#250 terrible serial SYRK performance

[R. Clint Whaley]

The 3.11.38 (and prior releases, not sure how far back) all have a stupid error that will roughly halve serial SYRK/HERK performance (may have some affect on parallel perf, not sure).

This is fixed in 3.11.39, but in the meantime, you can fix this error by adding the line

define USE_AMM 1

to ATLAS/include/atlas_kern3.h