static void gemvT2x16(const int M, const int N, const TYPE *A, const int lda,
const TYPE *X, const SCALAR beta, TYPE *Y)
{
int i;
const int M2 = (M>>1)<<1, N16 = (N>>4)<<4;
const int incAm = (N16 ? (lda<<1) - N16 + 16 : (lda<<1));
const int nr = (N16 ? N-N16+16 : N);
TYPE *stY = Y + M2;
const TYPE *stX = X + N16-16, *x;
const TYPE *A0 = A, *A1 = A + lda;
register TYPE y0, y1, x0, x1;
#ifndef ATL_PFD
#define ATL_PFD 12
#endif
if (M2)
{
do
{
#ifdef BETA0
y0 = y1 = ATL_rzero;
#elif defined(BETAX)
y0 = *Y * beta; y1 = beta * Y[1];
#else
y0 = *Y; y1 = Y[1];
#endif
x = X;
if (N16 > 16)
{
x0 = *x; x1 = x[1];
do
{
y0 += x0 * *A0; ATL_pfl1R(A0+ATL_PFD); ATL_pfl1R(A1+ATL_PFD);
ATL_pfl1R(A0+ATL_PFD+8); ATL_pfl1R(A1+ATL_PFD+8);
y1 += x0 * *A1; x0 = x[2];
y0 += x1 * A0[1];
y1 += x1 * A1[1]; x1 = x[3];
y0 += x0 * A0[2];
y1 += x0 * A1[2]; x0 = x[4];
y0 += x1 * A0[3];
y1 += x1 * A1[3]; x1 = x[5];
y0 += x0 * A0[4];
y1 += x0 * A1[4]; x0 = x[6];
y0 += x1 * A0[5];
y1 += x1 * A1[5]; x1 = x[7];
y0 += x0 * A0[6];
y1 += x0 * A1[6]; x0 = x[8];
y0 += x1 * A0[7];
y1 += x1 * A1[7]; x1 = x[9];
y0 += x0 * A0[8];
ATL_pfl1R(A0+ATL_PFD+4); ATL_pfl1R(A1+ATL_PFD+4);
ATL_pfl1R(A0+ATL_PFD+12); ATL_pfl1R(A1+ATL_PFD+12);
y1 += x0 * A1[8]; x0 = x[10];
y0 += x1 * A0[9];
y1 += x1 * A1[9]; x1 = x[11];
y0 += x0 * A0[10];
y1 += x0 * A1[10]; x0 = x[12];
y0 += x1 * A0[11];
y1 += x1 * A1[11]; x1 = x[13];
y0 += x0 * A0[12];
y1 += x0 * A1[12]; x0 = x[14];
y0 += x1 * A0[13];
y1 += x1 * A1[13]; x1 = x[15];
y0 += x0 * A0[14]; x += 16;
y1 += x0 * A1[14]; x0 = *x;
y0 += x1 * A0[15]; A0 += 16;
y1 += x1 * A1[15]; x1 = x[1]; A1 += 16;
}
while(x != stX);
}
for (i=0; i != nr; i++)
{
x0 = x[i];
y0 += x0 * A0[i];
y1 += x0 * A1[i];
}
A0 += incAm;
*Y = y0;
A1 += incAm;
Y[1] = y1;
Y += 2;
}
while (Y != stY);
}
if (M-M2) gemvT_Msmall(M-M2, N, A0, lda, X, beta, Y);
}
static void gemvT4x4(const int M, const int N, const TYPE *A, const int lda,
const TYPE *X, const SCALAR beta, TYPE *Y)
{
const int M4 = (M>>2)<<2, N4 = (N>>2)<<2;
const int nr = N-N4, incAm = (lda<<2) - N4;
const TYPE *stX = X + N4, *x;
const TYPE *A0 = A, *A1 = A + lda, *A2 = A1 + lda, *A3 = A2 + lda;
TYPE *stY = Y + M4;
register TYPE a00, a10, a20, a30, a01, a11, a21, a31;
register TYPE a02, a12, a22, a32, a03, a13, a23, a33;
register TYPE y0, y1, y2, y3, yy0, yy1, yy2, yy3;
