Repository navigation
Release GIL when calling mkl_lapack::orgqr - #2850
Merged
Merged
Conversation
Contributor
|
View rendered docs @ https://intelpython.github.io/dpnp/index.html |
Contributor
|
Array API standard conformance tests for dpnp=0.20.0dev6=py313h509198e_13 ran successfully. |
Collaborator
antonwolfy
marked this pull request as ready for review
April 13, 2026 20:37
antonwolfy
requested review from
ndgrigorian and
vlad-perevezentsev
as code owners
April 13, 2026 20:37
Co-authored-by: vlad-perevezentsev <vladislav.perevezentsev@intel.com>
vlad-perevezentsev
self-requested a review
April 14, 2026 14:00
vlad-perevezentsev
approved these changes
Apr 14, 2026
vlad-perevezentsev
left a comment
Contributor
There was a problem hiding this comment.
LGTM
Thank you @antonwolfy
github-actions Bot
added a commit
that referenced
this pull request
Apr 21, 2026
This PR fixes a deadlock in QR decomposition tests by adding GIL release
before `mkl_lapack::orgqr` call.
The hang occurred because:
1. dpnp/dpctl submits a `host_task` to manage Python object lifetimes
2. host_task needs to acquire the GIL to decrement reference counts
3. if the main thread holds the GIL during queue submission → deadlock
4. `orgqr` is currently implemented in oneMKL as GPU-to-Host reverse
offload:
```cpp
exec_q.submit([&](sycl::handler& cgh) {
cgh.depends_on(depends);
cgh.host_task([=]() { orgqr_host(...); });
}).wait();
```
As a solution PR proposes to release GIL using before calling the OneMKL
operations. The GIL is automatically reacquired when the function
returns (RAII). 3ab1216
5 of 7 tasks
abagusetty
added a commit
to abagusetty/dpnp
that referenced
this pull request
Aug 14, 2026
potrf, getrf, geqrf, gesv, gesvd, heevd, syevd, ungqr and the BLAS dot/gemm family block the calling thread for the whole oneMKL call while holding the GIL. That stalls every other Python thread, including the host tasks dpnp queues to manage object lifetimes, which reacquire the GIL to drop references. Completes the backport of IntelPythongh-2850, which applied the same fix to orgqr.cpp; gesv_batch, gesvd_batch, heevd_batch and syevd_batch already had it.
antonwolfy
pushed a commit
that referenced
this pull request
Aug 18, 2026
potrf, getrf, geqrf, gesv, gesvd, heevd, syevd, ungqr and the BLAS dot/gemm family block the calling thread for the whole oneMKL call while holding the GIL. That stalls every other Python thread, including the host tasks dpnp queues to manage object lifetimes, which reacquire the GIL to drop references. Completes the backport of #2850, which applied the same fix to orgqr.cpp; gesv_batch, gesvd_batch, heevd_batch and syevd_batch already had it.
github-actions Bot
added a commit
that referenced
this pull request
Aug 18, 2026
potrf, getrf, geqrf, gesv, gesvd, heevd, syevd, ungqr and the BLAS dot/gemm family block the calling thread for the whole oneMKL call while holding the GIL. That stalls every other Python thread, including the host tasks dpnp queues to manage object lifetimes, which reacquire the GIL to drop references. Completes the backport of #2850, which applied the same fix to orgqr.cpp; gesv_batch, gesvd_batch, heevd_batch and syevd_batch already had it. 7a2714b
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
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.
This PR fixes a deadlock in QR decomposition tests by adding GIL release before
mkl_lapack::orgqrcall.The hang occurred because:
host_taskto manage Python object lifetimesorgqris currently implemented in oneMKL as GPU-to-Host reverse offload:exec_q.submit([&](sycl::handler& cgh) { cgh.depends_on(depends); cgh.host_task([=]() { orgqr_host(...); }); }).wait();As a solution PR proposes to release GIL using before calling the OneMKL operations. The GIL is automatically reacquired when the function returns (RAII).