Switch NCCL to build from open source (version 2.3.5-5) by default.

Note to users manually patching ptxas from a later toolkit version:
Building NCCL requires the same version of ptxas and nvlink.

PiperOrigin-RevId: 215911973

5 files changed, 536 insertions, 366 deletions

diff --git a/third_party/nccl/LICENSE b/third_party/nccl/LICENSE
index 146d9b765c..b958518186 100644
--- a/third_party/nccl/LICENSE
+++ b/third_party/nccl/LICENSE
@@ -1,203 +1,30 @@
-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.
+ Copyright (c) 2015-2018, NVIDIA CORPORATION. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without
+ modification, are permitted provided that the following conditions
+ are met:
+  * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+  * Redistributions in binary form must reproduce the above copyright
+    notice, this list of conditions and the following disclaimer in the
+    documentation and/or other materials provided with the distribution.
+  * Neither the name of NVIDIA CORPORATION, Lawrence Berkeley National
+    Laboratory, the U.S. Department of Energy, nor the names of their
+    contributors may be used to endorse or promote products derived
+    from this software without specific prior written permission.
+
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
+ EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE COPYRIGHT OWNER OR
+ CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+ EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+ PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
+ OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+ The U.S. Department of Energy funded the development of this software
+ under subcontract 7078610 with Lawrence Berkeley National Laboratory.
diff --git a/third_party/nccl/archive.BUILD b/third_party/nccl/archive.BUILD
new file mode 100644
index 0000000000..f57f04c75e
--- /dev/null
+++ b/third_party/nccl/archive.BUILD
@@ -0,0 +1,179 @@
+# NVIDIA NCCL 2
+# A package of optimized primitives for collective multi-GPU communication.
+
+licenses(["restricted"])
+
+exports_files(["LICENSE.txt"])
+
+load(
+    "@local_config_nccl//:build_defs.bzl",
+    "device_link",
+    "gen_nccl_h",
+    "nccl_library",
+    "rdc_copts",
+)
+load(
+    "@local_config_cuda//cuda:build_defs.bzl",
+    "cuda_default_copts",
+)
+
+# Generate the nccl.h header file.
+gen_nccl_h(
+    name = "nccl_h",
+    output = "src/nccl.h",
+    template = "src/nccl.h.in",
+)
+
+nccl_library(
+    name = "src_hdrs",
+    hdrs = [
+        "src/nccl.h",
+        # src/include/common_coll.h #includes "collectives/collectives.h".
+        # All other #includes of collectives.h are patched in process_srcs.
+        "src/collectives/collectives.h",
+    ],
+    strip_include_prefix = "src",
+)
+
+nccl_library(
+    name = "include_hdrs",
+    hdrs = glob(["src/include/*.h"]),
+    strip_include_prefix = "src/include",
+)
+
+filegroup(
+    name = "device_hdrs",
+    srcs = glob(["src/collectives/device/*.h"]),
+)
+
+filegroup(
+    name = "device_srcs",
+    srcs = [
+        "src/collectives/device/all_gather.cu",
+        "src/collectives/device/all_reduce.cu",
+        "src/collectives/device/broadcast.cu",
+        "src/collectives/device/reduce.cu",
+        "src/collectives/device/reduce_scatter.cu",
+    ],
+)
+
+nccl_library(
+    name = "sum",
+    srcs = [
+        ":device_hdrs",
+        ":device_srcs",
+    ],
+    copts = ["-DNCCL_OP=0"] + rdc_copts(),
+    prefix = "sum_",
+    deps = [
+        ":src_hdrs",
+        ":include_hdrs",
+        "@local_config_cuda//cuda:cuda_headers",
+    ],
+    linkstatic = True,
+)
+
+nccl_library(
+    name = "prod",
+    srcs = [
+        ":device_hdrs",
+        ":device_srcs",
+    ],
+    copts = ["-DNCCL_OP=1"] + rdc_copts(),
+    prefix = "_prod",
+    deps = [
+        ":src_hdrs",
