pylibsparseir 0.1.0__cp310-cp310-macosx_15_0_arm64.whl → 0.6.0__cp310-cp310-macosx_15_0_arm64.whl

include/eigen3/Eigen/src/Core/products/GeneralMatrixMatrixTriangular.h

@@ -0,0 +1,317 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2009-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_GENERAL_MATRIX_MATRIX_TRIANGULAR_H
+#define EIGEN_GENERAL_MATRIX_MATRIX_TRIANGULAR_H
+
+namespace Eigen { 
+
+template<typename Scalar, typename Index, int StorageOrder, int UpLo, bool ConjLhs, bool ConjRhs>
+struct selfadjoint_rank1_update;
+
+namespace internal {
+
+/**********************************************************************
+* This file implements a general A * B product while
+* evaluating only one triangular part of the product.
+* This is a more general version of self adjoint product (C += A A^T)
+* as the level 3 SYRK Blas routine.
+**********************************************************************/
+
+// forward declarations (defined at the end of this file)
+template<typename LhsScalar, typename RhsScalar, typename Index, int mr, int nr, bool ConjLhs, bool ConjRhs, int ResInnerStride, int UpLo>
+struct tribb_kernel;
+  
+/* Optimized matrix-matrix product evaluating only one triangular half */
+template <typename Index,
+          typename LhsScalar, int LhsStorageOrder, bool ConjugateLhs,
+          typename RhsScalar, int RhsStorageOrder, bool ConjugateRhs,
+                              int ResStorageOrder, int ResInnerStride, int  UpLo, int Version = Specialized>
+struct general_matrix_matrix_triangular_product;
+
+// as usual if the result is row major => we transpose the product
+template <typename Index, typename LhsScalar, int LhsStorageOrder, bool ConjugateLhs,
+                          typename RhsScalar, int RhsStorageOrder, bool ConjugateRhs,
+                          int ResInnerStride, int  UpLo, int Version>
+struct general_matrix_matrix_triangular_product<Index,LhsScalar,LhsStorageOrder,ConjugateLhs,RhsScalar,RhsStorageOrder,ConjugateRhs,RowMajor,ResInnerStride,UpLo,Version>
+{
+  typedef typename ScalarBinaryOpTraits<LhsScalar, RhsScalar>::ReturnType ResScalar;
+  static EIGEN_STRONG_INLINE void run(Index size, Index depth,const LhsScalar* lhs, Index lhsStride,
+                                      const RhsScalar* rhs, Index rhsStride, ResScalar* res, Index resIncr, Index resStride,
+                                      const ResScalar& alpha, level3_blocking<RhsScalar,LhsScalar>& blocking)
+  {
+    general_matrix_matrix_triangular_product<Index,
+        RhsScalar, RhsStorageOrder==RowMajor ? ColMajor : RowMajor, ConjugateRhs,
+        LhsScalar, LhsStorageOrder==RowMajor ? ColMajor : RowMajor, ConjugateLhs,
+        ColMajor, ResInnerStride, UpLo==Lower?Upper:Lower>
+      ::run(size,depth,rhs,rhsStride,lhs,lhsStride,res,resIncr,resStride,alpha,blocking);
+  }
+};
+
+template <typename Index, typename LhsScalar, int LhsStorageOrder, bool ConjugateLhs,
+                          typename RhsScalar, int RhsStorageOrder, bool ConjugateRhs,
+                          int ResInnerStride, int  UpLo, int Version>
+struct general_matrix_matrix_triangular_product<Index,LhsScalar,LhsStorageOrder,ConjugateLhs,RhsScalar,RhsStorageOrder,ConjugateRhs,ColMajor,ResInnerStride,UpLo,Version>
+{
+  typedef typename ScalarBinaryOpTraits<LhsScalar, RhsScalar>::ReturnType ResScalar;
+  static EIGEN_STRONG_INLINE void run(Index size, Index depth,const LhsScalar* _lhs, Index lhsStride,
+                                      const RhsScalar* _rhs, Index rhsStride,
