Fix Apple Universal 2 (arm64/x86_64) builds

* In these types of builds we want arm64 and x86_64 (with AVX optimisations).
  However the way cmake works (with `CMAKE_OSX_ARCHITECTURES="arm64;x86_64"`
  means that we share one OpenEXRConfigInternal.h between both builds. So
  we have to have OPENEXR_IMF_HAVE_GCC_INLINE_ASM_AVX mean "AVX GCC asm is
  available if platform is x86", rather than "AVX GCC asm is available".
  Then we decide on AVX optimisations based on that #define and also the
  platform defines.

Signed-off-by: Harry Mallon <hjmallon@gmail.com>

OpenEXR/IlmImf/ImfDwaCompressorSimd.h

@@ -60,20 +60,6 @@ OPENEXR_IMF_INTERNAL_NAMESPACE_HEADER_ENTER
 #define _SSE_ALIGNMENT_MASK 0x0F
 #define _AVX_ALIGNMENT_MASK 0x1F
 
-//
-// Test if we should enable GCC inline asm paths for AVX
-//
-
-#ifdef OPENEXR_IMF_HAVE_GCC_INLINE_ASM_AVX 
-
-    #define IMF_HAVE_GCC_INLINEASM
-
-    #ifdef __LP64__
-        #define IMF_HAVE_GCC_INLINEASM_64
-    #endif /* __LP64__ */
-
-#endif /* OPENEXR_IMF_HAVE_GCC_INLINE_ASM_AVX */
-
 //
 // A simple 64-element array, aligned properly for SIMD access. 
 //
@@ -467,7 +453,7 @@ convertFloatToHalf64_f16c (unsigned short *dst, float *src)
     // I'll take the asm.
     //
 
-    #if defined IMF_HAVE_GCC_INLINEASM
+    #if defined IMF_HAVE_GCC_INLINEASM_X86
         __asm__
            ("vmovaps       (%0),     %%ymm0         \n"
             "vmovaps   0x20(%0),     %%ymm1         \n"
@@ -506,7 +492,7 @@ convertFloatToHalf64_f16c (unsigned short *dst, float *src)
            );
     #else
         convertFloatToHalf64_scalar (dst, src);
-    #endif /* IMF_HAVE_GCC_INLINEASM */
+    #endif /* IMF_HAVE_GCC_INLINEASM_X86 */
 }
 
 
@@ -683,7 +669,7 @@ fromHalfZigZag_scalar (unsigned short *src, float *dst)
 void 
 fromHalfZigZag_f16c (unsigned short *src, float *dst)
 {
-    #if defined IMF_HAVE_GCC_INLINEASM_64
+    #if defined IMF_HAVE_GCC_INLINEASM_X86_64
         __asm__
 
            /* x3 <- 0                    
@@ -835,7 +821,7 @@ fromHalfZigZag_f16c (unsigned short *src, float *dst)
 
     #else
         fromHalfZigZag_scalar(src, dst);
-    #endif /* defined IMF_HAVE_GCC_INLINEASM_64 */
+    #endif /* defined IMF_HAVE_GCC_INLINEASM_X86_64 */
 }
 
 
@@ -1607,7 +1593,7 @@ template <int zeroedRows>
 void
 dctInverse8x8_avx (float *data)
 {
-    #if defined IMF_HAVE_GCC_INLINEASM_64
+    #if defined IMF_HAVE_GCC_INLINEASM_X86_64
 
     /* The column-major version of M1, followed by the 
      * column-major version of M2:
@@ -1735,11 +1721,11 @@ dctInverse8x8_avx (float *data)
         } else {
             assert(false); // Invalid template instance parameter
         }
-    #else  /* IMF_HAVE_GCC_INLINEASM_64 */
+    #else  /* IMF_HAVE_GCC_INLINEASM_X86_64 */
 
         dctInverse8x8_scalar<zeroedRows>(data);
 
-    #endif /*  IMF_HAVE_GCC_INLINEASM_64 */
+    #endif /*  IMF_HAVE_GCC_INLINEASM_X86_64 */
 }
 
 

OpenEXR/IlmImf/ImfSystemSpecific.cpp

@@ -63,7 +63,7 @@ namespace {
 #endif // IMF_HAVE_SSE2 && __GNUC__
 
 
-#ifdef OPENEXR_IMF_HAVE_GCC_INLINE_ASM_AVX
+#ifdef IMF_HAVE_GCC_INLINEASM_X86
 
     void xgetbv(int n, int &eax, int &edx)
     {
@@ -74,14 +74,14 @@ namespace {
             : /* Clobber */);
     }
 
-#else //  OPENEXR_IMF_HAVE_GCC_INLINE_ASM_AVX
+#else //  IMF_HAVE_GCC_INLINEASM_X86
 
     void xgetbv(int n, int &eax, int &edx)
     {
         eax = edx = 0;
     }
 
-#endif //  OPENEXR_IMF_HAVE_GCC_INLINE_ASM_AVX
+#endif //  IMF_HAVE_GCC_INLINEASM_X86
 
 } // namespace 
 

OpenEXR/IlmImf/ImfSystemSpecific.h

@@ -39,9 +39,24 @@
 #include "ImfSimd.h"
 #include <stdlib.h>
 #include "ImfExport.h"
+#include "OpenEXRConfig.h"
+#include "OpenEXRConfigInternal.h"
 
 OPENEXR_IMF_INTERNAL_NAMESPACE_HEADER_ENTER
 
+//
+// Test if we should enable GCC inline asm paths for AVX
+//
+
+#if defined(OPENEXR_IMF_HAVE_GCC_INLINE_ASM_AVX) && (defined(_M_X64) || defined(__x86_64__))
+
+    #define IMF_HAVE_GCC_INLINEASM_X86
+
+    #ifdef __LP64__
+        #define IMF_HAVE_GCC_INLINEASM_X86_64
+    #endif /* __LP64__ */
+
+#endif /* OPENEXR_IMF_HAVE_GCC_INLINE_ASM_AVX */
 
 static unsigned long  systemEndianCheckValue   = 0x12345678;
 static unsigned long* systemEndianCheckPointer = &systemEndianCheckValue;

OpenEXR/config/CMakeLists.txt

@@ -20,12 +20,13 @@ int main()
     sysconf(_SC_NPROCESSORS_ONLN);
 }
 " OPENEXR_IMF_HAVE_SYSCONF_NPROCESSORS_ONLN
-)    
+)
 
 check_cxx_source_compiles(
 "
 int main()
 {
+#if defined(_M_X64) || defined(__x86_64__)
 #if defined(__SSE2__)
     int n   = 0;
     int eax = 0;
@@ -37,6 +38,7 @@ int main()
 #else
 #    error No SSE support enabled to query AVX support
 #endif
+#endif
 }
 " OPENEXR_IMF_HAVE_GCC_INLINE_ASM_AVX
 )