[AMD] Enable pingpong scheduling by default #5696
Open
+9
−1
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
Now it the pass selects possible pingpong options by the given problem. It's only enabled when we believe it improves the perfromance by applying the pingpong pattern. N.B. Still it can slow down a kernel significantly when register spills. It tries to avoid spilling gemm config but cannot guarantee when additional logic is involved.
antiagainst
requested changes
Jan 25, 2025
antiagainst
reviewed
Jan 25, 2025
|
Test fails, will fix it. |
|
#5718 fixes the failure. |
antiagainst
requested changes
Jan 30, 2025
| @@ -36,6 +36,14 @@ def get_gfx9_limits(lhsType, rhsType): | |||
| return lambda lhsType, rhsType: (1, 1, 1) if is_fma_supported(lhsType.scalar, rhsType.scalar) else (16, 16, 16) | |||
|
|
|||
|
|
|||
| def is_pingpong_enabled(arch): | |||
| if arch == 'gfx942': | |||
There was a problem hiding this comment.
Nit: Actually I feel something like the following would be simpler:
default = "1" if arch == "gfx942" else "0"
return os.getenv("TRITON_HIP_USE_BLOCK_PINGPONG", default) == "1"
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Outdated suggestions cannot be applied.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.
The pass selects possible pingpong options by the given problem. It's only enabled when we believe it improves the performance by applying the pingpong pattern. Now, we're enabling the pass by default on gfx942 arch where we believe it's beneficial.
N.B. Still it can slow down a kernel significantly when register spills. It tries to avoid spilling gemm config but cannot guarantee when additional logic is involved.