[AIROCMLIR-378] Add negative lit coverage for Rock/MIGraphX/MHAL

[bogdan-petkovic]

Motivation

This PR increases rocMLIR's code coverage toward the 85% target tracked in issue #1825 , building on the earlier MHAL coverage work in #2394. It focuses on the cheapest, highest-confidence wins: error branches of dialect op verifiers, attribute/type parsers, and the conv-transpose decomposition pass that were previously exercised only on their happy paths (so the failure branches showed up as miss/partial in Codecov). These are pure host-side checks that need no GPU, so they run in every CI configuration and harden the dialects against regressions in parsing/printing/verification.

Technical Details

Adds nine new -split-input-file -verify-diagnostics lit test files (71 cases total) that each trip a specific, previously-uncovered verifier or parser error branch. No source code changes — tests only. Coverage targets, by file: mlir/test/Dialect/Rock/transform_attr_invalid.mlir covers TransformAttr::verify, TransformMapAttr::verify, and TransformAttr::parse (every transform-type rule plus the unknown-name path); transforming_for_invalid.mlir covers TransformingForOp::parse/::verify; reduce_invalid.mlir covers ReduceOp::verify; op_verifier_invalid.mlir covers InsertSliceOp, GpuAllocOp, LiveInOp, LiveOutOp, GlobalStoreOp, InBoundsLoadOp, and InBoundsStoreOp; threadwise_blockwise_invalid.mlir covers ThreadwiseGemmOp, BlockwiseGemmOp, BlockwiseFillOp, and the remaining GlobalLoadToLDSOp branches; mlir/test/Dialect/MIGraphX/literal_unpack_invalid.mlir covers LiteralOp::verify/UnpackOp::verify; shaped_type_invalid.mlir covers MIXRShapedType::parse; mlir/test/Dialect/MHAL/attrs-invalid.mlir covers the unknown-type-name branches of TargetObjectAttr::parse and KernelPackageAttr::parse; and mlir/test/Conversion/RocmlirCustomTosaDecompose/rocmlir-custom-tosa-decompose-invalid.mlir covers the stride/out_pad/output-dim/bias branches of verifyConvTranspose. Where the diagnostics are emitted through the dialect-conversion driver (the decompose pass) or the attribute parser wrapper, the tests expect both the specific verifier message and the framework's follow-up diagnostic, matching the deterministic emission order. Each file's leading // COM: comments cite the exact source function being exercised.

Test Plan

PR CI, including the Codecov coverage stage which builds with -DLLVM_BUILD_INSTRUMENTED_COVERAGE=ON and runs the full check-rocmlir lit suite. Locally, every new file was validated by running its exact RUN command (rocmlir-opt ... -split-input-file -verify-diagnostics, and the decompose pass for the conversion test) against a local build; all 71 cases pass and the diagnostic expectations are deterministic across repeated runs.

Test Result

All 9 new lit files pass locally (71/71 cases). The targeted branches were confirmed reachable from textual IR, so they convert prior miss/partial lines into hits; the exact coverage delta will be reported by the Codecov bot on this PR.

Submission Checklist

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

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #2415      +/-   ##
===========================================
+ Coverage    82.57%   83.26%   +0.70%     
===========================================
  Files          120      120              
  Lines        42852    42828      -24     
  Branches      7110     7106       -4     
===========================================
+ Hits         35381    35660     +279     
+ Misses        4815     4613     -202     
+ Partials      2656     2555     -101     

Flag	Coverage Δ
gfx120x	83.13% <ø> (+0.60%)	⬆️
gfx950	83.04% <ø> (+0.70%)	⬆️
mfma	83.05% <ø> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.
see 31 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.