[alexioslyrakis-amd]

Summary

• Both Python and C++ read cfg.flat directly from the config, so the flat_mode parameter added in PR [PERF] gfx942 FMoE: a16 kernel dispatch for GLM-5 decode (T=16-256) #4558 was redundant
• The AITER_CHECK(cfg.flat == flat_mode) caused a crash on gfx950 because fmoe_g1u1 was never wired up to pass flat_mode, defaulting to 0 while the config had flat=1
• Removes flat_mode from get_heuristic_kernel, fmoe_g1u1, and fmoe_fp8_blockscale_g1u1 entirely
• Hardcodes cfg.flat == 0 in the heuristic filter to exclude flat kernels (they are always dispatched by explicit kernel name from the tuned config CSV)

Test plan

• Verify Qwen3.5-397B-A17B-MXFP4 accuracy test passes on gfx950 (reproduces the ATOM CI failure from ROCm/ATOM#1857)
• Verify no regression on gfx942 FMoE dispatch (flat and non-flat paths)

[github-actions]

🏷️ CI Guide

Runs automatically on every PR:

• ✅ Pre-checks (submodule verification, code formatting)
• ✅ Aiter op tests (gfx942 + gfx950)
• ✅ Triton tests on MI35X (only when aiter/ops/triton/** or related paths are changed)

Extended tests (opt-in via labels):

Label	Tests
ci:triton-300x	Run an additional Triton test job on MI300X in PRs; main branch always runs both MI35X and MI300X
ci:sglang	SGLang integration tests: DeepSeek-R1-MXFP4 accuracy, Qwen 3.5 accuracy
ci:atom	ATOM benchmark: DeepSeek-R1-0528, GPT-OSS-120B
ci:atom_full	ATOM accuracy suite for PR and main models from ATOM models_accuracy.json
ci:vllm	vLLM benchmark: GPT-OSS-120B, DeepSeek-R1-0528, Kimi-K2.5
ci:all	All standard extended tests (excludes ci:atom_full)

Only add ci:atom_full for FlyDSL or Triton upgrades.
Add labels via the sidebar or gh pr edit 4675 --add-label <label>