Commit Graph

201 Commits

Author SHA1 Message Date
36ac095ff8 Migrate PyTorch to C++17 (#85969)
With CUDA-10.2 gone we can finally do it!

This PR mostly contains build system related changes, invasive functional ones are to be followed.
Among many expected tweaks to the build system, here are few unexpected ones:
 - Force onnx_proto project to be updated to C++17 to avoid `duplicate symbols` error when compiled by gcc-7.5.0, as storage rule for `constexpr` changed in C++17, but gcc does not seem to follow it
 - Do not use `std::apply` on CUDA but rely on the built-in variant, as it results in test failures when CUDA runtime picks host rather than device function when `std::apply` is invoked from CUDA code.
 - `std::decay_t` -> `::std::decay_t` and `std::move`->`::std::move` as VC++ for some reason claims that `std` symbol is ambigious
 - Disable use of `std::aligned_alloc` on Android, as its `libc++` does not implement it.

Some prerequisites:
 - https://github.com/pytorch/pytorch/pull/89297
 - https://github.com/pytorch/pytorch/pull/89605
 - https://github.com/pytorch/pytorch/pull/90228
 - https://github.com/pytorch/pytorch/pull/90389
 - https://github.com/pytorch/pytorch/pull/90379
 - https://github.com/pytorch/pytorch/pull/89570
 - https://github.com/facebookincubator/gloo/pull/336
 - https://github.com/facebookincubator/gloo/pull/343
 - 919676fb32

Fixes https://github.com/pytorch/pytorch/issues/56055

Pull Request resolved: https://github.com/pytorch/pytorch/pull/85969
Approved by: https://github.com/ezyang, https://github.com/kulinseth
2022-12-08 02:27:48 +00:00
5b51ca6808 Update CUDA compiler matrix (#86360)
Switch GCC/Clang max versions to be exclusive as the `include/crt/host_config.h` checks the major version only for the upper bound. This allows to be less restrictive and match the checks in the aforementioned header.
Also update the versions using that header in the CUDA SDKs.

Follow up to #82860

I noticed this as PyTorch 1.12.1 with CUDA 11.3.1 and GCC 10.3 was failing in the `test_cpp_extensions*` tests.

