Repository navigation
State the roofline target in the benchmark-kernel skill - #401
Merged
Merged
Conversation
Every result reports %-of-roofline per arch against the ceiling for the kernel's dtype and board. rocm_profiler's KNOWN_HW holds only the fp16 peak and HBM bandwidth of one SKU per arch, so fp8 and MX formats need scaling, and MI325X and MI355X need their own peaks. gfx942 and gfx950 are optimised separately where the hardware differs: the backend choice goes in arch_caps, and tile and launch choices go in the kernel's own arch dispatch. Co-Authored-By: Claude <noreply@anthropic.com>
Security scan found no critical or high findings on lines changed by this pull requestView the security scan run for complete reports. “New” means the finding is located in a changed file on an added or modified line; it is not a base-versus-head findings delta. |
There was a problem hiding this comment.
🟢 Approval recommended
No active review issues are recorded for the documentation-only update to roofline reporting and per-architecture tuning guidance.
No open issues 🎉
What changed in this PR
This documentation-only PR defines per-architecture roofline targets for kernel optimization in the benchmark skill.
Changes:
- Requires results as a percentage of the roofline, using dtype- and board-specific ceilings.
- Directs separate gfx942 and gfx950 optimization where hardware differs, with backend selection and kernel tuning handled in their respective locations.
- Defines completion as approaching the ceiling or identifying the cause of the remaining gap.
| File | Description |
|---|---|
.claude/skills/benchmark-kernel/SKILL.md |
Adds roofline reporting requirements, per-architecture tuning guidance, and a stopping criterion. |
Review effort: Balanced
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.
Summary
States the performance target for kernel work in the
benchmark-kernelskill: every result is reported as %-of-roofline on each arch, and gfx942 and gfx950 are optimised separately wherever their hardware differs. A speedup alone does not say how much headroom is left.What changed
.claude/skills/benchmark-kernel/SKILL.md— new "The target is the roofline, per arch" section:rocm_profiler'sKNOWN_HWholds only the fp16 matrix peak and HBM bandwidth of one SKU per arch (MI300X, MI350X), so fp8/MX need scaling and MI325X/MI355X need their own peaks.arch_caps.py; tile and launch choices go in the kernel's own arch dispatch.Benchmark results
Not applicable: documentation only.
Test plan
pre-commit run --from-ref origin/amd-integration --to-ref HEAD