cmake: Detect AVX + allow build system to disable Intel intrinsics

CMakeLists.txt

@@ -367,6 +367,7 @@ option_string(SDL_ASSERTIONS "Enable internal sanity checks (auto/disabled/relea
 #set_option(SDL_DEPENDENCY_TRACKING "Use gcc -MMD -MT dependency tracking" ON)
 set_option(SDL_ASSEMBLY            "Enable assembly routines" ${SDL_ASSEMBLY_DEFAULT})
 dep_option(SDL_SSEMATH             "Allow GCC to use SSE floating point math" ON "SDL_ASSEMBLY;SDL_CPU_X86 OR SDL_CPU_X64" OFF)
+dep_option(SDL_AVX                 "Use AVX assembly routines" ON "SDL_ASSEMBLY;SDL_CPU_X86 OR SDL_CPU_X64" OFF)
 dep_option(SDL_SSE                 "Use SSE assembly routines" ON "SDL_ASSEMBLY;SDL_CPU_X86 OR SDL_CPU_X64" OFF)
 dep_option(SDL_SSE2                "Use SSE2 assembly routines" ON "SDL_ASSEMBLY;SDL_CPU_X86 OR SDL_CPU_X64" OFF)
 dep_option(SDL_SSE3                "Use SSE3 assembly routines" ON "SDL_ASSEMBLY;SDL_CPU_X86 OR SDL_CPU_X64" OFF)
@@ -711,6 +712,32 @@ if(SDL_ASSEMBLY)
     # TODO: Those all seem to be quite GCC specific - needs to be
     # reworked for better compiler support
     set(HAVE_ASSEMBLY TRUE)
+
+    if(SDL_AVX)
+      cmake_push_check_state()
+      set(CMAKE_REQUIRED_FLAGS "${CMAKE_REQUIRED_FLAGS} -mavx")
+      check_c_source_compiles("
+          #ifdef __MINGW32__
+          #include <_mingw.h>
+          #ifdef __MINGW64_VERSION_MAJOR
+          #include <intrin.h>
+          #else
+          #include <immintrin.h>
+          #endif
+          #else
+          #include <immintrin.h>
+          #endif
+          #ifndef __AVX__
+          #error Assembler CPP flag not enabled
+          #endif
+          int main(int argc, char **argv) { return 0; }" CPU_SUPPORTS_AVX)
+      cmake_pop_check_state()
+      if(CPU_SUPPORTS_AVX)
+        set(HAVE_AVX TRUE)
+        target_compile_options(sdl-build-options INTERFACE "-mavx")
+      endif()
+    endif()
+
     if(SDL_MMX)
       cmake_push_check_state()
       set(CMAKE_REQUIRED_FLAGS "${CMAKE_REQUIRED_FLAGS} -mmmx")
@@ -728,9 +755,10 @@ if(SDL_ASSEMBLY)
           #ifndef __MMX__
           #error Assembler CPP flag not enabled
           #endif
-          int main(int argc, char **argv) { return 0; }" HAVE_MMX)
+          int main(int argc, char **argv) { return 0; }" CPU_SUPPORTS_MMX)
       cmake_pop_check_state()
-      if(HAVE_MMX)
+      if(CPU_SUPPORTS_MMX)
+        set(HAVE_MMX TRUE)
         target_compile_options(sdl-build-options INTERFACE "-mmmx")
       endif()
     endif()
@@ -823,8 +851,6 @@ if(SDL_ASSEMBLY)
       set(HAVE_SSEMATH TRUE)
     endif()
 
-    check_include_file("immintrin.h" HAVE_IMMINTRIN_H)
-
     if(SDL_ALTIVEC)
       cmake_push_check_state()
       set(CMAKE_REQUIRED_FLAGS "${CMAKE_REQUIRED_FLAGS} -maltivec")
@@ -954,10 +980,46 @@ if(SDL_ASSEMBLY)
     if(SDL_SSE3)
       set(HAVE_SSE3 TRUE)
     endif()
-    check_include_file("immintrin.h" HAVE_IMMINTRIN_H)
+    if(SDL_AVX)
+      cmake_push_check_state()
+      # FIXME: should be CMAKE_REQUIRED_LINK_OPTIONS for CMake 3.14+
+      list(APPEND CMAKE_REQUIRED_LIBRARIES "/ARCH:AVX")
+      check_c_source_compiles("
+          #include <immintrin.h>
+          #ifndef __AVX__
+          #error Assembler CPP flag not enabled
+          #endif
+          int main(int argc, char **argv) { return 0; }" CPU_SUPPORTS_AVX)
+      cmake_pop_check_state()
+      if(CPU_SUPPORTS_AVX)
+        # FIXME: should be target_link_options for CMake 3.13+
+        target_link_libraries(sdl-build-options INTERFACE "/ARCH:AVX")
+        set(HAVE_AVX TRUE)
+      endif()
+    endif()
   endif()
 endif()
 
