diff --git a/Ray.cpp b/Ray.cpp
index 1d89b48f..8d283359 100644
--- a/Ray.cpp
+++ b/Ray.cpp
@@ -142,5 +142,6 @@ bool Ray::MatchDeviceNames(const char *name, const char *pattern) {
 
 const char *Ray::Version() { return "v0.3.0-unknown-commit"; }
 
-// Workaround for a strange clang behavior
+// Workaround for identical template instantiations merging causing illegal instruction error with Windows Clang
 template class std::vector<float, Ray::aligned_allocator<float, 64>>;
+template class std::vector<uint16_t>;
\ No newline at end of file
diff --git a/internal/CDFUtils.h b/internal/CDFUtils.h
index 152700a0..94e66303 100644
--- a/internal/CDFUtils.h
+++ b/internal/CDFUtils.h
@@ -6,7 +6,7 @@
 
 namespace Ray {
 // This mostly taken from Cycles source code
-template <typename Func> std::vector<float> CDFEvaluate(const int res, const float from, const float to, Func func) {
+template <typename Func> inline std::vector<float> CDFEvaluate(const int res, const float from, const float to, Func func) {
     const int cdf_count = res + 1;
     const float range = to - from;
 
@@ -30,8 +30,8 @@ template <typename Func> std::vector<float> CDFEvaluate(const int res, const flo
 std::vector<float> CDFInvert(int res, float from, float to, const std::vector<float> &cdf, bool make_symmetric);
 
 template <typename Func>
-std::vector<float> CDFInverted(const int res, const float from, const float to, Func func, const bool make_symmetric) {
-    std::vector<float> cdf = CDFEvaluate(res - 1, from, to, func);
+inline std::vector<float> CDFInverted(const int res, const float from, const float to, Func func, const bool make_symmetric) {
+    const std::vector<float> cdf = CDFEvaluate(res - 1, from, to, func);
     return CDFInvert(res, from, to, cdf, make_symmetric);
 }
 } // namespace Ray
diff --git a/internal/TextureUtilsNEON.cpp b/internal/TextureUtilsNEON.cpp
index e0501a85..5158554d 100644
--- a/internal/TextureUtilsNEON.cpp
+++ b/internal/TextureUtilsNEON.cpp
@@ -301,8 +301,8 @@ void InsetYCoCgBBox_NEON(uint8_t min_color[4], uint8_t max_color[4]) {
     min_col = vcombine_u8(vqmovun_s16(min_col), vqmovun_s16(min_col));
     max_col = vcombine_u8(vqmovun_s16(max_col), vqmovun_s16(max_col));
 
-    vst1q_lane_s32(reinterpret_cast<int32_t *>(min_color), min_col, 0);
-    vst1q_lane_s32(reinterpret_cast<int32_t *>(max_color), max_col, 0);
+    vst1q_lane_u8(min_color, min_col, 0);
+    vst1q_lane_u8(max_color, max_col, 0);
 }
 
 alignas(16) static const int16_t CoCgMask[] = {-1, 0, -1, 0, -1, 0, -1, 0};