Update CMake and use native CUDA language support #62445

peterbell10 · 2021-07-29T23:55:43Z

Stack from ghstack:

PyTorch currently uses the old style of compiling CUDA in CMake which is just a
bunch of scripts in FindCUDA.cmake. Newer versions support CUDA natively as
a language just like C++ or C.

Differential Revision:

cc @malfet @seemethere

Differential Revision: D31503350

PyTorch currently uses the old style of compiling CUDA in CMake which is just a bunch of scripts in `FindCUDA.cmake`. Newer versions support CUDA natively as a language just like C++ or C. [ghstack-poisoned]

facebook-github-bot · 2021-07-29T23:55:48Z

🔗 Helpful links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/62445
📄 Preview docs built from this PR

💊 CI failures summary and remediations

As of commit e53d9dd (more details on the Dr. CI page):

30/30 failures introduced in this PR

🕵️ 29 new failures recognized by patterns

The following CI failures do not appear to be due to upstream breakages:

binary_windows_wheel_3_8_cu102_nightly_build (1/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

CMake Error at C:/w/b/windows/conda/envs/py38/L...es/CMakeDetermineCUDACompiler.cmake:187 (message):

-- Performing Test CXX_HAS_AVX512_1 - Success
-- Current compiler supports avx2 extension. Will build perfkernels.
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS - Success
-- Current compiler supports avx512f extension. Will build fbgemm.
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY - Failed
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY - Failed
-- Found CUDA: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v10.2 (found version "10.2") 
CMake Error at C:/w/b/windows/conda/envs/py38/Library/share/cmake-3.19/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
  Couldn't find CUDA library root.
Call Stack (most recent call first):
  cmake/public/cuda.cmake:41 (enable_language)
  cmake/Dependencies.cmake:43 (include)
  CMakeLists.txt:646 (include)


-- Configuring incomplete, errors occurred!
See also "C:/w/b/windows/pytorch/build/CMakeFiles/CMakeOutput.log".
See also "C:/w/b/windows/pytorch/build/CMakeFiles/CMakeError.log".

binary_windows_libtorch_3_7_cu113_release_nightly_build (2/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

CMake Error at C:/w/b/windows/conda/envs/py37/L...es/CMakeDetermineCUDACompiler.cmake:187 (message):

-- Performing Test CXX_HAS_AVX512_1 - Success
-- Current compiler supports avx2 extension. Will build perfkernels.
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS - Success
-- Current compiler supports avx512f extension. Will build fbgemm.
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY - Failed
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY - Failed
-- Found CUDA: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v11.3 (found version "11.3") 
CMake Error at C:/w/b/windows/conda/envs/py37/Library/share/cmake-3.19/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
  Couldn't find CUDA library root.
Call Stack (most recent call first):
  cmake/public/cuda.cmake:41 (enable_language)
  cmake/Dependencies.cmake:43 (include)
  CMakeLists.txt:646 (include)


-- Configuring incomplete, errors occurred!
See also "C:/w/b/windows/pytorch/build/build/CMakeFiles/CMakeOutput.log".
See also "C:/w/b/windows/pytorch/build/build/CMakeFiles/CMakeError.log".

binary_windows_conda_3_7_cu113_nightly_build (3/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

CMake Error at C:/cb/pytorch_1000000000000/_bui...es/CMakeDetermineCUDACompiler.cmake:187 (message):

%SRC_DIR%>(IF EXIST C:\tools\ruby26\bin (SET "var=!var!;C:\tools\ruby26\bin" ) ) 

%SRC_DIR%>(IF EXIST C:\PROGRA~3\nvm (SET "var=!var!;C:\PROGRA~3\nvm" ) ) 

%SRC_DIR%>(IF EXIST C:\tools\MINICO~1 (SET "var=!var!;C:\tools\MINICO~1" ) ) 

CMake Warning at CMakeLists.txt:339 (message):
  TensorPipe cannot be used on Windows.  Set it to OFF


CMake Error at C:/cb/pytorch_1000000000000/_build_env/Library/share/cmake-3.19/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
  Couldn't find CUDA library root.
Call Stack (most recent call first):
  cmake/public/cuda.cmake:41 (enable_language)
  cmake/Dependencies.cmake:43 (include)
  CMakeLists.txt:646 (include)


%SRC_DIR%>(IF EXIST C:\tools\MINICO~1\Library\mingw-w64\bin (SET "var=!var!;C:\tools\MINICO~1\Library\mingw-w64\bin" ) ) 

%SRC_DIR%>(IF EXIST C:\tools\MINICO~1\Library\usr\bin (SET "var=!var!;C:\tools\MINICO~1\Library\usr\bin" ) )

binary_windows_conda_3_7_cu111_nightly_build (4/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

CMake Error at C:/cb/pytorch_1000000000000/_bui...es/CMakeDetermineCUDACompiler.cmake:187 (message):

%SRC_DIR%>(IF EXIST C:\tools\ruby26\bin (SET "var=!var!;C:\tools\ruby26\bin" ) ) 

%SRC_DIR%>(IF EXIST C:\PROGRA~3\nvm (SET "var=!var!;C:\PROGRA~3\nvm" ) ) 

%SRC_DIR%>(IF EXIST C:\tools\MINICO~1 (SET "var=!var!;C:\tools\MINICO~1" ) ) 

CMake Warning at CMakeLists.txt:339 (message):
  TensorPipe cannot be used on Windows.  Set it to OFF


CMake Error at C:/cb/pytorch_1000000000000/_build_env/Library/share/cmake-3.19/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
  Couldn't find CUDA library root.
Call Stack (most recent call first):
  cmake/public/cuda.cmake:41 (enable_language)
  cmake/Dependencies.cmake:43 (include)
  CMakeLists.txt:646 (include)


%SRC_DIR%>(IF EXIST C:\tools\MINICO~1\Library\mingw-w64\bin (SET "var=!var!;C:\tools\MINICO~1\Library\mingw-w64\bin" ) ) 

%SRC_DIR%>(IF EXIST C:\tools\MINICO~1\Library\usr\bin (SET "var=!var!;C:\tools\MINICO~1\Library\usr\bin" ) )

binary_windows_conda_3_8_cu102_nightly_build (5/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

CMake Error at C:/cb/pytorch_1000000000000/_bui...es/CMakeDetermineCUDACompiler.cmake:187 (message):

%SRC_DIR%>(IF EXIST C:\tools\ruby26\bin (SET "var=!var!;C:\tools\ruby26\bin" ) ) 

%SRC_DIR%>(IF EXIST C:\PROGRA~3\nvm (SET "var=!var!;C:\PROGRA~3\nvm" ) ) 

%SRC_DIR%>(IF EXIST C:\tools\MINICO~1 (SET "var=!var!;C:\tools\MINICO~1" ) ) 

