Intel MKL is threaded extensively for different domains. The threaded function list includes:

 Domain Threaded Functions Direct sparse solver PARDISO Interface RoutinesDSS Interface Routines LAPACK Linear equations, computational routines: - factorization: *getrf, *gbtrf, *potrf, *pptrf, *sytrf, *hetrf, *sptrf, *hptrf - solving: *gbtrs, *gttrs, *pptrs, *pbtrs, *pttrs, *sytrs, *sptrs, *hptrs, *tptrs, *tbtrsOrthogonal factorization, computational routines: *geqrf, *ormqr, *unmqr, *ormlq, *unmlq, *ormql, *unmql, *ormrq, *unmrq.Singular Value Decomposition, computational routines: *gebrd, *bdsqrSymmetric Eigenvalue Problems, computational routines: *sytrd, *hetrd, *sptrd, *hptrd, *steqr, *stedc BLAS Level1 BLAS: *axpy, *copy, *swap, ddot/sdot, drot/srotLevel2 BLAS: *gemv, *trmv, dsyr/ssyr, dsyr2/ssyr2, dsymv/ssymvAll Level 3 BLAS and all Sparse BLAS routines are threaded except Level 2 triangular solvers VML All Mathematical functions except the following are threaded:    Pack/Unpack family, Rounding family, Add, Sub, Mul, real Abs, Sqr FFT DFT transform are threaded except the following:1)Real 1D are not threaded 2)Split-complex 1Ds are not threaded 3)Small multidimensional transforms are not threaded

Note 1:  A number of other LAPACK routines, which are based on threaded LAPACK or BLAS routines, make effective use of parallelism: *gesv, *posv, *gels, *gesvd,*syev, *heev, etc.

Note 2: Level1 BLAS  and Level2 BLAS are threaded only for:
1) Intel® 64 architecture
2) Intel® CoreTM2 Duo and Intel® CoreTM i7 processors

Note 3: all data provided into this article are relevant to the latest MKL.

Note 4: The functions are not threaded when small vector/matrix size. For example, VML function are expected to be threaded for vector length > 16K.

Для получения подробной информации о возможностях оптимизации компилятора обратитесь к нашему Уведомлению об оптимизации.
Возможность комментирования русскоязычного контента была отключена. Узнать подробнее.