From 1d1b79d5f02070348c708dfa5132c8d3decac4a0 Mon Sep 17 00:00:00 2001
From: Enrique Martinez <73340871+emartinez-dev@users.noreply.github.com>
Date: Sat, 7 Feb 2026 00:12:36 +0100
Subject: [PATCH] Added ARM64 (Apple Silicon) support using sse2neon (#50)

* feat: implemented arm64 support for builds

* feat: explicitly add compiler flags when needed
---
 CMakeLists.txt                           | 37 ++++++++++++++++++++++++
 ps2xRuntime/include/ps2_runtime.h        | 10 +++++--
 ps2xRuntime/include/ps2_runtime_macros.h |  2 ++
 3 files changed, 47 insertions(+), 2 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index f57ce0d..0ef4d2a 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -6,6 +6,43 @@ set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
 
 option(PS2X_RUNTIME OFF)
 
+# ARM64 support using sse2neon
+if(CMAKE_SYSTEM_PROCESSOR MATCHES "arm64|aarch64|ARM64")
+    message(STATUS "ARM64 detected, fetching sse2neon")
+
+    include(FetchContent)
+    FetchContent_Declare(
+        sse2neon
+        GIT_REPOSITORY https://github.com/DLTcollab/sse2neon.git
+        GIT_TAG v1.9.1
+        GIT_SHALLOW TRUE
+    )
+    FetchContent_MakeAvailable(sse2neon)
+
+    include_directories(${sse2neon_SOURCE_DIR})
+    add_compile_definitions(USE_SSE2NEON)
+
+    if(APPLE)
+        # macOS ARM64 already uses optimal defaults
+        message(STATUS "macOS ARM64 - using default compiler flags")
+    elseif(MSVC)
+        # Windows ARM64 - MSVC already uses optimal defaults
+        message(STATUS "Windows ARM64 (MSVC) - using default compiler flags")
+    else()
+        # Linux ARM64 - add NEON flags
+        message(STATUS "Non-Apple ARM64 - adding NEON compiler flags")
+        add_compile_options(-march=armv8-a+fp+simd)
+
+        # Try to enable crypto and CRC extensions if supported
+        include(CheckCXXCompilerFlag)
+        check_cxx_compiler_flag("-march=armv8-a+fp+simd+crypto+crc" COMPILER_SUPPORTS_CRYPTO_CRC)
+        if(COMPILER_SUPPORTS_CRYPTO_CRC)
+            set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -march=armv8-a+fp+simd+crypto+crc")
+            message(STATUS "Crypto and CRC extensions enabled")
+        endif()
+    endif()
+endif()
+
 add_subdirectory("ps2xRecomp")
 
 add_subdirectory("ps2xRuntime")
diff --git a/ps2xRuntime/include/ps2_runtime.h b/ps2xRuntime/include/ps2_runtime.h
index db73007..4c3d82e 100644
--- a/ps2xRuntime/include/ps2_runtime.h
+++ b/ps2xRuntime/include/ps2_runtime.h
@@ -6,8 +6,14 @@
 #include <unordered_map>
 #include <string>
 #include <functional>
-#include <immintrin.h> // For SSE/AVX instructions
-#include <smmintrin.h>
+#if defined(_MSC_VER)
+    #include <intrin.h>
+#elif defined(USE_SSE2NEON)
+    #include "sse2neon.h"
+#else
+    #include <immintrin.h> // For SSE/AVX instructions
+    #include <smmintrin.h> // For SSE4.1 instructions
+#endif
 #include <atomic>
 #include <filesystem>
 #include <iostream>
diff --git a/ps2xRuntime/include/ps2_runtime_macros.h b/ps2xRuntime/include/ps2_runtime_macros.h
index 1576f1f..8f33073 100644
--- a/ps2xRuntime/include/ps2_runtime_macros.h
+++ b/ps2xRuntime/include/ps2_runtime_macros.h
@@ -3,6 +3,8 @@
 #include <cstdint>
 #if defined(_MSC_VER)
 	#include <intrin.h>
+#elif defined(USE_SSE2NEON)
+	#include "sse2neon.h"
 #else
 	#include <immintrin.h> // For SSE/AVX intrinsics
 #endif