+        ":include_hdrs",
+        "@local_config_cuda//cuda:cuda_headers",
+    ],
+    linkstatic = True,
+)
+
+nccl_library(
+    name = "min",
+    srcs = [
+        ":device_hdrs",
+        ":device_srcs",
+    ],
+    copts = ["-DNCCL_OP=2"] + rdc_copts(),
+    prefix = "min_",
+    deps = [
+        ":src_hdrs",
+        ":include_hdrs",
+        "@local_config_cuda//cuda:cuda_headers",
+    ],
+    linkstatic = True,
+)
+
+nccl_library(
+    name = "max",
+    srcs = [
+        ":device_hdrs",
+        ":device_srcs",
+    ],
+    copts = ["-DNCCL_OP=3"] + rdc_copts(),
+    prefix = "max_",
+    deps = [
+        ":src_hdrs",
+        ":include_hdrs",
+        "@local_config_cuda//cuda:cuda_headers",
+    ],
+    linkstatic = True,
+)
+
+nccl_library(
+    name = "functions",
+    srcs = [
+        ":device_hdrs",
+        "src/collectives/device/functions.cu",
+    ],
+    copts = rdc_copts(),
+    deps = [
+        ":src_hdrs",
+        ":include_hdrs",
+        "@local_config_cuda//cuda:cuda_headers",
+    ],
+    linkstatic = True,
+)
+
+device_link(
+    name = "device_code",
+    srcs = [
+        ":functions",
+        ":max",
+        ":min",
+        ":prod",
+        ":sum",
+    ],
+)
+
+# Primary NCCL target.
+nccl_library(
+    name = "nccl",
+    srcs = glob(
+        include = ["src/**/*.cu"],
+        # Exclude device-library code.
+        exclude = ["src/collectives/device/**"],
+    ) + [
+        # Required for header inclusion checking (see
+        # http://docs.bazel.build/versions/master/be/c-cpp.html#hdrs).
+        # Files in src/ which #include "nccl.h" load it from there rather than
+        # from the virtual includes directory.
+        "src/nccl.h",
+    ],
+    hdrs = ["src/nccl.h"],
+    include_prefix = "third_party/nccl",
+    strip_include_prefix = "src",
+    copts = cuda_default_copts(),
+    deps = [
+        ":device_code",
+        ":functions",
+        ":include_hdrs",
+        ":max",
+        ":min",
+        ":prod",
+        ":src_hdrs",
+        ":sum",
+    ],
+    visibility = ["//visibility:public"],
+)
diff --git a/third_party/nccl/build_defs.bzl.tpl b/third_party/nccl/build_defs.bzl.tpl
new file mode 100644
index 0000000000..ede1d3dad5
--- /dev/null
+++ b/third_party/nccl/build_defs.bzl.tpl
@@ -0,0 +1,210 @@
+"""Repository rule for NCCL."""
+
+load("@local_config_cuda//cuda:build_defs.bzl", "cuda_default_copts")
+
+def _gen_nccl_h_impl(ctx):
+    """Creates nccl.h from a template."""
+    ctx.actions.expand_template(
+        output = ctx.outputs.output,
+        template = ctx.file.template,
+        substitutions = {
+            "${nccl:Major}": "2",
+            "${nccl:Minor}": "3",
+            "${nccl:Patch}": "5",
+            "${nccl:Suffix}": "",
+            "${nccl:Version}": "2305",
+        },
+    )
+gen_nccl_h = rule(
+    implementation = _gen_nccl_h_impl,
+    attrs = {
+        "template": attr.label(allow_single_file = True),
+        "output": attr.output(),
+    },
+)
+"""Creates the NCCL header file."""
+
+
+def _process_srcs_impl(ctx):
+    """Appends .cc to .cu files, patches include directives."""
+    files = []
+    for src in ctx.files.srcs:
+        if not src.is_source:
+          # Process only once, specifically "src/nccl.h".
+          files.append(src)
+          continue
+        name = src.basename
+        if src.extension == "cu":
+            name = ctx.attr.prefix + name + ".cc"
+        file = ctx.actions.declare_file(name, sibling = src)
+        ctx.actions.expand_template(
+            output = file,
+            template = src,
+            substitutions = {
+                "\"collectives.h": "\"collectives/collectives.h",
+                "\"../collectives.h": "\"collectives/collectives.h",
+                "#if __CUDACC_VER_MAJOR__":
