Merge pull request #164 from ValeevGroup/163-convert-scalar-arguments…

…-to-btascontract-and-similar-functions-when-needed-to-the-type-of-the-result

support for mixed typing of scalars/data in blas/lapack-like functions

btas/generic/axpy_impl.h

@@ -52,8 +52,12 @@ template<> struct axpy_impl<true>
             _IteratorY itrY, const iterator_difference_t<_IteratorY>& incY,
       blas_lapack_impl_tag)
    { 
-     blas::axpy( Nsize, alpha, static_cast<const _T*>(&(*itrX)), incX, 
-                               static_cast<_T*>(&(*itrY)),       incY );
+      static_assert(std::is_same_v<iterator_value_t<_IteratorX>,iterator_value_t<_IteratorY>>,
+                    "mismatching iterator value types");
+      using T = iterator_value_t<_IteratorX>;
+
+     blas::axpy( Nsize, static_cast<T>(alpha), static_cast<const T*>(&(*itrX)), incX,
+                               static_cast<T*>(&(*itrY)),       incY );
    }
 #endif
 

btas/generic/dot_impl.h

@@ -110,19 +110,13 @@ struct dotc_impl<true> {
             _IteratorY itrY, const iterator_difference_t<_IteratorY>& incY,
       blas_lapack_impl_tag)
   {
-
-      using x_traits = std::iterator_traits<_IteratorX>;
-      using y_traits = std::iterator_traits<_IteratorY>;
-
-      using x_value_type = typename x_traits::value_type;
-      using y_value_type = typename y_traits::value_type;
-
-      using x_ptr_type = const x_value_type*;
-      using y_ptr_type = const y_value_type*;
+      static_assert(std::is_same_v<iterator_value_t<_IteratorX>,iterator_value_t<_IteratorY>>,
+                    "mismatching iterator value types");
+      using T = iterator_value_t<_IteratorX>;
 
       // XXX: DOTC == DOT in BLASPP
-      return blas::dot( Nsize, static_cast<x_ptr_type>(&(*itrX)), incX,
-                               static_cast<y_ptr_type>(&(*itrY)), incY );
+      return blas::dot( Nsize, static_cast<const T*>(&(*itrX)), incX,
+                               static_cast<const T*>(&(*itrY)), incY );
 
   }
 #endif
@@ -172,17 +166,12 @@ struct dotu_impl<true> {
       blas_lapack_impl_tag)
   {
 
-      using x_traits = std::iterator_traits<_IteratorX>;
-      using y_traits = std::iterator_traits<_IteratorY>;
-
-      using x_value_type = typename x_traits::value_type;
-      using y_value_type = typename y_traits::value_type;
-
-      using x_ptr_type = const x_value_type*;
-      using y_ptr_type = const y_value_type*;
+      static_assert(std::is_same_v<iterator_value_t<_IteratorX>,iterator_value_t<_IteratorY>>,
+                    "mismatching iterator value types");
+      using T = iterator_value_t<_IteratorX>;
 
-      return blas::dotu( Nsize, static_cast<x_ptr_type>(&(*itrX)), incX,
-                                static_cast<y_ptr_type>(&(*itrY)), incY );
+      return blas::dotu( Nsize, static_cast<const T*>(&(*itrX)), incX,
+                                static_cast<const T*>(&(*itrY)), incY );
 
   }
 #endif

btas/generic/gemm_impl.h

@@ -252,24 +252,16 @@ template<> struct gemm_impl<true>
       const unsigned long& LDC,
       blas_lapack_impl_tag)
    {
-
-      using a_traits = std::iterator_traits<_IteratorA>;
-      using b_traits = std::iterator_traits<_IteratorB>;
-      using c_traits = std::iterator_traits<_IteratorC>;
-
-      using a_value_type = typename a_traits::value_type;
-      using b_value_type = typename b_traits::value_type;
-      using c_value_type = typename c_traits::value_type;
-
-      using a_ptr_type = const a_value_type*;
-      using b_ptr_type = const b_value_type*;
-      using c_ptr_type =       c_value_type*;
-
-      blas::gemm( order, transA, transB, Msize, Nsize, Ksize, alpha,
-                  static_cast<a_ptr_type>(&(*itrA)), LDA,
-                  static_cast<b_ptr_type>(&(*itrB)), LDB,
-                  beta,
-                  static_cast<c_ptr_type>(&(*itrC)), LDC );
+      static_assert(std::is_same_v<iterator_value_t<_IteratorA>,iterator_value_t<_IteratorB>> &&
+                    std::is_same_v<iterator_value_t<_IteratorA>,iterator_value_t<_IteratorC>>,
+                    "mismatching iterator value types");
+      using T = iterator_value_t<_IteratorA>;
+
+      blas::gemm( order, transA, transB, Msize, Nsize, Ksize, static_cast<T>(alpha),
+                  static_cast<const T*>(&(*itrA)), LDA,
+                  static_cast<const T*>(&(*itrB)), LDB,
+                  static_cast<T>(beta),
+                  static_cast<T*>(&(*itrC)), LDC );
     }
 #endif
 

