From a16be41a8d76a677cd3e275e831c26d5b18cfc3b Mon Sep 17 00:00:00 2001
From: Csaba Kiraly <csaba.kiraly@gmail.com>
Date: Tue, 17 May 2022 13:00:08 +0200
Subject: [PATCH] add support for Mac M1, maybe other arm64+neon as well

There was already some NEON support, through a separate code
path. This version relies on the sse2neon library to add
Neon support

Signed-off-by: Csaba Kiraly <csaba.kiraly@gmail.com>
---
 .gitmodules       |  3 +++
 LeopardCommon.cpp |  2 ++
 LeopardCommon.h   | 11 +++++++++++
 sse2neon          |  1 +
 4 files changed, 17 insertions(+)
 create mode 100644 .gitmodules
 create mode 160000 sse2neon

diff --git a/.gitmodules b/.gitmodules
new file mode 100644
index 0000000..a11d0ff
--- /dev/null
+++ b/.gitmodules
@@ -0,0 +1,3 @@
+[submodule "sse2neon"]
+	path = sse2neon
+	url = https://github.com/DLTcollab/sse2neon
diff --git a/LeopardCommon.cpp b/LeopardCommon.cpp
index 76d87a3..9632915 100644
--- a/LeopardCommon.cpp
+++ b/LeopardCommon.cpp
@@ -107,6 +107,8 @@ static void _cpuid(unsigned int cpu_info[4U], const unsigned int cpu_info_type)
 #endif
 }
 
+#elif defined(LEO_USE_SSE2NEON)
+bool CpuHasSSSE3 = true;
 #endif // defined(LEO_TARGET_MOBILE)
 
 
diff --git a/LeopardCommon.h b/LeopardCommon.h
index 467f829..45bac3d 100644
--- a/LeopardCommon.h
+++ b/LeopardCommon.h
@@ -186,6 +186,11 @@
 // Unroll inner loops 4 times
 #define LEO_USE_VECTOR4_OPT
 
+// MacOS M1
+#if defined(__aarch64__)
+  #define LEO_USE_SSE2NEON
+  #define LEO_TARGET_MOBILE
+#endif
 
 //------------------------------------------------------------------------------
 // Debug
@@ -256,6 +261,8 @@
     // Note: MSVC currently only supports SSSE3 but not AVX2
     #include <tmmintrin.h> // SSSE3: _mm_shuffle_epi8
     #include <emmintrin.h> // SSE2
+#elif defined(LEO_USE_SSE2NEON)
+    #include "sse2neon/sse2neon.h"
 #endif // LEO_TARGET_MOBILE
 
 #if defined(HAVE_ARM_NEON_H)
@@ -270,6 +277,8 @@
     // Compiler-specific 128-bit SIMD register keyword
     #define LEO_M128 uint8x16_t
     #define LEO_TRY_NEON
+#elif defined(LEO_USE_SSE2NEON)
+    #define LEO_M128 __m128i
 #else
     #define LEO_M128 uint64_t
 # endif
@@ -335,6 +344,8 @@ void InitializeCPUArch();
 # endif
     // Does CPU support SSSE3?
     extern bool CpuHasSSSE3;
+#elif defined(LEO_USE_SSE2NEON)
+    extern bool CpuHasSSSE3;
 #endif // LEO_TARGET_MOBILE
 
 
diff --git a/sse2neon b/sse2neon
new file mode 160000
index 0000000..cad518a
--- /dev/null
+++ b/sse2neon
@@ -0,0 +1 @@
+Subproject commit cad518a93b326f0f644b7972d488d04eaa2b0475