Skip to content

[hipBLASLt] add tuned K=1024 grid entries for gfx1151 BF16 GEMMs used by VJEPA-2#9238

Open
serged-amd wants to merge 1 commit into
developfrom
users/serged/vjepa2_gfx1151_tuning
Open

[hipBLASLt] add tuned K=1024 grid entries for gfx1151 BF16 GEMMs used by VJEPA-2#9238
serged-amd wants to merge 1 commit into
developfrom
users/serged/vjepa2_gfx1151_tuning

Conversation

@serged-amd

@serged-amd serged-amd commented Jul 9, 2026

Copy link
Copy Markdown
Contributor

Motivation

The GEMM shapes used by the VJEPA-2 model have sub-optimal defaults which cause reduced model throughput.

This PR adds one and modifies two K=1024 grid points to significantly improve GEMM performance in this model.

Technical Details

Changes to gfx1151_Cijk_Alik_Bljk_BBS_BH_Bias_HAS_SAV_UserArgs.yaml:

Shape Old Solution New Solution Old TFLOPs New TFLOPs Speedup
3072×576×1024 829 (MT128x64x32) 1103 (MT96x128x32) 24.5 32.3 +32%
1024×576×1024 1073 (MT32x64x32) 1042 (MT128x64x64) 24.3 30.5 +26%
1024×576×4096 1035 (MT32x64x64) 1042 (MT128x64x64) 23.9 33.7 +41%

Grid entry changes:

  • Add [3072, 576, 1, 1024] -> [287, 32280.3]
  • Modify [1024, 576, 1, 1024] -> [226, 30463.3] (was [257, 6.93])
  • Modify [1024, 576, 1, 4096] -> [226, 33695.3] (was [219, 6.72])

Test Plan

  • Verify solution selection via HIPBLASLT_LOG_MASK=32
  • Benchmark V-JEPA2 ViT-L/384 model end-to-end

Test Result

V-JEPA2 ViT-L/384: 39.4 -> 44.3 FPS (+12%)

Solution selection verified:

  • 3072x576x1024: 829 -> 1103
  • 1024x576x1024: 1073 -> 1042
  • 1024x576x4096: 1035 -> 1042

Submission Checklist

Improve V-JEPA2 model throughput by 12% (39.4 -> 44.3 FPS)

Changes to gfx1151_Cijk_Alik_Bljk_BBS_BH_Bias_HAS_SAV_UserArgs.yaml:
- Add 3072×576×1024: solution 287 (24.5 -> 32.3 TFLOPs)
- Fix 1024×576×1024: solution 257→226 (24.3 -> 30.5 TFLOPs)
- Fix 1024×576×4096: solution 219→226 (23.9 -> 33.7 TFLOPs)
@serged-amd serged-amd requested a review from a team as a code owner July 9, 2026 19:21
@therock-pr-bot

therock-pr-bot Bot commented Jul 9, 2026

Copy link
Copy Markdown

❌ PR Check — Action Required

Check Status Details
🌿 Branch Name ✅ Pass
📝 PR Title/Description ❌ Fail Error: Title does not follow Conventional Commits style.
Expected: start with a valid type (feat, fix, docs, …).
Desired format: type(optional-scope): short description
───
Error: PR description must reference a JIRA ID, ISSUE ID, or a GitHub closing keyword.
Expected: include a JIRA ID / ISSUE ID line (separator : or -, or omitted; value may be a JIRA key, a number with/without #, or a link), OR a closing keyword + issue reference. Accepted examples:
JIRA ID : TESTAUTO-6039
JIRA ID - #330
JIRA ID #330
ISSUE ID : TESTUTO-3334
ISSUE ID #3334
ISSUE ID - TESTAUTO-3433
ISSUE ID : https://github.com/<org_name>/<repo_name>/issues/1234
Closes #10
Fixes octo-org/octo-repo#100
Resolves: #123
#123
https://github.com/<org_name>/<repo_name>/issues/123
Current: no valid JIRA/ISSUE/closing-keyword reference found
Forbidden Files ✅ Pass
🧪 Unit Test ✅ Pass PR does not contain code files — Unit Test auto-passed
🔎 pre-commit ✅ Pass
🚫 Draft PR 🔜 To Be Enabled
🚩 Feature Flag 🔜 To Be Enabled
📊 Code Coverage 🔜 To Be Enabled

⚠️ 1 policy check(s) failed. Please address the issues above before this PR can be Reviewed.

🚫 Please fix the failed policies

  • ❌ PR Title/Description

The Not ready to Review label was added to this PR. Once all policies pass, the label is removed automatically.

📖 Need help? See the Policy FAQ for details on every check and how to fix failures.

@therock-pr-bot

therock-pr-bot Bot commented Jul 9, 2026

Copy link
Copy Markdown

🚫 Please fix the failed policies before requesting reviews.

The following policy checks failed:

  • ❌ PR Title/Description

The Not ready to Review label has been added to this PR.
Once all policies pass, the label will be removed automatically.

@codecov-commenter

Copy link
Copy Markdown

Codecov Report

✅ All modified and coverable lines are covered by tests.

❌ Your project status has failed because the head coverage (76.92%) is below the target coverage (80.00%). You can increase the head coverage or adjust the target coverage.

Additional details and impacted files
@@             Coverage Diff             @@
##           develop    #9238      +/-   ##
===========================================
- Coverage    64.99%   64.99%   -0.00%     
===========================================
  Files         2685     2685              
  Lines       422106   422157      +51     
  Branches     62705    62717      +12     
===========================================
+ Hits        274342   274355      +13     
- Misses      127065   127102      +37     
- Partials     20699    20700       +1     
Flag Coverage Δ *Carryforward flag
TensileLite 34.86% <ø> (ø) Carriedforward from 436f490
hipBLAS 90.81% <ø> (ø) Carriedforward from 436f490
hipBLASLt 34.69% <ø> (-0.02%) ⬇️
hipCUB 82.68% <ø> (ø) Carriedforward from 436f490
hipDNN 86.15% <ø> (ø) Carriedforward from 436f490
hipFFT 50.88% <ø> (ø) Carriedforward from 436f490
hipRAND 76.12% <ø> (ø) Carriedforward from 436f490
hipSOLVER 69.18% <ø> (ø) Carriedforward from 436f490
hipSPARSE 86.10% <ø> (ø) Carriedforward from 436f490
rocBLAS 47.91% <ø> (ø) Carriedforward from 436f490
rocFFT 50.80% <ø> (ø) Carriedforward from 436f490
rocRAND 57.02% <ø> (ø) Carriedforward from 436f490
rocSOLVER 76.92% <ø> (ø) Carriedforward from 436f490
rocSPARSE 72.37% <ø> (ø) Carriedforward from 436f490
rocThrust 91.36% <ø> (ø) Carriedforward from 436f490

*This pull request uses carry forward flags. Click here to find out more.
see 3 files with indirect coverage changes

🚀 New features to boost your workflow:
  • ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
  • 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants