amd/blis
1
0
Fork 0
mirror of https://github.com/amd/blis.git synced 2026-04-20 07:38:53 +00:00
Code Issues Packages Projects Releases Wiki Activity

Handle edge cases, zero-filling in packm kernels.

Browse Source 
Details:
- Updated the API and semantics of packm kernels such that they must now
  handle edge cases, meaning that a c-by-k packm kernel must be able to
  pack edge cases that are fewer than c rows/columns and be able to
  zero-fill the remaining elements. They must also be able to zero-fill
  the equivalent region when copying fewer than k columns/rows (which is
  needed by trsm). The new packm kernel API is generally:

    void packm_kernel
         (
           conj_t           conja,
           dim_t            cdim,
           dim_t            n,
           dim_t            n_max,
           ctype*  restrict kappa,
           ctype*  restrict a, inc_t inca, inc_t lda,
           ctype*  restrict p,             inc_t ldp,
           cntx_t* restrict cntx
         );

  where cdim and n are the dimensions (short and long, respectively) of
  the submatrix being copied from the source matrix A, and n_max is the
  "full" long dimension (corresponding to the k dimension in gemm) of
  the micropanel. The "full" short dimension (corresponding to the
  register blocksize MR or NR) is not part of the API because it is
  known intrinsically by the packm kernel implementation. Thanks to
  Devin Matthews for prompting us to make this change (#282).
- Updated all reference packm kernels in ref_kernels/1m according to
  above changes, as well as all optimized packm kernels (which only
  consisted of those for knl).
- Bumped the major soname version number in 'so_version' to 2. At first
  I was considering leaving it unchanged, but I couldn't escape the
  reality that the packm kernel API is much closer to an expert API
  than it is some obscure helper function interface within the framework
  that nobody would ever notice.
- Removed reference packm kernels for mr/nr = 30. The only sub-config
  that would have been using those kernels is knc, which is likely no
  longer being used by very many people (if any). (This also mostly
  offset the larger object code footprint incurred by moving the edge-
  case handling into the individual packm kernels.)
- Fixed an obscure race condition for 3mh and 4mh induced methods in
  which those implementations were modifying the contexts stored in the
  gks rather than a local copy.
- Fixed a minor bug in the testsuite that prevented non-1m-based induced
  method implementations of trsm from executing.
This commit is contained in:
Field G. Van Zee
2018-12-12 15:22:59 -06:00
parent c534da62c0
commit f808d829c5
43 changed files with 9815 additions and 7105 deletions
Download Patch File Download Diff File Expand all files Collapse all files

