Closed hughperkins closed 8 years ago
Yes, that's right. AutoGemm is capable of generating row major kernels. However the gemm implementation can, for example, from row-major NT to column-major TN. So, AutoGemm only needs to generate column-major kernels for clBLAS.
Ah, ok, I see. The implementation has seemed to be using AutoGemm for row major kernels that I've passed to it, as far as I can see, which makes more sense now :-)
Per https://github.com/clMathLibraries/clBLAS/wiki/AutoGemm#autogemm-parameters , it looks like autogemm is only run for the orders specified in
orders
parameter ofAutoGemmParameters.py
. However, looking at AutoGemmParameters.py, orders is specified as, https://github.com/clMathLibraries/clBLAS/blob/master/src/library/blas/AutoGemm/AutoGemmParameters.py#L74 :Does this mean that AutoGemm is only activated for ColumnMajor matrices?