Add mkl::linalg_svd relative Ops #1511
Conversation
yucai-intel
commented
Mar 26, 2025
•
yucai-intel
commented
- linalg_svd
- linalg_svd.U
|
Test cases related to complex data types fail because onednn does not support it. |
yucai-intel
changed the title
There was a problem hiding this comment.
Please reactivate some skipped cases such as https://github.com/intel/torch-xpu-ops/blob/main/test/xpu/skip_list_common.py#L1397. Otherwise, CI cannot test with these svd cases.
CuiYifeng
force-pushed
the
slogdet
branch
4 times, most recently
from
56cebb3 to
9ac4024
Compare
CuiYifeng
force-pushed
the
slogdet
branch
2 times, most recently
from
e13dc4f to
da85fe6
Compare
There was a problem hiding this comment.
Pull Request Overview
This PR adds support for MKL-based SVD operations on the XPU backend, including new composite functions and updates to native function definitions for linalg_svd and related ops. Key changes include:
- New YAML definitions for _linalg_svd and linalg_svd (and their variants) exposing the compute_uv flag.
- Removal of several SVD-related skip tests in the xpu test skip list.
- New MKL-based implementation files for BatchLinearAlgebra (header and source) and updated XPU dispatch registration.
Reviewed Changes
Copilot reviewed 5 out of 6 changed files in this pull request and generated 1 comment.
Show a summary per file
| File | Description |
|---|---|
| yaml/native/native_functions.yaml | Added definitions for new SVD ops and composite functions. |
| test/xpu/skip_list_common.py | Removed skip tests for SVD float64 variants on XPU. |
| src/ATen/native/xpu/mkl/BatchLinearAlgebra.h | Declared the svd_mkl prototype. |
| src/ATen/native/xpu/mkl/BatchLinearAlgebra.cpp | Implemented the svd_mkl function using oneMKL calls. |
| src/ATen/native/xpu/BatchLinearAlgebra.cpp | Registered the XPU dispatch for the SVD kernel. |
Files not reviewed (1)
- src/ATen/native/xpu/XPUFallback.template: Language not supported
| @@ -9350,3 +9350,22 @@ | |||
| variants: function | |||
| dispatch: | |||
| XPU: _fft_r2c_xpu_out | |||
|
|
|||
| # This function is exposes the `compute_uv` flag, which is then used to implement `linalg.svd` and | |||
There was a problem hiding this comment.
Consider correcting the grammatical error in the comment. Perhaps change 'is exposes' to 'exposes'.
| # This function is exposes the `compute_uv` flag, which is then used to implement `linalg.svd` and | |
| # This function exposes the `compute_uv` flag, which is then used to implement `linalg.svd` and |
Copilot uses AI. Check for mistakes.
