Support multi-arch compilation on OSX with CMAKE_OSX_ARCHITECTURES

cmake/ArchDetect.cmake

@@ -1,48 +1,7 @@
-
 function(target_architecture output_arch output_arch_ver)
     if(APPLE AND CMAKE_OSX_ARCHITECTURES)
-        # On OS X we use CMAKE_OSX_ARCHITECTURES *if* it was set
-        # First let's normalize the order of the values
-
-        # Note that it's not possible to compile PowerPC applications if you are using
-        # the OS X SDK version 10.6 or later - you'll need 10.4/10.5 for that, so we
-        # disable it by default
-        # See this page for more information:
-        # http://stackoverflow.com/questions/5333490/how-can-we-restore-ppc-ppc64-as-well-as-full-10-4-10-5-sdk-support-to-xcode-4
-
-        # Architecture defaults to x86 or ppc on OS X 10.5 and earlier, depending on the CPU type detected at runtime.
-        # On OS X 10.6+ the default is x86_64 if the CPU supports it, x86 otherwise.
-
-        foreach(osx_arch ${CMAKE_OSX_ARCHITECTURES})
-            if("${osx_arch}" STREQUAL "ppc" AND ppc_support)
-                set(osx_arch_ppc TRUE)
-            elseif("${osx_arch}" STREQUAL "i386")
-                set(osx_arch_x86 TRUE)
-            elseif("${osx_arch}" STREQUAL "x86_64")
-                set(osx_arch_x86_64 TRUE)
-            elseif("${osx_arch}" STREQUAL "ppc64" AND ppc_support)
-                set(osx_arch_ppc64 TRUE)
-            else()
-                message(FATAL_ERROR "Invalid OS X arch name: ${osx_arch}")
-            endif()
-        endforeach()
-
-        # Now add all the architectures in our normalized order
-        if(osx_arch_ppc)
-            list(APPEND ARCH ppc)
-        endif()
-
-        if(osx_arch_x86)
-            list(APPEND ARCH x86)
-        endif()
-
-        if(osx_arch_x86_64)
-            list(APPEND ARCH x86_64)
-        endif()
-
-        if(osx_arch_ppc64)
-            list(APPEND ARCH ppc64)
-        endif()
+        set(ARCH "${CMAKE_OSX_ARCHITECTURES}")
+        set(ARCH_VER unknown)
     else()
 
         # Detect the architecture in a rather creative way...

cmake/ArchDetect.cpp

@@ -3,13 +3,7 @@
 #define TO_LITERAL_( string ) #string
 #define TO_LITERAL( string ) TO_LITERAL_( string )
 
-#ifdef TEST_FEATURE_SET_ACTIVE
-
-#if FASTSIMD_FEATURE_VALUE( TEST_FEATURE_SET_ACTIVE ) > 0
-static_assert( 0, "TEST_FEATURE_SET_ACTIVE_SUCCESS" );
-#endif
-
-#else
+#if !defined( TEST_FEATURE_SET_ACTIVE ) || FASTSIMD_FEATURE_VALUE( TEST_FEATURE_SET_ACTIVE ) > 0
 
 static_assert( 0, "FASTSIMD_ARCH<" TO_LITERAL( FASTSIMD_ARCH_NAME() ) ">" );
 

dispatch/cmake/ClassSIMD.cmake

@@ -120,26 +120,48 @@ function(fastsimd_create_dispatch_library simd_library_name)
 
