next up previous contents index
Next: The BLAS as the Up: Factors that Affect Performance Previous: Data Movement   Contents   Index


The nested loop structure of most linear algebra algorithms offers considerable scope for loop-based parallelism on shared memory machines. This is the principal type of parallelism that LAPACK at present aims to exploit. It can sometimes be generated automatically by a compiler, but often requires the insertion of compiler directives.

Susan Blackford