Add gfx950 BF16 GEMM tuning configs (16 shapes, 203 configs) - #55
Merged
Merged
Conversation
Tuned 203 configs (16 unique N,K shapes × 13 batch sizes) on MI355X using GemmTuner with --mp 8 --iters 200 --warmup 50. Shapes cover: DeepSeek-R1 (kv_proj, o_proj), GPT-OSS-120B, Qwen3-30B-A3B, Qwen3.5-35B-A3B, Qwen3-Next-80B, LFM-2-24B, small MoE gate projections. Best-of-3 backend selection (ASM vs hipBLASLt vs Triton) shows 23.6% geomean improvement over hipBLASLt alone across 826 tested configs. TODO: Remaining 76 (N,K) shapes need tuning (tuner hung after 16 shapes). Priority: N=28672 (Llama gate_up), N=36864 (DeepSeek gate_up), N=106496 (405B). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
🏷️ CI GuideRuns automatically on every PR:
Extended tests (opt-in via labels):
|
Added 21 shapes from hipBLASLt-only retry (bypassing ASM tile failures). Total: 36/92 unique (N,K) shapes tuned. Remaining 56 shapes need further investigation — filed ROCm#2515. Root causes: ASM tile alignment for non-standard N, grid overflow for large M with small tiles, tuner not falling through to hipBLASLt. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
BF16 GEMM tuning configs for MI355X (gfx950) covering 16 operator shapes.
23.6% geomean improvement over hipBLASLt alone. ASM wins 454/826 configs.
Upstream target: ROCm/aiter