I've been toying with different optimization switches and found that enabling some more agressive optimizations
FFLAGS=-O3 -xN -vec_report0 -parallel -openmp -fpp
(and also compiling cftisio and healpix with simliar options)
seems to speed up the code by quite a bit on a multiprocessor system. Does any one else have similar experiences? Is there any reason why the original Makefile used only
? Is there a concensus as to what the optimal choice of FFLAGS would be?
p.s. With either choice of FFLAGS, the output of "make check" differs slightly from the one provided.
WMAP3 likelihood code and optimization switches with Intel Compilers 8.1
1 post • Page 1 of 1