[all-commits] [llvm/llvm-project] 0060ff: [CUDA] Bump default GPU architecture to sm_35.

[Artem Belevich via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 0060fffc822261ff7350e34371c4456f363f866d
      https://github.com/llvm/llvm-project/commit/0060fffc822261ff7350e34371c4456f363f866d
  Author: Artem Belevich <tra at google.com>
  Date:   2021-08-23 (Mon, 23 Aug 2021)

  Changed paths:
    M clang/lib/Driver/Driver.cpp
    M clang/test/Driver/cuda-external-tools.cu
    M clang/test/Driver/cuda-march.cu
    R clang/test/Driver/cuda-options-freebsd.cu
    M clang/test/Driver/cuda-options.cu
    M clang/test/Driver/cuda-ptxas-path.cu
    M clang/test/Driver/lto.cu
    M clang/test/Driver/openmp-offload.c
    M clang/test/Driver/opt-record.c
    M clang/test/Driver/thinlto.cu

  Log Message:
  -----------
  [CUDA] Bump default GPU architecture to sm_35.

It's the oldest GPU architecture currently supported by all CUDA versions clang
can use.

Differential Revision: https://reviews.llvm.org/D108235


  Commit: 49d982d8cbbbb6e01b6f8e4f173ed6325beab08b
      https://github.com/llvm/llvm-project/commit/49d982d8cbbbb6e01b6f8e4f173ed6325beab08b
  Author: Artem Belevich <tra at google.com>
  Date:   2021-08-23 (Mon, 23 Aug 2021)

  Changed paths:
    M clang/include/clang/Basic/Cuda.h
    M clang/lib/Basic/Cuda.cpp
    M clang/lib/Basic/Targets/NVPTX.cpp
    M clang/lib/Driver/ToolChains/Cuda.cpp
    M llvm/lib/Target/NVPTX/NVPTX.td

  Log Message:
  -----------
  [CUDA] Add support for CUDA-11.4

Differential Revision: https://reviews.llvm.org/D108239


  Commit: 3db8e486e560183f064e31a228aada52fdeac5d6
      https://github.com/llvm/llvm-project/commit/3db8e486e560183f064e31a228aada52fdeac5d6
  Author: Artem Belevich <tra at google.com>
  Date:   2021-08-23 (Mon, 23 Aug 2021)

  Changed paths:
    M clang/include/clang/Basic/Cuda.h
    M clang/include/clang/Basic/DiagnosticDriverKinds.td
    M clang/lib/Basic/Cuda.cpp
    M clang/lib/Driver/ToolChains/Cuda.cpp
    M clang/lib/Driver/ToolChains/Cuda.h
    A clang/test/Driver/Inputs/CUDA-new/usr/local/cuda/bin/.keep
    A clang/test/Driver/Inputs/CUDA-new/usr/local/cuda/include/.keep
    A clang/test/Driver/Inputs/CUDA-new/usr/local/cuda/include/cuda.h
    A clang/test/Driver/Inputs/CUDA-new/usr/local/cuda/lib/.keep
    A clang/test/Driver/Inputs/CUDA-new/usr/local/cuda/lib64/.keep
    A clang/test/Driver/Inputs/CUDA-new/usr/local/cuda/nvvm/libdevice/libdevice.10.bc
    R clang/test/Driver/Inputs/CUDA-unknown/usr/local/cuda/version.txt
    A clang/test/Driver/Inputs/CUDA_80/usr/local/cuda/include/cuda.h
    R clang/test/Driver/Inputs/CUDA_80/usr/local/cuda/version.txt
    A clang/test/Driver/Inputs/CUDA_90/usr/local/cuda/include/cuda.h
    M clang/test/Driver/cuda-version-check.cu

  Log Message:
  -----------
  [CUDA] Improve CUDA version detection and diagnostics.

Always use cuda.h to detect CUDA version. It's a more universal approach
compared to version.txt which is no longer present in recent CUDA versions.

Split the 'unknown CUDA version' warning in two:

* when detected CUDA version is partially supported by clang. It's expected to
work in general, at the feature parity with the latest supported CUDA
version. and may be missing support for the new features/instructions/GPU
variants. Clang will issue a warning.

* when detected version is new. Recent CUDA versions have been working with
clang reasonably well, and will likely to work similarly to the partially
supported ones above. Or it may not work at all. Clang will issue a warning and
proceed as if the latest known CUDA version was detected.

Differential Revision: https://reviews.llvm.org/D108247


  Commit: ce4545db1d31f447bb42987099d691d5658da4bf
      https://github.com/llvm/llvm-project/commit/ce4545db1d31f447bb42987099d691d5658da4bf
  Author: Artem Belevich <tra at google.com>
  Date:   2021-08-23 (Mon, 23 Aug 2021)

  Changed paths:
    M clang/docs/ReleaseNotes.rst
    M clang/include/clang/Basic/Cuda.h

  Log Message:
  -----------
  [CUDA] Bump  the latest supported CUDA version to 11.4.

This should reduce the amount of noise issued by clang for the recent-ish CUDA
versions.

Clang still does not support all the features offered by NVCC, but is expected
to handle CUDA headers and produce binaries for all GPUs supported by NVCC.

Differential Revision: https://reviews.llvm.org/D108248


Compare: https://github.com/llvm/llvm-project/compare/10c982e0b3e6...ce4545db1d31