Example for CUDA 11.3.1 from the SDK header:

```
#if __GNUC__ > 11
// Error out
...
#if (__clang_major__ >= 12) || (__clang_major__ < 3) || ((__clang_major__ == 3) &&  (__clang_minor__ < 3))
// Error out
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/86360
Approved by: https://github.com/ezyang
2022-11-23 03:07:22 +00:00
575e02df53 Fix CUDNN_PATH handling on Windows (#88898)
Fixes https://github.com/pytorch/pytorch/issues/88873
Pull Request resolved: https://github.com/pytorch/pytorch/pull/88898
Approved by: https://github.com/kit1980
2022-11-11 21:19:26 +00:00
a7420d2ccb Hopper (sm90) support (#87736)
Essentially a followup of #87436

CC @xwang233 @ptrblck
Pull Request resolved: https://github.com/pytorch/pytorch/pull/87736
Approved by: https://github.com/xwang233, https://github.com/malfet
2022-11-09 01:49:50 +00:00
71fe069d98 ada lovelace (arch 8.9) support (#87436)
changes required to be able to compile https://github.com/pytorch/vision and https://github.com/nvidia/apex for `sm_89` architecture
Pull Request resolved: https://github.com/pytorch/pytorch/pull/87436
Approved by: https://github.com/ngimel
2022-10-24 21:25:36 +00:00
c28cdb53ea [BE] Delete BUILD_SPLIT_CUDA option (#87502)
As we are linking with cuDNN and cuBLAS dynamically for all configs anyway, as statically linked cuDNN is different library than dynamically linked one, increases default memory footprint, etc, and libtorch_cuda even if compiled for all GPU architectures is no longer approaching 2Gb binary size limit, so BUILD_SPLIT_CUDA can go away.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/87502
Approved by: https://github.com/atalman
2022-10-22 06:00:59 +00:00
fe87ae692f Fix check_compiler_ok_for_platform on non-English locales (#85891)
The function checks the output of e.g. `c++ -v` for "gcc version". But on another locale than English it might be "gcc-Version" which makes the check fail.
This causes the function to wrongly return false on systems where `c++` is a hardlink to `g++` and the current locale returns another output format.

Fix this by setting `LC_ALL=C`.

I found this as `test_utils.py` was failing in `test_cpp_compiler_is_ok`
Pull Request resolved: https://github.com/pytorch/pytorch/pull/85891
Approved by: https://github.com/ezyang
2022-09-29 18:36:36 +00:00
0183c1e336 Add __all__ to torch.utils submodules (#85331)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/85331
Approved by: https://github.com/albanD
2022-09-27 14:45:26 +00:00
1bf2371365 Rename path on Windows from lib/x64 to lib\x64 (#83417)
Use `os.path.join` to join path
Pull Request resolved: https://github.com/pytorch/pytorch/pull/83417
Approved by: https://github.com/ezyang
2022-08-15 14:47:19 +00:00
4618371da5 Integrate xdoctest - Rebased (#82797)
This is a new version of #15648 based on the latest master branch.

Unlike the previous PR where I fixed a lot of the doctests in addition to integrating xdoctest, I'm going to reduce the scope here. I'm simply going to integrate xdoctest, and then I'm going to mark all of the failing tests as "SKIP". This will let xdoctest run on the dashboards, provide some value, and still let the dashboards pass. I'll leave fixing the doctests themselves to another PR.

In my initial commit, I do the bare minimum to get something running with failing dashboards. The few tests that I marked as skip are causing segfaults. Running xdoctest results in 293 failed, 201 passed tests. The next commits will be to disable those tests. (unfortunately I don't have a tool that will insert the `#xdoctest: +SKIP` directive over every failing test, so I'm going to do this mostly manually.)

Fixes https://github.com/pytorch/pytorch/issues/71105

@ezyang
Pull Request resolved: https://github.com/pytorch/pytorch/pull/82797
Approved by: https://github.com/ezyang
2022-08-12 02:08:01 +00:00
737fa85dd2 Update CUDA compiler matrix (#82860)
Update CUDA compiler versions to match ones defined in
https://docs.nvidia.com/cuda/archive/11.4.1/cuda-installation-guide-linux/index.html#system-requirements
https://docs.nvidia.com/cuda/archive/11.5.0/cuda-installation-guide-linux/index.html#system-requirements
https://docs.nvidia.com/cuda/archive/11.6.0/cuda-installation-guide-linux/index.html#system-requirements
https://docs.nvidia.com/cuda/archive/11.7.0/cuda-installation-guide-linux/index.html#system-requirements

Special case 11.4.0, where maximum GCC supported version are similar to 11.3 rather that to 11.4.1+

Fixes https://github.com/pytorch/pytorch/issues/81039
Pull Request resolved: https://github.com/pytorch/pytorch/pull/82860
Approved by: https://github.com/huydhn
2022-08-06 00:46:30 +00:00
e849ed3d19 Redirect print messages to stderr in torch.utils.cpp_extension (#82097)
### Description
<!-- What did you change and why was it needed? -->

Listed in the commit message:

> The user may want to use `python3 -c "..."` to get the torch library
> path and the include path. Printing messages to stdout will mess up
> the output.

I'm using the command:

```bash
LIBTORCH_PATH="$(
    python3 -c 'print(":".join(__import__("torch.utils.cpp_extension", fromlist=[None]).library_paths()))'
)"
export LD_LIBRARY_PATH="${LIBTORCH_PATH}${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}"
```

To let the command line tools find the torch shared libraries. I think this would be a common use case for users who writing C/C++ extensions.

I got:

```console
$ LIBTORCH_PATH="$(python3 -c 'print(":".join(__import__("torch.utils.cpp_extension", fromlist=[None]).library_paths()))')"

$ export LD_LIBRARY_PATH="${LIBTORCH_PATH}${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}"

$ echo "LD_LIBRARY_PATH=${LD_LIBRARY_PATH}"
LD_LIBRARY_PATH=No CUDA runtime is found, using CUDA_HOME='/usr/local/cuda-11.6'
/opt/hostedtoolcache/Python/3.7.13/x64/lib/python3.7/site-packages/torch/lib:/usr/local/cuda-11.6/lib64:

