diff --git a/intern/cycles/util/math_float3.h b/intern/cycles/util/math_float3.h
index 63872e19c02..a1231977831 100644
--- a/intern/cycles/util/math_float3.h
+++ b/intern/cycles/util/math_float3.h
@@ -158,7 +158,7 @@ ccl_device_inline float3 operator/=(float3 &a, float f)
   return a = a * invf;
 }
 
-#  if !(defined(__KERNEL_METAL__) || defined(__KERNEL_CUDA__))
+#  if !(defined(__KERNEL_METAL__) || defined(__KERNEL_CUDA__) || defined(__KERNEL_HIP__))
 ccl_device_inline packed_float3 operator*=(packed_float3 &a, const float3 b)
 {
   a = float3(a) * b;
diff --git a/intern/cycles/util/types_float3.h b/intern/cycles/util/types_float3.h
index 303ffe81c00..47e17959e74 100644
--- a/intern/cycles/util/types_float3.h
+++ b/intern/cycles/util/types_float3.h
@@ -63,11 +63,10 @@ ccl_device_inline void print_float3(ccl_private const char *label, const float3
  * CPU SIMD instructions can be used. */
 #if defined(__KERNEL_METAL__)
 /* Metal has native packed_float3. */
-#elif defined(__KERNEL_CUDA__)
-/* CUDA float3 is already packed. */
+#elif defined(__KERNEL_CUDA__) || defined(__KERNEL_HIP__)
+/* CUDA and HIP float3 are 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(){};