|
libflame
revision_anchor
|
Functions | |
| FLA_Error | FLA_Fused_Her2_Ax_l_opt_var1 (FLA_Obj beta, FLA_Obj u, FLA_Obj z, FLA_Obj A, FLA_Obj x, FLA_Obj w) |
| FLA_Error | FLA_Fused_Her2_Ax_l_ops_var1 (int m_A, float *buff_beta, float *buff_u, int inc_u, float *buff_z, int inc_z, float *buff_A, int rs_A, int cs_A, float *buff_x, int inc_x, float *buff_w, int inc_w) |
| FLA_Error | FLA_Fused_Her2_Ax_l_opd_var1 (int m_A, double *buff_beta, double *buff_u, int inc_u, double *buff_z, int inc_z, double *buff_A, int rs_A, int cs_A, double *buff_x, int inc_x, double *buff_w, int inc_w) |
| FLA_Error | FLA_Fused_Her2_Ax_l_opc_var1 (int m_A, scomplex *buff_beta, scomplex *buff_u, int inc_u, scomplex *buff_z, int inc_z, scomplex *buff_A, int rs_A, int cs_A, scomplex *buff_x, int inc_x, scomplex *buff_w, int inc_w) |
| FLA_Error | FLA_Fused_Her2_Ax_l_opz_var1 (int m_A, dcomplex *buff_beta, dcomplex *buff_u, int inc_u, dcomplex *buff_z, int inc_z, dcomplex *buff_A, int rs_A, int cs_A, dcomplex *buff_x, int inc_x, dcomplex *buff_w, int inc_w) |
| FLA_Error FLA_Fused_Her2_Ax_l_opc_var1 | ( | int | m_A, |
| scomplex * | buff_beta, | ||
| scomplex * | buff_u, | ||
| int | inc_u, | ||
| scomplex * | buff_z, | ||
| int | inc_z, | ||
| scomplex * | buff_A, | ||
| int | rs_A, | ||
| int | cs_A, | ||
| scomplex * | buff_x, | ||
| int | inc_x, | ||
| scomplex * | buff_w, | ||
| int | inc_w | ||
| ) |
References bli_cdot(), bli_csetv(), BLIS_CONJUGATE, F77_caxpy(), FLA_ZERO, scomplex::imag, and scomplex::real.
Referenced by FLA_Fused_Her2_Ax_l_opt_var1(), and FLA_Tridiag_UT_l_step_ofc_var2().
{
scomplex* buff_0 = FLA_COMPLEX_PTR( FLA_ZERO );
int i;
bli_csetv( m_A,
buff_0,
buff_w, inc_w );
for ( i = 0; i < m_A; ++i )
{
scomplex* alpha11 = buff_A + (i )*cs_A + (i )*rs_A;
scomplex* a21 = buff_A + (i )*cs_A + (i+1)*rs_A;
scomplex* upsilon1 = buff_u + (i )*inc_u;
scomplex* u2 = buff_u + (i+1)*inc_u;
scomplex* zeta1 = buff_z + (i )*inc_z;
scomplex* z2 = buff_z + (i+1)*inc_z;
scomplex* chi1 = buff_x + (i )*inc_x;
scomplex* x2 = buff_x + (i+1)*inc_x;
scomplex* omega1 = buff_w + (i )*inc_w;
scomplex* w2 = buff_w + (i+1)*inc_w;
// scomplex* beta = buff_beta;
scomplex minus_conj_upsilon1;
scomplex minus_conj_zeta1;
scomplex temp;
int m_ahead = m_A - i - 1;
/*------------------------------------------------------------*/
// bli_ccopyconj( zeta1, &conj_zeta1 );
// bli_cmult3( beta, &conj_zeta1, &minus_conj_zeta1 );
// bli_cmult3( &minus_conj_zeta1, upsilon1, &temp );
// bli_cadd3( &temp, alpha11, alpha11 );
//bli_ccopyconj( upsilon1, &conj_upsilon1 );
//bli_cmult3( beta, &conj_upsilon1, &minus_conj_upsilon1 );
//bli_cmult3( &minus_conj_upsilon1, zeta1, &temp );
//bli_cadd3( &temp, alpha11, alpha11 );
minus_conj_zeta1.real = - zeta1->real;
minus_conj_zeta1.imag = - -zeta1->imag;
minus_conj_upsilon1.real = - upsilon1->real;
minus_conj_upsilon1.imag = - -upsilon1->imag;
alpha11->real -= zeta1->real * upsilon1->real - -zeta1->imag * upsilon1->imag +
zeta1->real * upsilon1->real - zeta1->imag * -upsilon1->imag;
alpha11->imag -= -zeta1->imag * upsilon1->real + zeta1->real * upsilon1->imag +
zeta1->imag * upsilon1->real + zeta1->real * -upsilon1->imag;
// bli_caxpyv( BLIS_NO_CONJUGATE,
// m_ahead,
// &minus_conj_zeta1,
// u2, inc_u,
// a21, rs_A );
F77_caxpy( &m_ahead,
