implementation/arm/neon+sve

CMakeLists.txt

@@ -1,4 +1,4 @@
-cmake_minimum_required(VERSION 3.30)
+cmake_minimum_required(VERSION 3.24)
 project(pernix VERSION 0.1.0 LANGUAGES CXX C)
 
 option(PERNIX_ENABLE_TESTS "Enable tests for pernix" on)
@@ -8,19 +8,49 @@ option(PERNIX_ENABLE_INSTALL "Enable installation of pernix" on)
 option(PERNIX_ENABLE_DOXYGEN "Build documentation with Doxygen." off)
 option(PERNIX_INSTALL_DOCS "Install documentation with pernix" off)
 
-option(PERNIX_DISABLE_BMI2 "Disable BMI2 optimizations" off)
-option(PERNIX_DISABLE_AVX2 "Disable AVX2 optimizations" off)
-option(PERNIX_DISABLE_AVX512 "Disable AVX512 optimizations" off)
+option(PERNIX_ENABLE_X86_BMI2 "Build x86 BMI2 backend" ON)
+option(PERNIX_ENABLE_X86_AVX2 "Build x86 AVX2 backend" ON)
+option(PERNIX_ENABLE_X86_AVX512VBMI "Build x86 AVX512-VBMI backend" ON)
+option(PERNIX_ENABLE_ARM64_NEON "Build arm64 NEON backend" ON)
+option(PERNIX_ENABLE_ARM64_SVE2 "Build arm64 SVE2 backend" ON)
 
 option(PERNIX_USE_SIMDE "Use SIMDe library for portable SIMD support" off)
 
 option(PERNIX_ENABLE_FORTRAN_BINDINGS "Build Fortran bindings for pernix" off)
 
 list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/cmake/")
 
+#[===[
+set(PERNIX_BUNDLE_SIMDE_FOR_INSTALL OFF)
 if (PERNIX_USE_SIMDE)
-    add_subdirectory(external/simde EXCLUDE_FROM_ALL)
+    if (PERNIX_SIMDE_PROVIDER STREQUAL "AUTO" OR PERNIX_SIMDE_PROVIDER STREQUAL "PACKAGE")
+        find_package(simde CONFIG QUIET)
+    endif ()
+
+    if (NOT TARGET simde::simde AND (PERNIX_SIMDE_PROVIDER STREQUAL "AUTO" OR PERNIX_SIMDE_PROVIDER STREQUAL "FETCH"))
+        include(FetchContent)
+        set(SIMDE_TEST_CMAKE_PACKAGING OFF CACHE BOOL "Test SIMDe CMake packaging" FORCE)
+        FetchContent_Declare(
+                simde
+                GIT_REPOSITORY https://github.com/simd-everywhere/simde.git
+                GIT_TAG f3e8262173b7089db9a9d57a9ecef8dd07ad9c97
+                GIT_PROGRESS TRUE
+                EXCLUDE_FROM_ALL
+        )
+        FetchContent_MakeAvailable(simde)
+        set(PERNIX_BUNDLE_SIMDE_FOR_INSTALL ON)
+    endif ()
+
+    if (NOT TARGET simde::simde AND DEFINED simde_SOURCE_DIR AND EXISTS "${simde_SOURCE_DIR}/simde")
+        add_library(simde::simde INTERFACE IMPORTED GLOBAL)
+        target_include_directories(simde::simde INTERFACE "${simde_SOURCE_DIR}")
+    endif ()
+
+    if (NOT TARGET simde::simde)
+        message(FATAL_ERROR "PERNIX_USE_SIMDE is enabled, but simde::simde was not found. Set PERNIX_SIMDE_PROVIDER=FETCH or install SIMDe's CMake package.")
+    endif ()
 endif ()
+]===]
 
 include(CTest)
 include(GitVersion)
@@ -40,51 +70,73 @@ else ()
 endif ()
 message(STATUS "Pernix version: ${VERSION}, normalized to ${NORMALIZED_VERSION}")
 
