Shadows/Highlights Sparmpask, speedup using SSE4 for native builds

2016-02-11 00:44:26 +01:00
parent a91d2c3dba
commit 7655fce8f9
2 changed files with 76 additions and 36 deletions
--- a/rtengine/helpersse2.h
+++ b/rtengine/helpersse2.h
@@ -39,6 +39,11 @@ typedef __m128i vint2;
 #define STVFU(x,y) _mm_storeu_ps(&x,y)
 #endif

+#if defined(__x86_64__) && defined(__AVX__)
+#define PERMUTEPS(a,mask) _mm_permute_ps(a,mask)
+#else
+#define PERMUTEPS(a,mask) _mm_shuffle_ps(a,a,mask)
+#endif

 static INLINE vfloat LC2VFU(float &a)
 {