wgmma is an instruction that performs warp-level matrix multiply-and-accumulate. It is only supported by Hopper and later CUDA GPUs:
Interesting tutorial: https://research.colfax-intl.com/cutlass-tutorial-wgmma-hopper/
wgmma is an instruction that performs warp-level matrix multiply-and-accumulate. It is only supported by Hopper and later CUDA GPUs:
Interesting tutorial: https://research.colfax-intl.com/cutlass-tutorial-wgmma-hopper/