+                                      ResScalar* _res, Index resIncr, Index resStride,
+                                      const ResScalar& alpha, level3_blocking<LhsScalar,RhsScalar>& blocking)
+  {
+    typedef gebp_traits<LhsScalar,RhsScalar> Traits;
+
+    typedef const_blas_data_mapper<LhsScalar, Index, LhsStorageOrder> LhsMapper;
+    typedef const_blas_data_mapper<RhsScalar, Index, RhsStorageOrder> RhsMapper;
+    typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor, Unaligned, ResInnerStride> ResMapper;
+    LhsMapper lhs(_lhs,lhsStride);
+    RhsMapper rhs(_rhs,rhsStride);
+    ResMapper res(_res, resStride, resIncr);
+
+    Index kc = blocking.kc();
+    Index mc = (std::min)(size,blocking.mc());
+
+    // !!! mc must be a multiple of nr:
+    if(mc > Traits::nr)
+      mc = (mc/Traits::nr)*Traits::nr;
+
+    std::size_t sizeA = kc*mc;
+    std::size_t sizeB = kc*size;
+
+    ei_declare_aligned_stack_constructed_variable(LhsScalar, blockA, sizeA, blocking.blockA());
+    ei_declare_aligned_stack_constructed_variable(RhsScalar, blockB, sizeB, blocking.blockB());
+
+    gemm_pack_lhs<LhsScalar, Index, LhsMapper, Traits::mr, Traits::LhsProgress, typename Traits::LhsPacket4Packing, LhsStorageOrder> pack_lhs;
+    gemm_pack_rhs<RhsScalar, Index, RhsMapper, Traits::nr, RhsStorageOrder> pack_rhs;
+    gebp_kernel<LhsScalar, RhsScalar, Index, ResMapper, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs> gebp;
+    tribb_kernel<LhsScalar, RhsScalar, Index, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs, ResInnerStride, UpLo> sybb;
+
+    for(Index k2=0; k2<depth; k2+=kc)
+    {
+      const Index actual_kc = (std::min)(k2+kc,depth)-k2;
+
+      // note that the actual rhs is the transpose/adjoint of mat
+      pack_rhs(blockB, rhs.getSubMapper(k2,0), actual_kc, size);
+
+      for(Index i2=0; i2<size; i2+=mc)
+      {
+        const Index actual_mc = (std::min)(i2+mc,size)-i2;
+
+        pack_lhs(blockA, lhs.getSubMapper(i2, k2), actual_kc, actual_mc);
+
+        // the selected actual_mc * size panel of res is split into three different part:
+        //  1 - before the diagonal => processed with gebp or skipped
+        //  2 - the actual_mc x actual_mc symmetric block => processed with a special kernel
+        //  3 - after the diagonal => processed with gebp or skipped
+        if (UpLo==Lower)
+          gebp(res.getSubMapper(i2, 0), blockA, blockB, actual_mc, actual_kc,
+               (std::min)(size,i2), alpha, -1, -1, 0, 0);
+
+        sybb(_res+resStride*i2 + resIncr*i2, resIncr, resStride, blockA, blockB + actual_kc*i2, actual_mc, actual_kc, alpha);
+
+        if (UpLo==Upper)
+        {
+          Index j2 = i2+actual_mc;
+          gebp(res.getSubMapper(i2, j2), blockA, blockB+actual_kc*j2, actual_mc,
+               actual_kc, (std::max)(Index(0), size-j2), alpha, -1, -1, 0, 0);
+        }
+      }
+    }
+  }
+};
+
+// Optimized packed Block * packed Block product kernel evaluating only one given triangular part
+// This kernel is built on top of the gebp kernel:
+// - the current destination block is processed per panel of actual_mc x BlockSize
+//   where BlockSize is set to the minimal value allowing gebp to be as fast as possible
+// - then, as usual, each panel is split into three parts along the diagonal,
+//   the sub blocks above and below the diagonal are processed as usual,
+//   while the triangular block overlapping the diagonal is evaluated into a
+//   small temporary buffer which is then accumulated into the result using a
+//   triangular traversal.