     foreach(simd_inl ${fastsimd_create_dispatch_library_SOURCES})
         foreach(feature_set ${fastsimd_create_dispatch_library_FEATURE_SETS})
-            try_run(
-                run_result_unused
-                compile_result_unused
-                "${CMAKE_BINARY_DIR}"
-                "${FastSIMD_SOURCE_DIR}/cmake/ArchDetect.cpp"
-                COMPILE_OUTPUT_VARIABLE COMPILE_OUTPUT
-                CMAKE_FLAGS CMAKE_OSX_ARCHITECTURES=${CMAKE_OSX_ARCHITECTURES}
-                COMPILE_DEFINITIONS -DTEST_FEATURE_SET_ACTIVE=${feature_set}
-            )
-
-            if ("${COMPILE_OUTPUT}" MATCHES "TEST_FEATURE_SET_ACTIVE_SUCCESS")
-                list(APPEND feature_set_list "FastSIMD::FeatureSet::${feature_set}")
-                list(APPEND feature_set_list_debug "${feature_set}")
-                fastsimd_add_feature_set_source(${simd_inl} ${feature_set} ${fastsimd_create_dispatch_library_RELAXED})
+            if(DEFINED CMAKE_OSX_ARCHITECTURES AND NOT "${feature_set}" STREQUAL "SCALAR")
+                # Loop through OSX arches and test compile on each separately
+                foreach(CMAKE_OSX_ARCHITECTURES ${CMAKE_OSX_ARCHITECTURES})
+                    #message(STATUS "${CMAKE_OSX_ARCHITECTURES} ${feature_set}")
+                    try_compile(
+                        compile_result_unused
+                        "${CMAKE_BINARY_DIR}"
+                        "${FastSIMD_SOURCE_DIR}/cmake/ArchDetect.cpp"
+                        OUTPUT_VARIABLE COMPILE_OUTPUT
+                        COMPILE_DEFINITIONS -DTEST_FEATURE_SET_ACTIVE=${feature_set}
+                    )
+
+                    #message(STATUS ${COMPILE_OUTPUT})
+                    if ("${COMPILE_OUTPUT}" MATCHES "FASTSIMD_ARCH<([^\"=]+)=([^>]+)")
+                        set(feature_arch_detect "FASTSIMD_CURRENT_ARCH_IS( ${CMAKE_MATCH_1} )")
+                        fastsimd_add_feature_set_source(${simd_inl} ${feature_set} ${fastsimd_create_dispatch_library_RELAXED})
+                        string(APPEND feature_set_list "#if ${feature_arch_detect}\n,FastSIMD::FeatureSet::${feature_set}\n#endif\n" )
+                        list(APPEND feature_set_list_debug "${feature_set}")
+                        break()
+                    endif()
+                endforeach()
+            else()
+                try_compile(
+                    compile_result_unused
+                    "${CMAKE_BINARY_DIR}"
+                    "${FastSIMD_SOURCE_DIR}/cmake/ArchDetect.cpp"
+                    OUTPUT_VARIABLE COMPILE_OUTPUT
+                    COMPILE_DEFINITIONS -DTEST_FEATURE_SET_ACTIVE=${feature_set}
+                )
+
+                #message(STATUS ${COMPILE_OUTPUT})
+                if ("${COMPILE_OUTPUT}" MATCHES "FASTSIMD_ARCH<([^\"=]+)=([^>]+)")
+                    set(feature_arch_detect "1")
+                    fastsimd_add_feature_set_source(${simd_inl} ${feature_set} ${fastsimd_create_dispatch_library_RELAXED})
+                    string(APPEND feature_set_list ",FastSIMD::FeatureSet::${feature_set}\n" )
+                    list(APPEND feature_set_list_debug "${feature_set}")
+                endif()
             endif()
         endforeach()
     endforeach()
 
     # Create array of compiled feature sets for lookup in FastSIMD::New()
-    string(REPLACE ";" ",\n" feature_set_list "${feature_set_list}")
     configure_file("${FastSIMD_SOURCE_DIR}/dispatch/cmake/simd_lib_config.h.in" "${simd_library_source_dir}/include/FastSIMD/${simd_library_name}_config.h")
 
     message(STATUS "FastSIMD: Created dispatch library \"${simd_library_name}\" with Feature Sets${relaxed_log_msg}: ${feature_set_list_debug}")

dispatch/cmake/feature_set_source.cpp.in

@@ -1,6 +1,9 @@
 #define FASTSIMD_MAX_FEATURE_SET ${feature_set}
+#include <FastSIMD/Utility/ArchDetect.h>
 
