int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *buffer)
{
BLASLONG i;
BLASLONG ix,iy;
BLASLONG jx,jy;
BLASLONG j;
BLASLONG j1;
BLASLONG j2;
BLASLONG m2;
FLOAT temp1;
FLOAT temp2;
FLOAT *xp, *yp;
FLOAT *a0,*a1,*a2,*a3;
FLOAT at0,at1,at2,at3;
FLOAT tmp1[4];
FLOAT tmp2[4];
#if 0
if( m != offset )
printf("Symv_U: m=%d offset=%d\n",m,offset);
#endif
BLASLONG m1 = m - offset;
BLASLONG mrange = m -m1;
if ( (inc_x!=1) || (inc_y!=1) || (mrange<16) )
{
jx = m1 * inc_x;
jy = m1 * inc_y;
for (j=m1; j<m; j++)
{
temp1 = alpha * x[jx];
temp2 = 0.0;
iy = 0;
ix = 0;
for (i=0; i<j; i++)
{
y[iy] += temp1 * a[j*lda+i];
temp2 += a[j*lda+i] * x[ix];
ix += inc_x;
iy += inc_y;
}
y[jy] += temp1 * a[j*lda+j] + alpha * temp2;
jx += inc_x;
jy += inc_y;
}
return(0);
}
xp = x;
yp = y;
m2 = m - ( mrange % 4 );
for (j=m1; j<m2; j+=4)
{
tmp1[0] = alpha * xp[j];
tmp1[1] = alpha * xp[j+1];
tmp1[2] = alpha * xp[j+2];
tmp1[3] = alpha * xp[j+3];
tmp2[0] = 0.0;
tmp2[1] = 0.0;
tmp2[2] = 0.0;
tmp2[3] = 0.0;
a0 = &a[j*lda];
a1 = a0+lda;
a2 = a1+lda;
a3 = a2+lda;
j1 = (j/8)*8;
if ( j1 )
dsymv_kernel_4x4(j1, a0, a1, a2, a3, xp, yp, tmp1, tmp2);
if ( j1 < j )
dsymv_kernel_1x4(j1, j, a0, a1, a2, a3, xp, yp, tmp1, tmp2);
j2 = 0;
for ( j1 = j ; j1 < j+4 ; j1++ )
{
temp1 = tmp1[j2];
temp2 = tmp2[j2];
a0 = &a[j1*lda];
for ( i=j ; i<j1; i++ )
{
yp[i] += temp1 * a0[i];
temp2 += a0[i] * xp[i];
}
y[j1] += temp1 * a0[j1] + alpha * temp2;
j2++;
}
}
for ( ; j<m; j++)
{
temp1 = alpha * xp[j];
temp2 = 0.0;
a0 = &a[j*lda];
FLOAT at0;
j1 = (j/8)*8;
if ( j1 )
dsymv_kernel_8x1(j1, a0, xp, yp, &temp1, &temp2);
for (i=j1 ; i<j; i++)
{
at0 = a0[i];
yp[i] += temp1 * at0;
temp2 += at0 * xp[i];
}
yp[j] += temp1 * a0[j] + alpha * temp2;
}
return(0);
}
int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *buffer)
{
BLASLONG i;
BLASLONG ix,iy;
BLASLONG jx,jy;
BLASLONG j;
FLOAT temp1;
FLOAT temp2;
FLOAT tmp1[4];
FLOAT tmp2[4];
FLOAT *ap[4];
#if 0
if ( m != offset )
printf("Symv_L: m=%d offset=%d\n",m,offset);
#endif
if ( (inc_x != 1) || (inc_y != 1) )
{
jx = 0;
jy = 0;
for (j=0; j<offset; j++)
{
temp1 = alpha * x[jx];
temp2 = 0.0;
y[jy] += temp1 * a[j*lda+j];
iy = jy;
ix = jx;
for (i=j+1; i<m; i++)
{
ix += inc_x;
iy += inc_y;
y[iy] += temp1 * a[j*lda+i];
temp2 += a[j*lda+i] * x[ix];
}
y[jy] += alpha * temp2;
