optimze-elect-one-compilation-hint

[yanqinz2]

Before submitting

• I agree to license this contribution under the terms of LICENSE.txt.
• I ran pre-commit run and committed any formatting changes.
• I added GitHub labels: one cat-*, one or more mod-*, and one orig-* (see label list).

Affected area

Summary

Remove repeated elect_sync call to optimize the compilation to avoid BSSY/BSYNC emission

Related issues

Non-optimal perf for small inference cases

API and compatibility impact

Testing

Summary by CodeRabbit

• Performance Improvements
  • Improved GPU kernel efficiency by reducing redundant thread-election operations across matrix multiplication and grouped computation workloads.
  • Preserved existing synchronization, data movement, and computation behavior while streamlining execution.
  • Improved blocked scale-factor handling for padded matrix dimensions.
  • Enhanced diagnostic reporting when reference scaled matrix multiplication is unavailable or fails.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 1e17ee66-02ee-4531-9ae0-f990c0fda148

📥 Commits

Reviewing files that changed from the base of the PR and between ab9efe1 and cbdc296.

📒 Files selected for processing (18)

• benchmark/gemm/frost/benchmark_block_scale_matmul.py
• benchmark/gemm/frost/benchmark_block_scale_matmul_swiglu.py
• python/cudnn/gemm/frost/kernel_templates/sm100_block_scale_matmul_1ctamma.py
• python/cudnn/gemm/frost/kernel_templates/sm100_block_scale_matmul_1ctamma_static.py
• python/cudnn/gemm/frost/kernel_templates/sm100_block_scale_matmul_2ctamma.py
• python/cudnn/gemm/frost/kernel_templates/sm100_block_scale_matmul_2ctamma_static.py
• python/cudnn/gemm/frost/kernel_templates/sm100_matmul_1ctamma.py
• python/cudnn/gemm/frost/kernel_templates/sm100_matmul_1ctamma_static.py
• python/cudnn/gemm/frost/kernel_templates/sm100_matmul_2ctamma.py
• python/cudnn/gemm/frost/kernel_templates/sm100_matmul_2ctamma_static.py
• python/cudnn/gemm/frost/kernel_templates/sm100_matmul_mainloop_1ctamma.py
• python/cudnn/gemm/frost/kernel_templates/sm100_matmul_mainloop_2ctamma.py
• python/cudnn/gemm/frost/kernel_templates/sm100_moe_grouped_block_scale_matmul_fwd_1ctamma.py
• python/cudnn/gemm/frost/kernel_templates/sm100_moe_grouped_block_scale_matmul_fwd_2ctamma.py
• python/cudnn/gemm/frost/kernel_templates/sm100_moe_grouped_matmul_fwd_1ctamma.py
• python/cudnn/gemm/frost/kernel_templates/sm100_moe_grouped_matmul_fwd_2ctamma.py
• python/cudnn/gemm/frost/kernel_templates/sm103_block_scale_matmul_1ctamma.py
• python/cudnn/gemm/frost/kernel_templates/sm103_block_scale_matmul_2ctamma.py

---

📝 Walkthrough

Walkthrough

The changes update blocked-scale benchmark shape handling and reference diagnostics. They also replace repeated nvvm.elect_sync() calls across FROST GEMM, block-scale GEMM, and MoE kernel templates with cached election predicates.

Changes

FROST kernel and benchmark updates

Layer / File(s)	Summary
Blocked scale reference preparation benchmark/gemm/frost/benchmark_block_scale_matmul.py, benchmark/gemm/frost/benchmark_block_scale_matmul_swiglu.py	Reference and SwiGLU scale tensors now use padded blocked-layout dimensions. The reference path uses separate swizzles, forwards verbosity, and reports explicit scaled_mm failures.
SM100 and SM103 election reuse python/cudnn/gemm/frost/kernel_templates/sm100_*, python/cudnn/gemm/frost/kernel_templates/sm103_*	Kernel templates compute elect_one once and reuse it for barrier, TMA, MMA, PDL, CLC, and epilogue operations.
MoE election reuse python/cudnn/gemm/frost/kernel_templates/sm100_moe_*	Grouped MoE kernels reuse one elected-thread predicate across data transfers, tensor-memory operations, scheduler signaling, synchronization, and epilogues.

Estimated code review effort: 4 (Complex) | ~45 minutes

Possibly related PRs

• NVIDIA/cudnn-frontend#476: Modifies the same FROST benchmark and SM100/SM103 kernel-template areas.

Suggested labels: cat-feature, orig-nv-eng, mod-cutedsl, mod-frost

Suggested reviewers: anerudhan

🚥 Pre-merge checks | ✅ 3 | ❌ 2

❌ Failed checks (2 warnings)

Check name	Status	Explanation	Resolution
Description check	⚠️ Warning	The description includes the summary and motivation but leaves the affected area, API impact, and testing details incomplete.	Select the affected area, state the API and compatibility impact, and provide exact testing commands with results or explain why testing was not performed.
Docstring Coverage	⚠️ Warning	Docstring coverage is 20.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (3 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title identifies the elect-one compilation optimization, which matches the main change despite a spelling error and awkward phrasing.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands.}

[Anerudhan]

@cudnn-ci-bot run frost

[cudnn-ci-bot]

🚀 Running mirror pipeline

Branch: cudnn-gh/pr-504-cbdc296
Pipeline: 61530048
Targets: frost