Merge branch 'master' into avijit/add-cpu-backend

9 files changed, 1193 insertions, 47 deletions

diff --git a/third_party/clang_toolchain/download_clang.bzl b/third_party/clang_toolchain/download_clang.bzl
index 5ef47cdd0d..e782739661 100644
--- a/third_party/clang_toolchain/download_clang.bzl
+++ b/third_party/clang_toolchain/download_clang.bzl
@@ -39,15 +39,15 @@ def download_clang(repo_ctx, out_folder):
 
     # 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 = "338452"
+    CLANG_REVISION = "340427"
     CLANG_SUB_REVISION = 1
 
     package_version = "%s-%s" % (CLANG_REVISION, CLANG_SUB_REVISION)
 
     checksums = {
-        "Linux_x64": "213ba23a0a9855ede5041f66661caa9c5c59a573ec60b82a31839f9a97f397bf",
-        "Mac": "4267774201f8cb50c25e081375e87038d58db80064a20a0d9d7fe57ea4357ece",
-        "Win": "a8a5d5b25443c099e2c20d1a0cdce2f1d17e2dba84de66a6dc6a239ce3e78c34",
+        "Linux_x64": "8a8f21fb624fc7be7e91e439a13114847185375bb932db51ba590174ecaf764b",
+        "Mac": "ba894536b7c8d37103a5ddba784f268d55e65bb2ea1200a2cf9f2ef1590eaacd",
+        "Win": "c3f5bd977266dfd011411c94a13e00974b643b70fb0225a5fb030f7f703fa474",
     }
 
     platform_folder = _get_platform_folder(repo_ctx.os.name)
diff --git a/third_party/gpus/crosstool/CROSSTOOL.tpl b/third_party/gpus/crosstool/CROSSTOOL.tpl
index 3972c96a2f..3189cf8e31 100644
--- a/third_party/gpus/crosstool/CROSSTOOL.tpl
+++ b/third_party/gpus/crosstool/CROSSTOOL.tpl
@@ -208,7 +208,7 @@ toolchain {
       action: "c++-link-dynamic-library"
       action: "c++-link-nodeps-dynamic-library"
       flag_group {
-        flag: "-B/usr/bin/"
+        %{linker_bin_path_flag}
       }
     }
   }
@@ -446,7 +446,7 @@ toolchain {
       action: "c++-link-dynamic-library"
       action: "c++-link-nodeps-dynamic-library"
       flag_group {
-        flag: "-B/usr/bin/"
+        %{linker_bin_path_flag}
       }
     }
   }
diff --git a/third_party/gpus/cuda_configure.bzl b/third_party/gpus/cuda_configure.bzl
index f6a39aeaf1..5648b1525a 100644
--- a/third_party/gpus/cuda_configure.bzl
+++ b/third_party/gpus/cuda_configure.bzl
@@ -1303,6 +1303,19 @@ def _create_local_cuda_repository(repository_ctx):
 
     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}"] = """
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 4cb8380938..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" "$(@D)/cuda/lib/libcudart.so.9.0" && cp "/usr/local/cuda-9.0/targets/x86_64-linux/lib/libcudart_static.a" "$(@D)/cuda/lib/libcudart_static.a" && cp "/usr/local/cuda-9.0/targets/x86_64-linux/lib/libcublas.so.9.0" "$(@D)/cuda/lib/libcublas.so.9.0" && cp "/usr/local/cuda-9.0/targets/x86_64-linux/lib/libcusolver.so.9.0" "$(@D)/cuda/lib/libcusolver.so.9.0" && cp "/usr/local/cuda-9.0/targets/x86_64-linux/lib/libcurand.so.9.0" "$(@D)/cuda/lib/libcurand.so.9.0" && cp "/usr/local/cuda-9.0/targets/x86_64-linux/lib/libcufft.so.9.0" "$(@D)/cuda/lib/libcufft.so.9.0" && cp "/usr/lib/x86_64-linux-gnu/libcudnn.so.7" "$(@D)/cuda/lib/libcudnn.so.7" && cp "/usr/local/cuda-9.0/extras/CUPTI/lib64/libcupti.so.9.0" "$(@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": [],
     }),
 )