Adaptive Approach for Basic Linear Algebra Subprograms

¨ Do a parameter study of the operation on the target machine, done once. ¨ Only generated code is on-chip multiply ¨ BLAS operation written in terms of generated on-chip multiply ¨ All tranpose cases coerced through data copy to 1 case of on-chip multiply ã Only 1 case generated per platform

N

K

N

A

M

C

B

M

*

K

NB