202
frame/include/level0/1m/bli_scal21ms_mxn.h Normal file
View File

@@ -0,0 +1,202 @@
/*
BLIS
An object-based framework for developing high-performance BLAS-like
libraries.
Copyright (C) 2014, The University of Texas at Austin
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are
met:
- Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
- Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in the
documentation and/or other materials provided with the distribution.
- Neither the name(s) of the copyright holder(s) nor the names of its
contributors may be used to endorse or promote products derived
from this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#ifndef BLIS_SCAL21MS_MXN_H
#define BLIS_SCAL21MS_MXN_H
// scal21ms_mxn
static void bli_cscal21ms_mxn
(
const pack_t schema,
const conj_t conjx,
const dim_t m,
const dim_t n,
scomplex* restrict alpha,
scomplex* restrict x, const inc_t rs_x, const inc_t cs_x,
scomplex* restrict y, const inc_t rs_y, const inc_t cs_y, const inc_t ld_y
)
{
dim_t i, j;
/* Handle 1e and 1r separately. */
if ( bli_is_1e_packed( schema ) )
{
scomplex* restrict y_ri = y;
scomplex* restrict y_ir = y + ld_y/2;
if ( bli_is_conj( conjx ) )
{
for ( j = 0; j < n; ++j )
for ( i = 0; i < m; ++i )
{
bli_cscal2j1es( *(alpha),
*(x + i*rs_x + j*cs_x),
*(y_ri + i*rs_y + j*cs_y),
*(y_ir + i*rs_y + j*cs_y) );
}
}
else /* if ( bli_is_noconj( conjx ) ) */
{
for ( j = 0; j < n; ++j )
for ( i = 0; i < m; ++i )
{
bli_cscal21es( *(alpha),
*(x + i*rs_x + j*cs_x),
*(y_ri + i*rs_y + j*cs_y),
*(y_ir + i*rs_y + j*cs_y) );
}
}
}
else /* if ( bli_is_1r_packed( schema ) ) */
{
inc_t rs_y2 = rs_y;
inc_t cs_y2 = cs_y;
/* Scale the non-unit stride by two for the 1r loop, which steps
in units of real (not complex) values. */
if ( rs_y2 == 1 ) { cs_y2 *= 2; }
else /* if ( cs_y2 == 1 ) */ { rs_y2 *= 2; }
float* restrict y_cast = ( float* )y;
float* restrict y_r = y_cast;
float* restrict y_i = y_cast + ld_y;
if ( bli_is_conj( conjx ) )
{
for ( j = 0; j < n; ++j )
for ( i = 0; i < m; ++i )
{
bli_cscal2j1rs( *(alpha),
*(x + i*rs_x + j*cs_x ),
*(y_r + i*rs_y2 + j*cs_y2),
*(y_i + i*rs_y2 + j*cs_y2) );
}
}
else /* if ( bli_is_noconj( conjx ) ) */
{
for ( j = 0; j < n; ++j )
for ( i = 0; i < m; ++i )
{
bli_cscal21rs( *(alpha),
*(x + i*rs_x + j*cs_x ),
*(y_r + i*rs_y2 + j*cs_y2),
*(y_i + i*rs_y2 + j*cs_y2) );
}
}
}
}
static void bli_zscal21ms_mxn
(
const pack_t schema,
const conj_t conjx,
const dim_t m,
const dim_t n,
dcomplex* restrict alpha,
dcomplex* restrict x, const inc_t rs_x, const inc_t cs_x,
dcomplex* restrict y, const inc_t rs_y, const inc_t cs_y, const inc_t ld_y
)
{
dim_t i, j;
/* Handle 1e and 1r separately. */
if ( bli_is_1e_packed( schema ) )
{
dcomplex* restrict y_ri = y;
dcomplex* restrict y_ir = y + ld_y/2;
if ( bli_is_conj( conjx ) )
{
for ( j = 0; j < n; ++j )
for ( i = 0; i < m; ++i )
{
bli_zscal2j1es( *(alpha),
*(x + i*rs_x + j*cs_x),
*(y_ri + i*rs_y + j*cs_y),
*(y_ir + i*rs_y + j*cs_y) );
}
}
else /* if ( bli_is_noconj( conjx ) ) */
{
for ( j = 0; j < n; ++j )
for ( i = 0; i < m; ++i )
{
bli_zscal21es( *(alpha),
*(x + i*rs_x + j*cs_x),
*(y_ri + i*rs_y + j*cs_y),
*(y_ir + i*rs_y + j*cs_y) );
}
}
}
else /* if ( bli_is_1r_packed( schema ) ) */
{
inc_t rs_y2 = rs_y;
inc_t cs_y2 = cs_y;
/* Scale the non-unit stride by two for the 1r loop, which steps
in units of real (not complex) values. */
if ( rs_y2 == 1 ) { cs_y2 *= 2; }
else /* if ( cs_y2 == 1 ) */ { rs_y2 *= 2; }
double* restrict y_cast = ( double* )y;
double* restrict y_r = y_cast;
double* restrict y_i = y_cast + ld_y;
if ( bli_is_conj( conjx ) )
{
for ( j = 0; j < n; ++j )
for ( i = 0; i < m; ++i )
{
bli_zscal2j1rs( *(alpha),
*(x + i*rs_x + j*cs_x ),
*(y_r + i*rs_y2 + j*cs_y2),
*(y_i + i*rs_y2 + j*cs_y2) );
}
}
else /* if ( bli_is_noconj( conjx ) ) */
{
for ( j = 0; j < n; ++j )
for ( i = 0; i < m; ++i )
{
bli_zscal21rs( *(alpha),
*(x + i*rs_x + j*cs_x ),
*(y_r + i*rs_y2 + j*cs_y2),
*(y_i + i*rs_y2 + j*cs_y2) );
}
}
}
}
#endif

