Support C++20 and beyond

.gitmodules

@@ -1,9 +1,6 @@
 [submodule "librapid/vendor/jitify"]
  path = librapid/vendor/jitify
  url = https://github.com/Pencilcaseman/jitify.git
-[submodule "librapid/vendor/Vc"]
- path = librapid/vendor/Vc
- url = https://github.com/Pencilcaseman/Vc.git
 [submodule "librapid/vendor/fmt"]
  path = librapid/vendor/fmt
  url = https://github.com/fmtlib/fmt.git
@@ -19,3 +16,6 @@
 [submodule "librapid/vendor/CLBlast"]
  path = librapid/vendor/CLBlast
  url = https://github.com/CNugteren/CLBlast.git
+[submodule "librapid/vendor/xsimd"]
+ path = librapid/vendor/xsimd
+ url = https://github.com/LibRapid/xsimd.git

CMakeLists.txt

@@ -2,7 +2,16 @@ cmake_minimum_required(VERSION 3.16)
 project(librapid)
 list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/cmake")
 cmake_policy(SET CMP0077 NEW)
-set(CMAKE_CXX_STANDARD 17)
+
+if (CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR)
+ message(STATUS "[ LIBRAPID ] LibRapid is a top-level project. Using C++23")
+ set(CMAKE_CXX_STANDARD 23)
+endif ()
+
+# LibRapid requires C++20 or later
+if (CMAKE_CXX_STANDARD LESS 20)
+ message(FATAL_ERROR "LibRapid requires C++20 or later")
+endif ()
 
 # Extract version information
 file(READ "version.txt" ver)
@@ -38,7 +47,7 @@ option(LIBRAPID_USE_OPENCL "Search for OpenCL and use it if possible" ON)
 option(LIBRAPID_USE_CUDA "Attempt to use CUDA" ON)
 option(LIBRAPID_USE_MULTIPREC "Include MPIR and MPFR in the LibRapid build" OFF)
 option(LIBRAPID_FAST_MATH "Use potentially less accurate operations to increase performance" OFF)
-option(LIBRAPID_NATIVE_ARCH "Use the native architecture of the system" OFF)
+option(LIBRAPID_NATIVE_ARCH "Use the native architecture of the system" ON)
 
 option(LIBRAPID_CUDA_DOUBLE_VECTOR_WIDTH "Preferred vector width for vectorised kernels" 2)
 option(LIBRAPID_CUDA_FLOAT_VECTOR_WIDTH "Preferred vector width for vectorised kernels" 4)
@@ -124,6 +133,12 @@ if (LIBRAPID_STRICT AND LIBRAPID_QUIET)
  message(FATAL_ERROR "LIBRAPID_STRICT and LIBRAPID_QUIET cannot be enabled at the same time")
 endif ()
 
+# SIMD instructions do not currently work on MacOS
+#if (IS_MACOS AND LIBRAPID_NATIVE_ARCH)
+# message(WARNING "SIMD instructions are not currently supported on MacOS. Disabling LIBRAPID_NATIVE_ARCH")
+# set(LIBRAPID_NATIVE_ARCH OFF)
+#endif ()
+
 if (LIBRAPID_STRICT)
  # Enable all warnings and treat them as errors
  if (MSVC)
@@ -415,19 +430,14 @@ endif ()
 
 # Add dependencies
 add_subdirectory("${CMAKE_CURRENT_SOURCE_DIR}/librapid/vendor/fmt")
-add_subdirectory("${CMAKE_CURRENT_SOURCE_DIR}/librapid/vendor/Vc")
-
-if (NOT MINGW)
- # scnlib does not support MinGW, since it does not implement std::from_chars, which is required by the library
- add_subdirectory("${CMAKE_CURRENT_SOURCE_DIR}/librapid/vendor/scnlib")
-else ()
- message(WARNING "[ LIBRAPID ] scnlib cannot be built by MinGW, so it will not be enabled")
- target_compile_definitions(${module_name} PUBLIC LIBRAPID_MINGW)
-endif ()
+# add_subdirectory("${CMAKE_CURRENT_SOURCE_DIR}/librapid/vendor/Vc")
+add_subdirectory("${CMAKE_CURRENT_SOURCE_DIR}/librapid/vendor/xsimd")
+add_subdirectory("${CMAKE_CURRENT_SOURCE_DIR}/librapid/vendor/scnlib")
 