CMake Warning at CMakeLists.txt:339 (message):
  TensorPipe cannot be used on Windows.  Set it to OFF


CMake Error at C:/cb/pytorch_1000000000000/_build_env/Library/share/cmake-3.19/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
  Couldn't find CUDA library root.
Call Stack (most recent call first):
  cmake/public/cuda.cmake:41 (enable_language)
  cmake/Dependencies.cmake:43 (include)
  CMakeLists.txt:646 (include)


%SRC_DIR%>(IF EXIST C:\tools\MINICO~1\Library\mingw-w64\bin (SET "var=!var!;C:\tools\MINICO~1\Library\mingw-w64\bin" ) ) 

%SRC_DIR%>(IF EXIST C:\tools\MINICO~1\Library\usr\bin (SET "var=!var!;C:\tools\MINICO~1\Library\usr\bin" ) )

binary_windows_libtorch_3_7_cu102_release_nightly_build (6/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

CMake Error at C:/w/b/windows/conda/envs/py37/L...es/CMakeDetermineCUDACompiler.cmake:187 (message):

-- Performing Test CXX_HAS_AVX512_1 - Success
-- Current compiler supports avx2 extension. Will build perfkernels.
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS - Success
-- Current compiler supports avx512f extension. Will build fbgemm.
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY - Failed
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY - Failed
-- Found CUDA: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v10.2 (found version "10.2") 
CMake Error at C:/w/b/windows/conda/envs/py37/Library/share/cmake-3.19/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
  Couldn't find CUDA library root.
Call Stack (most recent call first):
  cmake/public/cuda.cmake:41 (enable_language)
  cmake/Dependencies.cmake:43 (include)
  CMakeLists.txt:646 (include)


-- Configuring incomplete, errors occurred!
See also "C:/w/b/windows/pytorch/build/build/CMakeFiles/CMakeOutput.log".
See also "C:/w/b/windows/pytorch/build/build/CMakeFiles/CMakeError.log".

binary_windows_libtorch_3_7_cu113_debug_nightly_build (7/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

CMake Error at C:/w/b/windows/conda/envs/py37/L...es/CMakeDetermineCUDACompiler.cmake:187 (message):

-- Performing Test CXX_HAS_AVX512_1 - Success
-- Current compiler supports avx2 extension. Will build perfkernels.
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS - Success
-- Current compiler supports avx512f extension. Will build fbgemm.
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY - Failed
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY - Failed
-- Found CUDA: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v11.3 (found version "11.3") 
CMake Error at C:/w/b/windows/conda/envs/py37/Library/share/cmake-3.19/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
  Couldn't find CUDA library root.
Call Stack (most recent call first):
  cmake/public/cuda.cmake:41 (enable_language)
  cmake/Dependencies.cmake:43 (include)
  CMakeLists.txt:646 (include)


-- Configuring incomplete, errors occurred!
See also "C:/w/b/windows/pytorch/build/build/CMakeFiles/CMakeOutput.log".
See also "C:/w/b/windows/pytorch/build/build/CMakeFiles/CMakeError.log".

binary_windows_conda_3_9_cu113_nightly_build (8/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

CMake Error at C:/cb/pytorch_1000000000000/_bui...es/CMakeDetermineCUDACompiler.cmake:187 (message):

%SRC_DIR%>(IF EXIST C:\tools\ruby26\bin (SET "var=!var!;C:\tools\ruby26\bin" ) ) 

%SRC_DIR%>(IF EXIST C:\PROGRA~3\nvm (SET "var=!var!;C:\PROGRA~3\nvm" ) ) 

%SRC_DIR%>(IF EXIST C:\tools\MINICO~1 (SET "var=!var!;C:\tools\MINICO~1" ) ) 

CMake Warning at CMakeLists.txt:339 (message):
  TensorPipe cannot be used on Windows.  Set it to OFF


CMake Error at C:/cb/pytorch_1000000000000/_build_env/Library/share/cmake-3.19/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
  Couldn't find CUDA library root.
Call Stack (most recent call first):
  cmake/public/cuda.cmake:41 (enable_language)
  cmake/Dependencies.cmake:43 (include)
  CMakeLists.txt:646 (include)


%SRC_DIR%>(IF EXIST C:\tools\MINICO~1\Library\mingw-w64\bin (SET "var=!var!;C:\tools\MINICO~1\Library\mingw-w64\bin" ) ) 

%SRC_DIR%>(IF EXIST C:\tools\MINICO~1\Library\usr\bin (SET "var=!var!;C:\tools\MINICO~1\Library\usr\bin" ) )

binary_windows_wheel_3_7_cu113_nightly_build (9/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

CMake Error at C:/w/b/windows/conda/envs/py37/L...es/CMakeDetermineCUDACompiler.cmake:187 (message):

-- Performing Test CXX_HAS_AVX512_1 - Success
-- Current compiler supports avx2 extension. Will build perfkernels.
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS - Success
-- Current compiler supports avx512f extension. Will build fbgemm.
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY - Failed
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY - Failed
-- Found CUDA: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v11.3 (found version "11.3") 
CMake Error at C:/w/b/windows/conda/envs/py37/Library/share/cmake-3.19/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
  Couldn't find CUDA library root.
Call Stack (most recent call first):
  cmake/public/cuda.cmake:41 (enable_language)
  cmake/Dependencies.cmake:43 (include)
  CMakeLists.txt:646 (include)


-- Configuring incomplete, errors occurred!
See also "C:/w/b/windows/pytorch/build/CMakeFiles/CMakeOutput.log".
See also "C:/w/b/windows/pytorch/build/CMakeFiles/CMakeError.log".

binary_windows_wheel_3_9_cu102_nightly_build (10/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

CMake Error at C:/w/b/windows/conda/envs/py39/L...es/CMakeDetermineCUDACompiler.cmake:187 (message):

-- Performing Test CXX_HAS_AVX512_1 - Success
-- Current compiler supports avx2 extension. Will build perfkernels.
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS - Success
-- Current compiler supports avx512f extension. Will build fbgemm.
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY - Failed
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY - Failed
-- Found CUDA: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v10.2 (found version "10.2") 
CMake Error at C:/w/b/windows/conda/envs/py39/Library/share/cmake-3.19/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
  Couldn't find CUDA library root.
Call Stack (most recent call first):
  cmake/public/cuda.cmake:41 (enable_language)
  cmake/Dependencies.cmake:43 (include)
  CMakeLists.txt:646 (include)


-- Configuring incomplete, errors occurred!
See also "C:/w/b/windows/pytorch/build/CMakeFiles/CMakeOutput.log".
See also "C:/w/b/windows/pytorch/build/CMakeFiles/CMakeError.log".