246
frame/include/level0/1m/bli_set1ms_mxn.h
View File

@@ -49,128 +49,146 @@
components of packm. */ \
}
#define bli_cset1ms_mxn( schema, offm, offn, m, n, a, y, rs_y, cs_y, ld_y ) \
{ \
inc_t offm_local = offm; \
inc_t offn_local = offn; \
dim_t m_local = m; \
dim_t n_local = n; \
inc_t rs_y1 = rs_y; \
inc_t cs_y1 = cs_y; \
inc_t rs_y2 = rs_y; \
inc_t cs_y2 = cs_y; \
dim_t i, j; \
\
static void bli_cset1ms_mxn
(
const pack_t schema,
const dim_t offm,
const dim_t offn,
const dim_t m,
const dim_t n,
scomplex* restrict alpha,
scomplex* restrict y, const inc_t rs_y, const inc_t cs_y, const inc_t ld_y
)
{
inc_t offm_local = offm;
inc_t offn_local = offn;
dim_t m_local = m;
dim_t n_local = n;
inc_t rs_y1 = rs_y;
inc_t cs_y1 = cs_y;
inc_t rs_y2 = rs_y;
inc_t cs_y2 = cs_y;
dim_t i, j;
/* Optimization: The loops walk through y with unit stride if y is
column-stored. If y is row-stored, swap the dimensions and strides
to preserve unit stride movement. */ \
if ( cs_y == 1 ) \
{ \
bli_swap_incs( &offm_local, &offn_local ); \
bli_swap_dims( &m_local, &n_local ); \
bli_swap_incs( &rs_y1, &cs_y1 ); \
bli_swap_incs( &rs_y2, &cs_y2 ); \
} \
\
/* Handle 1e and 1r separately. */ \
if ( bli_is_1e_packed( schema ) ) \
{ \
scomplex* restrict y_off_ri = y + (offm_local )*rs_y1 \
+ (offn_local )*cs_y1; \
scomplex* restrict y_off_ir = y + (offm_local )*rs_y1 \
+ (offn_local )*cs_y1 + ld_y/2; \
\
for ( j = 0; j < n_local; ++j ) \
for ( i = 0; i < m_local; ++i ) \
{ \
bli_ccopy1es( *(a), \
*(y_off_ri + i*rs_y1 + j*cs_y1), \
*(y_off_ir + i*rs_y1 + j*cs_y1) ); \
} \
} \
else /* if ( bli_is_1r_packed( schema ) ) */ \
{ \
to preserve unit stride movement. */
if ( cs_y == 1 )
{
bli_swap_incs( &offm_local, &offn_local );
bli_swap_dims( &m_local, &n_local );
bli_swap_incs( &rs_y1, &cs_y1 );
bli_swap_incs( &rs_y2, &cs_y2 );
}
/* Handle 1e and 1r separately. */
if ( bli_is_1e_packed( schema ) )
{
scomplex* restrict y_off_ri = y + (offm_local )*rs_y1
+ (offn_local )*cs_y1;
scomplex* restrict y_off_ir = y + (offm_local )*rs_y1
+ (offn_local )*cs_y1 + ld_y/2;
for ( j = 0; j < n_local; ++j )
for ( i = 0; i < m_local; ++i )
{
bli_ccopy1es( *(alpha),
*(y_off_ri + i*rs_y1 + j*cs_y1),
*(y_off_ir + i*rs_y1 + j*cs_y1) );
}
}
else /* if ( bli_is_1r_packed( schema ) ) */
{
/* Scale the non-unit stride by two for the 1r loop, which steps
in units of real (not complex) values. */ \
if ( rs_y2 == 1 ) { cs_y2 *= 2; } \
else /* if ( cs_y2 == 1 ) */ { rs_y2 *= 2; } \
\
float* restrict y_cast = ( float* )y; \
float* restrict y_off_r = y_cast + (offm_local )*rs_y2 \
+ (offn_local )*cs_y2; \
float* restrict y_off_i = y_cast + (offm_local )*rs_y2 \
+ (offn_local )*cs_y2 + ld_y; \
\
for ( j = 0; j < n_local; ++j ) \
for ( i = 0; i < m_local; ++i ) \
{ \
bli_ccopy1rs( *(a), \
*(y_off_r + i*rs_y2 + j*cs_y2), \
*(y_off_i + i*rs_y2 + j*cs_y2) ); \
} \
} \
in units of real (not complex) values. */
if ( rs_y2 == 1 ) { cs_y2 *= 2; }
else /* if ( cs_y2 == 1 ) */ { rs_y2 *= 2; }
float* restrict y_cast = ( float* )y;
float* restrict y_off_r = y_cast + (offm_local )*rs_y2
+ (offn_local )*cs_y2;
float* restrict y_off_i = y_cast + (offm_local )*rs_y2
+ (offn_local )*cs_y2 + ld_y;
for ( j = 0; j < n_local; ++j )
for ( i = 0; i < m_local; ++i )
{
bli_ccopy1rs( *(alpha),
*(y_off_r + i*rs_y2 + j*cs_y2),