 target_compile_definitions(fmt PUBLIC FMT_HEADER_ONLY)
-target_compile_definitions(Vc PRIVATE Vc_HACK_OSTREAM_FOR_TTY)
-target_link_libraries(${module_name} PUBLIC fmt scn Vc)
+# target_compile_definitions(Vc PRIVATE Vc_HACK_OSTREAM_FOR_TTY)
+# target_link_libraries(${module_name} PUBLIC fmt scn Vc xsimd)
+target_link_libraries(${module_name} PUBLIC fmt scn xsimd)
 
 if (${LIBRAPID_USE_MULTIPREC})
  # Load MPIR
@@ -484,15 +494,19 @@ if (LIBRAPID_FAST_MATH)
  target_compile_definitions(${module_name} PUBLIC LIBRAPID_FAST_MATH)
 endif ()
 
-set(LIBRAPID_ARCH_FLAGS)
 if (LIBRAPID_NATIVE_ARCH)
  message(STATUS "[ LIBRAPID ] Compiling for native architecture")
- OptimizeForArchitecture()
- target_compile_options(${module_name} PUBLIC ${Vc_DEFINITIONS} ${Vc_ARCHITECTURE_FLAGS})
+
+ include(ArchDetect2)
+ target_compile_options(${module_name} PUBLIC ${LIBRAPID_ARCH_FLAGS})
  target_compile_definitions(${module_name} PUBLIC LIBRAPID_NATIVE_ARCH)
- set(LIBRAPID_ARCH_FLAGS ${Vc_DEFINITIONS} ${Vc_ARCHITECTURE_FLAGS})
- message(STATUS "[ LIBRAPID ] Additional Definitions: ${Vc_DEFINITIONS}")
- message(STATUS "[ LIBRAPID ] Supported flags: ${Vc_ARCHITECTURE_FLAGS}")
+
+ # OptimizeForArchitecture()
+ # target_compile_options(${module_name} PUBLIC ${Vc_DEFINITIONS} ${Vc_ARCHITECTURE_FLAGS})
+ # target_compile_definitions(${module_name} PUBLIC LIBRAPID_NATIVE_ARCH)
+ # set(LIBRAPID_ARCH_FLAGS ${Vc_DEFINITIONS} ${Vc_ARCHITECTURE_FLAGS})
+ # message(STATUS "[ LIBRAPID ] Additional Definitions: ${Vc_DEFINITIONS}")
+ # message(STATUS "[ LIBRAPID ] Supported flags: ${Vc_ARCHITECTURE_FLAGS}")
 endif ()
 
