/adl/sort.h", + "cuda/include/thrust/system/detail/adl/swap_ranges.h", + "cuda/include/thrust/system/detail/adl/tabulate.h", + "cuda/include/thrust/system/detail/adl/temporary_buffer.h", + "cuda/include/thrust/system/detail/adl/transform.h", + "cuda/include/thrust/system/detail/adl/transform_reduce.h", + "cuda/include/thrust/system/detail/adl/transform_scan.h", + "cuda/include/thrust/system/detail/adl/uninitialized_copy.h", + "cuda/include/thrust/system/detail/adl/uninitialized_fill.h", + "cuda/include/thrust/system/detail/adl/unique.h", + "cuda/include/thrust/system/detail/adl/unique_by_key.h", + "cuda/include/thrust/system/detail/bad_alloc.h", + "cuda/include/thrust/system/detail/errno.h", + "cuda/include/thrust/system/detail/error_category.inl", + "cuda/include/thrust/system/detail/error_code.inl", + "cuda/include/thrust/system/detail/error_condition.inl", + "cuda/include/thrust/system/detail/generic/adjacent_difference.h", + "cuda/include/thrust/system/detail/generic/adjacent_difference.inl", + "cuda/include/thrust/system/detail/generic/advance.h", + "cuda/include/thrust/system/detail/generic/advance.inl", + "cuda/include/thrust/system/detail/generic/binary_search.h", + "cuda/include/thrust/system/detail/generic/binary_search.inl", + "cuda/include/thrust/system/detail/generic/copy.h", + "cuda/include/thrust/system/detail/generic/copy.inl", + "cuda/include/thrust/system/detail/generic/copy_if.h", + "cuda/include/thrust/system/detail/generic/copy_if.inl", + "cuda/include/thrust/system/detail/generic/count.h", + "cuda/include/thrust/system/detail/generic/count.inl", + "cuda/include/thrust/system/detail/generic/distance.h", + "cuda/include/thrust/system/detail/generic/distance.inl", + "cuda/include/thrust/system/detail/generic/equal.h", + "cuda/include/thrust/system/detail/generic/equal.inl", + "cuda/include/thrust/system/detail/generic/extrema.h", + "cuda/include/thrust/system/detail/generic/extrema.inl", + "cuda/include/thrust/system/detail/generic/fill.h", + "cuda/include/thrust/system/detail/generic/find.h", + "cuda/include/thrust/system/detail/generic/find.inl", + "cuda/include/thrust/system/detail/generic/for_each.h", + "cuda/include/thrust/system/detail/generic/gather.h", + "cuda/include/thrust/system/detail/generic/gather.inl", + "cuda/include/thrust/system/detail/generic/generate.h", + "cuda/include/thrust/system/detail/generic/generate.inl", + "cuda/include/thrust/system/detail/generic/inner_product.h", + "cuda/include/thrust/system/detail/generic/inner_product.inl", + "cuda/include/thrust/system/detail/generic/logical.h", + "cuda/include/thrust/system/detail/generic/memory.h", + "cuda/include/thrust/system/detail/generic/memory.inl", + "cuda/include/thrust/system/detail/generic/merge.h", + "cuda/include/thrust/system/detail/generic/merge.inl", + "cuda/include/thrust/system/detail/generic/mismatch.h", + "cuda/include/thrust/system/detail/generic/mismatch.inl", + "cuda/include/thrust/system/detail/generic/partition.h", + "cuda/include/thrust/system/detail/generic/partition.inl", + "cuda/include/thrust/system/detail/generic/reduce.h", + "cuda/include/thrust/system/detail/generic/reduce.inl", + "cuda/include/thrust/system/detail/generic/reduce_by_key.h", + "cuda/include/thrust/system/detail/generic/reduce_by_key.inl", + "cuda/include/thrust/system/detail/generic/remove.h", + "cuda/include/thrust/system/detail/generic/remove.inl", + "cuda/include/thrust/system/detail/generic/replace.h", + "cuda/include/thrust/system/detail/generic/replace.inl", + "cuda/include/thrust/system/detail/generic/reverse.h", + "cuda/include/thrust/system/detail/generic/reverse.inl", + "cuda/include/thrust/system/detail/generic/scalar/binary_search.h", + "cuda/include/thrust/system/detail/generic/scalar/binary_search.inl", + "cuda/include/thrust/system/detail/generic/scan.h", + "cuda/include/thrust/system/detail/generic/scan.inl", + "cuda/include/thrust/system/detail/generic/scan_by_key.h", + "cuda/include/thrust/system/detail/generic/scan_by_key.inl", + "cuda/include/thrust/system/detail/generic/scatter.h", + "cuda/include/thrust/system/detail/generic/scatter.inl", + "cuda/include/thrust/system/detail/generic/select_system.h", + "cuda/include/thrust/system/detail/generic/sequence.h", + "cuda/include/thrust/system/detail/generic/sequence.inl", + "cuda/include/thrust/system/detail/generic/set_operations.h", + "cuda/include/thrust/system/detail/generic/set_operations.inl", + "cuda/include/thrust/system/detail/generic/sort.h", + "cuda/include/thrust/system/detail/generic/sort.inl", + "cuda/include/thrust/system/detail/generic/swap_ranges.h", + "cuda/include/thrust/system/detail/generic/swap_ranges.inl", + "cuda/include/thrust/system/detail/generic/tabulate.h", + "cuda/include/thrust/system/detail/generic/tabulate.inl", + "cuda/include/thrust/system/detail/generic/tag.h", + "cuda/include/thrust/system/detail/generic/temporary_buffer.h", + "cuda/include/thrust/system/detail/generic/temporary_buffer.inl", + "cuda/include/thrust/system/detail/generic/transform.h", + "cuda/include/thrust/system/detail/generic/transform.inl", + "cuda/include/thrust/system/detail/generic/transform_reduce.h", + "cuda/include/thrust/system/detail/generic/transform_reduce.inl", + "cuda/include/thrust/system/detail/generic/transform_scan.h", + "cuda/include/thrust/system/detail/generic/transform_scan.inl", + "cuda/include/thrust/system/detail/generic/type_traits.h", + "cuda/include/thrust/system/detail/generic/uninitialized_copy.h", + "cuda/include/thrust/system/detail/generic/uninitialized_copy.inl", + "cuda/include/thrust/system/detail/generic/uninitialized_fill.h", + "cuda/include/thrust/system/detail/generic/uninitialized_fill.inl", + "cuda/include/thrust/system/detail/generic/unique.h", + "cuda/include/thrust/system/detail/generic/unique.inl", + "cuda/include/thrust/system/detail/generic/unique_by_key.h", + "cuda/include/thrust/system/detail/generic/unique_by_key.inl", + "cuda/include/thrust/system/detail/internal/decompose.h", + "cuda/include/thrust/system/detail/sequential/adjacent_difference.h", + "cuda/include/thrust/system/detail/sequential/assign_value.h", + "cuda/include/thrust/system/detail/sequential/binary_search.h", + "cuda/include/thrust/system/detail/sequential/copy.h", + "cuda/include/thrust/system/detail/sequential/copy.inl", + "cuda/include/thrust/system/detail/sequential/copy_backward.h", + "cuda/include/thrust/system/detail/sequential/copy_if.h", + "cuda/include/thrust/system/detail/sequential/count.h", + "cuda/include/thrust/system/detail/sequential/equal.h", + "cuda/include/thrust/system/detail/sequential/execution_policy.h", + "cuda/include/thrust/system/detail/sequential/extrema.h", + "cuda/include/thrust/system/detail/sequential/fill.h", + "cuda/include/thrust/system/detail/sequential/find.h", + "cuda/include/thrust/system/detail/sequential/for_each.h", + "cuda/include/thrust/system/detail/sequential/gather.h", + "cuda/include/thrust/system/detail/sequential/general_copy.h", + "cuda/include/thrust/system/detail/sequential/generate.h", + "cuda/include/thrust/system/detail/sequential/get_value.h", + "cuda/include/thrust/system/detail/sequential/inner_product.h", + "cuda/include/thrust/system/detail/sequential/insertion_sort.h", + "cuda/include/thrust/system/detail/sequential/iter_swap.h", + "cuda/include/thrust/system/detail/sequential/logical.h", + "cuda/include/thrust/system/detail/sequential/malloc_and_free.h", + "cuda/include/thrust/system/detail/sequential/merge.h", + "cuda/include/thrust/system/detail/sequential/merge.inl", + "cuda/include/thrust/system/detail/sequential/mismatch.h", + "cuda/include/thrust/system/detail/sequential/partition.h", + "cuda/include/thrust/system/detail/sequential/reduce.h", + "cuda/include/thrust/system/detail/sequential/reduce_by_key.h", + "cuda/include/thrust/system/detail/sequential/remove.h", + "cuda/include/thrust/system/detail/sequential/replace.h", + "cuda/include/thrust/system/detail/sequential/reverse.h", + "cuda/include/thrust/system/detail/sequential/scan.h", + "cuda/include/thrust/system/detail/sequential/scan_by_key.h", + "cuda/include/thrust/system/detail/sequential/scatter.h", + "cuda/include/thrust/system/detail/sequential/sequence.h", + "cuda/include/thrust/system/detail/sequential/set_operations.h", + "cuda/include/thrust/system/detail/sequential/sort.h", + "cuda/include/thrust/system/detail/sequential/sort.inl", + "cuda/include/thrust/system/detail/sequential/stable_merge_sort.h", + "cuda/include/thrust/system/detail/sequential/stable_merge_sort.inl", + "cuda/include/thrust/system/detail/sequential/stable_primitive_sort.h", + "cuda/include/thrust/system/detail/sequential/stable_primitive_sort.inl", + "cuda/include/thrust/system/detail/sequential/stable_radix_sort.h", + "cuda/include/thrust/system/detail/sequential/stable_radix_sort.inl", + "cuda/include/thrust/system/detail/sequential/swap_ranges.h", + "cuda/include/thrust/system/detail/sequential/tabulate.h", + "cuda/include/thrust/system/detail/sequential/temporary_buffer.h", + "cuda/include/thrust/system/detail/sequential/transform.h", + "cuda/include/thrust/system/detail/sequential/transform_reduce.h", + "cuda/include/thrust/system/detail/sequential/transform_scan.h", + "cuda/include/thrust/system/detail/sequential/trivial_copy.h", + "cuda/include/thrust/system/detail/sequential/uninitialized_copy.h", + "cuda/include/thrust/system/detail/sequential/uninitialized_fill.h", + "cuda/include/thrust/system/detail/sequential/unique.h", + "cuda/include/thrust/system/detail/sequential/unique_by_key.h", + "cuda/include/thrust/system/detail/system_error.inl", + "cuda/include/thrust/system/error_code.h", + "cuda/include/thrust/system/omp/detail/adjacent_difference.h", + "cuda/include/thrust/system/omp/detail/assign_value.h", + "cuda/include/thrust/system/omp/detail/binary_search.h", + "cuda/include/thrust/system/omp/detail/copy.h", + "cuda/include/thrust/system/omp/detail/copy.inl", + "cuda/include/thrust/system/omp/detail/copy_if.h", + "cuda/include/thrust/system/omp/detail/copy_if.inl", + "cuda/include/thrust/system/omp/detail/count.h", + "cuda/include/thrust/system/omp/detail/default_decomposition.h", + "cuda/include/thrust/system/omp/detail/default_decomposition.inl", + "cuda/include/thrust/system/omp/detail/equal.h", + "cuda/include/thrust/system/omp/detail/execution_policy.h", + "cuda/include/thrust/system/omp/detail/extrema.h", + "cuda/include/thrust/system/omp/detail/fill.h", + "cuda/include/thrust/system/omp/detail/find.h", + "cuda/include/thrust/system/omp/detail/for_each.h", + "cuda/include/thrust/system/omp/detail/for_each.inl", + "cuda/include/thrust/system/omp/detail/gather.h", + "cuda/include/thrust/system/omp/detail/generate.h", + "cuda/include/thrust/system/omp/detail/get_value.h", + "cuda/include/thrust/system/omp/detail/inner_product.h", + "cuda/include/thrust/system/omp/detail/iter_swap.h", + "cuda/include/thrust/system/omp/detail/logical.h", + "cuda/include/thrust/system/omp/detail/malloc_and_free.h", + "cuda/include/thrust/system/omp/detail/memory.inl", + "cuda/include/thrust/system/omp/detail/merge.h", + "cuda/include/thrust/system/omp/detail/mismatch.h", + "cuda/include/thrust/system/omp/detail/par.h", + "cuda/include/thrust/system/omp/detail/partition.h", + "cuda/include/thrust/system/omp/detail/partition.inl", + "cuda/include/thrust/system/omp/detail/reduce.h", + "cuda/include/thrust/system/omp/detail/reduce.inl", + "cuda/include/thrust/system/omp/detail/reduce_by_key.h", + "cuda/include/thrust/system/omp/detail/reduce_by_key.inl", + "cuda/include/thrust/system/omp/detail/reduce_intervals.h", + "cuda/include/thrust/system/omp/detail/reduce_intervals.inl", + "cuda/include/thrust/system/omp/detail/remove.h", + "cuda/include/thrust/system/omp/detail/remove.inl", + "cuda/include/thrust/system/omp/detail/replace.h", + "cuda/include/thrust/system/omp/detail/reverse.h", + "cuda/include/thrust/system/omp/detail/scan.h", + "cuda/include/thrust/system/omp/detail/scan_by_key.h", + "cuda/include/thrust/system/omp/detail/scatter.h", + "cuda/include/thrust/system/omp/detail/sequence.h", + "cuda/include/thrust/system/omp/detail/set_operations.h", + "cuda/include/thrust/system/omp/detail/sort.h", + "cuda/include/thrust/system/omp/detail/sort.inl", + "cuda/include/thrust/system/omp/detail/swap_ranges.h", + "cuda/include/thrust/system/omp/detail/tabulate.h", + "cuda/include/thrust/system/omp/detail/temporary_buffer.h", + "cuda/include/thrust/system/omp/detail/transform.h", + "cuda/include/thrust/system/omp/detail/transform_reduce.h", + "cuda/include/thrust/system/omp/detail/transform_scan.h", + "cuda/include/thrust/system/omp/detail/uninitialized_copy.h", + "cuda/include/thrust/system/omp/detail/uninitialized_fill.h", + "cuda/include/thrust/system/omp/detail/unique.h", + "cuda/include/thrust/system/omp/detail/unique.inl", + "cuda/include/thrust/system/omp/detail/unique_by_key.h", + "cuda/include/thrust/system/omp/detail/unique_by_key.inl", + "cuda/include/thrust/system/omp/detail/vector.inl", + "cuda/include/thrust/system/omp/execution_policy.h", + "cuda/include/thrust/system/omp/memory.h", + "cuda/include/thrust/system/omp/vector.h", + "cuda/include/thrust/system/system_error.h", + "cuda/include/thrust/system/tbb/detail/adjacent_difference.h", + "cuda/include/thrust/system/tbb/detail/assign_value.h", + "cuda/include/thrust/system/tbb/detail/binary_search.h", + "cuda/include/thrust/system/tbb/detail/copy.h", + "cuda/include/thrust/system/tbb/detail/copy.inl", + "cuda/include/thrust/system/tbb/detail/copy_if.h", + "cuda/include/thrust/system/tbb/detail/copy_if.inl", + "cuda/include/thrust/system/tbb/detail/count.h", + "cuda/include/thrust/system/tbb/detail/equal.h", + "cuda/include/thrust/system/tbb/detail/execution_policy.h", + "cuda/include/thrust/system/tbb/detail/extrema.h", + "cuda/include/thrust/system/tbb/detail/fill.h", + "cuda/include/thrust/system/tbb/detail/find.h", + "cuda/include/thrust/system/tbb/detail/for_each.h", + "cuda/include/thrust/system/tbb/detail/for_each.inl", + "cuda/include/thrust/system/tbb/detail/gather.h", + "cuda/include/thrust/system/tbb/detail/generate.h", + "cuda/include/thrust/system/tbb/detail/get_value.h", + "cuda/include/thrust/system/tbb/detail/inner_product.h", + "cuda/include/thrust/system/tbb/detail/iter_swap.h", + "cuda/include/thrust/system/tbb/detail/logical.h", + "cuda/include/thrust/system/tbb/detail/malloc_and_free.h", + "cuda/include/thrust/system/tbb/detail/memory.inl", + "cuda/include/thrust/system/tbb/detail/merge.h", + "cuda/include/thrust/system/tbb/detail/merge.inl", + "cuda/include/thrust/system/tbb/detail/mismatch.h", + "cuda/include/thrust/system/tbb/detail/par.h", + "cuda/include/thrust/system/tbb/detail/partition.h", + "cuda/include/thrust/system/tbb/detail/partition.inl", + "cuda/include/thrust/system/tbb/detail/reduce.h", + "cuda/include/thrust/system/tbb/detail/reduce.inl", + "cuda/include/thrust/system/tbb/detail/reduce_by_key.h", + "cuda/include/thrust/system/tbb/detail/reduce_by_key.inl", + "cuda/include/thrust/system/tbb/detail/reduce_intervals.h", + "cuda/include/thrust/system/tbb/detail/remove.h", + "cuda/include/thrust/system/tbb/detail/remove.inl", + "cuda/include/thrust/system/tbb/detail/replace.h", + "cuda/include/thrust/system/tbb/detail/reverse.h", + "cuda/include/thrust/system/tbb/detail/scan.h", + "cuda/include/thrust/system/tbb/detail/scan.inl", + "cuda/include/thrust/system/tbb/detail/scan_by_key.h", + "cuda/include/thrust/system/tbb/detail/scatter.h", + "cuda/include/thrust/system/tbb/detail/sequence.h", + "cuda/include/thrust/system/tbb/detail/set_operations.h", + "cuda/include/thrust/system/tbb/detail/sort.h", + "cuda/include/thrust/system/tbb/detail/sort.inl", + "cuda/include/thrust/system/tbb/detail/swap_ranges.h", + "cuda/include/thrust/system/tbb/detail/tabulate.h", + "cuda/include/thrust/system/tbb/detail/temporary_buffer.h", + "cuda/include/thrust/system/tbb/detail/transform.h", + "cuda/include/thrust/system/tbb/detail/transform_reduce.h", + "cuda/include/thrust/system/tbb/detail/transform_scan.h", + "cuda/include/thrust/system/tbb/detail/uninitialized_copy.h", + "cuda/include/thrust/system/tbb/detail/uninitialized_fill.h", + "cuda/include/thrust/system/tbb/detail/unique.h", + "cuda/include/thrust/system/tbb/detail/unique.inl", + "cuda/include/thrust/system/tbb/detail/unique_by_key.h", + "cuda/include/thrust/system/tbb/detail/unique_by_key.inl", + "cuda/include/thrust/system/tbb/detail/vector.inl", + "cuda/include/thrust/system/tbb/execution_policy.h", + "cuda/include/thrust/system/tbb/memory.h", + "cuda/include/thrust/system/tbb/vector.h", + "cuda/include/thrust/system_error.h", + "cuda/include/thrust/tabulate.h", + "cuda/include/thrust/transform.h", + "cuda/include/thrust/transform_reduce.h", + "cuda/include/thrust/transform_scan.h", + "cuda/include/thrust/tuple.h", + "cuda/include/thrust/uninitialized_copy.h", + "cuda/include/thrust/uninitialized_fill.h", + "cuda/include/thrust/unique.h", + "cuda/include/thrust/version.h", + "cuda/include/vector_functions.h", + "cuda/include/vector_functions.hpp", + "cuda/include/vector_types.h", + ], + cmd = """ +if [ -d "$(@D)/extras" ]; then rm $(@D)/extras -drf; fi && if [ -d "$(@D)/include" ]; then rm $(@D)/include -drf; fi && if [ -d "$(@D)/lib" ]; then rm $(@D)/lib -drf; fi && if [ -d "$(@D)/nvvm" ]; then rm $(@D)/nvvm -drf; fi && cp "/usr/local/cuda-9.0/include/CL/cl.h" "$(@D)/cuda/include/CL/cl.h" && cp "/usr/local/cuda-9.0/include/CL/cl.hpp" "$(@D)/cuda/include/CL/cl.hpp" && cp "/usr/local/cuda-9.0/include/CL/cl_egl.h" "$(@D)/cuda/include/CL/cl_egl.h" && cp "/usr/local/cuda-9.0/include/CL/cl_ext.h" "$(@D)/cuda/include/CL/cl_ext.h" && cp "/usr/local/cuda-9.0/include/CL/cl_gl.h" "$(@D)/cuda/include/CL/cl_gl.h" && cp "/usr/local/cuda-9.0/include/CL/cl_gl_ext.h" "$(@D)/cuda/include/CL/cl_gl_ext.h" && cp "/usr/local/cuda-9.0/include/CL/cl_platform.h" "$(@D)/cuda/include/CL/cl_platform.h" && cp "/usr/local/cuda-9.0/include/CL/opencl.h" "$(@D)/cuda/include/CL/opencl.h" && cp "/usr/local/cuda-9.0/include/builtin_types.h" "$(@D)/cuda/include/builtin_types.h" && cp "/usr/local/cuda-9.0/include/channel_descriptor.h" "$(@D)/cuda/include/channel_descriptor.h" && cp "/usr/local/cuda-9.0/include/common_functions.h" "$(@D)/cuda/include/common_functions.h" && cp "/usr/local/cuda-9.0/include/cooperative_groups.h" "$(@D)/cuda/include/cooperative_groups.h" && cp "/usr/local/cuda-9.0/include/cooperative_groups_helpers.h" "$(@D)/cuda/include/cooperative_groups_helpers.h" && cp "/usr/local/cuda-9.0/include/crt/common_functions.h" "$(@D)/cuda/include/crt/common_functions.h" && cp "/usr/local/cuda-9.0/include/crt/device_double_functions.h" "$(@D)/cuda/include/crt/device_double_functions.h" && cp "/usr/local/cuda-9.0/include/crt/device_double_functions.hpp" "$(@D)/cuda/include/crt/device_double_functions.hpp" && cp "/usr/local/cuda-9.0/include/crt/device_functions.h" "$(@D)/cuda/include/crt/device_functions.h" && cp "/usr/local/cuda-9.0/include/crt/device_functions.hpp" "$(@D)/cuda/include/crt/device_functions.hpp" && cp "/usr/local/cuda-9.0/include/crt/func_macro.h" "$(@D)/cuda/include/crt/func_macro.h" && cp "/usr/local/cuda-9.0/include/crt/host_config.h" "$(@D)/cuda/include/crt/host_config.h" && cp "/usr/local/cuda-9.0/include/crt/host_defines.h" "$(@D)/cuda/include/crt/host_defines.h" && cp "/usr/local/cuda-9.0/include/crt/host_runtime.h" "$(@D)/cuda/include/crt/host_runtime.h" && cp "/usr/local/cuda-9.0/include/crt/math_functions.h" "$(@D)/cuda/include/crt/math_functions.h" && cp "/usr/local/cuda-9.0/include/crt/math_functions.hpp" "$(@D)/cuda/include/crt/math_functions.hpp" && cp "/usr/local/cuda-9.0/include/crt/mma.h" "$(@D)/cuda/include/crt/mma.h" && cp "/usr/local/cuda-9.0/include/crt/mma.hpp" "$(@D)/cuda/include/crt/mma.hpp" && cp "/usr/local/cuda-9.0/include/crt/nvfunctional" "$(@D)/cuda/include/crt/nvfunctional" && cp "/usr/local/cuda-9.0/include/crt/sm_70_rt.h" "$(@D)/cuda/include/crt/sm_70_rt.h" && cp "/usr/local/cuda-9.0/include/crt/sm_70_rt.hpp" "$(@D)/cuda/include/crt/sm_70_rt.hpp" && cp "/usr/local/cuda-9.0/include/crt/storage_class.h" "$(@D)/cuda/include/crt/storage_class.h" && cp "/usr/local/cuda-9.0/include/cuComplex.h" "$(@D)/cuda/include/cuComplex.h" && cp "/usr/local/cuda-9.0/include/cublas.h" "$(@D)/cuda/include/cublas.h" && cp "/usr/local/cuda-9.0/include/cublasXt.h" "$(@D)/cuda/include/cublasXt.h" && cp "/usr/local/cuda-9.0/include/cublas_api.h" "$(@D)/cuda/include/cublas_api.h" && cp "/usr/local/cuda-9.0/include/cublas_v2.h" "$(@D)/cuda/include/cublas_v2.h" && cp "/usr/local/cuda-9.0/include/cuda.h" "$(@D)/cuda/include/cuda.h" && cp "/usr/local/cuda-9.0/include/cudaEGL.h" "$(@D)/cuda/include/cudaEGL.h" && cp "/usr/local/cuda-9.0/include/cudaGL.h" "$(@D)/cuda/include/cudaGL.h" && cp "/usr/local/cuda-9.0/include/cudaProfiler.h" "$(@D)/cuda/include/cudaProfiler.h" && cp "/usr/local/cuda-9.0/include/cudaVDPAU.h" "$(@D)/cuda/include/cudaVDPAU.h" && cp "/usr/local/cuda-9.0/include/cuda_device_runtime_api.h" "$(@D)/cuda/include/cuda_device_runtime_api.h" && cp "/usr/local/cuda-9.0/include/cuda_fp16.h" "$(@D)/cuda/include/cuda_fp16.h" && cp "/usr/local/cuda-9.0/include/cuda_fp16.hpp" "$(@D)/cuda/include/cuda_fp16.hpp" && cp "/usr/local/cuda-9.0/include/cuda_gl_interop.h" "$(@D)/cuda/include/cuda_gl_interop.h" && cp "/usr/local/cuda-9.0/include/cuda_occupancy.h" "$(@D)/cuda/include/cuda_occupancy.h" && cp "/usr/local/cuda-9.0/include/cuda_profiler_api.h" "$(@D)/cuda/include/cuda_profiler_api.h" && cp "/usr/local/cuda-9.0/include/cuda_runtime.h" "$(@D)/cuda/include/cuda_runtime.h" && cp "/usr/local/cuda-9.0/include/cuda_runtime_api.h" "$(@D)/cuda/include/cuda_runtime_api.h" && cp "/usr/local/cuda-9.0/include/cuda_surface_types.h" "$(@D)/cuda/include/cuda_surface_types.h" && cp "/usr/local/cuda-9.0/include/cuda_texture_types.h" "$(@D)/cuda/include/cuda_texture_types.h" && cp "/usr/local/cuda-9.0/include/cuda_vdpau_interop.h" "$(@D)/cuda/include/cuda_vdpau_interop.h" && cp "/usr/local/cuda-9.0/include/cudalibxt.h" "$(@D)/cuda/include/cudalibxt.h" && cp "/usr/local/cuda-9.0/include/cufft.h" "$(@D)/cuda/include/cufft.h" && cp "/usr/local/cuda-9.0/include/cufftXt.h" "$(@D)/cuda/include/cufftXt.h" && cp "/usr/local/cuda-9.0/include/cufftw.h" "$(@D)/cuda/include/cufftw.h" && cp "/usr/local/cuda-9.0/include/curand.h" "$(@D)/cuda/include/curand.h" && cp "/usr/local/cuda-9.0/include/curand_discrete.h" "$(@D)/cuda/include/curand_discrete.h" && cp "/usr/local/cuda-9.0/include/curand_discrete2.h" "$(@D)/cuda/include/curand_discrete2.h" && cp "/usr/local/cuda-9.0/include/curand_globals.h" "$(@D)/cuda/include/curand_globals.h" && cp "/usr/local/cuda-9.0/include/curand_kernel.h" "$(@D)/cuda/include/curand_kernel.h" && cp "/usr/local/cuda-9.0/include/curand_lognormal.h" "$(@D)/cuda/include/curand_lognormal.h" && cp "/usr/local/cuda-9.0/include/curand_mrg32k3a.h" "$(@D)/cuda/include/curand_mrg32k3a.h" && cp "/usr/local/cuda-9.0/include/curand_mtgp32.h" "$(@D)/cuda/include/curand_mtgp32.h" && cp "/usr/local/cuda-9.0/include/curand_mtgp32_host.h" "$(@D)/cuda/include/curand_mtgp32_host.h" && cp "/usr/local/cuda-9.0/include/curand_mtgp32_kernel.h" "$(@D)/cuda/include/curand_mtgp32_kernel.h" && cp "/usr/local/cuda-9.0/include/curand_mtgp32dc_p_11213.h" "$(@D)/cuda/include/curand_mtgp32dc_p_11213.h" && cp "/usr/local/cuda-9.0/include/curand_normal.h" "$(@D)/cuda/include/curand_normal.h" && cp "/usr/local/cuda-9.0/include/curand_normal_static.h" "$(@D)/cuda/include/curand_normal_static.h" && cp "/usr/local/cuda-9.0/include/curand_philox4x32_x.h" "$(@D)/cuda/include/curand_philox4x32_x.h" && cp "/usr/local/cuda-9.0/include/curand_poisson.h" "$(@D)/cuda/include/curand_poisson.h" && cp "/usr/local/cuda-9.0/include/curand_precalc.h" "$(@D)/cuda/include/curand_precalc.h" && cp "/usr/local/cuda-9.0/include/curand_uniform.h" "$(@D)/cuda/include/curand_uniform.h" && cp "/usr/local/cuda-9.0/include/cusolverDn.h" "$(@D)/cuda/include/cusolverDn.h" && cp "/usr/local/cuda-9.0/include/cusolverRf.h" "$(@D)/cuda/include/cusolverRf.h" && cp "/usr/local/cuda-9.0/include/cusolverSp.h" "$(@D)/cuda/include/cusolverSp.h" && cp "/usr/local/cuda-9.0/include/cusolverSp_LOWLEVEL_PREVIEW.h" "$(@D)/cuda/include/cusolverSp_LOWLEVEL_PREVIEW.h" && cp "/usr/local/cuda-9.0/include/cusolver_common.h" "$(@D)/cuda/include/cusolver_common.h" && cp "/usr/local/cuda-9.0/include/cusparse.h" "$(@D)/cuda/include/cusparse.h" && cp "/usr/local/cuda-9.0/include/cusparse_v2.h" "$(@D)/cuda/include/cusparse_v2.h" && cp "/usr/local/cuda-9.0/include/device_atomic_functions.h" "$(@D)/cuda/include/device_atomic_functions.h" && cp "/usr/local/cuda-9.0/include/device_atomic_functions.hpp" "$(@D)/cuda/include/device_atomic_functions.hpp" && cp "/usr/local/cuda-9.0/include/device_double_functions.h" "$(@D)/cuda/include/device_double_functions.h" && cp "/usr/local/cuda-9.0/include/device_double_functions.hpp" "$(@D)/cuda/include/device_double_functions.hpp" && cp "/usr/local/cuda-9.0/include/device_functions.h" "$(@D)/cuda/include/device_functions.h" && cp "/usr/local/cuda-9.0/include/device_functions.hpp" "$(@D)/cuda/include/device_functions.hpp" && cp "/usr/local/cuda-9.0/include/device_functions_decls.h" "$(@D)/cuda/include/device_functions_decls.h" && cp "/usr/local/cuda-9.0/include/device_launch_parameters.h" "$(@D)/cuda/include/device_launch_parameters.h" && cp "/usr/local/cuda-9.0/include/device_types.h" "$(@D)/cuda/include/device_types.h" && cp "/usr/local/cuda-9.0/include/driver_functions.h" "$(@D)/cuda/include/driver_functions.h" && cp "/usr/local/cuda-9.0/include/driver_types.h" "$(@D)/cuda/include/driver_types.h" && cp "/usr/local/cuda-9.0/include/dynlink_cuda.h" "$(@D)/cuda/include/dynlink_cuda.h" && cp "/usr/local/cuda-9.0/include/dynlink_cuda_cuda.h" "$(@D)/cuda/include/dynlink_cuda_cuda.h" && cp "/usr/local/cuda-9.0/include/dynlink_cuviddec.h" "$(@D)/cuda/include/dynlink_cuviddec.h" && cp "/usr/local/cuda-9.0/include/dynlink_nvcuvid.h" "$(@D)/cuda/include/dynlink_nvcuvid.h" && cp "/usr/local/cuda-9.0/include/fatBinaryCtl.h" "$(@D)/cuda/include/fatBinaryCtl.h" && cp "/usr/local/cuda-9.0/include/fatbinary.h" "$(@D)/cuda/include/fatbinary.h" && cp "/usr/local/cuda-9.0/include/host_config.h" "$(@D)/cuda/include/host_config.h" && cp "/usr/local/cuda-9.0/include/host_defines.h" "$(@D)/cuda/include/host_defines.h" && cp "/usr/local/cuda-9.0/include/library_types.h" "$(@D)/cuda/include/library_types.h" && cp "/usr/local/cuda-9.0/include/math_constants.h" "$(@D)/cuda/include/math_constants.h" && cp "/usr/local/cuda-9.0/include/math_functions.h" "$(@D)/cuda/include/math_functions.h" && cp "/usr/local/cuda-9.0/include/math_functions.hpp" "$(@D)/cuda/include/math_functions.hpp" && cp "/usr/local/cuda-9.0/include/math_functions_dbl_ptx3.h" "$(@D)/cuda/include/math_functions_dbl_ptx3.h" && cp "/usr/local/cuda-9.0/include/math_functions_dbl_ptx3.hpp" "$(@D)/cuda/include/math_functions_dbl_ptx3.hpp" && cp "/usr/local/cuda-9.0/include/mma.h" "$(@D)/cuda/include/mma.h" && cp "/usr/local/cuda-9.0/include/npp.h" "$(@D)/cuda/include/npp.h" && cp "/usr/local/cuda-9.0/include/nppcore.h" "$(@D)/cuda/include/nppcore.h" && cp "/usr/local/cuda-9.0/include/nppdefs.h" "$(@D)/cuda/include/nppdefs.h" && cp "/usr/local/cuda-9.0/include/nppi.h" "$(@D)/cuda/include/nppi.h" && cp "/usr/local/cuda-9.0/include/nppi_arithmetic_and_logical_operations.h" "$(@D)/cuda/include/nppi_arithmetic_and_logical_operations.h" && cp "/usr/local/cuda-9.0/include/nppi_color_conversion.h" "$(@D)/cuda/include/nppi_color_conversion.h" && cp "/usr/local/cuda-9.0/include/nppi_compression_functions.h" "$(@D)/cuda/include/nppi_compression_functions.h" && cp "/usr/local/cuda-9.0/include/nppi_computer_vision.h" "$(@D)/cuda/include/nppi_computer_vision.h" && cp "/usr/local/cuda-9.0/include/nppi_data_exchange_and_initialization.h" "$(@D)/cuda/include/nppi_data_exchange_and_initialization.h" && cp "/usr/local/cuda-9.0/include/nppi_filtering_functions.h" "$(@D)/cuda/include/nppi_filtering_functions.h" && cp "/usr/local/cuda-9.0/include/nppi_geometry_transforms.h" "$(@D)/cuda/include/nppi_geometry_transforms.h" && cp "/usr/local/cuda-9.0/include/nppi_linear_transforms.h" "$(@D)/cuda/include/nppi_linear_transforms.h" && cp "/usr/local/cuda-9.0/include/nppi_morphological_operations.h" "$(@D)/cuda/include/nppi_morphological_operations.h" && cp "/usr/local/cuda-9.0/include/nppi_statistics_functions.h" "$(@D)/cuda/include/nppi_statistics_functions.h" && cp "/usr/local/cuda-9.0/include/nppi_support_functions.h" "$(@D)/cuda/include/nppi_support_functions.h" && cp "/usr/local/cuda-9.0/include/nppi_threshold_and_compare_operations.h" "$(@D)/cuda/include/nppi_threshold_and_compare_operations.h" && cp "/usr/local/cuda-9.0/include/npps.h" "$(@D)/cuda/include/npps.h" && cp "/usr/local/cuda-9.0/include/npps_arithmetic_and_logical_operations.h" "$(@D)/cuda/include/npps_arithmetic_and_logical_operations.h" && cp "/usr/local/cuda-9.0/include/npps_conversion_functions.h" "$(@D)/cuda/include/npps_conversion_functions.h" && cp "/usr/local/cuda-9.0/include/npps_filtering_functions.h" "$(@D)/cuda/include/npps_filtering_functions.h" && cp "/usr/local/cuda-9.0/include/npps_initialization.h" "$(@D)/cuda/include/npps_initialization.h" && cp "/usr/local/cuda-9.0/include/npps_statistics_functions.h" "$(@D)/cuda/include/npps_statistics_functions.h" && cp "/usr/local/cuda-9.0/include/npps_support_functions.h" "$(@D)/cuda/include/npps_support_functions.h" && cp "/usr/local/cuda-9.0/include/nppversion.h" "$(@D)/cuda/include/nppversion.h" && cp "/usr/local/cuda-9.0/include/nvToolsExt.h" "$(@D)/cuda/include/nvToolsExt.h" && cp "/usr/local/cuda-9.0/include/nvToolsExtCuda.h" "$(@D)/cuda/include/nvToolsExtCuda.h" && cp "/usr/local/cuda-9.0/include/nvToolsExtCudaRt.h" "$(@D)/cuda/include/nvToolsExtCudaRt.h" && cp "/usr/local/cuda-9.0/include/nvToolsExtMeta.h" "$(@D)/cuda/include/nvToolsExtMeta.h" && cp "/usr/local/cuda-9.0/include/nvToolsExtSync.h" "$(@D)/cuda/include/nvToolsExtSync.h" && cp "/usr/local/cuda-9.0/include/nvblas.h" "$(@D)/cuda/include/nvblas.h" && cp "/usr/local/cuda-9.0/include/nvfunctional" "$(@D)/cuda/include/nvfunctional" && cp "/usr/local/cuda-9.0/include/nvgraph.h" "$(@D)/cuda/include/nvgraph.h" && cp "/usr/local/cuda-9.0/include/nvml.h" "$(@D)/cuda/include/nvml.h" && cp "/usr/local/cuda-9.0/include/nvrtc.h" "$(@D)/cuda/include/nvrtc.h" && cp "/usr/local/cuda-9.0/include/sm_20_atomic_functions.h" "$(@D)/cuda/include/sm_20_atomic_functions.h" && cp "/usr/local/cuda-9.0/include/sm_20_atomic_functions.hpp" "$(@D)/cuda/include/sm_20_atomic_functions.hpp" && cp "/usr/local/cuda-9.0/include/sm_20_intrinsics.h" "$(@D)/cuda/include/sm_20_intrinsics.h" && cp "/usr/local/cuda-9.0/include/sm_20_intrinsics.hpp" "$(@D)/cuda/include/sm_20_intrinsics.hpp" && cp "/usr/local/cuda-9.0/include/sm_30_intrinsics.h" "$(@D)/cuda/include/sm_30_intrinsics.h" && cp "/usr/local/cuda-9.0/include/sm_30_intrinsics.hpp" "$(@D)/cuda/include/sm_30_intrinsics.hpp" && cp "/usr/local/cuda-9.0/include/sm_32_atomic_functions.h" "$(@D)/cuda/include/sm_32_atomic_functions.h" && cp "/usr/local/cuda-9.0/include/sm_32_atomic_functions.hpp" "$(@D)/cuda/include/sm_32_atomic_functions.hpp" && cp "/usr/local/cuda-9.0/include/sm_32_intrinsics.h" "$(@D)/cuda/include/sm_32_intrinsics.h" && cp "/usr/local/cuda-9.0/include/sm_32_intrinsics.hpp" "$(@D)/cuda/include/sm_32_intrinsics.hpp" && cp "/usr/local/cuda-9.0/include/sm_35_atomic_functions.h" "$(@D)/cuda/include/sm_35_atomic_functions.h" && cp "/usr/local/cuda-9.0/include/sm_35_intrinsics.h" "$(@D)/cuda/include/sm_35_intrinsics.h" && cp "/usr/local/cuda-9.0/include/sm_60_atomic_functions.h" "$(@D)/cuda/include/sm_60_atomic_functions.h" && cp "/usr/local/cuda-9.0/include/sm_60_atomic_functions.hpp" "$(@D)/cuda/include/sm_60_atomic_functions.hpp" && cp "/usr/local/cuda-9.0/include/sm_61_intrinsics.h" "$(@D)/cuda/include/sm_61_intrinsics.h" && cp "/usr/local/cuda-9.0/include/sm_61_intrinsics.hpp" "$(@D)/cuda/include/sm_61_intrinsics.hpp" && cp "/usr/local/cuda-9.0/include/sobol_direction_vectors.h" "$(@D)/cuda/include/sobol_direction_vectors.h" && cp "/usr/local/cuda-9.0/include/surface_functions.h" "$(@D)/cuda/include/surface_functions.h" && cp "/usr/local/cuda-9.0/include/surface_functions.hpp" "$(@D)/cuda/include/surface_functions.hpp" && cp "/usr/local/cuda-9.0/include/surface_indirect_functions.h" "$(@D)/cuda/include/surface_indirect_functions.h" && cp "/usr/local/cuda-9.0/include/surface_indirect_functions.hpp" "$(@D)/cuda/include/surface_indirect_functions.hpp" && cp "/usr/local/cuda-9.0/include/surface_types.h" "$(@D)/cuda/include/surface_types.h" && cp "/usr/local/cuda-9.0/include/texture_fetch_functions.h" "$(@D)/cuda/include/texture_fetch_functions.h" && cp "/usr/local/cuda-9.0/include/texture_fetch_functions.hpp" "$(@D)/cuda/include/texture_fetch_functions.hpp" && cp "/usr/local/cuda-9.0/include/texture_indirect_functions.h" "$(@D)/cuda/include/texture_indirect_functions.h" && cp "/usr/local/cuda-9.0/include/texture_indirect_functions.hpp" "$(@D)/cuda/include/texture_indirect_functions.hpp" && cp "/usr/local/cuda-9.0/include/texture_types.h" "$(@D)/cuda/include/texture_types.h" && cp "/usr/local/cuda-9.0/include/thrust/adjacent_difference.h" "$(@D)/cuda/include/thrust/adjacent_difference.h" && cp "/usr/local/cuda-9.0/include/thrust/advance.h" "$(@D)/cuda/include/thrust/advance.h" && cp "/usr/local/cuda-9.0/include/thrust/binary_search.h" "$(@D)/cuda/include/thrust/binary_search.h" && cp "/usr/local/cuda-9.0/include/thrust/complex.h" "$(@D)/cuda/include/thrust/complex.h" && cp "/usr/local/cuda-9.0/include/thrust/copy.h" "$(@D)/cuda/include/thrust/copy.h" && cp "/usr/local/cuda-9.0/include/thrust/count.h" "$(@D)/cuda/include/thrust/count.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/adjacent_difference.inl" "$(@D)/cuda/include/thrust/detail/adjacent_difference.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/advance.inl" "$(@D)/cuda/include/thrust/detail/advance.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/allocator/allocator_traits.h" "$(@D)/cuda/include/thrust/detail/allocator/allocator_traits.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/allocator/allocator_traits.inl" "$(@D)/cuda/include/thrust/detail/allocator/allocator_traits.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/allocator/copy_construct_range.h" "$(@D)/cuda/include/thrust/detail/allocator/copy_construct_range.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/allocator/copy_construct_range.inl" "$(@D)/cuda/include/thrust/detail/allocator/copy_construct_range.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/allocator/default_construct_range.h" "$(@D)/cuda/include/thrust/detail/allocator/default_construct_range.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/allocator/default_construct_range.inl" "$(@D)/cuda/include/thrust/detail/allocator/default_construct_range.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/allocator/destroy_range.h" "$(@D)/cuda/include/thrust/detail/allocator/destroy_range.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/allocator/destroy_range.inl" "$(@D)/cuda/include/thrust/detail/allocator/destroy_range.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/allocator/fill_construct_range.h" "$(@D)/cuda/include/thrust/detail/allocator/fill_construct_range.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/allocator/fill_construct_range.inl" "$(@D)/cuda/include/thrust/detail/allocator/fill_construct_range.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/allocator/malloc_allocator.h" "$(@D)/cuda/include/thrust/detail/allocator/malloc_allocator.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/allocator/malloc_allocator.inl" "$(@D)/cuda/include/thrust/detail/allocator/malloc_allocator.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/allocator/no_throw_allocator.h" "$(@D)/cuda/include/thrust/detail/allocator/no_throw_allocator.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/allocator/tagged_allocator.h" "$(@D)/cuda/include/thrust/detail/allocator/tagged_allocator.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/allocator/tagged_allocator.inl" "$(@D)/cuda/include/thrust/detail/allocator/tagged_allocator.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/allocator/temporary_allocator.h" "$(@D)/cuda/include/thrust/detail/allocator/temporary_allocator.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/allocator/temporary_allocator.inl" "$(@D)/cuda/include/thrust/detail/allocator/temporary_allocator.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/binary_search.inl" "$(@D)/cuda/include/thrust/detail/binary_search.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/complex/arithmetic.h" "$(@D)/cuda/include/thrust/detail/complex/arithmetic.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/complex/c99math.h" "$(@D)/cuda/include/thrust/detail/complex/c99math.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/complex/catrig.h" "$(@D)/cuda/include/thrust/detail/complex/catrig.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/complex/catrigf.h" "$(@D)/cuda/include/thrust/detail/complex/catrigf.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/complex/ccosh.h" "$(@D)/cuda/include/thrust/detail/complex/ccosh.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/complex/ccoshf.h" "$(@D)/cuda/include/thrust/detail/complex/ccoshf.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/complex/cexp.h" "$(@D)/cuda/include/thrust/detail/complex/cexp.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/complex/cexpf.h" "$(@D)/cuda/include/thrust/detail/complex/cexpf.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/complex/clog.h" "$(@D)/cuda/include/thrust/detail/complex/clog.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/complex/clogf.h" "$(@D)/cuda/include/thrust/detail/complex/clogf.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/complex/complex.inl" "$(@D)/cuda/include/thrust/detail/complex/complex.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/complex/cpow.h" "$(@D)/cuda/include/thrust/detail/complex/cpow.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/complex/cpowf.h" "$(@D)/cuda/include/thrust/detail/complex/cpowf.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/complex/cproj.h" "$(@D)/cuda/include/thrust/detail/complex/cproj.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/complex/csinh.h" "$(@D)/cuda/include/thrust/detail/complex/csinh.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/complex/csinhf.h" "$(@D)/cuda/include/thrust/detail/complex/csinhf.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/complex/csqrt.h" "$(@D)/cuda/include/thrust/detail/complex/csqrt.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/complex/csqrtf.h" "$(@D)/cuda/include/thrust/detail/complex/csqrtf.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/complex/ctanh.h" "$(@D)/cuda/include/thrust/detail/complex/ctanh.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/complex/ctanhf.h" "$(@D)/cuda/include/thrust/detail/complex/ctanhf.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/complex/math_private.h" "$(@D)/cuda/include/thrust/detail/complex/math_private.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/complex/stream.h" "$(@D)/cuda/include/thrust/detail/complex/stream.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/config.h" "$(@D)/cuda/include/thrust/detail/config.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/config/compiler.h" "$(@D)/cuda/include/thrust/detail/config/compiler.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/config/compiler_fence.h" "$(@D)/cuda/include/thrust/detail/config/compiler_fence.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/config/config.h" "$(@D)/cuda/include/thrust/detail/config/config.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/config/debug.h" "$(@D)/cuda/include/thrust/detail/config/debug.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/config/device_system.h" "$(@D)/cuda/include/thrust/detail/config/device_system.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/config/exec_check_disable.h" "$(@D)/cuda/include/thrust/detail/config/exec_check_disable.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/config/forceinline.h" "$(@D)/cuda/include/thrust/detail/config/forceinline.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/config/global_workarounds.h" "$(@D)/cuda/include/thrust/detail/config/global_workarounds.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/config/host_device.h" "$(@D)/cuda/include/thrust/detail/config/host_device.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/config/host_system.h" "$(@D)/cuda/include/thrust/detail/config/host_system.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/config/simple_defines.h" "$(@D)/cuda/include/thrust/detail/config/simple_defines.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/contiguous_storage.h" "$(@D)/cuda/include/thrust/detail/contiguous_storage.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/contiguous_storage.inl" "$(@D)/cuda/include/thrust/detail/contiguous_storage.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/copy.h" "$(@D)/cuda/include/thrust/detail/copy.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/copy.inl" "$(@D)/cuda/include/thrust/detail/copy.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/copy_if.h" "$(@D)/cuda/include/thrust/detail/copy_if.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/copy_if.inl" "$(@D)/cuda/include/thrust/detail/copy_if.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/count.inl" "$(@D)/cuda/include/thrust/detail/count.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/cstdint.h" "$(@D)/cuda/include/thrust/detail/cstdint.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/device_delete.inl" "$(@D)/cuda/include/thrust/detail/device_delete.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/device_free.inl" "$(@D)/cuda/include/thrust/detail/device_free.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/device_malloc.inl" "$(@D)/cuda/include/thrust/detail/device_malloc.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/device_new.inl" "$(@D)/cuda/include/thrust/detail/device_new.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/device_ptr.inl" "$(@D)/cuda/include/thrust/detail/device_ptr.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/device_reference.inl" "$(@D)/cuda/include/thrust/detail/device_reference.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/device_vector.inl" "$(@D)/cuda/include/thrust/detail/device_vector.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/dispatch/is_trivial_copy.h" "$(@D)/cuda/include/thrust/detail/dispatch/is_trivial_copy.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/distance.inl" "$(@D)/cuda/include/thrust/detail/distance.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/equal.inl" "$(@D)/cuda/include/thrust/detail/equal.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/execute_with_allocator.h" "$(@D)/cuda/include/thrust/detail/execute_with_allocator.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/execution_policy.h" "$(@D)/cuda/include/thrust/detail/execution_policy.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/extrema.inl" "$(@D)/cuda/include/thrust/detail/extrema.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/fill.inl" "$(@D)/cuda/include/thrust/detail/fill.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/find.inl" "$(@D)/cuda/include/thrust/detail/find.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/for_each.inl" "$(@D)/cuda/include/thrust/detail/for_each.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/function.h" "$(@D)/cuda/include/thrust/detail/function.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/functional.inl" "$(@D)/cuda/include/thrust/detail/functional.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/functional/actor.h" "$(@D)/cuda/include/thrust/detail/functional/actor.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/functional/actor.inl" "$(@D)/cuda/include/thrust/detail/functional/actor.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/functional/argument.h" "$(@D)/cuda/include/thrust/detail/functional/argument.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/functional/composite.h" "$(@D)/cuda/include/thrust/detail/functional/composite.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/functional/operators.h" "$(@D)/cuda/include/thrust/detail/functional/operators.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/functional/operators/arithmetic_operators.h" "$(@D)/cuda/include/thrust/detail/functional/operators/arithmetic_operators.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/functional/operators/assignment_operator.h" "$(@D)/cuda/include/thrust/detail/functional/operators/assignment_operator.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/functional/operators/bitwise_operators.h" "$(@D)/cuda/include/thrust/detail/functional/operators/bitwise_operators.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/functional/operators/compound_assignment_operators.h" "$(@D)/cuda/include/thrust/detail/functional/operators/compound_assignment_operators.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/functional/operators/logical_operators.h" "$(@D)/cuda/include/thrust/detail/functional/operators/logical_operators.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/functional/operators/operator_adaptors.h" "$(@D)/cuda/include/thrust/detail/functional/operators/operator_adaptors.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/functional/operators/relational_operators.h" "$(@D)/cuda/include/thrust/detail/functional/operators/relational_operators.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/functional/placeholder.h" "$(@D)/cuda/include/thrust/detail/functional/placeholder.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/functional/value.h" "$(@D)/cuda/include/thrust/detail/functional/value.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/gather.inl" "$(@D)/cuda/include/thrust/detail/gather.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/generate.inl" "$(@D)/cuda/include/thrust/detail/generate.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/get_iterator_value.h" "$(@D)/cuda/include/thrust/detail/get_iterator_value.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/host_vector.inl" "$(@D)/cuda/include/thrust/detail/host_vector.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/inner_product.inl" "$(@D)/cuda/include/thrust/detail/inner_product.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/integer_math.h" "$(@D)/cuda/include/thrust/detail/integer_math.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/integer_traits.h" "$(@D)/cuda/include/thrust/detail/integer_traits.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/internal_functional.h" "$(@D)/cuda/include/thrust/detail/internal_functional.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/logical.inl" "$(@D)/cuda/include/thrust/detail/logical.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/malloc_and_free.h" "$(@D)/cuda/include/thrust/detail/malloc_and_free.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/merge.inl" "$(@D)/cuda/include/thrust/detail/merge.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/minmax.h" "$(@D)/cuda/include/thrust/detail/minmax.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/mismatch.inl" "$(@D)/cuda/include/thrust/detail/mismatch.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/mpl/math.h" "$(@D)/cuda/include/thrust/detail/mpl/math.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/numeric_traits.h" "$(@D)/cuda/include/thrust/detail/numeric_traits.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/overlapped_copy.h" "$(@D)/cuda/include/thrust/detail/overlapped_copy.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/pair.inl" "$(@D)/cuda/include/thrust/detail/pair.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/partition.inl" "$(@D)/cuda/include/thrust/detail/partition.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/pointer.h" "$(@D)/cuda/include/thrust/detail/pointer.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/pointer.inl" "$(@D)/cuda/include/thrust/detail/pointer.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/range/head_flags.h" "$(@D)/cuda/include/thrust/detail/range/head_flags.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/range/tail_flags.h" "$(@D)/cuda/include/thrust/detail/range/tail_flags.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/raw_pointer_cast.h" "$(@D)/cuda/include/thrust/detail/raw_pointer_cast.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/raw_reference_cast.h" "$(@D)/cuda/include/thrust/detail/raw_reference_cast.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/reduce.inl" "$(@D)/cuda/include/thrust/detail/reduce.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/reference.h" "$(@D)/cuda/include/thrust/detail/reference.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/reference.inl" "$(@D)/cuda/include/thrust/detail/reference.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/reference_forward_declaration.h" "$(@D)/cuda/include/thrust/detail/reference_forward_declaration.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/remove.inl" "$(@D)/cuda/include/thrust/detail/remove.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/replace.inl" "$(@D)/cuda/include/thrust/detail/replace.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/reverse.inl" "$(@D)/cuda/include/thrust/detail/reverse.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/scan.inl" "$(@D)/cuda/include/thrust/detail/scan.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/scatter.inl" "$(@D)/cuda/include/thrust/detail/scatter.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/seq.h" "$(@D)/cuda/include/thrust/detail/seq.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/sequence.inl" "$(@D)/cuda/include/thrust/detail/sequence.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/set_operations.inl" "$(@D)/cuda/include/thrust/detail/set_operations.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/sort.inl" "$(@D)/cuda/include/thrust/detail/sort.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/static_assert.h" "$(@D)/cuda/include/thrust/detail/static_assert.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/static_map.h" "$(@D)/cuda/include/thrust/detail/static_map.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/swap.h" "$(@D)/cuda/include/thrust/detail/swap.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/swap.inl" "$(@D)/cuda/include/thrust/detail/swap.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/swap_ranges.inl" "$(@D)/cuda/include/thrust/detail/swap_ranges.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/tabulate.inl" "$(@D)/cuda/include/thrust/detail/tabulate.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/temporary_array.h" "$(@D)/cuda/include/thrust/detail/temporary_array.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/temporary_array.inl" "$(@D)/cuda/include/thrust/detail/temporary_array.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/temporary_buffer.h" "$(@D)/cuda/include/thrust/detail/temporary_buffer.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/transform.inl" "$(@D)/cuda/include/thrust/detail/transform.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/transform_reduce.inl" "$(@D)/cuda/include/thrust/detail/transform_reduce.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/transform_scan.inl" "$(@D)/cuda/include/thrust/detail/transform_scan.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/trivial_sequence.h" "$(@D)/cuda/include/thrust/detail/trivial_sequence.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/tuple.inl" "$(@D)/cuda/include/thrust/detail/tuple.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/tuple_meta_transform.h" "$(@D)/cuda/include/thrust/detail/tuple_meta_transform.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/tuple_transform.h" "$(@D)/cuda/include/thrust/detail/tuple_transform.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/type_traits.h" "$(@D)/cuda/include/thrust/detail/type_traits.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/type_traits/algorithm/intermediate_type_from_function_and_iterators.h" "$(@D)/cuda/include/thrust/detail/type_traits/algorithm/intermediate_type_from_function_and_iterators.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/type_traits/function_traits.h" "$(@D)/cuda/include/thrust/detail/type_traits/function_traits.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/type_traits/has_member_function.h" "$(@D)/cuda/include/thrust/detail/type_traits/has_member_function.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/type_traits/has_nested_type.h" "$(@D)/cuda/include/thrust/detail/type_traits/has_nested_type.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/type_traits/has_trivial_assign.h" "$(@D)/cuda/include/thrust/detail/type_traits/has_trivial_assign.