fengqing.lu@intel.com 7b074346e0 [Intel GPU] Support f32 intermediate dtype, headdim size <=576 and f32 causal mask for SDPA (#152091) ...

In OneDNN v3.7, SDPA has below defects:

1. The dtype of intermediate value is the same as QKV, while Pytorch uses FP32 dtype for intermediate value to make sure better accuracy.
2. Only support headdim size <= 256.
3. Don't support implict causal mask when QKV is FP32. We need to build an attention mask explicitly with aten ops.

In OneDNN v3.8, they have update for these defects. Since these are tiny changes, I decided to put them in single PR.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/152091
Approved by: https://github.com/EikanWang, https://github.com/guangyey, https://github.com/drisspg

2025-06-04 05:18:36 +00:00

..

External

Fix broken URLs (#152237)

2025-04-27 09:56:42 +00:00

Modules

[Intel GPU] Support f32 intermediate dtype, headdim size <=576 and f32 causal mask for SDPA (#152091)

2025-06-04 05:18:36 +00:00

Modules_CUDA_fix

Add scripts to check xrefs and urls (#151844)

2025-04-28 09:30:07 +00:00

public

ROCm Sparsity through HipSparseLT (#150578)

2025-05-31 02:03:40 +00:00

Allowlist.cmake

Replace whitelist with allowlist (#42067)

2020-07-28 08:01:16 -07:00

BuildVariables.cmake

Remove Caffe2_MAIN_LIBS (#38408)

2020-05-15 12:27:15 -07:00

Caffe2Config.cmake.in

xpu: improve error handling and reporting in XPU cmake files (#149353)

2025-03-20 02:00:39 +00:00

cmake_uninstall.cmake.in

Formatting cmake (to lowercase without space for if/elseif/else/endif) (#35521)

2020-03-27 14:25:17 -07:00

Codegen.cmake

Fix Codegen.cmake warning (#153023)

2025-05-07 12:45:20 +00:00

DebugHelper.cmake

Add debug helper function to check target property (#52093)

2021-02-11 15:37:14 -08:00

Dependencies.cmake

ROCm Sparsity through HipSparseLT (#150578)

2025-05-31 02:03:40 +00:00

FlatBuffers.cmake

[pytorch][PR] Add ability for a mobile::Module to save as flatbuffer (#70201)

2022-01-12 16:30:39 -08:00

IncludeSource.cpp.in

CMake: Include instead of copying cpu kernel files (#67656)

2021-11-30 19:13:53 -08:00

iOS.cmake

[executorch] Update iOS toolchain with a modern cmake syntax. (#115799)

2023-12-15 00:51:30 +00:00

Metal.cmake

[MPS] Support includes in metal objects (#145087)

2025-01-18 05:35:22 +00:00

MiscCheck.cmake

Remove CAFFE2_USE_EXCEPTION_PTR (#147247)

2025-03-06 02:56:23 +00:00

prioritized_text.txt

[Build] Add linker script optimization (#121975)

2024-04-09 20:22:25 +00:00

ProtoBuf.cmake

Revert "Use 3.27 as the minimum CMake version (#153153)"

2025-05-31 02:14:24 +00:00

ProtoBufPatch.cmake

Migrate PyTorch to C++17 (#85969)

2022-12-08 02:27:48 +00:00

Summary.cmake

Turn on compile with NVSHMEM (#154538)

2025-06-03 15:24:24 +00:00

TorchConfig.cmake.in

[Build] Remove pre-CXX11 ABI logic from build script (#149888)

2025-03-25 03:17:16 +00:00

TorchConfigVersion.cmake.in

Formatting cmake (to lowercase without space for if/elseif/else/endif) (#35521)

2020-03-27 14:25:17 -07:00

VulkanCodegen.cmake

[BE][CMake] Use FindPython module (#124613)

2024-05-29 13:17:35 +00:00

VulkanDependencies.cmake

[Vulkan] Remove GLSL Code Gen (#91912)

2023-01-10 20:29:47 +00:00