 # Add defines for CUDA vector widths

cmake/ArchDetect2.cmake

@@ -0,0 +1,243 @@
+INCLUDE(CheckCXXSourceRuns)
+
+set(COMPILER_GNU false)
+set(COMPILER_INTEL false)
+set(COMPILER_CLANG false)
+set(COMPILER_MSVC false)
+
+if (CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
+ set(COMPILER_GNU true)
+elseif (CMAKE_CXX_COMPILER_ID STREQUAL "Intel")
+ set(COMPILER_INTEL true)
+elseif (CMAKE_CXX_COMPILER_ID STREQUAL "Clang")
+ set(COMPILER_CLANG true)
+elseif (CMAKE_CXX_COMPILER_ID STREQUAL "MSVC")
+ set(COMPILER_MSVC true)
+else ()
+ # Unknown Compiler
+endif ()
+
+set(LIBRAPID_ARCH_FLAGS)
+set(LIBRAPID_ARCH_FOUND)
+
+# Function to test a given SIMD capability
+function(check_simd_capability FLAG_GNU FLAG_MSVC NAME TEST_SOURCE VAR)
+ set(CMAKE_REQUIRED_FLAGS)
+ if (COMPILER_GNU OR COMPILER_INTEL OR COMPILER_CLANG)
+ set(CMAKE_REQUIRED_FLAGS "${FLAG_GNU}")
+ elseif (COMPILER_MSVC) # reserve for WINDOWS
+ set(CMAKE_REQUIRED_FLAGS "${FLAG_MSVC}")
+ endif ()
+
+ CHECK_CXX_SOURCE_RUNS("${TEST_SOURCE}" ${VAR})
+
+ if (${${VAR}})
+ if (COMPILER_GNU OR COMPILER_INTEL OR COMPILER_CLANG)
+ # set(LIBRAPID_ARCH_FLAGS "${LIBRAPID_ARCH_FLAGS} ${FLAG_GNU}" PARENT_SCOPE)
+
+ list(APPEND LIBRAPID_ARCH_FLAGS ${FLAG_GNU})
+ set(LIBRAPID_ARCH_FLAGS ${LIBRAPID_ARCH_FLAGS} PARENT_SCOPE)
+
+ message(STATUS "[ LIBRAPID ] ${NAME} found: ${FLAG_GNU}")
+ elseif (MSVC)
+ # set(LIBRAPID_ARCH_FLAGS "${LIBRAPID_ARCH_FLAGS} ${FLAG_MSVC}" PARENT_SCOPE)
+
+ list(APPEND LIBRAPID_ARCH_FLAGS ${FLAG_MSVC})
+ set(LIBRAPID_ARCH_FLAGS ${LIBRAPID_ARCH_FLAGS} PARENT_SCOPE)
+
+ message(STATUS "[ LIBRAPID ] ${NAME} found: ${FLAG_MSVC}")
+ endif ()
+ set(LIBRAPID_ARCH_FOUND TRUE PARENT_SCOPE)
+ else ()
+ message(STATUS "[ LIBRAPID ] ${NAME} not found")
+ endif ()
+endfunction()
+
+# Check SSE2 (not a valid flag for MSVC)
+check_simd_capability("-msse2" "" "SSE2" "
+#include <emmintrin.h>
+int main() {
+ __m128i a = _mm_set_epi32 (-1, 2, -3, 4);
+ __m128i result = _mm_abs_epi32 (a);
+ return 0;
+}" SIMD_SSE2)
+
+# Check SSE3 (not a valid flag for MSVC)
+check_simd_capability("-msse3" "" "SSE3" "
+#include <pmmintrin.h>
+int main() {
+ __m128 a = _mm_set_ps (-1.0f, 2.0f, -3.0f, 4.0f);
+ __m128 b = _mm_set_ps (1.0f, 2.0f, 3.0f, 4.0f);
+ __m128 result = _mm_addsub_ps (a, b);
+ return 0;
+}" SIMD_SSE3)
+
+# Check SSSE3 (not a valid flag for MSVC)
+check_simd_capability("-mssse3" "" "SSSE3" "
+#include <tmmintrin.h>
+int main() {
+ __m128i a = _mm_set_epi8(-1, 2, -3, 4, -1, 2, -3, 4, -1, 2, -3, 4, -1, 2, -3, 4);
+ __m128i result = _mm_abs_epi8(a);
+ return 0;
+}" SIMD_SSSE3)
+
+# Check SSE4.1 (not a valid flag for MSVC)
+check_simd_capability("-msse4.1" "" "SSE4.1" "
+#include <smmintrin.h>
+int main() {
+ __m128i a = _mm_set_epi32(-1, 2, -3, 4);
+ __m128i result = _mm_abs_epi32(a);
+ return 0;
+}" SIMD_SSE4_1)
+
+# Check SSE4.2 (not a valid flag for MSVC)
+check_simd_capability("-msse4.2" "" "SSE4.2" "
+#include <nmmintrin.h>
+int main() {
+ __m128i a = _mm_set_epi32(-1, 2, -3, 4);
+ __m128i result = _mm_abs_epi32(a);
+ return 0;
+}" SIMD_SSE4_2)
+
+# Check AVX
+check_simd_capability("-mavx" "/arch:AVX" "AVX" "
+#include <immintrin.h>
+int main() {
+ __m256 a = _mm256_set_ps(-1.0f, 2.0f, -3.0f, 4.0f, -1.0f, 2.0f, -3.0f, 4.0f);
+ __m256 result = _mm256_abs_ps(a);
+ return 0;
+}" SIMD_AVX)
+
+# Check AVX2
+check_simd_capability("-mavx2" "/arch:AVX2" "AVX2" "
+#include <immintrin.h>
+int main() {
+ __m256i a = _mm256_set_epi32(-1, 2, -3, 4, -1, 2, -3, 4);
