From aeb103fb501ee54112c66aba184d5fe6e9918eda Mon Sep 17 00:00:00 2001
From: Xavier Hallade <me@ph0b.com>
Date: Mon, 8 Sep 2025 09:22:32 +0200
Subject: [PATCH] Cycles: Pack uint3/int3 structs for oneAPI

This recently changed after a fix in 28f93d544301b3dd8201350dc0cdd70bfdd3f29c
but we get better performance by ensuring int3 is packed instead.

Packing int3 currently gives a 7% speedup when rendering wdas_cloud on
Intel Arc B580.

Pull Request: https://projects.blender.org/blender/blender/pulls/145593
---
 intern/cycles/util/types_int3.h  | 9 +++++++--
 intern/cycles/util/types_uint3.h | 4 ++--
 2 files changed, 9 insertions(+), 4 deletions(-)

diff --git a/intern/cycles/util/types_int3.h b/intern/cycles/util/types_int3.h
index d170c066a0b..8e8af27b55e 100644
--- a/intern/cycles/util/types_int3.h
+++ b/intern/cycles/util/types_int3.h
@@ -9,7 +9,12 @@
 CCL_NAMESPACE_BEGIN
 
 #ifndef __KERNEL_NATIVE_VECTOR_TYPES__
+#  ifdef __KERNEL_ONEAPI__
+/* Keep structure packed for oneAPI. */
+struct int3
+#  else
 struct ccl_try_align(16) int3
+#  endif
 {
 #  ifdef __KERNEL_GPU__
   /* Compact structure on the GPU. */
@@ -98,8 +103,8 @@ ccl_device_inline void print_int3(const ccl_private char *label, const int3 a)
 
 #if defined(__KERNEL_METAL__)
 /* Metal has native packed_int3. */
-#elif defined(__KERNEL_CUDA__)
-/* CUDA is already packed. */
+#elif defined(__KERNEL_CUDA__) || defined(__KERNEL_ONEAPI__)
+/* CUDA/oneAPI int3 is already packed. */
 typedef int3 packed_int3;
 #else
 /* HIP int3 is not packed (https://github.com/ROCm-Developer-Tools/HIP/issues/706). */
diff --git a/intern/cycles/util/types_uint3.h b/intern/cycles/util/types_uint3.h
index 7b9dda8022b..e107a0af38b 100644
--- a/intern/cycles/util/types_uint3.h
+++ b/intern/cycles/util/types_uint3.h
@@ -36,8 +36,8 @@ ccl_device_inline uint3 make_uint3(const uint x, const uint y, uint z)
 
 #if defined(__KERNEL_METAL__)
 /* Metal has native packed_float3. */
-#elif defined(__KERNEL_CUDA__)
-/* CUDA uint3 is already packed. */
+#elif defined(__KERNEL_CUDA__) || defined(__KERNEL_ONEAPI__)
+/* CUDA/oneAPI uint3 is already packed. */
 using packed_uint3 = uint3;
 #else
 /* HIP uint3 is not packed (https://github.com/ROCm-Developer-Tools/HIP/issues/706). */