[hipBLASlt][CMS] CMS for TF32 128x128x32 NT

[emezh]

Motivation

Add CMS for TF32 128x128x32 NT.
Also, fix custom_mainloop_scheduling_tf32.yaml to use BiasTypeArgs: ['s'] for all configs.

Tensile, no CMS vs CMS

MNK = 2048,2048,8192

• Time: 12% improvement
• Efficiency: 44.7% --> 66%

Bench, Baseline vs CMS

MNK = 2048,2048,8192

• Time: no improvement (-15.4%)
• Efficiency: n/a - baseline and new CMS use different tiles/kernels

MNK = 2048,2048,4096

• Time: 5% improvement
• Efficiency: n/a - baseline and new CMS use different tiles/kernels

Test Result

Local tests:

• custom_mainloop_scheduling_tf32.yaml - pass
• test_CustomSchedule.py - pass
• test_CustomSchedule_LayoutAutoDetection.py - pass
• hipblaslt-test - pass

[==========] 22050 tests from 12 test suites ran. (1645862 ms total)
[  PASSED  ] 22050 tests.

• tensile with the ranges below - pass

          - Exact: [2048, 2048, 1, 8192]
          - Exact: [2048, 2048, 1, 4096]
          - Range: [[128], [128], [1], [64, 64, 256]]
          - Range: [[128], [128], [1], [1,1,64]]
          - Range: [[128], [128], [1], [32, 64, 256]]

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

AIGECORE-69

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.

❌ Your project status has failed because the head coverage (76.83%) 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    #4900   +/-   ##
========================================
  Coverage    65.94%   65.94%           
========================================
  Files         1718     1718           
  Lines       267197   267197           
  Branches     37045    37045           
========================================
  Hits        176194   176194           
  Misses       75448    75448           
  Partials     15555    15555           

Flag	Coverage Δ		*Carryforward flag
hipBLAS	90.67% <ø> (ø)		Carriedforward from aae6a45
hipBLASLt	43.55% <ø> (ø)
hipCUB	82.38% <ø> (ø)		Carriedforward from aae6a45
hipDNN	80.82% <ø> (ø)		Carriedforward from aae6a45
hipFFT	55.93% <ø> (ø)		Carriedforward from aae6a45
hipRAND	76.12% <ø> (ø)		Carriedforward from aae6a45
hipSOLVER	68.81% <ø> (ø)		Carriedforward from aae6a45
hipSPARSE	84.70% <ø> (ø)		Carriedforward from aae6a45
rocBLAS	47.97% <ø> (ø)		Carriedforward from aae6a45
rocFFT	52.91% <ø> (ø)		Carriedforward from aae6a45
rocRAND	57.06% <ø> (ø)		Carriedforward from aae6a45
rocSOLVER	76.83% <ø> (ø)		Carriedforward from aae6a45
rocSPARSE	71.53% <ø> (ø)		Carriedforward from aae6a45

*This pull request uses carry forward flags. Click here to find out more.

🚀 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.

[math-ci-webhook]

perfci run on commit 66fa26f

math-ci run