1 files changed, 35 insertions, 0 deletions
diff --git a/intern/cycles/util/types_float3.h b/intern/cycles/util/types_float3.h
index f990367e7b8..cafcfebf526 100644
--- a/intern/cycles/util/types_float3.h
+++ b/intern/cycles/util/types_float3.h
@@ -55,6 +55,41 @@ ccl_device_inline float3 make_float3(float x, float y, float z);
 ccl_device_inline void print_float3(const char *label, const float3 &a);
 #endif /* __KERNEL_GPU__ */
 
+/* Smaller float3 for storage. For math operations this must be converted to float3, so that on the
+ * CPU SIMD instructions can be used. */
+#if defined(__KERNEL_METAL__)
+/* Metal has native packed_float3. */
+#elif defined(__KERNEL_CUDA__)
+/* CUDA float3 is already packed. */
+typedef float3 packed_float3;
+#else
+/* HIP float3 is not packed (https://github.com/ROCm-Developer-Tools/HIP/issues/706). */
+struct packed_float3 {
+  ccl_device_inline_method packed_float3(){};
+
+  ccl_device_inline_method packed_float3(const float3 &a) : x(a.x), y(a.y), z(a.z)
+  {
+  }
+
+  ccl_device_inline_method operator float3() const
+  {
+    return make_float3(x, y, z);
+  }
+
+  ccl_device_inline_method packed_float3 &operator=(const float3 &a)
+  {
+    x = a.x;
+    y = a.y;
+    z = a.z;
+    return *this;
+  }
+
+  float x, y, z;
+};
+#endif
+
+static_assert(sizeof(packed_float3) == 12, "packed_float3 expected to be exactly 12 bytes");
+
 CCL_NAMESPACE_END
 
 #endif /* __UTIL_TYPES_FLOAT3_H__ */