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/type_traits/is_call_possible.h" "$(@D)/cuda/include/thrust/detail/type_traits/is_call_possible.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/type_traits/is_metafunction_defined.h" "$(@D)/cuda/include/thrust/detail/type_traits/is_metafunction_defined.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/type_traits/iterator/is_discard_iterator.h" "$(@D)/cuda/include/thrust/detail/type_traits/iterator/is_discard_iterator.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/type_traits/iterator/is_output_iterator.h" "$(@D)/cuda/include/thrust/detail/type_traits/iterator/is_output_iterator.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/type_traits/minimum_type.h" "$(@D)/cuda/include/thrust/detail/type_traits/minimum_type.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/type_traits/pointer_traits.h" "$(@D)/cuda/include/thrust/detail/type_traits/pointer_traits.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/type_traits/result_of_adaptable_function.h" "$(@D)/cuda/include/thrust/detail/type_traits/result_of_adaptable_function.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/uninitialized_copy.inl" "$(@D)/cuda/include/thrust/detail/uninitialized_copy.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/uninitialized_fill.inl" "$(@D)/cuda/include/thrust/detail/uninitialized_fill.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/unique.inl" "$(@D)/cuda/include/thrust/detail/unique.inl" && cp "/usr/local/cuda-9.0/include/thrust/detail/use_default.h" "$(@D)/cuda/include/thrust/detail/use_default.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/util/align.h" "$(@D)/cuda/include/thrust/detail/util/align.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/util/blocking.h" "$(@D)/cuda/include/thrust/detail/util/blocking.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/vector_base.h" "$(@D)/cuda/include/thrust/detail/vector_base.h" && cp "/usr/local/cuda-9.0/include/thrust/detail/vector_base.inl" "$(@D)/cuda/include/thrust/detail/vector_base.inl" && cp "/usr/local/cuda-9.0/include/thrust/device_allocator.h" "$(@D)/cuda/include/thrust/device_allocator.h" && cp "/usr/local/cuda-9.0/include/thrust/device_delete.h" "$(@D)/cuda/include/thrust/device_delete.h" && cp "/usr/local/cuda-9.0/include/thrust/device_free.h" "$(@D)/cuda/include/thrust/device_free.h" && cp "/usr/local/cuda-9.0/include/thrust/device_malloc.h" "$(@D)/cuda/include/thrust/device_malloc.h" && cp "/usr/local/cuda-9.0/include/thrust/device_malloc_allocator.h" "$(@D)/cuda/include/thrust/device_malloc_allocator.h" && cp "/usr/local/cuda-9.0/include/thrust/device_new.h" "$(@D)/cuda/include/thrust/device_new.h" && cp "/usr/local/cuda-9.0/include/thrust/device_new_allocator.h" "$(@D)/cuda/include/thrust/device_new_allocator.h" && cp "/usr/local/cuda-9.0/include/thrust/device_ptr.h" "$(@D)/cuda/include/thrust/device_ptr.h" && cp "/usr/local/cuda-9.0/include/thrust/device_reference.h" "$(@D)/cuda/include/thrust/device_reference.h" && cp "/usr/local/cuda-9.0/include/thrust/device_vector.h" "$(@D)/cuda/include/thrust/device_vector.h" && cp "/usr/local/cuda-9.0/include/thrust/distance.h" "$(@D)/cuda/include/thrust/distance.h" && cp "/usr/local/cuda-9.0/include/thrust/equal.h" "$(@D)/cuda/include/thrust/equal.h" && cp "/usr/local/cuda-9.0/include/thrust/execution_policy.h" "$(@D)/cuda/include/thrust/execution_policy.h" && cp "/usr/local/cuda-9.0/include/thrust/extrema.h" "$(@D)/cuda/include/thrust/extrema.h" && cp "/usr/local/cuda-9.0/include/thrust/fill.h" "$(@D)/cuda/include/thrust/fill.h" && cp "/usr/local/cuda-9.0/include/thrust/find.h" "$(@D)/cuda/include/thrust/find.h" && cp "/usr/local/cuda-9.0/include/thrust/for_each.h" "$(@D)/cuda/include/thrust/for_each.h" && cp "/usr/local/cuda-9.0/include/thrust/functional.h" "$(@D)/cuda/include/thrust/functional.h" && cp "/usr/local/cuda-9.0/include/thrust/gather.h" "$(@D)/cuda/include/thrust/gather.h" && cp "/usr/local/cuda-9.0/include/thrust/generate.h" "$(@D)/cuda/include/thrust/generate.h" && cp "/usr/local/cuda-9.0/include/thrust/host_vector.h" "$(@D)/cuda/include/thrust/host_vector.h" && cp "/usr/local/cuda-9.0/include/thrust/inner_product.h" "$(@D)/cuda/include/thrust/inner_product.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/constant_iterator.h" "$(@D)/cuda/include/thrust/iterator/constant_iterator.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/counting_iterator.h" "$(@D)/cuda/include/thrust/iterator/counting_iterator.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/any_assign.h" "$(@D)/cuda/include/thrust/iterator/detail/any_assign.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/any_system_tag.h" "$(@D)/cuda/include/thrust/iterator/detail/any_system_tag.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/constant_iterator_base.h" "$(@D)/cuda/include/thrust/iterator/detail/constant_iterator_base.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/counting_iterator.inl" "$(@D)/cuda/include/thrust/iterator/detail/counting_iterator.inl" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/device_system_tag.h" "$(@D)/cuda/include/thrust/iterator/detail/device_system_tag.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/discard_iterator_base.h" "$(@D)/cuda/include/thrust/iterator/detail/discard_iterator_base.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/distance_from_result.h" "$(@D)/cuda/include/thrust/iterator/detail/distance_from_result.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/host_system_tag.h" "$(@D)/cuda/include/thrust/iterator/detail/host_system_tag.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/is_iterator_category.h" "$(@D)/cuda/include/thrust/iterator/detail/is_iterator_category.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/is_trivial_iterator.h" "$(@D)/cuda/include/thrust/iterator/detail/is_trivial_iterator.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/iterator_adaptor_base.h" "$(@D)/cuda/include/thrust/iterator/detail/iterator_adaptor_base.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/iterator_category_to_system.h" "$(@D)/cuda/include/thrust/iterator/detail/iterator_category_to_system.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/iterator_category_to_traversal.h" "$(@D)/cuda/include/thrust/iterator/detail/iterator_category_to_traversal.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/iterator_category_with_system_and_traversal.h" "$(@D)/cuda/include/thrust/iterator/detail/iterator_category_with_system_and_traversal.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/iterator_facade_category.h" "$(@D)/cuda/include/thrust/iterator/detail/iterator_facade_category.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/iterator_traits.inl" "$(@D)/cuda/include/thrust/iterator/detail/iterator_traits.inl" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/iterator_traversal_tags.h" "$(@D)/cuda/include/thrust/iterator/detail/iterator_traversal_tags.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/join_iterator.h" "$(@D)/cuda/include/thrust/iterator/detail/join_iterator.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/minimum_category.h" "$(@D)/cuda/include/thrust/iterator/detail/minimum_category.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/minimum_system.h" "$(@D)/cuda/include/thrust/iterator/detail/minimum_system.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/normal_iterator.h" "$(@D)/cuda/include/thrust/iterator/detail/normal_iterator.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/permutation_iterator_base.h" "$(@D)/cuda/include/thrust/iterator/detail/permutation_iterator_base.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/retag.h" "$(@D)/cuda/include/thrust/iterator/detail/retag.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/reverse_iterator.inl" "$(@D)/cuda/include/thrust/iterator/detail/reverse_iterator.inl" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/reverse_iterator_base.h" "$(@D)/cuda/include/thrust/iterator/detail/reverse_iterator_base.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/tagged_iterator.h" "$(@D)/cuda/include/thrust/iterator/detail/tagged_iterator.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/transform_iterator.inl" "$(@D)/cuda/include/thrust/iterator/detail/transform_iterator.inl" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/transform_output_iterator.inl" "$(@D)/cuda/include/thrust/iterator/detail/transform_output_iterator.inl" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/tuple_of_iterator_references.h" "$(@D)/cuda/include/thrust/iterator/detail/tuple_of_iterator_references.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/universal_categories.h" "$(@D)/cuda/include/thrust/iterator/detail/universal_categories.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/zip_iterator.inl" "$(@D)/cuda/include/thrust/iterator/detail/zip_iterator.inl" && cp "/usr/local/cuda-9.0/include/thrust/iterator/detail/zip_iterator_base.h" "$(@D)/cuda/include/thrust/iterator/detail/zip_iterator_base.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/discard_iterator.h" "$(@D)/cuda/include/thrust/iterator/discard_iterator.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/iterator_adaptor.h" "$(@D)/cuda/include/thrust/iterator/iterator_adaptor.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/iterator_categories.h" "$(@D)/cuda/include/thrust/iterator/iterator_categories.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/iterator_facade.h" "$(@D)/cuda/include/thrust/iterator/iterator_facade.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/iterator_traits.h" "$(@D)/cuda/include/thrust/iterator/iterator_traits.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/permutation_iterator.h" "$(@D)/cuda/include/thrust/iterator/permutation_iterator.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/retag.h" "$(@D)/cuda/include/thrust/iterator/retag.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/reverse_iterator.h" "$(@D)/cuda/include/thrust/iterator/reverse_iterator.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/transform_iterator.h" "$(@D)/cuda/include/thrust/iterator/transform_iterator.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/transform_output_iterator.h" "$(@D)/cuda/include/thrust/iterator/transform_output_iterator.h" && cp "/usr/local/cuda-9.0/include/thrust/iterator/zip_iterator.h" "$(@D)/cuda/include/thrust/iterator/zip_iterator.h" && cp "/usr/local/cuda-9.0/include/thrust/logical.h" "$(@D)/cuda/include/thrust/logical.h" && cp "/usr/local/cuda-9.0/include/thrust/memory.h" "$(@D)/cuda/include/thrust/memory.h" && cp "/usr/local/cuda-9.0/include/thrust/merge.h" "$(@D)/cuda/include/thrust/merge.h" && cp "/usr/local/cuda-9.0/include/thrust/mismatch.h" "$(@D)/cuda/include/thrust/mismatch.h" && cp "/usr/local/cuda-9.0/include/thrust/pair.h" "$(@D)/cuda/include/thrust/pair.h" && cp "/usr/local/cuda-9.0/include/thrust/partition.h" "$(@D)/cuda/include/thrust/partition.h" && cp "/usr/local/cuda-9.0/include/thrust/random.h" "$(@D)/cuda/include/thrust/random.h" && cp "/usr/local/cuda-9.0/include/thrust/random/detail/discard_block_engine.inl" "$(@D)/cuda/include/thrust/random/detail/discard_block_engine.inl" && cp "/usr/local/cuda-9.0/include/thrust/random/detail/linear_congruential_engine.inl" "$(@D)/cuda/include/thrust/random/detail/linear_congruential_engine.inl" && cp "/usr/local/cuda-9.0/include/thrust/random/detail/linear_congruential_engine_discard.h" "$(@D)/cuda/include/thrust/random/detail/linear_congruential_engine_discard.h" && cp "/usr/local/cuda-9.0/include/thrust/random/detail/linear_feedback_shift_engine.inl" "$(@D)/cuda/include/thrust/random/detail/linear_feedback_shift_engine.inl" && cp "/usr/local/cuda-9.0/include/thrust/random/detail/linear_feedback_shift_engine_wordmask.h" "$(@D)/cuda/include/thrust/random/detail/linear_feedback_shift_engine_wordmask.h" && cp "/usr/local/cuda-9.0/include/thrust/random/detail/mod.h" "$(@D)/cuda/include/thrust/random/detail/mod.h" && cp "/usr/local/cuda-9.0/include/thrust/random/detail/normal_distribution.inl" "$(@D)/cuda/include/thrust/random/detail/normal_distribution.inl" && cp "/usr/local/cuda-9.0/include/thrust/random/detail/normal_distribution_base.h" "$(@D)/cuda/include/thrust/random/detail/normal_distribution_base.h" && cp "/usr/local/cuda-9.0/include/thrust/random/detail/random_core_access.h" "$(@D)/cuda/include/thrust/random/detail/random_core_access.h" && cp "/usr/local/cuda-9.0/include/thrust/random/detail/subtract_with_carry_engine.inl" "$(@D)/cuda/include/thrust/random/detail/subtract_with_carry_engine.inl" && cp "/usr/local/cuda-9.0/include/thrust/random/detail/uniform_int_distribution.inl" "$(@D)/cuda/include/thrust/random/detail/uniform_int_distribution.inl" && cp "/usr/local/cuda-9.0/include/thrust/random/detail/uniform_real_distribution.inl" "$(@D)/cuda/include/thrust/random/detail/uniform_real_distribution.inl" && cp "/usr/local/cuda-9.0/include/thrust/random/detail/xor_combine_engine.inl" "$(@D)/cuda/include/thrust/random/detail/xor_combine_engine.inl" && cp "/usr/local/cuda-9.0/include/thrust/random/detail/xor_combine_engine_max.h" "$(@D)/cuda/include/thrust/random/detail/xor_combine_engine_max.h" && cp "/usr/local/cuda-9.0/include/thrust/random/discard_block_engine.h" "$(@D)/cuda/include/thrust/random/discard_block_engine.h" && cp "/usr/local/cuda-9.0/include/thrust/random/linear_congruential_engine.h" "$(@D)/cuda/include/thrust/random/linear_congruential_engine.h" && cp "/usr/local/cuda-9.0/include/thrust/random/linear_feedback_shift_engine.h" "$(@D)/cuda/include/thrust/random/linear_feedback_shift_engine.h" && cp "/usr/local/cuda-9.0/include/thrust/random/normal_distribution.h" "$(@D)/cuda/include/thrust/random/normal_distribution.h" && cp "/usr/local/cuda-9.0/include/thrust/random/subtract_with_carry_engine.h" "$(@D)/cuda/include/thrust/random/subtract_with_carry_engine.h" && cp "/usr/local/cuda-9.0/include/thrust/random/uniform_int_distribution.h" "$(@D)/cuda/include/thrust/random/uniform_int_distribution.h" && cp "/usr/local/cuda-9.0/include/thrust/random/uniform_real_distribution.h" "$(@D)/cuda/include/thrust/random/uniform_real_distribution.h" && cp "/usr/local/cuda-9.0/include/thrust/random/xor_combine_engine.h" "$(@D)/cuda/include/thrust/random/xor_combine_engine.h" && cp "/usr/local/cuda-9.0/include/thrust/reduce.h" "$(@D)/cuda/include/thrust/reduce.h" && cp "/usr/local/cuda-9.0/include/thrust/remove.h" "$(@D)/cuda/include/thrust/remove.h" && cp "/usr/local/cuda-9.0/include/thrust/replace.h" "$(@D)/cuda/include/thrust/replace.h" && cp "/usr/local/cuda-9.0/include/thrust/reverse.h" "$(@D)/cuda/include/thrust/reverse.h" && cp "/usr/local/cuda-9.0/include/thrust/scan.h" "$(@D)/cuda/include/thrust/scan.h" && cp "/usr/local/cuda-9.0/include/thrust/scatter.h" "$(@D)/cuda/include/thrust/scatter.h" && cp "/usr/local/cuda-9.0/include/thrust/sequence.h" "$(@D)/cuda/include/thrust/sequence.h" && cp "/usr/local/cuda-9.0/include/thrust/set_operations.h" "$(@D)/cuda/include/thrust/set_operations.h" && cp "/usr/local/cuda-9.0/include/thrust/sort.h" "$(@D)/cuda/include/thrust/sort.h" && cp "/usr/local/cuda-9.0/include/thrust/swap.h" "$(@D)/cuda/include/thrust/swap.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/adjacent_difference.h" "$(@D)/cuda/include/thrust/system/cpp/detail/adjacent_difference.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/assign_value.h" "$(@D)/cuda/include/thrust/system/cpp/detail/assign_value.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/binary_search.h" "$(@D)/cuda/include/thrust/system/cpp/detail/binary_search.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/copy.h" "$(@D)/cuda/include/thrust/system/cpp/detail/copy.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/copy_if.h" "$(@D)/cuda/include/thrust/system/cpp/detail/copy_if.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/count.h" "$(@D)/cuda/include/thrust/system/cpp/detail/count.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/equal.h" "$(@D)/cuda/include/thrust/system/cpp/detail/equal.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/execution_policy.h" "$(@D)/cuda/include/thrust/system/cpp/detail/execution_policy.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/extrema.h" "$(@D)/cuda/include/thrust/system/cpp/detail/extrema.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/fill.h" "$(@D)/cuda/include/thrust/system/cpp/detail/fill.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/find.h" "$(@D)/cuda/include/thrust/system/cpp/detail/find.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/for_each.h" "$(@D)/cuda/include/thrust/system/cpp/detail/for_each.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/gather.h" "$(@D)/cuda/include/thrust/system/cpp/detail/gather.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/generate.h" "$(@D)/cuda/include/thrust/system/cpp/detail/generate.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/get_value.h" "$(@D)/cuda/include/thrust/system/cpp/detail/get_value.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/inner_product.h" "$(@D)/cuda/include/thrust/system/cpp/detail/inner_product.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/iter_swap.h" "$(@D)/cuda/include/thrust/system/cpp/detail/iter_swap.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/logical.h" "$(@D)/cuda/include/thrust/system/cpp/detail/logical.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/malloc_and_free.h" "$(@D)/cuda/include/thrust/system/cpp/detail/malloc_and_free.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/memory.inl" "$(@D)/cuda/include/thrust/system/cpp/detail/memory.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/merge.h" "$(@D)/cuda/include/thrust/system/cpp/detail/merge.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/mismatch.h" "$(@D)/cuda/include/thrust/system/cpp/detail/mismatch.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/par.h" "$(@D)/cuda/include/thrust/system/cpp/detail/par.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/partition.h" "$(@D)/cuda/include/thrust/system/cpp/detail/partition.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/reduce.h" "$(@D)/cuda/include/thrust/system/cpp/detail/reduce.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/reduce_by_key.h" "$(@D)/cuda/include/thrust/system/cpp/detail/reduce_by_key.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/remove.h" "$(@D)/cuda/include/thrust/system/cpp/detail/remove.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/replace.h" "$(@D)/cuda/include/thrust/system/cpp/detail/replace.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/reverse.h" "$(@D)/cuda/include/thrust/system/cpp/detail/reverse.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/scan.h" "$(@D)/cuda/include/thrust/system/cpp/detail/scan.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/scan_by_key.h" "$(@D)/cuda/include/thrust/system/cpp/detail/scan_by_key.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/scatter.h" "$(@D)/cuda/include/thrust/system/cpp/detail/scatter.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/sequence.h" "$(@D)/cuda/include/thrust/system/cpp/detail/sequence.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/set_operations.h" "$(@D)/cuda/include/thrust/system/cpp/detail/set_operations.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/sort.h" "$(@D)/cuda/include/thrust/system/cpp/detail/sort.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/swap_ranges.h" "$(@D)/cuda/include/thrust/system/cpp/detail/swap_ranges.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/tabulate.h" "$(@D)/cuda/include/thrust/system/cpp/detail/tabulate.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/temporary_buffer.h" "$(@D)/cuda/include/thrust/system/cpp/detail/temporary_buffer.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/transform.h" "$(@D)/cuda/include/thrust/system/cpp/detail/transform.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/transform_reduce.h" "$(@D)/cuda/include/thrust/system/cpp/detail/transform_reduce.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/transform_scan.h" "$(@D)/cuda/include/thrust/system/cpp/detail/transform_scan.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/uninitialized_copy.h" "$(@D)/cuda/include/thrust/system/cpp/detail/uninitialized_copy.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/uninitialized_fill.h" "$(@D)/cuda/include/thrust/system/cpp/detail/uninitialized_fill.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/unique.h" "$(@D)/cuda/include/thrust/system/cpp/detail/unique.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/unique_by_key.h" "$(@D)/cuda/include/thrust/system/cpp/detail/unique_by_key.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/detail/vector.inl" "$(@D)/cuda/include/thrust/system/cpp/detail/vector.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/execution_policy.h" "$(@D)/cuda/include/thrust/system/cpp/execution_policy.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/memory.h" "$(@D)/cuda/include/thrust/system/cpp/memory.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cpp/vector.h" "$(@D)/cuda/include/thrust/system/cpp/vector.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/config.h" "$(@D)/cuda/include/thrust/system/cuda/config.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/adjacent_difference.h" "$(@D)/cuda/include/thrust/system/cuda/detail/adjacent_difference.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/assign_value.h" "$(@D)/cuda/include/thrust/system/cuda/detail/assign_value.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/binary_search.h" "$(@D)/cuda/include/thrust/system/cuda/detail/binary_search.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/copy.h" "$(@D)/cuda/include/thrust/system/cuda/detail/copy.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/copy_if.h" "$(@D)/cuda/include/thrust/system/cuda/detail/copy_if.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/core/agent_launcher.h" "$(@D)/cuda/include/thrust/system/cuda/detail/core/agent_launcher.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/core/alignment.h" "$(@D)/cuda/include/thrust/system/cuda/detail/core/alignment.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/core/triple_chevron_launch.h" "$(@D)/cuda/include/thrust/system/cuda/detail/core/triple_chevron_launch.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/core/util.h" "$(@D)/cuda/include/thrust/system/cuda/detail/core/util.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/count.h" "$(@D)/cuda/include/thrust/system/cuda/detail/count.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cross_system.h" "$(@D)/cuda/include/thrust/system/cuda/detail/cross_system.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/agent/agent_histogram.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/agent/agent_histogram.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/agent/agent_radix_sort_downsweep.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/agent/agent_radix_sort_downsweep.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/agent/agent_radix_sort_upsweep.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/agent/agent_radix_sort_upsweep.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/agent/agent_reduce.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/agent/agent_reduce.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/agent/agent_reduce_by_key.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/agent/agent_reduce_by_key.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/agent/agent_rle.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/agent/agent_rle.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/agent/agent_scan.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/agent/agent_scan.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/agent/agent_segment_fixup.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/agent/agent_segment_fixup.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/agent/agent_select_if.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/agent/agent_select_if.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/agent/agent_spmv_csrt.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/agent/agent_spmv_csrt.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/agent/agent_spmv_orig.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/agent/agent_spmv_orig.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/agent/agent_spmv_row_based.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/agent/agent_spmv_row_based.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/agent/single_pass_scan_operators.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/agent/single_pass_scan_operators.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/block/block_adjacent_difference.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/block/block_adjacent_difference.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/block/block_discontinuity.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/block/block_discontinuity.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/block/block_exchange.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/block/block_exchange.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/block/block_histogram.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/block/block_histogram.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/block/block_load.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/block/block_load.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/block/block_radix_rank.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/block/block_radix_rank.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/block/block_radix_sort.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/block/block_radix_sort.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/block/block_raking_layout.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/block/block_raking_layout.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/block/block_reduce.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/block/block_reduce.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/block/block_scan.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/block/block_scan.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/block/block_shuffle.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/block/block_shuffle.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/block/block_store.