+template<typename LhsScalar, typename RhsScalar, typename Index, int mr, int nr, bool ConjLhs, bool ConjRhs, int ResInnerStride, int UpLo>
+struct tribb_kernel
+{
+  typedef gebp_traits<LhsScalar,RhsScalar,ConjLhs,ConjRhs> Traits;
+  typedef typename Traits::ResScalar ResScalar;
+
+  enum {
+    BlockSize  = meta_least_common_multiple<EIGEN_PLAIN_ENUM_MAX(mr,nr),EIGEN_PLAIN_ENUM_MIN(mr,nr)>::ret
+  };
+  void operator()(ResScalar* _res, Index resIncr, Index resStride, const LhsScalar* blockA, const RhsScalar* blockB, Index size, Index depth, const ResScalar& alpha)
+  {
+    typedef blas_data_mapper<ResScalar, Index, ColMajor, Unaligned, ResInnerStride> ResMapper;
+    typedef blas_data_mapper<ResScalar, Index, ColMajor, Unaligned> BufferMapper;
+    ResMapper res(_res, resStride, resIncr);
+    gebp_kernel<LhsScalar, RhsScalar, Index, ResMapper, mr, nr, ConjLhs, ConjRhs> gebp_kernel1;
+    gebp_kernel<LhsScalar, RhsScalar, Index, BufferMapper, mr, nr, ConjLhs, ConjRhs> gebp_kernel2;
+
+    Matrix<ResScalar,BlockSize,BlockSize,ColMajor> buffer((internal::constructor_without_unaligned_array_assert()));
+
+    // let's process the block per panel of actual_mc x BlockSize,
+    // again, each is split into three parts, etc.
+    for (Index j=0; j<size; j+=BlockSize)
+    {
+      Index actualBlockSize = std::min<Index>(BlockSize,size - j);
+      const RhsScalar* actual_b = blockB+j*depth;
+
+      if(UpLo==Upper)
+        gebp_kernel1(res.getSubMapper(0, j), blockA, actual_b, j, depth, actualBlockSize, alpha,
+                     -1, -1, 0, 0);
+      
+      // selfadjoint micro block
+      {
+        Index i = j;
+        buffer.setZero();
+        // 1 - apply the kernel on the temporary buffer
+        gebp_kernel2(BufferMapper(buffer.data(), BlockSize), blockA+depth*i, actual_b, actualBlockSize, depth, actualBlockSize, alpha,
+                     -1, -1, 0, 0);
+
+        // 2 - triangular accumulation
+        for(Index j1=0; j1<actualBlockSize; ++j1)
+        {
+          typename ResMapper::LinearMapper r = res.getLinearMapper(i,j+j1);
+          for(Index i1=UpLo==Lower ? j1 : 0;
+              UpLo==Lower ? i1<actualBlockSize : i1<=j1; ++i1)
+            r(i1) += buffer(i1,j1);
+        }
+      }
+
+      if(UpLo==Lower)
+      {
+        Index i = j+actualBlockSize;
+        gebp_kernel1(res.getSubMapper(i, j), blockA+depth*i, actual_b, size-i, 
+                     depth, actualBlockSize, alpha, -1, -1, 0, 0);
+      }
+    }
+  }
+};
+
+} // end namespace internal
+
+// high level API
+
+template<typename MatrixType, typename ProductType, int UpLo, bool IsOuterProduct>
+struct general_product_to_triangular_selector;
+
+
+template<typename MatrixType, typename ProductType, int UpLo>
+struct general_product_to_triangular_selector<MatrixType,ProductType,UpLo,true>
+{
+  static void run(MatrixType& mat, const ProductType& prod, const typename MatrixType::Scalar& alpha, bool beta)
+  {
+    typedef typename MatrixType::Scalar Scalar;
+    
+    typedef typename internal::remove_all<typename ProductType::LhsNested>::type Lhs;
+    typedef internal::blas_traits<Lhs> LhsBlasTraits;
+    typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhs;
+    typedef typename internal::remove_all<ActualLhs>::type _ActualLhs;
