Extend DispatchStub to support CUDA dispatch #9579
Closed
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This is part of pytorch#8919. It's separated to make it easier to merge the PR in pieces. There are a few major changes to DispatchStub - The environment variable ATEN_CPU_CAPABILITY overrides the CPU capability detection code (Previous ATEN_DISABLE_AVX/AVX2) - DispatchStub is defined in the generic native code instead of the CPU_CAPABILITY_DEFAULT kernel.
colesbury
requested review from
apaszke,
ezyang,
gchanan,
soumith and
zdevito
as code owners
Contributor
facebook-github-bot
left a comment
facebook-github-bot
left a comment
There was a problem hiding this comment.
@colesbury has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
gchanan
approved these changes
Jul 19, 2018
Contributor
facebook-github-bot
left a comment
facebook-github-bot
left a comment
There was a problem hiding this comment.
@colesbury has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
Contributor
facebook-github-bot
left a comment
facebook-github-bot
left a comment
There was a problem hiding this comment.
@colesbury has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
Contributor
|
If you're changing the capability variable name, please also edit ".jenkins/pytorch/test.sh" accordingly. |
zdevito
pushed a commit
to zdevito/ATen
that referenced
this pull request
Jul 19, 2018
Summary: This is a few files taken from pytorch/pytorch#8919. They're unchanged from the latest versions of that PR. ``` This is part of pytorch/pytorch#8919. It's separated to make it easier to merge the PR in pieces. There are a few major changes to DispatchStub - The environment variable ATEN_CPU_CAPABILITY overrides the CPU capability detection code (Previous ATEN_DISABLE_AVX/AVX2) - DispatchStub is defined in the generic native code instead of the CPU_CAPABILITY_DEFAULT kernel. ``` Pull Request resolved: pytorch/pytorch#9579 Differential Revision: D8909000 Pulled By: colesbury fbshipit-source-id: fdeb606270b06acdab3c01dba97ec9d81584ecc0
colesbury
added a commit
to colesbury/pytorch
that referenced
this pull request
Jul 20, 2018
This reverts commit bcf0bf4.
colesbury
added a commit
to colesbury/pytorch
that referenced
this pull request
Jul 21, 2018
This is a modification of the strategy from pytorch#8919 and pytorch#9579. Previously, the CPU architecture-specific kernels self-registered with the DispatchStub. When linking as part of a static library, this requires the flag --whole-archive to be passed to the linker to ensure that the object files for the kernels are included. Caffe2 and TensorFlow use that strategy. We ran into some issues with --whole-archive blowing up the binary size of some downstream projects in Facebook. This PR avoids --whole-archive for CPU kernels. The downside is that the generic code needs to be aware of whether kernels are compiled with AVX and with AVX2 (via HAVE_AVX_CPU_DEFINITION and HAVE_AVX2_CPU_DEFINITION). The CUDA kernels still self-register with DispatchStub because the CPU library is not aware of whether the CUDA library will be available at runtime. There are a few major changes to DispatchStub - The environment variable ATEN_CPU_CAPABILITY overrides the CPU capability detection code (Previous ATEN_DISABLE_AVX/AVX2) - DispatchStub is defined in the generic native code instead of the CPU_CAPABILITY_DEFAULT kernel.
facebook-github-bot
pushed a commit
that referenced
this pull request
Jul 23, 2018
Summary: This is a modification of the strategy from #8919 and #9579. ``` Previously, the CPU architecture-specific kernels self-registered with the DispatchStub. When linking as part of a static library, this requires the flag --whole-archive to be passed to the linker to ensure that the object files for the kernels are included. Caffe2 and TensorFlow use that strategy. We ran into some issues with --whole-archive blowing up the binary size of some downstream projects in Facebook. This PR avoids --whole-archive for CPU kernels. The downside is that the generic code needs to be aware of whether kernels are compiled with AVX and with AVX2 (via HAVE_AVX_CPU_DEFINITION and HAVE_AVX2_CPU_DEFINITION). The CUDA kernels still self-register with DispatchStub because the CPU library is not aware of whether the CUDA library will be available at runtime. There are a few major changes to DispatchStub - The environment variable ATEN_CPU_CAPABILITY overrides the CPU capability detection code (Previous ATEN_DISABLE_AVX/AVX2) - DispatchStub is defined in the generic native code instead of the CPU_CAPABILITY_DEFAULT kernel. ``` Pull Request resolved: #9664 Differential Revision: D8943350 Pulled By: colesbury fbshipit-source-id: 329229b0ee9ff94fc001b960287814bd734096ef
zdevito
pushed a commit
to zdevito/ATen
that referenced
this pull request
Jul 23, 2018
Summary: This is a modification of the strategy from pytorch/pytorch#8919 and pytorch/pytorch#9579. ``` Previously, the CPU architecture-specific kernels self-registered with the DispatchStub. When linking as part of a static library, this requires the flag --whole-archive to be passed to the linker to ensure that the object files for the kernels are included. Caffe2 and TensorFlow use that strategy. We ran into some issues with --whole-archive blowing up the binary size of some downstream projects in Facebook. This PR avoids --whole-archive for CPU kernels. The downside is that the generic code needs to be aware of whether kernels are compiled with AVX and with AVX2 (via HAVE_AVX_CPU_DEFINITION and HAVE_AVX2_CPU_DEFINITION). The CUDA kernels still self-register with DispatchStub because the CPU library is not aware of whether the CUDA library will be available at runtime. There are a few major changes to DispatchStub - The environment variable ATEN_CPU_CAPABILITY overrides the CPU capability detection code (Previous ATEN_DISABLE_AVX/AVX2) - DispatchStub is defined in the generic native code instead of the CPU_CAPABILITY_DEFAULT kernel. ``` Pull Request resolved: pytorch/pytorch#9664 Differential Revision: D8943350 Pulled By: colesbury fbshipit-source-id: 329229b0ee9ff94fc001b960287814bd734096ef