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/block/block_store.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/block/specializations/block_histogram_atomic.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/block/specializations/block_histogram_atomic.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/block/specializations/block_histogram_sort.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/block/specializations/block_histogram_sort.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/block/specializations/block_reduce_raking.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/block/specializations/block_reduce_raking.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/block/specializations/block_reduce_raking_commutative_only.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/block/specializations/block_reduce_raking_commutative_only.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/block/specializations/block_reduce_warp_reductions.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/block/specializations/block_reduce_warp_reductions.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/block/specializations/block_scan_raking.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/block/specializations/block_scan_raking.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/block/specializations/block_scan_warp_scans.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/block/specializations/block_scan_warp_scans.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/block/specializations/block_scan_warp_scans2.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/block/specializations/block_scan_warp_scans2.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/block/specializations/block_scan_warp_scans3.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/block/specializations/block_scan_warp_scans3.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/cub.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/cub.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/device/device_histogram.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/device/device_histogram.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/device/device_partition.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/device/device_partition.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/device/device_radix_sort.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/device/device_radix_sort.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/device/device_reduce.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/device/device_reduce.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/device/device_run_length_encode.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/device/device_run_length_encode.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/device/device_scan.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/device/device_scan.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/device/device_segmented_radix_sort.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/device/device_segmented_radix_sort.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/device/device_segmented_reduce.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/device/device_segmented_reduce.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/device/device_select.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/device/device_select.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/device/device_spmv.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/device/device_spmv.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_histogram.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_histogram.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_radix_sort.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_radix_sort.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_reduce.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_reduce.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_reduce_by_key.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_reduce_by_key.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_rle.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_rle.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_scan.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_scan.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_select_if.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_select_if.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_spmv_csrt.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_spmv_csrt.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_spmv_orig.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_spmv_orig.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_spmv_row_based.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/device/dispatch/dispatch_spmv_row_based.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/grid/grid_barrier.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/grid/grid_barrier.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/grid/grid_even_share.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/grid/grid_even_share.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/grid/grid_mapping.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/grid/grid_mapping.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/grid/grid_queue.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/grid/grid_queue.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/host/mutex.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/host/mutex.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/iterator/arg_index_input_iterator.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/iterator/arg_index_input_iterator.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/iterator/cache_modified_input_iterator.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/iterator/cache_modified_input_iterator.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/iterator/cache_modified_output_iterator.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/iterator/cache_modified_output_iterator.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/iterator/constant_input_iterator.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/iterator/constant_input_iterator.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/iterator/counting_input_iterator.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/iterator/counting_input_iterator.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/iterator/discard_output_iterator.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/iterator/discard_output_iterator.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/iterator/tex_obj_input_iterator.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/iterator/tex_obj_input_iterator.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/iterator/tex_ref_input_iterator.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/iterator/tex_ref_input_iterator.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/iterator/transform_input_iterator.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/iterator/transform_input_iterator.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/thread/thread_load.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/thread/thread_load.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/thread/thread_operators.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/thread/thread_operators.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/thread/thread_reduce.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/thread/thread_reduce.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/thread/thread_scan.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/thread/thread_scan.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/thread/thread_search.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/thread/thread_search.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/thread/thread_store.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/thread/thread_store.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/util_allocator.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/util_allocator.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/util_arch.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/util_arch.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/util_debug.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/util_debug.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/util_device.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/util_device.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/util_macro.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/util_macro.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/util_namespace.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/util_namespace.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/util_ptx.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/util_ptx.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/util_type.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/util_type.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/warp/specializations/warp_reduce_shfl.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/warp/specializations/warp_reduce_shfl.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/warp/specializations/warp_reduce_smem.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/warp/specializations/warp_reduce_smem.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/warp/specializations/warp_scan_shfl.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/warp/specializations/warp_scan_shfl.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/warp/specializations/warp_scan_smem.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/warp/specializations/warp_scan_smem.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/warp/warp_reduce.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/warp/warp_reduce.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/cub/warp/warp_scan.cuh" "$(@D)/cuda/include/thrust/system/cuda/detail/cub/warp/warp_scan.cuh" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/equal.h" "$(@D)/cuda/include/thrust/system/cuda/detail/equal.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/error.inl" "$(@D)/cuda/include/thrust/system/cuda/detail/error.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/execution_policy.h" "$(@D)/cuda/include/thrust/system/cuda/detail/execution_policy.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/extrema.h" "$(@D)/cuda/include/thrust/system/cuda/detail/extrema.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/fill.h" "$(@D)/cuda/include/thrust/system/cuda/detail/fill.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/find.h" "$(@D)/cuda/include/thrust/system/cuda/detail/find.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/for_each.h" "$(@D)/cuda/include/thrust/system/cuda/detail/for_each.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/gather.h" "$(@D)/cuda/include/thrust/system/cuda/detail/gather.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/generate.h" "$(@D)/cuda/include/thrust/system/cuda/detail/generate.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/get_value.h" "$(@D)/cuda/include/thrust/system/cuda/detail/get_value.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/guarded_cuda_runtime_api.h" "$(@D)/cuda/include/thrust/system/cuda/detail/guarded_cuda_runtime_api.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/guarded_driver_types.h" "$(@D)/cuda/include/thrust/system/cuda/detail/guarded_driver_types.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/inner_product.h" "$(@D)/cuda/include/thrust/system/cuda/detail/inner_product.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/internal/copy_cross_system.h" "$(@D)/cuda/include/thrust/system/cuda/detail/internal/copy_cross_system.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/internal/copy_device_to_device.h" "$(@D)/cuda/include/thrust/system/cuda/detail/internal/copy_device_to_device.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/iter_swap.h" "$(@D)/cuda/include/thrust/system/cuda/detail/iter_swap.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/logical.h" "$(@D)/cuda/include/thrust/system/cuda/detail/logical.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/malloc_and_free.h" "$(@D)/cuda/include/thrust/system/cuda/detail/malloc_and_free.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/memory.inl" "$(@D)/cuda/include/thrust/system/cuda/detail/memory.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/memory_buffer.h" "$(@D)/cuda/include/thrust/system/cuda/detail/memory_buffer.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/merge.h" "$(@D)/cuda/include/thrust/system/cuda/detail/merge.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/mismatch.h" "$(@D)/cuda/include/thrust/system/cuda/detail/mismatch.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/par.h" "$(@D)/cuda/include/thrust/system/cuda/detail/par.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/par_to_seq.h" "$(@D)/cuda/include/thrust/system/cuda/detail/par_to_seq.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/parallel_for.h" "$(@D)/cuda/include/thrust/system/cuda/detail/parallel_for.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/partition.h" "$(@D)/cuda/include/thrust/system/cuda/detail/partition.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/reduce.h" "$(@D)/cuda/include/thrust/system/cuda/detail/reduce.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/reduce_by_key.h" "$(@D)/cuda/include/thrust/system/cuda/detail/reduce_by_key.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/remove.h" "$(@D)/cuda/include/thrust/system/cuda/detail/remove.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/replace.h" "$(@D)/cuda/include/thrust/system/cuda/detail/replace.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/reverse.h" "$(@D)/cuda/include/thrust/system/cuda/detail/reverse.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/scan.h" "$(@D)/cuda/include/thrust/system/cuda/detail/scan.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/scan_by_key.h" "$(@D)/cuda/include/thrust/system/cuda/detail/scan_by_key.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/scatter.h" "$(@D)/cuda/include/thrust/system/cuda/detail/scatter.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/sequence.h" "$(@D)/cuda/include/thrust/system/cuda/detail/sequence.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/set_operations.h" "$(@D)/cuda/include/thrust/system/cuda/detail/set_operations.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/sort.h" "$(@D)/cuda/include/thrust/system/cuda/detail/sort.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/swap_ranges.h" "$(@D)/cuda/include/thrust/system/cuda/detail/swap_ranges.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/tabulate.h" "$(@D)/cuda/include/thrust/system/cuda/detail/tabulate.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/temporary_buffer.h" "$(@D)/cuda/include/thrust/system/cuda/detail/temporary_buffer.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/terminate.h" "$(@D)/cuda/include/thrust/system/cuda/detail/terminate.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/transform.h" "$(@D)/cuda/include/thrust/system/cuda/detail/transform.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/transform_reduce.h" "$(@D)/cuda/include/thrust/system/cuda/detail/transform_reduce.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/transform_scan.h" "$(@D)/cuda/include/thrust/system/cuda/detail/transform_scan.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/uninitialized_copy.h" "$(@D)/cuda/include/thrust/system/cuda/detail/uninitialized_copy.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/uninitialized_fill.h" "$(@D)/cuda/include/thrust/system/cuda/detail/uninitialized_fill.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/unique.h" "$(@D)/cuda/include/thrust/system/cuda/detail/unique.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/unique_by_key.h" "$(@D)/cuda/include/thrust/system/cuda/detail/unique_by_key.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/util.h" "$(@D)/cuda/include/thrust/system/cuda/detail/util.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/detail/vector.inl" "$(@D)/cuda/include/thrust/system/cuda/detail/vector.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/error.h" "$(@D)/cuda/include/thrust/system/cuda/error.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/execution_policy.h" "$(@D)/cuda/include/thrust/system/cuda/execution_policy.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/experimental/pinned_allocator.h" "$(@D)/cuda/include/thrust/system/cuda/experimental/pinned_allocator.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/memory.h" "$(@D)/cuda/include/thrust/system/cuda/memory.h" && cp "/usr/local/cuda-9.0/include/thrust/system/cuda/vector.h" "$(@D)/cuda/include/thrust/system/cuda/vector.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/adjacent_difference.h" "$(@D)/cuda/include/thrust/system/detail/adl/adjacent_difference.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/assign_value.h" "$(@D)/cuda/include/thrust/system/detail/adl/assign_value.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/binary_search.h" "$(@D)/cuda/include/thrust/system/detail/adl/binary_search.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/copy.h" "$(@D)/cuda/include/thrust/system/detail/adl/copy.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/copy_if.h" "$(@D)/cuda/include/thrust/system/detail/adl/copy_if.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/count.h" "$(@D)/cuda/include/thrust/system/detail/adl/count.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/equal.h" "$(@D)/cuda/include/thrust/system/detail/adl/equal.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/extrema.h" "$(@D)/cuda/include/thrust/system/detail/adl/extrema.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/fill.h" "$(@D)/cuda/include/thrust/system/detail/adl/fill.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/find.h" "$(@D)/cuda/include/thrust/system/detail/adl/find.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/for_each.h" "$(@D)/cuda/include/thrust/system/detail/adl/for_each.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/gather.h" "$(@D)/cuda/include/thrust/system/detail/adl/gather.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/generate.h" "$(@D)/cuda/include/thrust/system/detail/adl/generate.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/get_value.h" "$(@D)/cuda/include/thrust/system/detail/adl/get_value.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/inner_product.h" "$(@D)/cuda/include/thrust/system/detail/adl/inner_product.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/iter_swap.h" "$(@D)/cuda/include/thrust/system/detail/adl/iter_swap.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/logical.h" "$(@D)/cuda/include/thrust/system/detail/adl/logical.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/malloc_and_free.h" "$(@D)/cuda/include/thrust/system/detail/adl/malloc_and_free.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/merge.h" "$(@D)/cuda/include/thrust/system/detail/adl/merge.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/mismatch.h" "$(@D)/cuda/include/thrust/system/detail/adl/mismatch.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/partition.h" "$(@D)/cuda/include/thrust/system/detail/adl/partition.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/reduce.h" "$(@D)/cuda/include/thrust/system/detail/adl/reduce.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/reduce_by_key.h" "$(@D)/cuda/include/thrust/system/detail/adl/reduce_by_key.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/remove.h" "$(@D)/cuda/include/thrust/system/detail/adl/remove.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/replace.h" "$(@D)/cuda/include/thrust/system/detail/adl/replace.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/reverse.h" "$(@D)/cuda/include/thrust/system/detail/adl/reverse.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/scan.h" "$(@D)/cuda/include/thrust/system/detail/adl/scan.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/scan_by_key.h" "$(@D)/cuda/include/thrust/system/detail/adl/scan_by_key.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/scatter.h" "$(@D)/cuda/include/thrust/system/detail/adl/scatter.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/sequence.h" "$(@D)/cuda/include/thrust/system/detail/adl/sequence.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/set_operations.h" "$(@D)/cuda/include/thrust/system/detail/adl/set_operations.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/sort.h" "$(@D)/cuda/include/thrust/system/detail/adl/sort.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/swap_ranges.h" "$(@D)/cuda/include/thrust/system/detail/adl/swap_ranges.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/tabulate.h" "$(@D)/cuda/include/thrust/system/detail/adl/tabulate.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/temporary_buffer.h" "$(@D)/cuda/include/thrust/system/detail/adl/temporary_buffer.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/transform.h" "$(@D)/cuda/include/thrust/system/detail/adl/transform.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/transform_reduce.h" "$(@D)/cuda/include/thrust/system/detail/adl/transform_reduce.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/transform_scan.h" "$(@D)/cuda/include/thrust/system/detail/adl/transform_scan.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/uninitialized_copy.h" "$(@D)/cuda/include/thrust/system/detail/adl/uninitialized_copy.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/uninitialized_fill.h" "$(@D)/cuda/include/thrust/system/detail/adl/uninitialized_fill.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/unique.h" "$(@D)/cuda/include/thrust/system/detail/adl/unique.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/adl/unique_by_key.h" "$(@D)/cuda/include/thrust/system/detail/adl/unique_by_key.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/bad_alloc.h" "$(@D)/cuda/include/thrust/system/detail/bad_alloc.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/errno.h" "$(@D)/cuda/include/thrust/system/detail/errno.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/error_category.inl" "$(@D)/cuda/include/thrust/system/detail/error_category.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/error_code.inl" "$(@D)/cuda/include/thrust/system/detail/error_code.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/error_condition.inl" "$(@D)/cuda/include/thrust/system/detail/error_condition.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/adjacent_difference.h" "$(@D)/cuda/include/thrust/system/detail/generic/adjacent_difference.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/adjacent_difference.inl" "$(@D)/cuda/include/thrust/system/detail/generic/adjacent_difference.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/advance.h" "$(@D)/cuda/include/thrust/system/detail/generic/advance.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/advance.inl" "$(@D)/cuda/include/thrust/system/detail/generic/advance.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/binary_search.h" "$(@D)/cuda/include/thrust/system/detail/generic/binary_search.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/binary_search.inl" "$(@D)/cuda/include/thrust/system/detail/generic/binary_search.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/copy.h" "$(@D)/cuda/include/thrust/system/detail/generic/copy.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/copy.inl" "$(@D)/cuda/include/thrust/system/detail/generic/copy.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/copy_if.h" "$(@D)/cuda/include/thrust/system/detail/generic/copy_if.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/copy_if.inl" "$(@D)/cuda/include/thrust/system/detail/generic/copy_if.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/count.h" "$(@D)/cuda/include/thrust/system/detail/generic/count.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/count.inl" "$(@D)/cuda/include/thrust/system/detail/generic/count.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/distance.h" "$(@D)/cuda/include/thrust/system/detail/generic/distance.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/distance.inl" "$(@D)/cuda/include/thrust/system/detail/generic/distance.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/equal.h" "$(@D)/cuda/include/thrust/system/detail/generic/equal.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/equal.inl" "$(@D)/cuda/include/thrust/system/detail/generic/equal.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/extrema.h" "$(@D)/cuda/include/thrust/system/detail/generic/extrema.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/extrema.inl" "$(@D)/cuda/include/thrust/system/detail/generic/extrema.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/fill.h" "$(@D)/cuda/include/thrust/system/detail/generic/fill.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/find.h" "$(@D)/cuda/include/thrust/system/detail/generic/find.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/find.inl" "$(@D)/cuda/include/thrust/system/detail/generic/find.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/for_each.h" "$(@D)/cuda/include/thrust/system/detail/generic/for_each.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/gather.h" "$(@D)/cuda/include/thrust/system/detail/generic/gather.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/gather.inl" "$(@D)/cuda/include/thrust/system/detail/generic/gather.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/generate.h" "$(@D)/cuda/include/thrust/system/detail/generic/generate.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/generate.inl" "$(@D)/cuda/include/thrust/system/detail/generic/generate.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/inner_product.h" "$(@D)/cuda/include/thrust/system/detail/generic/inner_product.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/inner_product.inl" "$(@D)/cuda/include/thrust/system/detail/generic/inner_product.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/logical.h" "$(@D)/cuda/include/thrust/system/detail/generic/logical.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/memory.h" "$(@D)/cuda/include/thrust/system/detail/generic/memory.