+    typename internal::add_const_on_value_type<ActualLhs>::type actualLhs = LhsBlasTraits::extract(prod.lhs());
+    
+    typedef typename internal::remove_all<typename ProductType::RhsNested>::type Rhs;
+    typedef internal::blas_traits<Rhs> RhsBlasTraits;
+    typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhs;
+    typedef typename internal::remove_all<ActualRhs>::type _ActualRhs;
+    typename internal::add_const_on_value_type<ActualRhs>::type actualRhs = RhsBlasTraits::extract(prod.rhs());
+
+    Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs().derived()) * RhsBlasTraits::extractScalarFactor(prod.rhs().derived());
+
+    if(!beta)
+      mat.template triangularView<UpLo>().setZero();
+
+    enum {
+      StorageOrder = (internal::traits<MatrixType>::Flags&RowMajorBit) ? RowMajor : ColMajor,
+      UseLhsDirectly = _ActualLhs::InnerStrideAtCompileTime==1,
+      UseRhsDirectly = _ActualRhs::InnerStrideAtCompileTime==1
+    };
+    
+    internal::gemv_static_vector_if<Scalar,Lhs::SizeAtCompileTime,Lhs::MaxSizeAtCompileTime,!UseLhsDirectly> static_lhs;
+    ei_declare_aligned_stack_constructed_variable(Scalar, actualLhsPtr, actualLhs.size(),
+      (UseLhsDirectly ? const_cast<Scalar*>(actualLhs.data()) : static_lhs.data()));
+    if(!UseLhsDirectly) Map<typename _ActualLhs::PlainObject>(actualLhsPtr, actualLhs.size()) = actualLhs;
+    
+    internal::gemv_static_vector_if<Scalar,Rhs::SizeAtCompileTime,Rhs::MaxSizeAtCompileTime,!UseRhsDirectly> static_rhs;
+    ei_declare_aligned_stack_constructed_variable(Scalar, actualRhsPtr, actualRhs.size(),
+      (UseRhsDirectly ? const_cast<Scalar*>(actualRhs.data()) : static_rhs.data()));
+    if(!UseRhsDirectly) Map<typename _ActualRhs::PlainObject>(actualRhsPtr, actualRhs.size()) = actualRhs;
+    
+    
+    selfadjoint_rank1_update<Scalar,Index,StorageOrder,UpLo,
+                              LhsBlasTraits::NeedToConjugate && NumTraits<Scalar>::IsComplex,
+                              RhsBlasTraits::NeedToConjugate && NumTraits<Scalar>::IsComplex>
+          ::run(actualLhs.size(), mat.data(), mat.outerStride(), actualLhsPtr, actualRhsPtr, actualAlpha);
+  }
+};
+
+template<typename MatrixType, typename ProductType, int UpLo>
+struct general_product_to_triangular_selector<MatrixType,ProductType,UpLo,false>
+{
+  static void run(MatrixType& mat, const ProductType& prod, const typename MatrixType::Scalar& alpha, bool beta)
+  {
+    typedef typename internal::remove_all<typename ProductType::LhsNested>::type Lhs;
+    typedef internal::blas_traits<Lhs> LhsBlasTraits;
+    typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhs;
+    typedef typename internal::remove_all<ActualLhs>::type _ActualLhs;
+    typename internal::add_const_on_value_type<ActualLhs>::type actualLhs = LhsBlasTraits::extract(prod.lhs());
+    
+    typedef typename internal::remove_all<typename ProductType::RhsNested>::type Rhs;
+    typedef internal::blas_traits<Rhs> RhsBlasTraits;
+    typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhs;
+    typedef typename internal::remove_all<ActualRhs>::type _ActualRhs;
+    typename internal::add_const_on_value_type<ActualRhs>::type actualRhs = RhsBlasTraits::extract(prod.rhs());
+
