[SM90] Change register allocation for TileN=208 to avoid spills #2219
Conversation
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
With the usual register allocation (producer 40, consumer 232) compiling Gemm with tile shape 256 x 208 (cooperative) or 128 x 208 (pingpong) show lots of register spilling (e.g. ~3000 bytes spill). For this case we can change the register allocation to producer 24, consumer 240, which avoids spills.
|
This is awesome! Thank you :D |
|
Were you able to AB test perf results by any chance? |
|
Perfwise nothing changes for the usual tile shapes since we keep reg allocation the same for those. For tile shape 256 x 208 it's obv a lot faster without the massive spill (5-10x). In some niche cases where 256 x 208 fits just right and there's minimal wave quantization 256 x 208 can be a good choice. |
hwu36
approved these changes
Apr 21, 2025
andralex
pushed a commit
to andralex/cutlass
that referenced
this pull request
Jun 14, 2025
…IA#2219) With the usual register allocation (producer 40, consumer 232) compiling Gemm with tile shape 256 x 208 (cooperative) or 128 x 208 (pingpong) show lots of register spilling (e.g. ~3000 bytes spill). For this case we can change the register allocation to producer 24, consumer 240, which avoids spills.
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Outdated suggestions cannot be applied.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.
With the usual register allocation (producer 40, consumer 232) compiling Gemm with tile shape 256 x 208 (cooperative) or 128 x 208 (pingpong) show lots of register spilling (e.g. ~3000 bytes spill). For this case we can change the register allocation to producer 24, consumer 240, which avoids spills.
Cc @thakkarV @hwu36