Support compiling with -std=c++* (and -std=gnu++*)

.github/workflows/ci.yml

@@ -31,7 +31,7 @@ jobs:
         run: |
             cmake . -G "Unix Makefiles" -DBUILD_TESTS=ON -DCMAKE_BUILD_TYPE=${{matrix.config}} -DCMAKE_CXX_FLAGS="-Wall -Wextra -pedantic -Werror"
             grep "^OpenMP:INTERNAL=1$" CMakeCache.txt
-            grep "^int128.*:INTERNAL=1$" CMakeCache.txt
+            grep "^int128_STL.*:INTERNAL=1$" CMakeCache.txt
             cmake --build . --parallel --verbose
       - name: CTest (unit tests)
         run: ctest -j2
@@ -118,7 +118,7 @@ jobs:
         run: |
             cmake . -DBUILD_TESTS=ON -DCMAKE_BUILD_TYPE=${{matrix.config}} -DCMAKE_CXX_FLAGS="-Wall -Wextra -pedantic -Werror"
             grep "^OpenMP:INTERNAL=1$" CMakeCache.txt
-            grep "^int128.*:INTERNAL=1$" CMakeCache.txt
+            grep "^int128_STL.*:INTERNAL=1$" CMakeCache.txt
             cmake --build . --parallel --verbose
       - name: CTest (unit tests)
         run: ctest -j2
@@ -152,7 +152,7 @@ jobs:
         run: |
             cmake . -DBUILD_TESTS=ON -DCMAKE_BUILD_TYPE=${{matrix.config}} -DCMAKE_CXX_FLAGS="-mpopcnt -Wall -Wextra -pedantic -Werror"
             grep "^OpenMP:INTERNAL=1$" CMakeCache.txt
-            grep "^int128.*:INTERNAL=1$" CMakeCache.txt
+            grep "^int128_STL.*:INTERNAL=1$" CMakeCache.txt
             cmake --build . --parallel --verbose
       - name: CTest (unit tests)
         run: ctest -j2
@@ -252,7 +252,7 @@ jobs:
         run: |
             cmake . -DBUILD_TESTS=ON -DBUILD_MANPAGE=ON -DCMAKE_BUILD_TYPE=${{matrix.config}} -DCMAKE_CXX_FLAGS="-Wall -Wextra -pedantic -Werror"
             grep "^OpenMP.*:INTERNAL=1$" CMakeCache.txt
-            grep "^int128.*:INTERNAL=1$" CMakeCache.txt
+            grep "^int128_STL.*:INTERNAL=1$" CMakeCache.txt
             cmake --build . --parallel --verbose
       - name: CTest (unit tests)
         run: ctest -j2
@@ -290,7 +290,7 @@ jobs:
         run: |
             cmake . -DBUILD_TESTS=ON -DBUILD_MANPAGE=ON -DCMAKE_BUILD_TYPE=${{matrix.config}} -DCMAKE_CXX_FLAGS="-mpopcnt -Wall -Wextra -pedantic -Werror"
             grep "^OpenMP.*:INTERNAL=1$" CMakeCache.txt
-            grep "^int128.*:INTERNAL=1$" CMakeCache.txt
+            grep "^int128_STL.*:INTERNAL=1$" CMakeCache.txt
             cmake --build . --parallel --verbose
       - name: CTest (unit tests)
         run: ctest -j2
@@ -326,7 +326,7 @@ jobs:
         run: |
             LDFLAGS="-L$(brew --prefix libomp)/lib" cmake . -DBUILD_TESTS=ON -DCMAKE_BUILD_TYPE=${{matrix.config}} -DCMAKE_CXX_FLAGS="-I$(brew --prefix libomp)/include -Wall -Wextra -pedantic -Werror"
             grep "^OpenMP.*:INTERNAL=1$" CMakeCache.txt
-            grep "^int128.*:INTERNAL=1$" CMakeCache.txt
+            grep "^int128_STL.*:INTERNAL=1$" CMakeCache.txt
             cmake --build . --parallel --verbose
       - name: CTest (unit tests)
         run: ctest -j2
@@ -353,7 +353,7 @@ jobs:
         run: |
             cmake . -DBUILD_TESTS=ON
             grep "^OpenMP:INTERNAL=1$" CMakeCache.txt
-            grep "^int128.*:INTERNAL=1$" CMakeCache.txt
+            grep "^int128_STL.*:INTERNAL=1$" CMakeCache.txt
             cmake --build . --parallel --verbose
       - name: CTest (unit tests)
         run: ctest -j2

