[amd-vkale]

Motivation

Week three of the range replay test roadmap (source/docs/conceptual/range_replay/range_replay_test_plan.md): C3 and C4. A range whose tracked allocation set changes between the entry snapshot and close must decline with ALLOCATION_CHANGED_IN_RANGE rather than restore into memory the application has since repurposed; an empty range must close with NO_DISPATCH. Both are bookkeeping-covered in unit tests; this PR drives them on hardware and checks that no pass runs and the application's own result is intact.

Stacked on #13272 (week two: C8).

Technical Details

Extends tests/range-replay-declines/ (C1/C2) with three new RR_DECLINE_MODE values:

• hip-malloc (C3). One binding kernel takes the entry snapshot; hipMalloc inside the range changes the tracked set; CLOSE reports ALLOCATION_CHANGED_IN_RANGE with one observed dispatch and zero passes; acc == 1.
• hip-free (C3). Same shape, but hipFree of a pre-range allocation after the binding kernel.
• empty (C4). begin then end with no dispatch; CLOSE reports NO_DISPATCH with zero dispatches.

The tool now takes RR_DECLINE_EXPECT_DISPATCHES per mode so C1/C2 still expect one dispatch while C4 expects zero. The roadmap coverage row and missing-tests preamble record C3 and C4.

Issue Tracking

Related: #10864 (range replay, whose allocation-set check and empty-range CLOSE this exercises).

Test Plan

• test-range-replay-decline-hip-malloc, test-range-replay-decline-hip-free, test-range-replay-decline-empty on the mi325 CI jobs (plus the existing C1/C2 modes still registered from this suite).
• Locally, with no GPU or ROCm on the build machine:
  • clang-format 11 and cmake-format are clean;
  • full HIP device compile and -fsyntax-only against HIP headers were not available here (no hip/hip_runtime.h / no libstdc++ for a CMake compiler test).

Test Result

CI for this PR is pending. Nothing could be run on a GPU locally.

Week two (#13272, 5758d889a0): C8 (test-range-replay-concurrent-graph-launch) PASSED on Core rhel-8.8, rhel-9.5, sles-15.6, and on ubuntu-22.04's main suite. Ubuntu Core was red for other tests (not C8); coverage hit limits; Multi-Arch summary failed / cancelled stages.

#10864 is unchanged at 0844152fe6; its Core reds remain pre-existing flakes / bot timeouts (rocshmem, samples, therock-pr-bot), not range-replay regressions.

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/rocm-systems/blob/develop/CONTRIBUTING.md.