binary_windows_wheel_3_6_cu113_nightly_build (11/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

CMake Error at C:/w/b/windows/conda/envs/py36/L...es/CMakeDetermineCUDACompiler.cmake:187 (message):

-- Performing Test CXX_HAS_AVX512_1 - Success
-- Current compiler supports avx2 extension. Will build perfkernels.
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS - Success
-- Current compiler supports avx512f extension. Will build fbgemm.
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY - Failed
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY - Failed
-- Found CUDA: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v11.3 (found version "11.3") 
CMake Error at C:/w/b/windows/conda/envs/py36/Library/share/cmake-3.19/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
  Couldn't find CUDA library root.
Call Stack (most recent call first):
  cmake/public/cuda.cmake:41 (enable_language)
  cmake/Dependencies.cmake:43 (include)
  CMakeLists.txt:646 (include)


-- Configuring incomplete, errors occurred!
See also "C:/w/b/windows/pytorch/build/CMakeFiles/CMakeOutput.log".
See also "C:/w/b/windows/pytorch/build/CMakeFiles/CMakeError.log".

binary_windows_wheel_3_6_cu102_nightly_build (12/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

CMake Error at C:/w/b/windows/conda/envs/py36/L...es/CMakeDetermineCUDACompiler.cmake:187 (message):

-- Performing Test CXX_HAS_AVX512_1 - Success
-- Current compiler supports avx2 extension. Will build perfkernels.
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS - Success
-- Current compiler supports avx512f extension. Will build fbgemm.
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY - Failed
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY - Failed
-- Found CUDA: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v10.2 (found version "10.2") 
CMake Error at C:/w/b/windows/conda/envs/py36/Library/share/cmake-3.19/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
  Couldn't find CUDA library root.
Call Stack (most recent call first):
  cmake/public/cuda.cmake:41 (enable_language)
  cmake/Dependencies.cmake:43 (include)
  CMakeLists.txt:646 (include)


-- Configuring incomplete, errors occurred!
See also "C:/w/b/windows/pytorch/build/CMakeFiles/CMakeOutput.log".
See also "C:/w/b/windows/pytorch/build/CMakeFiles/CMakeError.log".

binary_windows_wheel_3_8_cu113_nightly_build (13/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

CMake Error at C:/w/b/windows/conda/envs/py38/L...es/CMakeDetermineCUDACompiler.cmake:187 (message):

-- Performing Test CXX_HAS_AVX512_1 - Success
-- Current compiler supports avx2 extension. Will build perfkernels.
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS - Success
-- Current compiler supports avx512f extension. Will build fbgemm.
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY - Failed
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY - Failed
-- Found CUDA: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v11.3 (found version "11.3") 
CMake Error at C:/w/b/windows/conda/envs/py38/Library/share/cmake-3.19/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
  Couldn't find CUDA library root.
Call Stack (most recent call first):
  cmake/public/cuda.cmake:41 (enable_language)
  cmake/Dependencies.cmake:43 (include)
  CMakeLists.txt:646 (include)


-- Configuring incomplete, errors occurred!
See also "C:/w/b/windows/pytorch/build/CMakeFiles/CMakeOutput.log".
See also "C:/w/b/windows/pytorch/build/CMakeFiles/CMakeError.log".

binary_windows_wheel_3_9_cu113_nightly_build (14/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

CMake Error at C:/w/b/windows/conda/envs/py39/L...es/CMakeDetermineCUDACompiler.cmake:187 (message):

-- Performing Test CXX_HAS_AVX512_1 - Success
-- Current compiler supports avx2 extension. Will build perfkernels.
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS - Success
-- Current compiler supports avx512f extension. Will build fbgemm.
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY - Failed
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY - Failed
-- Found CUDA: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v11.3 (found version "11.3") 
CMake Error at C:/w/b/windows/conda/envs/py39/Library/share/cmake-3.19/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
  Couldn't find CUDA library root.
Call Stack (most recent call first):
  cmake/public/cuda.cmake:41 (enable_language)
  cmake/Dependencies.cmake:43 (include)
  CMakeLists.txt:646 (include)


-- Configuring incomplete, errors occurred!
See also "C:/w/b/windows/pytorch/build/CMakeFiles/CMakeOutput.log".
See also "C:/w/b/windows/pytorch/build/CMakeFiles/CMakeError.log".

binary_windows_conda_3_9_cu111_nightly_build (15/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

CMake Error at C:/cb/pytorch_1000000000000/_bui...es/CMakeDetermineCUDACompiler.cmake:187 (message):

%SRC_DIR%>(IF EXIST C:\tools\ruby26\bin (SET "var=!var!;C:\tools\ruby26\bin" ) ) 

%SRC_DIR%>(IF EXIST C:\PROGRA~3\nvm (SET "var=!var!;C:\PROGRA~3\nvm" ) ) 

%SRC_DIR%>(IF EXIST C:\tools\MINICO~1 (SET "var=!var!;C:\tools\MINICO~1" ) ) 

CMake Warning at CMakeLists.txt:339 (message):
  TensorPipe cannot be used on Windows.  Set it to OFF


CMake Error at C:/cb/pytorch_1000000000000/_build_env/Library/share/cmake-3.19/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
  Couldn't find CUDA library root.
Call Stack (most recent call first):
  cmake/public/cuda.cmake:41 (enable_language)
  cmake/Dependencies.cmake:43 (include)
  CMakeLists.txt:646 (include)


%SRC_DIR%>(IF EXIST C:\tools\MINICO~1\Library\mingw-w64\bin (SET "var=!var!;C:\tools\MINICO~1\Library\mingw-w64\bin" ) ) 

%SRC_DIR%>(IF EXIST C:\tools\MINICO~1\Library\usr\bin (SET "var=!var!;C:\tools\MINICO~1\Library\usr\bin" ) )

binary_windows_libtorch_3_7_cu111_release_nightly_build (16/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

CMake Error at C:/w/b/windows/conda/envs/py37/L...es/CMakeDetermineCUDACompiler.cmake:187 (message):

-- Performing Test CXX_HAS_AVX512_1 - Success
-- Current compiler supports avx2 extension. Will build perfkernels.
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS - Success
-- Current compiler supports avx512f extension. Will build fbgemm.
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY - Failed
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY - Failed
-- Found CUDA: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v11.1 (found version "11.1") 
CMake Error at C:/w/b/windows/conda/envs/py37/Library/share/cmake-3.19/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
  Couldn't find CUDA library root.
Call Stack (most recent call first):
  cmake/public/cuda.cmake:41 (enable_language)
  cmake/Dependencies.cmake:43 (include)
  CMakeLists.txt:646 (include)


