diff --git a/CMakeLists.txt b/CMakeLists.txt
index 7026665..39c8ab0 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -978,6 +978,8 @@ elseif(WIN32)
 		# MSVC11 and MSVC12 need _ALLOW_KEYWORD_MACROS to build
 		if(MSVC11 OR MSVC12)
 			add_definitions(/D_ALLOW_KEYWORD_MACROS)
+			#TODO: Eigen cannot detect SSE3+ instructions with MSVC
+			add_definitions(/DEIGEN_VECTORIZE_SSE3 /DEIGEN_VECTORIZE_SSSE3 /DEIGEN_VECTORIZE_SSE4_1 /DEIGEN_VECTORIZE_SSE4_2)
 		endif()
 
 		set(CMAKE_CXX_FLAGS "/nologo /J /Gd /EHsc /MP" CACHE STRING "MSVC MT C++ flags " FORCE)
diff --git a/intern/cycles/util/util_math.h b/intern/cycles/util/util_math.h
index 2e73639..bc3b50b 100644
--- a/intern/cycles/util/util_math.h
+++ b/intern/cycles/util/util_math.h
@@ -31,7 +31,6 @@
 #include <float.h>
 #include <math.h>
 #include <stdio.h>
-
 #endif
 
 #include "util_types.h"
@@ -655,8 +654,8 @@ template<> __forceinline const float4 shuffle<0, 1, 0, 1>(const float4& b)
 ccl_device_inline float4 operator-(const float4& a)
 {
 #ifdef __KERNEL_SSE__
-	__m128 mask = _mm_castsi128_ps(_mm_set1_epi32(0x80000000));
-	return _mm_xor_ps(a.m128, mask);
+	const __m128 mask = _mm_castsi128_ps(_mm_setr_epi32(0x80000000, 0x80000000, 0x80000000, 0x80000000));
+	return _mm_xor_ps(a, mask);
 #else
 	return make_float4(-a.x, -a.y, -a.z, -a.w);
 #endif
@@ -665,7 +664,7 @@ ccl_device_inline float4 operator-(const float4& a)
 ccl_device_inline float4 operator*(const float4& a, const float4& b)
 {
 #ifdef __KERNEL_SSE__
-	return _mm_mul_ps(a.m128, b.m128);
+	return _mm_mul_ps(a, b);
 #else
 	return make_float4(a.x*b.x, a.y*b.y, a.z*b.z, a.w*b.w);
 #endif
@@ -688,13 +687,23 @@ ccl_device_inline float4 operator*(float f, const float4& a)
 ccl_device_inline float4 rcp(const float4& a)
 {
 #ifdef __KERNEL_SSE__
-	float4 r = _mm_rcp_ps(a.m128);
+	const float4 r = _mm_rcp_ps(a);
 	return _mm_sub_ps(_mm_add_ps(r, r), _mm_mul_ps(_mm_mul_ps(r, r), a));
 #else
 	return make_float4(1.0f/a.x, 1.0f/a.y, 1.0f/a.z, 1.0f/a.w);
 #endif
 }
 
+ccl_device_inline float4 abs(const float4& a)
+{
+#ifdef __KERNEL_SSE__
+	const float4  mask = _mm_castsi128_ps(_mm_setr_epi32(0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF));
+	return _mm_and_ps(a,mask);
+#else
+	return make_float4(fabs(a.x)), fabs(a.y), fabs(a.z), fabs(a.w));
+#endif
+}
+
 ccl_device_inline float4 operator/(const float4& a, float f)
 {
 	return a * (1.0f/f);
@@ -703,7 +712,8 @@ ccl_device_inline float4 operator/(const float4& a, float f)
 ccl_device_inline float4 operator/(const float4& a, const float4& b)
 {
 #ifdef __KERNEL_SSE__
-	return a * rcp(b);
+	/* return a * rcp(b); */
+	return _mm_div_ps(a, b);
 #else
 	return make_float4(a.x/b.x, a.y/b.y, a.z/b.z, a.w/b.w);
 #endif
@@ -713,7 +723,7 @@ ccl_device_inline float4 operator/(const float4& a, const float4& b)
 ccl_device_inline float4 operator+(const float4& a, const float4& b)
 {
 #ifdef __KERNEL_SSE__
-	return _mm_add_ps(a.m128, b.m128);
+	return _mm_add_ps(a, b);
 #else
 	return make_float4(a.x+b.x, a.y+b.y, a.z+b.z, a.w+b.w);
 #endif
@@ -722,7 +732,7 @@ ccl_device_inline float4 operator+(const float4& a, const float4& b)
 ccl_device_inline float4 operator-(const float4& a, const float4& b)
 {
 #ifdef __KERNEL_SSE__
-	return _mm_sub_ps(a.m128, b.m128);
+	return _mm_sub_ps(a, b);
 #else
 	return make_float4(a.x-b.x, a.y-b.y, a.z-b.z, a.w-b.w);
 #endif
@@ -782,9 +792,14 @@ ccl_device_inline bool operator==(const float4 a, const float4 b)
 ccl_device_inline float4 cross(const float4& a, const float4& b)
 {
 #ifdef __KERNEL_SSE__
-	return (shuffle<1,2,0,0>(a)*shuffle<2,0,1,0>(b)) - (shuffle<2,0,1,0>(a)*shuffle<1,2,0,0>(b));
+	/* return (shuffle<1,2,0,0>(a)*shuffle<2,0,1,0>(b)) - (shuffle<2,0,1,0>(a)*shuffle<1,2,0,0>(b));*/
+	/* Same thing, expanded */
+	return _mm_sub_ps(_mm_mul_ps(_mm_shuffle_ps(a, a, 9), _mm_shuffle_ps(b, b, 18)),
+		_mm_mul_ps(_mm_shuffle_ps(b, b, 9), _mm_shuffle_ps(a, a, 18)));
+
+
 #else
-	return make_float4(a.y*b.z - a.z*b.y, a.z*b.x - a.x*b.z, a.x*b.y - a.y*b.x, 0.0f);
+	return make_float4(a.y*b.z - a.z*b.y, a.z*b.a - a.x*b.z, a.x*b.y - a.y*b.x, 0.0f);
 #endif
 }
 
