tune: a8w8 gemm tuning for Qwen3.5 MXFP4-AttnFP8 model

[yuzho-amd]

Motivation

Adds a8w8_bpreshuffle FP8 GEMM tuning entries for the Qwen3.5-397B-A17B-MXFP4-AttnFP8-V2
model

N × K	module	#shapes	M range
64 × 4096	GatedDeltaNet in_proj_ba (linear-attn b/a gate)	108	1 .. 131072
4096 × 4096	full-attn o_proj	57	24 .. 504
8704 × 4096	full-attn qkv_proj (attn_output_gate)	38	24 .. 1984
10240 × 4096	GatedDeltaNet in_proj_qkvz (linear-attn qkvz)	26	24 .. 248

Technical Details

Test Plan

Test Result

Submission Checklist

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

[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 4396 --add-label <label>

[Copilot]

Pull request overview

This PR adds a large set of tuned a8w8_bpreshuffle GEMM config rows intended to accelerate FP8 GEMMs for the Qwen3.5-397B MXFP4-AttnFP8 model on gfx950.

Changes:

• Appends new gfx950 tuning rows covering multiple (N, K) groups (notably 64×4096, 4096×4096, 8704×4096, 10240×4096) across a wide M range.
• Introduces additional kernel selections spanning ck, cktile, and flydsl implementations for those shapes.

---

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

[yzhou103]

model configs should be added to the specific csv in model_configs dir

[yzhou103]

please move the configs to csv in model_configs

[yuzho-amd]

please move the configs to csv in model_configs

done. please review it.

[yzhou103]

please move the configs to csv in model_configs

done. please review it.

There is already a related csv in aiter. aiter/configs/model_configs/a8w8_bpreshuffle_tuned_gemm_qwen3_5_397b_a17b_mxfp4_attnfp8.csv(#4317)

[yuzho-amd]

please move the configs to csv in model_configs

done. please review it.

There is already a related csv in aiter. aiter/configs/model_configs/a8w8_bpreshuffle_tuned_gemm_qwen3_5_397b_a17b_mxfp4_attnfp8.csv(#4317)

sorry. please review it.

updated