-- Configuring incomplete, errors occurred!
See also "C:/w/b/windows/pytorch/build/build/CMakeFiles/CMakeOutput.log".
See also "C:/w/b/windows/pytorch/build/build/CMakeFiles/CMakeError.log".

binary_windows_conda_3_9_cu102_nightly_build (17/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

CMake Error at C:/cb/pytorch_1000000000000/_bui...es/CMakeDetermineCUDACompiler.cmake:187 (message):

%SRC_DIR%>(IF EXIST C:\tools\ruby26\bin (SET "var=!var!;C:\tools\ruby26\bin" ) ) 

%SRC_DIR%>(IF EXIST C:\PROGRA~3\nvm (SET "var=!var!;C:\PROGRA~3\nvm" ) ) 

%SRC_DIR%>(IF EXIST C:\tools\MINICO~1 (SET "var=!var!;C:\tools\MINICO~1" ) ) 

CMake Warning at CMakeLists.txt:339 (message):
  TensorPipe cannot be used on Windows.  Set it to OFF


CMake Error at C:/cb/pytorch_1000000000000/_build_env/Library/share/cmake-3.19/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
  Couldn't find CUDA library root.
Call Stack (most recent call first):
  cmake/public/cuda.cmake:41 (enable_language)
  cmake/Dependencies.cmake:43 (include)
  CMakeLists.txt:646 (include)


%SRC_DIR%>(IF EXIST C:\tools\MINICO~1\Library\mingw-w64\bin (SET "var=!var!;C:\tools\MINICO~1\Library\mingw-w64\bin" ) ) 

%SRC_DIR%>(IF EXIST C:\tools\MINICO~1\Library\usr\bin (SET "var=!var!;C:\tools\MINICO~1\Library\usr\bin" ) )

binary_windows_wheel_3_6_cu111_nightly_build (18/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

CMake Error at C:/w/b/windows/conda/envs/py36/L...es/CMakeDetermineCUDACompiler.cmake:187 (message):

-- Performing Test CXX_HAS_AVX512_1 - Success
-- Current compiler supports avx2 extension. Will build perfkernels.
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS - Success
-- Current compiler supports avx512f extension. Will build fbgemm.
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY - Failed
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY - Failed
-- Found CUDA: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v11.1 (found version "11.1") 
CMake Error at C:/w/b/windows/conda/envs/py36/Library/share/cmake-3.19/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
  Couldn't find CUDA library root.
Call Stack (most recent call first):
  cmake/public/cuda.cmake:41 (enable_language)
  cmake/Dependencies.cmake:43 (include)
  CMakeLists.txt:646 (include)


-- Configuring incomplete, errors occurred!
See also "C:/w/b/windows/pytorch/build/CMakeFiles/CMakeOutput.log".
See also "C:/w/b/windows/pytorch/build/CMakeFiles/CMakeError.log".

binary_windows_wheel_3_7_cu102_nightly_build (19/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

CMake Error at C:/w/b/windows/conda/envs/py37/L...es/CMakeDetermineCUDACompiler.cmake:187 (message):

-- Performing Test CXX_HAS_AVX512_1 - Success
-- Current compiler supports avx2 extension. Will build perfkernels.
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS - Success
-- Current compiler supports avx512f extension. Will build fbgemm.
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY - Failed
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY - Failed
-- Found CUDA: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v10.2 (found version "10.2") 
CMake Error at C:/w/b/windows/conda/envs/py37/Library/share/cmake-3.19/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
  Couldn't find CUDA library root.
Call Stack (most recent call first):
  cmake/public/cuda.cmake:41 (enable_language)
  cmake/Dependencies.cmake:43 (include)
  CMakeLists.txt:646 (include)


-- Configuring incomplete, errors occurred!
See also "C:/w/b/windows/pytorch/build/CMakeFiles/CMakeOutput.log".
See also "C:/w/b/windows/pytorch/build/CMakeFiles/CMakeError.log".

binary_windows_wheel_3_9_cu111_nightly_build (20/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

CMake Error at C:/w/b/windows/conda/envs/py39/L...es/CMakeDetermineCUDACompiler.cmake:187 (message):

-- Performing Test CXX_HAS_AVX512_1 - Success
-- Current compiler supports avx2 extension. Will build perfkernels.
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS - Success
-- Current compiler supports avx512f extension. Will build fbgemm.
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY - Failed
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY - Failed
-- Found CUDA: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v11.1 (found version "11.1") 
CMake Error at C:/w/b/windows/conda/envs/py39/Library/share/cmake-3.19/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
  Couldn't find CUDA library root.
Call Stack (most recent call first):
  cmake/public/cuda.cmake:41 (enable_language)
  cmake/Dependencies.cmake:43 (include)
  CMakeLists.txt:646 (include)


-- Configuring incomplete, errors occurred!
See also "C:/w/b/windows/pytorch/build/CMakeFiles/CMakeOutput.log".
See also "C:/w/b/windows/pytorch/build/CMakeFiles/CMakeError.log".

binary_windows_wheel_3_8_cu111_nightly_build (21/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

CMake Error at C:/w/b/windows/conda/envs/py38/L...es/CMakeDetermineCUDACompiler.cmake:187 (message):

-- Performing Test CXX_HAS_AVX512_1 - Success
-- Current compiler supports avx2 extension. Will build perfkernels.
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS - Success
-- Current compiler supports avx512f extension. Will build fbgemm.
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY - Failed
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY - Failed
-- Found CUDA: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v11.1 (found version "11.1") 
CMake Error at C:/w/b/windows/conda/envs/py38/Library/share/cmake-3.19/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
  Couldn't find CUDA library root.
Call Stack (most recent call first):
  cmake/public/cuda.cmake:41 (enable_language)
  cmake/Dependencies.cmake:43 (include)
  CMakeLists.txt:646 (include)


-- Configuring incomplete, errors occurred!
See also "C:/w/b/windows/pytorch/build/CMakeFiles/CMakeOutput.log".
See also "C:/w/b/windows/pytorch/build/CMakeFiles/CMakeError.log".

binary_windows_wheel_3_7_cu111_nightly_build (22/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

CMake Error at C:/w/b/windows/conda/envs/py37/L...es/CMakeDetermineCUDACompiler.cmake:187 (message):

-- Performing Test CXX_HAS_AVX512_1 - Success
-- Current compiler supports avx2 extension. Will build perfkernels.
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS - Success
-- Current compiler supports avx512f extension. Will build fbgemm.
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY - Failed
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY - Failed
-- Found CUDA: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v11.1 (found version "11.1") 
CMake Error at C:/w/b/windows/conda/envs/py37/Library/share/cmake-3.19/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
  Couldn't find CUDA library root.
Call Stack (most recent call first):
  cmake/public/cuda.cmake:41 (enable_language)
  cmake/Dependencies.cmake:43 (include)
  CMakeLists.txt:646 (include)