CMakeLists.txt

@@ -90,19 +90,6 @@ if(NOT STATICALLY_LINK_LIBPRIMECOUNT)
     endif()
 endif()
 
-# Set default build type to Release ##################################
-
-if(NOT CMAKE_VERSION VERSION_LESS 3.9)
-    get_property(isMultiConfig GLOBAL PROPERTY GENERATOR_IS_MULTI_CONFIG)
-elseif(CMAKE_CONFIGURATION_TYPES)
-    set(isMultiConfig TRUE)
-endif()
-
-if(NOT isMultiConfig AND NOT CMAKE_BUILD_TYPE)
-    set(CMAKE_BUILD_TYPE Release CACHE STRING
-    "Choose the type of build, options are: None Debug Release RelWithDebInfo MinSizeRel." FORCE)
-endif()
-
 # primecount binary source files #####################################
 
 set(BIN_SRC src/app/CmdOptions.cpp
@@ -154,20 +141,24 @@ set(LIB_SRC src/api.cpp
             src/gourdon/SegmentedPiTable.cpp
             src/gourdon/Sigma.cpp)
 
-# Required includes ##################################################
+# Use libdivide.h (fast integer divison) #############################
 
-include(GNUInstallDirs)
+if(WITH_LIBDIVIDE)
+    set(LIB_SRC ${LIB_SRC} src/deleglise-rivat/S2_easy_libdivide.cpp)
+    set(LIB_SRC ${LIB_SRC} src/gourdon/AC_libdivide.cpp)
+else()
+    set(LIB_SRC ${LIB_SRC} src/deleglise-rivat/S2_easy.cpp)
+    set(LIB_SRC ${LIB_SRC} src/gourdon/AC.cpp)
+endif()
 
-# Use -Wno-uninitialized with GCC compiler ###########################
+# Enable __float128 support (requires libquadmath) ###################
 
-# GCC's -Wuninitialized enabled with -Wall -pedantic causes
-# false postive warnings in libprimecount:
-# https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107287
-if(CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
-    set(WNO_UNINITIALIZED "-Wno-uninitialized")
+if(WITH_FLOAT128)
+    set(LIB_QUADMATH "quadmath")
+    set(HAVE_FLOAT128 "HAVE_FLOAT128")
 endif()
 
-# libprimecount compile definitions ##################################
+# Use 32-bit integer division ########################################
 
 # Check at runtime if the dividend and divisor are < 2^32 and
 # if so use 32-bit integer division instead of 64-bit integer
@@ -177,11 +168,13 @@ if(WITH_DIV32)
     set(ENABLE_DIV32 "ENABLE_DIV32")
 endif()
 
-# Enable expensive debugging assertions.
-# These assertions enable e.g. bounds checks
-# for the Vector and Array types.
-if(CMAKE_BUILD_TYPE STREQUAL "Debug")
-    set(ENABLE_ASSERT "ENABLE_ASSERT")
+# Use -Wno-uninitialized with GCC compiler ###########################
+
+# GCC's -Wuninitialized enabled with -Wall -pedantic causes
+# false postive warnings in libprimecount:
+# https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107287
+if(CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
+    set(WNO_UNINITIALIZED "-Wno-uninitialized")
 endif()
 
