Enable x86 CPU vectorization on windows [submodule sleef] #118980

xuhancn · 2024-02-02T07:46:39Z

Enable VEC on Windows OS.

Fix some type defination gap between Windows and Linux.
Fix some operator not support on Windows, such as [], /.
Enable static sleef library build on Windows.
Disable unsupported function overloading on MSVC.
Upgrade submodule sleef lib, which fixed build issue on Windows.
Fixed bazel build issues.
Fix test app not link to sleef on Windows.

Note: If rebuild fail after pulled this PR, please sync sleef submodule by run:

git submodule sync
git submodule update --init --recursive

cc @peterjc123 @mszhanyi @skyline75489 @nbcsm @vladimir-aubrecht @iremyux @Blackhex @cristianPanaite @jgong5 @mingfeima @XiaobingSuper @sanchitintel @ashokei @jingxu10 @frank-wei

pytorch-bot · 2024-02-02T07:46:43Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/118980

📄 Preview Python docs built from this PR
📄 Preview C++ docs built from this PR
❓ Need help or want to give feedback on the CI? Visit the bot commands wiki or our office hours

Note: Links to docs will display an error until the docs builds have been completed.

❌ 1 New Failure, 2 Unrelated Failures

As of commit 4336fb0 with merge base a001b4b ():

NEW FAILURE - The following job has failed:

linux-binary-libtorch-cxx11-abi / libtorch-rocm6_0-shared-with-deps-cxx11-abi-build / build (gh)
Process completed with exit code 1.

FLAKY - The following jobs failed but were likely due to flakiness present on trunk:

pull / linux-jammy-py3.8-gcc11 / test (default, 2, 3, linux.2xlarge) (gh)
inductor/test_torchinductor_opinfo.py::TestInductorOpInfoCPU::test_comprehensive_fft_ihfftn_cpu_int32
pull / linux-jammy-py3.8-gcc11 / test (default, 3, 3, linux.2xlarge) (gh)
inductor/test_torchinductor_opinfo.py::TestInductorOpInfoCPU::test_comprehensive_fft_rfft2_cpu_int32

This comment was automatically generated by Dr. CI and updates every 15 minutes.

aten/src/ATen/native/cpu/GridSamplerKernel.cpp

ezyang · 2024-02-06T03:02:46Z

@pytorchbot merge

pytorchmergebot · 2024-02-06T03:04:45Z

Merge failed

Reason: This PR needs a release notes: label
If your changes are user facing and intended to be a part of release notes, please use a label starting with release notes:.

If not, please add the topic: not user facing label.

To add a label, you can comment to pytorchbot, for example
@pytorchbot label "topic: not user facing"

For more information, see
https://github.com/pytorch/pytorch/wiki/PyTorch-AutoLabel-Bot#why-categorize-for-release-notes-and-how-does-it-work.

Details for Dev Infra team

Raised by workflow job

xuhancn · 2024-02-06T03:21:52Z

@pytorchbot label "topic: not user facing"

xuhancn · 2024-02-06T03:28:29Z

@pytorchbot merge

pytorchmergebot · 2024-02-06T03:30:41Z

Merge started

Your change will be merged once all checks pass (ETA 0-4 Hours).

Learn more about merging in the wiki.

Questions? Feedback? Please reach out to the PyTorch DevX Team

Advanced Debugging

Check the merge workflow status
here

pytorchmergebot · 2024-02-06T03:30:54Z

Merge failed

Reason: 1 mandatory check(s) failed. The first few are:

pull / linux-jammy-py3-clang12-executorch / test (executorch, 1, 1, linux.2xlarge)

Dig deeper by viewing the failures on hud

Details for Dev Infra team

Raised by workflow job

Failing merge rule: Core Maintainers

ezyang · 2024-02-06T03:40:54Z

@pytorchbot merge -i

pytorchmergebot · 2024-02-06T03:45:43Z

Merge started

Your change will be merged while ignoring the following 1 checks: .github/workflows/pull.yml / linux-jammy-py3-clang12-executorch / test (executorch, 1, 1, linux.2xlarge)

Learn more about merging in the wiki.

