[Bugfix][Benchmark] Fix compiler API usage in Softmax and RMSNorm benchmarks

[tangzzycc]

Motivation

The shared comparison sweep does not report FlyDSL timing results for Softmax and RMSNorm.

Both paths call the removed top-level flydsl.compile API. The resulting AttributeError is caught by run_compare_sweep, which records flydsl_gpu_us=None, so the failure appears as a missing - result instead of an explicit error.

Technical Details

• Import flydsl.compiler as flyc.
• Use flyc.compile for the Softmax and RMSNorm benchmark paths.
• Reuse the same compiler import in the LayerNorm path.

Test Plan

Test Result

Submission Checklist

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

[Copilot]

🟢 Approval recommended

The change is a narrow, verified API-call fix in benchmark code that aligns with the existing flydsl.compiler.compile export surface and removes the known failure mode.

Pull request overview

Fixes FlyDSL benchmark timing gaps for Softmax and RMSNorm by updating the benchmarks to use the current compiler entry point (flydsl.compiler.compile) instead of the removed top-level flydsl.compile, preventing silent AttributeError-driven “missing result” reporting in the comparison sweep.

Changes:

• Switch Softmax and RMSNorm benchmark compilation from flydsl.compile(...) to flydsl.compiler.compile(...) (imported as flyc.compile(...)).
• Reuse the same flydsl.compiler as flyc import across Softmax, LayerNorm, and RMSNorm paths.
• Update the docstring to reflect the new API usage.

File summaries

File	Description
tests/kernels/benchmark_common.py	Updates benchmark compilation calls to the supported flydsl.compiler.compile API so Softmax/RMSNorm timings are reported correctly.

Review details

• Files reviewed: 1/1 changed files
• Comments generated: 0
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.