-- Configuring incomplete, errors occurred!
See also "C:/w/b/windows/pytorch/build/CMakeFiles/CMakeOutput.log".
See also "C:/w/b/windows/pytorch/build/CMakeFiles/CMakeError.log".

binary_windows_conda_3_8_cu113_nightly_build (23/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

CMake Error at C:/cb/pytorch_1000000000000/_bui...es/CMakeDetermineCUDACompiler.cmake:187 (message):

%SRC_DIR%>(IF EXIST C:\tools\ruby26\bin (SET "var=!var!;C:\tools\ruby26\bin" ) ) 

%SRC_DIR%>(IF EXIST C:\PROGRA~3\nvm (SET "var=!var!;C:\PROGRA~3\nvm" ) ) 

%SRC_DIR%>(IF EXIST C:\tools\MINICO~1 (SET "var=!var!;C:\tools\MINICO~1" ) ) 

CMake Warning at CMakeLists.txt:339 (message):
  TensorPipe cannot be used on Windows.  Set it to OFF


CMake Error at C:/cb/pytorch_1000000000000/_build_env/Library/share/cmake-3.19/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
  Couldn't find CUDA library root.
Call Stack (most recent call first):
  cmake/public/cuda.cmake:41 (enable_language)
  cmake/Dependencies.cmake:43 (include)
  CMakeLists.txt:646 (include)


%SRC_DIR%>(IF EXIST C:\tools\MINICO~1\Library\mingw-w64\bin (SET "var=!var!;C:\tools\MINICO~1\Library\mingw-w64\bin" ) ) 

%SRC_DIR%>(IF EXIST C:\tools\MINICO~1\Library\usr\bin (SET "var=!var!;C:\tools\MINICO~1\Library\usr\bin" ) )

binary_windows_conda_3_7_cu102_nightly_build (24/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

CMake Error at C:/cb/pytorch_1000000000000/_bui...es/CMakeDetermineCUDACompiler.cmake:187 (message):

%SRC_DIR%>(IF EXIST C:\tools\ruby26\bin (SET "var=!var!;C:\tools\ruby26\bin" ) ) 

%SRC_DIR%>(IF EXIST C:\PROGRA~3\nvm (SET "var=!var!;C:\PROGRA~3\nvm" ) ) 

%SRC_DIR%>(IF EXIST C:\tools\MINICO~1 (SET "var=!var!;C:\tools\MINICO~1" ) ) 

CMake Warning at CMakeLists.txt:339 (message):
  TensorPipe cannot be used on Windows.  Set it to OFF


CMake Error at C:/cb/pytorch_1000000000000/_build_env/Library/share/cmake-3.19/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
  Couldn't find CUDA library root.
Call Stack (most recent call first):
  cmake/public/cuda.cmake:41 (enable_language)
  cmake/Dependencies.cmake:43 (include)
  CMakeLists.txt:646 (include)


%SRC_DIR%>(IF EXIST C:\tools\MINICO~1\Library\mingw-w64\bin (SET "var=!var!;C:\tools\MINICO~1\Library\mingw-w64\bin" ) ) 

%SRC_DIR%>(IF EXIST C:\tools\MINICO~1\Library\usr\bin (SET "var=!var!;C:\tools\MINICO~1\Library\usr\bin" ) )

binary_windows_conda_3_6_cu113_nightly_build (25/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

CMake Error at C:/cb/pytorch_1000000000000/_bui...es/CMakeDetermineCUDACompiler.cmake:187 (message):

%SRC_DIR%>(IF EXIST C:\tools\ruby26\bin (SET "var=!var!;C:\tools\ruby26\bin" ) ) 

%SRC_DIR%>(IF EXIST C:\PROGRA~3\nvm (SET "var=!var!;C:\PROGRA~3\nvm" ) ) 

%SRC_DIR%>(IF EXIST C:\tools\MINICO~1 (SET "var=!var!;C:\tools\MINICO~1" ) ) 

CMake Warning at CMakeLists.txt:339 (message):
  TensorPipe cannot be used on Windows.  Set it to OFF


CMake Error at C:/cb/pytorch_1000000000000/_build_env/Library/share/cmake-3.19/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
  Couldn't find CUDA library root.
Call Stack (most recent call first):
  cmake/public/cuda.cmake:41 (enable_language)
  cmake/Dependencies.cmake:43 (include)
  CMakeLists.txt:646 (include)


%SRC_DIR%>(IF EXIST C:\tools\MINICO~1\Library\mingw-w64\bin (SET "var=!var!;C:\tools\MINICO~1\Library\mingw-w64\bin" ) ) 

%SRC_DIR%>(IF EXIST C:\tools\MINICO~1\Library\usr\bin (SET "var=!var!;C:\tools\MINICO~1\Library\usr\bin" ) )

binary_windows_libtorch_3_7_cu111_debug_nightly_build (26/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

CMake Error at C:/w/b/windows/conda/envs/py37/L...es/CMakeDetermineCUDACompiler.cmake:187 (message):

-- Performing Test CXX_HAS_AVX512_1 - Success
-- Current compiler supports avx2 extension. Will build perfkernels.
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS - Success
-- Current compiler supports avx512f extension. Will build fbgemm.
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY - Failed
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY - Failed
-- Found CUDA: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v11.1 (found version "11.1") 
CMake Error at C:/w/b/windows/conda/envs/py37/Library/share/cmake-3.19/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
  Couldn't find CUDA library root.
Call Stack (most recent call first):
  cmake/public/cuda.cmake:41 (enable_language)
  cmake/Dependencies.cmake:43 (include)
  CMakeLists.txt:646 (include)


-- Configuring incomplete, errors occurred!
See also "C:/w/b/windows/pytorch/build/build/CMakeFiles/CMakeOutput.log".
See also "C:/w/b/windows/pytorch/build/build/CMakeFiles/CMakeError.log".

binary_windows_conda_3_8_cu111_nightly_build (27/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

CMake Error at C:/cb/pytorch_1000000000000/_bui...es/CMakeDetermineCUDACompiler.cmake:187 (message):

%SRC_DIR%>(IF EXIST C:\tools\ruby26\bin (SET "var=!var!;C:\tools\ruby26\bin" ) ) 

%SRC_DIR%>(IF EXIST C:\PROGRA~3\nvm (SET "var=!var!;C:\PROGRA~3\nvm" ) ) 

%SRC_DIR%>(IF EXIST C:\tools\MINICO~1 (SET "var=!var!;C:\tools\MINICO~1" ) ) 

CMake Warning at CMakeLists.txt:339 (message):
  TensorPipe cannot be used on Windows.  Set it to OFF