register TYPE x0, x1, x2, x3;
if (N4 > 4)
{
if (M4)
{
do
{
a00 = *A0; a01 = *A1; a02 = *A2; a03 = *A3;
a10 = A0[1]; a11 = A1[1]; a12 = A2[1]; a13 = A3[1];
a20 = A0[2]; a21 = A1[2]; a22 = A2[2]; a23 = A3[2];
a30 = A0[3]; a31 = A1[3]; a32 = A2[3]; a33 = A3[3];
A0 += 4; A1 += 4; A2 += 4; A3 += 4;
#ifdef BETA0
y0 = y1 = y2 = y3 = yy0 = yy1 = yy2 = yy3 = ATL_rzero;
#elif defined(BETAX)
y0 = beta;
yy0 = *Y; yy1 = Y[1]; yy2 = Y[2]; yy3 = Y[3];
yy0 *= y0; yy1 *= y0; yy2 *= y0; yy3 *= y0;
y0 = y1 = y2 = y3 = ATL_rzero;
#else
yy0 = *Y; yy1 = Y[1]; yy2 = Y[2]; yy3 = Y[3];
y0 = y1 = y2 = y3 = ATL_rzero;
#endif
x0 = *X; x1 = X[1]; x2 = X[2]; x3 = X[3]; x = X + 4;
do
{
y0 += a00 * x0; a00 = *A0;
y1 += a01 * x0; a01 = *A1;
yy0 += a10 * x1; a10 = A0[1];
yy1 += a11 * x1; a11 = A1[1];
y2 += a02 * x0; a02 = *A2;
y3 += a03 * x0; a03 = *A3; x0 = *x;
yy2 += a12 * x1; a12 = A2[1];
yy3 += a13 * x1; a13 = A3[1]; x1 = x[1];
y0 += a20 * x2; a20 = A0[2];
y1 += a21 * x2; a21 = A1[2];
yy0 += a30 * x3; a30 = A0[3]; A0 += 4;
yy1 += a31 * x3; a31 = A1[3]; A1 += 4;
y2 += a22 * x2; a22 = A2[2];
y3 += a23 * x2; a23 = A3[2]; x2 = x[2];
yy2 += a32 * x3; a32 = A2[3]; A2 += 4;
yy3 += a33 * x3; a33 = A3[3]; A3 += 4; x3 = x[3]; x += 4;
}
while (x != stX);
y0 += a00 * x0;
y1 += a01 * x0;
yy0 += a10 * x1;
yy1 += a11 * x1;
y2 += a02 * x0;
y3 += a03 * x0;
yy2 += a12 * x1;
yy3 += a13 * x1;
y0 += a20 * x2;
y1 += a21 * x2;
yy0 += a30 * x3;
yy1 += a31 * x3;
y2 += a22 * x2;
y3 += a23 * x2;
yy2 += a32 * x3;
yy3 += a33 * x3;
switch(nr)
{
case 1:
x0 = *x;
y0 += *A0 * x0;
y1 += *A1 * x0;
y2 += *A2 * x0;
y3 += *A3 * x0;
break;
case 2:
x0 = *x;
x1 = x[1];
y0 += *A0 * x0;
y1 += *A1 * x0;
yy0 += A0[1] * x1;
yy1 += A1[1] * x1;
y2 += *A2 * x0;
y3 += *A3 * x0;
yy2 += A2[1] * x1;
yy3 += A3[1] * x1;
break;
case 3:
x0 = *x;
x1 = x[1];
x2 = x[2];
y0 += *A0 * x0;
y1 += *A1 * x0;
yy0 += A0[1] * x1;
yy1 += A1[1] * x1;
y2 += *A2 * x0;
y3 += *A3 * x0;
yy2 += A2[1] * x1;
yy3 += A3[1] * x1;
y0 += A0[2] * x2;
y1 += A1[2] * x2;
y2 += A2[2] * x2;
y3 += A3[2] * x2;
break;
default:;
}
A0 += incAm;
y0 += yy0;
A1 += incAm;
y1 += yy1;
A2 += incAm;
y2 += yy2;
A3 += incAm;
y3 += yy3;
*Y = y0;
Y[1] = y1;
Y[2] = y2;
Y[3] = y3;
Y += 4;
}
while (Y != stY);
}
if (M-M4) gemvT_Msmall(M-M4, N, A0, lda, X, beta, Y);
}
else if (M) gemvT_Nsmall(M, N, A, lda, X, beta, Y);
}
