Merge pull request #526 from Cytnx-dev/remove-uni-omp-pr

Remove UNI_OMP guards surrounding `#pragma omp...`

CMakeLists.txt

@@ -19,7 +19,6 @@ set(CMAKE_CXX_STANDARD 17)
 set(CMAKE_CXX_STANDARD_REQUIRED ON)
 set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
 
-# set(CMAKE_CXX_EXTENSIONS OFF)
 message(STATUS " Generator: ${CMAKE_GENERATOR}")
 message(STATUS " Build Target: ${CMAKE_SYSTEM_NAME}-${CMAKE_SYSTEM_PROCESSOR}")
 message(STATUS " Installation Prefix: ${CMAKE_INSTALL_PREFIX}")
@@ -95,7 +94,8 @@ option(DEV_MODE "Build testing dev_test.cpp with cytnx" OFF)
 # #####################################################################
 project(CYTNX VERSION ${CYTNX_VERSION} LANGUAGES CXX C)
 
-# C++ uses link-time optimization anyway; this enables additionally -flto=auto, for parallel compilation
+# C++ uses link-time optimization anyway; this enables additionally -flto=auto,
+# for parallel compilation
 set(CMAKE_INTERPROCEDURAL_OPTIMIZATION TRUE)
 
 add_library(cytnx STATIC)
@@ -117,20 +117,9 @@ target_include_directories(cytnx SYSTEM
   ${Boost_INCLUDE_DIRS}
 )
 
-# set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wformat=0 -Wno-c++11-narrowing -w")
-# set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wformat=0 -w -fsized-deallocation -fsanitize=address")
-
 target_compile_options(cytnx PUBLIC -Wformat=0 -w -fsized-deallocation)
-
-#set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wformat=0 -w -fsized-deallocation -diag-disable=10441")
-
-# set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -D_LIBCPP_DISABLE_AVAILABILITY")
-# set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -D_LIBCPP_ENABLE_CXX17_REMOVED_UNARY_BINARY_FUNCTION")
 target_compile_features(cytnx PUBLIC cxx_std_17)
 
-# target_compile_definitions(cytnx PUBLIC _LIBCPP_DISABLE_AVAILABILITY)
-# target_compile_definitions(cytnx PUBLIC _LIBCPP_ENABLE_CXX17_REMOVED_UNARY_BINARY_FUNCTION)
-# set_property(TARGET cytnx PROPERTY CXX_STANDARD 11)
 add_subdirectory(src)
 
 if(USE_DEBUG)
@@ -140,23 +129,14 @@ if(USE_DEBUG)
   set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fsanitize=address")
 endif()
 
-if(RUN_TESTS)
-  # Include CTest explicitly to suppress the error that DartConfiguration.tcl
-  # is not found
-  include(CTest)
-  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -g -fprofile-arcs -ftest-coverage")
-  target_link_libraries(cytnx PUBLIC "-lgcov --coverage")
-  add_subdirectory(tests)
-  #add_subdirectory(bm_tests)
-endif()
-
 include(GNUInstallDirs)
 
 if(USE_ICPC)
   target_compile_definitions(cytnx PUBLIC UNI_ICPC)
   # This option is to disable the warning of icpc
   target_compile_options(cytnx PUBLIC -diag-disable=10441)
-  # This option is to maintain consistency of the floating point operation, also try to improve the performance
+  # This option is to maintain consistency of the floating point operation, also
+  # try to improve the performance
   target_compile_options(cytnx PUBLIC -fp-model consistent -fimf-use-svml)
 endif()
 
@@ -207,36 +187,24 @@ endif() # Backend torch
 # ## Get Gtest & benchmark
 # #####################################################################
 if(RUN_TESTS)
-  find_package(GTest REQUIRED)
-  #find_package(benchmark REQUIRED)
-
-
-  #include(FetchContent)
-  #FetchContent_Declare(
-  #  googletest
-  #  URL https://github.com/google/googletest/archive/609281088cfefc76f9d0ce82e1ff6c30cc3591e5.zip
-  #)
+  # Include CTest explicitly to suppress the error that DartConfiguration.tcl
+  # is not found
+  include(CTest)
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -g -fprofile-arcs -ftest-coverage")
+  target_link_libraries(cytnx PUBLIC "-lgcov --coverage")
+  # "-fopenmp" flag is appended to CMAKE_CXX_FLAGS in CytnxBKNDCMakeLists.cmake.
+  # To make the executalbe targets defined under tests/ link OpenMP correctly,
+  # tests/ must be added after including CytnxBKNDCMakeLists.cmake.
+  add_subdirectory(tests)
+  #add_subdirectory(bm_tests)
 
