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

optimized ddot.c for increments != 1

This commit is contained in:
Werner Saar 2015-04-24 10:56:55 +02:00
parent 3c09cea4b2
commit 3a67daa954
1 changed files with 26 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

27
kernel/x86_64/ddot.c
View File

@ -101,15 +101,40 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
} }
FLOAT temp1 = 0.0;
FLOAT temp2 = 0.0;
BLASLONG n1 = n & -4;
while(i < n1)
{
FLOAT m1 = y[iy] * x[ix] ;
FLOAT m2 = y[iy+inc_y] * x[ix+inc_x] ;
FLOAT m3 = y[iy+2*inc_y] * x[ix+2*inc_x] ;
FLOAT m4 = y[iy+3*inc_y] * x[ix+3*inc_x] ;
ix += inc_x*4 ;
iy += inc_y*4 ;
temp1 += m1+m3;
temp2 += m2+m4;
i+=4 ;
}
while(i < n) while(i < n)
{ {
dot += y[iy] * x[ix] ; temp1 += y[iy] * x[ix] ;
ix += inc_x ; ix += inc_x ;
iy += inc_y ; iy += inc_y ;
i++ ; i++ ;
} }
dot = temp1 + temp2;
return(dot); return(dot);
} }