+#if ${feature_arch_detect}
 #include <FastSIMD/${simd_library_name}_config.h>
 
 #include "${FastSIMD_SOURCE_DIR}/dispatch/impl/DispatchClassImpl.h"
 #include "${simd_inl_full}"
+#endif

dispatch/cmake/simd_lib_config.h.in

@@ -6,7 +6,7 @@ namespace FastSIMD
 {
 namespace ${simd_library_name}
 {
-using CompiledFeatureSets = FeatureSetList<
+using CompiledFeatureSets = FeatureSetList<0
 ${feature_set_list}>;
 }
 }

examples/dispatch_library/CMakeLists.txt

@@ -1,5 +1,5 @@
 
-fastsimd_create_dispatch_library(example_simd SOURCES "example.inl")
+fastsimd_create_dispatch_library(simd_example_dispatch_library SOURCES "example.inl")
 
 add_executable(example_dispatch_library "main.cpp")
-target_link_libraries(example_dispatch_library PRIVATE FastSIMD example_simd)
+target_link_libraries(example_dispatch_library PRIVATE FastSIMD simd_example_dispatch_library)

include/FastSIMD/ToolSet/Generic/Functions.h

@@ -206,7 +206,7 @@ namespace FS
         }
         else
         {
-            return Register<T, N, SIMD> { SelectHighBit( mask.v0, ifTrue.v0, ifFalse.v0 ), SelectHighBit( mask.v1, ifTrue.v1, ifFalse.v1 ) };
+            return Register<T, N, SIMD>{ SelectHighBit( mask.v0, ifTrue.v0, ifFalse.v0 ), SelectHighBit( mask.v1, ifTrue.v1, ifFalse.v1 ) };
         }
     }
 
@@ -220,7 +220,7 @@ namespace FS
         }
         else
         {
-            return Register<T, N, SIMD> { Increment( a.v0 ), Increment( a.v1 ) };
+            return Register<T, N, SIMD>{ Increment( a.v0 ), Increment( a.v1 ) };
         }
     }
 
@@ -234,7 +234,7 @@ namespace FS
         }
         else
         {
-            return Register<T, N, SIMD> { Decrement( a.v0 ), Decrement( a.v1 ) };
+            return Register<T, N, SIMD>{ Decrement( a.v0 ), Decrement( a.v1 ) };
         }
     }
 
@@ -290,7 +290,7 @@ namespace FS
         }
         else
         {        
-            return Register<T, N, SIMD> { FNMulSub( a.v0, b.v0, c.v0 ), FNMulSub( a.v1, b.v1, c.v1 ) };
+            return Register<T, N, SIMD>{ FNMulSub( a.v0, b.v0, c.v0 ), FNMulSub( a.v1, b.v1, c.v1 ) };
         }
     }
 
@@ -304,7 +304,7 @@ namespace FS
         }
         else
         {
-            return Register<T, N, SIMD> { BitwiseAndNot( a.v0, b.v0 ), BitwiseAndNot( a.v1, b.v1 ) };
+            return Register<T, N, SIMD>{ BitwiseAndNot( a.v0, b.v0 ), BitwiseAndNot( a.v1, b.v1 ) };
         }
     }    
 
@@ -429,7 +429,7 @@ namespace FS
         }
         else
         {        
-            return Register<T, N, SIMD> { MaskedMul( mask.v0, a.v0, b.v0 ), MaskedMul( mask.v1, a.v1, b.v1 ) };
+            return Register<T, N, SIMD>{ MaskedMul( mask.v0, a.v0, b.v0 ), MaskedMul( mask.v1, a.v1, b.v1 ) };
         }
     }
 
@@ -442,7 +442,7 @@ namespace FS
         }
         else
         {        
-            return Register<T, N, SIMD> { InvMaskedMul( mask.v0, a.v0, b.v0 ), InvMaskedMul( mask.v1, a.v1, b.v1 ) };
+            return Register<T, N, SIMD>{ InvMaskedMul( mask.v0, a.v0, b.v0 ), InvMaskedMul( mask.v1, a.v1, b.v1 ) };
         }
     }
 
@@ -517,7 +517,14 @@ namespace FS
     template<typename T, std::size_t N, FastSIMD::FeatureSet SIMD>
     FS_FORCEINLINE Register<T, N, SIMD> Sin( const Register<T, N, SIMD>& a )
     {
-        return Cos( Register<T, N, SIMD>( (T)1.57079632679 ) - a );
+        if constexpr( IsNativeV<Register<T, N, SIMD>> )
+        {
+            return Cos( Register<T, N, SIMD>( (T)1.57079632679 ) - a );
+        }
+        else
+        {
+            return Register<T, N, SIMD>{ Sin( a.v0 ), Sin( a.v1 ) };
+        }
     }
 
     template<std::size_t N, FastSIMD::FeatureSet SIMD>

include/FastSIMD/Utility/FeatureSetList.h

@@ -3,11 +3,11 @@
 
 namespace FastSIMD
 {
-    template<FeatureSet...>
+    template<int, FeatureSet...>
     struct FeatureSetList;
 
     template<FeatureSet HEAD>
-    struct FeatureSetList<HEAD>
+    struct FeatureSetList<0, HEAD>
     {
         static constexpr FeatureSet AsArray[] = { HEAD };
         static constexpr FeatureSet Minimum = HEAD;
@@ -18,14 +18,14 @@ namespace FastSIMD
     };
 
     template<FeatureSet HEAD, FeatureSet... TAIL>
-    struct FeatureSetList<HEAD, TAIL...>
+    struct FeatureSetList<0, HEAD, TAIL...>
     {
         static constexpr FeatureSet AsArray[] = { HEAD, TAIL... };
         static constexpr FeatureSet Minimum = HEAD;
-        static constexpr FeatureSet Maximum = FeatureSetList<TAIL...>::Maximum;
+        static constexpr FeatureSet Maximum = FeatureSetList<0, TAIL...>::Maximum;
 
         template<FeatureSet L>
-        static constexpr FeatureSet NextAfter = (L == HEAD) ? FeatureSetList<TAIL...>::Minimum : FeatureSetList<TAIL...>::template NextAfter<L>;
+        static constexpr FeatureSet NextAfter = (L == HEAD) ? FeatureSetList<0, TAIL...>::Minimum : FeatureSetList<0, TAIL...>::template NextAfter<L>;
     };
 
 }

tests/CMakeLists.txt

@@ -1,9 +1,9 @@
 
-fastsimd_create_dispatch_library(test_simd SOURCES "test.inl" FEATURE_SETS SCALAR SSE2 SSE41 AVX2 AVX512 NEON AARCH64 WASM)
-fastsimd_create_dispatch_library(test_simd_relaxed RELAXED SOURCES "test.inl" FEATURE_SETS SCALAR SSE2 SSE41 AVX2 AVX512 NEON AARCH64 WASM)
+fastsimd_create_dispatch_library(simd_test SOURCES "test.inl" FEATURE_SETS SCALAR SSE2 SSE41 AVX2 AVX512 NEON AARCH64 WASM)
+fastsimd_create_dispatch_library(simd_test_relaxed RELAXED SOURCES "test.inl" FEATURE_SETS SCALAR SSE2 SSE41 AVX2 AVX512 NEON AARCH64 WASM)
 
 add_executable(test "test.cpp")
-target_link_libraries(test PRIVATE FastSIMD test_simd test_simd_relaxed)
+target_link_libraries(test PRIVATE FastSIMD simd_test simd_test_relaxed)
 
 if(CMAKE_SYSTEM_NAME STREQUAL "Emscripten")
   set(CMAKE_EXECUTABLE_SUFFIX ".html")