[inductor][cpp] BF16 AMX micro-gemm support #127195
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
[ghstack-poisoned]
pytorch-bot
bot
added
the
module: cpu
This was referenced May 27, 2024
jgong5
changed the title
Collaborator
Author
@jansel Thanks for the comment. The existing UTs already cover this well when they run on the CPUs that support AMX. Now I added a new one to make sure AMX micro-gemm is selected when the CPU has the support. Please review. |
jansel
approved these changes
Jun 20, 2024
Collaborator
Author
|
@pytorchbot merge |
Collaborator
Merge startedYour 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 |
Collaborator
|
The merge job was canceled or timed out. This most often happen if two merge requests were issued for the same PR, or if merge job was waiting for more than 6 hours for tests to finish. In later case, please do not hesitate to reissue the merge command |
Collaborator
Author
|
@pytorchbot merge |
Collaborator
Merge startedYour 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 |
Collaborator
|
Thanks @xwang233 for catching this, looks like this is causing the nightly build for ARM to fail since 6/22 - https://github.com/pytorch/pytorch/actions/runs/9623875645/job/26546874031. |
pytorchmergebot
pushed a commit
that referenced
this pull request
Jun 25, 2024
This PR fixes the build error on s390 after #127195. The following is the log of the build on s390x. This is because `SYS_arch_prctl` is not defined on s390x. ``` ... [792/2147] Building CXX object caffe2/CMakeFiles/torch_cpu.dir/__/aten/src/ATen/cpu/FlushDenormal.cpp.o [793/2147] Building CXX object caffe2/CMakeFiles/torch_cpu.dir/__/aten/src/ATen/cpu/Utils.cpp.o FAILED: caffe2/CMakeFiles/torch_cpu.dir/__/aten/src/ATen/cpu/Utils.cpp.o /usr/bin/c++ -DAT_PER_OPERATOR_HEADERS -DCAFFE2_BUILD_MAIN_LIB -DFLASHATTENTION_DISABLE_ALIBI -DFMT_HEADER_ONLY=1 -DHAVE_MALLOC_USABLE_SIZE=1 -DHAVE_MMAP=1 -DHAVE_SHM_OPEN=1 -DHAVE_SHM_UNLINK=1 -DMINIZ_DISABLE_ZIP_READER_CRC32_CHECKS -DONNXIFI_ENABLE_EXT=1 -DONNX_ML=1 -DONNX_NAMESPACE=onnx_torch -DUSE_DISTRIBUTED -DUSE_EXTERNAL_MZCRC -DUSE_RPC -DUSE_TENSORPIPE -D_FILE_OFFSET_BITS=64 -Dtorch_cpu_EXPORTS -I/pytorch/build/aten/src -I/pytorch/aten/src -I/pytorch/build -I/pytorch -I/pytorch/cmake/../third_party/benchmark/include -I/pytorch/third_party/onnx -I/pytorch/build/third_party/onnx -I/pytorch/third_party/foxi -I/pytorch/build/third_party/foxi -I/pytorch/torch/csrc/api -I/pytorch/torch/csrc/api/include -I/pytorch/caffe2/aten/src/TH -I/pytorch/build/caffe2/aten/src/TH -I/pytorch/build/caffe2/aten/src -I/pytorch/build/caffe2/../aten/src -I/pytorch/torch/csrc -I/pytorch/third_party/miniz-2.1.0 -I/pytorch/third_party/kineto/libkineto/include -I/pytorch/third_party/kineto/libkineto/src -I/pytorch/third_party/cpp-httplib -I/pytorch/aten/src/ATen/.. -I/pytorch/c10/.. -I/pytorch/third_party/FP16/include -I/pytorch/third_party/tensorpipe -I/pytorch/build/third_party/tensorpipe -I/pytorch/third_party/tensorpipe/third_party/libnop/include -I/pytorch/third_party/fmt/include -I/pytorch/third_party/flatbuffers/include -isystem /pytorch/cmake/../third_party/tensorpipe/third_party/libuv/include -isystem /pytorch/cmake/../third_party/googletest/googlemock/include -isystem /pytorch/cmake/../third_party/googletest/googletest/include -isystem /pytorch/third_party/protobuf/src -isystem /pytorch/cmake/../third_party/eigen -isystem /pytorch/build/include -Wno-maybe-uninitialized -Wno-uninitialized -Wno-free-nonheap-object -Wno-nonnull -D_GLIBCXX_USE_CXX11_ABI=1 -fvisibility-inlines-hidden -DNDEBUG -DUSE_KINETO -DLIBKINETO_NOCUPTI -DLIBKINETO_NOROCTRACER -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -O2 -fPIC -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Werror=range-loop-construct -Werror=bool-operation -Wnarrowing -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wno-unused-parameter -Wno-unused-function -Wno-unused-result -Wno-strict-overflow -Wno-strict-aliasing -Wno-stringop-overflow -Wsuggest-override -Wno-psabi -Wno-error=pedantic -Wno-error=old-style-cast -Wno-missing-braces -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Wno-stringop-overflow -O3 -DNDEBUG -DNDEBUG -fPIC -DTORCH_USE_LIBUV -Wall -Wextra -Wdeprecated -Wno-unused-parameter -Wno-missing-field-initializers -Wno-unknown-pragmas -Wno-type-limits -Wno-array-bounds -Wno-strict-overflow -Wno-strict-aliasing -Wunused-function -Wno-maybe-uninitialized -fvisibility=hidden -O2 -fopenmp -MD -MT caffe2/CMakeFiles/torch_cpu.dir/__/aten/src/ATen/cpu/Utils.cpp.o -MF caffe2/CMakeFiles/torch_cpu.dir/__/aten/src/ATen/cpu/Utils.cpp.o.d -o caffe2/CMakeFiles/torch_cpu.dir/__/aten/src/ATen/cpu/Utils.cpp.o -c /pytorch/aten/src/ATen/cpu/Utils.cpp /pytorch/aten/src/ATen/cpu/Utils.cpp: In function 'bool at::cpu::init_amx()': /pytorch/aten/src/ATen/cpu/Utils.cpp:60:21: error: 'SYS_arch_prctl' was not declared in this scope; did you mean 'SYS_prctl'? 60 | long rc = syscall(SYS_arch_prctl, ARCH_REQ_XCOMP_PERM, XFEATURE_XTILEDATA); | ^~~~~~~~~~~~~~ | SYS_prctl [794/2147] Building CXX object caffe2/CMakeFiles/torch_cpu.dir/__/aten/src/ATen/native/Integration.cpp.o [795/2147] Building CXX object caffe2/CMakeFiles/torch_cpu.dir/__/aten/src/ATen/native/GridSampler.cpp.o [796/2147] Building CXX object caffe2/CMakeFiles/torch_cpu.dir/__/aten/src/ATen/detail/CPUGuardImpl.cpp.o [797/2147] Building CXX object caffe2/CMakeFiles/torch_cpu.dir/__/aten/src/ATen/ThreadLocalState.cpp.o [798/2147] Building CXX object caffe2/CMakeFiles/vec_test_all_types_DEFAULT.dir/__/aten/src/ATen/test/vec_test_all_types.cpp.o [799/2147] Building CXX object caffe2/CMakeFiles/torch_cpu.dir/__/aten/src/ATen/Utils.cpp.o [800/2147] Building CXX object caffe2/CMakeFiles/torch_cpu.dir/__/aten/src/ATen/VmapModeRegistrations.cpp.o [801/2147] Building CXX object caffe2/CMakeFiles/torch_cpu.dir/__/aten/src/ATen/ZeroTensorFallback.cpp.o [802/2147] Building CXX object caffe2/CMakeFiles/torch_cpu.dir/__/aten/src/ATen/autocast_mode.cpp.o ninja: build stopped: subcommand failed. Building wheel torch-2.5.0a0+git94dc325 -- Building version 2.5.0a0+git94dc325 cmake -GNinja -DBUILD_CAFFE2=0 -DBUILD_PYTHON=True -DBUILD_TEST=True -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/pytorch/torch -DCMAKE_PREFIX_PATH=/usr/local/lib/python3.10/dist-packages -DPython_EXECUTABLE=/usr/bin/python3 -DTORCH_BUILD_VERSION=2.5.0a0+git94dc325 -DUSE_GLOO=0 -DUSE_NUMPY=True /pytorch cmake --build . --target install --config Release Build step 'Execute shell' marked build as failure ... ``` Pull Request resolved: #129326 Approved by: https://github.com/Skylion007, https://github.com/eqy
Collaborator
|
aarch64 CI builds are also failing. Instead of adding (!defined) for every variant of arm, could someone please guard x86_64 AMX specific changes with an appropriate macro. |
pytorchmergebot
pushed a commit
that referenced
this pull request
Jun 25, 2024
Trying to mitigate aarch64 and s390 nightly failures as per this comment: #127195 (comment) Fixes #129443 Pull Request resolved: #129479 Approved by: https://github.com/nWEIdia, https://github.com/malfet
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Stack from ghstack (oldest at bottom):
This PR adds the intrinsics based micro-gemm for BF16 using Advanced Matrix eXtension (AMX) instructions available in Intel 4th and 5th Xeon processors. A compilation check is added to
codecache.pyto check the validity of the compiler support. Also, since AMX requires an initialization in the Linux kernel to extra register states, an initialization function is added to do that and triggered viacodecache.py.Performance speedups with >=10% on BF16 AMP, max_autotune vs. no autotune, measured on Intel(R) Xeon(R) Platinum 8488C:
Static shapes
Single-threaded
Multi-threaded
Dynamic shapes
Single-threaded
Multi-threaded
No perf regressions.
cc @mingfeima @XiaobingSuper @sanchitintel @ashokei @jingxu10 @voznesenskym @penguinwu @EikanWang @Guobing-Chen @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @peterbell10 @ipiszy @yf225 @chenyang78 @kadeng @muchulee8 @ColinPeppler @amjames @desertfire @chauhang