replace rocprof with rocprofv2 for the tune gemm script #613
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
zhanglx13
reviewed
Jul 17, 2024
scripts/amd/gemm/tune_gemm.py
Outdated
| @@ -392,9 +392,14 @@ def main(): | |||
|
|
|||
|
|
|||
| def extract_kernel_time(M, N, K, config, df, bias_size): | |||
| # Correct the header by removing 'sig' and 'obj' to reduce number from 21 to 19 | |||
| # once the bug is fixed, we should not need below two lines | |||
There was a problem hiding this comment.
Can you add the issue of rocprof here so that people know what you are referring to
ROCm/rocprofiler#144
zhanglx13
reviewed
Jul 17, 2024
| @@ -409,7 +414,7 @@ def profile_batch_kernels(M, N, K, gpuid, gpus, jobs, verbose): | |||
| kernel_name = generated_kernel_name(M, N, K, jobId) | |||
| if verbose: | |||
| print(f"profiling {kernel_name} on GPU {gpuid}") | |||
| run_bash_command_wrapper(f"rocprof --stats -o results-{jobId}.csv python {kernel_name}", capture=(verbose < 2)) | |||
| run_bash_command_wrapper(f"rocprofv2 --plugin file --plugin-version 1 --kernel-trace -o {jobId} python {generated_kernel_name(M, N, K, jobId)}", capture=(verbose < 2)) | |||
There was a problem hiding this comment.
generated_kernel_name(M, N, K, jobId) is the same as kernel_name, the latter is simpler.
And why did you change the output from results-{jobId}.csv to {jobId}? Is it automatically expanded to results_{jobId}.csv?
There was a problem hiding this comment.
yes, it is automatically expanded.
|
Have you checked performance on a few shapes to confirm before and after is the same? |
|
yes, I did @vgokhale Here is with tune_streamk using rocprof and tune_streamk using rocprofv2 tune_gemm with rocprof on smc300x tune_gemm with rocprofv2 on smc300x |
zhanglx13
approved these changes
Jul 17, 2024
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.
rocprofv2 is much faster than rocprof
rocprofv2 naturally support python
for 8192x8192x8192, it reduce tuning time from more than an hour to 5.22 mins with full tuning space.