GFX1250 Gluon MoE A4W4 Kernel

[farlukas]

Motivation

Port MoE A4W4 kernel from Triton to Gluon for GFX1250.

Technical Details

Follows the same function signature as Triton version but using TDM features in Gluon.

Test Plan

Added backend parameter to the Triton unit tests to switch between Triton and Gluon.

pytest op_tests/triton_tests/moe/test_moe_gemm_a4w4.py # runs both triton and gluon tests
pytest op_tests/triton_tests/moe/test_moe_gemm_a4w4.py -k "gluon" # runs just gluon tests
pytest op_tests/triton_tests/moe/test_moe_gemm_a4w4.py -k "triton" # runs just triton tests

Test Result

The following unit tests has passed:

• Base (no HBM swizzling, no gather, no scatter, no gammas, no activations, no fused quant)
• With HBM swizzling
• With gather
• With scatter
• With gammas
• With swiglu activation
• With fused quant

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 (only when aiter/ops/triton/** or related paths are changed)

Extended tests (opt-in via labels):

Label	Tests
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 2337 --add-label <label>