Add gfx1151 flash tuning database for hdim=64, fp16/bf16#194
Open
amd-callumm wants to merge 1 commit into
Open
Add gfx1151 flash tuning database for hdim=64, fp16/bf16#194amd-callumm wants to merge 1 commit into
amd-callumm wants to merge 1 commit into
Conversation
Tuning database coverage:
- Head dimension: 64
- Sequence lengths: all Q × K combos where Q,K ∈ {16, 32, 64, 128, 256, 512, 1024, 2048}
- Data types: float16, bfloat16
- Attention modes: causal and non-causal
- Bias types: (0, 1)
Performance improvements (attn_fwd kernel vs. default):
- Average speedup: 14.7%
- Range: -1.7% to 53.5%
- Best improvements: large shapes with bias (2048×1024: 53.5%, 2048×512: 37.5%)
Validation:
- Build and run tuning test libraries at levels 0 and 1
- Level 1: 14,712 tests skipped (hdim > 256)
- All other tests passed
Collaborator
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Motivation
Start tuning flash attention on gfx1151 for performance.
For now, only head dim 64 and 16-bit data types are included. I intend to do broader/more complete runs, but they will take a while, so in the meantime, this covers some desirable use cases.
Technical Details
Tuning database coverage:
Performance improvements (attn_fwd kernel vs. default):
Test Plan
Build and run tuning test libraries and run level 0 and 1 test suites
Test Result
Level 1 tests: 14,712 tests skipped (hdim > 256)
All other tests passed.