The Meson Build System http://mesonbuild.com
You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
42 lines
903 B
42 lines
903 B
#include<simdconfig.h> |
|
#include<simdfuncs.h> |
|
#include<stdint.h> |
|
|
|
/* |
|
* FIXME add proper runtime detection for VS. |
|
*/ |
|
|
|
#ifdef _MSC_VER |
|
#include<intrin.h> |
|
int avx2_available(void) { |
|
return 0; |
|
} |
|
#else |
|
#include<immintrin.h> |
|
#include<cpuid.h> |
|
|
|
#if defined(__APPLE__) |
|
int avx2_available(void) { return 0; } |
|
#else |
|
int avx2_available(void) { |
|
return __builtin_cpu_supports("avx2"); |
|
} |
|
#endif |
|
#endif |
|
|
|
void increment_avx2(float arr[4]) { |
|
double darr[4]; |
|
darr[0] = arr[0]; |
|
darr[1] = arr[1]; |
|
darr[2] = arr[2]; |
|
darr[3] = arr[3]; |
|
__m256d val = _mm256_loadu_pd(darr); |
|
__m256d one = _mm256_set1_pd(1.0); |
|
__m256d result = _mm256_add_pd(val, one); |
|
_mm256_storeu_pd(darr, result); |
|
one = _mm256_permute4x64_pd(one, 66); /* A no-op, just here to use AVX2. */ |
|
arr[0] = (float)darr[0]; |
|
arr[1] = (float)darr[1]; |
|
arr[2] = (float)darr[2]; |
|
arr[3] = (float)darr[3]; |
|
}
|
|
|