search

JeffersonLab / qphix

QCD for Intel Xeon Phi and Xeon processors
http://jeffersonlab.github.io/qphix/
Other
13 stars 11 forks source link

Some OpenMP SIMD pragmas in blas_new_c do not lead to correct vectorisation #117

Open kostrzewa opened 6 years ago

kostrzewa commented 6 years ago 
libs/QPhiX/include/qphix/blas_new_c.h(893): warning #15552: loop was not vectorized with "simd"
libs/QPhiX/include/qphix/blas_new_c.h(893): warning #15552: loop was not vectorized with "simd"
libs/QPhiX/include/qphix/blas_new_c.h(893): warning #15552: loop was not vectorized with "simd"
libs/QPhiX/include/qphix/blas_new_c.h(893): warning #15552: loop was not vectorized with "simd"
libs/QPhiX/include/qphix/blas_new_c.h(893): warning #15552: loop was not vectorized with "simd"
libs/QPhiX/include/qphix/blas_new_c.h(893): warning #15552: loop was not vectorized with "simd"
libs/QPhiX/include/qphix/blas_new_c.h(966): warning #15552: loop was not vectorized with "simd"
libs/QPhiX/include/qphix/blas_new_c.h(966): warning #15552: loop was not vectorized with "simd"
libs/QPhiX/include/qphix/blas_new_c.h(966): warning #15552: loop was not vectorized with "simd"
libs/QPhiX/include/qphix/blas_new_c.h(966): warning #15552: loop was not vectorized with "simd"
libs/QPhiX/include/qphix/blas_new_c.h(966): warning #15552: loop was not vectorized with "simd"
libs/QPhiX/include/qphix/blas_new_c.h(966): warning #15552: loop was not vectorized with "simd"
libs/QPhiX/include/qphix/blas_new_c.h(966): warning #15552: loop was not vectorized with "simd"
libs/QPhiX/include/qphix/blas_new_c.h(966): warning #15552: loop was not vectorized with "simd"
libs/QPhiX/include/qphix/blas_new_c.h(966): warning #15552: loop was not vectorized with "simd"
libs/QPhiX/include/qphix/blas_new_c.h(966): warning #15552: loop was not vectorized with "simd"
libs/QPhiX/include/qphix/blas_new_c.h(966): warning #15552: loop was not vectorized with "simd"
libs/QPhiX/include/qphix/blas_new_c.h(966): warning #15552: loop was not vectorized with "simd"

Compilation done using Intel 18.0.2.199 Build 20180210 for KNL.

kostrzewa commented 6 years ago

Could it be that this leads to a significant performance regression in the mixed precision solver? Assigning @azrael417 since this was recently changed.

kostrzewa commented 6 years ago

Actually, it seems that performance is unchanged.