ARM performance library comparisons about openblas HOT 4 OPEN

I don't have data for that right now , best you do the comparison on the hardware and functions you want to use. I expect performance will be pretty similar for GEMM, but OpenBLAS does not yet have SVE kernels for every function where it would make sense (e.g. complex dot product has fairly poor performance)

from openblas.

I'd think DOT and AXPY, as a number of other BLAS functions can/will be implemented in terms of them. Maybe TRMM for completeness (will not necessarily be close to GEMM). Of course M1 is a bit special as it does not provide SVE support - on the other hand, if you (can) include Apple's own Accelerate library in your comparison you gain access to the (officially) undocumented matrix math coprocessor (AMX).

from openblas.

I will probably run some test on M1 chips since both can be easily installed on my Mac. What other functions you suggest I can test except GEMM?

Thanks,

Jianshu

from openblas.