-  # For Windows: Prevent overriding the parent project's compiler/linker settings
   set(gtest_force_shared_crt ON CACHE BOOL "" FORCE)
-
-  #FetchContent_Declare(
-  #  googlebenchmark
-  #  GIT_REPOSITORY https://github.com/google/benchmark.git
-  #  GIT_TAG v1.7.1) # need master for benchmark::benchmark
-  #set(BENCHMARK_ENABLE_TESTING off) # to suppress benchmark internal tests
-
-  #FetchContent_MakeAvailable(
-  #  googletest
-  #  googlebenchmark)
 endif()
 
 # #######
 # ## Python wrapper
 # #######
 IF(BUILD_PYTHON)
-  # PythonInterp is removed in newer version of cmake
-  # find_package(PythonInterp REQUIRED)
   find_package (Python COMPONENTS Interpreter Development)
   find_package(pybind11 REQUIRED)
 
@@ -406,7 +374,6 @@ message(STATUS "    CXX Flags: ${CMAKE_CXX_FLAGS}")
 message(STATUS "    BLAS and LAPACK Libraries: ${LAPACK_LIBRARIES}")
 message(STATUS "    Link libraries: ${CYTNX_LINK_Libs}")
 
-# # this need to append into the linkflags.tmp
 FILE(APPEND "${CMAKE_BINARY_DIR}/linkflags.tmp" "${LAPACK_LIBRARIES}\n" "")
 FILE(APPEND "${CMAKE_BINARY_DIR}/cxxflags.tmp" "${CMAKE_CXX_FLAGS}\n" "")
 

CytnxBKNDCMakeLists.cmake

@@ -261,10 +261,6 @@ if(USE_OMP)
 else()
     message( STATUS " Build OMP Support: NO")
 
-    if(USE_MKL)
-        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fopenmp")
-    endif()
-
     if(USE_HPTT)
         find_package( OpenMP )
         if ( OPENMP_FOUND )

src/Accessor.cpp

@@ -222,10 +222,8 @@ namespace cytnx {
       pos.clear();
       pos.resize(this->idx_list.size());
       len = pos.size();
-// cout << "list in accessor len:" <<len << endl;
-#ifdef UNI_OMP
-  #pragma omp parallel for schedule(dynamic)
-#endif
+      // cout << "list in accessor len:" <<len << endl;
+#pragma omp parallel for schedule(dynamic)
       for (cytnx_uint64 i = 0; i < this->idx_list.size(); i++) {
         // checking:
         if (this->idx_list[i] < 0) {

src/Bond.cpp

@@ -139,9 +139,7 @@ namespace cytnx {
           new_qnums = std::vector<std::vector<cytnx_int64>>(Dnew_qnums,
                                                             std::vector<cytnx_int64>(this->Nsym()));
 
-#ifdef UNI_OMP
-  #pragma omp parallel for schedule(dynamic)
-#endif
+#pragma omp parallel for schedule(dynamic)
           for (cytnx_uint64 d = 0; d < new_qnums.size(); d++) {
             for (cytnx_uint32 i = 0; i < this->Nsym(); i++) {
               if (bd_in->type() != this->type()) {
@@ -169,9 +167,7 @@ namespace cytnx {
         new_qnums =
           std::vector<std::vector<cytnx_int64>>(this->_dim, std::vector<cytnx_int64>(this->Nsym()));
 
-#ifdef UNI_OMP
-  #pragma omp parallel for schedule(dynamic)
-#endif
+#pragma omp parallel for schedule(dynamic)
         for (cytnx_uint64 d = 0; d < this->_dim; d++) {
           for (cytnx_uint32 i = 0; i < this->Nsym(); i++) {
             if (bd_in->type() != this->type()) {
@@ -252,9 +248,7 @@ namespace cytnx {
           new_qnums = std::vector<std::vector<cytnx_int64>>(Dnew_qnums,
                                                             std::vector<cytnx_int64>(this->Nsym()));
 
-#ifdef UNI_OMP
-  #pragma omp parallel for schedule(dynamic)
-#endif
+#pragma omp parallel for schedule(dynamic)
           for (cytnx_uint64 d = 0; d < new_qnums.size(); d++) {
             for (cytnx_uint32 i = 0; i < this->Nsym(); i++) {
               this->_syms[i].combine_rule_(new_qnums[d][i],
@@ -276,9 +270,7 @@ namespace cytnx {
         new_qnums =
           std::vector<std::vector<cytnx_int64>>(this->_dim, std::vector<cytnx_int64>(this->Nsym()));
 