 # Check if int128_t is supported #####################################
@@ -192,34 +185,39 @@ include("${PROJECT_SOURCE_DIR}/cmake/int128_t.cmake")
 
 if(WITH_MULTIARCH)
     include("${PROJECT_SOURCE_DIR}/cmake/multiarch_avx512_vpopcnt.cmake")
-    if(multiarch_avx512_vpopcnt)
-        set(ENABLE_MULTIARCH_AVX512_BMI2 "ENABLE_MULTIARCH_AVX512_BMI2")
-    endif()
 endif()
 
-# Use libdivide.h (fast integer divison) #############################
+# Check for OpenMP ###################################################
 
-if(WITH_LIBDIVIDE)
-    set(LIB_SRC ${LIB_SRC} src/deleglise-rivat/S2_easy_libdivide.cpp)
-    set(LIB_SRC ${LIB_SRC} src/gourdon/AC_libdivide.cpp)
-else()
-    set(LIB_SRC ${LIB_SRC} src/deleglise-rivat/S2_easy.cpp)
-    set(LIB_SRC ${LIB_SRC} src/gourdon/AC.cpp)
+if(WITH_OPENMP)
+    include("${PROJECT_SOURCE_DIR}/cmake/OpenMP.cmake")
 endif()
 
-# Enable __float128 support (requires libquadmath) ###################
+# Set default build type to Release ##################################
 
-if(WITH_FLOAT128)
-    set(LIB_QUADMATH "quadmath")
-    set(HAVE_FLOAT128 "HAVE_FLOAT128")
+# We do this after the C++ compilation tests above (include *.cmake)
+# so that those tests are compiled without optimization flags
+# which should speed up the cmake configure step.
+
+if(NOT CMAKE_VERSION VERSION_LESS 3.9)
+    get_property(isMultiConfig GLOBAL PROPERTY GENERATOR_IS_MULTI_CONFIG)
+elseif(CMAKE_CONFIGURATION_TYPES)
+    set(isMultiConfig TRUE)
 endif()
 
-# Check for OpenMP ###################################################
+if(NOT isMultiConfig AND NOT CMAKE_BUILD_TYPE)
+    set(CMAKE_BUILD_TYPE Release CACHE STRING
+    "Choose the type of build, options are: None Debug Release RelWithDebInfo MinSizeRel." FORCE)
+endif()
 
-if(WITH_OPENMP)
-    include("${PROJECT_SOURCE_DIR}/cmake/OpenMP.cmake")
+if(CMAKE_BUILD_TYPE STREQUAL "Debug")
+    set(ENABLE_ASSERT "ENABLE_ASSERT")
 endif()
 
+# Required includes ##################################################
+
+include(GNUInstallDirs)
+
 # libprimesieve ######################################################
 
 # By default the libprimesieve dependency is built from source
@@ -259,7 +257,7 @@ if(BUILD_SHARED_LIBS)
     set_target_properties(libprimecount PROPERTIES SOVERSION ${PRIMECOUNT_VERSION_MAJOR})
     set_target_properties(libprimecount PROPERTIES VERSION ${PRIMECOUNT_VERSION})
     target_compile_options(libprimecount PRIVATE "${WNO_UNINITIALIZED}")
-    target_compile_definitions(libprimecount PRIVATE "${HAVE_FLOAT128}" "${DISABLE_INT128}" "${ENABLE_DIV32}" "${ENABLE_ASSERT}" "${ENABLE_MULTIARCH_AVX512_BMI2}")
+    target_compile_definitions(libprimecount PRIVATE "${HAVE_FLOAT128}" "${DISABLE_INT128}" "${ENABLE_DIV32}" "${ENABLE_ASSERT}" "${ENABLE_MULTIARCH_AVX512_BMI2}" "${ENABLE_INT128_STL_PATCH}" "${ENABLE_INT128_OPENMP_PATCH}")
     target_link_libraries(libprimecount PRIVATE primesieve::primesieve "${LIB_OPENMP}" "${LIB_QUADMATH}" "${LIB_ATOMIC}")
 
