From 800ms to ~25ms: harness-driven optimization of a CUDA matmul kernel(github.com)
3 pointsby icyace5 hours ago1 comment