btas/generic/gemv_impl.h

@@ -151,11 +151,16 @@ template<> struct gemv_impl<true>
       blas_lapack_impl_tag)
    {
 
-     blas::gemv( order, transA, Msize, Nsize, alpha,
-                 static_cast<const _T*>(&(*itrA)), LDA,
-                 static_cast<const _T*>(&(*itrX)), incX,
-                 beta,
-                 static_cast<      _T*>(&(*itrY)), incY );
+     static_assert(std::is_same_v<iterator_value_t<_IteratorX>,iterator_value_t<_IteratorY>> &&
+                   std::is_same_v<iterator_value_t<_IteratorX>,iterator_value_t<_IteratorA>>,
+                   "mismatching iterator value types");
+     using T = iterator_value_t<_IteratorX>;
+
+     blas::gemv( order, transA, Msize, Nsize, static_cast<T>(alpha),
+                 static_cast<const T*>(&(*itrA)), LDA,
+                 static_cast<const T*>(&(*itrX)), incX,
+                 static_cast<T>(beta),
+                 static_cast<      T*>(&(*itrY)), incY );
 
    }
 #endif

btas/generic/ger_impl.h

@@ -79,11 +79,15 @@ template<> struct ger_impl<true>
       const unsigned long& LDA,
       blas_lapack_impl_tag)
    {
-
-     blas::geru( order, Msize, Nsize, alpha,
-                 static_cast<const _T*>(&(*itrX)), incX,
-                 static_cast<const _T*>(&(*itrY)), incY,
-                 static_cast<      _T*>(&*(itrA)), LDA );
+     static_assert(std::is_same_v<iterator_value_t<_IteratorX>,iterator_value_t<_IteratorY>> &&
+                   std::is_same_v<iterator_value_t<_IteratorX>,iterator_value_t<_IteratorA>>,
+                   "mismatching iterator value types");
+     using T = iterator_value_t<_IteratorX>;
+
+     blas::geru( order, Msize, Nsize, static_cast<T>(alpha),
+                 static_cast<const T*>(&(*itrX)), incX,
+                 static_cast<const T*>(&(*itrY)), incY,
+                 static_cast<      T*>(&*(itrA)), LDA );
    }
 #endif
 

btas/generic/gesvd_impl.h

@@ -73,8 +73,6 @@ template<> struct gesvd_impl<true>
      if( inplaceU and inplaceVt ) 
        BTAS_EXCEPTION("SVD cannot return both vectors inplace");
 
-
-
      value_type dummy;
      value_type* A = static_cast<value_type*>(&(*itrA));
      value_type* U = (needU and not inplaceU) ?

btas/generic/scal_impl.h

@@ -44,7 +44,8 @@ template<> struct scal_impl<true>
             _IteratorX itrX, const iterator_difference_t<_IteratorX>& incX,
             blas_lapack_impl_tag)
    {
-     blas::scal( Nsize, alpha, static_cast<_T*>(&(*itrX)), incX );
+     using T = iterator_value_t<_IteratorX>;
+     blas::scal( Nsize, static_cast<T>(alpha), static_cast<T*>(&(*itrX)), incX );
    }
 #endif
 

btas/type_traits.h

@@ -275,6 +275,9 @@ namespace btas {
 
   // Convienience traits
   template <typename _Iterator>
+  using iterator_value_t =
+      typename std::iterator_traits<_Iterator>::value_type;
+  template <typename _Iterator>
   using iterator_difference_t = 
     typename std::iterator_traits<_Iterator>::difference_type;
 