$ ls -alh "${LIBTORCH_PATH}"
ls: cannot access 'No CUDA runtime is found, using CUDA_HOME='\''/usr/local/cuda-11.6'\'''$'\n''/opt/hostedtoolcache/Python/3.7.13/x64/lib/python3.7/site-packages/torch/lib': No such file or directory
```

This PR prints messages in `torch.utils.cpp_extension` to `stderr`, which allows users to get correct result using `VAR="$(python3 -c '...')"`

### Issue
<!-- Link to Issue ticket or RFP -->

N/A

### Testing
<!-- How did you test your change? -->

N/A
Pull Request resolved: https://github.com/pytorch/pytorch/pull/82097
Approved by: https://github.com/ezyang
2022-07-25 21:55:15 +00:00
95c148e502 [BE] Turn _check_cuda_version into a function (#81603)
It was class method, but does not use any of the class properties/called other class methods
Pull Request resolved: https://github.com/pytorch/pytorch/pull/81603
Approved by: https://github.com/ezyang
2022-07-17 05:49:39 +00:00
7e274964d3 [BE] Disamntle pyramid of doom in _check_cuda_version (#81602)
Replace `if stmt: doSmth; else: raise_or_return` with `if not stmt: raise_or_return; doSmth`
Pull Request resolved: https://github.com/pytorch/pytorch/pull/81602
Approved by: https://github.com/ezyang
2022-07-17 05:49:39 +00:00
71ee384924 [ROCm] Use torch._C._cuda_getArchFlags to get list of gfx archs pytorch was built for (#80498)
*even if no GPUs are available*

When building PyTorch extensions for ROCm Pytorch, if the user doesn't specify a list of archs using PYTORCH_ROCM_ARCH env var, we would like to use the list of gfx archs that PyTorch was built for as the default value. To do this successfully even in an environment where no GPUs are available eg. a build-only CPU node, we need to be able to get the list of archs. `torch.cuda.get_arch_list()` doesn't work here because it calls `torch.cuda.available()` first: 0922cc024e/torch/cuda/__init__.py (L463), which will return `False` if no GPUs are available, resulting in an empty list being returned by `torch.cuda.get_arch_list()`. To get around this issue, we call the underlying API `torch._C._cuda_getArchFlags()`
Pull Request resolved: https://github.com/pytorch/pytorch/pull/80498
Approved by: https://github.com/ezyang, https://github.com/malfet
2022-07-07 16:06:12 +00:00
ec4be38ba9 Revert "To add hipify_torch as a submodule in pytorch/third_party (#74704)"
This reverts commit 93b0fec39dd112d5c06106ad0186d55d61f1531a.

Reverted https://github.com/pytorch/pytorch/pull/74704 on behalf of https://github.com/malfet due to broke torchvision
2022-06-21 23:54:00 +00:00
93b0fec39d To add hipify_torch as a submodule in pytorch/third_party (#74704)
`hipify_torch` as a submodule in `pytorch/third_party`
Pull Request resolved: https://github.com/pytorch/pytorch/pull/74704
Approved by: https://github.com/jeffdaily, https://github.com/malfet
2022-06-21 18:56:49 +00:00
ef0332e36d Allow relocatable device code linking in pytorch CUDA extensions (#78225)
Close https://github.com/pytorch/pytorch/issues/57543

Doc: check `Relocatable device code linking:` in https://docs-preview.pytorch.org/78225/cpp_extension.html#torch.utils.cpp_extension.CUDAExtension
Pull Request resolved: https://github.com/pytorch/pytorch/pull/78225
Approved by: https://github.com/ezyang, https://github.com/malfet
2022-06-02 21:35:56 +00:00
fb0f285638 [lint] upgrade mypy to latest version
Fixes https://github.com/pytorch/pytorch/issues/75927.

Had to fix some bugs and add some ignores.

To check if clean:
```
lintrunner --paths-cmd='git grep -Il .' --take MYPY,MYPYSTRICT
```

Pull Request resolved: https://github.com/pytorch/pytorch/pull/76753

Approved by: https://github.com/malfet
2022-05-03 20:51:34 +00:00
3d7428d9ac Revert "[lint] upgrade mypy to latest version"
This reverts commit 9bf18aab94943f5352604a39340ad57ad4d0c5a4.

Reverted https://github.com/pytorch/pytorch/pull/76753 on behalf of https://github.com/suo
2022-05-03 20:01:18 +00:00
9bf18aab94 [lint] upgrade mypy to latest version
Fixes https://github.com/pytorch/pytorch/issues/75927.

Had to fix some bugs and add some ignores.

To check if clean:
```
lintrunner --paths-cmd='git grep -Il .' --take MYPY,MYPYSTRICT
```

Pull Request resolved: https://github.com/pytorch/pytorch/pull/76753

Approved by: https://github.com/malfet
2022-05-03 19:43:28 +00:00
7422ccea8b Hipify fixes for a successful DeepSpeed build
These commits are required to build DeepSpeed on ROCm without the hipify errors.

a41829d9ed
663c718462

cc: @jeffdaily

Pull Request resolved: https://github.com/pytorch/pytorch/pull/76141
Approved by: https://github.com/jeffdaily, https://github.com/pruthvistony, https://github.com/albanD
2022-04-28 13:19:59 +00:00
9562aedb58 ROCm: add HIP_HOME/include,lib in cpp_extensions (#75548)
Summary:
hip/hip_runtime.h and libamdhip64.so may be required to compile
extension such as torch_ucc. They are in $ROCM_HOME/hip by default,
and may not be symlinked to $ROCM_HOME/include and $ROCM_HOME/lib.
This commit defines $ROCM_HOME/hip as $HIP_HOME, and adds its include
and lib paths when building hipified extension.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/75548

Test Plan:
## Verify OSS pytorch + TorchUCC on an AMD GPU machine (MI100)
- step 1. Install OSS pytorch
```
export ROCM_PATH=/opt/rocm-4.5.2
git clone https://github.com/pytorch/pytorch.git
cd pytorch
python3 tools/amd_build/build_amd.py

