KNL is now available on Stampede. KNL supports AVX-512F, AVX_512CD, AVX-512ER and AVX-512PF. Since the only division within an inner loop occurs in mocks/DDrppi_mocks, 512ER is only applicable there. There are no prefetch calls within the code yet, I will need to use Vtune to check if prefetch helps performance in the current incarnation.
Might as well add AVX2 abilities, since the login nodes for the KNL cluster are Haswell chips.
KNL is now available on Stampede. KNL supports
AVX-512F
,AVX_512CD
,AVX-512ER
andAVX-512PF
. Since the only division within an inner loop occurs inmocks/DDrppi_mocks
,512ER
is only applicable there. There are no prefetch calls within the code yet, I will need to use Vtune to check if prefetch helps performance in the current incarnation.Might as well add
AVX2
abilities, since the login nodes for the KNL cluster are Haswell chips.