Questions? Feedback? Please reach out to the PyTorch DevX Team

Advanced Debugging

Check the merge workflow status
here

xuhancn · 2024-03-29T07:29:54Z

merge PR avoid code getting stale, @huydhn please update fb_code sleef to shibatch/sleef@60e76d2, it should be works well. CC: @malfet

atalman · 2024-03-30T14:19:10Z

@pytorchbot revert -m "Breaks internal build" -c ghfirst

atalman · 2024-03-30T14:20:04Z

Please note multiple jobs are broken. here are some of the errors:

xplat/caffe2/aten/src/ATen/cpu/vec/vec256/vec256_bfloat16.h:495:16: error: cannot initialize a parameter of type '__m256 (*)(__m256)' with an lvalue of type 'const __m256 (__m256)': different return type ('__m256' (vector of 8 'float' values) vs 'const __m256' (vector of 8 'float' values))
    return map(Sleef_logf8_u10);
               ^~~~~~~~~~~~~~~
xplat/caffe2/aten/src/ATen/native/cpu/UnaryOpsKernel.cpp:157:52: note: in instantiation of member function 'at::vec::Vectorized16<c10::BFloat16>::log' requested here
                return (x_vec / (kOneVec - x_vec)).log();
                                                   ^
xplat/caffe2/aten/src/ATen/cpu/vec/vec256/vec256_bfloat16.h:281:58: note: passing argument to parameter 'vop' here
  Vectorized<T> map(SLEEF_CONST __m256 (*SLEEF_CONST_OLD vop)(__m256)) const {
                                                         ^
xplat/caffe2/aten/src/ATen/cpu/vec/vec256/vec256_bfloat16.h:495:16: error: cannot initialize a parameter of type '__m256 (*)(__m256)' with an lvalue of type 'const __m256 (__m256)': different return type ('__m256' (vector of 8 'float' values) vs 'const __m256' (vector of 8 'float' values))
    return map(Sleef_logf8_u10);
               ^~~~~~~~~~~~~~~
xplat/caffe2/aten/src/ATen/native/cpu/UnaryOpsKernel.cpp:157:52: note: in instantiation of member function 'at::vec::Vectorized16<c10::Half>::log' requested here
                return (x_vec / (kOneVec - x_vec)).log();
                                                   ^
xplat/caffe2/aten/src/ATen/cpu/vec/vec256/vec256_bfloat16.h:281:58: note: passing argument to parameter 'vop' here
  Vectorized<T> map(SLEEF_CONST __m256 (*SLEEF_CONST_OLD vop)(__m256)) const {
                                                         ^
xplat/caffe2/aten/src/ATen/cpu/vec/vec256/vec256_bfloat16.h:510:16: error: cannot initialize a parameter of type '__m256 (*)(__m256)' with an lvalue of type 'const __m256 (__m256)': different return type ('__m256' (vector of 8 'float' values) vs 'const __m256' (vector of 8 'float' values))
    return map(Sleef_sinhf8_u10);
               ^~~~~~~~~~~~~~~~
xplat/caffe2/aten/src/ATen/native/cpu/UnaryOpsKernel.cpp:371:60: note: in instantiation of member function 'at::vec::Vectorized16<c10::BFloat16>::sinh' requested here
        [=](Vectorized<scalar_t> self_vec){return self_vec.sinh();});
                                                           ^
xplat/caffe2/aten/src/ATen/cpu/vec/vec256/vec256_bfloat16.h:281:58: note: passing argument to parameter 'vop' here
  Vectorized<T> map(SLEEF_CONST __m256 (*SLEEF_CONST_OLD vop)(__m256)) const {
                                                         ^
xplat/caffe2/aten/src/ATen/cpu/vec/vec256/vec256_bfloat16.h:510:16: error: cannot initialize a parameter of type '__m256 (*)(__m256)' with an lvalue of type 'const __m256 (__m256)': different return type ('__m256' (vector of 8 'float' values) vs 'const __m256' (vector of 8 'float' values))
    return map(Sleef_sinhf8_u10);
               ^~~~~~~~~~~~~~~~
xplat/caffe2/aten/src/ATen/native/cpu/UnaryOpsKernel.cpp:371:60: note: in instantiation of member function 'at::vec::Vectorized16<c10::Half>::sinh' requested here
        [=](Vectorized<scalar_t> self_vec){return self_vec.sinh();});
                                                           ^
xplat/caffe2/aten/src/ATen/cpu/vec/vec256/vec256_bfloat16.h:281:58: note: passing argument to parameter 'vop' here
  Vectorized<T> map(SLEEF_CONST __m256 (*SLEEF_CONST_OLD vop)(__m256)) const {
                                                         ^
xplat/caffe2/aten/src/ATen/cpu/vec/vec256/vec256_bfloat16.h:516:16: error: cannot initialize a parameter of type '__m256 (*)(__m256)' with an lvalue of type 'const __m256 (__m256)': different return type ('__m256' (vector of 8 'float' values) vs 'const __m256' (vector of 8 'float' values))
    return map(Sleef_coshf8_u10);
               ^~~~~~~~~~~~~~~~
xplat/caffe2/aten/src/ATen/native/cpu/UnaryOpsKernel.cpp:380:60: note: in instantiation of member function 'at::vec::Vectorized16<c10::BFloat16>::cosh' requested here
        [=](Vectorized<scalar_t> self_vec){return self_vec.cosh();});
                                                           ^
xplat/caffe2/aten/src/ATen/cpu/vec/vec256/vec256_bfloat16.h:281:58: note: passing argument to parameter 'vop' here
  Vectorized<T> map(SLEEF_CONST __m256 (*SLEEF_CONST_OLD vop)(__m256)) const {
                                                         ^
xplat/caffe2/aten/src/ATen/cpu/vec/vec256/vec256_bfloat16.h:516:16: error: cannot initialize a parameter of type '__m256 (*)(__m256)' with an lvalue of type 'const __m256 (__m256)': different return type ('__m256' (vector of 8 'float' values) vs 'const __m256' (vector of 8 'float' values))
    return map(Sleef_coshf8_u10);
               ^~~~~~~~~~~~~~~~
xplat/caffe2/aten/src/ATen/native/cpu/UnaryOpsKernel.cpp:380:60: note: in instantiation of member function 'at::vec::Vectorized16<c10::Half>::cosh' requested here
        [=](Vectorized<scalar_t> self_vec){return self_vec.cosh();});

