Using kernel specific max work group size instead of device max work … #542
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Outdated suggestions cannot be applied.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.
…group size.
Max work group size of kernel is not a static and device related only property. It now in SYCL depends on driver/compiler implementation. Device max work group means the probable max work group size allowd by the device. But actual max work group size depends on driver/compiler implementation, like compilaton optimization. Using kernel specific max work group size could get actual max work group allowed correctly. For example, on Xe, if compiler chooses SIMD16 and large GRF (32 HW threads per SS), the actual max work group size will be 512 (16 * 32), not 1024 queried by device::info::max_work_group_size.