Does llama.cpp provide quant weights in one column of Matrix A?

[ErvinXie]

I am researching quant of llama.cpp. It seems all quant types quant weights in one row of Matrix A and weights in one column of Matrix B when calculating C = A x B. Does llama.cpp provide SGEMM of column major quant?

[ErvinXie]

According to this paper, per channel quant work better for K state. Per channel quant is a column major quant of Matrix A. So i wonder whether llama.cpp support this per channel quant.