[SYCL] do not set zeroed GlobalOffset in kernel parameters #20082
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
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
bratpiorka
force-pushed
the
rrudnick_kernel_offset_opt
branch
from
3ab05aa to
2b0efb7
Compare
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
bratpiorka
force-pushed
the
rrudnick_kernel_offset_opt
branch
from
2b0efb7 to
d89391f
Compare
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
|
for "SYCL :: Basic/submit_time.cpp" fail there is an issue #20248 @intel/sycl-graphs-reviewers @intel/llvm-reviewers-runtime please review |
Comment on lines
2613
to
2615
| const bool HasOffset = NDRDesc.GlobalOffset[0] != 0 || | ||
| NDRDesc.GlobalOffset[1] != 0 || | ||
| NDRDesc.GlobalOffset[2] != 0; |
There was a problem hiding this comment.
Wouldn't this be dependent on the dimensionality? I.e. say I have a 2-dimensional launch, wouldn't the offset be zero of either of the first 2 dimensions are 0? Also, in a case like that, it shouldn't matter what the 3rd dimension is, right?
What I am suggesting is basically:
Suggested change
| const bool HasOffset = NDRDesc.GlobalOffset[0] != 0 || | |
| NDRDesc.GlobalOffset[1] != 0 || | |
| NDRDesc.GlobalOffset[2] != 0; | |
| const bool HasOffset = NDRDesc.GlobalOffset[0] != 0 && | |
| (NDRDesc.Dims < 2 || NDRDesc.GlobalOffset[1] != 0) && | |
| (NDRDesc.Dims < 3 || NDRDesc.GlobalOffset[2] != 0); |
There was a problem hiding this comment.
right - applied in both places where "HasOffset" is calculated and additionally to the "EnforcedLocalSize" flag, which uses the same code pattern
bratpiorka
force-pushed
the
rrudnick_kernel_offset_opt
branch
from
d89391f to
f068bce
Compare
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
bratpiorka
had a problem deploying
to
WindowsCILock
GitHub Actions
Failure
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
bratpiorka
had a problem deploying
to
WindowsCILock
GitHub Actions
Failure
bratpiorka
had a problem deploying
to
WindowsCILock
GitHub Actions
Failure
bratpiorka
had a problem deploying
to
WindowsCILock
GitHub Actions
Failure
mmichel11
reviewed
Oct 2, 2025
| Adapter.call_nocheck<UrApiKind::urCommandBufferAppendKernelLaunchExp>( | ||
| CommandBuffer, UrKernel, NDRDesc.Dims, &NDRDesc.GlobalOffset[0], | ||
| CommandBuffer, UrKernel, NDRDesc.Dims, | ||
| HasOffset ? &NDRDesc.GlobalOffset[0] : nullptr, |
There was a problem hiding this comment.
The specification of urCommandBufferAppendKernelLaunchExp seems to suggest that passing nullptr to the global offset should return UR_RESULT_ERROR_INVALID_NULL_POINTER. Does the UR implementation just not align with the specification here?
There was a problem hiding this comment.
You're right—I've updated the specification, so now global offset is optional. This is consistent with EnqueueKernelLaunch, where this parameter is also optional.
bratpiorka
had a problem deploying
to
WindowsCILock
GitHub Actions
Failure
bratpiorka
had a problem deploying
to
WindowsCILock
GitHub Actions
Failure
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
bratpiorka
force-pushed
the
rrudnick_kernel_offset_opt
branch
from
f068bce to
6b54f63
Compare
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
pbalcer
approved these changes
Oct 6, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 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.
Do not set a zeroed GlobalOffset in kernel parameters - the Unified Runtime layer will handle this correctly.
Note that there is already a similar optimization in https://github.com/intel/llvm/blob/sycl/sycl/source/detail/scheduler/commands.cpp#L2477