*(y_off_i + i*rs_y2 + j*cs_y2) );
}
}
}
#define bli_zset1ms_mxn( schema, offm, offn, m, n, a, y, rs_y, cs_y, ld_y ) \
{ \
inc_t offm_local = offm; \
inc_t offn_local = offn; \
dim_t m_local = m; \
dim_t n_local = n; \
inc_t rs_y1 = rs_y; \
inc_t cs_y1 = cs_y; \
inc_t rs_y2 = rs_y; \
inc_t cs_y2 = cs_y; \
dim_t i, j; \
\
static void bli_zset1ms_mxn
(
const pack_t schema,
const dim_t offm,
const dim_t offn,
const dim_t m,
const dim_t n,
dcomplex* restrict alpha,
dcomplex* restrict y, const inc_t rs_y, const inc_t cs_y, const inc_t ld_y
)
{
inc_t offm_local = offm;
inc_t offn_local = offn;
dim_t m_local = m;
dim_t n_local = n;
inc_t rs_y1 = rs_y;
inc_t cs_y1 = cs_y;
inc_t rs_y2 = rs_y;
inc_t cs_y2 = cs_y;
dim_t i, j;
/* Optimization: The loops walk through y with unit stride if y is
column-stored. If y is row-stored, swap the dimensions and strides
to preserve unit stride movement. */ \
if ( cs_y == 1 ) \
{ \
bli_swap_incs( &offm_local, &offn_local ); \
bli_swap_dims( &m_local, &n_local ); \
bli_swap_incs( &rs_y1, &cs_y1 ); \
bli_swap_incs( &rs_y2, &cs_y2 ); \
} \
\
/* Handle 1e and 1r separately. */ \
if ( bli_is_1e_packed( schema ) ) \
{ \
dcomplex* restrict y_off_ri = y + (offm_local )*rs_y1 \
+ (offn_local )*cs_y1; \
dcomplex* restrict y_off_ir = y + (offm_local )*rs_y1 \
+ (offn_local )*cs_y1 + ld_y/2; \
\
for ( j = 0; j < n_local; ++j ) \
for ( i = 0; i < m_local; ++i ) \
{ \
bli_zcopy1es( *(a), \
*(y_off_ri + i*rs_y1 + j*cs_y1), \
*(y_off_ir + i*rs_y1 + j*cs_y1) ); \
} \
} \
else /* if ( bli_is_1r_packed( schema ) ) */ \
{ \
to preserve unit stride movement. */
if ( cs_y == 1 )
{
bli_swap_incs( &offm_local, &offn_local );
bli_swap_dims( &m_local, &n_local );
bli_swap_incs( &rs_y1, &cs_y1 );
bli_swap_incs( &rs_y2, &cs_y2 );
}
/* Handle 1e and 1r separately. */
if ( bli_is_1e_packed( schema ) )
{
dcomplex* restrict y_off_ri = y + (offm_local )*rs_y1
+ (offn_local )*cs_y1;
dcomplex* restrict y_off_ir = y + (offm_local )*rs_y1
+ (offn_local )*cs_y1 + ld_y/2;
for ( j = 0; j < n_local; ++j )
for ( i = 0; i < m_local; ++i )
{
bli_zcopy1es( *(alpha),
*(y_off_ri + i*rs_y1 + j*cs_y1),
*(y_off_ir + i*rs_y1 + j*cs_y1) );
}
}
else /* if ( bli_is_1r_packed( schema ) ) */
{
/* Scale the non-unit stride by two for the 1r loop, which steps
in units of real (not complex) values. */ \
if ( rs_y2 == 1 ) { cs_y2 *= 2; } \
else /* if ( cs_y2 == 1 ) */ { rs_y2 *= 2; } \
\
double* restrict y_cast = ( double* )y; \
double* restrict y_off_r = y_cast + (offm_local )*rs_y2 \
+ (offn_local )*cs_y2; \
double* restrict y_off_i = y_cast + (offm_local )*rs_y2 \
+ (offn_local )*cs_y2 + ld_y; \
\
for ( j = 0; j < n_local; ++j ) \
for ( i = 0; i < m_local; ++i ) \
{ \
bli_zcopy1rs( *(a), \
*(y_off_r + i*rs_y2 + j*cs_y2), \
*(y_off_i + i*rs_y2 + j*cs_y2) ); \
} \
} \
in units of real (not complex) values. */
if ( rs_y2 == 1 ) { cs_y2 *= 2; }
else /* if ( cs_y2 == 1 ) */ { rs_y2 *= 2; }
double* restrict y_cast = ( double* )y;
double* restrict y_off_r = y_cast + (offm_local )*rs_y2
+ (offn_local )*cs_y2;
double* restrict y_off_i = y_cast + (offm_local )*rs_y2
+ (offn_local )*cs_y2 + ld_y;
for ( j = 0; j < n_local; ++j )
for ( i = 0; i < m_local; ++i )
{
bli_zcopy1rs( *(alpha),
*(y_off_r + i*rs_y2 + j*cs_y2),
*(y_off_i + i*rs_y2 + j*cs_y2) );
}
}
}
#endif

