Add CPU instruction set requirements

CMakeLists.txt

@@ -24,6 +24,18 @@ if (WIN32)
     set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} "${CMAKE_SOURCE_DIR}/lib/")
 endif()
 
+# Check CPU requirements
+include(cmake/ZREDetectSSE42.cmake)
+message(STATUS "SUPPORTS_SSE4.2 = ${SUPPORTS_SSE42}")
+include(cmake/ZREDetectNEON.cmake)
+message(STATUS "SUPPORTS_NEON = ${SUPPORTS_NEON}")
+
+if(CMAKE_SIZEOF_VOID_P EQUAL 8 AND CMAKE_SYSTEM_PROCESSOR MATCHES "x86_64|amd64|AMD64" AND NOT SUPPORTS_AVX)
+    message(FATAL_ERROR "FATAL: A CPU supporting SSE4.2 is required.")
+elseif(NOT SUPPORTS_NEON)
+    message(FATAL_ERROR "FATAL: A CPU supporting NEON is required.")
+endif()
+
 set(RT64_STATIC TRUE)
 add_subdirectory(${CMAKE_SOURCE_DIR}/lib/rt64 ${CMAKE_BINARY_DIR}/rt64)
 

cmake/ZREDetectNEON.cmake

@@ -0,0 +1,4 @@
+try_compile(SUPPORTS_NEON
+  "${CMAKE_BINARY_DIR}"
+  "${CMAKE_SOURCE_DIR}/cmake/checks/cpu_neon.cpp"
+)

cmake/ZREDetectSSE42.cmake

@@ -0,0 +1,4 @@
+try_compile(SUPPORTS_SSE42
+  "${CMAKE_BINARY_DIR}"
+  "${CMAKE_SOURCE_DIR}/cmake/checks/cpu_sse42.cpp"
+)

cmake/checks/cpu_neon.cpp

@@ -0,0 +1,29 @@
+#include <stdio.h>
+
+#if defined _WIN32 && (defined(_M_ARM) || defined(_M_ARM64))
+#include <Intrin.h>
+#include <arm_neon.h>
+#define CV_NEON 1
+#elif defined(__ARM_NEON)
+#include <arm_neon.h>
+#define CV_NEON 1
+#endif
+
+void test_aliased_type(const uint8x16_t &a) {}
+void test_aliased_type(const int8x16_t &a) {}
+
+#if defined CV_NEON
+int test() {
+  const float src[] = {0.0f, 0.0f, 0.0f, 0.0f};
+  float32x4_t val = vld1q_f32((const float32_t *)(src));
+  return (int)vgetq_lane_f32(val, 0);
+}
+#else
+#error "NEON is not supported"
+#endif
+
+int main() {
+  auto result = test();
+  printf("NEON supported and test function executed. Result: %d\n", result);
+  return 0;
+}

cmake/checks/cpu_sse42.cpp

@@ -0,0 +1,12 @@
+#include <stdio.h>
+#include <nmmintrin.h>
+
+int test() {
+    unsigned int res = _mm_crc32_u8(1, 2);
+}
+
+int main() {
+    auto result = test();
+    printf("SSE4.2 supported and test function executed. Result: %d\n", result);
+    return 0;
+}