-set(BENCHMARK_CXX_STANDARD 20)
+if (NOT CMAKE_CXX_COMPILER_ID MATCHES "GNU|Clang|Intel")
+    message(FATAL_ERROR "Unsupported compiler: ${CMAKE_CXX_COMPILER_ID}")
+endif ()
 
-set(CMAKE_CXX_STANDARD ${BENCHMARK_CXX_STANDARD})
-set(CMAKE_CXX_STANDARD_REQUIRED YES)
-set(CMAKE_CXX_EXTENSIONS OFF)
+include(CheckCXXCompilerFlag)
+set(PERNIX_PRIVATE_COMPILE_OPTIONS)
+foreach (PERNIX_CXX_FLAG
+        -Wall
+        -Wextra
+        -Wshadow
+        -Wfloat-equal
+        -Wold-style-cast
+        -Wconversion
+        -fstrict-aliasing
+        -Wno-ignored-attributes
+)
+    string(MAKE_C_IDENTIFIER "PERNIX_HAS_CXX_FLAG_${PERNIX_CXX_FLAG}" PERNIX_CXX_FLAG_VARIABLE)
+    check_cxx_compiler_flag("${PERNIX_CXX_FLAG}" "${PERNIX_CXX_FLAG_VARIABLE}")
+    if (${PERNIX_CXX_FLAG_VARIABLE})
+        list(APPEND PERNIX_PRIVATE_COMPILE_OPTIONS "${PERNIX_CXX_FLAG}")
+    else ()
+        message(STATUS "Compiler flag not supported: ${PERNIX_CXX_FLAG}")
+    endif ()
+endforeach ()
 
-include(AddCXXCompilerFlag)
-if (MSVC)
-    message(FATAL_ERROR "MSVC compiler is not supported")
-else ()
-    add_cxx_compiler_flag(-Wall)
-    add_cxx_compiler_flag(-Wextra)
-    add_cxx_compiler_flag(-Wshadow)
-    add_cxx_compiler_flag(-Wfloat-equal)
-    add_cxx_compiler_flag(-Wold-style-cast)
-    add_cxx_compiler_flag(-Wconversion)
-    add_cxx_compiler_flag(-fstrict-aliasing)
-    add_cxx_compiler_flag(-Wno-ignored-attributes)
-
-    if (PERNIX_ENABLE_LTO)
-        add_cxx_compiler_flag(-flto=auto)
-        add_cxx_compiler_flag(-Wno-lto-type-mismatch)
-        if ("${CMAKE_CXX_COMPILER_ID}" STREQUAL "GNU")
-            find_program(GCC_AR gcc-ar)
-            if (GCC_AR)
-                set(CMAKE_AR ${GCC_AR})
-            endif ()
-            find_program(GCC_RANLIB gcc-ranlib)
-            if (GCC_RANLIB)
-                set(CMAKE_RANLIB ${GCC_RANLIB})
-            endif ()
-        elseif ("${CMAKE_CXX_COMPILER_ID}" MATCHES "Clang")
-            find_program(LLVM_AR llvm-ar)
-            if (LLVM_AR)
-                set(CMAKE_AR ${LLVM_AR})
-            endif ()
-            find_program(LLVM_RANLIB llvm-ranlib)
-            if (LLVM_RANLIB)
-                set(CMAKE_RANLIB ${LLVM_RANLIB})
-            endif ()
+if (PERNIX_ENABLE_LTO)
+    include(CheckIPOSupported)
+    check_ipo_supported(RESULT PERNIX_IPO_SUPPORTED OUTPUT PERNIX_IPO_ERROR)
+    if (NOT PERNIX_IPO_SUPPORTED)
+        message(FATAL_ERROR "PERNIX_ENABLE_LTO is enabled, but IPO/LTO is not supported: ${PERNIX_IPO_ERROR}")
+    endif ()
+
+    check_cxx_compiler_flag("-Wno-lto-type-mismatch" PERNIX_HAS_CXX_FLAG_WNO_LTO_TYPE_MISMATCH)
+    if (PERNIX_HAS_CXX_FLAG_WNO_LTO_TYPE_MISMATCH)
+        list(APPEND PERNIX_PRIVATE_COMPILE_OPTIONS "-Wno-lto-type-mismatch")
+    endif ()
+
+    if ("${CMAKE_CXX_COMPILER_ID}" STREQUAL "GNU")
+        find_program(GCC_AR gcc-ar)
+        if (GCC_AR)
+            set(CMAKE_AR ${GCC_AR})
+        endif ()
+        find_program(GCC_RANLIB gcc-ranlib)
+        if (GCC_RANLIB)
+            set(CMAKE_RANLIB ${GCC_RANLIB})
+        endif ()
+    elseif ("${CMAKE_CXX_COMPILER_ID}" MATCHES "Clang")
+        find_program(LLVM_AR llvm-ar)
+        if (LLVM_AR)
+            set(CMAKE_AR ${LLVM_AR})
+        endif ()
+        find_program(LLVM_RANLIB llvm-ranlib)
+        if (LLVM_RANLIB)
+            set(CMAKE_RANLIB ${LLVM_RANLIB})
         endif ()
     endif ()
 endif ()
 
