From b389e93e4f3f0de53ded778c62ef14c223e0621d Mon Sep 17 00:00:00 2001
From: Igor Freire <igor@blockstream.com>
Date: Thu, 15 Apr 2021 15:04:04 -0300
Subject: [PATCH] Review intrinsics includes and macros on gf256

- Define the LINUX_ARM macro automatically based on ARM flags set by the
  compiler.
- Detect ARM Neon based on compiler flags. Take both __ARM_NEON and
  __ARM_NEON__ into account.
- Fix the condition for inclusion of AVX2, as discussed in Issue #31.
- Reorganize some macro definitions for better readability. For example,
  use an isolated if-elif-else conditional to define GF256_M128.
- Throw error if SSE2 is not available when building for a non-mobile
  target.
- Remove the unused GF256_ALIGNED_ACCESSES macro.
---
 CMakeLists.txt |  3 ---
 gf256.cpp      |  2 +-
 gf256.h        | 48 ++++++++++++++++++++++--------------------------
 3 files changed, 23 insertions(+), 30 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 7852b3f..a0e6a16 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -59,9 +59,6 @@ else()
     set(CMAKE_CXX_FLAGS "-Wall -Wextra")
     set(CMAKE_CXX_FLAGS_DEBUG "-g -O0 -march=native")
     set(CMAKE_CXX_FLAGS_RELEASE "-O3 -march=native")
-    if (CMAKE_SYSTEM_PROCESSOR MATCHES "aarch64")
-        add_definitions(-DLINUX_ARM=1)
-    endif()
 endif()
 
 add_library(wirehair ${LIB_SOURCE_FILES})
diff --git a/gf256.cpp b/gf256.cpp
index 510146b..66aa9f0 100644
--- a/gf256.cpp
+++ b/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/gf256.h b/gf256.h
index 71134d2..3eee8b8 100644
--- a/gf256.h
+++ b/gf256.h
@@ -53,50 +53,46 @@
 //------------------------------------------------------------------------------
 // Platform/Architecture
 
+#if defined(__ARM_ARCH) || defined(__ARM_NEON) || defined(__ARM_NEON__)
+    #define LINUX_ARM
+#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) && !defined(GF256_TARGET_MOBILE)
+#if defined(__AVX2__) && (!defined (_MSC_VER) || _MSC_VER >= 1900)
     #define GF256_TRY_AVX2 /* 256-bit */
     #include <immintrin.h>
     #define GF256_ALIGN_BYTES 32
+    #define GF256_M256 __m256i
 #else // __AVX2__
     #define GF256_ALIGN_BYTES 16
 #endif // __AVX2__
 
-#if !defined(GF256_TARGET_MOBILE)
-    // Note: MSVC currently only supports SSSE3 but not AVX2
-    #include <tmmintrin.h> // SSSE3: _mm_shuffle_epi8
+#if defined(__SSE2__)
     #include <emmintrin.h> // SSE2
-#endif // GF256_TARGET_MOBILE
+#elif !defined(GF256_TARGET_MOBILE)
+    #error "SSE2 is required for non-mobile target"
+#endif // __SSE2__
 
-#if defined(HAVE_ARM_NEON_H)
-    #include <arm_neon.h>
-#endif // HAVE_ARM_NEON_H
-
-#if defined(GF256_TARGET_MOBILE)
+#if defined(__SSE3__)
+    #define GF256_TRY_SSE3
+    #include <tmmintrin.h> // SSSE3: _mm_shuffle_epi8
+#endif // __SSE3__
 
-    #define GF256_ALIGNED_ACCESSES /* Inputs must be aligned to GF256_ALIGN_BYTES */
+#if defined(__ARM_NEON) || defined(__ARM_NEON__)
+    #include <arm_neon.h>
+    #define GF256_TRY_NEON
+#endif // __ARM_NEON__
 
-# if defined(HAVE_ARM_NEON_H)
-    // Compiler-specific 128-bit SIMD register keyword
+// Compiler-specific 128-bit SIMD register keyword
+#if defined(GF256_TARGET_MOBILE) && defined(GF256_TRY_NEON)
     #define GF256_M128 uint8x16_t
-    #define GF256_TRY_NEON
+#elif defined(__SSE2__) || defined(GF256_TRY_SSE3)
+    #define GF256_M128 __m128i
 #else
     #define GF256_M128 uint64_t
-# endif
-
-#else // GF256_TARGET_MOBILE
-
-    // Compiler-specific 128-bit SIMD register keyword
-    #define GF256_M128 __m128i
-
-#endif // GF256_TARGET_MOBILE
-
-#ifdef GF256_TRY_AVX2
-    // Compiler-specific 256-bit SIMD register keyword
-    #define GF256_M256 __m256i
 #endif
 
 // Compiler-specific C++11 restrict keyword