54
frame/include/level0/bli_set0s_mxn.h
View File

@@ -41,42 +41,36 @@
// - The first char encodes the type of x.
// - The second char encodes the type of y.
#define bli_sset0s_mxn( m, n, y, rs_y, cs_y ) \
{ \
dim_t _i, _j; \
\
for ( _j = 0; _j < n; ++_j ) \
for ( _i = 0; _i < m; ++_i ) \
bli_sset0s( *(y + _i*rs_y + _j*cs_y) ); \
static void bli_sset0s_mxn( const dim_t m, const dim_t n,
float* restrict y, const inc_t rs_y, const inc_t cs_y )
{
for ( dim_t j = 0; j < n; ++j )
for ( dim_t i = 0; i < m; ++i )
bli_sset0s( *(y + i*rs_y + j*cs_y) );
}
#define bli_dset0s_mxn( m, n, y, rs_y, cs_y ) \
{ \
dim_t _i, _j; \
\
for ( _j = 0; _j < n; ++_j ) \
for ( _i = 0; _i < m; ++_i ) \
bli_dset0s( *(y + _i*rs_y + _j*cs_y) ); \
static void bli_dset0s_mxn( const dim_t m, const dim_t n,
double* restrict y, const inc_t rs_y, const inc_t cs_y )
{
for ( dim_t j = 0; j < n; ++j )
for ( dim_t i = 0; i < m; ++i )
bli_dset0s( *(y + i*rs_y + j*cs_y) );
}
#define bli_cset0s_mxn( m, n, y, rs_y, cs_y ) \
{ \
dim_t _i, _j; \
\
for ( _j = 0; _j < n; ++_j ) \
for ( _i = 0; _i < m; ++_i ) \
bli_cset0s( *(y + _i*rs_y + _j*cs_y) ); \
static void bli_cset0s_mxn( const dim_t m, const dim_t n,
scomplex* restrict y, const inc_t rs_y, const inc_t cs_y )
{
for ( dim_t j = 0; j < n; ++j )
for ( dim_t i = 0; i < m; ++i )
bli_cset0s( *(y + i*rs_y + j*cs_y) );
}
#define bli_zset0s_mxn( m, n, y, rs_y, cs_y ) \
{ \
dim_t _i, _j; \
\
for ( _j = 0; _j < n; ++_j ) \
for ( _i = 0; _i < m; ++_i ) \
bli_zset0s( *(y + _i*rs_y + _j*cs_y) ); \
static void bli_zset0s_mxn( const dim_t m, const dim_t n,
dcomplex* restrict y, const inc_t rs_y, const inc_t cs_y )
{
for ( dim_t j = 0; j < n; ++j )
for ( dim_t i = 0; i < m; ++i )
bli_zset0s( *(y + i*rs_y + j*cs_y) );
}
#endif