+                    "#if defined __CUDACC_VER_MAJOR__ && __CUDACC_VER_MAJOR__",
+                # Substitutions are applied in order.
+                "std::nullptr_t": "nullptr_t",
+                "nullptr_t": "std::nullptr_t",
+            },
+        )
+        files.append(file)
+    return [DefaultInfo(files = depset(files))]
+_process_srcs = rule(
+    implementation = _process_srcs_impl,
+    attrs = {
+        "srcs": attr.label_list(allow_files = True),
+        "prefix": attr.string(default = ""),
+    },
+)
+"""Processes the NCCL srcs so they can be compiled with bazel and clang."""
+
+
+def nccl_library(name, srcs=None, hdrs=None, prefix=None, **kwargs):
+    """Processes the srcs and hdrs and creates a cc_library."""
+
+    _process_srcs(
+        name = name + "_srcs",
+        srcs = srcs,
+        prefix = prefix,
+    )
+    _process_srcs(
+        name = name + "_hdrs",
+        srcs = hdrs,
+    )
+
+    native.cc_library(
+        name = name,
+        srcs = [name + "_srcs"] if srcs else [],
+        hdrs = [name + "_hdrs"] if hdrs else [],
+        **kwargs
+    )
+
+
+def rdc_copts():
+    """Returns copts for compiling relocatable device code."""
+
+    # The global functions can not have a lower register count than the
+    # device functions. This is enforced by setting a fixed register count.
+    # https://github.com/NVIDIA/nccl/blob/f93fe9bfd94884cec2ba711897222e0df5569a53/makefiles/common.mk#L48
+    maxrregcount = "-maxrregcount=96"
+
+    return cuda_default_copts() + select({
+          "@local_config_cuda//cuda:using_nvcc": [
+              "-nvcc_options",
+              "relocatable-device-code=true",
+              "-nvcc_options",
+              "ptxas-options=" + maxrregcount,
+          ],
+          "@local_config_cuda//cuda:using_clang": [
+              "-fcuda-rdc",
+              "-Xcuda-ptxas",
+              maxrregcount,
+          ],
+          "//conditions:default": [],
+      }) + ["-fvisibility=hidden"]
+
+
+def _filter_impl(ctx):
+    suffix = ctx.attr.suffix
+    files = [src for src in ctx.files.srcs if src.path.endswith(suffix)]
+    return [DefaultInfo(files = depset(files))]
+_filter = rule(
+    implementation = _filter_impl,
+    attrs = {
+        "srcs": attr.label_list(allow_files = True),
+        "suffix": attr.string(),
+    },
+)
+"""Filters the srcs to the ones ending with suffix."""
+
+
+def _gen_link_src_impl(ctx):
+    ctx.actions.expand_template(
+        output = ctx.outputs.output,
+        template = ctx.file.template,
+        substitutions = {
+            "REGISTERLINKBINARYFILE": '"%s"' % ctx.file.register_hdr.short_path,
+            "FATBINFILE": '"%s"' % ctx.file.fatbin_hdr.short_path,
+        },
+    )
+_gen_link_src = rule(
+    implementation = _gen_link_src_impl,
+    attrs = {
+        "register_hdr": attr.label(allow_single_file = True),
+        "fatbin_hdr": attr.label(allow_single_file = True),
+        "template": attr.label(allow_single_file = True),
+        "output": attr.output(),
+    },
+)
+"""Patches the include directives for the link.stub file."""
+
+
+def device_link(name, srcs):
+    """Links seperately compiled relocatable device code into a cc_library."""
+
+    # From .a and .pic.a archives, just use the latter.
+    _filter(
+        name = name + "_pic_a",
+        srcs = srcs,
+        suffix = ".pic.a",
+    )
+
+    # Device-link to cubins for each architecture.
+    images = []
+    cubins = []
+    for arch in %{gpu_architectures}:
+        cubin = "%s_%s.cubin" % (name, arch)
+        register_hdr = "%s_%s.h" % (name, arch)
+        nvlink = "@local_config_nccl//:nvlink"
+        cmd = ("$(location %s) --cpu-arch=X86_64 " % nvlink +
+            "--arch=%s $(SRCS) " % arch +
+            "--register-link-binaries=$(location %s) " % register_hdr +
+            "--output-file=$(location %s)" % cubin)