USE_NCCL=0 USE_RCCL=0 USE_KINETO=0 with-proxy python3 setup.py develop
USE_NCCL=0 USE_RCCL=0 USE_KINETO=0 with-proxy python3 setup.py install
```

- step2. Install torchUCC extension
```
# /opt/rocm-4.5.2/include/hip does not exist, need include /opt/rocm-4.5.2/hip/include at compile time
export ROCM_PATH=/opt/rocm-4.5.2
export RCCL_INSTALL_DIR=/opt/rccl-rocm-rel-4.4-rdc
git clone https://github.com/facebookresearch/torch_ucc.git
cd torch_ucc
UCX_HOME=$RCCL_INSTALL_DIR UCC_HOME=$RCCL_INSTALL_DIR WITH_CUDA=$ROCM_PATH python setup.py
```
Build log before fix (error "hip/hip_runtime.h: No such file or directory"): P493038915
Build log after fix: P493037572

Reviewed By: ezyang

Differential Revision: D35506098

Pulled By: minsii

fbshipit-source-id: 76cbb6d4eaa6549a00898c9d9ebaca47a55330e9
(cherry picked from commit d684c080edf1fbd293e3321151976812c1da8533)
2022-04-19 20:51:37 +00:00
7a243ddd19 Add import to importlib.abc
Fixes #70525

```
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-3-334d309cf512> in <module>
----> 1 lltm_cpp = load(name="lltm_cpp", sources=["lltm.cpp"])

/usr/lib/python3.10/site-packages/torch/utils/cpp_extension.py in load(name, sources, extra_cflags, extra_cuda_cflags, extra_ldflags, extra_include_paths, build_directory, verbose, with_cuda, is_python_module, is_standalone, keep_intermediates)
   1122                 verbose=True)
   1123     '''
-> 1124     return _jit_compile(
   1125         name,
   1126         [sources] if isinstance(sources, str) else sources,

/usr/lib/python3.10/site-packages/torch/utils/cpp_extension.py in _jit_compile(name, sources, extra_cflags, extra_cuda_cflags, extra_ldflags, extra_include_paths, build_directory, verbose, with_cuda, is_python_module, is_standalone, keep_intermediates)
   1360         return _get_exec_path(name, build_directory)
   1361
-> 1362     return _import_module_from_library(name, build_directory, is_python_module)
   1363
   1364

/usr/lib/python3.10/site-packages/torch/utils/cpp_extension.py in _import_module_from_library(module_name, path, is_python_module)
   1751         spec = importlib.util.spec_from_file_location(module_name, filepath)
   1752         module = importlib.util.module_from_spec(spec)
-> 1753         assert isinstance(spec.loader, importlib.abc.Loader)
   1754         spec.loader.exec_module(module)
   1755         return module