&minus_conj_zeta1,
u2, &inc_u,
a21, &rs_A );
// bli_caxpyv( BLIS_NO_CONJUGATE,
// m_ahead,
// &minus_conj_upsilon1,
// z2, inc_z,
// a21, rs_A );
F77_caxpy( &m_ahead,
&minus_conj_upsilon1,
z2, &inc_z,
a21, &rs_A );
// bli_cmult3( alpha11, chi1, &temp );
// bli_cadd3( &temp, omega1, omega1 );
omega1->real += alpha11->real * chi1->real - alpha11->imag * chi1->imag;
omega1->imag += alpha11->imag * chi1->real + alpha11->real * chi1->imag;
bli_cdot( BLIS_CONJUGATE,
m_ahead,
a21, rs_A,
x2, inc_x,
&temp );
// bli_cadd3( &temp, omega1, omega1 );
omega1->real += temp.real;
omega1->imag += temp.imag;
// bli_caxpyv( BLIS_NO_CONJUGATE,
// m_ahead,
// chi1,
// a21, rs_A,
// w2, inc_w );
F77_caxpy( &m_ahead,
chi1,
a21, &rs_A,
w2, &inc_w );
/*------------------------------------------------------------*/
}
return FLA_SUCCESS;
}
| FLA_Error FLA_Fused_Her2_Ax_l_opd_var1 | ( | int | m_A, |
| double * | buff_beta, | ||
| double * | buff_u, | ||
| int | inc_u, | ||
| double * | buff_z, | ||
| int | inc_z, | ||
| double * | buff_A, | ||
| int | rs_A, | ||
| int | cs_A, | ||
| double * | buff_x, | ||
| int | inc_x, | ||
| double * | buff_w, | ||
| int | inc_w | ||
| ) |
References bli_daxpyv2bdotaxpy(), bli_dsetv(), and FLA_ZERO.
Referenced by FLA_Fused_Her2_Ax_l_opt_var1(), and FLA_Tridiag_UT_l_step_ofd_var2().
{
double* buff_0 = FLA_DOUBLE_PTR( FLA_ZERO );
int i;
bli_dsetv( m_A,
buff_0,
buff_w, inc_w );
for ( i = 0; i < m_A; ++i )
{
double* alpha11 = buff_A + (i )*cs_A + (i )*rs_A;
double* a21 = buff_A + (i )*cs_A + (i+1)*rs_A;
double* upsilon1 = buff_u + (i )*inc_u;
double* u2 = buff_u + (i+1)*inc_u;
double* zeta1 = buff_z + (i )*inc_z;
double* z2 = buff_z + (i+1)*inc_z;
double* chi1 = buff_x + (i )*inc_x;
double* x2 = buff_x + (i+1)*inc_x;
double* omega1 = buff_w + (i )*inc_w;
double* w2 = buff_w + (i+1)*inc_w;
// double* beta = buff_beta;
double minus_conj_upsilon1;
double minus_conj_zeta1;
double temp;
int m_ahead = m_A - i - 1;
/*------------------------------------------------------------*/
// bli_dcopyconj( zeta1, &conj_zeta1 );
// bli_dmult3( beta, &conj_zeta1, &minus_conj_zeta1 );
// bli_dmult3( &minus_conj_zeta1, upsilon1, &temp );
// bli_dadd3( &temp, alpha11, alpha11 );
//bli_dcopyconj( upsilon1, &conj_upsilon1 );
//bli_dmult3( beta, &conj_upsilon1, &minus_conj_upsilon1 );
//bli_dmult3( &minus_conj_upsilon1, zeta1, &temp );
//bli_dadd3( &temp, alpha11, alpha11 );
minus_conj_zeta1 = - *zeta1;
minus_conj_upsilon1 = - *upsilon1;
*alpha11 -= 2.0 * *zeta1 * *upsilon1;
// bli_dmult3( alpha11, chi1, &temp );
// bli_dadd3( &temp, omega1, omega1 );
*omega1 += *alpha11 * *chi1;
bli_daxpyv2bdotaxpy( m_ahead,
&minus_conj_zeta1,
u2, inc_u,
&minus_conj_upsilon1,
z2, inc_z,
a21, rs_A,
x2, inc_x,
chi1,
&temp,
w2, inc_w );
// bli_dadd3( &temp, omega1, omega1 );
*omega1 += temp;
/*------------------------------------------------------------*/
}
return FLA_SUCCESS;
}
| FLA_Error FLA_Fused_Her2_Ax_l_ops_var1 | ( | int | m_A, |
| float * | buff_beta, | ||
| float * | buff_u, | ||
| int | inc_u, | ||
| float * | buff_z, | ||
| int | inc_z, | ||
| float * | buff_A, | ||
| int | rs_A, | ||
| int | cs_A, | ||
| float * | buff_x, | ||
| int | inc_x, | ||
| float * | buff_w, | ||
| int | inc_w | ||
| ) |
References bli_ssetv(), F77_saxpy(), F77_sdot(), and FLA_ZERO.
