diff --git a/src/nvtt/icbc.h b/src/nvtt/icbc.h
index 56bf3b7..4abc98f 100644
--- a/src/nvtt/icbc.h
+++ b/src/nvtt/icbc.h
@@ -44,6 +44,11 @@ namespace icbc {
 #include <assert.h>
 #endif
 
+#if ICBC_USE_SIMD
+#include <xmmintrin.h>
+#include <emmintrin.h>
+#endif
+
 namespace icbc {
 
 ///////////////////////////////////////////////////////////////////////////////////////////////////
@@ -223,11 +228,13 @@ inline bool equal(Vector3 a, Vector3 b, float epsilon) {
 
 #if ICBC_USE_SIMD
 
-#include <xmmintrin.h>
-#include <emmintrin.h>
-
 #define SIMD_INLINE inline
-#define SIMD_NATIVE __forceinline
+#if __GNUC__
+// Also need "inline" to avoid "function body can be overwritten at link time" errors.
+#   define SIMD_NATIVE __attribute__((always_inline)) inline
+#else // _MSC_VER
+#   define SIMD_NATIVE __forceinline
+#endif
 
 class SimdVector
 {
@@ -2196,12 +2203,12 @@ bool ClusterFit::fastCompress3(Vector3 * start, Vector3 * end)
 
         for (int c1 = 0; c1 <= count - c0; c1++, i++)
         {
-            const SimdVector constants = SimdVector((const float *)&s_threeElement[i]);
-
-            const SimdVector alpha2_sum = constants.splatX();
-            const SimdVector beta2_sum = constants.splatY();
-            const SimdVector alphabeta_sum = constants.splatZ();
-            const SimdVector factor = constants.splatW();
+            const SimdVector constants = SimdVector((const float *)&s_threeElement[i]);
+
+            const SimdVector alpha2_sum = constants.splatX();
+            const SimdVector beta2_sum = constants.splatY();
+            const SimdVector alphabeta_sum = constants.splatZ();
+            const SimdVector factor = constants.splatW();
 
             const SimdVector alphax_sum = multiplyAdd(x1, half, x0);
             const SimdVector betax_sum = m_xsum - alphax_sum;
@@ -2283,15 +2290,15 @@ bool ClusterFit::fastCompress4(Vector3 * start, Vector3 * end)
 
             for (int c2 = 0; c2 <= 16 - c0 - c1; c2++, i++)
             {
-                const SimdVector constants = SimdVector((const float *)&s_fourElement[i]); 
-
-                const SimdVector alpha2_sum = constants.splatX();
-                const SimdVector beta2_sum = constants.splatY();
-                const SimdVector alphabeta_sum = constants.splatZ();
-                const SimdVector factor = constants.splatW();
-                
-                const SimdVector alphax_sum = multiplyAdd(x2, onethird, multiplyAdd(x1, twothirds, x0));
-                const SimdVector betax_sum = m_xsum - alphax_sum;
+                const SimdVector constants = SimdVector((const float *)&s_fourElement[i]); 
+
+                const SimdVector alpha2_sum = constants.splatX();
+                const SimdVector beta2_sum = constants.splatY();
+                const SimdVector alphabeta_sum = constants.splatZ();
+                const SimdVector factor = constants.splatW();
+                
+                const SimdVector alphax_sum = multiplyAdd(x2, onethird, multiplyAdd(x1, twothirds, x0));
+                const SimdVector betax_sum = m_xsum - alphax_sum;
 
                 SimdVector a = negativeMultiplySubtract(betax_sum, alphabeta_sum, alphax_sum*beta2_sum) * factor;
                 SimdVector b = negativeMultiplySubtract(alphax_sum, alphabeta_sum, betax_sum*alpha2_sum) * factor;
@@ -2560,10 +2567,10 @@ bool ClusterFit::fastCompress3(Vector3 * start, Vector3 * end)
 
         for (uint c1 = 0; c1 <= count - c0; c1++, i++)
         {
-            float const alpha2_sum = s_threeElement[i].alpha2_sum;
-            float const beta2_sum = s_threeElement[i].beta2_sum;
-            float const alphabeta_sum = s_threeElement[i].alphabeta_sum;
-            float const factor = s_threeElement[i].factor;
+            float const alpha2_sum = s_threeElement[i].alpha2_sum;
+            float const beta2_sum = s_threeElement[i].beta2_sum;
+            float const alphabeta_sum = s_threeElement[i].alphabeta_sum;
+            float const factor = s_threeElement[i].factor;
 
             Vector3 const alphax_sum = x0 + x1 * 0.5f;
             Vector3 const betax_sum = m_xsum - alphax_sum;
@@ -2647,11 +2654,11 @@ bool ClusterFit::fastCompress4(Vector3 * start, Vector3 * end)
 
             for (uint c2 = 0; c2 <= count - c0 - c1; c2++, i++)
             {
-                float const alpha2_sum = s_fourElement[i].alpha2_sum;
-                float const beta2_sum = s_fourElement[i].beta2_sum;
-                float const alphabeta_sum = s_fourElement[i].alphabeta_sum;
-                float const factor = s_fourElement[i].factor;
-
+                float const alpha2_sum = s_fourElement[i].alpha2_sum;
+                float const beta2_sum = s_fourElement[i].beta2_sum;
+                float const alphabeta_sum = s_fourElement[i].alphabeta_sum;
+                float const factor = s_fourElement[i].factor;
+
                 Vector3 const alphax_sum = x0 + x1 * (2.0f / 3.0f) + x2 * (1.0f / 3.0f);
                 Vector3 const betax_sum = m_xsum - alphax_sum;