@@ -797,14 +812,40 @@ ccl_device_inline bool is_zero(const float4& a)
 #endif
 }
 
+ccl_device_inline float float4x(const float4& a)
+{
+#ifdef __KERNEL_SSE__
+#ifdef __KERNEL_SSE3__
+#if defined(_MSC_VER) && defined(_WIN64) && !defined(__INTEL_COMPILER)
+	return a.m128.m128_f32[0];
+#elif defined(_MSC_VER) && !defined(__INTEL_COMPILER)
+	/* The temporary variable fixes an internal a possible compilation error in vs <= 2008 and a wrong-result bug in vs 2010 */
+	float x = _mm_cvtss_f32(a); 
+	return x;
+#else 	/* Other compiler*/
+	return _mm_cvtss_f32(a);
+#endif /* compilers */
+#endif /* __KERNEL_SSE3__ */
+#else
+	return a.x;
+#endif /* _KERNEL_SSE__ */
+}
+
 ccl_device_inline float reduce_add(const float4& a)
 {
 #ifdef __KERNEL_SSE__
+#ifdef __KERNEL_SSE3__
+	float4 h = _mm_hadd_ps(a, a);
+	return float4x(_mm_hadd_ps(h, h));
+#elif defined(__KERNEL_SSE2__)
+	float4 h = _mm_add_ps(a, _mm_movehl_ps(a, a));
+	return float4x(_mm_add_ss(h, _mm_shuffle_ps(h, h, 1)));
+	/* Original: 
 	float4 h = shuffle<1,0,3,2>(a) + a;
-	return _mm_cvtss_f32(shuffle<2,3,0,1>(h) + h); /* todo: efficiency? */
-#else
+	return _mm_cvtss_f32(shuffle<2,3,0,1>(h) + h); */ /* todo: efficiency? */
+#endif /* __KERNEL_SSE3__ */
 	return ((a.x + a.y) + (a.z + a.w));
-#endif
+#endif /*  __KERNEL_SSE__ */
 }
 
 ccl_device_inline float average(const float4& a)
diff --git a/intern/cycles/util/util_optimization.h b/intern/cycles/util/util_optimization.h
index f901513..763f733 100644
--- a/intern/cycles/util/util_optimization.h
+++ b/intern/cycles/util/util_optimization.h
@@ -78,7 +78,7 @@
  * This is disabled code for an experiment to use SSE types globally for types
  * such as float3 and float4. Currently this gives an overall slowdown. */
 
-#if 0
+#if 1
 #define __KERNEL_SSE__
 #ifndef __KERNEL_SSE2__
 #define __KERNEL_SSE2__
diff --git a/intern/cycles/util/util_types.h b/intern/cycles/util/util_types.h
index d5f67ab..b79106b 100644
--- a/intern/cycles/util/util_types.h
+++ b/intern/cycles/util/util_types.h
@@ -90,6 +90,11 @@ CCL_NAMESPACE_BEGIN
  * Also vector types, named to be compatible with OpenCL builtin types, while
  * working for CUDA and C++ too. */
 
+#ifdef _MSC_VER
+#pragma pack(push,16) /* Must ensure class & union 16-B aligned */
+#endif  /* _MSC_VER */
+
+
 /* Shorter Unsigned Names */
 
 #ifndef __KERNEL_OPENCL__
@@ -450,6 +455,10 @@ ccl_device_inline int4 make_int4(const float3& f)
 
 #endif
 
+#ifdef _MSC_VER
+#pragma pack(pop)
+#endif  /* _MSC_VER */
+
 CCL_NAMESPACE_END
 
 #endif /* __UTIL_TYPES_H__ */