+ __m256i result = _mm256_abs_epi32(a);
+ return 0;
+}" SIMD_AVX2)
+
+# Check AVX512F
+check_simd_capability("-mavx512f" "/arch:AVX512" "AVX512F" "
+#include <immintrin.h>
+int main() {
+ __m512i a = _mm512_set_epi32(-1, 2, -3, 4, -1, 2, -3, 4, -1, 2, -3, 4, -1, 2, -3, 4);
+ __m512i result = _mm512_abs_epi32(a);
+ return 0;
+}" SIMD_AVX512F)
+
+# Check AVX512BW
+check_simd_capability("-mavx512bw" "/arch:AVX512" "AVX512BW" "
+#include <immintrin.h>
+int main() {
+ __m512i a = _mm512_set_epi64(-1, 2, -3, 4, -1, 2, -3, 4);
+ __m512i result = _mm512_abs_epi8(a);
+ return 0;
+}" SIMD_AVX512BW)
+
+# Check AVX512CD
+check_simd_capability("-mavx512cd" "/arch:AVX512" "AVX512CD" "
+#include <immintrin.h>
+int main() {
+ __m512i a = _mm512_set_epi64(-1, 2, -3, 4, -1, 2, -3, 4);
+ __m512i result = _mm512_conflict_epi64(a);
+ return 0;
+}" SIMD_AVX512CD)
+
+# Check AVX512DQ
+check_simd_capability("-mavx512dq" "/arch:AVX512" "AVX512DQ" "
+#include <immintrin.h>
+int main() {
+ __m512d a = _mm512_set_pd(-1.0, 2.0, -3.0, 4.0, -1.0, 2.0, -3.0, 4.0);
+ __m512d result = _mm512_abs_pd(a);
+ return 0;
+}" SIMD_AVX512DQ)
+
+# Check AVX512ER
+check_simd_capability("-mavx512er" "/arch:AVX512" "AVX512ER" "
+#include <immintrin.h>
+int main() {
+ __m512d a = _mm512_set_pd(-1.0, 2.0, -3.0, 4.0, -1.0, 2.0, -3.0, 4.0);
+ __m512d result = _mm512_exp_pd(a);
+ return 0;
+}" SIMD_AVX512ER)
+
+# Check AVX512PF
+check_simd_capability("-mavx512pf" "/arch:AVX512" "AVX512PF" "
+#include <immintrin.h>
+int main() {
+ __m512 a = _mm512_set_ps(-1.0f, 2.0f, -3.0f, 4.0f, -1.0f, 2.0f, -3.0f, 4.0f);
+ __m512 result = _mm512_exp_ps(a);
+ return 0;
+}" SIMD_AVX512PF)
+
+# ARM
+check_simd_capability("-march=armv7-a" "" "ARMv7" "
+#include <arm_neon.h>
+int main() {
+ int32x4_t a = vdupq_n_s32(1);
+ int32x4_t b = vdupq_n_s32(2);
+ int32x4_t result = vaddq_s32(a, b);
+ return 0;
+}" SIMD_ARMv7)
+
+check_simd_capability("-march=armv8-a" "" "ARMv8" "
+#include <arm_neon.h>
+int main() {
+ int32x4_t a = vdupq_n_s32(1);
+ int32x4_t b = vdupq_n_s32(2);
+ int32x4_t result = vaddq_s32(a, b);
+ return 0;
+}" SIMD_ARMv8)
+
+# ARM64
+check_simd_capability("-march=armv8.1-a" "" "ARMv8.1" "
+#include <arm_neon.h>
+int main() {
+ int32x4_t a = vdupq_n_s32(1);
+ int32x4_t b = vdupq_n_s32(2);
+ int32x4_t result = vaddq_s32(a, b);
+ return 0;
+}" SIMD_ARMv8_1)
+
+check_simd_capability("-march=armv8.2-a" "" "ARMv8.2" "
+#include <arm_neon.h>
+int main() {
+ int32x4_t a = vdupq_n_s32(1);
+ int32x4_t b = vdupq_n_s32(2);
+ int32x4_t result = vaddq_s32(a, b);
+ return 0;
+}" SIMD_ARMv8_2)
+
+check_simd_capability("-march=armv8.3-a" "" "ARMv8.3" "
+#include <arm_neon.h>
+int main() {
+ int32x4_t a = vdupq_n_s32(1);
+ int32x4_t b = vdupq_n_s32(2);
+ int32x4_t result = vaddq_s32(a, b);
+ return 0;
+}" SIMD_ARMv8_3)
+
+check_simd_capability("-march=armv8.4-a" "" "ARMv8.4" "
+#include <arm_neon.h>
+int main() {
+ int32x4_t a = vdupq_n_s32(1);
+ int32x4_t b = vdupq_n_s32(2);
+ int32x4_t result = vaddq_s32(a, b);
+ return 0;
+}" SIMD_ARMv8_4)
+
+check_simd_capability("-march=armv8.5-a" "" "ARMv8.5" "
+#include <arm_neon.h>
+int main() {
+ int32x4_t a = vdupq_n_s32(1);
+ int32x4_t b = vdupq_n_s32(2);
+ int32x4_t result = vaddq_s32(a, b);
+ return 0;
+}" SIMD_ARMv8_5)
+
+if (LIBRAPID_ARCH_FOUND)
+ message(STATUS "[ LIBRAPID ] Architecture Flags: ${LIBRAPID_ARCH_FLAGS}")
+else()
+ message(STATUS "[ LIBRAPID ] Architecture Flags Not Found")
+endif()