+    typename ProductType::Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs().derived()) * RhsBlasTraits::extractScalarFactor(prod.rhs().derived());
+
+    if(!beta)
+      mat.template triangularView<UpLo>().setZero();
+
+    enum {
+      IsRowMajor = (internal::traits<MatrixType>::Flags&RowMajorBit) ? 1 : 0,
+      LhsIsRowMajor = _ActualLhs::Flags&RowMajorBit ? 1 : 0,
+      RhsIsRowMajor = _ActualRhs::Flags&RowMajorBit ? 1 : 0,
+      SkipDiag = (UpLo&(UnitDiag|ZeroDiag))!=0
+    };
+
+    Index size = mat.cols();
+    if(SkipDiag)
+      size--;
+    Index depth = actualLhs.cols();
+
+    typedef internal::gemm_blocking_space<IsRowMajor ? RowMajor : ColMajor,typename Lhs::Scalar,typename Rhs::Scalar,
+          MatrixType::MaxColsAtCompileTime, MatrixType::MaxColsAtCompileTime, _ActualRhs::MaxColsAtCompileTime> BlockingType;
+
+    BlockingType blocking(size, size, depth, 1, false);
+
+    internal::general_matrix_matrix_triangular_product<Index,
+      typename Lhs::Scalar, LhsIsRowMajor ? RowMajor : ColMajor, LhsBlasTraits::NeedToConjugate,
+      typename Rhs::Scalar, RhsIsRowMajor ? RowMajor : ColMajor, RhsBlasTraits::NeedToConjugate,
+      IsRowMajor ? RowMajor : ColMajor, MatrixType::InnerStrideAtCompileTime, UpLo&(Lower|Upper)>
+      ::run(size, depth,
+            &actualLhs.coeffRef(SkipDiag&&(UpLo&Lower)==Lower ? 1 : 0,0), actualLhs.outerStride(),
+            &actualRhs.coeffRef(0,SkipDiag&&(UpLo&Upper)==Upper ? 1 : 0), actualRhs.outerStride(),
+            mat.data() + (SkipDiag ? (bool(IsRowMajor) != ((UpLo&Lower)==Lower) ? mat.innerStride() : mat.outerStride() ) : 0),
+            mat.innerStride(), mat.outerStride(), actualAlpha, blocking);
+  }
+};
+
+template<typename MatrixType, unsigned int UpLo>
+template<typename ProductType>
+EIGEN_DEVICE_FUNC TriangularView<MatrixType,UpLo>& TriangularViewImpl<MatrixType,UpLo,Dense>::_assignProduct(const ProductType& prod, const Scalar& alpha, bool beta)
+{
+  EIGEN_STATIC_ASSERT((UpLo&UnitDiag)==0, WRITING_TO_TRIANGULAR_PART_WITH_UNIT_DIAGONAL_IS_NOT_SUPPORTED);
+  eigen_assert(derived().nestedExpression().rows() == prod.rows() && derived().cols() == prod.cols());
+
+  general_product_to_triangular_selector<MatrixType, ProductType, UpLo, internal::traits<ProductType>::InnerSize==1>::run(derived().nestedExpression().const_cast_derived(), prod, alpha, beta);
+
+  return derived();
+}
+
+} // end namespace Eigen
+
+#endif // EIGEN_GENERAL_MATRIX_MATRIX_TRIANGULAR_H

include/eigen3/Eigen/src/Core/products/GeneralMatrixMatrixTriangular_BLAS.h

@@ -0,0 +1,145 @@
+/*
+ Copyright (c) 2011, Intel Corporation. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without modification,
+ are permitted provided that the following conditions are met:
+
+ * Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+ * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+ * Neither the name of Intel Corporation nor the names of its contributors may
+   be used to endorse or promote products derived from this software without
+   specific prior written permission.
+
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+ ********************************************************************************
+ *   Content : Eigen bindings to BLAS F77
+ *   Level 3 BLAS SYRK/HERK implementation.
