QR operator utilizing XPU. #2399
Open
+193
−1
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
Contributor
|
Contributor
|
There were no tests on the skip lists, as QR was silently falling back to CPU. Maybe after removing it, some tests will start to fail now |
CuiYifeng
requested changes
Nov 28, 2025
Contributor
CuiYifeng
left a comment
CuiYifeng
left a comment
There was a problem hiding this comment.
Please check comments and fix related failed cases.
Comment on lines
+69
to
+83
| TORCH_IMPL_FUNC(linalg_qr_xpu_out)(const Tensor& A, | ||
| std::string_view mode, | ||
| const Tensor & Q, | ||
| const Tensor & R) { | ||
| #if defined(USE_ONEMKL_XPU) | ||
| xpu::linalg_qr_kernel(A, mode, Q, R); | ||
| #else | ||
| auto A_cpu = A.to(at::kCPU); | ||
| auto Q_cpu = at::empty_like(Q, at::kCPU); | ||
| auto R_cpu = at::empty_like(R, at::kCPU); | ||
| at::cpu::linalg_qr_out(Q_cpu, R_cpu, A_cpu, mode); | ||
| Q.copy_(Q_cpu); | ||
| R.copy_(R_cpu); | ||
| #endif // USE_ONEMKL_XPU | ||
| } |
Contributor
There was a problem hiding this comment.
My suggestion is to register geqrf_kerenl_xpu/orgqr_kernel_xpu to geqrf_stub/orgqr_stub, which allows us to reuse op level code in stock Pytorch and reuse these two kernels in future.
Comment on lines
+9446
to
+9456
| - func: linalg_qr(Tensor A, str mode='reduced') -> (Tensor Q, Tensor R) | ||
| python_module: linalg | ||
| variants: function | ||
| structured_delegate: linalg_qr.out | ||
|
|
||
| - func: linalg_qr.out(Tensor A, str mode='reduced', *, Tensor(a!) Q, Tensor(b!) R) -> (Tensor(a!) Q, Tensor(b!) R) | ||
| python_module: linalg | ||
| structured: True | ||
| dispatch: | ||
| XPU: linalg_qr_xpu_out | ||
|
|
Contributor
There was a problem hiding this comment.
Please refer to https://github.com/pytorch/pytorch/blob/main/aten/src/ATen/native/native_functions.yaml#L14623 and consider stub mentioned above.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 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.
fixes #1900. The implementation utilizes two mkl::lapack libraries,
geqrf and geqrf for recovering pure Q matrix. Since torch and lapack use different storage formats, a hard transposition (memory layout not only stride ) was necessary. The iteraion over batch utilizes internal memory layout of processed data.