Repository navigation
Conversation
kekaczma
force-pushed
the
fix-launch-bounds-lambda-attrs
branch
from
October 15, 2026 19:23
c2237a4 to
c90d585
Compare
… on kernel lambdas Fixes intel#23322, two problems with expressing CUDA's __launch_bounds__ in SYCL that were discovered when revisiting the guidance from intel#8080: 1. intel::min_work_groups_per_cu and intel::max_work_groups_per_mp lacked SupportsNonconformingLambdaSyntax, so placing them on a kernel lambda (the usual SYCL attribute placement, after the parameter list) was rejected with "attribute cannot be applied to types", even though their documentation says they apply to "a device function/lambda function". 2. intel::max_work_group_size was removed as an FPGA attribute by PR intel#21864, but on NVPTX/AMDGPU it was also the documented way to set maxntid launch bounds. Existing code using it now only gets a generic unknown-attribute warning and silently loses the launch bound. Add a dedicated diagnostic, gated to SYCL device compilation, that points users at the max_work_group_size kernel property from sycl_ext_oneapi_kernel_properties instead.
bratpiorka
marked this pull request as ready for review
October 16, 2026 08:54
This branch has not been deployed
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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 two issues reported in #23322 with expressing CUDA launch bounds in SYCL:
intel::min_work_groups_per_cuandintel::max_work_groups_per_mpattributes on kernel lambdas using SYCL's attribute placement syntax.intel::max_work_group_sizeattribute, directing users to themax_work_group_sizekernel property instead.Add regression tests for kernel-lambda attribute placement and the new diagnostic.