Using Intel® MKL from Eigen
Since Eigen version 3.1 and later, users can benefit from built-in Intel® Math Kernel Library (MKL) optimizations with an installed copy of Intel MKL 10.3 (or later).
Intel MKL provides highly optimized multi-threaded mathematical routines for x86-compatible architectures. Intel MKL is available on Linux, Mac and Windows for both Intel64 and IA32 architectures.
Using Intel MKL through Eigen is easy:
- define the
EIGEN_USE_MKL_ALL
macro before including any Eigen's header - link your program to MKL libraries (see the MKL linking advisor)
- on a 64bits system, you must use the LP64 interface (not the ILP64 one)
When doing so, a number of Eigen's algorithms are silently substituted with calls to Intel MKL routines. These substitutions apply only for Dynamic or large enough objects with one of the following four standard scalar types: float
, double
, complex<float>
, and complex<double>
. Operations on other scalar types or mixing reals and complexes will continue to use the built-in algorithms.
In addition you can choose which parts will be substituted by defining one or multiple of the following macros:
EIGEN_USE_BLAS | Enables the use of external BLAS level 2 and 3 routines |
EIGEN_USE_LAPACKE | Enables the use of external Lapack routines via the Lapacke C interface to Lapack |
EIGEN_USE_LAPACKE_STRICT | Same as EIGEN_USE_LAPACKE but algorithm of lower robustness are disabled.This currently concerns only JacobiSVD which otherwise would be replaced by gesvd that is less robust than Jacobi rotations. |
EIGEN_USE_MKL_VML | Enables the use of Intel VML (vector operations) |
EIGEN_USE_MKL_ALL | Defines EIGEN_USE_BLAS , EIGEN_USE_LAPACKE , and EIGEN_USE_MKL_VML |
The EIGEN_USE_BLAS
and EIGEN_USE_LAPACKE*
macros can be combined with EIGEN_USE_MKL
to explicitly tell Eigen that the underlying BLAS/Lapack implementation is Intel MKL. The main effect is to enable MKL direct call feature (MKL_DIRECT_CALL
). This may help to increase performance of some MKL BLAS (?GEMM, ?GEMV, ?TRSM, ?AXPY and ?DOT) and LAPACK (LU, Cholesky and QR) routines for very small matrices. MKL direct call can be disabled by defining EIGEN_MKL_NO_DIRECT_CALL
.
Note that the BLAS and LAPACKE backends can be enabled for any F77 compatible BLAS and LAPACK libraries. See this page for the details.
Finally, the PARDISO sparse solver shipped with Intel MKL can be used through the PardisoLU, PardisoLLT and PardisoLDLT classes of the PardisoSupport module.
The following table summarizes the list of functions covered by EIGEN_USE_MKL_VML:
Code example | MKL routines |
---|---|
v2=v1.array().sin(); v2=v1.array().asin(); v2=v1.array().cos(); v2=v1.array().acos(); v2=v1.array().tan(); v2=v1.array().exp(); v2=v1.array().log(); v2=v1.array().sqrt(); v2=v1.array().square(); v2=v1.array().pow(1.5); | v?Sin v?Asin v?Cos v?Acos v?Tan v?Exp v?Ln v?Sqrt v?Sqr v?Powx |
In the examples, v1 and v2 are dense vectors.
Links
- Intel MKL can be purchased and downloaded here.
- Intel MKL is also bundled with Intel Composer XE.