bf16 asm mha: enable doubleq and kv reverse to improve perf

[tingchen988]

Motivation

[gfx1250] fmha fwd bf16 hdim128/64 support per tensor bf16 for gfx1250: enable doubleq and kv reverse to improve perf

Technical Details

change a runtime arg in csrc/py_itfs_cu/asm_fmha_fwd_with_sink_varlen.cu

Test Plan

python3 op_tests/test_fmha_fwd_with_sink_varlen_asm.py

Test Result

Function test:
pytest op_tests/test_fmha_fwd_with_sink_varlen_asm.py::test_fmha_fwd_with_sink_varlen_asm_correctness -v -s
test pass

Perf test:
pytest op_tests/test_fmha_fwd_with_sink_varlen_asm.py::test_fmha_fwd_with_sink_varlen_asm_perf -v -s
perf would be improved by ~2% with doubleq and kv reverse

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 4068 --add-label <label>

[valarLip]

no tg num change?

[tingchen988]

tg num would be automatically changed according to args.opt in line 252.