jramseyer
pushed a commit
to jramseyer/pytorch
that referenced
this pull request
Jul 30, 2018
Summary: This is a few files taken from pytorch#8919. They're unchanged from the latest versions of that PR. ``` This is part of pytorch#8919. It's separated to make it easier to merge the PR in pieces. There are a few major changes to DispatchStub - The environment variable ATEN_CPU_CAPABILITY overrides the CPU capability detection code (Previous ATEN_DISABLE_AVX/AVX2) - DispatchStub is defined in the generic native code instead of the CPU_CAPABILITY_DEFAULT kernel. ``` Pull Request resolved: pytorch#9579 Differential Revision: D8909000 Pulled By: colesbury fbshipit-source-id: fdeb606270b06acdab3c01dba97ec9d81584ecc0
jramseyer
pushed a commit
to jramseyer/pytorch
that referenced
this pull request
Jul 30, 2018
…ytorch#9614) Summary: This reverts commit bcf0bf4. The commit was causing issues for some internal FB projects. Pull Request resolved: pytorch#9614 Reviewed By: Yangqing Differential Revision: D8929552 Pulled By: colesbury fbshipit-source-id: ae9026ad8762a4c5de401273694b4c878fc241a6
jramseyer
pushed a commit
to jramseyer/pytorch
that referenced
this pull request
Jul 30, 2018
Summary: This is a modification of the strategy from pytorch#8919 and pytorch#9579. ``` Previously, the CPU architecture-specific kernels self-registered with the DispatchStub. When linking as part of a static library, this requires the flag --whole-archive to be passed to the linker to ensure that the object files for the kernels are included. Caffe2 and TensorFlow use that strategy. We ran into some issues with --whole-archive blowing up the binary size of some downstream projects in Facebook. This PR avoids --whole-archive for CPU kernels. The downside is that the generic code needs to be aware of whether kernels are compiled with AVX and with AVX2 (via HAVE_AVX_CPU_DEFINITION and HAVE_AVX2_CPU_DEFINITION). The CUDA kernels still self-register with DispatchStub because the CPU library is not aware of whether the CUDA library will be available at runtime. There are a few major changes to DispatchStub - The environment variable ATEN_CPU_CAPABILITY overrides the CPU capability detection code (Previous ATEN_DISABLE_AVX/AVX2) - DispatchStub is defined in the generic native code instead of the CPU_CAPABILITY_DEFAULT kernel. ``` Pull Request resolved: pytorch#9664 Differential Revision: D8943350 Pulled By: colesbury fbshipit-source-id: 329229b0ee9ff94fc001b960287814bd734096ef
goodlux
pushed a commit
to goodlux/pytorch
that referenced
this pull request
Aug 15, 2018
Summary: This is a few files taken from pytorch#8919. They're unchanged from the latest versions of that PR. ``` This is part of pytorch#8919. It's separated to make it easier to merge the PR in pieces. There are a few major changes to DispatchStub - The environment variable ATEN_CPU_CAPABILITY overrides the CPU capability detection code (Previous ATEN_DISABLE_AVX/AVX2) - DispatchStub is defined in the generic native code instead of the CPU_CAPABILITY_DEFAULT kernel. ``` Pull Request resolved: pytorch#9579 Differential Revision: D8909000 Pulled By: colesbury fbshipit-source-id: fdeb606270b06acdab3c01dba97ec9d81584ecc0
goodlux
pushed a commit
to goodlux/pytorch
that referenced
this pull request
Aug 15, 2018
…ytorch#9614) Summary: This reverts commit bcf0bf4. The commit was causing issues for some internal FB projects. Pull Request resolved: pytorch#9614 Reviewed By: Yangqing Differential Revision: D8929552 Pulled By: colesbury fbshipit-source-id: ae9026ad8762a4c5de401273694b4c878fc241a6
goodlux
pushed a commit
to goodlux/pytorch
that referenced
this pull request
Aug 15, 2018
Summary: This is a modification of the strategy from pytorch#8919 and pytorch#9579. ``` Previously, the CPU architecture-specific kernels self-registered with the DispatchStub. When linking as part of a static library, this requires the flag --whole-archive to be passed to the linker to ensure that the object files for the kernels are included. Caffe2 and TensorFlow use that strategy. We ran into some issues with --whole-archive blowing up the binary size of some downstream projects in Facebook. This PR avoids --whole-archive for CPU kernels. The downside is that the generic code needs to be aware of whether kernels are compiled with AVX and with AVX2 (via HAVE_AVX_CPU_DEFINITION and HAVE_AVX2_CPU_DEFINITION). The CUDA kernels still self-register with DispatchStub because the CPU library is not aware of whether the CUDA library will be available at runtime. There are a few major changes to DispatchStub - The environment variable ATEN_CPU_CAPABILITY overrides the CPU capability detection code (Previous ATEN_DISABLE_AVX/AVX2) - DispatchStub is defined in the generic native code instead of the CPU_CAPABILITY_DEFAULT kernel. ``` Pull Request resolved: pytorch#9664 Differential Revision: D8943350 Pulled By: colesbury fbshipit-source-id: 329229b0ee9ff94fc001b960287814bd734096ef
pruthvistony
pushed a commit
to ROCm/pytorch
that referenced
this pull request
Oct 28, 2024
…code to disable warnings for ROCm builds (#1642) Fixes pytorch#9579
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This is a few files taken from #8919. They're unchanged from the latest versions of that PR.