From 5756d8b8d4deb1319e1d6360bca868f1946d7a05 Mon Sep 17 00:00:00 2001
From: Sergey Sharybin <sergey@blender.org>
Date: Thu, 27 Apr 2023 11:05:22 +0200
Subject: [PATCH 1/2] Cycles: Support newer version of sse2neon

Since the version v1.5.0 of sse2neon the functionality for denormals
flushing is implemented in the library. This commit makes it so the
_MM_SET_FLUSH_ZERO_MODE and _MM_SET_DENORMALS_ZERO_MODE are used from
the ss2neon if available.

This solves macro re-definition when a newer sse2neon is used.

The change is implemented in a way that both current and new sse2neon
are supported.
---
 intern/cycles/util/simd.h | 20 ++++++++++++++------
 1 file changed, 14 insertions(+), 6 deletions(-)

diff --git a/intern/cycles/util/simd.h b/intern/cycles/util/simd.h
index 03783abd20f..7da0a0f8390 100644
--- a/intern/cycles/util/simd.h
+++ b/intern/cycles/util/simd.h
@@ -33,11 +33,19 @@
     _MM_SET_FLUSH_ZERO_MODE(_MM_FLUSH_ZERO_ON); \
     _MM_SET_DENORMALS_ZERO_MODE(_MM_DENORMALS_ZERO_ON);
 #elif defined(__aarch64__) || defined(_M_ARM64)
-#  define _MM_FLUSH_ZERO_ON 24
-#  define __get_fpcr(__fpcr) __asm__ __volatile__("mrs %0,fpcr" : "=r"(__fpcr))
-#  define __set_fpcr(__fpcr) __asm__ __volatile__("msr fpcr,%0" : : "ri"(__fpcr))
-#  define SIMD_SET_FLUSH_TO_ZERO set_fz(_MM_FLUSH_ZERO_ON);
-#  define SIMD_GET_FLUSH_TO_ZERO get_fz(_MM_FLUSH_ZERO_ON)
+/* The get/set denormals to zero was implemented in sse2neon v1.5.0.
+ * Keep the compatibility code until the minimum library version is increased. */
+#  if defined(_MM_SET_FLUSH_ZERO_MODE)
+#    define SIMD_SET_FLUSH_TO_ZERO \
+      _MM_SET_FLUSH_ZERO_MODE(_MM_FLUSH_ZERO_ON); \
+      _MM_SET_DENORMALS_ZERO_MODE(_MM_DENORMALS_ZERO_ON);
+#  else
+#    define _MM_FLUSH_ZERO_ON 24
+#    define __get_fpcr(__fpcr) __asm__ __volatile__("mrs %0,fpcr" : "=r"(__fpcr))
+#    define __set_fpcr(__fpcr) __asm__ __volatile__("msr fpcr,%0" : : "ri"(__fpcr))
+#    define SIMD_SET_FLUSH_TO_ZERO set_fz(_MM_FLUSH_ZERO_ON);
+#    define SIMD_GET_FLUSH_TO_ZERO get_fz(_MM_FLUSH_ZERO_ON)
+#  endif
 #else
 #  define SIMD_SET_FLUSH_TO_ZERO
 #endif
@@ -111,7 +119,7 @@ static struct StepTy {
 } step ccl_attr_maybe_unused;
 
 #endif
-#if defined(__aarch64__) || defined(_M_ARM64)
+#if (defined(__aarch64__) || defined(_M_ARM64)) && !defined(_MM_SET_FLUSH_ZERO_MODE)
 __forceinline int set_fz(uint32_t flag)
 {
   uint64_t old_fpcr, new_fpcr;
-- 
2.30.2


From abade6ea69a83a852bdcd949e97b411b3dac35f2 Mon Sep 17 00:00:00 2001
From: Sergey Sharybin <sergey@blender.org>
Date: Thu, 27 Apr 2023 11:12:25 +0200
Subject: [PATCH 2/2] Deps: Bump version of sse2neon

This brings the following improvements:

- Implementation of _MM_SET_FLUSH_ZERO_MODE and _MM_SET_DENORMALS_ZERO_MODE
- Implementation of _mm_round_ss

Does not seem to be any performance impact with just this upgrade.
---
 build_files/build_environment/cmake/versions.cmake | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/build_files/build_environment/cmake/versions.cmake b/build_files/build_environment/cmake/versions.cmake
index d32ac42930c..34565b15cf5 100644
--- a/build_files/build_environment/cmake/versions.cmake
+++ b/build_files/build_environment/cmake/versions.cmake
@@ -579,9 +579,9 @@ set(ZSTD_HASH_TYPE SHA256)
 set(ZSTD_FILE zstd-${ZSTD_VERSION}.tar.gz)
 set(ZSTD_CPE "cpe:2.3:a:facebook:zstandard:${ZSTD_VERSION}:*:*:*:*:*:*:*")
 
-set(SSE2NEON_VERSION fe5ff00bb8d19b327714a3c290f3e2ce81ba3525)
+set(SSE2NEON_VERSION v1.6.0)
 set(SSE2NEON_URI https://github.com/DLTcollab/sse2neon/archive/${SSE2NEON_VERSION}.tar.gz)
-set(SSE2NEON_HASH 0780253525d299c31775ef95853698d03db9c7739942af8570000f4a25a5d605)
+set(SSE2NEON_HASH 06f4693219deccb91b457135d836fc514a1c0a57e9fa66b143982901d2d19677)
 set(SSE2NEON_HASH_TYPE SHA256)
 set(SSE2NEON_FILE sse2neon-${SSE2NEON_VERSION}.tar.gz)
 
-- 
2.30.2