[ROCm][Perf] Fuse MXFP8 activation quant into the 1x32 dot_scaled linear for M <= 16 - #63915

[ROCm][Perf] Fuse MXFP8 activation quant into the 1x32 dot_scaled linear for M <= 16#63915

aoshen02 merged 3 commits into

vllm-project:mainvllm-project/vllm:mainfrom

RuibinCheung:cursor/rocm-mxfp8-linear-fused-act-quant-45a8RuibinCheung/vllm:cursor/rocm-mxfp8-linear-fused-act-quant-45a8Copy head branch name to clipboard