evaleev / libint

Libint: high-performance library for computing Gaussian integrals in quantum mechanics
Other
227 stars 95 forks source link

Problem with __FMA4__ intrinsics on ORNL Titan #95

Closed lukebroskop closed 6 years ago

lukebroskop commented 6 years ago

ORNL Titan - AMD bulldozer used: PrgEnv-gnu environment

COMPILATION ERROR:

/lustre/atlas/scratch/lukebr/csc249/libint/libintClone/src/lib/libint/../../../include/libint2/util/vector_x86.h:840:39: error: '_mm256_facc_pd' was not declared in this scope d.d = _mm256_facc_pd(a.d, b.d, c.d);

SOLUTION: in file "libint/include/libint2/util/vector_x86.h" see diff: diff ~/opt/build/libint/libintClone/include/libint2/util/vector_x86.h include/libint2/util/vector_x86.h 840c840 d.d = _mm256_facc_pd(a.d, b.d, c.d); --> d.d = _mm256_macc_pd(a.d, b.d, c.d); 845c845 d.d = _mm256_fsub_pd(a.d, b.d, c.d); --> d.d = _mm256_msub_pd(a.d, b.d, c.d);