AttributeError: module 'importlib' has no attribute 'abc'
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/75736
Approved by: https://github.com/ezyang
2022-04-14 03:32:30 +00:00
86deecd7be Check clang++/g++ version when compiling CUDA extensions (#63230)
Summary:
See https://github.com/pytorch/pytorch/issues/55267

Pull Request resolved: https://github.com/pytorch/pytorch/pull/63230

Reviewed By: soulitzer

Differential Revision: D34159119

Pulled By: malfet

fbshipit-source-id: 6eef7582388bf6a42dcc1d82b6e4b1f40f418dd7
(cherry picked from commit 2056d0a0be7951602de22f8d3b4efc28dd71b6c2)
2022-02-24 08:32:32 +00:00
46f9e16afe Documenting cuda 11.5 windows issue (#73013)
Summary:
Adding documentation about compiling extension with CUDA 11.5 and Windows

Example of failure: https://github.com/pytorch/pytorch/runs/4408796098?check_suite_focus=true

 Note: Don't use torch/extension.h In CUDA 11.5 under windows in your C++ code:
    Use aten instead of torch interface in all cuda 11.5 code under windows. It has been failing with errors, due to a bug in nvcc.
    Example use:
        >>> #include <ATen/ATen.h>
        >>> at::Tensor SigmoidAlphaBlendForwardCuda(....)
    Instead of:
        >>> #include <torch/extension.h>
        >>> torch::Tensor SigmoidAlphaBlendForwardCuda(...)
    Currently open issue for nvcc bug: https://github.com/pytorch/pytorch/issues/69460
    Complete Workaround code example: cb170ac024

Pull Request resolved: https://github.com/pytorch/pytorch/pull/73013

Reviewed By: malfet, seemethere

Differential Revision: D34306134

Pulled By: atalman

fbshipit-source-id: 3c5b9d7a89c91bd1920dc63dbd356e45dc48a8bd
(cherry picked from commit 87098e7f17fca1b98c90fafe2dde1defb6633f49)
2022-02-19 02:34:59 +00:00
8dfdc3df82 [ROCm] Refactor how to specify AMD gpu targets using PYTORCH_ROCM_ARCH (#61706)
Summary:
Remove all hardcoded AMD gfx targets

PyTorch build and Magma build will use rocm_agent_enumerator as
backup if PYTORCH_ROCM_ARCH env var is not defined

PyTorch extensions will use same gfx targets as the PyTorch build,
unless PYTORCH_ROCM_ARCH env var is defined

torch.cuda.get_arch_list() now works for ROCm builds

PyTorch CI dockers will continue to be built for gfx900 and gfx906 for now.

PYTORCH_ROCM_ARCH env var can be a space or semicolon separated list of gfx archs eg. "gfx900 gfx906" or "gfx900;gfx906"
cc jeffdaily sunway513 jithunnair-amd ROCmSupport KyleCZH

Pull Request resolved: https://github.com/pytorch/pytorch/pull/61706

Reviewed By: seemethere

Differential Revision: D32735862

Pulled By: malfet

fbshipit-source-id: 3170e445e738e3ce373203e1e4ae99c84e645d7d
2021-12-13 15:41:40 -08:00
bede18b061 Add support for C++ frontend wrapper on Linux (#69094)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/69094

Partially addresses https://github.com/pytorch/pytorch/issues/68768

Test Plan: Imported from OSS

Reviewed By: seemethere

Differential Revision: D32730079

Pulled By: malfet

fbshipit-source-id: 854e4215ff66e087bdf354fed7a17e87f2649c87
2021-12-02 16:47:00 -08:00
c08e95dd9c Introduce IS_LINUX and IS_MACOS global vars (#69093)
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/69093

Test Plan: Imported from OSS

Reviewed By: samdow

Differential Revision: D32730080

Pulled By: malfet

fbshipit-source-id: aa3f218d09814b4edd96b01c7b57b85fd58c47fc
2021-12-01 09:47:38 -08:00
f6f1b580f8 Fix mypy in cpp_extension.py (#69101)
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/69101