pytorchmergebot · 2024-03-30T14:20:50Z

@pytorchbot successfully started a revert job. Check the current status here.
Questions? Feedback? Please reach out to the PyTorch DevX Team

…18980)" This reverts commit d94db5f. Reverted #118980 on behalf of https://github.com/atalman due to Breaks internal build ([comment](#118980 (comment)))

pytorchmergebot · 2024-03-30T14:21:00Z

@xuhancn your PR has been successfully reverted.

xuhancn · 2024-03-30T18:42:14Z

@pytorchmergebot merge

pytorchmergebot · 2024-03-30T18:43:59Z

Merge started

Your change will be merged once all checks pass (ETA 0-4 Hours).

Learn more about merging in the wiki.

Questions? Feedback? Please reach out to the PyTorch DevX Team

Advanced Debugging

Check the merge workflow status
here

pytorchmergebot · 2024-03-30T18:59:37Z

Merge failed

Reason: 1 jobs have failed, first few of them are: trunk / linux-focal-rocm6.0-py3.8 / build

Details for Dev Infra team

Raised by workflow job

xuhancn · 2024-03-31T03:05:14Z

@pytorchmergebot merge -i

pytorchmergebot · 2024-03-31T03:07:12Z

Merge started

Your change will be merged while ignoring the following 3 checks: pull / linux-jammy-py3.8-gcc11 / test (default, 2, 3, linux.2xlarge), pull / linux-jammy-py3.8-gcc11 / test (default, 3, 3, linux.2xlarge), linux-binary-libtorch-cxx11-abi / libtorch-rocm6_0-shared-with-deps-cxx11-abi-build / build

Learn more about merging in the wiki.

