status-im
/
leopard
mirror of https://github.com/status-im/leopard.git
2
0
Fork 0
Code Issues Projects Releases Wiki Activity

Merge pull request #1 from cskiraly/arm64 

add support for Apple M1 and other arm64+Neon architectures
This commit is contained in:
Csaba Kiraly 2022-05-19 10:38:18 +02:00 committed by GitHub
parent b58d1eaf59 a16be41a8d
commit 20eb7c8bc5
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
4 changed files with 17 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
.gitmodules vendored Normal file
View File

@ -0,0 +1,3 @@
[submodule "sse2neon"]
path = sse2neon
url = https://github.com/DLTcollab/sse2neon

2
LeopardCommon.cpp
View File

@ -107,6 +107,8 @@ static void _cpuid(unsigned int cpu_info[4U], const unsigned int cpu_info_type)
#endif #endif
} }
#elif defined(LEO_USE_SSE2NEON)
bool CpuHasSSSE3 = true;
#endif // defined(LEO_TARGET_MOBILE) #endif // defined(LEO_TARGET_MOBILE)

11
LeopardCommon.h
View File

@ -186,6 +186,11 @@
// Unroll inner loops 4 times // Unroll inner loops 4 times
#define LEO_USE_VECTOR4_OPT #define LEO_USE_VECTOR4_OPT
// MacOS M1
#if defined(__aarch64__)
#define LEO_USE_SSE2NEON
#define LEO_TARGET_MOBILE
#endif
//------------------------------------------------------------------------------ //------------------------------------------------------------------------------
// Debug // Debug
@ -256,6 +261,8 @@
// Note: MSVC currently only supports SSSE3 but not AVX2 // Note: MSVC currently only supports SSSE3 but not AVX2
#include <tmmintrin.h> // SSSE3: _mm_shuffle_epi8 #include <tmmintrin.h> // SSSE3: _mm_shuffle_epi8
#include <emmintrin.h> // SSE2 #include <emmintrin.h> // SSE2
#elif defined(LEO_USE_SSE2NEON)
#include "sse2neon/sse2neon.h"
#endif // LEO_TARGET_MOBILE #endif // LEO_TARGET_MOBILE
#if defined(HAVE_ARM_NEON_H) #if defined(HAVE_ARM_NEON_H)
@ -270,6 +277,8 @@
// Compiler-specific 128-bit SIMD register keyword // Compiler-specific 128-bit SIMD register keyword
#define LEO_M128 uint8x16_t #define LEO_M128 uint8x16_t
#define LEO_TRY_NEON #define LEO_TRY_NEON
#elif defined(LEO_USE_SSE2NEON)
#define LEO_M128 __m128i
#else #else
#define LEO_M128 uint64_t #define LEO_M128 uint64_t
# endif # endif
@ -335,6 +344,8 @@ void InitializeCPUArch();
# endif # endif
// Does CPU support SSSE3? // Does CPU support SSSE3?
extern bool CpuHasSSSE3; extern bool CpuHasSSSE3;
#elif defined(LEO_USE_SSE2NEON)
extern bool CpuHasSSSE3;
#endif // LEO_TARGET_MOBILE #endif // LEO_TARGET_MOBILE

1
sse2neon Submodule

@ -0,0 +1 @@
Subproject commit cad518a93b326f0f644b7972d488d04eaa2b0475