Referenced by FLA_Fused_Her2_Ax_l_opt_var1(), and FLA_Tridiag_UT_l_step_ofs_var2().
{
float* buff_0 = FLA_FLOAT_PTR( FLA_ZERO );
int i;
bli_ssetv( m_A,
buff_0,
buff_w, inc_w );
for ( i = 0; i < m_A; ++i )
{
float* alpha11 = buff_A + (i )*cs_A + (i )*rs_A;
float* a21 = buff_A + (i )*cs_A + (i+1)*rs_A;
float* upsilon1 = buff_u + (i )*inc_u;
float* u2 = buff_u + (i+1)*inc_u;
float* zeta1 = buff_z + (i )*inc_z;
float* z2 = buff_z + (i+1)*inc_z;
float* chi1 = buff_x + (i )*inc_x;
float* x2 = buff_x + (i+1)*inc_x;
float* omega1 = buff_w + (i )*inc_w;
float* w2 = buff_w + (i+1)*inc_w;
// float* beta = buff_beta;
float minus_conj_upsilon1;
float minus_conj_zeta1;
float temp;
int m_ahead = m_A - i - 1;
/*------------------------------------------------------------*/
// bli_scopyconj( zeta1, &conj_zeta1 );
// bli_smult3( beta, &conj_zeta1, &minus_conj_zeta1 );
// bli_smult3( &minus_conj_zeta1, upsilon1, &temp );
// bli_sadd3( &temp, alpha11, alpha11 );
//bli_scopyconj( upsilon1, &conj_upsilon1 );
//bli_smult3( beta, &conj_upsilon1, &minus_conj_upsilon1 );
//bli_smult3( &minus_conj_upsilon1, zeta1, &temp );
//bli_sadd3( &temp, alpha11, alpha11 );
minus_conj_zeta1 = - *zeta1;
minus_conj_upsilon1 = - *upsilon1;
*alpha11 -= 2.0F * *zeta1 * *upsilon1;
// bli_saxpyv( BLIS_NO_CONJUGATE,
// m_ahead,
// &minus_conj_zeta1,
// u2, inc_u,
// a21, rs_A );
F77_saxpy( &m_ahead,
&minus_conj_zeta1,
u2, &inc_u,
a21, &rs_A );
// bli_saxpyv( BLIS_NO_CONJUGATE,
// m_ahead,
// &minus_conj_upsilon1,
// z2, inc_z,
// a21, rs_A );
F77_saxpy( &m_ahead,
&minus_conj_upsilon1,
z2, &inc_z,
a21, &rs_A );
// bli_smult3( alpha11, chi1, &temp );
// bli_sadd3( &temp, omega1, omega1 );
*omega1 += *alpha11 * *chi1;
// bli_sdot( BLIS_CONJUGATE,
// m_ahead,
// a21, rs_A,
// x2, inc_x,
// &temp );
temp = F77_sdot( &m_ahead,
a21, &rs_A,
x2, &inc_x );
// bli_sadd3( &temp, omega1, omega1 );
*omega1 += temp;
// bli_saxpyv( BLIS_NO_CONJUGATE,
// m_ahead,
// chi1,
// a21, rs_A,
// w2, inc_w );
F77_saxpy( &m_ahead,
chi1,
a21, &rs_A,
w2, &inc_w );
/*------------------------------------------------------------*/
}
return FLA_SUCCESS;
}
| FLA_Error FLA_Fused_Her2_Ax_l_opt_var1 | ( | FLA_Obj | beta, |
| FLA_Obj | u, | ||
| FLA_Obj | z, | ||
| FLA_Obj | A, | ||
| FLA_Obj | x, | ||
| FLA_Obj | w | ||
| ) |
References FLA_Fused_Her2_Ax_l_opc_var1(), FLA_Fused_Her2_Ax_l_opd_var1(), FLA_Fused_Her2_Ax_l_ops_var1(), FLA_Fused_Her2_Ax_l_opz_var1(), FLA_Obj_col_stride(), FLA_Obj_datatype(), FLA_Obj_length(), FLA_Obj_row_stride(), and FLA_Obj_vector_inc().
