Always use size 16 sub-groups in single work-group radix sort if supported #1833
Conversation
This file contains 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
This change is added to avoid a bug where IGC cannot compile SIMD32 kernels with -O0 compilation flags. No performance impact is observed. Signed-off-by: Matthew Michel <matthew.michel@intel.com>
Signed-off-by: Matthew Michel <matthew.michel@intel.com>
…ely" This reverts commit 7572f84. After experimentation with cold cache benchmarks, benefit of up to ~15% was observed requiring size 16 sub-groups for the larger single work-group cases. IGC compiles the kernels with sub-group size 32 here so leaving the requirement is needed to maximize performance.
|
There is a comment on line 819 of |
Signed-off-by: Matthew Michel <matthew.michel@intel.com>
danhoeflinger
approved these changes
Sep 11, 2024
There was a problem hiding this comment.
LGTM, but I'd like @MikeDvorskiy to also be able to look at this.
I made a few tweaks to the comment to match the new behavior. |
MikeDvorskiy
approved these changes
Sep 13, 2024
This pull request was closed.
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Outdated suggestions cannot be applied.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.
Previously, an IGC workaround has caused kernels compiled with
[[sycl::reqd_sub_group_size(32)]]to be silently compiled with size 16 sub-groups with-O0on certain devices. To comply with the SYCL spec, this case will now throw a synchronous exception at JIT time.Single work-group radix sort uses a sub-group sizes of 32 for the smallest inputs and 16 for larger inputs to minimize register pressure. I have experimented with the current mainline version, a version only using sub-group size of 16, and a version removing the requirement altogether. The first two versions performed similarly with no measurable difference, and explicitly requiring a sub-group size of 16 offered up to a ~15% benefit over the version with no requirement at all.
By using only sub-group sizes of 16 in single work-group radix sort, we are able to avoid the IGC issue while not impacting performance.