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/memory.inl" "$(@D)/cuda/include/thrust/system/detail/generic/memory.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/merge.h" "$(@D)/cuda/include/thrust/system/detail/generic/merge.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/merge.inl" "$(@D)/cuda/include/thrust/system/detail/generic/merge.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/mismatch.h" "$(@D)/cuda/include/thrust/system/detail/generic/mismatch.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/mismatch.inl" "$(@D)/cuda/include/thrust/system/detail/generic/mismatch.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/partition.h" "$(@D)/cuda/include/thrust/system/detail/generic/partition.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/partition.inl" "$(@D)/cuda/include/thrust/system/detail/generic/partition.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/reduce.h" "$(@D)/cuda/include/thrust/system/detail/generic/reduce.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/reduce.inl" "$(@D)/cuda/include/thrust/system/detail/generic/reduce.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/reduce_by_key.h" "$(@D)/cuda/include/thrust/system/detail/generic/reduce_by_key.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/reduce_by_key.inl" "$(@D)/cuda/include/thrust/system/detail/generic/reduce_by_key.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/remove.h" "$(@D)/cuda/include/thrust/system/detail/generic/remove.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/remove.inl" "$(@D)/cuda/include/thrust/system/detail/generic/remove.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/replace.h" "$(@D)/cuda/include/thrust/system/detail/generic/replace.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/replace.inl" "$(@D)/cuda/include/thrust/system/detail/generic/replace.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/reverse.h" "$(@D)/cuda/include/thrust/system/detail/generic/reverse.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/reverse.inl" "$(@D)/cuda/include/thrust/system/detail/generic/reverse.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/scalar/binary_search.h" "$(@D)/cuda/include/thrust/system/detail/generic/scalar/binary_search.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/scalar/binary_search.inl" "$(@D)/cuda/include/thrust/system/detail/generic/scalar/binary_search.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/scan.h" "$(@D)/cuda/include/thrust/system/detail/generic/scan.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/scan.inl" "$(@D)/cuda/include/thrust/system/detail/generic/scan.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/scan_by_key.h" "$(@D)/cuda/include/thrust/system/detail/generic/scan_by_key.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/scan_by_key.inl" "$(@D)/cuda/include/thrust/system/detail/generic/scan_by_key.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/scatter.h" "$(@D)/cuda/include/thrust/system/detail/generic/scatter.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/scatter.inl" "$(@D)/cuda/include/thrust/system/detail/generic/scatter.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/select_system.h" "$(@D)/cuda/include/thrust/system/detail/generic/select_system.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/sequence.h" "$(@D)/cuda/include/thrust/system/detail/generic/sequence.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/sequence.inl" "$(@D)/cuda/include/thrust/system/detail/generic/sequence.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/set_operations.h" "$(@D)/cuda/include/thrust/system/detail/generic/set_operations.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/set_operations.inl" "$(@D)/cuda/include/thrust/system/detail/generic/set_operations.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/sort.h" "$(@D)/cuda/include/thrust/system/detail/generic/sort.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/sort.inl" "$(@D)/cuda/include/thrust/system/detail/generic/sort.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/swap_ranges.h" "$(@D)/cuda/include/thrust/system/detail/generic/swap_ranges.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/swap_ranges.inl" "$(@D)/cuda/include/thrust/system/detail/generic/swap_ranges.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/tabulate.h" "$(@D)/cuda/include/thrust/system/detail/generic/tabulate.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/tabulate.inl" "$(@D)/cuda/include/thrust/system/detail/generic/tabulate.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/tag.h" "$(@D)/cuda/include/thrust/system/detail/generic/tag.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/temporary_buffer.h" "$(@D)/cuda/include/thrust/system/detail/generic/temporary_buffer.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/temporary_buffer.inl" "$(@D)/cuda/include/thrust/system/detail/generic/temporary_buffer.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/transform.h" "$(@D)/cuda/include/thrust/system/detail/generic/transform.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/transform.inl" "$(@D)/cuda/include/thrust/system/detail/generic/transform.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/transform_reduce.h" "$(@D)/cuda/include/thrust/system/detail/generic/transform_reduce.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/transform_reduce.inl" "$(@D)/cuda/include/thrust/system/detail/generic/transform_reduce.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/transform_scan.h" "$(@D)/cuda/include/thrust/system/detail/generic/transform_scan.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/transform_scan.inl" "$(@D)/cuda/include/thrust/system/detail/generic/transform_scan.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/type_traits.h" "$(@D)/cuda/include/thrust/system/detail/generic/type_traits.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/uninitialized_copy.h" "$(@D)/cuda/include/thrust/system/detail/generic/uninitialized_copy.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/uninitialized_copy.inl" "$(@D)/cuda/include/thrust/system/detail/generic/uninitialized_copy.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/uninitialized_fill.h" "$(@D)/cuda/include/thrust/system/detail/generic/uninitialized_fill.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/uninitialized_fill.inl" "$(@D)/cuda/include/thrust/system/detail/generic/uninitialized_fill.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/unique.h" "$(@D)/cuda/include/thrust/system/detail/generic/unique.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/unique.inl" "$(@D)/cuda/include/thrust/system/detail/generic/unique.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/unique_by_key.h" "$(@D)/cuda/include/thrust/system/detail/generic/unique_by_key.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/generic/unique_by_key.inl" "$(@D)/cuda/include/thrust/system/detail/generic/unique_by_key.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/internal/decompose.h" "$(@D)/cuda/include/thrust/system/detail/internal/decompose.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/adjacent_difference.h" "$(@D)/cuda/include/thrust/system/detail/sequential/adjacent_difference.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/assign_value.h" "$(@D)/cuda/include/thrust/system/detail/sequential/assign_value.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/binary_search.h" "$(@D)/cuda/include/thrust/system/detail/sequential/binary_search.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/copy.h" "$(@D)/cuda/include/thrust/system/detail/sequential/copy.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/copy.inl" "$(@D)/cuda/include/thrust/system/detail/sequential/copy.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/copy_backward.h" "$(@D)/cuda/include/thrust/system/detail/sequential/copy_backward.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/copy_if.h" "$(@D)/cuda/include/thrust/system/detail/sequential/copy_if.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/count.h" "$(@D)/cuda/include/thrust/system/detail/sequential/count.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/equal.h" "$(@D)/cuda/include/thrust/system/detail/sequential/equal.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/execution_policy.h" "$(@D)/cuda/include/thrust/system/detail/sequential/execution_policy.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/extrema.h" "$(@D)/cuda/include/thrust/system/detail/sequential/extrema.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/fill.h" "$(@D)/cuda/include/thrust/system/detail/sequential/fill.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/find.h" "$(@D)/cuda/include/thrust/system/detail/sequential/find.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/for_each.h" "$(@D)/cuda/include/thrust/system/detail/sequential/for_each.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/gather.h" "$(@D)/cuda/include/thrust/system/detail/sequential/gather.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/general_copy.h" "$(@D)/cuda/include/thrust/system/detail/sequential/general_copy.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/generate.h" "$(@D)/cuda/include/thrust/system/detail/sequential/generate.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/get_value.h" "$(@D)/cuda/include/thrust/system/detail/sequential/get_value.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/inner_product.h" "$(@D)/cuda/include/thrust/system/detail/sequential/inner_product.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/insertion_sort.h" "$(@D)/cuda/include/thrust/system/detail/sequential/insertion_sort.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/iter_swap.h" "$(@D)/cuda/include/thrust/system/detail/sequential/iter_swap.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/logical.h" "$(@D)/cuda/include/thrust/system/detail/sequential/logical.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/malloc_and_free.h" "$(@D)/cuda/include/thrust/system/detail/sequential/malloc_and_free.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/merge.h" "$(@D)/cuda/include/thrust/system/detail/sequential/merge.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/merge.inl" "$(@D)/cuda/include/thrust/system/detail/sequential/merge.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/mismatch.h" "$(@D)/cuda/include/thrust/system/detail/sequential/mismatch.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/partition.h" "$(@D)/cuda/include/thrust/system/detail/sequential/partition.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/reduce.h" "$(@D)/cuda/include/thrust/system/detail/sequential/reduce.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/reduce_by_key.h" "$(@D)/cuda/include/thrust/system/detail/sequential/reduce_by_key.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/remove.h" "$(@D)/cuda/include/thrust/system/detail/sequential/remove.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/replace.h" "$(@D)/cuda/include/thrust/system/detail/sequential/replace.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/reverse.h" "$(@D)/cuda/include/thrust/system/detail/sequential/reverse.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/scan.h" "$(@D)/cuda/include/thrust/system/detail/sequential/scan.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/scan_by_key.h" "$(@D)/cuda/include/thrust/system/detail/sequential/scan_by_key.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/scatter.h" "$(@D)/cuda/include/thrust/system/detail/sequential/scatter.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/sequence.h" "$(@D)/cuda/include/thrust/system/detail/sequential/sequence.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/set_operations.h" "$(@D)/cuda/include/thrust/system/detail/sequential/set_operations.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/sort.h" "$(@D)/cuda/include/thrust/system/detail/sequential/sort.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/sort.inl" "$(@D)/cuda/include/thrust/system/detail/sequential/sort.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/stable_merge_sort.h" "$(@D)/cuda/include/thrust/system/detail/sequential/stable_merge_sort.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/stable_merge_sort.inl" "$(@D)/cuda/include/thrust/system/detail/sequential/stable_merge_sort.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/stable_primitive_sort.h" "$(@D)/cuda/include/thrust/system/detail/sequential/stable_primitive_sort.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/stable_primitive_sort.inl" "$(@D)/cuda/include/thrust/system/detail/sequential/stable_primitive_sort.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/stable_radix_sort.h" "$(@D)/cuda/include/thrust/system/detail/sequential/stable_radix_sort.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/stable_radix_sort.inl" "$(@D)/cuda/include/thrust/system/detail/sequential/stable_radix_sort.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/swap_ranges.h" "$(@D)/cuda/include/thrust/system/detail/sequential/swap_ranges.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/tabulate.h" "$(@D)/cuda/include/thrust/system/detail/sequential/tabulate.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/temporary_buffer.h" "$(@D)/cuda/include/thrust/system/detail/sequential/temporary_buffer.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/transform.h" "$(@D)/cuda/include/thrust/system/detail/sequential/transform.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/transform_reduce.h" "$(@D)/cuda/include/thrust/system/detail/sequential/transform_reduce.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/transform_scan.h" "$(@D)/cuda/include/thrust/system/detail/sequential/transform_scan.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/trivial_copy.h" "$(@D)/cuda/include/thrust/system/detail/sequential/trivial_copy.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/uninitialized_copy.h" "$(@D)/cuda/include/thrust/system/detail/sequential/uninitialized_copy.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/uninitialized_fill.h" "$(@D)/cuda/include/thrust/system/detail/sequential/uninitialized_fill.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/unique.h" "$(@D)/cuda/include/thrust/system/detail/sequential/unique.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/sequential/unique_by_key.h" "$(@D)/cuda/include/thrust/system/detail/sequential/unique_by_key.h" && cp "/usr/local/cuda-9.0/include/thrust/system/detail/system_error.inl" "$(@D)/cuda/include/thrust/system/detail/system_error.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/error_code.h" "$(@D)/cuda/include/thrust/system/error_code.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/adjacent_difference.h" "$(@D)/cuda/include/thrust/system/omp/detail/adjacent_difference.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/assign_value.h" "$(@D)/cuda/include/thrust/system/omp/detail/assign_value.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/binary_search.h" "$(@D)/cuda/include/thrust/system/omp/detail/binary_search.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/copy.h" "$(@D)/cuda/include/thrust/system/omp/detail/copy.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/copy.inl" "$(@D)/cuda/include/thrust/system/omp/detail/copy.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/copy_if.h" "$(@D)/cuda/include/thrust/system/omp/detail/copy_if.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/copy_if.inl" "$(@D)/cuda/include/thrust/system/omp/detail/copy_if.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/count.h" "$(@D)/cuda/include/thrust/system/omp/detail/count.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/default_decomposition.h" "$(@D)/cuda/include/thrust/system/omp/detail/default_decomposition.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/default_decomposition.inl" "$(@D)/cuda/include/thrust/system/omp/detail/default_decomposition.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/equal.h" "$(@D)/cuda/include/thrust/system/omp/detail/equal.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/execution_policy.h" "$(@D)/cuda/include/thrust/system/omp/detail/execution_policy.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/extrema.h" "$(@D)/cuda/include/thrust/system/omp/detail/extrema.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/fill.h" "$(@D)/cuda/include/thrust/system/omp/detail/fill.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/find.h" "$(@D)/cuda/include/thrust/system/omp/detail/find.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/for_each.h" "$(@D)/cuda/include/thrust/system/omp/detail/for_each.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/for_each.inl" "$(@D)/cuda/include/thrust/system/omp/detail/for_each.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/gather.h" "$(@D)/cuda/include/thrust/system/omp/detail/gather.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/generate.h" "$(@D)/cuda/include/thrust/system/omp/detail/generate.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/get_value.h" "$(@D)/cuda/include/thrust/system/omp/detail/get_value.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/inner_product.h" "$(@D)/cuda/include/thrust/system/omp/detail/inner_product.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/iter_swap.h" "$(@D)/cuda/include/thrust/system/omp/detail/iter_swap.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/logical.h" "$(@D)/cuda/include/thrust/system/omp/detail/logical.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/malloc_and_free.h" "$(@D)/cuda/include/thrust/system/omp/detail/malloc_and_free.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/memory.inl" "$(@D)/cuda/include/thrust/system/omp/detail/memory.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/merge.h" "$(@D)/cuda/include/thrust/system/omp/detail/merge.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/mismatch.h" "$(@D)/cuda/include/thrust/system/omp/detail/mismatch.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/par.h" "$(@D)/cuda/include/thrust/system/omp/detail/par.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/partition.h" "$(@D)/cuda/include/thrust/system/omp/detail/partition.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/partition.inl" "$(@D)/cuda/include/thrust/system/omp/detail/partition.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/reduce.h" "$(@D)/cuda/include/thrust/system/omp/detail/reduce.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/reduce.inl" "$(@D)/cuda/include/thrust/system/omp/detail/reduce.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/reduce_by_key.h" "$(@D)/cuda/include/thrust/system/omp/detail/reduce_by_key.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/reduce_by_key.inl" "$(@D)/cuda/include/thrust/system/omp/detail/reduce_by_key.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/reduce_intervals.h" "$(@D)/cuda/include/thrust/system/omp/detail/reduce_intervals.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/reduce_intervals.inl" "$(@D)/cuda/include/thrust/system/omp/detail/reduce_intervals.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/remove.h" "$(@D)/cuda/include/thrust/system/omp/detail/remove.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/remove.inl" "$(@D)/cuda/include/thrust/system/omp/detail/remove.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/replace.h" "$(@D)/cuda/include/thrust/system/omp/detail/replace.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/reverse.h" "$(@D)/cuda/include/thrust/system/omp/detail/reverse.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/scan.h" "$(@D)/cuda/include/thrust/system/omp/detail/scan.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/scan_by_key.h" "$(@D)/cuda/include/thrust/system/omp/detail/scan_by_key.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/scatter.h" "$(@D)/cuda/include/thrust/system/omp/detail/scatter.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/sequence.h" "$(@D)/cuda/include/thrust/system/omp/detail/sequence.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/set_operations.h" "$(@D)/cuda/include/thrust/system/omp/detail/set_operations.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/sort.h" "$(@D)/cuda/include/thrust/system/omp/detail/sort.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/sort.inl" "$(@D)/cuda/include/thrust/system/omp/detail/sort.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/swap_ranges.h" "$(@D)/cuda/include/thrust/system/omp/detail/swap_ranges.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/tabulate.h" "$(@D)/cuda/include/thrust/system/omp/detail/tabulate.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/temporary_buffer.h" "$(@D)/cuda/include/thrust/system/omp/detail/temporary_buffer.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/transform.h" "$(@D)/cuda/include/thrust/system/omp/detail/transform.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/transform_reduce.h" "$(@D)/cuda/include/thrust/system/omp/detail/transform_reduce.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/transform_scan.h" "$(@D)/cuda/include/thrust/system/omp/detail/transform_scan.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/uninitialized_copy.h" "$(@D)/cuda/include/thrust/system/omp/detail/uninitialized_copy.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/uninitialized_fill.h" "$(@D)/cuda/include/thrust/system/omp/detail/uninitialized_fill.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/unique.h" "$(@D)/cuda/include/thrust/system/omp/detail/unique.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/unique.inl" "$(@D)/cuda/include/thrust/system/omp/detail/unique.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/unique_by_key.h" "$(@D)/cuda/include/thrust/system/omp/detail/unique_by_key.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/unique_by_key.inl" "$(@D)/cuda/include/thrust/system/omp/detail/unique_by_key.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/detail/vector.inl" "$(@D)/cuda/include/thrust/system/omp/detail/vector.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/execution_policy.h" "$(@D)/cuda/include/thrust/system/omp/execution_policy.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/memory.h" "$(@D)/cuda/include/thrust/system/omp/memory.h" && cp "/usr/local/cuda-9.0/include/thrust/system/omp/vector.h" "$(@D)/cuda/include/thrust/system/omp/vector.h" && cp "/usr/local/cuda-9.0/include/thrust/system/system_error.h" "$(@D)/cuda/include/thrust/system/system_error.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/adjacent_difference.h" "$(@D)/cuda/include/thrust/system/tbb/detail/adjacent_difference.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/assign_value.h" "$(@D)/cuda/include/thrust/system/tbb/detail/assign_value.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/binary_search.h" "$(@D)/cuda/include/thrust/system/tbb/detail/binary_search.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/copy.h" "$(@D)/cuda/include/thrust/system/tbb/detail/copy.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/copy.inl" "$(@D)/cuda/include/thrust/system/tbb/detail/copy.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/copy_if.h" "$(@D)/cuda/include/thrust/system/tbb/detail/copy_if.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/copy_if.inl" "$(@D)/cuda/include/thrust/system/tbb/detail/copy_if.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/count.h" "$(@D)/cuda/include/thrust/system/tbb/detail/count.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/equal.h" "$(@D)/cuda/include/thrust/system/tbb/detail/equal.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/execution_policy.h" "$(@D)/cuda/include/thrust/system/tbb/detail/execution_policy.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/extrema.h" "$(@D)/cuda/include/thrust/system/tbb/detail/extrema.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/fill.h" "$(@D)/cuda/include/thrust/system/tbb/detail/fill.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/find.h" "$(@D)/cuda/include/thrust/system/tbb/detail/find.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/for_each.h" "$(@D)/cuda/include/thrust/system/tbb/detail/for_each.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/for_each.inl" "$(@D)/cuda/include/thrust/system/tbb/detail/for_each.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/gather.h" "$(@D)/cuda/include/thrust/system/tbb/detail/gather.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/generate.h" "$(@D)/cuda/include/thrust/system/tbb/detail/generate.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/get_value.h" "$(@D)/cuda/include/thrust/system/tbb/detail/get_value.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/inner_product.h" "$(@D)/cuda/include/thrust/system/tbb/detail/inner_product.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/iter_swap.h" "$(@D)/cuda/include/thrust/system/tbb/detail/iter_swap.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/logical.h" "$(@D)/cuda/include/thrust/system/tbb/detail/logical.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/malloc_and_free.h" "$(@D)/cuda/include/thrust/system/tbb/detail/malloc_and_free.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/memory.inl" "$(@D)/cuda/include/thrust/system/tbb/detail/memory.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/merge.h" "$(@D)/cuda/include/thrust/system/tbb/detail/merge.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/merge.inl" "$(@D)/cuda/include/thrust/system/tbb/detail/merge.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/mismatch.h" "$(@D)/cuda/include/thrust/system/tbb/detail/mismatch.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/par.h" "$(@D)/cuda/include/thrust/system/tbb/detail/par.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/partition.h" "$(@D)/cuda/include/thrust/system/tbb/detail/partition.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/partition.inl" "$(@D)/cuda/include/thrust/system/tbb/detail/partition.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/reduce.h" "$(@D)/cuda/include/thrust/system/tbb/detail/reduce.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/reduce.inl" "$(@D)/cuda/include/thrust/system/tbb/detail/reduce.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/reduce_by_key.h" "$(@D)/cuda/include/thrust/system/tbb/detail/reduce_by_key.