{
/*
Effective computation:
A = A + beta * ( u * z' + z * u' );
w = A * x;
*/
FLA_Datatype datatype;
int m_A;
int rs_A, cs_A;
int inc_u, inc_z, inc_x, inc_w;
datatype = FLA_Obj_datatype( A );
m_A = FLA_Obj_length( A );
rs_A = FLA_Obj_row_stride( A );
cs_A = FLA_Obj_col_stride( A );
inc_u = FLA_Obj_vector_inc( u );
inc_z = FLA_Obj_vector_inc( z );
inc_x = FLA_Obj_vector_inc( x );
inc_w = FLA_Obj_vector_inc( w );
switch ( datatype )
{
case FLA_FLOAT:
{
float* buff_A = FLA_FLOAT_PTR( A );
float* buff_u = FLA_FLOAT_PTR( u );
float* buff_z = FLA_FLOAT_PTR( z );
float* buff_x = FLA_FLOAT_PTR( x );
float* buff_w = FLA_FLOAT_PTR( w );
float* buff_beta = FLA_FLOAT_PTR( beta );
FLA_Fused_Her2_Ax_l_ops_var1( m_A,
buff_beta,
buff_u, inc_u,
buff_z, inc_z,
buff_A, rs_A, cs_A,
buff_x, inc_x,
buff_w, inc_w );
break;
}
case FLA_DOUBLE:
{
double* buff_A = FLA_DOUBLE_PTR( A );
double* buff_u = FLA_DOUBLE_PTR( u );
double* buff_z = FLA_DOUBLE_PTR( z );
double* buff_x = FLA_DOUBLE_PTR( x );
double* buff_w = FLA_DOUBLE_PTR( w );
double* buff_beta = FLA_DOUBLE_PTR( beta );
FLA_Fused_Her2_Ax_l_opd_var1( m_A,
buff_beta,
buff_u, inc_u,
buff_z, inc_z,
buff_A, rs_A, cs_A,
buff_x, inc_x,
buff_w, inc_w );
break;
}
case FLA_COMPLEX:
{
scomplex* buff_A = FLA_COMPLEX_PTR( A );
scomplex* buff_u = FLA_COMPLEX_PTR( u );
scomplex* buff_z = FLA_COMPLEX_PTR( z );
scomplex* buff_x = FLA_COMPLEX_PTR( x );
scomplex* buff_w = FLA_COMPLEX_PTR( w );
scomplex* buff_beta = FLA_COMPLEX_PTR( beta );
FLA_Fused_Her2_Ax_l_opc_var1( m_A,
buff_beta,
buff_u, inc_u,
buff_z, inc_z,
buff_A, rs_A, cs_A,
buff_x, inc_x,
buff_w, inc_w );
break;
}
case FLA_DOUBLE_COMPLEX:
{
dcomplex* buff_A = FLA_DOUBLE_COMPLEX_PTR( A );
dcomplex* buff_u = FLA_DOUBLE_COMPLEX_PTR( u );
dcomplex* buff_z = FLA_DOUBLE_COMPLEX_PTR( z );
dcomplex* buff_x = FLA_DOUBLE_COMPLEX_PTR( x );
dcomplex* buff_w = FLA_DOUBLE_COMPLEX_PTR( w );
dcomplex* buff_beta = FLA_DOUBLE_COMPLEX_PTR( beta );
FLA_Fused_Her2_Ax_l_opz_var1( m_A,
buff_beta,
buff_u, inc_u,
buff_z, inc_z,
buff_A, rs_A, cs_A,
buff_x, inc_x,
buff_w, inc_w );
break;
}
}
return FLA_SUCCESS;
}
| FLA_Error FLA_Fused_Her2_Ax_l_opz_var1 | ( | int | m_A, |
| dcomplex * | buff_beta, | ||
| dcomplex * | buff_u, | ||
| int | inc_u, | ||
| dcomplex * | buff_z, | ||
| int | inc_z, | ||
| dcomplex * | buff_A, | ||
| int | rs_A, | ||
| int | cs_A, | ||
| dcomplex * | buff_x, | ||
| int | inc_x, | ||
| dcomplex * | buff_w, | ||
| int | inc_w | ||
| ) |
References bli_z0(), bli_zaxpyv2b(), bli_zdotaxpy(), bli_zsetv(), dcomplex::imag, and dcomplex::real.