+ ********************************************************************************
+*/
+
+#ifndef EIGEN_GENERAL_MATRIX_MATRIX_TRIANGULAR_BLAS_H
+#define EIGEN_GENERAL_MATRIX_MATRIX_TRIANGULAR_BLAS_H
+
+namespace Eigen {
+
+namespace internal {
+
+template <typename Index, typename Scalar, int AStorageOrder, bool ConjugateA, int ResStorageOrder, int UpLo>
+struct general_matrix_matrix_rankupdate :
+       general_matrix_matrix_triangular_product<
+         Index,Scalar,AStorageOrder,ConjugateA,Scalar,AStorageOrder,ConjugateA,ResStorageOrder,1,UpLo,BuiltIn> {};
+
+
+// try to go to BLAS specialization
+#define EIGEN_BLAS_RANKUPDATE_SPECIALIZE(Scalar) \
+template <typename Index, int LhsStorageOrder, bool ConjugateLhs, \
+                          int RhsStorageOrder, bool ConjugateRhs, int  UpLo> \
+struct general_matrix_matrix_triangular_product<Index,Scalar,LhsStorageOrder,ConjugateLhs, \
+               Scalar,RhsStorageOrder,ConjugateRhs,ColMajor,1,UpLo,Specialized> { \
+  static EIGEN_STRONG_INLINE void run(Index size, Index depth,const Scalar* lhs, Index lhsStride, \
+                          const Scalar* rhs, Index rhsStride, Scalar* res, Index resIncr, Index resStride, Scalar alpha, level3_blocking<Scalar, Scalar>& blocking) \
+  { \
+    if ( lhs==rhs && ((UpLo&(Lower|Upper))==UpLo) ) { \
+      general_matrix_matrix_rankupdate<Index,Scalar,LhsStorageOrder,ConjugateLhs,ColMajor,UpLo> \
+      ::run(size,depth,lhs,lhsStride,rhs,rhsStride,res,resStride,alpha,blocking); \
+    } else { \
+      general_matrix_matrix_triangular_product<Index, \
+        Scalar, LhsStorageOrder, ConjugateLhs, \
+        Scalar, RhsStorageOrder, ConjugateRhs, \
+        ColMajor, 1, UpLo, BuiltIn> \
+      ::run(size,depth,lhs,lhsStride,rhs,rhsStride,res,resIncr,resStride,alpha,blocking); \
+    } \
+  } \
+};
+
+EIGEN_BLAS_RANKUPDATE_SPECIALIZE(double)
+EIGEN_BLAS_RANKUPDATE_SPECIALIZE(float)
+// TODO handle complex cases
+// EIGEN_BLAS_RANKUPDATE_SPECIALIZE(dcomplex)
+// EIGEN_BLAS_RANKUPDATE_SPECIALIZE(scomplex)
+
+// SYRK for float/double
+#define EIGEN_BLAS_RANKUPDATE_R(EIGTYPE, BLASTYPE, BLASFUNC) \
+template <typename Index, int AStorageOrder, bool ConjugateA, int  UpLo> \
+struct general_matrix_matrix_rankupdate<Index,EIGTYPE,AStorageOrder,ConjugateA,ColMajor,UpLo> { \
+  enum { \
+    IsLower = (UpLo&Lower) == Lower, \
+    LowUp = IsLower ? Lower : Upper, \
+    conjA = ((AStorageOrder==ColMajor) && ConjugateA) ? 1 : 0 \
+  }; \
+  static EIGEN_STRONG_INLINE void run(Index size, Index depth,const EIGTYPE* lhs, Index lhsStride, \
+                          const EIGTYPE* /*rhs*/, Index /*rhsStride*/, EIGTYPE* res, Index resStride, EIGTYPE alpha, level3_blocking<EIGTYPE, EIGTYPE>& /*blocking*/) \
+  { \
+  /* typedef Matrix<EIGTYPE, Dynamic, Dynamic, RhsStorageOrder> MatrixRhs;*/ \
+\
+   BlasIndex lda=convert_index<BlasIndex>(lhsStride), ldc=convert_index<BlasIndex>(resStride), n=convert_index<BlasIndex>(size), k=convert_index<BlasIndex>(depth); \
+   char uplo=((IsLower) ? 'L' : 'U'), trans=((AStorageOrder==RowMajor) ? 'T':'N'); \