+if(NOT HAVE_AVX)
+  set(SDL_DISABLE_AVX 1)
+endif()
+
+if(NOT HAVE_MMX)
+  set(SDL_DISABLE_MMX 1)
+endif()
+
+if(NOT HAVE_SSE)
+  set(SDL_DISABLE_SSE 1)
+endif()
+
+if(NOT HAVE_SSE2)
+  set(SDL_DISABLE_SSE2 1)
+endif()
+
+if(NOT HAVE_SSE3)
+  set(SDL_DISABLE_SSE3 1)
+endif()
+
 # TODO: Can't deactivate on FreeBSD? w/o LIBC, SDL_stdinc.h can't define
 # anything.
 if(SDL_LIBC)

include/SDL3/SDL_intrin.h

@@ -101,21 +101,20 @@ _m_prefetch(void *__P)
 #include <lasxintrin.h>
 #define __LASX__
 #endif
-#if defined(HAVE_IMMINTRIN_H) && !defined(SDL_DISABLE_IMMINTRIN_H)
+#if defined(__AVX__) && !defined(SDL_DISABLE_AVX)
 #include <immintrin.h>
-#else
-#if defined(__MMX__) && !defined(SDL_DISABLE_MMINTRIN_H)
+#endif
+#if defined(__MMX__) && !defined(SDL_DISABLE_MMX)
 #include <mmintrin.h>
 #endif
-#if defined(__SSE__) && !defined(SDL_DISABLE_XMMINTRIN_H)
+#if defined(__SSE__) && !defined(SDL_DISABLE_SSE)
 #include <xmmintrin.h>
 #endif
-#if defined(__SSE2__) && !defined(SDL_DISABLE_EMMINTRIN_H)
+#if defined(__SSE2__) && !defined(SDL_DISABLE_SSE2)
 #include <emmintrin.h>
 #endif
-#if defined(__SSE3__) && !defined(SDL_DISABLE_PMMINTRIN_H)
+#if defined(__SSE3__) && !defined(SDL_DISABLE_SSE3)
 #include <pmmintrin.h>
 #endif
-#endif /* HAVE_IMMINTRIN_H */
 
 #endif /* SDL_intrin_h_ */

include/build_config/SDL_build_config.h.cmake

@@ -226,18 +226,6 @@
 #cmakedefine HAVE_LIBUSB 1
 #cmakedefine HAVE_O_CLOEXEC 1
 
-/* Apple platforms might be building universal binaries, where Intel builds
-   can use immintrin.h but other architectures can't. */
-#ifdef __APPLE__
-#  if defined(__has_include) && (defined(__i386__) || defined(__x86_64))
-#    if __has_include(<immintrin.h>)
-#       define HAVE_IMMINTRIN_H 1
-#    endif
-#  endif
-#else  /* non-Apple platforms can use the normal CMake check for this. */
-#cmakedefine HAVE_IMMINTRIN_H 1
-#endif
-
 #cmakedefine HAVE_LIBUDEV_H 1
 #cmakedefine HAVE_LIBSAMPLERATE_H 1
 #cmakedefine HAVE_LIBDECOR_H  1
@@ -597,4 +585,11 @@ typedef unsigned int uintptr_t;
 #endif /* Visual Studio 2008 */
 #endif /* !_STDINT_H_ && !HAVE_STDINT_H */
 
+/* Configure use of intrinsics */
+
+#cmakedefine SDL_DISABLE_SSE 1
+#cmakedefine SDL_DISABLE_SSE2 1
+#cmakedefine SDL_DISABLE_SSE3 1
+#cmakedefine SDL_DISABLE_AVX 1
+
 #endif /* SDL_build_config_h_ */

include/build_config/SDL_build_config_macos.h

@@ -138,8 +138,8 @@
 #define HAVE_SYSCTLBYNAME 1
 
 #if defined(__has_include) && (defined(__i386__) || defined(__x86_64))
-# if __has_include(<immintrin.h>)
-#   define HAVE_IMMINTRIN_H 1
+# if !__has_include(<immintrin.h>)
+#   define SDL_DISABLE_AVX 1
 # endif
 #endif
 

include/build_config/SDL_build_config_windows.h

@@ -103,11 +103,12 @@ typedef unsigned int uintptr_t;
 #define HAVE_TPCSHRD_H 1
 #define HAVE_SENSORSAPI_H 1
 #if (defined(_M_IX86) || defined(_M_X64) || defined(_M_AMD64)) && (defined(_MSC_VER) && _MSC_VER >= 1600)
-#define HAVE_IMMINTRIN_H 1
 #elif defined(__has_include) && (defined(__i386__) || defined(__x86_64))
-# if __has_include(<immintrin.h>)
-#   define HAVE_IMMINTRIN_H 1
+# if !__has_include(<immintrin.h>)
+#   define SDL_DISABLE_AVX 1
 # endif
+#else
+# define SDL_DISABLE_AVX 1
 #endif
 