Test Plan: Imported from OSS

Reviewed By: atalman, janeyx99

Differential Revision: D32730081

Pulled By: malfet

fbshipit-source-id: 76ace65b51850b74b175a3c4688c05e107873e8d
2021-11-30 16:01:55 -08:00
78f970568c Add dummy op to use instead of searchsorted (#66964)
Summary:
Would help unblock https://github.com/pytorch/pytorch/issues/66818 if this actually works

Pull Request resolved: https://github.com/pytorch/pytorch/pull/66964

Reviewed By: mruberry

Differential Revision: D31817942

Pulled By: janeyx99

fbshipit-source-id: 9e9a2bcb0c0479ec7000ab8760a2e64bf0e85e95
2021-10-21 12:56:22 -07:00
085e2f7bdd [ROCm] Changes not to rely on CUDA_VERSION or HIP_VERSION (#65610)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/65610

- Replace HIP_PLATFORM_HCC with USE_ROCM
- Dont rely on CUDA_VERSION or HIP_VERSION and use USE_ROCM and ROCM_VERSION.

- In the next PR
   - Will be removing the mapping from CUDA_VERSION to HIP_VERSION and CUDA to HIP in hipify.
   - HIP_PLATFORM_HCC is deprecated, so will add HIP_PLATFORM_AMD to support HIP host code compilation on gcc.

cc jeffdaily sunway513 jithunnair-amd ROCmSupport amathews-amd

Reviewed By: jbschlosser

Differential Revision: D30909053

Pulled By: ezyang

fbshipit-source-id: 224a966ebf1aaec79beccbbd686fdf3d49267e06
2021-09-29 09:55:43 -07:00
e6dc7bc61b Subprocess encoding fixes for cpp extension (#63756)
Summary:
Fixes https://github.com/pytorch/pytorch/issues/63584

Pull Request resolved: https://github.com/pytorch/pytorch/pull/63756

Reviewed By: bdhirsh

Differential Revision: D30485046

Pulled By: ezyang

fbshipit-source-id: 4f0ac383da4e8843e2a602dceae85f389d7434ee
2021-08-24 10:46:11 -07:00
9679fa7f30 Update cpp_extension.py (#61484)
Summary:
By default, majority of Python-3.[6789] installation comes with `pkg_resources.packaging` version 16.8 (or `setuptool` older than 49.6.0), which does not have major/minor properties on Version package, as one can observe in https://github.com/pypa/setuptools/blob/v49.5.0/pkg_resources/_vendor/packaging/version.py
On the other hand, compare operators exists, so why not use it to check for version equality

Fixes https://github.com/pytorch/pytorch/issues/61036

Pull Request resolved: https://github.com/pytorch/pytorch/pull/61484

Reviewed By: walterddr, seemethere

Differential Revision: D29643883

Pulled By: malfet

fbshipit-source-id: 3db9168c1b009ac3a278709083ea8c5b417471b8
2021-07-13 07:11:58 -07:00
2d0c6e60a7 going back to use packaging.version.parse instead (#61053)
Summary:
I think this may be related to https://app.circleci.com/pipelines/github/pytorch/vision/9352/workflows/9c8afb1c-6157-4c82-a5c8-105c5adac57d/jobs/687003

Apparently `pkg_resource.parse_version` returns a type of `pkg_resources.extern.packaging.version.Version` instead of `packaging.version.Version` and seems on some older version of the setuptools it doesn't support `.major/minor` operation. changing it back to using `packaging.version.parse`

Pull Request resolved: https://github.com/pytorch/pytorch/pull/61053

Test Plan: CI

Reviewed By: samestep

Differential Revision: D29494322

Pulled By: walterddr

fbshipit-source-id: 294572a10b167677440d7404e5ebe007ab59d299
2021-06-30 16:23:59 -07:00
d46eb77b04 Improve CUDA extension building error/warning messages (#59665)
Summary:
See https://github.com/pytorch/pytorch/issues/55267

Pull Request resolved: https://github.com/pytorch/pytorch/pull/59665

Reviewed By: mruberry

Differential Revision: D29462248

Pulled By: ezyang

fbshipit-source-id: 9de13a284a14a7cd24200b9684151ce652e1eb1e
2021-06-29 13:03:30 -07:00
6322f66878 Add python version and cuda-specific folder to store extensions (#60592)
Summary:
See https://github.com/pytorch/pytorch/issues/55267

