Merge branch 'master' into kmp5/feature/cp-bcd

.github/workflows/cmake.yml

@@ -74,14 +74,12 @@ jobs:
         fi
         # install MKL if want vendor linalg
         if [ "${{matrix.linalg}}" = "vendor" ]; then
-          wget https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS-2019.PUB 
-          sudo apt-key add GPG-PUB-KEY-INTEL-SW-PRODUCTS-2019.PUB
-          rm GPG-PUB-KEY-INTEL-SW-PRODUCTS-2019.PUB
-          sudo sh -c 'echo deb https://apt.repos.intel.com/mkl all main > /etc/apt/sources.list.d/intel-mkl.list'
-          sudo apt-get update
-          sudo apt-get install intel-mkl-64bit-2019.4-070
+          sudo sh -c 'wget -O - https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB | gpg --dearmor > /usr/share/keyrings/oneapi-archive-keyring.gpg'
+          sudo sh -c 'echo "deb [signed-by=/usr/share/keyrings/oneapi-archive-keyring.gpg] https://apt.repos.intel.com/oneapi all main" > /etc/apt/sources.list.d/oneAPI.list'
+          sudo apt-get -yq update
+          sudo apt-get install intel-oneapi-mkl-devel
           echo "BLAS_PREFERENCE_LIST=IntelMKL" >> $GITHUB_ENV
-          echo "MKLROOT=/opt/intel/mkl" >> $GITHUB_ENV
+          echo "MKLROOT=/opt/intel/oneapi/mkl/latest" >> $GITHUB_ENV
         else
           echo "BLAS_PREFERENCE_LIST=ReferenceBLAS" >> $GITHUB_ENV
         fi

CMakeLists.txt

@@ -2,6 +2,9 @@
 
 cmake_minimum_required (VERSION 3.14.0)  # FetchContent_MakeAvailable
 
+# Preload versions/tags of all dependencies ====================================
+include(external/versions.cmake)
+
 ###############################################################################
 # CMake defaults to address key pain points
 ###############################################################################
@@ -24,6 +27,7 @@ FetchContent_Declare(
         vg_cmake_kit
         QUIET
         GIT_REPOSITORY      https://github.com/ValeevGroup/kit-cmake.git
+        GIT_TAG             ${BTAS_TRACKED_VGCMAKEKIT_TAG}
         SOURCE_DIR ${${VG_CMAKE_KIT_PREFIX_DIR}}/cmake/vg
         BINARY_DIR ${${VG_CMAKE_KIT_PREFIX_DIR}}/cmake/vg-build
         SUBBUILD_DIR ${${VG_CMAKE_KIT_PREFIX_DIR}}/cmake/vg-subbuild
@@ -60,6 +64,7 @@ include(FeatureSummary)
 include(RedefaultableOption)
 include(CMakePackageConfigHelpers)
 include(AddCustomTargetSubproject)
+include(CMakePushCheckState)
 include(CTest) # defines BUILD_TESTING option
 
 # Configure options
@@ -77,28 +82,27 @@ add_feature_info("TARGET_MAX_INDEX_RANK=${TARGET_MAX_INDEX_RANK}" TRUE "default
 
 set(TARGET_ARCH "${CMAKE_SYSTEM_NAME}-${CMAKE_SYSTEM_PROCESSOR}")
 
-include(CMakePushCheckState)
-include(GNUInstallDirs)
-include(AppendFlags)
-
 ##########################
-# Standard build variables
+# INSTALL variables
 ##########################
-set(BTAS_INSTALL_BINDIR "bin"
+include(GNUInstallDirs)
+set(BTAS_INSTALL_BINDIR "${CMAKE_INSTALL_BINDIR}"
     CACHE PATH "BTAS BIN install directory")
-set(BTAS_INSTALL_INCLUDEDIR "include"
+set(BTAS_INSTALL_INCLUDEDIR "${CMAKE_INSTALL_INCLUDEDIR}"
     CACHE PATH "BTAS INCLUDE install directory")
-set(BTAS_INSTALL_LIBDIR "lib"
+set(BTAS_INSTALL_LIBDIR "${CMAKE_INSTALL_LIBDIR}"
     CACHE PATH "BTAS LIB install directory")
-set(BTAS_INSTALL_SHAREDIR "share/BTAS/${BTAS_MAJOR_VERSION}.${BTAS_MINOR_VERSION}.${BTAS_MICRO_VERSION}"
-    CACHE PATH "BTAS SHARE install directory")
-set(BTAS_INSTALL_DATADIR "${BTAS_INSTALL_SHAREDIR}/data"
+set(BTAS_INSTALL_DATADIR "${CMAKE_INSTALL_DATAROOTDIR}/BTAS/${BTAS_EXT_VERSION}"
     CACHE PATH "BTAS DATA install directory")
-set(BTAS_INSTALL_DOCDIR "${BTAS_INSTALL_SHAREDIR}/doc"
+set(BTAS_INSTALL_DOCDIR "${BTAS_INSTALL_DATADIR}/doc"
     CACHE PATH "BTAS DOC install directory")
-set(BTAS_INSTALL_CMAKEDIR "lib/cmake/BTAS"
+set(BTAS_INSTALL_CMAKEDIR "${CMAKE_INSTALL_LIBDIR}/cmake/BTAS"
     CACHE PATH "BTAS CMAKE install directory")
 
+##########################
+# Standard build variables
+##########################
+include(AppendFlags)
 # Get standard build variables from the environment if they have not already been set
 if(NOT CMAKE_C_FLAGS OR NOT DEFINED CMAKE_C_FLAGS)
   set(CMAKE_C_FLAGS "$ENV{CPPFLAGS}")

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;
 

external/boost.cmake

@@ -21,9 +21,9 @@ if (NOT TARGET Boost::boost OR NOT TARGET Boost::serialization)
   # try config first
   # OPTIONAL_COMPONENTS in FindBoost available since 3.11
   cmake_minimum_required(VERSION 3.11.0)
-  find_package(Boost CONFIG OPTIONAL_COMPONENTS ${Boost_BTAS_DEPS_LIBRARIES})
+  find_package(Boost ${BTAS_TRACKED_BOOST_VERSION} CONFIG OPTIONAL_COMPONENTS ${Boost_BTAS_DEPS_LIBRARIES})
   if (NOT TARGET Boost::boost)
-    find_package(Boost OPTIONAL_COMPONENTS ${Boost_BTAS_DEPS_LIBRARIES})
+    find_package(Boost ${BTAS_TRACKED_BOOST_VERSION} OPTIONAL_COMPONENTS ${Boost_BTAS_DEPS_LIBRARIES})
     if (TARGET Boost::boost)
       set(Boost_USE_CONFIG FALSE)
     endif(TARGET Boost::boost)

external/versions.cmake

@@ -0,0 +1,7 @@
+set(BTAS_TRACKED_VGCMAKEKIT_TAG d6746098e63deab4032309c4455bb084a17ff51a)
+
+# likely can use earlier, but
+# - as of oct 2023 tested with 1.71 and up only
+# - avoids the need to avoid 1.70 in which Boost.Container is broken
+# - matches the version provided by https://github.com/Orphis/boost-cmake as of oct 2023
+set(BTAS_TRACKED_BOOST_VERSION 1.71)

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++)