     target_compile_features(libprimecount
@@ -284,7 +282,7 @@ if(BUILD_STATIC_LIBS)
     add_library(libprimecount-static STATIC ${LIB_SRC})
     set_target_properties(libprimecount-static PROPERTIES OUTPUT_NAME primecount)
     target_compile_options(libprimecount-static PRIVATE "${WNO_UNINITIALIZED}")
-    target_compile_definitions(libprimecount-static PRIVATE "${HAVE_FLOAT128}" "${DISABLE_INT128}" "${ENABLE_DIV32}" "${ENABLE_ASSERT}" "${ENABLE_MULTIARCH_AVX512_BMI2}")
+    target_compile_definitions(libprimecount-static PRIVATE "${HAVE_FLOAT128}" "${DISABLE_INT128}" "${ENABLE_DIV32}" "${ENABLE_ASSERT}" "${ENABLE_MULTIARCH_AVX512_BMI2}" "${ENABLE_INT128_STL_PATCH}" "${ENABLE_INT128_OPENMP_PATCH}")
     target_link_libraries(libprimecount-static PRIVATE primesieve::primesieve "${LIB_OPENMP}" "${LIB_QUADMATH}" "${LIB_ATOMIC}")
 
     if(WITH_MSVC_CRT_STATIC)
@@ -320,7 +318,7 @@ endif()
 if(BUILD_PRIMECOUNT)
     add_executable(primecount ${BIN_SRC})
     target_link_libraries(primecount PRIVATE primecount::primecount primesieve::primesieve)
-    target_compile_definitions(primecount PRIVATE "${DISABLE_INT128}" "${ENABLE_DIV32}" "${ENABLE_ASSERT}")
+    target_compile_definitions(primecount PRIVATE "${DISABLE_INT128}" "${ENABLE_DIV32}" "${ENABLE_ASSERT}" "${ENABLE_INT128_STL_PATCH}" "${ENABLE_INT128_OPENMP_PATCH}")
     target_compile_features(primecount PRIVATE cxx_auto_type)
     install(TARGETS primecount DESTINATION ${CMAKE_INSTALL_BINDIR})
 

ChangeLog

@@ -1,9 +1,14 @@
-Changes in primecount-7.13, 2024-04-04
+Changes in primecount-7.13, 2024-04-08
 
+* Delete ci-sage.yml, it has not been updated in 3 years and I
+  don't want to maintain it. I feel like this CI script should
+  be part of SageMath, not primecount.
 * Added AVX512 runtime dispatching for x86 CPUs.
 * CMakeLists.txt: New WITH_MULTIARCH option (default ON).
 * Sieve_count.cpp: New AVX512 popcount algorithm for x86 CPUs.
 * Sieve_count.cpp: New ARM SVE popcount algorithm.
+* int128.cmake: Improve int128_t support for Windows.
+* OpenMP.cmake: Improve LLVM/Clang OpenMP detection.
 
 Changes in primecount-7.12, 2024-04-01
 

cmake/OpenMP.cmake

@@ -1,3 +1,6 @@
+# Check if OpenMP supports 128-bit integers out of the box
+# or if we have to link against libatomic.
+
 include(CheckCXXSourceCompiles)
 include(CMakePushCheckState)
 
@@ -8,8 +11,8 @@ if(NOT (OpenMP_FOUND OR OpenMP_CXX_FOUND))
     message(STATUS "Performing Test OpenMP - Failed")
 endif()
 
-# Check if libatomic is required (usually by Clang)
-# to support 128-bit integers with OpenMP.
+# CMake has found OpenMP, now we need to check
+# if OpenMP supports 128-bit integers.
 if(OpenMP_FOUND OR OpenMP_CXX_FOUND)
     cmake_push_check_state()
     set(CMAKE_REQUIRED_INCLUDES "${PROJECT_SOURCE_DIR}/include")
@@ -24,7 +27,7 @@ if(OpenMP_FOUND OR OpenMP_CXX_FOUND)
         set(CMAKE_REQUIRED_DEFINITIONS "-D${DISABLE_INT128}")
     endif()
 