CMake Error at C:/cb/pytorch_1000000000000/_build_env/Library/share/cmake-3.19/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
  Couldn't find CUDA library root.
Call Stack (most recent call first):
  cmake/public/cuda.cmake:41 (enable_language)
  cmake/Dependencies.cmake:43 (include)
  CMakeLists.txt:646 (include)


%SRC_DIR%>(IF EXIST C:\tools\MINICO~1\Library\mingw-w64\bin (SET "var=!var!;C:\tools\MINICO~1\Library\mingw-w64\bin" ) ) 

%SRC_DIR%>(IF EXIST C:\tools\MINICO~1\Library\usr\bin (SET "var=!var!;C:\tools\MINICO~1\Library\usr\bin" ) )

binary_windows_conda_3_6_cu102_nightly_build (28/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

CMake Error at C:/cb/pytorch_1000000000000/_bui...es/CMakeDetermineCUDACompiler.cmake:187 (message):

%SRC_DIR%>(IF EXIST C:\tools\ruby26\bin (SET "var=!var!;C:\tools\ruby26\bin" ) ) 

%SRC_DIR%>(IF EXIST C:\PROGRA~3\nvm (SET "var=!var!;C:\PROGRA~3\nvm" ) ) 

%SRC_DIR%>(IF EXIST C:\tools\MINICO~1 (SET "var=!var!;C:\tools\MINICO~1" ) ) 

CMake Warning at CMakeLists.txt:339 (message):
  TensorPipe cannot be used on Windows.  Set it to OFF


CMake Error at C:/cb/pytorch_1000000000000/_build_env/Library/share/cmake-3.19/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
  Couldn't find CUDA library root.
Call Stack (most recent call first):
  cmake/public/cuda.cmake:41 (enable_language)
  cmake/Dependencies.cmake:43 (include)
  CMakeLists.txt:646 (include)


%SRC_DIR%>(IF EXIST C:\tools\MINICO~1\Library\mingw-w64\bin (SET "var=!var!;C:\tools\MINICO~1\Library\mingw-w64\bin" ) ) 

%SRC_DIR%>(IF EXIST C:\tools\MINICO~1\Library\usr\bin (SET "var=!var!;C:\tools\MINICO~1\Library\usr\bin" ) )

binary_windows_libtorch_3_7_cu102_debug_nightly_build (29/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

CMake Error at C:/w/b/windows/conda/envs/py37/L...es/CMakeDetermineCUDACompiler.cmake:187 (message):

-- Performing Test CXX_HAS_AVX512_1 - Success
-- Current compiler supports avx2 extension. Will build perfkernels.
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS
-- Performing Test CAFFE2_COMPILER_SUPPORTS_AVX512_EXTENSIONS - Success
-- Current compiler supports avx512f extension. Will build fbgemm.
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_VISIBILITY - Failed
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY
-- Performing Test COMPILER_SUPPORTS_HIDDEN_INLINE_VISIBILITY - Failed
-- Found CUDA: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v10.2 (found version "10.2") 
CMake Error at C:/w/b/windows/conda/envs/py37/Library/share/cmake-3.19/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
  Couldn't find CUDA library root.
Call Stack (most recent call first):
  cmake/public/cuda.cmake:41 (enable_language)
  cmake/Dependencies.cmake:43 (include)
  CMakeLists.txt:646 (include)


-- Configuring incomplete, errors occurred!
See also "C:/w/b/windows/pytorch/build/build/CMakeFiles/CMakeOutput.log".
See also "C:/w/b/windows/pytorch/build/build/CMakeFiles/CMakeError.log".

1 failure not recognized by patterns:

Job	Step	Action
^{binary_windows_conda_3_6_cu111_nightly_build}	^Unknown	🔁 rerun

This comment was automatically generated by Dr. CI (expand for details).

Follow this link to opt-out of these comments for your Pull Requests.

Please report bugs/suggestions to the (internal) Dr. CI Users group.

Click here to manually regenerate this comment.

PyTorch currently uses the old style of compiling CUDA in CMake which is just a bunch of scripts in `FindCUDA.cmake`. Newer versions support CUDA natively as a language just like C++ or C. ghstack-source-id: b88b94a Pull Request resolved: #62445

PyTorch currently uses the old style of compiling CUDA in CMake which is just a bunch of scripts in `FindCUDA.cmake`. Newer versions support CUDA natively as a language just like C++ or C. [ghstack-poisoned]

PyTorch currently uses the old style of compiling CUDA in CMake which is just a bunch of scripts in `FindCUDA.cmake`. Newer versions support CUDA natively as a language just like C++ or C. ghstack-source-id: 5cbca68 Pull Request resolved: #62445

peterbell10 · 2021-07-30T14:57:55Z

@malfet do you know if requiring a newer CMake version is okay? 3.12 was released in 2018 so it's 3 years old at this point.

One major benefit from the new style is CMake understands the dependencies far better. Previously, all .cu files waited for libtorch_cpu to link before starting to compile, and then all the cpp files in libtorch_cuda and libtorch_python had to wait for all .cu files to finish compiling. That creates slow build times where cores are sat idle:

Compare with a trace with the changes from this PR, where the build time is down from 14 minutes to 10 minutes:

malfet · 2021-07-30T17:27:16Z

@peterbell10 this sounds great, but moving to 3.12 is a bit challenging, as cmake-3.10 is the only version available in ubuntu-18.04 How hard would it be to have ifdefs for old vs new cmake versions?

CMakeLists.txt

malfet · 2021-07-30T17:33:04Z

When PR is finalized, please add an explanation why you are replacing list(ADD with string(APPEND, as this change was previously done to support compilation in folders with spaces
Also, some explanation would be good why all those extra -Wno-this-or-that are added to the code?

peterbell10 · 2021-07-30T18:07:11Z

cmake/public/utils.cmake

        -Wall
        -Wextra
        -Wno-unused-parameter
+        -Wno-unused-variable


Target compile flags are propagated to all languages by default, so we now have stricter warning settings on CUDA code. These -Wno- settings are added to CMAKE_CXX_FLAGS in the root CMakeLists.txt which of course isn't propagated to CUDA.

pytorch/CMakeLists.txt

Lines 690 to 716 in 73ba166

string(APPEND CMAKE_CXX_FLAGS " -O2 -fPIC")

string(APPEND CMAKE_CXX_FLAGS " -Wno-narrowing")

