void Mjoin(PATL,ger1_a1_x1_yX)
(const int M, const int N, const SCALAR alpha, const TYPE *X, const int incX,
const TYPE *Y, const int incY, TYPE *A, const int lda)
{
int i, j;
const int incy = incY<<2;
const TYPE *Y1= Y + incY, *Y2 = Y1 + incY, *Y3 = Y2 + incY;
const TYPE *x;
TYPE *A0 = A, *A1 = A + lda, *A2 = A1 + lda, *A3 = A2 + lda;
const int M4 = (M>>2)<<2, N4 = (N>>2)<<2, incAn = (lda<<2) - M4;
register TYPE x0, x1, x2, x3, y0, y1, y2, y3;
if (M4)
{
for (j=N4; j; j -= 4)
{
y0 = *Y; Y += incy;
y1 = *Y1; Y1 += incy;
y2 = *Y2; Y2 += incy;
y3 = *Y3; Y3 += incy;
x = X;
for (i=M4; i; i -= 4)
{
x0 = *x; x1 = x[1]; x2 = x[2]; x3 = x[3];
*A0 += y0 * x0;
x += 4;
*A1 += y1 * x0;
*A2 += y2 * x0;
*A3 += y3 * x0;
A0[1] += y0 * x1;
A1[1] += y1 * x1;
A2[1] += y2 * x1;
A3[1] += y3 * x1;
A0[2] += y0 * x2;
A1[2] += y1 * x2;
A2[2] += y2 * x2;
A3[2] += y3 * x2;
A0[3] += y0 * x3; A0 += 4;
A1[3] += y1 * x3; A1 += 4;
A2[3] += y2 * x3; A2 += 4;
A3[3] += y3 * x3; A3 += 4;
}
switch(M-M4)
{
case 1:
x0 = *x;
*A0 += y0 * x0;
*A1 += y1 * x0;
*A2 += y2 * x0;
*A3 += y3 * x0;
break;
case 2:
x0 = *x; x1 = x[1];
*A0 += y0 * x0;
*A1 += y1 * x0;
*A2 += y2 * x0;
*A3 += y3 * x0;
A0[1] += y0 * x1;
A1[1] += y1 * x1;
A2[1] += y2 * x1;
A3[1] += y3 * x1;
break;
case 3:
x0 = *x; x1 = x[1]; x2 = x[2];
*A0 += y0 * x0;
*A1 += y1 * x0;
*A2 += y2 * x0;
*A3 += y3 * x0;
A0[1] += y0 * x1;
A1[1] += y1 * x1;
A2[1] += y2 * x1;
A3[1] += y3 * x1;
A0[2] += y0 * x2;
A1[2] += y1 * x2;
A2[2] += y2 * x2;
A3[2] += y3 * x2;
break;
}
A0 += incAn;
A1 += incAn;
A2 += incAn;
A3 += incAn;
}
if (N-N4) ger_Nle4(M, N-N4, X, Y, incY, A0, lda);
}
else ger_Mle8(M, N, X, Y, incY, A, lda);
}
void Mjoin(PATL,ger1_a1_x1_yX)
(const int M, const int N, const SCALAR alpha, const TYPE *X, const int incX,
const TYPE *Y, const int incY, TYPE *A, const int lda)
{
int i, j;
const int incy = incY<<2;
const TYPE *Y1= Y + incY, *Y2 = Y1 + incY, *Y3 = Y2 + incY;
const TYPE *x;
TYPE *A0 = A, *A1 = A + lda, *A2 = A1 + lda, *A3 = A2 + lda;
const int M8 = ((M-1)>>3)<<3, mr = M-M8-1;
const int N4 = (N>>2)<<2, incAn = (lda<<2) - M + 1;
register TYPE m0, m1, m2, m3, x0, y0, y1, y2, y3;
if (M8)
{
for (j=N4; j; j -= 4)
{
y0 = *Y; y1 = *Y1; y2 = *Y2; y3 = *Y3;
x0 = *X; x = X + 1;
m0 = y0 * x0; Y += incy;
m1 = y1 * x0; Y1 += incy;
m2 = y2 * x0; Y2 += incy;
m3 = y3 * x0; Y3 += incy;
for (i=M8; i; i -= 8)
{
x0 = *x;
*A0 += m0; m0 = y0 * x0;
*A1 += m1; m1 = y1 * x0;
*A2 += m2; m2 = y2 * x0;
*A3 += m3; m3 = y3 * x0;
x0 = x[1];
A0[1] += m0; m0 = y0 * x0;
A1[1] += m1; m1 = y1 * x0;
A2[1] += m2; m2 = y2 * x0;
A3[1] += m3; m3 = y3 * x0;
x0 = x[2];
A0[2] += m0; m0 = y0 * x0;
A1[2] += m1; m1 = y1 * x0;
A2[2] += m2; m2 = y2 * x0;
A3[2] += m3; m3 = y3 * x0;
x0 = x[3];
A0[3] += m0; m0 = y0 * x0;
A1[3] += m1; m1 = y1 * x0;
A2[3] += m2; m2 = y2 * x0;
A3[3] += m3; m3 = y3 * x0;
x0 = x[4];
A0[4] += m0; m0 = y0 * x0;
A1[4] += m1; m1 = y1 * x0;
A2[4] += m2; m2 = y2 * x0;
A3[4] += m3; m3 = y3 * x0;
x0 = x[5];
A0[5] += m0; m0 = y0 * x0;
A1[5] += m1; m1 = y1 * x0;
A2[5] += m2; m2 = y2 * x0;
A3[5] += m3; m3 = y3 * x0;
x0 = x[6];
A0[6] += m0; m0 = y0 * x0;
A1[6] += m1; m1 = y1 * x0;
A2[6] += m2; m2 = y2 * x0;
A3[6] += m3; m3 = y3 * x0;
x0 = x[7]; x += 8;
A0[7] += m0; m0 = y0 * x0; A0 += 8;
A1[7] += m1; m1 = y1 * x0; A1 += 8;
A2[7] += m2; m2 = y2 * x0; A2 += 8;
A3[7] += m3; m3 = y3 * x0; A3 += 8;
}
if (mr)
{
for (i=mr; i; i--)
{
x0 = *x++;
*A0++ += m0; m0 = y0 * x0;
*A1++ += m1; m1 = y1 * x0;
*A2++ += m2; m2 = y2 * x0;
*A3++ += m3; m3 = y3 * x0;
}
}
*A0 += m0; A0 += incAn;
*A1 += m1; A1 += incAn;
*A2 += m2; A2 += incAn;
*A3 += m3; A3 += incAn;
}
if (N-N4) ger_Nle4(M, N-N4, X, Y, incY, A0, lda);
}
else ger_Mle8(M, N, X, Y, incY, A, lda);
}