+   EIGTYPE beta(1); \
+   BLASFUNC(&uplo, &trans, &n, &k, (const BLASTYPE*)&numext::real_ref(alpha), lhs, &lda, (const BLASTYPE*)&numext::real_ref(beta), res, &ldc); \
+  } \
+};
+
+// HERK for complex data
+#define EIGEN_BLAS_RANKUPDATE_C(EIGTYPE, BLASTYPE, RTYPE, BLASFUNC) \
+template <typename Index, int AStorageOrder, bool ConjugateA, int  UpLo> \
+struct general_matrix_matrix_rankupdate<Index,EIGTYPE,AStorageOrder,ConjugateA,ColMajor,UpLo> { \
+  enum { \
+    IsLower = (UpLo&Lower) == Lower, \
+    LowUp = IsLower ? Lower : Upper, \
+    conjA = (((AStorageOrder==ColMajor) && ConjugateA) || ((AStorageOrder==RowMajor) && !ConjugateA)) ? 1 : 0 \
+  }; \
+  static EIGEN_STRONG_INLINE void run(Index size, Index depth,const EIGTYPE* lhs, Index lhsStride, \
+                          const EIGTYPE* /*rhs*/, Index /*rhsStride*/, EIGTYPE* res, Index resStride, EIGTYPE alpha, level3_blocking<EIGTYPE, EIGTYPE>& /*blocking*/) \
+  { \
+   typedef Matrix<EIGTYPE, Dynamic, Dynamic, AStorageOrder> MatrixType; \
+\
+   BlasIndex lda=convert_index<BlasIndex>(lhsStride), ldc=convert_index<BlasIndex>(resStride), n=convert_index<BlasIndex>(size), k=convert_index<BlasIndex>(depth); \
+   char uplo=((IsLower) ? 'L' : 'U'), trans=((AStorageOrder==RowMajor) ? 'C':'N'); \
+   RTYPE alpha_, beta_; \
+   const EIGTYPE* a_ptr; \
+\
+   alpha_ = alpha.real(); \
+   beta_ = 1.0; \
+/* Copy with conjugation in some cases*/ \
+   MatrixType a; \
+   if (conjA) { \
+     Map<const MatrixType, 0, OuterStride<> > mapA(lhs,n,k,OuterStride<>(lhsStride)); \
+     a = mapA.conjugate(); \
+     lda = a.outerStride(); \
+     a_ptr = a.data(); \
+   } else a_ptr=lhs; \
+   BLASFUNC(&uplo, &trans, &n, &k, &alpha_, (BLASTYPE*)a_ptr, &lda, &beta_, (BLASTYPE*)res, &ldc); \
+  } \
+};
+
+#ifdef EIGEN_USE_MKL
+EIGEN_BLAS_RANKUPDATE_R(double, double, dsyrk)
+EIGEN_BLAS_RANKUPDATE_R(float,  float,  ssyrk)
+#else
+EIGEN_BLAS_RANKUPDATE_R(double, double, dsyrk_)
+EIGEN_BLAS_RANKUPDATE_R(float,  float,  ssyrk_)
+#endif
+
+// TODO hanlde complex cases
+// EIGEN_BLAS_RANKUPDATE_C(dcomplex, double, double, zherk_)
+// EIGEN_BLAS_RANKUPDATE_C(scomplex, float,  float, cherk_)
+
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_GENERAL_MATRIX_MATRIX_TRIANGULAR_BLAS_H

include/eigen3/Eigen/src/Core/products/GeneralMatrixMatrix_BLAS.h

@@ -0,0 +1,124 @@
+/*
+ Copyright (c) 2011, Intel Corporation. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without modification,
+ are permitted provided that the following conditions are met:
+
+ * Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+ * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+ * Neither the name of Intel Corporation nor the names of its contributors may
+   be used to endorse or promote products derived from this software without
+   specific prior written permission.
+
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+ ********************************************************************************
+ *   Content : Eigen bindings to BLAS F77
+ *   General matrix-matrix product functionality based on ?GEMM.