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/reduce_by_key.inl" "$(@D)/cuda/include/thrust/system/tbb/detail/reduce_by_key.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/reduce_intervals.h" "$(@D)/cuda/include/thrust/system/tbb/detail/reduce_intervals.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/remove.h" "$(@D)/cuda/include/thrust/system/tbb/detail/remove.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/remove.inl" "$(@D)/cuda/include/thrust/system/tbb/detail/remove.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/replace.h" "$(@D)/cuda/include/thrust/system/tbb/detail/replace.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/reverse.h" "$(@D)/cuda/include/thrust/system/tbb/detail/reverse.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/scan.h" "$(@D)/cuda/include/thrust/system/tbb/detail/scan.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/scan.inl" "$(@D)/cuda/include/thrust/system/tbb/detail/scan.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/scan_by_key.h" "$(@D)/cuda/include/thrust/system/tbb/detail/scan_by_key.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/scatter.h" "$(@D)/cuda/include/thrust/system/tbb/detail/scatter.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/sequence.h" "$(@D)/cuda/include/thrust/system/tbb/detail/sequence.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/set_operations.h" "$(@D)/cuda/include/thrust/system/tbb/detail/set_operations.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/sort.h" "$(@D)/cuda/include/thrust/system/tbb/detail/sort.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/sort.inl" "$(@D)/cuda/include/thrust/system/tbb/detail/sort.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/swap_ranges.h" "$(@D)/cuda/include/thrust/system/tbb/detail/swap_ranges.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/tabulate.h" "$(@D)/cuda/include/thrust/system/tbb/detail/tabulate.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/temporary_buffer.h" "$(@D)/cuda/include/thrust/system/tbb/detail/temporary_buffer.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/transform.h" "$(@D)/cuda/include/thrust/system/tbb/detail/transform.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/transform_reduce.h" "$(@D)/cuda/include/thrust/system/tbb/detail/transform_reduce.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/transform_scan.h" "$(@D)/cuda/include/thrust/system/tbb/detail/transform_scan.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/uninitialized_copy.h" "$(@D)/cuda/include/thrust/system/tbb/detail/uninitialized_copy.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/uninitialized_fill.h" "$(@D)/cuda/include/thrust/system/tbb/detail/uninitialized_fill.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/unique.h" "$(@D)/cuda/include/thrust/system/tbb/detail/unique.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/unique.inl" "$(@D)/cuda/include/thrust/system/tbb/detail/unique.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/unique_by_key.h" "$(@D)/cuda/include/thrust/system/tbb/detail/unique_by_key.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/unique_by_key.inl" "$(@D)/cuda/include/thrust/system/tbb/detail/unique_by_key.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/detail/vector.inl" "$(@D)/cuda/include/thrust/system/tbb/detail/vector.inl" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/execution_policy.h" "$(@D)/cuda/include/thrust/system/tbb/execution_policy.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/memory.h" "$(@D)/cuda/include/thrust/system/tbb/memory.h" && cp "/usr/local/cuda-9.0/include/thrust/system/tbb/vector.h" "$(@D)/cuda/include/thrust/system/tbb/vector.h" && cp "/usr/local/cuda-9.0/include/thrust/system_error.h" "$(@D)/cuda/include/thrust/system_error.h" && cp "/usr/local/cuda-9.0/include/thrust/tabulate.h" "$(@D)/cuda/include/thrust/tabulate.h" && cp "/usr/local/cuda-9.0/include/thrust/transform.h" "$(@D)/cuda/include/thrust/transform.h" && cp "/usr/local/cuda-9.0/include/thrust/transform_reduce.h" "$(@D)/cuda/include/thrust/transform_reduce.h" && cp "/usr/local/cuda-9.0/include/thrust/transform_scan.h" "$(@D)/cuda/include/thrust/transform_scan.h" && cp "/usr/local/cuda-9.0/include/thrust/tuple.h" "$(@D)/cuda/include/thrust/tuple.h" && cp "/usr/local/cuda-9.0/include/thrust/uninitialized_copy.h" "$(@D)/cuda/include/thrust/uninitialized_copy.h" && cp "/usr/local/cuda-9.0/include/thrust/uninitialized_fill.h" "$(@D)/cuda/include/thrust/uninitialized_fill.h" && cp "/usr/local/cuda-9.0/include/thrust/unique.h" "$(@D)/cuda/include/thrust/unique.h" && cp "/usr/local/cuda-9.0/include/thrust/version.h" "$(@D)/cuda/include/thrust/version.h" && cp "/usr/local/cuda-9.0/include/vector_functions.h" "$(@D)/cuda/include/vector_functions.h" && cp "/usr/local/cuda-9.0/include/vector_functions.hpp" "$(@D)/cuda/include/vector_functions.hpp" && cp "/usr/local/cuda-9.0/include/vector_types.h" "$(@D)/cuda/include/vector_types.h" + """, +) + +genrule( + name = "cuda-nvvm", + outs = [ + "cuda/nvvm/libdevice/libdevice.10.bc", + ], + cmd = """ +if [ -d "$(@D)/extras" ]; then rm $(@D)/extras -drf; fi && if [ -d "$(@D)/include" ]; then rm $(@D)/include -drf; fi && if [ -d "$(@D)/lib" ]; then rm $(@D)/lib -drf; fi && if [ -d "$(@D)/nvvm" ]; then rm $(@D)/nvvm -drf; fi && cp "/usr/local/cuda-9.0/nvvm/libdevice/libdevice.10.bc" "$(@D)//libdevice.10.bc" + """, +) + +genrule( + name = "cuda-extras", + outs = [ + "cuda/extras/CUPTI/include/GL/gl.h", + "cuda/extras/CUPTI/include/GL/glew.h", + "cuda/extras/CUPTI/include/GL/glext.h", + "cuda/extras/CUPTI/include/GL/glu.h", + "cuda/extras/CUPTI/include/GL/glut.h", + "cuda/extras/CUPTI/include/GL/glx.h", + "cuda/extras/CUPTI/include/GL/glxext.h", + "cuda/extras/CUPTI/include/GL/wglew.h", + "cuda/extras/CUPTI/include/GL/wglext.h", + "cuda/extras/CUPTI/include/cuda_stdint.h", + "cuda/extras/CUPTI/include/cupti.h", + "cuda/extras/CUPTI/include/cupti_activity.h", + "cuda/extras/CUPTI/include/cupti_callbacks.h", + "cuda/extras/CUPTI/include/cupti_driver_cbid.h", + "cuda/extras/CUPTI/include/cupti_events.h", + "cuda/extras/CUPTI/include/cupti_metrics.h", + "cuda/extras/CUPTI/include/cupti_nvtx_cbid.h", + "cuda/extras/CUPTI/include/cupti_result.h", + "cuda/extras/CUPTI/include/cupti_runtime_cbid.h", + "cuda/extras/CUPTI/include/cupti_version.h", + "cuda/extras/CUPTI/include/generated_cudaGL_meta.h", + "cuda/extras/CUPTI/include/generated_cudaVDPAU_meta.h", + "cuda/extras/CUPTI/include/generated_cuda_gl_interop_meta.h", + "cuda/extras/CUPTI/include/generated_cuda_meta.h", + "cuda/extras/CUPTI/include/generated_cuda_runtime_api_meta.h", + "cuda/extras/CUPTI/include/generated_cuda_vdpau_interop_meta.h", + "cuda/extras/CUPTI/include/generated_nvtx_meta.h", + "cuda/extras/CUPTI/include/openacc/cupti_openacc.h", + ], + cmd = """ +if [ -d "$(@D)/extras" ]; then rm $(@D)/extras -drf; fi && if [ -d "$(@D)/include" ]; then rm $(@D)/include -drf; fi && if [ -d "$(@D)/lib" ]; then rm $(@D)/lib -drf; fi && if [ -d "$(@D)/nvvm" ]; then rm $(@D)/nvvm -drf; fi && cp "/usr/local/cuda-9.0/extras/CUPTI/include/GL/gl.h" "$(@D)/cuda/extras/CUPTI/include/GL/gl.h" && cp "/usr/local/cuda-9.0/extras/CUPTI/include/GL/glew.h" "$(@D)/cuda/extras/CUPTI/include/GL/glew.h" && cp "/usr/local/cuda-9.0/extras/CUPTI/include/GL/glext.h" "$(@D)/cuda/extras/CUPTI/include/GL/glext.h" && cp "/usr/local/cuda-9.0/extras/CUPTI/include/GL/glu.h" "$(@D)/cuda/extras/CUPTI/include/GL/glu.h" && cp "/usr/local/cuda-9.0/extras/CUPTI/include/GL/glut.h" "$(@D)/cuda/extras/CUPTI/include/GL/glut.h" && cp "/usr/local/cuda-9.0/extras/CUPTI/include/GL/glx.h" "$(@D)/cuda/extras/CUPTI/include/GL/glx.h" && cp "/usr/local/cuda-9.0/extras/CUPTI/include/GL/glxext.h" "$(@D)/cuda/extras/CUPTI/include/GL/glxext.h" && cp "/usr/local/cuda-9.0/extras/CUPTI/include/GL/wglew.h" "$(@D)/cuda/extras/CUPTI/include/GL/wglew.h" && cp "/usr/local/cuda-9.0/extras/CUPTI/include/GL/wglext.h" "$(@D)/cuda/extras/CUPTI/include/GL/wglext.h" && cp "/usr/local/cuda-9.0/extras/CUPTI/include/cuda_stdint.h" "$(@D)/cuda/extras/CUPTI/include/cuda_stdint.h" && cp "/usr/local/cuda-9.0/extras/CUPTI/include/cupti.h" "$(@D)/cuda/extras/CUPTI/include/cupti.h" && cp "/usr/local/cuda-9.0/extras/CUPTI/include/cupti_activity.h" "$(@D)/cuda/extras/CUPTI/include/cupti_activity.h" && cp "/usr/local/cuda-9.0/extras/CUPTI/include/cupti_callbacks.h" "$(@D)/cuda/extras/CUPTI/include/cupti_callbacks.h" && cp "/usr/local/cuda-9.0/extras/CUPTI/include/cupti_driver_cbid.h" "$(@D)/cuda/extras/CUPTI/include/cupti_driver_cbid.h" && cp "/usr/local/cuda-9.0/extras/CUPTI/include/cupti_events.h" "$(@D)/cuda/extras/CUPTI/include/cupti_events.h" && cp "/usr/local/cuda-9.0/extras/CUPTI/include/cupti_metrics.h" "$(@D)/cuda/extras/CUPTI/include/cupti_metrics.h" && cp "/usr/local/cuda-9.0/extras/CUPTI/include/cupti_nvtx_cbid.h" "$(@D)/cuda/extras/CUPTI/include/cupti_nvtx_cbid.h" && cp "/usr/local/cuda-9.0/extras/CUPTI/include/cupti_result.h" "$(@D)/cuda/extras/CUPTI/include/cupti_result.h" && cp "/usr/local/cuda-9.0/extras/CUPTI/include/cupti_runtime_cbid.h" "$(@D)/cuda/extras/CUPTI/include/cupti_runtime_cbid.h" && cp "/usr/local/cuda-9.0/extras/CUPTI/include/cupti_version.h" "$(@D)/cuda/extras/CUPTI/include/cupti_version.h" && cp "/usr/local/cuda-9.0/extras/CUPTI/include/generated_cudaGL_meta.h" "$(@D)/cuda/extras/CUPTI/include/generated_cudaGL_meta.h" && cp "/usr/local/cuda-9.0/extras/CUPTI/include/generated_cudaVDPAU_meta.h" "$(@D)/cuda/extras/CUPTI/include/generated_cudaVDPAU_meta.h" && cp "/usr/local/cuda-9.0/extras/CUPTI/include/generated_cuda_gl_interop_meta.h" "$(@D)/cuda/extras/CUPTI/include/generated_cuda_gl_interop_meta.h" && cp "/usr/local/cuda-9.0/extras/CUPTI/include/generated_cuda_meta.h" "$(@D)/cuda/extras/CUPTI/include/generated_cuda_meta.h" && cp "/usr/local/cuda-9.0/extras/CUPTI/include/generated_cuda_runtime_api_meta.h" "$(@D)/cuda/extras/CUPTI/include/generated_cuda_runtime_api_meta.h" && cp "/usr/local/cuda-9.0/extras/CUPTI/include/generated_cuda_vdpau_interop_meta.h" "$(@D)/cuda/extras/CUPTI/include/generated_cuda_vdpau_interop_meta.h" && cp "/usr/local/cuda-9.0/extras/CUPTI/include/generated_nvtx_meta.h" "$(@D)/cuda/extras/CUPTI/include/generated_nvtx_meta.h" && cp "/usr/local/cuda-9.0/extras/CUPTI/include/openacc/cupti_openacc.h" "$(@D)/cuda/extras/CUPTI/include/openacc/cupti_openacc.h" + """, +) + +genrule( + name = "cuda-lib", + outs = [ + "cuda/lib/libcuda.so", + "cuda/lib/libcudart.so.9.0", + "cuda/lib/libcudart_static.a", + "cuda/lib/libcublas.so.9.0", + "cuda/lib/libcusolver.so.9.0", + "cuda/lib/libcurand.so.9.0", + "cuda/lib/libcufft.so.9.0", + "cuda/lib/libcudnn.so.7", + "cuda/lib/libcupti.so.9.0", + ], + cmd = """ +if [ -d "$(@D)/extras" ]; then rm $(@D)/extras -drf; fi && if [ -d "$(@D)/include" ]; then rm $(@D)/include -drf; fi && if [ -d "$(@D)/lib" ]; then rm $(@D)/lib -drf; fi && if [ -d "$(@D)/nvvm" ]; then rm $(@D)/nvvm -drf; fi && cp "/usr/local/cuda-9.0/targets/x86_64-linux/lib/stubs/libcuda.so" "$(@D)/cuda/lib/libcuda.so" && cp "/usr/local/cuda-9.0/targets/x86_64-linux/lib/libcudart.so.9.0.176" "$(@D)/cuda/lib/libcudart.so.9.0" && cp "/usr/local/cuda-9.0/targets/x86_64-linux/lib/libcudart_static.a" "$(@D)/cuda/lib/libcudart_static.a" && cp "/usr/local/cuda-9.0/targets/x86_64-linux/lib/libcublas.so.9.0.480" "$(@D)/cuda/lib/libcublas.so.9.0" && cp "/usr/local/cuda-9.0/targets/x86_64-linux/lib/libcusolver.so.9.0.176" "$(@D)/cuda/lib/libcusolver.so.9.0" && cp "/usr/local/cuda-9.0/targets/x86_64-linux/lib/libcurand.so.9.0.176" "$(@D)/cuda/lib/libcurand.so.9.0" && cp "/usr/local/cuda-9.0/targets/x86_64-linux/lib/libcufft.so.9.0.176" "$(@D)/cuda/lib/libcufft.so.9.0" && cp "/usr/lib/x86_64-linux-gnu/libcudnn.so.7.2.1" "$(@D)/cuda/lib/libcudnn.so.7" && cp "/usr/local/cuda-9.0/extras/CUPTI/lib64/libcupti.so.9.0.176" "$(@D)/cuda/lib/libcupti.so.9.0" + """, +) + +genrule( + name = "cudnn-include", + outs = [ + "cuda/include/cudnn.h", + ], + cmd = """ +if [ -d "$(@D)/extras" ]; then rm $(@D)/extras -drf; fi && if [ -d "$(@D)/include" ]; then rm $(@D)/include -drf; fi && if [ -d "$(@D)/lib" ]; then rm $(@D)/lib -drf; fi && if [ -d "$(@D)/nvvm" ]; then rm $(@D)/nvvm -drf; fi && cp "/usr/include/cudnn.h" "$(@D)/cudnn.h" + """, +) diff --git a/third_party/toolchains/preconfig/ubuntu14.04/cuda9.0-cudnn7/cuda/build_defs.bzl b/third_party/toolchains/preconfig/ubuntu14.04/cuda9.0-cudnn7/cuda/build_defs.bzl new file mode 100755 index 0000000000..5c6703aab4 --- /dev/null +++ b/third_party/toolchains/preconfig/ubuntu14.04/cuda9.0-cudnn7/cuda/build_defs.bzl @@ -0,0 +1,33 @@ +# Macros for building CUDA code. +def if_cuda(if_true, if_false = []): + """Shorthand for select()'ing on whether we're building with CUDA. + + Returns a select statement which evaluates to if_true if we're building + with CUDA enabled. Otherwise, the select statement evaluates to if_false. + + """ + return select({ + "@local_config_cuda//cuda:using_nvcc": if_true, + "@local_config_cuda//cuda:using_clang": if_true, + "//conditions:default": if_false + }) + + +def cuda_default_copts(): + """Default options for all CUDA compilations.""" + return if_cuda(["-x", "cuda", "-DGOOGLE_CUDA=1"] + []) + + +def cuda_is_configured(): + """Returns true if CUDA was enabled during the configure process.""" + return True + +def if_cuda_is_configured(x): + """Tests if the CUDA was enabled during the configure process. + + Unlike if_cuda(), this does not require that we are building with + --config=cuda. Used to allow non-CUDA code to depend on CUDA libraries. + """ + if cuda_is_configured(): + return x + return [] diff --git a/third_party/toolchains/preconfig/ubuntu14.04/cuda9.0-cudnn7/cuda/cuda/cuda_config.h b/third_party/toolchains/preconfig/ubuntu14.04/cuda9.0-cudnn7/cuda/cuda/cuda_config.h new file mode 100755 index 0000000000..5d0d3013a9 --- /dev/null +++ b/third_party/toolchains/preconfig/ubuntu14.04/cuda9.0-cudnn7/cuda/cuda/cuda_config.h @@ -0,0 +1,26 @@ +/* Copyright 2015 The TensorFlow Authors. All Rights Reserved. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +==============================================================================*/ + +#ifndef CUDA_CUDA_CONFIG_H_ +#define CUDA_CUDA_CONFIG_H_ + +#define TF_CUDA_CAPABILITIES CudaVersion("3.0") + +#define TF_CUDA_VERSION "9.0" +#define TF_CUDNN_VERSION "7" + +#define TF_CUDA_TOOLKIT_PATH "/usr/local/cuda-9.0" + +#endif // CUDA_CUDA_CONFIG_H_ diff --git a/third_party/toolchains/preconfig/ubuntu14.04/gcc-nvcc/BUILD b/third_party/toolchains/preconfig/ubuntu14.04/gcc-nvcc/BUILD new file mode 100755 index 0000000000..a56b4513fb --- /dev/null +++ b/third_party/toolchains/preconfig/ubuntu14.04/gcc-nvcc/BUILD @@ -0,0 +1,73 @@ +licenses(["restricted"]) + +package(default_visibility = ["//visibility:public"]) + +cc_toolchain_suite( + name = "toolchain", + toolchains = { + "local|compiler": ":cc-compiler-local", + "darwin|compiler": ":cc-compiler-darwin", + "x64_windows|msvc-cl": ":cc-compiler-windows", + }, +) + +cc_toolchain( + name = "cc-compiler-local", + all_files = ":crosstool_wrapper_driver_is_not_gcc", + compiler_files = ":empty", + cpu = "local", + dwp_files = ":empty", + dynamic_runtime_libs = [":empty"], + linker_files = ":crosstool_wrapper_driver_is_not_gcc", + objcopy_files = ":empty", + static_runtime_libs = [":empty"], + strip_files = ":empty", + # To support linker flags that need to go to the start of command line + # we need the toolchain to support parameter files. Parameter files are + # last on the command line and contain all shared libraries to link, so all + # regular options will be left of them. + supports_param_files = 1, +) + +cc_toolchain( + name = "cc-compiler-darwin", + all_files = ":crosstool_wrapper_driver_is_not_gcc", + compiler_files = ":empty", + cpu = "darwin", + dwp_files = ":empty", + dynamic_runtime_libs = [":empty"], + linker_files = ":crosstool_wrapper_driver_is_not_gcc", + objcopy_files = ":empty", + static_runtime_libs = [":empty"], + strip_files = ":empty", + supports_param_files = 0, +) + +cc_toolchain( + name = "cc-compiler-windows", + all_files = ":windows_msvc_wrapper_files", + compiler_files = ":empty", + cpu = "x64_windows", + dwp_files = ":empty", + dynamic_runtime_libs = [":empty"], + linker_files = ":windows_msvc_wrapper_files", + objcopy_files = ":empty", + static_runtime_libs = [":empty"], + strip_files = ":empty", + supports_param_files = 1, +) + +filegroup( + name = "empty", + srcs = [], +) + +filegroup( + name = "crosstool_wrapper_driver_is_not_gcc", + srcs = ["clang/bin/crosstool_wrapper_driver_is_not_gcc"], +) + +filegroup( + name = "windows_msvc_wrapper_files", + srcs = glob(["windows/msvc_*"]), +) diff --git a/third_party/toolchains/preconfig/ubuntu14.04/gcc-nvcc/CROSSTOOL b/third_party/toolchains/preconfig/ubuntu14.04/gcc-nvcc/CROSSTOOL new file mode 100755 index 0000000000..a14eceacbb --- /dev/null +++ b/third_party/toolchains/preconfig/ubuntu14.04/gcc-nvcc/CROSSTOOL @@ -0,0 +1,1410 @@ +major_version: "local" +minor_version: "" +default_target_cpu: "same_as_host" + +default_toolchain { + cpu: "k8" + toolchain_identifier: "local_linux" +} +default_toolchain { + cpu: "piii" + toolchain_identifier: "local_linux" +} +default_toolchain { + cpu: "arm" + toolchain_identifier: "local_linux" +} +default_toolchain { + cpu: "darwin" + toolchain_identifier: "local_darwin" +} +default_toolchain { + cpu: "ppc" + toolchain_identifier: "local_linux" +} +default_toolchain { + cpu: "x64_windows" + toolchain_identifier: "local_windows" +} + +toolchain { + abi_version: "local" + abi_libc_version: "local" + compiler: "compiler" + host_system_name: "local" + needsPic: true + target_libc: "local" + target_cpu: "local" + target_system_name: "local" + toolchain_identifier: "local_linux" + + feature { + name: "c++11" + flag_set { + action: "c++-compile" + flag_group { + flag: "-std=c++11" + } + } + } + + feature { + name: "stdlib" + flag_set { + action: "c++-link-executable" + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "-lstdc++" + } + } + } + + feature { + name: "determinism" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + # Make C++ compilation deterministic. Use linkstamping instead of these + # compiler symbols. + flag: "-Wno-builtin-macro-redefined" + flag: "-D__DATE__=\"redacted\"" + flag: "-D__TIMESTAMP__=\"redacted\"" + flag: "-D__TIME__=\"redacted\"" + } + } + } + + feature { + name: "alwayslink" + flag_set { + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + action: "c++-link-executable" + flag_group { + flag: "-Wl,-no-as-needed" + } + } + } + + # This feature will be enabled for builds that support pic by bazel. + feature { + name: "pic" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + expand_if_all_available: "pic" + flag: "-fPIC" + } + flag_group { + expand_if_none_available: "pic" + flag: "-fPIE" + } + } + } + + # Security hardening on by default. + feature { + name: "hardening" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + # Conservative choice; -D_FORTIFY_SOURCE=2 may be unsafe in some cases. + # We need to undef it before redefining it as some distributions now + # have it enabled by default. + flag: "-U_FORTIFY_SOURCE" + flag: "-D_FORTIFY_SOURCE=1" + flag: "-fstack-protector" + } + } + flag_set { + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "-Wl,-z,relro,-z,now" + } + } + flag_set { + action: "c++-link-executable" + flag_group { + flag: "-pie" + flag: "-Wl,-z,relro,-z,now" + } + } + } + + feature { + name: "warnings" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + # All warnings are enabled. Maybe enable -Werror as well? + flag: "-Wall" + + } + } + } + + # Keep stack frames for debugging, even in opt mode. + feature { + name: "frame-pointer" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + flag: "-fno-omit-frame-pointer" + } + } + } + + feature { + name: "build-id" + flag_set { + action: "c++-link-executable" + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + flag_group { + # Stamp the binary with a unique identifier. + flag: "-Wl,--build-id=md5" + flag: "-Wl,--hash-style=gnu" + } + } + } + + feature { + name: "no-canonical-prefixes" + flag_set { + action: "c-compile" + action: "c++-compile" + action: "c++-link-executable" + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag:"-no-canonical-prefixes" + } + } + } + + feature { + name: "disable-assertions" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + flag: "-DNDEBUG" + } + } + } + + feature { + name: "linker-bin-path" + + flag_set { + action: "c++-link-executable" + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "-B/usr/bin" + } + } + } + + feature { + name: "common" + implies: "stdlib" + implies: "c++11" + implies: "determinism" + implies: "alwayslink" + implies: "hardening" + implies: "warnings" + implies: "frame-pointer" + implies: "build-id" + implies: "no-canonical-prefixes" + implies: "linker-bin-path" + } + + feature { + name: "opt" + implies: "common" + implies: "disable-assertions" + + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + # No debug symbols. + # Maybe we should enable https://gcc.gnu.org/wiki/DebugFission for opt + # or even generally? However, that can't happen here, as it requires + # special handling in Bazel. + flag: "-g0" + + # Conservative choice for -O + # -O3 can increase binary size and even slow down the resulting binaries. + # Profile first and / or use FDO if you need better performance than this. + flag: "-O2" + + # Removal of unused code and data at link time (can this increase binary size in some cases?). + flag: "-ffunction-sections" + flag: "-fdata-sections" + } + } + flag_set { + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + action: "c++-link-executable" + flag_group { + flag: "-Wl,--gc-sections" + } + } + } + + feature { + name: "fastbuild" + implies: "common" + } + + feature { + name: "dbg" + implies: "common" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + flag: "-g" + } + } + } + + # Set clang as a C/C++ compiler. + tool_path { name: "gcc" path: "clang/bin/crosstool_wrapper_driver_is_not_gcc" } + + # Use the default system toolchain for everything else. + tool_path { name: "ar" path: "/usr/bin/ar" } + tool_path { name: "compat-ld" path: "/usr/bin/ld" } + tool_path { name: "cpp" path: "/usr/bin/cpp" } + tool_path { name: "dwp" path: "/usr/bin/dwp" } + tool_path { name: "gcov" path: "/usr/bin/gcov" } + tool_path { name: "ld" path: "/usr/bin/ld" } + tool_path { name: "nm" path: "/usr/bin/nm" } + tool_path { name: "objcopy" path: "/usr/bin/objcopy" } + tool_path { name: "objdump" path: "/usr/bin/objdump" } + tool_path { name: "strip" path: "/usr/bin/strip" } + + # Enabled dynamic linking. + linking_mode_flags { mode: DYNAMIC } + + + cxx_builtin_include_directory: "/" +} + +toolchain { + abi_version: "local" + abi_libc_version: "local" + compiler: "compiler" + host_system_name: "local" + needsPic: true + target_libc: "macosx" + target_cpu: "darwin" + target_system_name: "local" + toolchain_identifier: "local_darwin" + feature { + name: "c++11" + flag_set { + action: "c++-compile" + flag_group { + flag: "-std=c++11" + } + } + } + + feature { + name: "stdlib" + flag_set { + action: "c++-link-executable" + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "-lc++" + } + } + } + + feature { + name: "determinism" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + # Make C++ compilation deterministic. Use linkstamping instead of these + # compiler symbols. + flag: "-Wno-builtin-macro-redefined" + flag: "-D__DATE__=\"redacted\"" + flag: "-D__TIMESTAMP__=\"redacted\"" + flag: "-D__TIME__=\"redacted\"" + } + } + } + + # This feature will be enabled for builds that support pic by bazel. + feature { + name: "pic" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + expand_if_all_available: "pic" + flag: "-fPIC" + } + flag_group { + expand_if_none_available: "pic" + flag: "-fPIE" + } + } + } + + # Security hardening on by default. + feature { + name: "hardening" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + # Conservative choice; -D_FORTIFY_SOURCE=2 may be unsafe in some cases. + # We need to undef it before redefining it as some distributions now + # have it enabled by default. + flag: "-U_FORTIFY_SOURCE" + flag: "-D_FORTIFY_SOURCE=1" + flag: "-fstack-protector" + } + } + flag_set { + action: "c++-link-executable" + flag_group { + flag: "-pie" + } + } + } + + feature { + name: "warnings" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + # All warnings are enabled. Maybe enable -Werror as well? + flag: "-Wall" + + } + } + } + + # Keep stack frames for debugging, even in opt mode. + feature { + name: "frame-pointer" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + flag: "-fno-omit-frame-pointer" + } + } + } + + feature { + name: "no-canonical-prefixes" + flag_set { + action: "c-compile" + action: "c++-compile" + action: "c++-link-executable" + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag:"-no-canonical-prefixes" + } + } + } + + feature { + name: "disable-assertions" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + flag: "-DNDEBUG" + } + } + } + + feature { + name: "linker-bin-path" + + flag_set { + action: "c++-link-executable" + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "-B/usr/bin" + } + } + } + + feature { + name: "undefined-dynamic" + flag_set { + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + action: "c++-link-executable" + flag_group { + flag: "-undefined" + flag: "dynamic_lookup" + } + } + } + + feature { + name: "common" + implies: "stdlib" + implies: "c++11" + implies: "determinism" + implies: "hardening" + implies: "warnings" + implies: "frame-pointer" + implies: "no-canonical-prefixes" + implies: "linker-bin-path" + implies: "undefined-dynamic" + } + + feature { + name: "opt" + implies: "common" + implies: "disable-assertions" + + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + # No debug symbols. + # Maybe we should enable https://gcc.gnu.org/wiki/DebugFission for opt + # or even generally? However, that can't happen here, as it requires + # special handling in Bazel. + flag: "-g0" + + # Conservative choice for -O + # -O3 can increase binary size and even slow down the resulting binaries. + # Profile first and / or use FDO if you need better performance than this. + flag: "-O2" + + # Removal of unused code and data at link time (can this increase binary size in some cases?). + flag: "-ffunction-sections" + flag: "-fdata-sections" + } + } + } + + feature { + name: "fastbuild" + implies: "common" + } + + feature { + name: "dbg" + implies: "common" + flag_set { + action: "c-compile" + action: "c++-compile" + flag_group { + flag: "-g" + } + } + } + + # Set clang as a C/C++ compiler. + tool_path { name: "gcc" path: "clang/bin/crosstool_wrapper_driver_is_not_gcc" } + + # Use the default system toolchain for everything else. + tool_path { name: "ar" path: "/usr/bin/libtool" } + tool_path { name: "compat-ld" path: "/usr/bin/ld" } + tool_path { name: "cpp" path: "/usr/bin/cpp" } + tool_path { name: "dwp" path: "/usr/bin/dwp" } + tool_path { name: "gcov" path: "/usr/bin/gcov" } + tool_path { name: "ld" path: "/usr/bin/ld" } + tool_path { name: "nm" path: "/usr/bin/nm" } + tool_path { name: "objcopy" path: "/usr/bin/objcopy" } + tool_path { name: "objdump" path: "/usr/bin/objdump" } + tool_path { name: "strip" path: "/usr/bin/strip" } + + # Enabled dynamic linking. + linking_mode_flags { mode: DYNAMIC } + + + cxx_builtin_include_directory: "/" +} + +toolchain { + toolchain_identifier: "local_windows" + host_system_name: "local" + target_system_name: "local" + + abi_version: "local" + abi_libc_version: "local" + target_cpu: "x64_windows" + compiler: "msvc-cl" + target_libc: "msvcrt" + + + + tool_path { + name: "ar" + path: "" + } + tool_path { + name: "ml" + path: "" + } + tool_path { + name: "cpp" + path: "" + } + tool_path { + name: "gcc" + path: "" + } + tool_path { + name: "gcov" + path: "wrapper/bin/msvc_nop.bat" + } + tool_path { + name: "ld" + path: "" + } + tool_path { + name: "nm" + path: "wrapper/bin/msvc_nop.bat" + } + tool_path { + name: "objcopy" + path: "wrapper/bin/msvc_nop.bat" + } + tool_path { + name: "objdump" + path: "wrapper/bin/msvc_nop.bat" + } + tool_path { + name: "strip" + path: "wrapper/bin/msvc_nop.bat" + } + supports_interface_shared_objects: true + + # TODO(pcloudy): Review those flags below, they should be defined by cl.exe + compiler_flag: "/DCOMPILER_MSVC" + + # Don't define min/max macros in windows.h. + compiler_flag: "/DNOMINMAX" + + # Platform defines. + compiler_flag: "/D_WIN32_WINNT=0x0600" + # Turn off warning messages. + compiler_flag: "/D_CRT_SECURE_NO_DEPRECATE" + compiler_flag: "/D_CRT_SECURE_NO_WARNINGS" + compiler_flag: "/D_SILENCE_STDEXT_HASH_DEPRECATION_WARNINGS" + + # Useful options to have on for compilation. + # Increase the capacity of object files to 2^32 sections. + compiler_flag: "/bigobj" + # Allocate 500MB for precomputed headers. + compiler_flag: "/Zm500" + # Use unsigned char by default. + compiler_flag: "/J" + # Use function level linking. + compiler_flag: "/Gy" + # Use string pooling. + compiler_flag: "/GF" + # Catch C++ exceptions only and tell the compiler to assume that functions declared + # as extern "C" never throw a C++ exception. + compiler_flag: "/EHsc" + + # Globally disabled warnings. + # Don't warn about elements of array being be default initialized. + compiler_flag: "/wd4351" + # Don't warn about no matching delete found. + compiler_flag: "/wd4291" + # Don't warn about diamond inheritance patterns. + compiler_flag: "/wd4250" + # Don't warn about insecure functions (e.g. non _s functions). + compiler_flag: "/wd4996" + + linker_flag: "/MACHINE:X64" + + feature { + name: "no_legacy_features" + } + + # Suppress startup banner. + feature { + name: "nologo" + flag_set { + action: "c-compile" + action: "c++-compile" + action: "c++-module-compile" + action: "c++-module-codegen" + action: "c++-header-parsing" + action: "assemble" + action: "preprocess-assemble" + action: "c++-link-executable" + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + action: "c++-link-static-library" + flag_group { + flag: "/nologo" + } + } + } + + feature { + name: 'has_configured_linker_path' + } + + # This feature indicates strip is not supported, building stripped binary will just result a copy of orignial binary + feature { + name: 'no_stripping' + } + + # This feature indicates this is a toolchain targeting Windows. + feature { + name: 'targets_windows' + implies: 'copy_dynamic_libraries_to_binary' + enabled: true + } + + feature { + name: 'copy_dynamic_libraries_to_binary' + } + + action_config { + config_name: 'assemble' + action_name: 'assemble' + tool { + tool_path: '' + } + implies: 'compiler_input_flags' + implies: 'compiler_output_flags' + implies: 'nologo' + implies: 'msvc_env' + implies: 'sysroot' + } + + action_config { + config_name: 'preprocess-assemble' + action_name: 'preprocess-assemble' + tool { + tool_path: '' + } + implies: 'compiler_input_flags' + implies: 'compiler_output_flags' + implies: 'nologo' + implies: 'msvc_env' + implies: 'sysroot' + } + + action_config { + config_name: 'c-compile' + action_name: 'c-compile' + tool { + tool_path: '' + } + implies: 'compiler_input_flags' + implies: 'compiler_output_flags' + implies: 'legacy_compile_flags' + implies: 'nologo' + implies: 'msvc_env' + implies: 'parse_showincludes' + implies: 'user_compile_flags' + implies: 'sysroot' + implies: 'unfiltered_compile_flags' + } + + action_config { + config_name: 'c++-compile' + action_name: 'c++-compile' + tool { + tool_path: '' + } + implies: 'compiler_input_flags' + implies: 'compiler_output_flags' + implies: 'legacy_compile_flags' + implies: 'nologo' + implies: 'msvc_env' + implies: 'parse_showincludes' + implies: 'user_compile_flags' + implies: 'sysroot' + implies: 'unfiltered_compile_flags' + } + + action_config { + config_name: 'c++-link-executable' + action_name: 'c++-link-executable' + tool { + tool_path: '' + } + implies: 'nologo' + implies: 'linkstamps' + implies: 'output_execpath_flags' + implies: 'input_param_flags' + implies: 'user_link_flags' + implies: 'legacy_link_flags' + implies: 'linker_subsystem_flag' + implies: 'linker_param_file' + implies: 'msvc_env' + implies: 'no_stripping' + } + + action_config { + config_name: 'c++-link-dynamic-library' + action_name: 'c++-link-dynamic-library' + tool { + tool_path: '' + } + implies: 'nologo' + implies: 'shared_flag' + implies: 'linkstamps' + implies: 'output_execpath_flags' + implies: 'input_param_flags' + implies: 'user_link_flags' + implies: 'legacy_link_flags' + implies: 'linker_subsystem_flag' + implies: 'linker_param_file' + implies: 'msvc_env' + implies: 'no_stripping' + implies: 'has_configured_linker_path' + implies: 'def_file' + } + + action_config { + config_name: 'c++-link-nodeps-dynamic-library' + action_name: 'c++-link-nodeps-dynamic-library' + tool { + tool_path: '' + } + implies: 'nologo' + implies: 'shared_flag' + implies: 'linkstamps' + implies: 'output_execpath_flags' + implies: 'input_param_flags' + implies: 'user_link_flags' + implies: 'legacy_link_flags' + implies: 'linker_subsystem_flag' + implies: 'linker_param_file' + implies: 'msvc_env' + implies: 'no_stripping' + implies: 'has_configured_linker_path' + implies: 'def_file' + } + + action_config { + config_name: 'c++-link-static-library' + action_name: 'c++-link-static-library' + tool { + tool_path: '' + } + implies: 'nologo' + implies: 'archiver_flags' + implies: 'input_param_flags' + implies: 'linker_param_file' + implies: 'msvc_env' + } + + # TODO(b/65151735): Remove legacy_compile_flags feature when legacy fields are + # not used in this crosstool + feature { + name: 'legacy_compile_flags' + flag_set { + expand_if_all_available: 'legacy_compile_flags' + action: 'preprocess-assemble' + action: 'c-compile' + action: 'c++-compile' + action: 'c++-header-parsing' + action: 'c++-module-compile' + action: 'c++-module-codegen' + flag_group { + iterate_over: 'legacy_compile_flags' + flag: '%{legacy_compile_flags}' + } + } + } + + feature { + name: "msvc_env" + env_set { + action: "c-compile" + action: "c++-compile" + action: "c++-module-compile" + action: "c++-module-codegen" + action: "c++-header-parsing" + action: "assemble" + action: "preprocess-assemble" + action: "c++-link-executable" + action: "c++-link-dynamic-library" + action: "c++-link-nodeps-dynamic-library" + action: "c++-link-static-library" + env_entry { + key: "PATH" + value: "" + } + env_entry { + key: "INCLUDE" + value: "" + } + env_entry { + key: "LIB" + value: "" + } + env_entry { + key: "TMP" + value: "" + } + env_entry { + key: "TEMP" + value: "" + } + } + } + + feature { + name: 'include_paths' + flag_set { + action: "assemble" + action: 'preprocess-assemble' + action: 'c-compile' + action: 'c++-compile' + action: 'c++-header-parsing' + action: 'c++-module-compile' + flag_group { + iterate_over: 'quote_include_paths' + flag: '/I%{quote_include_paths}' + } + flag_group { + iterate_over: 'include_paths' + flag: '/I%{include_paths}' + } + flag_group { + iterate_over: 'system_include_paths' + flag: '/I%{system_include_paths}' + } + } + } + + feature { + name: "preprocessor_defines" + flag_set { + action: "assemble" + action: "preprocess-assemble" + action: "c-compile" + action: "c++-compile" + action: "c++-header-parsing" + action: "c++-module-compile" + flag_group { + flag: "/D%{preprocessor_defines}" + iterate_over: "preprocessor_defines" + } + } + } + + # Tell Bazel to parse the output of /showIncludes + feature { + name: 'parse_showincludes' + flag_set { + action: 'preprocess-assemble' + action: 'c-compile' + action: 'c++-compile' + action: 'c++-module-compile' + action: 'c++-header-parsing' + flag_group { + flag: "/showIncludes" + } + } + } + + + feature { + name: 'generate_pdb_file' + requires: { + feature: 'dbg' + } + requires: { + feature: 'fastbuild' + } + } + + feature { + name: 'shared_flag' + flag_set { + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: '/DLL' + } + } + } + + feature { + name: 'linkstamps' + flag_set { + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + expand_if_all_available: 'linkstamp_paths' + flag_group { + iterate_over: 'linkstamp_paths' + flag: '%{linkstamp_paths}' + } + } + } + + feature { + name: 'output_execpath_flags' + flag_set { + expand_if_all_available: 'output_execpath' + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: '/OUT:%{output_execpath}' + } + } + } + + feature { + name: 'archiver_flags' + flag_set { + expand_if_all_available: 'output_execpath' + action: 'c++-link-static-library' + flag_group { + flag: '/OUT:%{output_execpath}' + } + } + } + + feature { + name: 'input_param_flags' + flag_set { + expand_if_all_available: 'interface_library_output_path' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "/IMPLIB:%{interface_library_output_path}" + } + } + flag_set { + expand_if_all_available: 'libopts' + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + iterate_over: 'libopts' + flag: '%{libopts}' + } + } + flag_set { + expand_if_all_available: 'libraries_to_link' + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + action: 'c++-link-static-library' + flag_group { + iterate_over: 'libraries_to_link' + flag_group { + expand_if_equal: { + variable: 'libraries_to_link.type' + value: 'object_file_group' + } + iterate_over: 'libraries_to_link.object_files' + flag_group { + flag: '%{libraries_to_link.object_files}' + } + } + flag_group { + expand_if_equal: { + variable: 'libraries_to_link.type' + value: 'object_file' + } + flag_group { + flag: '%{libraries_to_link.name}' + } + } + flag_group { + expand_if_equal: { + variable: 'libraries_to_link.type' + value: 'interface_library' + } + flag_group { + flag: '%{libraries_to_link.name}' + } + } + flag_group { + expand_if_equal: { + variable: 'libraries_to_link.type' + value: 'static_library' + } + flag_group { + expand_if_false: 'libraries_to_link.is_whole_archive' + flag: '%{libraries_to_link.name}' + } + flag_group { + expand_if_true: 'libraries_to_link.is_whole_archive' + flag: '/WHOLEARCHIVE:%{libraries_to_link.name}' + } + } + } + } + } + + # Since this feature is declared earlier in the CROSSTOOL than + # "user_link_flags", this feature will be applied prior to it anwyhere they + # are both implied. And since "user_link_flags" contains the linkopts from + # the build rule, this allows the user to override the /SUBSYSTEM in the BUILD + # file. + feature { + name: 'linker_subsystem_flag' + flag_set { + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: '/SUBSYSTEM:CONSOLE' + } + } + } + + # The "user_link_flags" contains user-defined linkopts (from build rules) + # so it should be defined after features that declare user-overridable flags. + # For example the "linker_subsystem_flag" defines a default "/SUBSYSTEM" flag + # but we want to let the user override it, therefore "link_flag_subsystem" is + # defined earlier in the CROSSTOOL file than "user_link_flags". + feature { + name: 'user_link_flags' + flag_set { + expand_if_all_available: 'user_link_flags' + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + iterate_over: 'user_link_flags' + flag: '%{user_link_flags}' + } + } + } + feature { + name: 'legacy_link_flags' + flag_set { + expand_if_all_available: 'legacy_link_flags' + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + iterate_over: 'legacy_link_flags' + flag: '%{legacy_link_flags}' + } + } + } + + feature { + name: 'linker_param_file' + flag_set { + expand_if_all_available: 'linker_param_file' + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + action: 'c++-link-static-library' + flag_group { + flag: '@%{linker_param_file}' + } + } + } + + feature { + name: 'static_link_msvcrt' + } + + feature { + name: 'static_link_msvcrt_no_debug' + flag_set { + action: 'c-compile' + action: 'c++-compile' + flag_group { + flag: "/MT" + } + } + flag_set { + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "/DEFAULTLIB:libcmt.lib" + } + } + requires: { feature: 'fastbuild'} + requires: { feature: 'opt'} + } + + feature { + name: 'dynamic_link_msvcrt_no_debug' + flag_set { + action: 'c-compile' + action: 'c++-compile' + flag_group { + flag: "/MD" + } + } + flag_set { + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "/DEFAULTLIB:msvcrt.lib" + } + } + requires: { feature: 'fastbuild'} + requires: { feature: 'opt'} + } + + feature { + name: 'static_link_msvcrt_debug' + flag_set { + action: 'c-compile' + action: 'c++-compile' + flag_group { + flag: "/MTd" + } + } + flag_set { + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "/DEFAULTLIB:libcmtd.lib" + } + } + requires: { feature: 'dbg'} + } + + feature { + name: 'dynamic_link_msvcrt_debug' + flag_set { + action: 'c-compile' + action: 'c++-compile' + flag_group { + flag: "/MDd" + } + } + flag_set { + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "/DEFAULTLIB:msvcrtd.lib" + } + } + requires: { feature: 'dbg'} + } + + feature { + name: 'dbg' + flag_set { + action: 'c-compile' + action: 'c++-compile' + flag_group { + flag: "/Od" + flag: "/Z7" + flag: "/DDEBUG" + } + } + flag_set { + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "/DEBUG:FULL" + flag: "/INCREMENTAL:NO" + } + } + implies: 'generate_pdb_file' + } + + feature { + name: 'fastbuild' + flag_set { + action: 'c-compile' + action: 'c++-compile' + flag_group { + flag: "/Od" + flag: "/Z7" + flag: "/DDEBUG" + } + } + flag_set { + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "/DEBUG:FASTLINK" + flag: "/INCREMENTAL:NO" + } + } + implies: 'generate_pdb_file' + } + + feature { + name: 'opt' + flag_set { + action: 'c-compile' + action: 'c++-compile' + flag_group { + flag: "/O2" + flag: "/DNDEBUG" + } + } + } + + feature { + name: 'user_compile_flags' + flag_set { + expand_if_all_available: 'user_compile_flags' + action: 'preprocess-assemble' + action: 'c-compile' + action: 'c++-compile' + action: 'c++-header-parsing' + action: 'c++-module-compile' + action: 'c++-module-codegen' + flag_group { + iterate_over: 'user_compile_flags' + flag: '%{user_compile_flags}' + } + } + } + + feature { + name: 'sysroot' + flag_set { + expand_if_all_available: 'sysroot' + action: 'assemble' + action: 'preprocess-assemble' + action: 'c-compile' + action: 'c++-compile' + action: 'c++-header-parsing' + action: 'c++-module-compile' + action: 'c++-module-codegen' + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + iterate_over: 'sysroot' + flag: '--sysroot=%{sysroot}' + } + } + } + + feature { + name: 'unfiltered_compile_flags' + flag_set { + expand_if_all_available: 'unfiltered_compile_flags' + action: 'preprocess-assemble' + action: 'c-compile' + action: 'c++-compile' + action: 'c++-header-parsing' + action: 'c++-module-compile' + action: 'c++-module-codegen' + flag_group { + iterate_over: 'unfiltered_compile_flags' + flag: '%{unfiltered_compile_flags}' + } + } + } + + feature { + name: 'compiler_output_flags' + flag_set { + action: 'assemble' + flag_group { + expand_if_all_available: 'output_file' + expand_if_none_available: 'output_assembly_file' + expand_if_none_available: 'output_preprocess_file' + flag: '/Fo%{output_file}' + flag: '/Zi' + } + } + flag_set { + action: 'preprocess-assemble' + action: 'c-compile' + action: 'c++-compile' + action: 'c++-header-parsing' + action: 'c++-module-compile' + action: 'c++-module-codegen' + flag_group { + expand_if_all_available: 'output_file' + expand_if_none_available: 'output_assembly_file' + expand_if_none_available: 'output_preprocess_file' + flag: '/Fo%{output_file}' + } + flag_group { + expand_if_all_available: 'output_file' + expand_if_all_available: 'output_assembly_file' + flag: '/Fa%{output_file}' + } + flag_group { + expand_if_all_available: 'output_file' + expand_if_all_available: 'output_preprocess_file' + flag: '/P' + flag: '/Fi%{output_file}' + } + } + } + + feature { + name: 'compiler_input_flags' + flag_set { + action: 'assemble' + action: 'preprocess-assemble' + action: 'c-compile' + action: 'c++-compile' + action: 'c++-header-parsing' + action: 'c++-module-compile' + action: 'c++-module-codegen' + flag_group { + expand_if_all_available: 'source_file' + flag: '/c' + flag: '%{source_file}' + } + } + } + + feature { + name : 'def_file', + flag_set { + expand_if_all_available: 'def_file_path' + action: 'c++-link-executable' + action: 'c++-link-dynamic-library' + action: "c++-link-nodeps-dynamic-library" + flag_group { + flag: "/DEF:%{def_file_path}" + # We can specify a different DLL name in DEF file, /ignore:4070 suppresses + # the warning message about DLL name doesn't match the default one. + # See https://msdn.microsoft.com/en-us/library/sfkk2fz7.aspx + flag: "/ignore:4070" + } + } + } + + feature { + name: 'windows_export_all_symbols' + } + + feature { + name: 'no_windows_export_all_symbols' + } + + linking_mode_flags { mode: DYNAMIC } +} diff --git a/third_party/toolchains/preconfig/ubuntu14.04/gcc-nvcc/clang/bin/crosstool_wrapper_driver_is_not_gcc b/third_party/toolchains/preconfig/ubuntu14.04/gcc-nvcc/clang/bin/crosstool_wrapper_driver_is_not_gcc new file mode 100755 index 0000000000..63893d3722 --- /dev/null +++ b/third_party/toolchains/preconfig/ubuntu14.04/gcc-nvcc/clang/bin/crosstool_wrapper_driver_is_not_gcc @@ -0,0 +1,264 @@ +#!/usr/bin/env python +# Copyright 2015 The TensorFlow Authors. All Rights Reserved. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +# ============================================================================== + +"""Crosstool wrapper for compiling CUDA programs. + +SYNOPSIS: + crosstool_wrapper_is_not_gcc [options passed in by cc_library() + or cc_binary() rule] + +DESCRIPTION: + This script is expected to be called by the cc_library() or cc_binary() bazel + rules. When the option "-x cuda" is present in the list of arguments passed + to this script, it invokes the nvcc CUDA compiler. Most arguments are passed + as is as a string to --compiler-options of nvcc. When "-x cuda" is not + present, this wrapper invokes hybrid_driver_is_not_gcc with the input + arguments as is. + +NOTES: + Changes to the contents of this file must be propagated from + //third_party/gpus/crosstool/crosstool_wrapper_is_not_gcc to + //third_party/gpus/crosstool/v*/*/clang/bin/crosstool_wrapper_is_not_gcc +""" + +from __future__ import print_function + +__author__ = 'keveman@google.com (Manjunath Kudlur)' + +from argparse import ArgumentParser +import os +import subprocess +import re +import sys +import pipes + +# Template values set by cuda_autoconf. +CPU_COMPILER = ('/usr/bin/gcc') +GCC_HOST_COMPILER_PATH = ('/usr/bin/gcc') + +NVCC_PATH = '/usr/local/cuda-9.0/bin/nvcc' +PREFIX_DIR = os.path.dirname(GCC_HOST_COMPILER_PATH) +NVCC_VERSION = '9.0' + +def Log(s): + print('gpus/crosstool: {0}'.format(s)) + + +def GetOptionValue(argv, option): + """Extract the list of values for option from the argv list. + + Args: + argv: A list of strings, possibly the argv passed to main(). + option: The option whose value to extract, without the leading '-'. + + Returns: + A list of values, either directly following the option, + (eg., -opt val1 val2) or values collected from multiple occurrences of + the option (eg., -opt val1 -opt val2). + """ + + parser = ArgumentParser() + parser.add_argument('-' + option, nargs='*', action='append') + args, _ = parser.parse_known_args(argv) + if not args or not vars(args)[option]: + return [] + else: + return sum(vars(args)[option], []) + + +def GetHostCompilerOptions(argv): + """Collect the -isystem, -iquote, and --sysroot option values from argv. + + Args: + argv: A list of strings, possibly the argv passed to main(). + + Returns: + The string that can be used as the --compiler-options to nvcc. + """ + + parser = ArgumentParser() + parser.add_argument('-isystem', nargs='*', action='append') + parser.add_argument('-iquote', nargs='*', action='append') + parser.add_argument('--sysroot', nargs=1) + parser.add_argument('-g', nargs='*', action='append') + parser.add_argument('-fno-canonical-system-headers', action='store_true') + + args, _ = parser.parse_known_args(argv) + + opts = '' + + if args.isystem: + opts += ' -isystem ' + ' -isystem '.join(sum(args.isystem, [])) + if args.iquote: + opts += ' -iquote ' + ' -iquote '.join(sum(args.iquote, [])) + if args.g: + opts += ' -g' + ' -g'.join(sum(args.g, [])) + if args.fno_canonical_system_headers: + opts += ' -fno-canonical-system-headers' + if args.sysroot: + opts += ' --sysroot ' + args.sysroot[0] + + return opts + +def _update_options(nvcc_options): + if NVCC_VERSION in ("7.0",): + return nvcc_options + + update_options = { "relaxed-constexpr" : "expt-relaxed-constexpr" } + return [ update_options[opt] if opt in update_options else opt + for opt in nvcc_options ] + +def GetNvccOptions(argv): + """Collect the -nvcc_options values from argv. + + Args: + argv: A list of strings, possibly the argv passed to main(). + + Returns: + The string that can be passed directly to nvcc. + """ + + parser = ArgumentParser() + parser.add_argument('-nvcc_options', nargs='*', action='append') + + args, _ = parser.parse_known_args(argv) + + if args.nvcc_options: + options = _update_options(sum(args.nvcc_options, [])) + return ' '.join(['--'+a for a in options]) + return '' + + +def InvokeNvcc(argv, log=False): + """Call nvcc with arguments assembled from argv. + + Args: + argv: A list of strings, possibly the argv passed to main(). + log: True if logging is requested. + + Returns: + The return value of calling os.system('nvcc ' + args) + """ + + host_compiler_options = GetHostCompilerOptions(argv) + nvcc_compiler_options = GetNvccOptions(argv) + opt_option = GetOptionValue(argv, 'O') + m_options = GetOptionValue(argv, 'm') + m_options = ''.join([' -m' + m for m in m_options if m in ['32', '64']]) + include_options = GetOptionValue(argv, 'I') + out_file = GetOptionValue(argv, 'o') + depfiles = GetOptionValue(argv, 'MF') + defines = GetOptionValue(argv, 'D') + defines = ''.join([' -D' + define for define in defines]) + undefines = GetOptionValue(argv, 'U') + undefines = ''.join([' -U' + define for define in undefines]) + std_options = GetOptionValue(argv, 'std') + # currently only c++11 is supported by Cuda 7.0 std argument + nvcc_allowed_std_options = ["c++11"] + std_options = ''.join([' -std=' + define + for define in std_options if define in nvcc_allowed_std_options]) + + # The list of source files get passed after the -c option. I don't know of + # any other reliable way to just get the list of source files to be compiled. + src_files = GetOptionValue(argv, 'c') + + # Pass -w through from host to nvcc, but don't do anything fancier with + # warnings-related flags, since they're not necessarily the same across + # compilers. + warning_options = ' -w' if '-w' in argv else '' + + if len(src_files) == 0: + return 1 + if len(out_file) != 1: + return 1 + + opt = (' -O2' if (len(opt_option) > 0 and int(opt_option[0]) > 0) + else ' -g -G') + + includes = (' -I ' + ' -I '.join(include_options) + if len(include_options) > 0 + else '') + + # Unfortunately, there are other options that have -c prefix too. + # So allowing only those look like C/C++ files. + src_files = [f for f in src_files if + re.search('\.cpp$|\.cc$|\.c$|\.cxx$|\.C$', f)] + srcs = ' '.join(src_files) + out = ' -o ' + out_file[0] + + supported_cuda_compute_capabilities = [ "3.0" ] + nvccopts = '-D_FORCE_INLINES ' + for capability in supported_cuda_compute_capabilities: + capability = capability.replace('.', '') + nvccopts += r'-gencode=arch=compute_%s,\"code=sm_%s,compute_%s\" ' % ( + capability, capability, capability) + nvccopts += ' ' + nvcc_compiler_options + nvccopts += undefines + nvccopts += defines + nvccopts += std_options + nvccopts += m_options + nvccopts += warning_options + + if depfiles: + # Generate the dependency file + depfile = depfiles[0] + cmd = (NVCC_PATH + ' ' + nvccopts + + ' --compiler-options "' + host_compiler_options + '"' + + ' --compiler-bindir=' + GCC_HOST_COMPILER_PATH + + ' -I .' + + ' -x cu ' + opt + includes + ' ' + srcs + ' -M -o ' + depfile) + if log: Log(cmd) + exit_status = os.system(cmd) + if exit_status != 0: + return exit_status + + cmd = (NVCC_PATH + ' ' + nvccopts + + ' --compiler-options "' + host_compiler_options + ' -fPIC"' + + ' --compiler-bindir=' + GCC_HOST_COMPILER_PATH + + ' -I .' + + ' -x cu ' + opt + includes + ' -c ' + srcs + out) + + # TODO(zhengxq): for some reason, 'gcc' needs this help to find 'as'. + # Need to investigate and fix. + cmd = 'PATH=' + PREFIX_DIR + ':$PATH ' + cmd + if log: Log(cmd) + return os.system(cmd) + + +def main(): + parser = ArgumentParser() + parser.add_argument('-x', nargs=1) + parser.add_argument('--cuda_log', action='store_true') + args, leftover = parser.parse_known_args(sys.argv[1:]) + + if args.x and args.x[0] == 'cuda': + if args.cuda_log: Log('-x cuda') + leftover = [pipes.quote(s) for s in leftover] + if args.cuda_log: Log('using nvcc') + return InvokeNvcc(leftover, log=args.cuda_log) + + # Strip our flags before passing through to the CPU compiler for files which + # are not -x cuda. We can't just pass 'leftover' because it also strips -x. + # We not only want to pass -x to the CPU compiler, but also keep it in its + # relative location in the argv list (the compiler is actually sensitive to + # this). + cpu_compiler_flags = [flag for flag in sys.argv[1:] + if not flag.startswith(('--cuda_log'))] + + return subprocess.call([CPU_COMPILER] + cpu_compiler_flags) + +if __name__ == '__main__': + sys.exit(main()) diff --git a/third_party/toolchains/preconfig/ubuntu14.04/gcc-nvcc/windows/msvc_wrapper_for_nvcc.bat b/third_party/toolchains/preconfig/ubuntu14.04/gcc-nvcc/windows/msvc_wrapper_for_nvcc.bat new file mode 100755 index 0000000000..e896e654fd --- /dev/null +++ b/third_party/toolchains/preconfig/ubuntu14.04/gcc-nvcc/windows/msvc_wrapper_for_nvcc.bat @@ -0,0 +1,20 @@ +:: Copyright 2015 The TensorFlow Authors. All Rights Reserved. +:: +:: Licensed under the Apache License, Version 2.0 (the "License"); +:: you may not use this file except in compliance with the License. +:: You may obtain a copy of the License at +:: +:: http://www.apache.org/licenses/LICENSE-2.0 +:: +:: Unless required by applicable law or agreed to in writing, software +:: distributed under the License is distributed on an "AS IS" BASIS, +:: WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +:: See the License for the specific language governing permissions and +:: limitations under the License. +:: ============================================================================= + +:: Invoke msvc_wrapper_for_nvcc.py, which is located in the same directory. +@echo OFF +set arg0=%~0 +for %%F in ("%arg0%") do set DRIVER_BIN=%%~dpF +"/usr/bin/python3" -B "%DRIVER_BIN%\msvc_wrapper_for_nvcc.py" %* diff --git a/third_party/toolchains/preconfig/ubuntu14.04/gcc-nvcc/windows/msvc_wrapper_for_nvcc.py b/third_party/toolchains/preconfig/ubuntu14.04/gcc-nvcc/windows/msvc_wrapper_for_nvcc.py new file mode 100755 index 0000000000..859b3196d5 --- /dev/null +++ b/third_party/toolchains/preconfig/ubuntu14.04/gcc-nvcc/windows/msvc_wrapper_for_nvcc.py @@ -0,0 +1,192 @@ +#!