It does if isa(A, StridedVecOrMat), they have unit stride along the first dimension, and the element type is supported by BLAS (Float32, Float64, Complex64, Complex128). These are all simply restrictions of BLAS, not limitations of julia.
When those don't hold, there are at least cache-optimized implementations of a few algorithms, like matrix-matrix multiplication, already in Base. Not as fast as BLAS, but not terrible either. How are you creating your subarrays? Best, --Tim On Monday, April 04, 2016 11:19:34 AM Gregor Robinson wrote: > Yep. But, as far as I can tell, BLAS doesn't play with SubMatrix. > > On Monday, April 4, 2016 at 6:11:52 AM UTC-6, Tim Holy wrote: > > Have you tried `sub`? > > > > --Tim