# Eigen fails to build with some versions, so convert this to a warning

# Details at http://eigen.tuxfamily.org/bz/show_bug.cgi?id=1459

string(APPEND CMAKE_CXX_FLAGS " -Wall")

string(APPEND CMAKE_CXX_FLAGS " -Wextra")

string(APPEND CMAKE_CXX_FLAGS " -Werror=return-type")

string(APPEND CMAKE_CXX_FLAGS " -Wno-missing-field-initializers")

string(APPEND CMAKE_CXX_FLAGS " -Wno-type-limits")

string(APPEND CMAKE_CXX_FLAGS " -Wno-array-bounds")

string(APPEND CMAKE_CXX_FLAGS " -Wno-unknown-pragmas")

string(APPEND CMAKE_CXX_FLAGS " -Wno-sign-compare")

string(APPEND CMAKE_CXX_FLAGS " -Wno-unused-parameter")

string(APPEND CMAKE_CXX_FLAGS " -Wno-unused-variable")

string(APPEND CMAKE_CXX_FLAGS " -Wno-unused-function")

string(APPEND CMAKE_CXX_FLAGS " -Wno-unused-result")

string(APPEND CMAKE_CXX_FLAGS " -Wno-unused-local-typedefs")

string(APPEND CMAKE_CXX_FLAGS " -Wno-strict-overflow")

string(APPEND CMAKE_CXX_FLAGS " -Wno-strict-aliasing")

string(APPEND CMAKE_CXX_FLAGS " -Wno-error=deprecated-declarations")

if(CMAKE_COMPILER_IS_GNUCXX AND NOT (CMAKE_CXX_COMPILER_VERSION VERSION_LESS 7.0.0))

string(APPEND CMAKE_CXX_FLAGS " -Wno-stringop-overflow")

endif()

if(CMAKE_COMPILER_IS_GNUCXX)

# Suppress "The ABI for passing parameters with 64-byte alignment has changed in GCC 4.6"

string(APPEND CMAKE_CXX_FLAGS " -Wno-psabi")

endif()

peterbell10 · 2021-07-30T18:07:47Z

moving to 3.12 is a bit challenging, as cmake-3.10 is the only version available in ubuntu-18.04

3.10 should be possible. I'll have a go later today.

PyTorch currently uses the old style of compiling CUDA in CMake which is just a bunch of scripts in `FindCUDA.cmake`. Newer versions support CUDA natively as a language just like C++ or C. [ghstack-poisoned]

PyTorch currently uses the old style of compiling CUDA in CMake which is just a bunch of scripts in `FindCUDA.cmake`. Newer versions support CUDA natively as a language just like C++ or C. Differential Revision: [](https://our.internmc.facebook.com/intern/diff/) cc malfet seemethere [ghstack-poisoned]

…se native CUDA language support" PyTorch currently uses the old style of compiling CUDA in CMake which is just a bunch of scripts in `FindCUDA.cmake`. Newer versions support CUDA natively as a language just like C++ or C. Differential Revision: [](https://our.internmc.facebook.com/intern/diff/) cc malfet seemethere [ghstack-poisoned]

peterbell10 · 2021-09-30T14:40:28Z

@malfet any updates?

PyTorch currently uses the old style of compiling CUDA in CMake which is just a bunch of scripts in `FindCUDA.cmake`. Newer versions support CUDA natively as a language just like C++ or C. Differential Revision: [](https://our.internmc.facebook.com/intern/diff/) cc malfet seemethere [ghstack-poisoned]

peterbell10 · 2021-10-08T13:41:11Z

@malfet PTAL. Rebased on master and the docker builds are succeeding again. The only test failure is the periodic build that's also failing with the same error in #66258 so not related.

malfet · 2021-10-08T13:47:07Z

@malfet has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

malfet · 2021-10-11T23:31:16Z

I don't know what am I doing wrong with the tool, but it stripped authorship attribution from this PR when it got landed :(

peterbell10 · 2021-10-12T00:35:59Z

@ezyang any ideas?

ezyang · 2021-10-12T14:03:52Z

The authorship never showed up on the initial import. @malfet if you can pull out the rage for that invocation that would be a big help.

Separately, we can add a note in the release notes reattributing this commit properly, it's the least we can do. Who is RM for 1.10?

WIP: Update CMake and use native CUDA language support

eed3060

PyTorch currently uses the old style of compiling CUDA in CMake which is just a bunch of scripts in `FindCUDA.cmake`. Newer versions support CUDA natively as a language just like C++ or C. [ghstack-poisoned]

facebook-github-bot added the cla signed label Jul 29, 2021

pytorchbot added the open source label Jul 30, 2021

Update on "WIP: Update CMake and use native CUDA language support"

c997020

PyTorch currently uses the old style of compiling CUDA in CMake which is just a bunch of scripts in `FindCUDA.cmake`. Newer versions support CUDA natively as a language just like C++ or C. [ghstack-poisoned]

Update on "WIP: Update CMake and use native CUDA language support"

109ba8b

PyTorch currently uses the old style of compiling CUDA in CMake which is just a bunch of scripts in `FindCUDA.cmake`. Newer versions support CUDA natively as a language just like C++ or C. [ghstack-poisoned]

peterbell10 requested a review from malfet July 30, 2021 15:20

peterbell10 added the module: build Build system issues label Jul 30, 2021

malfet reviewed Jul 30, 2021

View reviewed changes

CMakeLists.txt Show resolved Hide resolved

peterbell10 commented Jul 30, 2021

View reviewed changes

Update on "WIP: Update CMake and use native CUDA language support"

0691759

PyTorch currently uses the old style of compiling CUDA in CMake which is just a bunch of scripts in `FindCUDA.cmake`. Newer versions support CUDA natively as a language just like C++ or C. [ghstack-poisoned]

peterbell10 requested review from jeffdaily and jithunnair-amd as code owners July 30, 2021 19:33

Update on "WIP: Update CMake and use native CUDA language support"

75e3285

PyTorch currently uses the old style of compiling CUDA in CMake which is just a bunch of scripts in `FindCUDA.cmake`. Newer versions support CUDA natively as a language just like C++ or C. [ghstack-poisoned]

Update on "WIP: Update CMake and use native CUDA language support"

20d8f02

PyTorch currently uses the old style of compiling CUDA in CMake which is just a bunch of scripts in `FindCUDA.cmake`. Newer versions support CUDA natively as a language just like C++ or C. [ghstack-poisoned]

Update on "WIP: Update CMake and use native CUDA language support"

007f607

PyTorch currently uses the old style of compiling CUDA in CMake which is just a bunch of scripts in `FindCUDA.cmake`. Newer versions support CUDA natively as a language just like C++ or C. [ghstack-poisoned]

Update on "WIP: Update CMake and use native CUDA language support"

