Skip to content

Enable mbsk spmm #2119

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 6 commits into from
Jun 3, 2025
Merged

Enable mbsk spmm #2119

merged 6 commits into from
Jun 3, 2025

Conversation

vin-huang
Copy link
Collaborator

  1. Added requiredSynchornizerSizes()
  2. Enable MBSK for SPMM
  3. Add MBSK test for SPMM

@vin-huang vin-huang added the gfx950 run CI on gfx950 label May 24, 2025
@vin-huang vin-huang force-pushed the enable_mbsk_spmm branch 4 times, most recently from d3cb107 to 94935bb Compare May 26, 2025 08:27
@vin-huang vin-huang marked this pull request as ready for review May 26, 2025 08:28
@vin-huang
Copy link
Collaborator Author

The Azure CI was failed due to cannot find proper device.

Hip error: 'no ROCm-capable device is detected'(100) at /agent/_work/1/s/library/src/amd_detail/hipblaslt.cpp:171

@vin-huang vin-huang force-pushed the enable_mbsk_spmm branch 2 times, most recently from 90eaf6b to c1c8201 Compare May 27, 2025 17:19
KKyang
KKyang previously approved these changes May 28, 2025
@@ -867,14 +867,19 @@ def GSUSynccodegen(self, writer, kernel, tmpVgpr, tmpVgprSize, tmpVgprDynamic, b
vmcnt = SyncloadedData = SyncloadedData -1
module.add(SWaitCnt(lgkmcnt=lgkmcnt, vmcnt=vmcnt, vscnt=vscnt, comment="(wait for buffer ready)"))

if ((gwvw % 2) == 1):
if kernel["ProblemType"]["DataType"].isInt8() or kernel["ProblemType"]["DataType"].isInt32():
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

should we check kernel["ProblemType"]["ComputeDataType"].isInt32() ?

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

In the original case, which didn't consider the output datatype is int32 from mfma/smfma instruction.
So, here I use kernel["ProblemType"]["DataType"] to identify the output datatype.

@vin-huang vin-huang merged commit e7b145e into ROCm:develop Jun 3, 2025
25 of 33 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
gfx950 run CI on gfx950
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants