add a "using MKL" documentation page, add a minimal documentation of PARDISO wrapper classes, refine a bit the EIGEN_USE_* logic

Eigen/src/Core/util/MKL_support.h

@@ -37,15 +37,19 @@
   #ifndef EIGEN_USE_BLAS
     #define EIGEN_USE_BLAS
   #endif
-  #ifndef EIGEN_USE_MKL_CLAPACK
-    #define EIGEN_USE_MKL_CLAPACK
+  #ifndef EIGEN_USE_LAPACKE
+    #define EIGEN_USE_LAPACKE
   #endif
   #ifndef EIGEN_USE_MKL_VML
     #define EIGEN_USE_MKL_VML
   #endif
 #endif
 
-#if defined(EIGEN_USE_MKL_CLAPACK) || defined(EIGEN_USE_MKL_VML)
+#ifdef EIGEN_USE_LAPACKE_STRICT
+  #define EIGEN_USE_LAPACKE
+#endif
+
+#if defined(EIGEN_USE_BLAS) || defined(EIGEN_USE_LAPACKE) || defined(EIGEN_USE_MKL_VML)
   #define EIGEN_USE_MKL
 #endif
 
@@ -55,17 +59,6 @@
 #include <mkl_lapacke.h>
 #define EIGEN_MKL_VML_THRESHOLD 128
 
-#elif defined EIGEN_USE_BLAS
-
-#error Currently EIGEN_USE_BLAS requires Intel MKL. If you want to use MKL's BLAS and only it, then define EIGEN_USE_MKL too.
-
-#include "../../misc/blas.h"
-#define MKL_INT int
-
-#endif
-
-#if defined(EIGEN_USE_MKL) || defined(EIGEN_USE_BLAS)
-
 namespace Eigen {
 
 typedef std::complex<double> dcomplex;
@@ -83,7 +76,6 @@ static inline void assign_conj_scalar_eig2mkl(MKLType& mklScalar, const EigenTyp
   mklScalar=eigenScalar;
 }
 
-#ifdef EIGEN_USE_MKL
 template <>
 inline void assign_scalar_eig2mkl<MKL_Complex16,dcomplex>(MKL_Complex16& mklScalar, const dcomplex& eigenScalar) {
   mklScalar.real=eigenScalar.real();
@@ -108,8 +100,6 @@ inline void assign_conj_scalar_eig2mkl<MKL_Complex8,scomplex>(MKL_Complex8& mklS
   mklScalar.imag=-eigenScalar.imag();
 }
 
-#endif
-
 } // end namespace internal
 
 } // end namespace Eigen

Eigen/src/PARDISOSupport/PARDISOSupport.h

@@ -312,6 +312,18 @@ bool PardisoImpl<Base>::_solve(const MatrixBase<BDerived> &b,
 }
 
 
+/** \ingroup PARDISOSupport_Module
+  * \class PardisoLU
+  * \brief A sparse direct LU factorization and solver based on the PARDISO library
+  *
+  * This class allows to solve for A.X = B sparse linear problems via a direct LU factorization
+  * using the Intel MKL PARDISO library. The sparse matrix A must be squared and invertible.
+  * The vectors or matrices X and B can be either dense or sparse.
+  *
+  * \tparam _MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
+  *
+  * \sa \ref TutorialSparseDirectSolvers
+  */
 template<typename MatrixType>
 class PardisoLU : public PardisoImpl< PardisoLU<MatrixType> >
 {
@@ -340,6 +352,18 @@ class PardisoLU : public PardisoImpl< PardisoLU<MatrixType> >
     }
 };
 
+/** \ingroup PARDISOSupport_Module
+  * \class PardisoLLT
+  * \brief A sparse direct Cholesky (LLT) factorization and solver based on the PARDISO library
+  *
+  * This class allows to solve for A.X = B sparse linear problems via a LL^T Cholesky factorization
+  * using the Intel MKL PARDISO library. The sparse matrix A must be selfajoint and positive definite.
+  * The vectors or matrices X and B can be either dense or sparse.
+  *
+  * \tparam _MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
+  *
+  * \sa \ref TutorialSparseDirectSolvers
+  */
 template<typename MatrixType>
 class PardisoLLT : public PardisoImpl< PardisoLLT<MatrixType> >
 {
@@ -368,6 +392,18 @@ class PardisoLLT : public PardisoImpl< PardisoLLT<MatrixType> >
     }
 };
 
+/** \ingroup PARDISOSupport_Module
+  * \class PardisoLDLT
+  * \brief A sparse direct Cholesky (LLT) factorization and solver based on the PARDISO library
+  *
+  * This class allows to solve for A.X = B sparse linear problems via a LDL^T Cholesky factorization
+  * using the Intel MKL PARDISO library. The sparse matrix A must be selfajoint and positive definite.
+  * The vectors or matrices X and B can be either dense or sparse.
+  *
+  * \tparam _MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
+  *
+  * \sa \ref TutorialSparseDirectSolvers
+  */
 template<typename MatrixType>
 class PardisoLDLT : public PardisoImpl< PardisoLDLT<MatrixType> >
 {