jx += inc_x;
jy += inc_y;
}
return(0);
}
BLASLONG offset1 = (offset/4)*4;
for (j=0; j<offset1; j+=4)
{
tmp1[0] = alpha * x[j];
tmp1[1] = alpha * x[j+1];
tmp1[2] = alpha * x[j+2];
tmp1[3] = alpha * x[j+3];
tmp2[0] = 0.0;
tmp2[1] = 0.0;
tmp2[2] = 0.0;
tmp2[3] = 0.0;
ap[0] = &a[j*lda];
ap[1] = ap[0] + lda;
ap[2] = ap[1] + lda;
ap[3] = ap[2] + lda;
y[j] += tmp1[0] * ap[0][j];
y[j+1] += tmp1[1] * ap[1][j+1];
y[j+2] += tmp1[2] * ap[2][j+2];
y[j+3] += tmp1[3] * ap[3][j+3];
BLASLONG from = j+1;
if ( m - from >=12 )
{
BLASLONG m2 = (m/4)*4;
for (i=j+1; i<j+4; i++)
{
y[i] += tmp1[0] * ap[0][i];
tmp2[0] += ap[0][i] * x[i];
}
for (i=j+2; i<j+4; i++)
{
y[i] += tmp1[1] * ap[1][i];
tmp2[1] += ap[1][i] * x[i];
}
for (i=j+3; i<j+4; i++)
{
y[i] += tmp1[2] * ap[2][i];
tmp2[2] += ap[2][i] * x[i];
}
if ( m2 > j+4 )
dsymv_kernel_4x4(j+4,m2,ap,x,y,tmp1,tmp2);
for (i=m2; i<m; i++)
{
y[i] += tmp1[0] * ap[0][i];
tmp2[0] += ap[0][i] * x[i];
y[i] += tmp1[1] * ap[1][i];
tmp2[1] += ap[1][i] * x[i];
y[i] += tmp1[2] * ap[2][i];
tmp2[2] += ap[2][i] * x[i];
y[i] += tmp1[3] * ap[3][i];
tmp2[3] += ap[3][i] * x[i];
}
}
else
{
for (i=j+1; i<j+4; i++)
{
y[i] += tmp1[0] * ap[0][i];
tmp2[0] += ap[0][i] * x[i];
}
for (i=j+2; i<j+4; i++)
{
y[i] += tmp1[1] * ap[1][i];
tmp2[1] += ap[1][i] * x[i];
}
for (i=j+3; i<j+4; i++)
{
y[i] += tmp1[2] * ap[2][i];
tmp2[2] += ap[2][i] * x[i];
}
for (i=j+4; i<m; i++)
{
y[i] += tmp1[0] * ap[0][i];
tmp2[0] += ap[0][i] * x[i];
y[i] += tmp1[1] * ap[1][i];
tmp2[1] += ap[1][i] * x[i];
y[i] += tmp1[2] * ap[2][i];
tmp2[2] += ap[2][i] * x[i];
y[i] += tmp1[3] * ap[3][i];
tmp2[3] += ap[3][i] * x[i];
}
}
y[j] += alpha * tmp2[0];
y[j+1] += alpha * tmp2[1];
y[j+2] += alpha * tmp2[2];
y[j+3] += alpha * tmp2[3];
}
for (j=offset1; j<offset; j++)
{
temp1 = alpha * x[j];
temp2 = 0.0;
y[j] += temp1 * a[j*lda+j];
BLASLONG from = j+1;
if ( m - from >=8 )
{
BLASLONG j1 = ((from + 4)/4)*4;
BLASLONG j2 = (m/4)*4;
for (i=from; i<j1; i++)
{
y[i] += temp1 * a[j*lda+i];
temp2 += a[j*lda+i] * x[i];
}
for (i=j1; i<j2; i++)
{
y[i] += temp1 * a[j*lda+i];
temp2 += a[j*lda+i] * x[i];
}
for (i=j2; i<m; i++)
{
y[i] += temp1 * a[j*lda+i];
temp2 += a[j*lda+i] * x[i];
}
}
else
{
for (i=from; i<m; i++)
{
y[i] += temp1 * a[j*lda+i];
temp2 += a[j*lda+i] * x[i];
}
}
y[j] += alpha * temp2;
}
return(0);
}
