[ROCm-Profiler] Add lds stall preset

[diptorupd]

Adds one more preset to the rocm_profiler to capture LDS stall cycles.

[Copilot]

Pull request overview

Adds a new ROCm profiler counter preset aimed at diagnosing LDS/VMEM stall behavior, and introduces automatic derived stall-metric reporting when the required PMCs are present in the collected counter CSVs.

Changes:

• Added a new lds_stall 3-pass rocprofv3 preset to capture LDS/VMEM stall + latency counters under gfx942 constraints.
• Added automatic post-processing to print an LDS/VMEM stall analysis table when stall counters are detected.
• Updated the FA2 prefill ROCm benchmark documentation to include the new preset and the derived metrics it prints.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
rocm_profiler/rocm_profiler.py	Adds lds_stall preset and implements stall table detection + printing after profiling.
benchmarks/rocm_benchmarks/bench_fa2_prefill.py	Documents the new lds_stall preset and its derived metrics in the usage header.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.