Pull Request resolved: https://github.com/pytorch/pytorch/pull/60592

Reviewed By: albanD

Differential Revision: D29353368

Pulled By: ezyang

fbshipit-source-id: 1fbcd021f1030132c0f950f33ce4a3a2fef351e0
2021-06-25 10:27:04 -07:00
0a0e024648 use importlib instead of imp as it support python 3.5+ (#57160)
Summary:
Prevent some annoying deprecation warning when importing cpp_extensions

Pull Request resolved: https://github.com/pytorch/pytorch/pull/57160

Reviewed By: astaff

Differential Revision: D28096751

Pulled By: albanD

fbshipit-source-id: f169ad4c4945b0fff54c0339052a29f95b9f1831
2021-05-03 05:56:25 -07:00
4b96fc060b Remove distutils (#57040)
Summary:
[distutils](https://docs.python.org/3/library/distutils.html) is on its way out and will be deprecated-on-import for Python 3.10+ and removed in Python 3.12 (see [PEP 632](https://www.python.org/dev/peps/pep-0632/)). There's no reason for us to keep it around since all the functionality we want from it can be found in `setuptools` / `sysconfig`. `setuptools` includes a copy of most of `distutils` (which is fine to use according to the PEP), that it uses under the hood, so this PR also uses that in some places.

Fixes #56527
Pull Request resolved: https://github.com/pytorch/pytorch/pull/57040

Pulled By: driazati

Reviewed By: nikithamalgifb

Differential Revision: D28051356

fbshipit-source-id: 1ca312219032540e755593e50da0c9e23c62d720
2021-04-29 12:10:11 -07:00
75024e228c Add lint for unqualified type: ignore (#56290)
Summary:
The other half of https://github.com/pytorch/pytorch/issues/56272.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/56290

Test Plan:
CI should pass on the tip of this PR, and we know that the lint works because the following CI runs (before this PR was finished) failed:

- https://github.com/pytorch/pytorch/runs/2384511062
- https://github.com/pytorch/pytorch/actions/runs/765036024

Reviewed By: seemethere

Differential Revision: D27867219

Pulled By: samestep

fbshipit-source-id: e648f07b6822867e70833e23ddafe7fb7eaca235
2021-04-21 08:07:23 -07:00
b383b63550 [ROCm] Updating ROCM_HOME handling for >ROCm 4.0 (#55968)
Summary:
- This change is required to handle the case when hipcc is
  updated to the latest using update-alternatives.
- Update-alternatives support for few ROCm binaries is available
  from ROCm 4.1 onwards.
- This change doesnt not affect any previous versions of ROCm.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/55968

Reviewed By: mruberry

Differential Revision: D27785123

Pulled By: ezyang

fbshipit-source-id: 8467e468d8d51277fab9b0c8cbd57e80bbcfc7f7
2021-04-15 07:48:36 -07:00
0b8bd22614 Fix bug with rebuilding extensions every import (#56015)
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/56015

Reviewed By: mruberry

Differential Revision: D27765934

Pulled By: ezyang

fbshipit-source-id: 65cace951fce5f2284ab91d8bd687ac89a2311fb
2021-04-14 13:25:01 -07:00
e5b97777e3 [ROCm] allow PYTORCH_ROCM_ARCH in cpp_extension.py (#54341)
Summary:
Allows extensions to override ROCm gfx arch targets.  Reuses the same env var used during cmake build for consistency.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/54341

Reviewed By: bdhirsh

Differential Revision: D27244010

Pulled By: heitorschueroff

fbshipit-source-id: 279e1a41ee395a0596aa7f696b6e908cf7f5bb83
2021-03-23 13:06:00 -07:00
65087dd1d4 Fix broken link from load_inline to new test location (#53701)
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/53701

Reviewed By: jbschlosser

Differential Revision: D27047406

Pulled By: ezyang

fbshipit-source-id: 0be6e669cf41527d3ffeb101e5f36db07e41b4af
2021-03-15 13:53:15 -07:00
44ff79d849 Automatically set BUILD_SPLIT_CUDA for cpp exts (#52503)
Summary:
Fixes https://github.com/pytorch/vision/pull/3418#issuecomment-781673110

Pull Request resolved: https://github.com/pytorch/pytorch/pull/52503

Reviewed By: malfet

Differential Revision: D26546857

Pulled By: janeyx99

fbshipit-source-id: a100b408e7cd28695145a1dda7f2fa081bb7f21f
2021-02-19 12:22:55 -08:00
550c965b2e Re-enable test_standalone_load for Windows 11.1 (#51596)
Summary:
This fixes the previous erroring out by adding stricter conditions in cpp_extension.py.