5dabecf

PyTorch currently uses the old style of compiling CUDA in CMake which is just a bunch of scripts in `FindCUDA.cmake`. Newer versions support CUDA natively as a language just like C++ or C. [ghstack-poisoned]

peterbell10 added 3 commits September 23, 2021 16:07

peterbell10 mentioned this pull request Sep 23, 2021

Sparse SoftMax: Remove unused variables #65539

Closed

peterbell10 added 3 commits October 6, 2021 18:33

peterbell10 mentioned this pull request Oct 7, 2021

Docker: Fix valgrind install #66262

Closed

peterbell10 added 2 commits October 7, 2021 21:51

malfet approved these changes Oct 8, 2021

View reviewed changes

facebook-github-bot closed this in c373387 Oct 11, 2021

facebook-github-bot deleted the gh/peterbell10/108/head branch October 15, 2021 14:24

mszhanyi mentioned this pull request Nov 29, 2021

CUDA sources are not cached with sccache #37928

Closed

lanking520 mentioned this pull request Apr 19, 2022

PyTorch on Windows GPU compile failure: [No CUDA toolset found] #76010

Closed

ngimel mentioned this pull request Apr 19, 2022

Profiler SegFault on a single Linear layer #75369

Closed

peterbell10 mentioned this pull request Apr 20, 2022

Eliminate uses of deprecated FindCUDA.cmake #76082

Open

3 tasks

bhack mentioned this pull request Sep 23, 2024

Tensor.nonzero fails on GPU for tensors containing more than INT_MAX elements #51871

Closed

	string(APPEND CMAKE_CXX_FLAGS " -O2 -fPIC")
	string(APPEND CMAKE_CXX_FLAGS " -Wno-narrowing")
	# Eigen fails to build with some versions, so convert this to a warning
	# Details at http://eigen.tuxfamily.org/bz/show_bug.cgi?id=1459
	string(APPEND CMAKE_CXX_FLAGS " -Wall")
	string(APPEND CMAKE_CXX_FLAGS " -Wextra")
	string(APPEND CMAKE_CXX_FLAGS " -Werror=return-type")
	string(APPEND CMAKE_CXX_FLAGS " -Wno-missing-field-initializers")
	string(APPEND CMAKE_CXX_FLAGS " -Wno-type-limits")
	string(APPEND CMAKE_CXX_FLAGS " -Wno-array-bounds")
	string(APPEND CMAKE_CXX_FLAGS " -Wno-unknown-pragmas")
	string(APPEND CMAKE_CXX_FLAGS " -Wno-sign-compare")
	string(APPEND CMAKE_CXX_FLAGS " -Wno-unused-parameter")
	string(APPEND CMAKE_CXX_FLAGS " -Wno-unused-variable")
	string(APPEND CMAKE_CXX_FLAGS " -Wno-unused-function")
	string(APPEND CMAKE_CXX_FLAGS " -Wno-unused-result")
	string(APPEND CMAKE_CXX_FLAGS " -Wno-unused-local-typedefs")
	string(APPEND CMAKE_CXX_FLAGS " -Wno-strict-overflow")
	string(APPEND CMAKE_CXX_FLAGS " -Wno-strict-aliasing")
	string(APPEND CMAKE_CXX_FLAGS " -Wno-error=deprecated-declarations")
	if(CMAKE_COMPILER_IS_GNUCXX AND NOT (CMAKE_CXX_COMPILER_VERSION VERSION_LESS 7.0.0))
	string(APPEND CMAKE_CXX_FLAGS " -Wno-stringop-overflow")
	endif()
	if(CMAKE_COMPILER_IS_GNUCXX)
	# Suppress "The ABI for passing parameters with 64-byte alignment has changed in GCC 4.6"
	string(APPEND CMAKE_CXX_FLAGS " -Wno-psabi")
	endif()

Update CMake and use native CUDA language support #62445

Update CMake and use native CUDA language support #62445

Uh oh!

Conversation

peterbell10 commented Jul 29, 2021 • edited by malfet Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

facebook-github-bot commented Jul 29, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful links

💊 CI failures summary and remediations

🕵️ 29 new failures recognized by patterns

binary_windows_wheel_3_8_cu102_nightly_build (1/29)

binary_windows_libtorch_3_7_cu113_release_nightly_build (2/29)

binary_windows_conda_3_7_cu113_nightly_build (3/29)

binary_windows_conda_3_7_cu111_nightly_build (4/29)

binary_windows_conda_3_8_cu102_nightly_build (5/29)

binary_windows_libtorch_3_7_cu102_release_nightly_build (6/29)

binary_windows_libtorch_3_7_cu113_debug_nightly_build (7/29)

binary_windows_conda_3_9_cu113_nightly_build (8/29)

binary_windows_wheel_3_7_cu113_nightly_build (9/29)

binary_windows_wheel_3_9_cu102_nightly_build (10/29)

binary_windows_wheel_3_6_cu113_nightly_build (11/29)

binary_windows_wheel_3_6_cu102_nightly_build (12/29)

binary_windows_wheel_3_8_cu113_nightly_build (13/29)

binary_windows_wheel_3_9_cu113_nightly_build (14/29)

binary_windows_conda_3_9_cu111_nightly_build (15/29)

binary_windows_libtorch_3_7_cu111_release_nightly_build (16/29)

binary_windows_conda_3_9_cu102_nightly_build (17/29)

binary_windows_wheel_3_6_cu111_nightly_build (18/29)

binary_windows_wheel_3_7_cu102_nightly_build (19/29)

binary_windows_wheel_3_9_cu111_nightly_build (20/29)

binary_windows_wheel_3_8_cu111_nightly_build (21/29)

binary_windows_wheel_3_7_cu111_nightly_build (22/29)

binary_windows_conda_3_8_cu113_nightly_build (23/29)

binary_windows_conda_3_7_cu102_nightly_build (24/29)

binary_windows_conda_3_6_cu113_nightly_build (25/29)

binary_windows_libtorch_3_7_cu111_debug_nightly_build (26/29)

binary_windows_conda_3_8_cu111_nightly_build (27/29)

binary_windows_conda_3_6_cu102_nightly_build (28/29)

binary_windows_libtorch_3_7_cu102_debug_nightly_build (29/29)

1 failure not recognized by patterns:

Uh oh!

peterbell10 commented Jul 30, 2021

Uh oh!

malfet commented Jul 30, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

malfet commented Jul 30, 2021

Uh oh!

peterbell10 Jul 30, 2021

Choose a reason for hiding this comment

Uh oh!

peterbell10 commented Jul 30, 2021

Uh oh!

peterbell10 commented Sep 30, 2021

Uh oh!

peterbell10 commented Oct 8, 2021

Uh oh!

malfet commented Oct 8, 2021

Uh oh!

malfet commented Oct 11, 2021

Uh oh!

peterbell10 commented Oct 12, 2021

Uh oh!

ezyang commented Oct 12, 2021

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

peterbell10 commented Jul 29, 2021 •

edited by malfet

Loading

facebook-github-bot commented Jul 29, 2021 •

edited

Loading

malfet commented Jul 30, 2021 •

edited

Loading