-    # Check if compiles without libatomic
+    # Check if OpenMP supports 128-bit integers
     check_cxx_source_compiles("
         #include <int128_t.hpp>
         #include <omp.h>
@@ -42,8 +45,9 @@ if(OpenMP_FOUND OR OpenMP_CXX_FOUND)
             return 0;
         }" OpenMP)
 
-    # Our code requires libatomic to compile
     if(NOT OpenMP)
+        # Try if OpenMP works if we link against libatomic.
+        # This is sometimes required for LLVM/Clang.
         find_library(LIB_ATOMIC NAMES atomic atomic.so.1 libatomic.so.1)
 
         if(NOT LIB_ATOMIC)
@@ -58,7 +62,7 @@ if(OpenMP_FOUND OR OpenMP_CXX_FOUND)
 
         set(CMAKE_REQUIRED_LIBRARIES "${CMAKE_REQUIRED_LIBRARIES}" "${LIB_ATOMIC}")
 
-        # Check if compiles with libatomic
+        # Check if OpenMP compiles with libatomic
         check_cxx_source_compiles("
             #include <int128_t.hpp>
             #include <omp.h>
@@ -78,13 +82,42 @@ if(OpenMP_FOUND OR OpenMP_CXX_FOUND)
 
         if(NOT OpenMP_with_libatomic)
             set(LIB_ATOMIC "")
+
+            if (NOT DISABLE_INT128)
+                # As a last resort check if OpenMP supports int128_t if
+                # we include our <int128_OpenMP_patch.hpp> header.
+                # In this case OpenMP will use critical sections instead
+                # of atomics for 128-bit integers which is slightly less
+                # efficient. (required for LLVM/Clang on Windows)
+                check_cxx_source_compiles("
+                    #include <int128_t.hpp>
+                    #include <int128_OpenMP_patch.hpp>
+                    #include <omp.h>
+                    #include <stdint.h>
+                    #include <iostream>
+                    int main(int, char** argv) {
+                        using primecount::maxint_t;
+                        uintptr_t n = (uintptr_t) argv;
+                        maxint_t sum = (maxint_t) n;
+                        int iters = (int) n;
+                        #pragma omp parallel for reduction(+: sum)
+                        for (int i = 0; i < iters; i++)
+                            sum += (i / 3) * omp_get_thread_num();
+                        std::cout << (long) sum;
+                        return 0;
+                    }" OpenMP_int128_patch)
+
+                if(OpenMP_int128_patch)
+                    set(ENABLE_INT128_OPENMP_PATCH "ENABLE_INT128_OPENMP_PATCH")
+                endif()
+            endif()
         endif()
     endif()
 
     cmake_pop_check_state()
 
     # OpenMP has been tested successfully, enable it
-    if(OpenMP OR OpenMP_with_libatomic)
+    if(OpenMP OR OpenMP_with_libatomic OR OpenMP_int128_patch)
         if(TARGET OpenMP::OpenMP_CXX)
             set(LIB_OPENMP "OpenMP::OpenMP_CXX")
         else()
@@ -99,7 +132,7 @@ if(OpenMP_FOUND OR OpenMP_CXX_FOUND)
 endif()
 
 # OpenMP test has failed, print warning message
-if(NOT OpenMP AND NOT OpenMP_with_libatomic)
+if(NOT OpenMP AND NOT OpenMP_with_libatomic AND NOT OpenMP_int128_patch)
     if (CMAKE_CXX_COMPILER_ID MATCHES "Clang|LLVM")
         message(WARNING "Install the OpenMP library (libomp) to enable multithreading in primecount!")
     elseif (CMAKE_CXX_COMPILER_ID STREQUAL "GNU")