Referenced by FLA_Fused_Her2_Ax_l_opt_var1(), and FLA_Tridiag_UT_l_step_ofz_var2().
{
dcomplex zero = bli_z0();
int i;
bli_zsetv( m_A,
&zero,
buff_w, inc_w );
for ( i = 0; i < m_A; ++i )
{
dcomplex* restrict alpha11 = buff_A + (i )*cs_A + (i )*rs_A;
dcomplex* restrict a21 = buff_A + (i )*cs_A + (i+1)*rs_A;
dcomplex* restrict upsilon1 = buff_u + (i )*inc_u;
dcomplex* restrict u2 = buff_u + (i+1)*inc_u;
dcomplex* restrict zeta1 = buff_z + (i )*inc_z;
dcomplex* restrict z2 = buff_z + (i+1)*inc_z;
dcomplex* restrict chi1 = buff_x + (i )*inc_x;
dcomplex* restrict x2 = buff_x + (i+1)*inc_x;
dcomplex* restrict omega1 = buff_w + (i )*inc_w;
dcomplex* restrict w2 = buff_w + (i+1)*inc_w;
//dcomplex* restrict beta = buff_beta;
dcomplex minus_conj_upsilon1;
dcomplex minus_conj_zeta1;
dcomplex temp;
dcomplex ze1;
dcomplex up1;
dcomplex a11;
dcomplex om1;
dcomplex ch1;
int m_ahead = m_A - i - 1;
/*------------------------------------------------------------*/
// bli_zcopyconj( zeta1, &conj_zeta1 );
// bli_zmult3( beta, &conj_zeta1, &minus_conj_zeta1 );
// bli_zmult3( &minus_conj_zeta1, upsilon1, &temp );
// bli_zadd3( &temp, alpha11, alpha11 );
//bli_zcopyconj( upsilon1, &conj_upsilon1 );
//bli_zmult3( beta, &conj_upsilon1, &minus_conj_upsilon1 );
//bli_zmult3( &minus_conj_upsilon1, zeta1, &temp );
//bli_zadd3( &temp, alpha11, alpha11 );
minus_conj_zeta1.real = - zeta1->real;
minus_conj_zeta1.imag = - -zeta1->imag;
minus_conj_upsilon1.real = - upsilon1->real;
minus_conj_upsilon1.imag = - -upsilon1->imag;
ze1 = *zeta1;
up1 = *upsilon1;
a11 = *alpha11;
om1 = *omega1;
ch1 = *chi1;
//alpha11->real -= zeta1->real * upsilon1->real - -zeta1->imag * upsilon1->imag +
// zeta1->real * upsilon1->real - zeta1->imag * -upsilon1->imag;
//alpha11->imag -= -zeta1->imag * upsilon1->real + zeta1->real * upsilon1->imag +
// zeta1->imag * upsilon1->real + zeta1->real * -upsilon1->imag;
a11.real -= ze1.real * up1.real - -ze1.imag * up1.imag +
up1.real * ze1.real - -up1.imag * ze1.imag;
a11.imag -= ze1.real * up1.imag + -ze1.imag * up1.real +
up1.real * ze1.imag + -up1.imag * ze1.real;
// bli_zmult3( alpha11, chi1, &temp );
// bli_zadd3( &temp, omega1, omega1 );
//omega1->real += alpha11->real * chi1->real - alpha11->imag * chi1->imag;
//omega1->imag += alpha11->imag * chi1->real + alpha11->real * chi1->imag;
om1.real += a11.real * ch1.real - a11.imag * ch1.imag;
om1.imag += a11.imag * ch1.real + a11.real * ch1.imag;
*alpha11 = a11;
*omega1 = om1;
/*
bli_zaxpyv2bdotaxpy( m_ahead,
&minus_conj_zeta1,
u2, inc_u,
&minus_conj_upsilon1,
z2, inc_z,
a21, rs_A,
x2, inc_x,
chi1,
&temp,
w2, inc_w );
*/
bli_zaxpyv2b( m_ahead,
&minus_conj_zeta1,
&minus_conj_upsilon1,
u2, inc_u,
z2, inc_z,
a21, rs_A );
bli_zdotaxpy( m_ahead,
a21, rs_A,
x2, inc_x,
chi1,
&temp,
w2, inc_w );
// bli_zadd3( &temp, omega1, omega1 );
omega1->real += temp.real;
omega1->imag += temp.imag;
/*------------------------------------------------------------*/
}
return FLA_SUCCESS;
}
1.7.6.1