+        native.genrule(
+            name = "%s_%s" % (name, arch),
+            outs = [register_hdr, cubin],
+            srcs = [name + "_pic_a"],
+            cmd = cmd,
+            tools = [nvlink],
+        )
+        images.append("--image=profile=%s,file=$(location %s)" % (arch, cubin))
+        cubins.append(cubin)
+
+    # Generate fatbin header from all cubins.
+    fatbin_hdr = name + ".fatbin.h"
+    fatbinary = "@local_config_nccl//:cuda/bin/fatbinary"
+    cmd = ("PATH=$$CUDA_TOOLKIT_PATH/bin:$$PATH " + # for bin2c
+          "$(location %s) -64 --cmdline=--compile-only --link " % fatbinary +
+          "--compress-all %s --create=%%{name}.fatbin " % " ".join(images) +
+          "--embedded-fatbin=$@")
+    native.genrule(
+        name = name + "_fatbin_h",
+        outs = [fatbin_hdr],
+        srcs = cubins,
+        cmd = cmd,
+        tools = [fatbinary],
+    )
+
+    # Generate the source file #including the headers generated above.
+    _gen_link_src(
+        name = name + "_cc",
+        # Include just the last one, they are equivalent.
+        register_hdr = register_hdr,
+        fatbin_hdr = fatbin_hdr,
+        template = "@local_config_nccl//:cuda/bin/crt/link.stub",
+        output = name + ".cc",
+    )
+
+    # Compile the source file into the cc_library.
+    native.cc_library(
+        name = name,
+        srcs = [name + "_cc"],
+        textual_hdrs = [register_hdr, fatbin_hdr],
+        deps = [
+            "@local_config_cuda//cuda:cuda_headers",
+            "@local_config_cuda//cuda:cudart_static",
+        ],
+    )
diff --git a/third_party/nccl/nccl_archive.BUILD b/third_party/nccl/nccl_archive.BUILD
deleted file mode 100644
index a05899e38d..0000000000
--- a/third_party/nccl/nccl_archive.BUILD
+++ /dev/null
@@ -1,68 +0,0 @@
-# NVIDIA nccl
-# A package of optimized primitives for collective multi-GPU communication.
-
-licenses(["notice"])  # BSD
-
-exports_files(["LICENSE.txt"])
-
-load("@local_config_cuda//cuda:build_defs.bzl", "cuda_default_copts", "if_cuda")
-
-SRCS = [
-    "src/all_gather.cu",
-    "src/all_reduce.cu",
-    "src/broadcast.cu",
-    "src/core.cu",
-    "src/libwrap.cu",
-    "src/reduce.cu",
-    "src/reduce_scatter.cu",
-]
-
-# Copy .cu to .cu.cc so they can be in srcs of cc_library.
-[
-    genrule(
-        name = "gen_" + src,
-        srcs = [src],
-        outs = [src + ".cc"],
-        cmd = "cp $(location " + src + ") $(location " + src + ".cc)",
-    )
-    for src in SRCS
-]
-
-SRCS_CU_CC = [src + ".cc" for src in SRCS]
-
-cc_library(
-    name = "nccl",
-    srcs = if_cuda(SRCS_CU_CC + glob(["src/*.h"])),
-    hdrs = if_cuda(["src/nccl.h"]),
-    copts = [
-        "-DCUDA_MAJOR=0",
-        "-DCUDA_MINOR=0",
-        "-DNCCL_MAJOR=0",
-        "-DNCCL_MINOR=0",
-        "-DNCCL_PATCH=0",
-        "-Iexternal/nccl_archive/src",
-        "-O3",
-    ] + cuda_default_copts(),
-    include_prefix = "third_party/nccl",
-    linkopts = select({
-        "@org_tensorflow//tensorflow:android": [
-            "-pie",
-        ],
-        "@org_tensorflow//tensorflow:darwin": [
-            "-Wl,-framework",
-            "-Wl,CoreFoundation",
-            "-Wl,-framework",
-            "-Wl,Security",
-        ],
-        "@org_tensorflow//tensorflow:ios": [],
-        "@org_tensorflow//tensorflow:windows": [
-            "-DEFAULTLIB:ws2_32.lib",
-        ],
-        "//conditions:default": [
-            "-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
index d78fe8f3aa..7f00df0962 100644
--- a/third_party/nccl/nccl_configure.bzl
+++ b/third_party/nccl/nccl_configure.bzl
@@ -11,12 +11,16 @@
 load(
     "//third_party/gpus:cuda_configure.bzl",
     "auto_configure_fail",
+    "compute_capabilities",
+    "cuda_toolkit_path",
     "find_cuda_define",
     "matches_version",
 )
 