unittest/tensor_blas_test.cc

@@ -112,7 +112,7 @@ TEST_CASE("Tensor Scal")
         Tensor<double> T(4,2,6,5);
         T.generate([](){ return randomReal<double>(); });
         Tensor<double> Tbak=T;
-        double d = randomReal<double>();
+        const auto d = randomReal<float>();  // N.B. use different types for scalar and tensor
         scal(d,T);
         double res=0;
         for(auto i : T.range()) res+=std::abs(T(i)-Tbak(i)*d);
@@ -136,10 +136,11 @@ TEST_CASE("Tensor Scal")
         Tensor<std::complex<double>> T(4,2,6,5);
         T.generate([](){ return randomCplx<double>(); });
         Tensor<std::complex<double>> Tbak=T;
-        std::complex<double> d = randomCplx<double>();
+        const auto d = randomCplx<float>();  // N.B. use different types for scalar and tensor
         scal(d,T);
         double res=0;
-        for(auto i : T.range()) res+=std::abs(T(i)-Tbak(i)*d);
+        std::complex<double> d_double(d);
+        for(auto i : T.range()) res+=std::abs(T(i)-Tbak(i)*d_double);
         CHECK(res < eps_double);
         }
 
@@ -166,7 +167,7 @@ TEST_CASE("Tensor Axpy")
         X.generate([](){ return randomReal<double>(); });
         Y.generate([](){ return randomReal<double>(); });
         Tensor<double> Ybak=Y;
-        double alpha = randomReal<double>();
+        const auto alpha = randomReal<float>();  // N.B. use different types for scalar and tensor
         axpy(alpha,X,Y);
         double res=0;
         for(auto i : Y.range()) res+=std::abs(Ybak(i)+X(i)*alpha-Y(i));
@@ -228,7 +229,7 @@ TEST_CASE("Tensor Ger")
         X.generate([](){ return randomReal<double>(); });
         Y.generate([](){ return randomReal<double>(); });
         Tensor<double> Abak=A;
-        double a = randomReal<double>();
+        const auto a = randomReal<float>();  // N.B. use different types for scalar and tensor
         ger(a,X,Y,A);
         double res=0;
         for(auto i : A.range()) res+=std::abs(a*X(i[0],i[1])*Y(i[2],i[3])+Abak(i)-A(i));
@@ -296,8 +297,8 @@ TEST_CASE("Tensor Gemv")
         A.generate([](){ return randomReal<double>(); });
         X.generate([](){ return randomReal<double>(); });
         Y.generate([](){ return randomReal<double>(); });
-        double alpha = randomReal<double>();
-        double beta = randomReal<double>();
+        const auto alpha = randomReal<float>();  // N.B. use different types for scalar and tensor
+        const auto beta = randomReal<float>();  // N.B. use different types for scalar and tensor
         Tensor<double> Ytest=Y;
         scal(beta,Ytest);
         for(long i=0;i<A.extent(0);i++)
@@ -681,8 +682,8 @@ TEST_CASE("Contraction")
         A.generate([](){ return randomReal<double>(); });
         B.generate([](){ return randomReal<double>(); });
         C.generate([](){ return randomReal<double>(); });
-        double alpha = randomReal<double>();
-        double beta = randomReal<double>();
+        const auto alpha = randomReal<float>();  // N.B. use different types for scalar and tensor
+        const auto beta = randomReal<float>();  // N.B. use different types for scalar and tensor
         Ctest=C;
         scal(beta,Ctest);
         contract(alpha,A,{'i','j','k'},B,{'k','j','l','m'},beta,C,{'i','m','l'});
@@ -700,7 +701,7 @@ TEST_CASE("Contraction")
         Tensor<double> D;
         Tensor<double> Dtest(2,4,6);
         Dtest.fill(0.0);
-        contract(alpha,A,{'i','j','k'},B,{'k','j','l','m'},0.0,D,{'i','m','l'});
+        contract(alpha,A,{'i','j','k'},B,{'k','j','l','m'},static_cast<decltype(alpha)>(0.0),D,{'i','m','l'});
         for(long i=0;i<A.extent(0);i++)
         for(long j=0;j<A.extent(1);j++)
         for(long k=0;k<A.extent(2);k++)