floraachy
/
OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

saxpy_haswell: Add AVX512 support 

avx512 support fits nicely in the C+intrinsics code and gets a
speed improvement for vectors where the saxpy operation is not fully
memory bound
This commit is contained in:
Arjan van de Ven 2018-08-05 17:50:16 +00:00
parent 06ea72f5a5
commit 850b73dbb9
1 changed files with 28 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

28
kernel/x86_64/saxpy_microk_haswell-2.c
View File

@ -44,6 +44,34 @@ static void saxpy_kernel_16( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
__alpha = _mm256_broadcastss_ps(_mm_load_ss(alpha)); __alpha = _mm256_broadcastss_ps(_mm_load_ss(alpha));
#ifdef __AVX512CD__
BLASLONG n64;
__m512 __alpha5;
__alpha5 = _mm512_broadcastss_ps(_mm_load_ss(alpha));
n64 = n & ~63;
for (; i < n64; i+= 64) {
__m512 y0, y16, y32, y48;
y0 = _mm512_loadu_ps(&y[i + 0]);
y16 = _mm512_loadu_ps(&y[i + 16]);
y32 = _mm512_loadu_ps(&y[i + 32]);
y48 = _mm512_loadu_ps(&y[i + 48]);
y0 += __alpha5 * _mm512_loadu_ps(&x[i + 0]);
y16 += __alpha5 * _mm512_loadu_ps(&x[i + 16]);
y32 += __alpha5 * _mm512_loadu_ps(&x[i + 32]);
y48 += __alpha5 * _mm512_loadu_ps(&x[i + 48]);
_mm512_storeu_ps(&y[i + 0], y0);
_mm512_storeu_ps(&y[i + 16], y16);
_mm512_storeu_ps(&y[i + 32], y32);
_mm512_storeu_ps(&y[i + 48], y48);
}
#endif
for (; i < n; i+= 32) { for (; i < n; i+= 32) {
__m256 y0, y8, y16, y24; __m256 y0, y8, y16, y24;