Update wirehair

Including my custom CMakeLists.
sz3 · Feb 18, 2022 · 39b37d9 · 39b37d9
1 parent 82ed9e9
commit 39b37d9
Show file tree

Hide file tree

Showing 5 changed files with 17 additions and 25 deletions.
diff --git a/src/third_party_lib/wirehair/CMakeLists.txt b/src/third_party_lib/wirehair/CMakeLists.txt
@@ -61,9 +61,6 @@ endif()
 if(MSVC)
 else()
  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${ARCH_NATIVE}")
- if (CMAKE_SYSTEM_PROCESSOR MATCHES "aarch64")
- add_definitions(-DLINUX_ARM=1)
- endif()
 endif()
 
 add_library(wirehair ${LIB_SOURCE_FILES})

diff --git a/src/third_party_lib/wirehair/WirehairTools.cpp b/src/third_party_lib/wirehair/WirehairTools.cpp
@@ -30,6 +30,7 @@
 #include "WirehairTools.h"
 
 #include <cmath>
+#include <cstdlib>
 
 #ifdef _MSC_VER
 #include <intrin.h> // _BitScanReverse

diff --git a/src/third_party_lib/wirehair/gf256.cpp b/src/third_party_lib/wirehair/gf256.cpp
@@ -200,7 +200,7 @@ static bool gf256_self_test()
 #endif
 
 #if defined(GF256_TRY_NEON)
-# if defined(IOS) && defined(__ARM_NEON__)
+# if defined(IOS) && (defined(__ARM_NEON) || defined(__ARM_NEON__))
 // Requires iPhone 5S or newer
 static const bool CpuHasNeon = true;
 static const bool CpuHasNeon64 = true;

diff --git a/src/third_party_lib/wirehair/gf256.h b/src/third_party_lib/wirehair/gf256.h
@@ -53,11 +53,17 @@
 //------------------------------------------------------------------------------
 // Platform/Architecture
 
+#if defined(__ARM_ARCH) || defined(__ARM_NEON) || defined(__ARM_NEON__)
+ #if !defined IOS
+ #define LINUX_ARM
+ #endif
+#endif
+
 #if defined(ANDROID) || defined(IOS) || defined(LINUX_ARM) || defined(__powerpc__) || defined(__s390__)
  #define GF256_TARGET_MOBILE
 #endif // ANDROID
 
-#if defined(__AVX2__) || (defined (_MSC_VER) && _MSC_VER >= 1900)
+#if defined(__AVX2__) && (!defined (_MSC_VER) || _MSC_VER >= 1900)
  #define GF256_TRY_AVX2 /* 256-bit */
  #include <immintrin.h>
  #define GF256_ALIGN_BYTES 32
@@ -66,36 +72,28 @@
 #endif // __AVX2__
 
 #if !defined(GF256_TARGET_MOBILE)
- // Note: MSVC currently only supports SSSE3 but not AVX2
  #include <tmmintrin.h> // SSSE3: _mm_shuffle_epi8
  #include <emmintrin.h> // SSE2
 #endif // GF256_TARGET_MOBILE
 
-#if defined(HAVE_ARM_NEON_H)
+#if defined(__ARM_NEON) || defined(__ARM_NEON__)
  #include <arm_neon.h>
-#endif // HAVE_ARM_NEON_H
+ #define GF256_TRY_NEON
+#endif
 
+// Compiler-specific 128-bit SIMD register keyword
 #if defined(GF256_TARGET_MOBILE)
-
- #define GF256_ALIGNED_ACCESSES /* Inputs must be aligned to GF256_ALIGN_BYTES */
-
-# if defined(HAVE_ARM_NEON_H)
- // Compiler-specific 128-bit SIMD register keyword
+#if defined(GF256_TRY_NEON)
  #define GF256_M128 uint8x16_t
- #define GF256_TRY_NEON
 #else
  #define GF256_M128 uint64_t
-# endif
-
+#endif // GF256_TRY_NEON
 #else // GF256_TARGET_MOBILE
-
- // Compiler-specific 128-bit SIMD register keyword
  #define GF256_M128 __m128i
-
 #endif // GF256_TARGET_MOBILE
 
+// Compiler-specific 256-bit SIMD register keyword
 #ifdef GF256_TRY_AVX2
- // Compiler-specific 256-bit SIMD register keyword
  #define GF256_M256 __m256i
 #endif
 
@@ -272,10 +270,6 @@ static GF256_FORCE_INLINE void gf256_div_mem(void * GF256_RESTRICT vz,
 //------------------------------------------------------------------------------
 // Misc Operations
 
-/// Swap two memory buffers in-place
-extern void gf256_memswap(void * GF256_RESTRICT vx, void * GF256_RESTRICT vy, int bytes);
-
-
 #ifdef __cplusplus
 }
 #endif // __cplusplus

diff --git a/src/third_party_lib/wirehair/wirehair.cpp b/src/third_party_lib/wirehair/wirehair.cpp
@@ -153,7 +153,7 @@ WIREHAIR_EXPORT WirehairCodec wirehair_decoder_create(
 )
 {
  // If input is invalid:
- if (messageBytes < 1 || blockBytes < 1) {
+ if (!m_init || messageBytes < 1 || blockBytes < 1) {
  return nullptr;
  }