Questions? Feedback? Please reach out to the PyTorch DevX Team

Advanced Debugging

Check the merge workflow status
here

huydhn · 2024-03-31T07:46:30Z

@atalman I'm importing the sleef module to unblocking this internally, I have the diff to import sleef at D55567582. We can rebase this on top to see if it fixes the build issue.

…torch#118980)" This reverts commit 56451cd.

…8980) Enable VEC on Windows OS. 1. Fix some type defination gap between Windows and Linux. 2. Fix some operator not support on Windows, such as [], /. 3. Enable static sleef library build on Windows. 4. Disable unsupported function overloading on MSVC. 5. Upgrade submodule sleef lib, which fixed build issue on Windows. 6. Fixed bazel build issues. 7. Fix test app not link to sleef on Windows. Note: If rebuild fail after pulled this PR, please sync `sleef` submodule by run: ```cmd git submodule sync git submodule update --init --recursive ``` Pull Request resolved: pytorch#118980 Approved by: https://github.com/jgong5, https://github.com/ezyang, https://github.com/malfet

Subsequent PR to #118980, fix sleef build warning. submodule sleef, include this sleef PR: shibatch/sleef#514 Pull Request resolved: #122168 Approved by: https://github.com/jgong5, https://github.com/ezyang

…18980)" This reverts commit 2b06098. Reverted #118980 on behalf of https://github.com/zou3519 due to This caused build failures for 2+ pytorch devs, so we're reverting it to be safe ([comment](#118980 (comment)))

…torch#118980)" This reverts commit d94db5f. Reverted pytorch#118980 on behalf of https://github.com/atalman due to Breaks internal build ([comment](pytorch#118980 (comment)))

…8980) Enable VEC on Windows OS. 1. Fix some type defination gap between Windows and Linux. 2. Fix some operator not support on Windows, such as [], /. 3. Enable static sleef library build on Windows. 4. Disable unsupported function overloading on MSVC. 5. Upgrade submodule sleef lib, which fixed build issue on Windows. 6. Fixed bazel build issues. 7. Fix test app not link to sleef on Windows. Note: If rebuild fail after pulled this PR, please sync `sleef` submodule by run: ```cmd git submodule sync git submodule update --init --recursive ``` Pull Request resolved: pytorch#118980 Approved by: https://github.com/jgong5, https://github.com/ezyang, https://github.com/malfet

Enable VEC on Windows OS. 1. Fix some type defination gap between Windows and Linux. 2. Fix some operator not support on Windows, such as [], /. 3. Enable static sleef library build on Windows. 4. Disable unsupported function overloading on MSVC. 5. Upgrade submodule sleef lib, which fixed build issue on Windows. 6. Fixed bazel build issues. 7. Fix test app not link to sleef on Windows. Pull Request resolved: #118980 Approved by: https://github.com/jgong5, https://github.com/ezyang, https://github.com/malfet

…18980)" This reverts commit aa74a8b. Reverted #118980 on behalf of https://github.com/huydhn due to Sorry for revert your change one more time but the hard part is that it breaks lot of internal builds ([comment](#118980 (comment)))

…ytorch#118980)" This reverts commit 7e3fc10.

pytorchbot added the open source label Feb 2, 2024

github-actions bot added the module: cpu CPU specific problem (e.g., perf, algorithm) label Feb 2, 2024

jgong5 reviewed Feb 3, 2024

View reviewed changes

xuhancn added module: intel Specific to x86 architecture intel This tag is for PR from Intel labels Feb 5, 2024

xuhancn self-assigned this Feb 5, 2024

xuhancn added the module: windows Windows support for PyTorch label Feb 5, 2024

xuhancn requested a review from jgong5 February 5, 2024 09:50

xuhancn marked this pull request as ready for review February 5, 2024 09:50

jgong5 approved these changes Feb 5, 2024

View reviewed changes

xuhancn requested review from malfet and ezyang February 5, 2024 12:43

xuhancn mentioned this pull request Feb 5, 2024

Bad performance of stock model on Windows compared to Linux #62387

Open

ezyang approved these changes Feb 6, 2024

View reviewed changes

pytorch-bot bot added the ciflow/trunk Trigger trunk jobs on your pull request label Feb 6, 2024

pytorchmergebot added the merging label Feb 6, 2024

pytorchmergebot removed the merging label Feb 6, 2024

pytorch-bot bot added the topic: not user facing topic category label Feb 6, 2024

pytorchmergebot added the merging label Feb 6, 2024

pytorchmergebot removed the merging label Feb 6, 2024

pytorchmergebot added the merging label Feb 6, 2024

pytorchmergebot reopened this Mar 30, 2024

try to fix fb_code sleef related build fail.