docs/source/cmakeIntegration.md

@@ -172,7 +172,7 @@ but may cause some functions to return slightly incorrect results due to lower p
 ### ``LIBRAPID_NATIVE_ARCH``
 
 ```
-DEFAULT: OFF
+DEFAULT: ON
 ```
 
 Enabling this flag compiles librapid with the most advanced instruction set available on the system. This can lead to

examples/example-vector-1.cpp

@@ -5,6 +5,7 @@ namespace lrc = librapid;
 auto main() -> int {
  fmt::print("LibRapid Example -- Vector 1\n");
 
+#if 0 // Currently broken -- switching SIMD backend
  // Create a 3 dimensional vector
  lrc::Vec3d myVector(2, 3, 4);
  lrc::Vec3d myOtherVector(10, 5, 8);
@@ -52,6 +53,7 @@ auto main() -> int {
  fmt::print("One vector: {}\n", one);
  fmt::print("Full vector: {}\n", full);
  fmt::print("Random vector: {:.3f}\n", random);
+#endif
 
  return 0;
 }

librapid/include/librapid/array/array.hpp

@@ -6,11 +6,11 @@
 #include "storage.hpp"
 
 #if defined(LIBRAPID_HAS_OPENCL)
-# include "../OpenCL/openclStorage.hpp"
+# include "../OpenCL/openclStorage.hpp"
 #endif // LIBRAPID_HAS_OPENCL
 
 #if defined(LIBRAPID_HAS_CUDA)
-# include "../cuda/cudaStorage.hpp"
+# include "../cuda/cudaStorage.hpp"
 #endif // LIBRAPID_HAS_CUDA
 
 #include "arrayTypeDef.hpp"