This issue is an evolving document describing the design and implementation of using sparse matrix instructions for optimizing skinny GEMM on AMDGPU in IREE. Using FP8 as an example: currently skinny ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results