-include_directories(${PROJECT_SOURCE_DIR}/include)
+set(PERNIX_TARGET_IS_X86 OFF)
+if (CMAKE_SYSTEM_PROCESSOR MATCHES "^(x86_64|AMD64|i[3-6]86|i686)$")
+    set(PERNIX_TARGET_IS_X86 ON)
+endif ()
+
+set(PERNIX_TARGET_IS_ARM64 OFF)
+if (CMAKE_SYSTEM_PROCESSOR MATCHES "^(aarch64|arm64|ARM64)$")
+    set(PERNIX_TARGET_IS_ARM64 ON)
+endif ()
 
 add_subdirectory(src)
 
@@ -97,4 +149,4 @@ endif ()
 if (PERNIX_ENABLE_TESTS)
     enable_testing()
     add_subdirectory(tests)
-endif ()
+endif ()

cmake/pernixConfig.cmake.in

@@ -0,0 +1,18 @@
+@PACKAGE_INIT@
+
+if (@PERNIX_USE_SIMDE@)
+    find_package(simde CONFIG QUIET)
+    if (NOT TARGET simde::simde AND @PERNIX_BUNDLE_SIMDE_FOR_INSTALL@)
+        add_library(simde::simde INTERFACE IMPORTED)
+        target_include_directories(simde::simde INTERFACE "${PACKAGE_PREFIX_DIR}/include")
+    endif ()
+    if (NOT TARGET simde::simde)
+        set(pernix_FOUND FALSE)
+        set(pernix_NOT_FOUND_MESSAGE "pernix was built with SIMDe support, but simde::simde was not found. Install SIMDe's CMake package or use a Pernix package built with bundled SIMDe headers.")
+        return()
+    endif ()
+endif ()
+
+include("${CMAKE_CURRENT_LIST_DIR}/pernixTargets.cmake")
+
+check_required_components(pernix)

include/pernix/compat.h

@@ -0,0 +1,45 @@
+#ifndef PERNIX_COMPAT_H
+#define PERNIX_COMPAT_H
+
+#include <cstdint>
+#include <cmath>
+#include <cstddef>
+
+#ifndef __always_inline
+#if defined(__GNUC__) || defined(__clang__)
+#define __always_inline inline __attribute__((always_inline))
+#elif defined(_MSC_VER)
+#define __always_inline __forceinline
+#else
+#define __always_inline inline
+#endif
+#endif
+
+#if defined(_WIN32) && defined(PERNIX_SHARED)
+#if defined(PERNIX_BUILD_LIB)
+#define PERNIX_API __declspec(dllexport)
+#else
+#define PERNIX_API __declspec(dllimport)
+#endif
+#else
+#define PERNIX_API
+#endif
+
+// Convenient type declarations
+typedef uint8_t u8;
+typedef uint16_t u16;
+typedef uint32_t u32;
+typedef uint64_t u64;
+typedef uintptr_t uptr;
+
+typedef int8_t i8;
+typedef int16_t i16;
+typedef int32_t i32;
+typedef int64_t i64;
+
+typedef float_t f32;
+typedef double_t f64;
+
+typedef size_t usize;
+
+#endif //PERNIX_COMPAT_H