-_NCCL_INSTALL_PATH = "NCCL_INSTALL_PATH"
+_CUDA_TOOLKIT_PATH = "CUDA_TOOLKIT_PATH"
 _NCCL_HDR_PATH = "NCCL_HDR_PATH"
+_NCCL_INSTALL_PATH = "NCCL_INSTALL_PATH"
+_TF_CUDA_COMPUTE_CAPABILITIES = "TF_CUDA_COMPUTE_CAPABILITIES"
 _TF_NCCL_VERSION = "TF_NCCL_VERSION"
 _TF_NCCL_CONFIG_REPO = "TF_NCCL_CONFIG_REPO"
 
@@ -37,6 +41,12 @@ cc_library(
 """
 
 _NCCL_ARCHIVE_BUILD_CONTENT = """
+exports_files([
+    "cuda/bin/crt/link.stub",
+    "cuda/bin/fatbinary",
+    "nvlink",
+])
+
 filegroup(
   name = "LICENSE",
   data = ["@nccl_archive//:LICENSE.txt"],
@@ -50,113 +60,125 @@ alias(
 )
 """
 
-# 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 _label(file):
+    return Label("//third_party/nccl:{}".format(file))
 
 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.
+    """Finds the NCCL header on the system.
 
-  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
+    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_hdr_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 = repository_ctx.path("%s/nccl.h" % nccl_hdr_path)
-  if not header_path.exists:
-    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
-
+    """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_hdr_path: The NCCL header path.
+      nccl_version: The expected NCCL version.
+
+    Returns:
+      A string containing the library version of NCCL.
+    """
+    header_path = repository_ctx.path("%s/nccl.h" % nccl_hdr_path)
+    if not header_path.exists:
+        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 _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()
-    nccl_hdr_path = repository_ctx.os.environ[_NCCL_HDR_PATH].strip()
-    _check_nccl_version(repository_ctx, nccl_install_path, nccl_hdr_path, nccl_version)
-    repository_ctx.template("BUILD", _NCCL_LOCAL_BUILD_TEMPLATE, {
-        "%{version}": nccl_version,
-        "%{install_path}": nccl_install_path,
-        "%{hdr_path}": nccl_hdr_path,
-    })
-
+    """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", _label("remote.BUILD.tpl"), {
+            "%{target}": repository_ctx.os.environ[_TF_NCCL_CONFIG_REPO],
+        })
+        return
+
+    nccl_version = repository_ctx.os.environ[_TF_NCCL_VERSION].strip()
+    if nccl_version == "":
+        # Alias to open source build from @nccl_archive.
+        repository_ctx.file("BUILD", _NCCL_ARCHIVE_BUILD_CONTENT)
+
+        # TODO(csigg): implement and reuse in cuda_configure.bzl.
+        gpu_architectures = [
+            "sm_" + capability.replace(".", "")
+            for capability in compute_capabilities(repository_ctx)
+        ]
+
+        # Round-about way to make the list unique.
+        gpu_architectures = dict(zip(gpu_architectures, gpu_architectures)).keys()
+        repository_ctx.template("build_defs.bzl", _label("build_defs.bzl.tpl"), {
+            "%{gpu_architectures}": str(gpu_architectures),
+        })
+
+        repository_ctx.symlink(cuda_toolkit_path(repository_ctx), "cuda")
+
+        # Temporary work-around for setups which symlink ptxas to a newer
+        # version. The versions of nvlink and ptxas need to agree, so we find
+        # nvlink next to the real location of ptxas. This is only temporary and
+        # will be removed again soon.
+        nvlink_dir = repository_ctx.path("cuda/bin/ptxas").realpath.dirname
+        repository_ctx.symlink(nvlink_dir.get_child("nvlink"), "nvlink")
+    else:
+        # Create target for locally installed NCCL.
+        nccl_install_path = repository_ctx.os.environ[_NCCL_INSTALL_PATH].strip()
+        nccl_hdr_path = repository_ctx.os.environ[_NCCL_HDR_PATH].strip()
+        _check_nccl_version(repository_ctx, nccl_install_path, nccl_hdr_path, nccl_version)
+        repository_ctx.template("BUILD", _label("system.BUILD.tpl"), {
+            "%{version}": nccl_version,
+            "%{install_path}": nccl_install_path,
+            "%{hdr_path}": nccl_hdr_path,
+        })
 
 nccl_configure = repository_rule(
-    implementation=_nccl_configure_impl,
-    environ=[
-        _NCCL_INSTALL_PATH,
+    implementation = _nccl_configure_impl,
+    environ = [
+        _CUDA_TOOLKIT_PATH,
         _NCCL_HDR_PATH,
+        _NCCL_INSTALL_PATH,
         _TF_NCCL_VERSION,
+        _TF_CUDA_COMPUTE_CAPABILITIES,
+        _TF_NCCL_CONFIG_REPO,
     ],
 )
 """Detects and configures the NCCL configuration.