-#ifdef UNI_OMP
-  #pragma omp parallel for schedule(dynamic)
-#endif
+#pragma omp parallel for schedule(dynamic)
         for (cytnx_uint64 d = 0; d < this->_dim; d++) {
           for (cytnx_uint32 i = 0; i < this->Nsym(); i++) {
             this->_syms[i].combine_rule_(new_qnums[d][i],
@@ -470,9 +462,7 @@ namespace cytnx {
   std::vector<std::vector<cytnx_int64>> Bond_impl::calc_reverse_qnums() {
     std::vector<std::vector<cytnx_int64>> out(this->_qnums.size(),
                                               std::vector<cytnx_int64>(this->_syms.size()));
-#ifdef UNI_OMP
-  #pragma omp parallel for
-#endif
+#pragma omp parallel for
     for (cytnx_uint64 i = 0; i < out.size(); i++) {
       for (int j = 0; j < out[i].size(); j++) {
         this->_syms[j].reverse_rule_(out[i][j], this->_qnums[i][j]);

src/Device.cpp

@@ -1,4 +1,7 @@
 #include "Device.hpp"
+
+#include <thread>
+
 #include "cytnx_error.hpp"
 #ifdef UNI_OMP
   #include <omp.h>
@@ -7,7 +10,7 @@
 using namespace std;
 namespace cytnx {
 
-  Device_class::Device_class() : Ngpus(0), Ncpus(1) {
+  Device_class::Device_class() : Ngpus(0), Ncpus(std::thread::hardware_concurrency()) {
     // cout << "init_device class!" << endl;
 #ifdef UNI_GPU
 
@@ -38,17 +41,7 @@ namespace cytnx {
         }
       }
     }
-
-#endif
-
-#ifdef UNI_OMP
-  #pragma omp parallel
-    {
-      if (omp_get_thread_num() == 0) {
-        Ncpus = omp_get_num_threads();
-      }
-    }
-#endif
+#endif  // UNI_GPU
   };
 
   Device_class::~Device_class(){

src/SparseUniTensor.cpp

@@ -168,9 +168,7 @@ namespace cytnx {
       for (cytnx_uint64 i = 0; i < cb_inbonds.size(); i++) {
         N_sym = cb_inbonds[i].Nsym();
 
-#ifdef UNI_OMP
-  #pragma omp parallel for schedule(dynamic)
-#endif
+#pragma omp parallel for schedule(dynamic)
         for (cytnx_uint64 d = 0; d < N_sym * cb_inbonds[i].dim(); d++) {
           cb_inbonds[i].qnums()[cytnx_uint64(d / N_sym)][d % N_sym] *=
             cb_inbonds[i].type() * bondType::BD_KET;
@@ -187,9 +185,7 @@ namespace cytnx {
       for (cytnx_uint64 i = 0; i < cb_outbonds.size(); i++) {
         N_sym = cb_outbonds[i].Nsym();
 
-#ifdef UNI_OMP
-  #pragma omp parallel for schedule(dynamic)
-#endif
+#pragma omp parallel for schedule(dynamic)
         for (cytnx_uint64 d = 0; d < N_sym * cb_outbonds[i].dim(); d++) {
           cb_outbonds[i].qnums()[cytnx_uint64(d / N_sym)][d % N_sym] *=
             cb_outbonds[i].type() * bondType::BD_BRA;

src/Symmetry.cpp

@@ -82,9 +82,7 @@ namespace cytnx {
                                         const std::vector<cytnx_int64> &inL,
                                         const std::vector<cytnx_int64> &inR) {
     out.resize(inL.size() * inR.size());
-#ifdef UNI_OMP
-  #pragma omp parallel for schedule(dynamic)
-#endif
+#pragma omp parallel for schedule(dynamic)
     for (cytnx_uint64 i = 0; i < out.size(); i++) {
       out[i] = inL[cytnx_uint64(i / inR.size())] + inR[i % inR.size()];
     }
@@ -146,9 +144,7 @@ namespace cytnx {
                                         const std::vector<cytnx_int64> &inL,
                                         const std::vector<cytnx_int64> &inR) {
     out.resize(inL.size() * inR.size());
-#ifdef UNI_OMP
-  #pragma omp parallel for schedule(dynamic)
-#endif
+#pragma omp parallel for schedule(dynamic)
     for (cytnx_uint64 i = 0; i < out.size(); i++) {
       out[i] = (inL[cytnx_uint64(i / inR.size())] + inR[i % inR.size()]) % (this->n);
     }

src/backend/Storage_base.cpp

@@ -989,9 +989,7 @@ namespace cytnx {
   void Storage_base::_cpy_bool(void *ptr, const std::vector<cytnx_bool> &vin) {
     bool *tmp = static_cast<bool *>(ptr);
 
