Add mixed precision (FP16) support for ROCm #7663
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
|
/test mini |
|
Thanks, @jglaser! Your change looks good to me. Can I ask you to fix just static-checks? |
I also checked |
|
/test mini |
|
Thanks @takagi for moving this along. Looks like two CUDA related tests are still failing, however, I can't see the test results without logging in |
|
CI failures are not related. Our CI requires logging in for security reasons. |
|
Thanks, @jglaser! |
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Outdated suggestions cannot be applied.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.
This PR adds mixed precision support for matrix-matrix multiplies on ROCm, which is available in hipBLAS through
hipblasGemmEx. On MI200 GPUs, this utilizes the MFMA (matrix fused-multiply add) instructions, which are intended as analogues of the NVIDIA tensor cores.For
cupy.float16arrays, we allow the use of the hipBLAS primitive, which was previously disabled, and fp16 values were casted to single precision, albeit issuing a warning. I am not entirely sure if the reason for this was the lack of hardware and/or API support in earlier versions of the AMD libraries, therefore a version check for hipBLAS might be required. It appears though that support has been stable since ROCm 4.2.0 (https://github.com/ROCmSoftwarePlatform/hipBLAS/blob/develop/CHANGELOG.md).To verify that the code now uses
rocblas_gemm_exunder the hood, enable rocBLAS runtime tracing with the unit test like this