search

Bruce-Lee-LY / cuda_hgemm

Several optimization methods of half-precision general matrix multiplication (HGEMM) using tensor core with WMMA API and MMA PTX instruction.
MIT License
290 stars 66 forks source link

为什么B矩阵要transpose？ #10

Closed luliyucoordinate closed 5 months ago

luliyucoordinate commented 6 months ago

我注意到这里的B是T的layout，为什么这样呢？我采用N的layout，如下：

image

采用padding 16的方式，然后B reg采用row-major。然后这种做法在wmma_async_stage3.cu代码A100下测试，会有10%的性能损失。这是为什么？这里面有什么说法吗🤣