-#ifdef UNI_OMP
-  #pragma omp parallel for schedule(dynamic)
-#endif
+#pragma omp parallel for schedule(dynamic)
     for (cytnx_uint64 i = 0; i < vin.size(); i++) {
       tmp[i] = vin[i];
     }

src/backend/linalg_internal_cpu/Abs_internal.cpp

@@ -14,9 +14,7 @@ namespace cytnx {
       cytnx_double *_out = (cytnx_double *)out->Mem;
       cytnx_complex128 *_ten = (cytnx_complex128 *)ten->Mem;
 
-#ifdef UNI_OMP
-  #pragma omp parallel for schedule(dynamic)
-#endif
+#pragma omp parallel for schedule(dynamic)
       for (cytnx_uint64 n = 0; n < Nelem; n++) {
         _out[n] = std::abs(_ten[n]);
       }
@@ -27,9 +25,7 @@ namespace cytnx {
       cytnx_float *_out = (cytnx_float *)out->Mem;
       cytnx_complex64 *_ten = (cytnx_complex64 *)ten->Mem;
 
-#ifdef UNI_OMP
-  #pragma omp parallel for schedule(dynamic)
-#endif
+#pragma omp parallel for schedule(dynamic)
       for (cytnx_uint64 n = 0; n < Nelem; n++) {
         _out[n] = std::abs(_ten[n]);
       }
@@ -40,9 +36,7 @@ namespace cytnx {
       cytnx_double *_out = (cytnx_double *)out->Mem;
       cytnx_double *_ten = (cytnx_double *)ten->Mem;
 
-#ifdef UNI_OMP
-  #pragma omp parallel for schedule(dynamic)
-#endif
+#pragma omp parallel for schedule(dynamic)
       for (cytnx_uint64 n = 0; n < Nelem; n++) {
         _out[n] = std::abs(_ten[n]);
       }
@@ -53,9 +47,7 @@ namespace cytnx {
       cytnx_float *_out = (cytnx_float *)out->Mem;
       cytnx_float *_ten = (cytnx_float *)ten->Mem;
 
-#ifdef UNI_OMP
-  #pragma omp parallel for schedule(dynamic)
-#endif
+#pragma omp parallel for schedule(dynamic)
       for (cytnx_uint64 n = 0; n < Nelem; n++) {
         _out[n] = std::abs(_ten[n]);
       }
@@ -67,9 +59,7 @@ namespace cytnx {
       cytnx_int64 *_out = (cytnx_int64 *)out->Mem;
       cytnx_int64 *_ten = (cytnx_int64 *)ten->Mem;
 
-#ifdef UNI_OMP
-  #pragma omp parallel for schedule(dynamic)
-#endif
+#pragma omp parallel for schedule(dynamic)
       for (cytnx_uint64 n = 0; n < Nelem; n++) {
         _out[n] = std::abs(_ten[n]);
       }
@@ -81,9 +71,7 @@ namespace cytnx {
       cytnx_int32 *_out = (cytnx_int32 *)out->Mem;
       cytnx_int32 *_ten = (cytnx_int32 *)ten->Mem;
 
-#ifdef UNI_OMP
-  #pragma omp parallel for schedule(dynamic)
-#endif
+#pragma omp parallel for schedule(dynamic)
       for (cytnx_uint64 n = 0; n < Nelem; n++) {
         _out[n] = std::abs(_ten[n]);
       }
@@ -95,9 +83,7 @@ namespace cytnx {
       cytnx_int16 *_out = (cytnx_int16 *)out->Mem;
       cytnx_int16 *_ten = (cytnx_int16 *)ten->Mem;
 
-#ifdef UNI_OMP
-  #pragma omp parallel for schedule(dynamic)
-#endif
+#pragma omp parallel for schedule(dynamic)
       for (cytnx_uint64 n = 0; n < Nelem; n++) {
         _out[n] = std::abs(cytnx_double(_ten[n]));
       }