To test, run a split torch_cuda build on Windows with export BUILD_SPLIT_CUDA=ON && python setup.py develop and then run the following test: python test/test_utils.py TestStandaloneCPPJIT.test_load_standalone. It should pass.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/51596

Reviewed By: malfet

Differential Revision: D26213816

Pulled By: janeyx99

fbshipit-source-id: a752ce7f9ab9d73dcf56f952bed2f2e040614443
2021-02-03 08:58:34 -08:00
f7313b3105 Fix Python.h discovery logic on some MacOS platforms (#51586)
Summary:
On all non-Windows platforms we should use 'posix_prefix' schema to discover location of Python.h header

Pull Request resolved: https://github.com/pytorch/pytorch/pull/51586

Reviewed By: ezyang

Differential Revision: D26208684

Pulled By: malfet

fbshipit-source-id: bafa6d79de42231629960c642d535f1fcf7a427f
2021-02-02 21:38:37 -08:00
88af2149e1 Add build option to split torch_cuda library into torch_cuda_cu and torch_cuda_cpp (#49050)
Summary:
Because of the size of our `libtorch_cuda.so`, linking with other hefty binaries presents a problem where 32bit relocation markers are too small and end up overflowing. This PR attempts to break up `torch_cuda` into `torch_cuda_cu` and `torch_cuda_cpp`.

`torch_cuda_cu`: all the files previously in `Caffe2_GPU_SRCS` that are
* pure `.cu` files in `aten`match
* all the BLAS files
* all the THC files, except for THCAllocator.cpp, THCCachingHostAllocator.cpp and THCGeneral.cpp
* all files in`detail`
* LegacyDefinitions.cpp and LegacyTHFunctionsCUDA.cpp
* Register*CUDA.cpp
* CUDAHooks.cpp
* CUDASolver.cpp
* TensorShapeCUDA.cpp

`torch_cuda_cpp`: all other files in `Caffe2_GPU_SRCS`

Accordingly, TORCH_CUDA_API and TORCH_CUDA_BUILD_MAIN_LIB usages are getting split as well to TORCH_CUDA_CU_API and TORCH_CUDA_CPP_API.

To test this locally, you can run `export BUILD_SPLIT_CUDA=ON && python setup.py develop`. In your `build/lib` folder, you should find binaries for both `torch_cuda_cpp` and `torch_cuda_cu`. To see that the SPLIT_CUDA option was toggled, you can grep the Summary of running cmake and make sure `Split CUDA` is ON.

This build option is tested on CI for CUDA 11.1 builds (linux for now, but windows soon).

Pull Request resolved: https://github.com/pytorch/pytorch/pull/49050

Reviewed By: walterddr

Differential Revision: D26114310

Pulled By: janeyx99

fbshipit-source-id: 0180f2519abb5a9cdde16a6fb7dd3171cff687a6
2021-02-01 18:42:35 -08:00
327539ca79 Fix bug in hipify if include_dirs is not specified in setup.py (#50703)
Summary:
Bugs:
1) would introduce -I* in compile commands
2) wouldn't hipify source code directly in build_dir, only one level down or more

Pull Request resolved: https://github.com/pytorch/pytorch/pull/50703

Reviewed By: mrshenli

Differential Revision: D25949070

Pulled By: ngimel

fbshipit-source-id: 018c2a056b68019a922e20e5db2eb8435ad147fe
2021-01-19 16:30:17 -08:00
e29082b2a6 Run mypy over test/test_utils.py (#50278)
Summary:
_resubmission of gh-49654, which was reverted due to a cross-merge conflict_

This caught one incorrect annotation in `cpp_extension.load`.

xref gh-16574.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/50278

Reviewed By: walterddr

Differential Revision: D25865278

Pulled By: ezyang

fbshipit-source-id: 25489191628af5cf9468136db36f5a0f72d9d54d
2021-01-11 08:16:23 -08:00