+ ********************************************************************************
+*/
+
+#ifndef EIGEN_GENERAL_MATRIX_MATRIX_BLAS_H
+#define EIGEN_GENERAL_MATRIX_MATRIX_BLAS_H
+
+namespace Eigen { 
+
+namespace internal {
+
+/**********************************************************************
+* This file implements general matrix-matrix multiplication using BLAS
+* gemm function via partial specialization of
+* general_matrix_matrix_product::run(..) method for float, double,
+* std::complex<float> and std::complex<double> types
+**********************************************************************/
+
+// gemm specialization
+
+#define GEMM_SPECIALIZATION(EIGTYPE, EIGPREFIX, BLASTYPE, BLASFUNC) \
+template< \
+  typename Index, \
+  int LhsStorageOrder, bool ConjugateLhs, \
+  int RhsStorageOrder, bool ConjugateRhs> \
+struct general_matrix_matrix_product<Index,EIGTYPE,LhsStorageOrder,ConjugateLhs,EIGTYPE,RhsStorageOrder,ConjugateRhs,ColMajor,1> \
+{ \
+typedef gebp_traits<EIGTYPE,EIGTYPE> Traits; \
+\
+static void run(Index rows, Index cols, Index depth, \
+  const EIGTYPE* _lhs, Index lhsStride, \
+  const EIGTYPE* _rhs, Index rhsStride, \
+  EIGTYPE* res, Index resIncr, Index resStride, \
+  EIGTYPE alpha, \
+  level3_blocking<EIGTYPE, EIGTYPE>& /*blocking*/, \
+  GemmParallelInfo<Index>* /*info = 0*/) \
+{ \
+  using std::conj; \
+\
+  EIGEN_ONLY_USED_FOR_DEBUG(resIncr); \
+  eigen_assert(resIncr == 1); \
+  char transa, transb; \
+  BlasIndex m, n, k, lda, ldb, ldc; \
+  const EIGTYPE *a, *b; \
+  EIGTYPE beta(1); \
+  MatrixX##EIGPREFIX a_tmp, b_tmp; \
+\
+/* Set transpose options */ \
+  transa = (LhsStorageOrder==RowMajor) ? ((ConjugateLhs) ? 'C' : 'T') : 'N'; \
+  transb = (RhsStorageOrder==RowMajor) ? ((ConjugateRhs) ? 'C' : 'T') : 'N'; \
+\
+/* Set m, n, k */ \
+  m = convert_index<BlasIndex>(rows);  \
+  n = convert_index<BlasIndex>(cols);  \
+  k = convert_index<BlasIndex>(depth); \
+\
+/* Set lda, ldb, ldc */ \
+  lda = convert_index<BlasIndex>(lhsStride); \
+  ldb = convert_index<BlasIndex>(rhsStride); \
+  ldc = convert_index<BlasIndex>(resStride); \
+\
+/* Set a, b, c */ \
+  if ((LhsStorageOrder==ColMajor) && (ConjugateLhs)) { \
+    Map<const MatrixX##EIGPREFIX, 0, OuterStride<> > lhs(_lhs,m,k,OuterStride<>(lhsStride)); \
+    a_tmp = lhs.conjugate(); \
+    a = a_tmp.data(); \
+    lda = convert_index<BlasIndex>(a_tmp.outerStride()); \
+  } else a = _lhs; \
+\
+  if ((RhsStorageOrder==ColMajor) && (ConjugateRhs)) { \
+    Map<const MatrixX##EIGPREFIX, 0, OuterStride<> > rhs(_rhs,k,n,OuterStride<>(rhsStride)); \
+    b_tmp = rhs.conjugate(); \
+    b = b_tmp.data(); \
+    ldb = convert_index<BlasIndex>(b_tmp.outerStride()); \
+  } else b = _rhs; \
+\
+  BLASFUNC(&transa, &transb, &m, &n, &k, (const BLASTYPE*)&numext::real_ref(alpha), (const BLASTYPE*)a, &lda, (const BLASTYPE*)b, &ldb, (const BLASTYPE*)&numext::real_ref(beta), (BLASTYPE*)res, &ldc); \
+}};
+
+#ifdef EIGEN_USE_MKL
+GEMM_SPECIALIZATION(double,   d,  double, dgemm)
+GEMM_SPECIALIZATION(float,    f,  float,  sgemm)
+GEMM_SPECIALIZATION(dcomplex, cd, MKL_Complex16, zgemm)
+GEMM_SPECIALIZATION(scomplex, cf, MKL_Complex8,  cgemm)
+#else
+GEMM_SPECIALIZATION(double,   d,  double, dgemm_)
+GEMM_SPECIALIZATION(float,    f,  float,  sgemm_)
+GEMM_SPECIALIZATION(dcomplex, cd, double, zgemm_)
+GEMM_SPECIALIZATION(scomplex, cf, float,  cgemm_)
+#endif
+
+} // end namespase internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_GENERAL_MATRIX_MATRIX_BLAS_H