Optimize fused_dynamic_mxfp4_quant_moe_sort_hip in small M

[junhaha666]

test on MI355: python3 op_tests/test_moe_sorting_mxfp4.py -dim1 7168 -dim2 256 -ek 256,8
Stage1:
before:

now:

Stage2:
before:

now:

[github-actions]

🏷️ CI Guide

Runs automatically on every PR:

• ✅ Pre-checks (submodule verification, code formatting)
• ✅ Aiter op tests (gfx942 + gfx950)
• ✅ Triton tests (only when aiter/ops/triton/** or related paths are changed)

Extended tests (opt-in via labels):

Label	Tests
ci:triton-355	Run Triton tests on MI355 in addition to MI325
ci:sglang	SGLang integration tests
ci:atom	ATOM benchmark (DeepSeek-R1 + GPT-OSS)
ci:vllm	vLLM benchmark
ci:all	All of the above

Add labels via the sidebar or gh pr edit 2700 --add-label <label>

[Copilot]

Pull request overview

Optimizes the HIP fused MXFP4 quantization + MoE scale sorting path for smaller M by changing the sorted_ids access pattern and adjusting the Python-side heuristic for when to use the fused kernel.

Changes:

• Update mxfp4_quant_moe_sort_kernel to use a strided indexing scheme within block_m for sorted_ids traversal.
• Pass tgs_per_block_m into the kernel and derive it from block_m / sub_block_m.
• Adjust the stage2 dispatch threshold in fused_dynamic_mxfp4_quant_moe_sort to account for topk.

Reviewed changes

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

File	Description
csrc/kernels/quant_kernels.cu	Implements strided sorted_ids traversal and adds tgs_per_block_m plumbing for the fused HIP kernel.
aiter/ops/quant.py	Updates the stage2 cutoff heuristic for selecting the fused HIP path.

---

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