[CUDA] Optimize Transpose3DKernel #10891
Merged
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
pengwa
reviewed
Mar 17, 2022
| @@ -7,44 +7,52 @@ | |||
| namespace onnxruntime { | |||
| namespace cuda { | |||
|
|
|||
| constexpr unsigned int TILE_DIM = 16; | |||
| constexpr unsigned int NUM_ELE_PER_THREAD = 4; | |||
There was a problem hiding this comment.
Have we ever tried other values like 2/8/16?
There was a problem hiding this comment.
And we expect same improvement on ROCM?
There was a problem hiding this comment.
Tried 4 and 8, make no much difference. Most of our other kernels use 4, so follow the same number here. Didn't check the perf on ROCm.
lavanyax
pushed a commit
to intel/onnxruntime
that referenced
this pull request
Mar 29, 2022
seddonm1
pushed a commit
to seddonm1/onnxruntime
that referenced
this pull request
May 15, 2022
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Outdated suggestions cannot be applied.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.
Optimize CUDA Transpose3DKernel by handling more elements in one thread.
Take Transpose([3072,64,512], [0,21]) as example (which is from one of our models):
in A100, the kernel takes:
in V100:
Also run more perf tests on different input tensor size, no perf regression observed. The optimization works better on FP16 than FP32, and large tensors.