fix(tensilelite): address AIHPBLAS-3868 characterization gaps

[newling]

JIRA ID : AIHPBLAS-3868

Motivation

TensileLite's configuration-driven tests exercise many GPU assembly-generation paths without requiring a GPU. The stack ending in #11970 increased that coverage, but most saved results recorded only the solution-derived kernel name and emitter return code. An emitted instruction could therefore change incorrectly without failing the corresponding test, while unrelated solution-state changes could still churn the saved kernel names.

The stack also admitted one unsupported MX scale layout until kernel emission, selected only the first problem group from multi-group YAML files without recording that choice, and contained repeated test code that generated the same kernels twice.

Technical Details

This PR follows #11970 and addresses the five findings from the stack review.

• Config-driven saved results now include a digest of the emitted opcode set. This detects added, removed, or replaced instruction kinds while ignoring unstable register allocation and instruction ordering.
• Rejection tests compare their exact diagnostic multisets instead of saving only a zero solution count.
• Unsupported WMMA_V3 in-memory-swizzled MX local reads are rejected during solution derivation. A defensive runtime check remains for callers that bypass derivation.
• Set-cover cases record an explicit BenchmarkProblems index in their parameter lists and test names.
• Shared helpers run each ordinary configuration once, removing 36 duplicate generation passes and repeated assertion code.
• The coverage rebaseline documents every historical reduction and updates three reproduced floors after removing invalid pre-exception coverage.

Test Plan

Run the changed TensileLite unit and configuration-driven characterization tests, the complete characterization suite under coverage, the strict YAML corpus check, and the per-file coverage gate. Automated GitHub checks must provide the final gfx1250 compiler coverage.

Test Result

• Changed test selection: 267 passed.
• Current-develop merge-tree codegen suite: 416 passed, 1 expected failure, 1 unexpected pass.
• Full characterization coverage: 3,256 passed, 2 expected failures, 1 unexpected pass; 891 saved results passed.
• Strict YAML corpus: 419 passed.
• Per-file coverage gate: passed.
• Flake8 and diff whitespace checks: passed.
• Full local unit coverage could not complete because the installed ROCm 7.1 device libraries do not support gfx1250. Automated GitHub checks are pending.

Submission Checklist

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

Risk level

Medium (3/5): the production change rejects one unsupported gfx1250 MX layout earlier in solution derivation. Valid layouts are unchanged, but the generator path and a broad set of saved characterization results require gfx1250 validation in automated GitHub checks.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Additional details and impacted files

@@                            Coverage Diff                            @@
##           users/davidd-amd/mut-v2-combined-godfile   #12004   +/-   ##
=========================================================================
  Coverage                                     70.67%   70.68%           
=========================================================================
  Files                                          2810     2810           
  Lines                                        462487   462487           
  Branches                                      68090    68090           
=========================================================================
+ Hits                                         326859   326866    +7     
+ Misses                                       112070   112066    -4     
+ Partials                                      23558    23555    -3     

Flag	Coverage Δ		*Carryforward flag
TensileLite-CPP	46.45% <ø> (ø)		Carriedforward from c9a14dc
TensileLite-Unit	76.94% <ø> (ø)		Carriedforward from c9a14dc
hipBLAS	90.62% <ø> (ø)		Carriedforward from c9a14dc
hipBLASLt	35.27% <ø> (+0.03%)	⬆️
hipCUB	82.68% <ø> (ø)		Carriedforward from c9a14dc
hipDNN	87.02% <ø> (ø)		Carriedforward from c9a14dc
hipFFT	42.66% <ø> (ø)		Carriedforward from c9a14dc
hipRAND	76.12% <ø> (ø)		Carriedforward from c9a14dc
hipSOLVER	68.92% <ø> (ø)		Carriedforward from c9a14dc
hipSPARSE	86.99% <ø> (ø)		Carriedforward from c9a14dc
rocBLAS	48.31% <ø> (ø)		Carriedforward from c9a14dc
rocFFT	51.72% <ø> (ø)		Carriedforward from c9a14dc
rocRAND	56.90% <ø> (ø)		Carriedforward from c9a14dc
rocSOLVER	76.83% <ø> (ø)		Carriedforward from c9a14dc
rocSPARSE	74.61% <ø> (ø)		Carriedforward from c9a14dc
rocThrust	91.60% <ø> (ø)		Carriedforward from c9a14dc

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

Files with missing lines	Coverage Δ
...blaslt/tensilelite/Tensile/Components/LocalRead.py	56.05% <ø> (ø)
...lt/tensilelite/Tensile/SolutionStructs/Solution.py	59.17% <ø> (ø)

... and 1 file 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.