 /* This is disabled by default to avoid C runtime dependencies and manifest requirements */

include/build_config/SDL_build_config_wingdk.h

@@ -48,11 +48,12 @@
 #define HAVE_TPCSHRD_H 1
 #define HAVE_SENSORSAPI_H 1
 #if (defined(_M_IX86) || defined(_M_X64) || defined(_M_AMD64)) && (defined(_MSC_VER) && _MSC_VER >= 1600)
-#define HAVE_IMMINTRIN_H 1
 #elif defined(__has_include) && (defined(__i386__) || defined(__x86_64))
-# if __has_include(<immintrin.h>)
-#   define HAVE_IMMINTRIN_H 1
+# if !__has_include(<immintrin.h>)
+#  define SDL_DISABLE_AVX 1
 # endif
+#else
+# define SDL_DISABLE_AVX 1
 #endif
 
 /* This is disabled by default to avoid C runtime dependencies and manifest requirements */

include/build_config/SDL_build_config_xbox.h

@@ -48,11 +48,12 @@
 /*#define HAVE_TPCSHRD_H  1*/
 /*#define HAVE_SENSORSAPI_H 1*/
 #if (defined(_M_IX86) || defined(_M_X64) || defined(_M_AMD64)) && (defined(_MSC_VER) && _MSC_VER >= 1600)
-#define HAVE_IMMINTRIN_H 1
 #elif defined(__has_include) && (defined(__i386__) || defined(__x86_64))
-# if __has_include(<immintrin.h>)
-#   define HAVE_IMMINTRIN_H 1
+# if !__has_include(<immintrin.h>)
+#  define SDL_DISABLE_AVX 1
 # endif
+#else
+# define SDL_DISABLE_AVX 1
 #endif
 
 /* This is disabled by default to avoid C runtime dependencies and manifest requirements */

src/SDL_internal.h

@@ -185,6 +185,46 @@
 
 #include <SDL3/SDL.h>
 #include <SDL3/SDL_intrin.h>
+
+
+#ifdef __ARM_NEON
+#define HAVE_NEON_INTRINSICS 1
+#endif
+
+#if defined(__MMX__) && !defined(SDL_DISABLE_MMX)
+#define HAVE_MMX_INTRINSICS 1
+#endif
+
+#if defined(__SSE__) && !defined(SDL_DISABLE_SSE)
+#define HAVE_SSE_INTRINSICS 1
+#endif
+
+#if defined(__SSE2__) && !defined(SDL_DISABLE_SSE2)
+#define HAVE_SSE2_INTRINSICS 1
+#endif
+
+#if defined(__SSE3__) && !defined(SDL_DISABLE_SSE3)
+#define HAVE_SSE3_INTRINSICS 1
+#endif
+
+#if defined(__AVX__) && !defined(SDL_DISABLE_AVX)
+#define HAVE_AVX_INTRINSICS 1
+#endif
+
+#if defined __clang__
+#if (!__has_attribute(target))
+#undef HAVE_AVX_INTRINSICS
+#endif
+#if (defined(_MSC_VER) || defined(__SCE__)) && !defined(__AVX__)
+#undef HAVE_AVX_INTRINSICS
+#endif
+#elif defined __GNUC__
+#if (__GNUC__ < 4) || (__GNUC__ == 4 && __GNUC_MINOR__ < 9)
+#undef HAVE_AVX_INTRINSICS
+#endif
+#endif
+
+
 #define SDL_MAIN_NOIMPL /* don't drag in header-only implementation of SDL_main */
 #include <SDL3/SDL_main.h>
 

src/audio/SDL_audiocvt.c

@@ -29,35 +29,6 @@
 
 #define DEBUG_AUDIOSTREAM 0
 
-#ifdef __ARM_NEON
-#define HAVE_NEON_INTRINSICS 1
-#endif
-
-#ifdef __SSE__
-#define HAVE_SSE_INTRINSICS 1
-#endif
-
-#ifdef __SSE3__
-#define HAVE_SSE3_INTRINSICS 1
-#endif
-
-#if defined(HAVE_IMMINTRIN_H) && !defined(SDL_DISABLE_IMMINTRIN_H)
-#define HAVE_AVX_INTRINSICS 1
-#endif
-#if defined __clang__
-#if (!__has_attribute(target))
-#undef HAVE_AVX_INTRINSICS
-#endif
-#if (defined(_MSC_VER) || defined(__SCE__)) && !defined(__AVX__)
-#undef HAVE_AVX_INTRINSICS
-#endif
-#elif defined __GNUC__
-#if (__GNUC__ < 4) || (__GNUC__ == 4 && __GNUC_MINOR__ < 9)
-#undef HAVE_AVX_INTRINSICS
-#endif
-#endif
-
-
 /**
  * Initialize an SDL_AudioCVT structure for conversion.
  *

src/audio/SDL_audiotypecvt.c

@@ -27,10 +27,6 @@
 #define HAVE_NEON_INTRINSICS 1
 #endif
 
-#ifdef __SSE2__
-#define HAVE_SSE2_INTRINSICS 1
-#endif
-
 #if defined(__x86_64__) && HAVE_SSE2_INTRINSICS
 #define NEED_SCALAR_CONVERTER_FALLBACKS 0 /* x86_64 guarantees SSE2. */
 #elif __MACOS__ && HAVE_SSE2_INTRINSICS