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

optimized saxpy.c for increments != 1

This commit is contained in:
Werner Saar 2015-04-24 11:52:59 +02:00
parent 0273966abb
commit dee100d0e4
1 changed files with 25 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

27
kernel/x86_64/saxpy.c
View File

@ -76,9 +76,9 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS
{ {
#if defined(SANDYBRIDGE) #if defined(SANDYBRIDGE)
int n1 = n & -64; BLASLONG n1 = n & -64;
#else #else
int n1 = n & -32; BLASLONG n1 = n & -32;
#endif #endif
if ( n1 ) if ( n1 )
@ -97,6 +97,29 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS
} }
BLASLONG n1 = n & -4;
while(i < n1)
{
FLOAT m1 = da * x[ix] ;
FLOAT m2 = da * x[ix+inc_x] ;
FLOAT m3 = da * x[ix+2*inc_x] ;
FLOAT m4 = da * x[ix+3*inc_x] ;
y[iy] += m1 ;
y[iy+inc_y] += m2 ;
y[iy+2*inc_y] += m3 ;
y[iy+3*inc_y] += m4 ;
ix += inc_x*4 ;
iy += inc_y*4 ;
i+=4 ;
}
while(i < n) while(i < n)
{ {