Cycles: add better specializations for SSE shuffle function and few more wrappers.

1 files changed, 16 insertions, 1 deletions

diff --git a/intern/cycles/util/util_sseb.h b/intern/cycles/util/util_sseb.h
index ea38e53b59f..6e669701f3b 100644
--- a/intern/cycles/util/util_sseb.h
+++ b/intern/cycles/util/util_sseb.h
@@ -119,14 +119,29 @@ template<size_t i0, size_t i1, size_t i2, size_t i3> __forceinline const sseb sh
 	return _mm_shuffle_epi32(a, _MM_SHUFFLE(i3, i2, i1, i0));
 }
 
+template<> __forceinline const sseb shuffle<0, 1, 0, 1>( const sseb& a ) {
+	return _mm_movelh_ps(a, a);
+}
+
+template<> __forceinline const sseb shuffle<2, 3, 2, 3>( const sseb& a ) {
+	return _mm_movehl_ps(a, a);
+}
+
 template<size_t i0, size_t i1, size_t i2, size_t i3> __forceinline const sseb shuffle( const sseb& a, const sseb& b ) {
 	return _mm_shuffle_ps(a, b, _MM_SHUFFLE(i3, i2, i1, i0));
 }
 
+template<> __forceinline const sseb shuffle<0, 1, 0, 1>( const sseb& a, const sseb& b ) {
+	return _mm_movelh_ps(a, b);
+}
+
+template<> __forceinline const sseb shuffle<2, 3, 2, 3>( const sseb& a, const sseb& b ) {
+	return _mm_movehl_ps(b, a);
+}
+
 #if defined(__KERNEL_SSE3__)
 template<> __forceinline const sseb shuffle<0, 0, 2, 2>( const sseb& a ) { return _mm_moveldup_ps(a); }
 template<> __forceinline const sseb shuffle<1, 1, 3, 3>( const sseb& a ) { return _mm_movehdup_ps(a); }
-template<> __forceinline const sseb shuffle<0, 1, 0, 1>( const sseb& a ) { return _mm_castpd_ps(_mm_movedup_pd (a)); }
 #endif
 
 #if defined(__KERNEL_SSE41__)