From ee258c390821efb41a5f40c95a8fbace25773598 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Aras=20Pranckevi=C4=8Dius?= <aras@nesnausk.org>
Date: Tue, 21 Mar 2023 01:45:06 +0200
Subject: [PATCH] Fix ARMv7 build by making recent ZIP NEON optimizations be
 ARMv8 (aarch64) only (#1366)

Should fix #1365. Recent PR (#1348) added NEON accelerated code paths
for ZIP filtering. But that code uses several instructions that are
ARMv8 (aarch64) only, and thus fail building on 32-bit ARM (armv7)
platforms. Make these optimizations only kick in when building
for 64-bit ARM platforms.

Signed-off-by: Aras Pranckevicius <aras@nesnausk.org>
---
 src/lib/OpenEXR/ImfSimd.h          | 4 ++++
 src/lib/OpenEXR/ImfZip.cpp         | 6 +++---
 src/lib/OpenEXRCore/internal_zip.c | 8 ++++----
 3 files changed, 11 insertions(+), 7 deletions(-)

diff --git a/src/lib/OpenEXR/ImfSimd.h b/src/lib/OpenEXR/ImfSimd.h
index bfdaed7f34..9918c12173 100644
--- a/src/lib/OpenEXR/ImfSimd.h
+++ b/src/lib/OpenEXR/ImfSimd.h
@@ -47,6 +47,10 @@
 #    define IMF_HAVE_NEON
 #endif
 
+#if defined(__aarch64__)
+#    define IMF_HAVE_NEON_AARCH64 1
+#endif
+
 extern "C" {
 #ifdef IMF_HAVE_SSE2
     #include <emmintrin.h>
diff --git a/src/lib/OpenEXR/ImfZip.cpp b/src/lib/OpenEXR/ImfZip.cpp
index 98477ba594..92896eadf1 100644
--- a/src/lib/OpenEXR/ImfZip.cpp
+++ b/src/lib/OpenEXR/ImfZip.cpp
@@ -165,7 +165,7 @@ reconstruct_sse41 (char* buf, size_t outSize)
 
 #endif
 
-#ifdef IMF_HAVE_NEON
+#ifdef IMF_HAVE_NEON_AARCH64
 
 void
 reconstruct_neon (char* buf, size_t outSize)
@@ -265,7 +265,7 @@ interleave_sse2 (const char* source, size_t outSize, char* out)
 
 #endif
 
-#ifdef IMF_HAVE_NEON
+#ifdef IMF_HAVE_NEON_AARCH64
 
 void
 interleave_neon (const char* source, size_t outSize, char* out)
@@ -387,7 +387,7 @@ Zip::initializeFuncs ()
     }
 #endif
 
-#ifdef IMF_HAVE_NEON
+#ifdef IMF_HAVE_NEON_AARCH64
     reconstruct = reconstruct_neon;
     interleave = interleave_neon;
 #endif
diff --git a/src/lib/OpenEXRCore/internal_zip.c b/src/lib/OpenEXRCore/internal_zip.c
index e829e6d7ce..1374cab0d0 100644
--- a/src/lib/OpenEXRCore/internal_zip.c
+++ b/src/lib/OpenEXRCore/internal_zip.c
@@ -24,8 +24,8 @@
 #    define IMF_HAVE_SSE4_1 1
 #    include <smmintrin.h>
 #endif
-#if defined(__ARM_NEON)
-#    define IMF_HAVE_NEON 1
+#if defined(__aarch64__)
+#    define IMF_HAVE_NEON_AARCH64 1
 #    include <arm_neon.h>
 #endif
 
@@ -78,7 +78,7 @@ reconstruct (uint8_t* buf, uint64_t outSize)
         prev      = d;
     }
 }
-#elif defined(IMF_HAVE_NEON)
+#elif defined(IMF_HAVE_NEON_AARCH64)
 static void
 reconstruct (uint8_t* buf, uint64_t outSize)
 {
@@ -174,7 +174,7 @@ interleave (uint8_t* out, const uint8_t* source, uint64_t outSize)
         *(sOut++) = (i % 2 == 0) ? *(t1++) : *(t2++);
 }
 
-#elif defined(IMF_HAVE_NEON)
+#elif defined(IMF_HAVE_NEON_AARCH64)
 static void
 interleave (uint8_t* out, const uint8_t* source, uint64_t outSize)
 {