Port SIMD to Arm Neon

.github/workflows/CI.yml

@@ -1,6 +1,6 @@
 name: CI
 
-on: [push, pull_request]
+on: [push, pull_request, workflow_dispatch]
 
 jobs:
   build:
@@ -88,3 +88,9 @@ jobs:
         cmake --build build_avs -j 2
         cmake -S . -B build_vs -G Ninja -DENABLE_DAV1D=OFF -DENABLE_MFX=OFF -DENABLE_XML2=OFF -DBUILD_AVS_PLUGIN=OFF -DENABLE_VPX=OFF
         cmake --build build_vs -j 2
+    - name: Upload dylib
+      if: matrix.os == ${{'macos-latest'}}
+      uses: actions/upload-artifact@v3
+      with:
+        name: libvslsmashsource
+        path: build_vs/liblsmashsource.1.dylib

AviUtl/colorspace_simd.c

@@ -27,10 +27,17 @@
 #include "../common/utils.h"
 #include "../common/lwsimd.h"
 
+
+
+#if defined(__x86_64__) || defined(_M_X64) || defined(__i386) || defined(_M_IX86)
+#include <tmmintrin.h>
 #ifdef __GNUC__
 #pragma GCC target ("ssse3")
 #endif
-#include <tmmintrin.h>
+#elif defined(__arm__) || defined(__aarch64__)
+#include "../common/sse2neon.h"
+#endif
+
 /* SSSE3 version of func convert_yv12i_to_yuy2 */
 void LW_FUNC_ALIGN convert_yv12i_to_yuy2_ssse3
 (
@@ -204,10 +211,15 @@ void LW_FUNC_ALIGN convert_yv12i_to_yuy2_ssse3
     }
 }
 
+
+#if defined(__x86_64__) || defined(_M_X64) || defined(__i386) || defined(_M_IX86)
+#include <smmintrin.h>
 #ifdef __GNUC__
 #pragma GCC target ("sse4.1")
 #endif
-#include <smmintrin.h>
+#elif defined(__arm__) || defined(__aarch64__)
+#include "../common/sse2neon.h"
+#endif
 
 /* the inner loop branch should be deleted by forced inline expansion and "bit_depth" constant propagation. */
 static void LW_FUNC_ALIGN LW_FORCEINLINE convert_yuv420ple_i_to_yuv444p16le_sse41

AviUtl/lwcolor_simd.c

@@ -33,10 +33,15 @@ typedef unsigned char BYTE;
 typedef unsigned short USHORT;
 #endif
 
+
+#if defined(__x86_64__) || defined(_M_X64) || defined(__i386) || defined(_M_IX86)
+#include <smmintrin.h>
 #ifdef __GNUC__
 #pragma GCC target ("sse4.1")
 #endif
-#include <smmintrin.h>
+#elif defined(__arm__) || defined(__aarch64__)
+#include "sse2neon.h"
+#endif
 
 static LW_FORCEINLINE void fill_rgb_buffer_sse41( BYTE *rgb_buffer, BYTE *lw48_ptr )
 {

CMakeLists.txt

@@ -27,6 +27,12 @@ option(ENABLE_VPX "Enable libvpx support" ON)
 message(STATUS "Enable libvpx support: ${ENABLE_VPX}.")
 
 cmake_host_system_information(RESULT sse2 QUERY HAS_SSE2)
+
+if(CMAKE_SYSTEM_PROCESSOR MATCHES "arm*")
+    # sse2neon
+    set(sse2 ON)
+endif()
+
 option(ENABLE_SSE2 "Enable SSE2 support" ${sse2})
 message(STATUS "Enable SSE2 support: ${ENABLE_SSE2}.")
 
@@ -220,11 +226,12 @@ if (NOT CMAKE_GENERATOR MATCHES "Visual Studio")
         target_compile_definitions(LSMASHSource PRIVATE DEBUG_BUILD)
     else (build_type STREQUAL release)
         target_compile_definitions(LSMASHSource PRIVATE RELEASE_BUILD)
+        target_compile_options(LSMASHSource PRIVATE -Ofast -funroll-loops)
     endif()
-
     message(STATUS "Build type - ${CMAKE_BUILD_TYPE}")
 endif()
 
+
 if (${sse2})
     target_compile_definitions(LSMASHSource PRIVATE SSE2_ENABLED=1)
 endif()

VapourSynth/meson.build

@@ -56,6 +56,8 @@ deps = [
   version_h
 ]
 
+add_project_arguments('-Ofast', '-funroll-loops', language: 'c')
+
 if host_machine.cpu_family().startswith('x86')
   add_project_arguments('-mfpmath=sse', '-msse2', language: 'c')
 endif

common/planar_yuv_sse2.c

@@ -1,4 +1,8 @@
+#if defined(__x86_64__) || defined(_M_X64) || defined(__i386) || defined(_M_IX86)
 #include <emmintrin.h>
+#elif defined(__arm__) || defined(__aarch64__)
+#include "sse2neon.h"
+#endif
 #include  <stdint.h>
 
 static inline __m128i _MM_PACKUS_EPI32(const __m128i* low, const __m128i* high)