#include #include #ifdef _MSC_VER #include int sse3_available(void) { return 1; } #else #include #include #include #if defined(__APPLE__) int sse3_available(void) { return 1; } #else int sse3_available(void) { return __builtin_cpu_supports("sse3"); } #endif #endif void increment_sse3(float arr[4]) { ALIGN_16 double darr[4]; __m128d val1 = _mm_set_pd(arr[0], arr[1]); __m128d val2 = _mm_set_pd(arr[2], arr[3]); __m128d one = _mm_set_pd(1.0, 1.0); __m128d result = _mm_add_pd(val1, one); _mm_store_pd(darr, result); result = _mm_add_pd(val2, one); _mm_store_pd(&darr[2], result); result = _mm_hadd_pd(val1, val2); /* This does nothing. Only here so we use an SSE3 instruction. */ arr[0] = (float)darr[1]; arr[1] = (float)darr[0]; arr[2] = (float)darr[3]; arr[3] = (float)darr[2]; }