3rdparty(mlas): add missing power (ppc64le) kernel headers #29516
Closes: https://github.com/opencv/opencv/issues/29465
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
The previous commit bdf348c13a introduced
MLAS support including the s390x backend, but accidentally omitted the
required internal header files. This causes a compilation failure on
s390x:
fatal error: SgemmKernelZVECTOR.h: No such file or directory
This PR restores the build by adding the following missing headers from
upstream (microsoft/onnxruntime):
- FgemmKernelZVECTOR.h
- SgemmKernelZVECTOR.h
This is the same class of bug as #29422 (loongarch64), which was fixed
by PR #29423 for that architecture only.
MlasThreadedBufAlloc() and its ThreadedBufHolder guarded the
_aligned_malloc / _aligned_free path on _MSC_VER. Non-MSVC Windows
toolchains (MinGW, clang) therefore fell through to posix_memalign(),
which the Windows CRT does not provide, so the build failed with
"'posix_memalign' was not declared in this scope".
Guard the aligned-allocation path on _WIN32 instead, in both the holder
declaration (mlasi.h) and definition (platform.cpp) so the unique_ptr
deleter type stays consistent across translation units, and include
<malloc.h> on Windows since MinGW only declares the _aligned_* functions
there. The non-Windows posix_memalign / aligned_alloc fallback is
unchanged.
Fixes#29350
Fix MLAS 32-bit x86 build by integrating missing upstream assembly files #29226
### Pull Request Readiness Checklist
This resolves the 32-bit x86 build failure for MLAS. Related to: https://github.com/opencv/opencv/pull/29218
* Added the required `x86` assembly files and headers from upstream.
* Added `__x86_64__` guards around the AMX `syscall` and the FMA3/AVX512F kernel assignments to prevent 32-bit compilation crashes.
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
Attention graph fusion with MLAS FlashAttention #29126
Performance numbers for Owl-v2 model on intel i9:
```
ORT: Average inference time over 10 runs: 1411.55 ms (min 1399.75, max 1438.89)
NEW: Average inference time over 10 runs: 1078 ms (min 1048.04, max 1110.61)
```
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
Added MLAS third party module and integrated into GeMM path #28934
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
Alexander Smorkalov
Abhishek Gola
Anu Oguntayo
吴小白
vrooomy
ShivaPriyanShanmuga
omrope79
Alexander Smorkalov