Merge pull request #15 from steinwurf/fix-aarch64-neon-detection

Fix NEON detection on AArch64

Author: jpihl

.github/workflows/cpp-internal.yml

@@ -84,25 +84,26 @@ jobs:
     container:
       image: ghcr.io/steinwurf/build-images/zig-cpp
       options: --user 0:0
-      volumes:
-        - /root/.ssh:/root/.ssh
     steps:
       - name: Checkout source code
         uses: actions/checkout@v4
-      - name: Install dependencies
+        with:
+          persist-credentials: false
+      - name: Configure Github Authentication
         run: |
-          apk add --no-cache \
-            python3 \
-            py3-pip \
-            git \
-            cmake \
-            build-base
-      - name: Waf Clean
-        run: python3 waf clean --no_resolve
+          git config --global credential.helper 'store'
+          git credential approve <<EOF
+          protocol=https
+          host=github.com
+          username=x-access-token
+          password=${{ secrets.GH_ACCESS_TOKEN }}
+          EOF
       - name: Waf Configure with Zig Toolchain
-        run: python3 waf configure --git_protocol=git@ --cmake_toolchain=./resolve_symlinks/toolchains/zig-toolchain-x86_64-linux-musl.cmake --cmake_verbose
+        run: python3 waf configure --git_protocol=https:// --cmake_toolchain=./resolve_symlinks/toolchains/zig-toolchain-x86_64-linux-musl.cmake --cmake_verbose
       - name: Waf Build with Zig Toolchain
-        run: python3 waf build --run_tests
+        run: python3 waf build
+      - name: Waf Run Tests
+        run: python3 waf --run_tests
 
   macos_cmake:
     timeout-minutes: 45

CMakeLists.txt

@@ -60,6 +60,7 @@ target_link_libraries(aybabtu PRIVATE steinwurf::platform)
 
 # Check Accelerations
 include(CheckCXXCompilerFlag)
+include(CheckCXXSourceCompiles)
 
 if(CMAKE_C_COMPILER_ID MATCHES "GNU|AppleClang|Clang")
   # # SSE2 check_cxx_compiler_flag("-Werror -msse2" HAS_SSE2)
@@ -89,12 +90,21 @@ if(CMAKE_C_COMPILER_ID MATCHES "GNU|AppleClang|Clang")
     set_source_files_properties(${avx2} PROPERTIES COMPILE_FLAGS -mavx2)
   endif()
 
+
   # NEON
-  check_cxx_compiler_flag("-Werror -mfpu=neon" HAS_NEON)
+  if(CMAKE_SYSTEM_PROCESSOR MATCHES "^(aarch64|AARCH64|arm64|ARM64)$")
+    check_cxx_source_compiles(
+      "int main()\n{\n#if defined(__ARM_NEON) || defined(__ARM_NEON__) || defined(__aarch64__)\n  return 0;\n#else\n#error NEON is required\n#endif\n}\n"
+      HAS_NEON)
+  else()
+    check_cxx_compiler_flag("-Werror -mfpu=neon" HAS_NEON)
+  endif()
 
   if(HAS_NEON)
     file(GLOB_RECURSE neon ./src/*neon.cpp)
-    set_source_files_properties(${neon} PROPERTIES COMPILE_FLAGS -mfpu=neon)
+    if(NOT CMAKE_SYSTEM_PROCESSOR MATCHES "^(aarch64|AARCH64|arm64|ARM64)$")
+      set_source_files_properties(${neon} PROPERTIES COMPILE_FLAGS -mfpu=neon)
+    endif()
   endif()
 elseif(CMAKE_C_COMPILER_ID MATCHES "MSVC")
   # AVX2
@@ -129,7 +139,7 @@ if(${CMAKE_PROJECT_NAME} STREQUAL ${PROJECT_NAME})
   # Build test executable
   file(GLOB_RECURSE aybabtu_tests_sources ./test/*.cpp)
   add_executable(aybabtu_tests ${aybabtu_tests_sources})
-  target_link_libraries(aybabtu_tests 
+  target_link_libraries(aybabtu_tests
     gtest_main
     steinwurf::aybabtu
     steinwurf::cpuid

NEWS.rst

@@ -6,7 +6,8 @@ every change, see the Git log.
 
 Latest
 ------
-* tbd
+* Patch: Fixed NEON detection on AArch64 by checking for compiler intrinsics instead of
+  relying on the unavailable -mfpu=neon flag.
 
 5.0.0
 -----