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

Optimize the performance of daxpy by using universal intrinsics

Browse Source
This commit is contained in:
Qiyu8
2020-09-22 10:38:35 +08:00
parent 0f112077e6
commit 325b539c26
5 changed files with 131 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files
kernel/x86_64/daxpy.c
+23 -16
View File
@@ -45,28 +45,35 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "daxpy_microk_sandy-2.c"
#endif
#ifndef HAVE_KERNEL_8
#include"../simd/intrin.h"
static void daxpy_kernel_8(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
void daxpy_kernel_8(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
{
BLASLONG register i = 0;
FLOAT a = *alpha;
#if V_SIMD
v_f32 __alpha, tmp;
__alpha = v_setall_f32(*alpha);
const int vstep = v_nlanes_f32;
for (; i < n; i += vstep) {
tmp = v_add_f32(v_loadu_f32(y + i), v_mul_f32(__alpha, v_loadu_f32( x + i )));
v_storeu_f32(y + i, tmp);
}
#else
while(i < n)
{
y[i] += a * x[i];
y[i+1] += a * x[i+1];
y[i+2] += a * x[i+2];
y[i+3] += a * x[i+3];
y[i+4] += a * x[i+4];
y[i+5] += a * x[i+5];
y[i+6] += a * x[i+6];
y[i+7] += a * x[i+7];
i+=8 ;
}
{
y[i] += a * x[i];
y[i+1] += a * x[i+1];
y[i+2] += a * x[i+2];
y[i+3] += a * x[i+3];
y[i+4] += a * x[i+4];
y[i+5] += a * x[i+5];
y[i+6] += a * x[i+6];
y[i+7] += a * x[i+7];
i+=8 ;
}
#endif
}
#endif