WMMA guide for AMD RDNA 4 architecture GPUs - part 1

Practical guide to fusing GEMMs on AMD RDNA™ 4 architecture, covering WMMA layout, a transpose-by-swapping A/B technique, HIP sample code, and hipBLAS-verified results used in Llama.cpp.
评论
?
参与讨论

Practical guide to fusing GEMMs on AMD RDNA™ 4 architecture, covering WMMA layout, a transpose-by-swapping A/B technique, HIP sample code, and hipBLAS-verified results used in Llama.cpp.