Cycles: optimized SSE BVH traversal now also works with SSE2 CPUs, so all the

way back to Pentium 4, using a slightly less efficient instruction.

Also ensure /Ox is used for Visual Studio for RelWithDebInfo builds.

1 files changed, 46 insertions, 5 deletions

diff --git a/intern/cycles/util/util_types.h b/intern/cycles/util/util_types.h
index d0116186ee8..4fd1e9d8807 100644
--- a/intern/cycles/util/util_types.h
+++ b/intern/cycles/util/util_types.h
@@ -76,8 +76,8 @@
 #define __KERNEL_SSE2__
 #endif
 
-#ifndef __KERNEL_SSSE3__
-#define __KERNEL_SSSE3__
+#ifndef __KERNEL_SSE3__
+#define __KERNEL_SSE3__
 #endif
 
 #ifndef __KERNEL_SSSE3__
@@ -100,7 +100,7 @@
 #include <xmmintrin.h> /* SSE 1 */
 #include <emmintrin.h> /* SSE 2 */
 
-#ifdef __KERNEL_SSSE3__
+#ifdef __KERNEL_SSE3__
 #include <pmmintrin.h> /* SSE 3 */
 #endif
 #ifdef __KERNEL_SSSE3__
@@ -491,15 +491,56 @@ __device_inline int4 make_int4(const float3& f)
 
 #endif
 
-#ifdef __KERNEL_SSSE3__
+#ifdef __KERNEL_SSE2__
 
 /* SSE shuffle utility functions */
 
-__device_inline const __m128 shuffle8(const __m128& a, const __m128i& shuf)
+#ifdef __KERNEL_SSSE3__
+
+/* faster version for SSSE3 */
+typedef __m128i shuffle_swap_t;
+
+__device_inline const shuffle_swap_t shuffle_swap_identity(void)
+{
+	return _mm_set_epi8(15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0);
+}
+
+__device_inline const shuffle_swap_t shuffle_swap_swap(void)
+{
+	return _mm_set_epi8(7, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8);
+}
+
+__device_inline const __m128 shuffle_swap(const __m128& a, const shuffle_swap_t& shuf)
 {
 	return _mm_castsi128_ps(_mm_shuffle_epi8(_mm_castps_si128(a), shuf));
 }
 
+#else
+
+/* somewhat slower version for SSE3 */
+typedef int shuffle_swap_t;
+
+__device_inline const shuffle_swap_t shuffle_swap_identity(void)
+{
+	return 0;
+}
+
+__device_inline const shuffle_swap_t shuffle_swap_swap(void)
+{
+	return 1;
+}
+
+__device_inline const __m128 shuffle_swap(const __m128& a, shuffle_swap_t shuf)
+{
+	/* shuffle value must be a constant, so we need to branch */
+	if(shuf)
+		return _mm_shuffle_ps(a, a, _MM_SHUFFLE(1, 0, 3, 2));
+	else
+		return _mm_shuffle_ps(a, a, _MM_SHUFFLE(3, 2, 1, 0));
+}
+
+#endif
+
 template<size_t i0, size_t i1, size_t i2, size_t i3> __device_inline const __m128 shuffle(const __m128& a, const __m128& b)
 {
 	return _mm_shuffle_ps(a, b, _MM_SHUFFLE(i3, i2, i1, i0));