4336fb0

pytorchmergebot added the merging label Mar 30, 2024

pytorchmergebot removed the merging label Mar 30, 2024

pytorchmergebot added the merging label Mar 31, 2024

pytorchmergebot closed this in 56451cd Mar 31, 2024

pytorchmergebot removed the merging label Mar 31, 2024

xuhancn mentioned this pull request Apr 1, 2024

[Don't Merge] add sleef shared lib support. #123085

Closed

ozanMSFT added a commit to iremyux/pytorch that referenced this pull request Apr 19, 2024

Revert "Enable x86 CPU vectorization on windows [submodule sleef] (py…

7e3fc10

…torch#118980)" This reverts commit 56451cd.

iremyux added a commit to iremyux/pytorch that referenced this pull request May 1, 2024

Reapply "Enable x86 CPU vectorization on windows [submodule sleef] (p…

3126537

…ytorch#118980)" This reverts commit 7e3fc10.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Enable x86 CPU vectorization on windows [submodule sleef] #118980

Enable x86 CPU vectorization on windows [submodule sleef] #118980

xuhancn commented Feb 2, 2024 •

edited

pytorch-bot bot commented Feb 2, 2024 •

edited

ezyang commented Feb 6, 2024

pytorchmergebot commented Feb 6, 2024

xuhancn commented Feb 6, 2024

xuhancn commented Feb 6, 2024

pytorchmergebot commented Feb 6, 2024

pytorchmergebot commented Feb 6, 2024

ezyang commented Feb 6, 2024

pytorchmergebot commented Feb 6, 2024

xuhancn commented Mar 29, 2024

atalman commented Mar 30, 2024

atalman commented Mar 30, 2024

pytorchmergebot commented Mar 30, 2024

pytorchmergebot commented Mar 30, 2024

xuhancn commented Mar 30, 2024

pytorchmergebot commented Mar 30, 2024

pytorchmergebot commented Mar 30, 2024

xuhancn commented Mar 31, 2024

pytorchmergebot commented Mar 31, 2024

huydhn commented Mar 31, 2024

Enable x86 CPU vectorization on windows [submodule sleef] #118980

Enable x86 CPU vectorization on windows [submodule sleef] #118980

Conversation

xuhancn commented Feb 2, 2024 • edited

pytorch-bot bot commented Feb 2, 2024 • edited

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/118980

❌ 1 New Failure, 2 Unrelated Failures

ezyang commented Feb 6, 2024

pytorchmergebot commented Feb 6, 2024

Merge failed

xuhancn commented Feb 6, 2024

xuhancn commented Feb 6, 2024

pytorchmergebot commented Feb 6, 2024

Merge started

pytorchmergebot commented Feb 6, 2024

Merge failed

ezyang commented Feb 6, 2024

pytorchmergebot commented Feb 6, 2024

Merge started

xuhancn commented Mar 29, 2024

atalman commented Mar 30, 2024

atalman commented Mar 30, 2024

pytorchmergebot commented Mar 30, 2024

pytorchmergebot commented Mar 30, 2024

xuhancn commented Mar 30, 2024

pytorchmergebot commented Mar 30, 2024

Merge started

pytorchmergebot commented Mar 30, 2024

Merge failed

xuhancn commented Mar 31, 2024

pytorchmergebot commented Mar 31, 2024

Merge started

huydhn commented Mar 31, 2024

xuhancn commented Feb 2, 2024 •

edited

pytorch-bot bot commented Feb 2, 2024 •

edited