/usr/bin/env python +# Copyright 2015 The TensorFlow Authors. All Rights Reserved. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +# ============================================================================== + +"""Crosstool wrapper for compiling CUDA programs with nvcc on Windows. + +DESCRIPTION: + This script is the Windows version of //third_party/gpus/crosstool/crosstool_wrapper_is_not_gcc +""" + +from __future__ import print_function + +from argparse import ArgumentParser +import os +import subprocess +import re +import sys +import pipes + +# Template values set by cuda_autoconf. +CPU_COMPILER = ('/usr/bin/gcc') +GCC_HOST_COMPILER_PATH = ('/usr/bin/gcc') + +NVCC_PATH = '/usr/local/cuda-9.0/bin/nvcc' +NVCC_VERSION = '9.0' +NVCC_TEMP_DIR = "C:\\Windows\\Temp\\nvcc_inter_files_tmp_dir" +supported_cuda_compute_capabilities = [ "3.0" ] + +def Log(s): + print('gpus/crosstool: {0}'.format(s)) + + +def GetOptionValue(argv, option): + """Extract the list of values for option from options. + + Args: + option: The option whose value to extract, without the leading '/'. + + Returns: + 1. A list of values, either directly following the option, + (eg., /opt val1 val2) or values collected from multiple occurrences of + the option (eg., /opt val1 /opt val2). + 2. The leftover options. + """ + + parser = ArgumentParser(prefix_chars='/') + parser.add_argument('/' + option, nargs='*', action='append') + args, leftover = parser.parse_known_args(argv) + if args and vars(args)[option]: + return (sum(vars(args)[option], []), leftover) + return ([], leftover) + +def _update_options(nvcc_options): + if NVCC_VERSION in ("7.0",): + return nvcc_options + + update_options = { "relaxed-constexpr" : "expt-relaxed-constexpr" } + return [ update_options[opt] if opt in update_options else opt + for opt in nvcc_options ] + +def GetNvccOptions(argv): + """Collect the -nvcc_options values from argv. + + Args: + argv: A list of strings, possibly the argv passed to main(). + + Returns: + 1. The string that can be passed directly to nvcc. + 2. The leftover options. + """ + + parser = ArgumentParser() + parser.add_argument('-nvcc_options', nargs='*', action='append') + + args, leftover = parser.parse_known_args(argv) + + if args.nvcc_options: + options = _update_options(sum(args.nvcc_options, [])) + return (['--' + a for a in options], leftover) + return ([], leftover) + + +def InvokeNvcc(argv, log=False): + """Call nvcc with arguments assembled from argv. + + Args: + argv: A list of strings, possibly the argv passed to main(). + log: True if logging is requested. + + Returns: + The return value of calling os.system('nvcc ' + args) + """ + + src_files = [f for f in argv if + re.search('\.cpp$|\.cc$|\.c$|\.cxx$|\.C$', f)] + if len(src_files) == 0: + raise Error('No source files found for cuda compilation.') + + out_file = [ f for f in argv if f.startswith('/Fo') ] + if len(out_file) != 1: + raise Error('Please sepecify exactly one output file for cuda compilation.') + out = ['-o', out_file[0][len('/Fo'):]] + + nvcc_compiler_options, argv = GetNvccOptions(argv) + + opt_option, argv = GetOptionValue(argv, 'O') + opt = ['-g', '-G'] + if (len(opt_option) > 0 and opt_option[0] != 'd'): + opt = ['-O2'] + + include_options, argv = GetOptionValue(argv, 'I') + includes = ["-I " + include for include in include_options] + + defines, argv = GetOptionValue(argv, 'D') + defines = ['-D' + define for define in defines] + + undefines, argv = GetOptionValue(argv, 'U') + undefines = ['-U' + define for define in undefines] + + # The rest of the unrecongized options should be passed to host compiler + host_compiler_options = [option for option in argv if option not in (src_files + out_file)] + + m_options = ["-m64"] + + nvccopts = ['-D_FORCE_INLINES'] + for capability in supported_cuda_compute_capabilities: + capability = capability.replace('.', '') + nvccopts += [r'-gencode=arch=compute_%s,"code=sm_%s,compute_%s"' % ( + capability, capability, capability)] + nvccopts += nvcc_compiler_options + nvccopts += undefines + nvccopts += defines + nvccopts += m_options + nvccopts += ['--compiler-options="' + " ".join(host_compiler_options) + '"'] + nvccopts += ['-x', 'cu'] + opt + includes + out + ['-c'] + src_files + # If we don't specify --keep-dir, nvcc will generate intermediate files under TEMP + # Put them under NVCC_TEMP_DIR instead, then Bazel can ignore files under NVCC_TEMP_DIR during dependency check + # http://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html#options-for-guiding-compiler-driver + # Different actions are sharing NVCC_TEMP_DIR, so we cannot remove it if the directory already exists. + if os.path.isfile(NVCC_TEMP_DIR): + os.remove(NVCC_TEMP_DIR) + if not os.path.exists(NVCC_TEMP_DIR): + os.makedirs(NVCC_TEMP_DIR) + nvccopts += ['--keep', '--keep-dir', NVCC_TEMP_DIR] + cmd = [NVCC_PATH] + nvccopts + if log: + Log(cmd) + proc = subprocess.Popen(cmd, + stdout=sys.stdout, + stderr=sys.stderr, + env=os.environ.copy(), + shell=True) + proc.wait() + return proc.returncode + +def main(): + parser = ArgumentParser() + parser.add_argument('-x', nargs=1) + parser.add_argument('--cuda_log', action='store_true') + args, leftover = parser.parse_known_args(sys.argv[1:]) + + if args.x and args.x[0] == 'cuda': + if args.cuda_log: Log('-x cuda') + leftover = [pipes.quote(s) for s in leftover] + if args.cuda_log: Log('using nvcc') + return InvokeNvcc(leftover, log=args.cuda_log) + + # Strip our flags before passing through to the CPU compiler for files which + # are not -x cuda. We can't just pass 'leftover' because it also strips -x. + # We not only want to pass -x to the CPU compiler, but also keep it in its + # relative location in the argv list (the compiler is actually sensitive to + # this). + cpu_compiler_flags = [flag for flag in sys.argv[1:] + if not flag.startswith(('--cuda_log')) + and not flag.startswith(('-nvcc_options'))] + + return subprocess.call([CPU_COMPILER] + cpu_compiler_flags) + +if __name__ == '__main__': + sys.exit(main()) diff --git a/third_party/toolchains/preconfig/ubuntu14.04/nccl2/BUILD b/third_party/toolchains/preconfig/ubuntu14.04/nccl2/BUILD new file mode 100755 index 0000000000..96ed60d3cf --- /dev/null +++ b/third_party/toolchains/preconfig/ubuntu14.04/nccl2/BUILD @@ -0,0 +1,25 @@ +filegroup( + name = "LICENSE", + visibility = ["//visibility:public"], +) + +cc_library( + name = "nccl", + srcs = ["libnccl.so.2"], + hdrs = ["nccl.h"], + include_prefix = "third_party/nccl", + visibility = ["//visibility:public"], + deps = [ + "@local_config_cuda//cuda:cuda_headers", + ], +) + +genrule( + name = "nccl-files", + outs = [ + "libnccl.so.2", + "nccl.h", + ], + cmd = """cp "/usr/include/nccl.h" "$(@D)/nccl.h" && + cp "/usr/lib/libnccl.so.2" "$(@D)/libnccl.so.2" """, +) diff --git a/third_party/toolchains/preconfig/ubuntu14.04/nccl2/WORKSPACE b/third_party/toolchains/preconfig/ubuntu14.04/nccl2/WORKSPACE new file mode 100644 index 0000000000..1e6662ac91 --- /dev/null +++ b/third_party/toolchains/preconfig/ubuntu14.04/nccl2/WORKSPACE @@ -0,0 +1,2 @@ +# DO NOT EDIT: automatically generated WORKSPACE file for nccl_configure rule +workspace(name = "local_config_nccl") diff --git a/third_party/toolchains/preconfig/ubuntu14.04/py3/BUILD b/third_party/toolchains/preconfig/ubuntu14.04/py3/BUILD new file mode 100755 index 0000000000..e021df9e1e --- /dev/null +++ b/third_party/toolchains/preconfig/ubuntu14.04/py3/BUILD @@ -0,0 +1,176 @@ +licenses(["restricted"]) + +package(default_visibility = ["//visibility:public"]) + +# To build Python C/C++ extension on Windows, we need to link to python import library pythonXY.lib +# See https://docs.python.org/3/extending/windows.html +cc_import( + name = "python_lib", + interface_library = select({ + ":windows": ":python_import_lib", + # A placeholder for Unix platforms which makes --no_build happy. + "//conditions:default": "not-existing.lib", + }), + system_provided = 1, +) + +cc_library( + name = "python_headers", + hdrs = [":python_include"], + includes = ["python_include"], + deps = select({ + ":windows": [":python_lib"], + "//conditions:default": [], + }), +) + +cc_library( + name = "numpy_headers", + hdrs = [":numpy_include"], + includes = ["numpy_include"], +) + +config_setting( + name = "windows", + values = {"cpu": "x64_windows"}, + visibility = ["//visibility:public"], +) + +genrule( + name = "python_include", + outs = [ + "python_include/Python-ast.h", + "python_include/Python.h", + "python_include/abstract.h", + "python_include/accu.h", + "python_include/asdl.h", + "python_include/ast.h", + "python_include/bitset.h", + "python_include/bltinmodule.h", + "python_include/boolobject.h", + "python_include/bytearrayobject.h", + "python_include/bytes_methods.h", + "python_include/bytesobject.h", + "python_include/cellobject.h", + "python_include/ceval.h", + "python_include/classobject.h", + "python_include/code.h", + "python_include/codecs.h", + "python_include/compile.h", + "python_include/complexobject.h", + "python_include/datetime.h", + "python_include/descrobject.h", + "python_include/dictobject.h", + "python_include/dtoa.h", + "python_include/dynamic_annotations.h", + "python_include/enumobject.h", + "python_include/errcode.h", + "python_include/eval.h", + "python_include/fileobject.h", + "python_include/fileutils.h", + "python_include/floatobject.h", + "python_include/frameobject.h", + "python_include/funcobject.h", + "python_include/genobject.h", + "python_include/graminit.h", + "python_include/grammar.h", + "python_include/import.h", + "python_include/intrcheck.h", + "python_include/iterobject.h", + "python_include/listobject.h", + "python_include/longintrepr.h", + "python_include/longobject.h", + "python_include/marshal.h", + "python_include/memoryobject.h", + "python_include/metagrammar.h", + "python_include/methodobject.h", + "python_include/modsupport.h", + "python_include/moduleobject.h", + "python_include/namespaceobject.h", + "python_include/node.h", + "python_include/object.h", + "python_include/objimpl.h", + "python_include/opcode.h", + "python_include/osdefs.h", + "python_include/parsetok.h", + "python_include/patchlevel.h", + "python_include/pgen.h", + "python_include/pgenheaders.h", + "python_include/py_curses.h", + "python_include/pyarena.h", + "python_include/pyatomic.h", + "python_include/pycapsule.h", + "python_include/pyconfig.h", + "python_include/pyctype.h", + "python_include/pydebug.h", + "python_include/pyerrors.h", + "python_include/pyexpat.h", + "python_include/pyfpe.h", + "python_include/pygetopt.h", + "python_include/pyhash.h", + "python_include/pymacconfig.h", + "python_include/pymacro.h", + "python_include/pymath.h", + "python_include/pymem.h", + "python_include/pyport.h", + "python_include/pystate.h", + "python_include/pystrcmp.h", + "python_include/pystrtod.h", + "python_include/pythonrun.h", + "python_include/pythread.h", + "python_include/pytime.h", + "python_include/rangeobject.h", + "python_include/setobject.h", + "python_include/sliceobject.h", + "python_include/structmember.h", + "python_include/structseq.h", + "python_include/symtable.h", + "python_include/sysmodule.h", + "python_include/token.h", + "python_include/traceback.h", + "python_include/tupleobject.h", + "python_include/typeslots.h", + "python_include/ucnhash.h", + "python_include/unicodeobject.h", + "python_include/warnings.h", + "python_include/weakrefobject.h", + ], + cmd = """ +cp "/usr/include/python3.4m/Python-ast.h" "$(@D)/python_include/Python-ast.h" && cp "/usr/include/python3.4m/Python.h" "$(@D)/python_include/Python.h" && cp "/usr/include/python3.4m/abstract.h" "$(@D)/python_include/abstract.h" && cp "/usr/include/python3.4m/accu.h" "$(@D)/python_include/accu.h" && cp "/usr/include/python3.4m/asdl.h" "$(@D)/python_include/asdl.h" && cp "/usr/include/python3.4m/ast.h" "$(@D)/python_include/ast.h" && cp "/usr/include/python3.4m/bitset.h" "$(@D)/python_include/bitset.h" && cp "/usr/include/python3.4m/bltinmodule.h" "$(@D)/python_include/bltinmodule.h" && cp "/usr/include/python3.4m/boolobject.h" "$(@D)/python_include/boolobject.h" && cp "/usr/include/python3.4m/bytearrayobject.h" "$(@D)/python_include/bytearrayobject.h" && cp "/usr/include/python3.4m/bytes_methods.h" "$(@D)/python_include/bytes_methods.h" && cp "/usr/include/python3.4m/bytesobject.h" "$(@D)/python_include/bytesobject.h" && cp "/usr/include/python3.4m/cellobject.h" "$(@D)/python_include/cellobject.h" && cp "/usr/include/python3.4m/ceval.h" "$(@D)/python_include/ceval.h" && cp "/usr/include/python3.4m/classobject.h" "$(@D)/python_include/classobject.h" && cp "/usr/include/python3.4m/code.h" "$(@D)/python_include/code.h" && cp "/usr/include/python3.4m/codecs.h" "$(@D)/python_include/codecs.h" && cp "/usr/include/python3.4m/compile.h" "$(@D)/python_include/compile.h" && cp "/usr/include/python3.4m/complexobject.h" "$(@D)/python_include/complexobject.h" && cp "/usr/include/python3.4m/datetime.h" "$(@D)/python_include/datetime.h" && cp "/usr/include/python3.4m/descrobject.h" "$(@D)/python_include/descrobject.h" && cp "/usr/include/python3.4m/dictobject.h" "$(@D)/python_include/dictobject.h" && cp "/usr/include/python3.4m/dtoa.h" "$(@D)/python_include/dtoa.h" && cp "/usr/include/python3.4m/dynamic_annotations.h" "$(@D)/python_include/dynamic_annotations.h" && cp "/usr/include/python3.4m/enumobject.h" "$(@D)/python_include/enumobject.h" && cp "/usr/include/python3.4m/errcode.h" "$(@D)/python_include/errcode.h" && cp "/usr/include/python3.4m/eval.h" "$(@D)/python_include/eval.h" && cp "/usr/include/python3.4m/fileobject.h" "$(@D)/python_include/fileobject.h" && cp "/usr/include/python3.4m/fileutils.h" "$(@D)/python_include/fileutils.h" && cp "/usr/include/python3.4m/floatobject.h" "$(@D)/python_include/floatobject.h" && cp "/usr/include/python3.4m/frameobject.h" "$(@D)/python_include/frameobject.h" && cp "/usr/include/python3.4m/funcobject.h" "$(@D)/python_include/funcobject.h" && cp "/usr/include/python3.4m/genobject.h" "$(@D)/python_include/genobject.h" && cp "/usr/include/python3.4m/graminit.h" "$(@D)/python_include/graminit.h" && cp "/usr/include/python3.4m/grammar.h" "$(@D)/python_include/grammar.h" && cp "/usr/include/python3.4m/import.h" "$(@D)/python_include/import.h" && cp "/usr/include/python3.4m/intrcheck.h" "$(@D)/python_include/intrcheck.h" && cp "/usr/include/python3.4m/iterobject.h" "$(@D)/python_include/iterobject.h" && cp "/usr/include/python3.4m/listobject.h" "$(@D)/python_include/listobject.h" && cp "/usr/include/python3.4m/longintrepr.h" "$(@D)/python_include/longintrepr.h" && cp "/usr/include/python3.4m/longobject.h" "$(@D)/python_include/longobject.h" && cp "/usr/include/python3.4m/marshal.h" "$(@D)/python_include/marshal.h" && cp "/usr/include/python3.4m/memoryobject.h" "$(@D)/python_include/memoryobject.h" && cp "/usr/include/python3.4m/metagrammar.h" "$(@D)/python_include/metagrammar.h" && cp "/usr/include/python3.4m/methodobject.h" "$(@D)/python_include/methodobject.h" && cp "/usr/include/python3.4m/modsupport.h" "$(@D)/python_include/modsupport.h" && cp "/usr/include/python3.4m/moduleobject.h" "$(@D)/python_include/moduleobject.h" && cp "/usr/include/python3.4m/namespaceobject.h" "$(@D)/python_include/namespaceobject.h" && cp "/usr/include/python3.4m/node.h" "$(@D)/python_include/node.h" && cp "/usr/include/python3.4m/object.h" "$(@D)/python_include/object.h" && cp "/usr/include/python3.4m/objimpl.h" "$(@D)/python_include/objimpl.h" && cp "/usr/include/python3.4m/opcode.h" "$(@D)/python_include/opcode.h" && cp "/usr/include/python3.4m/osdefs.h" "$(@D)/python_include/osdefs.h" && cp "/usr/include/python3.4m/parsetok.h" "$(@D)/python_include/parsetok.h" && cp "/usr/include/python3.4m/patchlevel.h" "$(@D)/python_include/patchlevel.h" && cp "/usr/include/python3.4m/pgen.h" "$(@D)/python_include/pgen.h" && cp "/usr/include/python3.4m/pgenheaders.h" "$(@D)/python_include/pgenheaders.h" && cp "/usr/include/python3.4m/py_curses.h" "$(@D)/python_include/py_curses.h" && cp "/usr/include/python3.4m/pyarena.h" "$(@D)/python_include/pyarena.h" && cp "/usr/include/python3.4m/pyatomic.h" "$(@D)/python_include/pyatomic.h" && cp "/usr/include/python3.4m/pycapsule.h" "$(@D)/python_include/pycapsule.h" && cp "/usr/include/python3.4m/pyconfig.h" "$(@D)/python_include/pyconfig.h" && cp "/usr/include/python3.4m/pyctype.h" "$(@D)/python_include/pyctype.h" && cp "/usr/include/python3.4m/pydebug.h" "$(@D)/python_include/pydebug.h" && cp "/usr/include/python3.4m/pyerrors.h" "$(@D)/python_include/pyerrors.h" && cp "/usr/include/python3.4m/pyexpat.h" "$(@D)/python_include/pyexpat.h" && cp "/usr/include/python3.4m/pyfpe.h" "$(@D)/python_include/pyfpe.h" && cp "/usr/include/python3.4m/pygetopt.h" "$(@D)/python_include/pygetopt.h" && cp "/usr/include/python3.4m/pyhash.h" "$(@D)/python_include/pyhash.h" && cp "/usr/include/python3.4m/pymacconfig.h" "$(@D)/python_include/pymacconfig.h" && cp "/usr/include/python3.4m/pymacro.h" "$(@D)/python_include/pymacro.h" && cp "/usr/include/python3.4m/pymath.h" "$(@D)/python_include/pymath.h" && cp "/usr/include/python3.4m/pymem.h" "$(@D)/python_include/pymem.h" && cp "/usr/include/python3.4m/pyport.h" "$(@D)/python_include/pyport.h" && cp "/usr/include/python3.4m/pystate.h" "$(@D)/python_include/pystate.h" && cp "/usr/include/python3.4m/pystrcmp.h" "$(@D)/python_include/pystrcmp.h" && cp "/usr/include/python3.4m/pystrtod.h" "$(@D)/python_include/pystrtod.h" && cp "/usr/include/python3.4m/pythonrun.h" "$(@D)/python_include/pythonrun.h" && cp "/usr/include/python3.4m/pythread.h" "$(@D)/python_include/pythread.h" && cp "/usr/include/python3.4m/pytime.h" "$(@D)/python_include/pytime.h" && cp "/usr/include/python3.4m/rangeobject.h" "$(@D)/python_include/rangeobject.h" && cp "/usr/include/python3.4m/setobject.h" "$(@D)/python_include/setobject.h" && cp "/usr/include/python3.4m/sliceobject.h" "$(@D)/python_include/sliceobject.h" && cp "/usr/include/python3.4m/structmember.h" "$(@D)/python_include/structmember.h" && cp "/usr/include/python3.4m/structseq.h" "$(@D)/python_include/structseq.h" && cp "/usr/include/python3.4m/symtable.h" "$(@D)/python_include/symtable.h" && cp "/usr/include/python3.4m/sysmodule.h" "$(@D)/python_include/sysmodule.h" && cp "/usr/include/python3.4m/token.h" "$(@D)/python_include/token.h" && cp "/usr/include/python3.4m/traceback.h" "$(@D)/python_include/traceback.h" && cp "/usr/include/python3.4m/tupleobject.h" "$(@D)/python_include/tupleobject.h" && cp "/usr/include/python3.4m/typeslots.h" "$(@D)/python_include/typeslots.h" && cp "/usr/include/python3.4m/ucnhash.h" "$(@D)/python_include/ucnhash.h" && cp "/usr/include/python3.4m/unicodeobject.h" "$(@D)/python_include/unicodeobject.h" && cp "/usr/include/python3.4m/warnings.h" "$(@D)/python_include/warnings.h" && cp "/usr/include/python3.4m/weakrefobject.h" "$(@D)/python_include/weakrefobject.h" + """, +) + +genrule( + name = "numpy_include", + outs = [ + "numpy_include/numpy/__multiarray_api.h", + "numpy_include/numpy/__ufunc_api.h", + "numpy_include/numpy/_neighborhood_iterator_imp.h", + "numpy_include/numpy/_numpyconfig.h", + "numpy_include/numpy/arrayobject.h", + "numpy_include/numpy/arrayscalars.h", + "numpy_include/numpy/halffloat.h", + "numpy_include/numpy/multiarray_api.txt", + "numpy_include/numpy/ndarrayobject.h", + "numpy_include/numpy/ndarraytypes.h", + "numpy_include/numpy/noprefix.h", + "numpy_include/numpy/npy_1_7_deprecated_api.h", + "numpy_include/numpy/npy_3kcompat.h", + "numpy_include/numpy/npy_common.h", + "numpy_include/numpy/npy_cpu.h", + "numpy_include/numpy/npy_endian.h", + "numpy_include/numpy/npy_interrupt.h", + "numpy_include/numpy/npy_math.h", + "numpy_include/numpy/npy_no_deprecated_api.h", + "numpy_include/numpy/npy_os.h", + "numpy_include/numpy/numpyconfig.h", + "numpy_include/numpy/old_defines.h", + "numpy_include/numpy/oldnumeric.h", + "numpy_include/numpy/ufunc_api.txt", + "numpy_include/numpy/ufuncobject.h", + "numpy_include/numpy/utils.h", + ], + cmd = """ +cp "/usr/local/lib/python3.4/dist-packages/numpy/core/include/numpy/__multiarray_api.h" "$(@D)/numpy_include/numpy/__multiarray_api.h" && cp "/usr/local/lib/python3.4/dist-packages/numpy/core/include/numpy/__ufunc_api.h" "$(@D)/numpy_include/numpy/__ufunc_api.h" && cp "/usr/local/lib/python3.4/dist-packages/numpy/core/include/numpy/_neighborhood_iterator_imp.h" "$(@D)/numpy_include/numpy/_neighborhood_iterator_imp.h" && cp "/usr/local/lib/python3.4/dist-packages/numpy/core/include/numpy/_numpyconfig.h" "$(@D)/numpy_include/numpy/_numpyconfig.h" && cp "/usr/local/lib/python3.4/dist-packages/numpy/core/include/numpy/arrayobject.h" "$(@D)/numpy_include/numpy/arrayobject.h" && cp "/usr/local/lib/python3.4/dist-packages/numpy/core/include/numpy/arrayscalars.h" "$(@D)/numpy_include/numpy/arrayscalars.h" && cp "/usr/local/lib/python3.4/dist-packages/numpy/core/include/numpy/halffloat.h" "$(@D)/numpy_include/numpy/halffloat.h" && cp "/usr/local/lib/python3.4/dist-packages/numpy/core/include/numpy/multiarray_api.txt" "$(@D)/numpy_include/numpy/multiarray_api.txt" && cp "/usr/local/lib/python3.4/dist-packages/numpy/core/include/numpy/ndarrayobject.h" "$(@D)/numpy_include/numpy/ndarrayobject.h" && cp "/usr/local/lib/python3.4/dist-packages/numpy/core/include/numpy/ndarraytypes.h" "$(@D)/numpy_include/numpy/ndarraytypes.h" && cp "/usr/local/lib/python3.4/dist-packages/numpy/core/include/numpy/noprefix.h" "$(@D)/numpy_include/numpy/noprefix.h" && cp "/usr/local/lib/python3.4/dist-packages/numpy/core/include/numpy/npy_1_7_deprecated_api.h" "$(@D)/numpy_include/numpy/npy_1_7_deprecated_api.h" && cp "/usr/local/lib/python3.4/dist-packages/numpy/core/include/numpy/npy_3kcompat.h" "$(@D)/numpy_include/numpy/npy_3kcompat.h" && cp "/usr/local/lib/python3.4/dist-packages/numpy/core/include/numpy/npy_common.h" "$(@D)/numpy_include/numpy/npy_common.h" && cp "/usr/local/lib/python3.4/dist-packages/numpy/core/include/numpy/npy_cpu.h" "$(@D)/numpy_include/numpy/npy_cpu.h" && cp "/usr/local/lib/python3.4/dist-packages/numpy/core/include/numpy/npy_endian.h" "$(@D)/numpy_include/numpy/npy_endian.h" && cp "/usr/local/lib/python3.4/dist-packages/numpy/core/include/numpy/npy_interrupt.h" "$(@D)/numpy_include/numpy/npy_interrupt.h" && cp "/usr/local/lib/python3.4/dist-packages/numpy/core/include/numpy/npy_math.h" "$(@D)/numpy_include/numpy/npy_math.h" && cp "/usr/local/lib/python3.4/dist-packages/numpy/core/include/numpy/npy_no_deprecated_api.h" "$(@D)/numpy_include/numpy/npy_no_deprecated_api.h" && cp "/usr/local/lib/python3.4/dist-packages/numpy/core/include/numpy/npy_os.h" "$(@D)/numpy_include/numpy/npy_os.h" && cp "/usr/local/lib/python3.4/dist-packages/numpy/core/include/numpy/numpyconfig.h" "$(@D)/numpy_include/numpy/numpyconfig.h" && cp "/usr/local/lib/python3.4/dist-packages/numpy/core/include/numpy/old_defines.h" "$(@D)/numpy_include/numpy/old_defines.h" && cp "/usr/local/lib/python3.4/dist-packages/numpy/core/include/numpy/oldnumeric.h" "$(@D)/numpy_include/numpy/oldnumeric.h" && cp "/usr/local/lib/python3.4/dist-packages/numpy/core/include/numpy/ufunc_api.txt" "$(@D)/numpy_include/numpy/ufunc_api.txt" && cp "/usr/local/lib/python3.4/dist-packages/numpy/core/include/numpy/ufuncobject.h" "$(@D)/numpy_include/numpy/ufuncobject.h" && cp "/usr/local/lib/python3.4/dist-packages/numpy/core/include/numpy/utils.h" "$(@D)/numpy_include/numpy/utils.h" + """, +) diff --git a/third_party/toolchains/preconfig/ubuntu14.04/py3/WORKSPACE b/third_party/toolchains/preconfig/ubuntu14.04/py3/WORKSPACE new file mode 100644 index 0000000000..1d298fefa3 --- /dev/null +++ b/third_party/toolchains/preconfig/ubuntu14.04/py3/WORKSPACE @@ -0,0 +1,2 @@ +# DO NOT EDIT: automatically generated WORKSPACE file for python_configure rule +workspace(name = "local_config_python") diff --git a/third_party/zlib.BUILD b/third_party/zlib.BUILD index d164ee719c..33694eaaae 100644 --- a/third_party/zlib.BUILD +++ b/third_party/zlib.BUILD @@ -2,18 +2,6 @@ package(default_visibility = ["//visibility:public"]) licenses(["notice"]) # BSD/MIT-like license (for zlib) -config_setting( - name = "windows", - values = {"cpu": "x64_windows"}, - visibility = ["//visibility:public"], -) - -config_setting( - name = "windows_msvc", - values = {"cpu": "x64_windows_msvc"}, - visibility = ["//visibility:public"], -) - cc_library( name = "zlib", srcs = [ @@ -45,8 +33,7 @@ cc_library( ], hdrs = ["zlib.h"], copts = select({ - ":windows": [], - ":windows_msvc": [], + "@org_tensorflow//tensorflow:windows": [], "//conditions:default": [ "-Wno-shift-negative-value", "-DZ_HAVE_UNISTD_H", |