projectNe10/Ne10

sgemm on Ne10

Opened this issue · 2 comments

Only 22, 33, 4*4 sgemm are supported on Ne10, what the other sizes of matmul?

The sgemm is an important function in BLAS. Ne10 does not provide it but there are some handy functions that can be combined into a sgemm.

Can you elaborate on those other functions to implement sgemm? Is mat_mul the only sgemm equivalent?