[Feature]: AITER fused-moe to support inter_dim 384 for amd/MiniMax-M2.1-MXFP4 tp=4 (used in vLLM)

[hongxiayang]

(1) Context: vllm issue report vllm-project/vllm#35637

Sample error message:

RuntimeError: wrong! device_gemm with the specified compilation parameters does not support this GEMM problem

(2) Tuning for fused_moe with inter_dim 384 failed.

Sample entry for tuning:

token,model_dim,inter_dim,expert,topk,act_type,dtype,q_dtype_a,q_dtype_w,q_type,use_g1u1,doweight_stage1
2048,3072,384,256,8,ActivationType.Silu,torch.bfloat16,torch.float4_e2m1fn_x2,torch.float4_e2m1fn_x2,QuantType.per_1x32,1,0

(3) Feature request:
Supporting this use case in AITER can enable vLLM to continue using the aiter-fused-moe for better performance.

Thank you.

Operating System

No response

GPU

gfx950

ROCm Component

No response

[hongxiayang]

cc @valarLip