173
frame/include/level0/ri/bli_scal2ris_mxn.h Normal file
View File

@@ -0,0 +1,173 @@
/*
BLIS
An object-based framework for developing high-performance BLAS-like
libraries.
Copyright (C) 2014, The University of Texas at Austin
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are
met:
- Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
- Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in the
documentation and/or other materials provided with the distribution.
- Neither the name(s) of the copyright holder(s) nor the names of its
contributors may be used to endorse or promote products derived
from this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#ifndef BLIS_SCAL2RIS_MXN_H
#define BLIS_SCAL2RIS_MXN_H
// scal2ris_mxn
static void bli_cscal2ris_mxn
(
const conj_t conjx,
const dim_t m,
const dim_t n,
scomplex* restrict alpha,
scomplex* restrict x, const inc_t rs_x, const inc_t cs_x,
scomplex* restrict y, const inc_t rs_y, const inc_t cs_y, const inc_t is_y
)
{
float* restrict alpha_r = ( float* )alpha; \
float* restrict alpha_i = ( float* )alpha + 1; \
float* restrict x_r = ( float* )x; \
float* restrict x_i = ( float* )x + 1; \
float* restrict y_r = ( float* )y; \
float* restrict y_i = ( float* )y + is_y; \
const dim_t incx2 = 2*rs_x; \
const dim_t ldx2 = 2*cs_x; \
/* Treat the micro-panel as panel_dim x panel_len and column-stored
(unit row stride). */ \
if ( bli_is_conj( conjx ) )
{
for ( dim_t j = 0; j < n; ++j )
for ( dim_t i = 0; i < m; ++i )
{
float* restrict chi11_r = x_r + (i )*incx2 + (j )*ldx2;
float* restrict chi11_i = x_i + (i )*incx2 + (j )*ldx2;
float* restrict psi11_r = y_r + (i )*1 + (j )*cs_y;
float* restrict psi11_i = y_i + (i )*1 + (j )*cs_y;
bli_cscal2jris
(
*alpha_r,
*alpha_i,
*chi11_r,
*chi11_i,
*psi11_r,
*psi11_i
);
}
}
else /* if ( bli_is_noconj( conjx ) ) */
{
for ( dim_t j = 0; j < n; ++j )
for ( dim_t i = 0; i < m; ++i )
{
float* restrict chi11_r = x_r + (i )*incx2 + (j )*ldx2;
float* restrict chi11_i = x_i + (i )*incx2 + (j )*ldx2;
float* restrict psi11_r = y_r + (i )*1 + (j )*cs_y;
float* restrict psi11_i = y_i + (i )*1 + (j )*cs_y;
bli_cscal2ris
(
*alpha_r,
*alpha_i,
*chi11_r,
*chi11_i,
*psi11_r,
*psi11_i
);
}
}
}
static void bli_zscal2ris_mxn
(
const conj_t conjx,
const dim_t m,
const dim_t n,
dcomplex* restrict alpha,
dcomplex* restrict x, const inc_t rs_x, const inc_t cs_x,
dcomplex* restrict y, const inc_t rs_y, const inc_t cs_y, const inc_t is_y
)
{
double* restrict alpha_r = ( double* )alpha; \
double* restrict alpha_i = ( double* )alpha + 1; \
double* restrict x_r = ( double* )x; \
double* restrict x_i = ( double* )x + 1; \
double* restrict y_r = ( double* )y; \
double* restrict y_i = ( double* )y + is_y; \
const dim_t incx2 = 2*rs_x; \
const dim_t ldx2 = 2*cs_x; \
/* Treat the micro-panel as panel_dim x panel_len and column-stored
(unit row stride). */ \
if ( bli_is_conj( conjx ) )
{
for ( dim_t j = 0; j < n; ++j )
for ( dim_t i = 0; i < m; ++i )
{
double* restrict chi11_r = x_r + (i )*incx2 + (j )*ldx2;
double* restrict chi11_i = x_i + (i )*incx2 + (j )*ldx2;
double* restrict psi11_r = y_r + (i )*1 + (j )*cs_y;
double* restrict psi11_i = y_i + (i )*1 + (j )*cs_y;
bli_zscal2jris
(
*alpha_r,
*alpha_i,
*chi11_r,
*chi11_i,
*psi11_r,
*psi11_i
);
}
}
else /* if ( bli_is_noconj( conjx ) ) */
{
for ( dim_t j = 0; j < n; ++j )
for ( dim_t i = 0; i < m; ++i )
{
double* restrict chi11_r = x_r + (i )*incx2 + (j )*ldx2;
double* restrict chi11_i = x_i + (i )*incx2 + (j )*ldx2;
double* restrict psi11_r = y_r + (i )*1 + (j )*cs_y;
double* restrict psi11_i = y_i + (i )*1 + (j )*cs_y;
bli_zscal2ris
(
*alpha_r,
*alpha_i,
*chi11_r,
*chi11_i,
*psi11_r,
*psi11_i
);
}
}
}
#endif

183
frame/include/level0/ri3/bli_scal2ri3s_mxn.h Normal file
View File

@@ -0,0 +1,183 @@
/*
BLIS
An object-based framework for developing high-performance BLAS-like
libraries.
Copyright (C) 2014, The University of Texas at Austin
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are
met:
- Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
- Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in the
documentation and/or other materials provided with the distribution.
- Neither the name(s) of the copyright holder(s) nor the names of its
contributors may be used to endorse or promote products derived
from this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#ifndef BLIS_SCAL2RI3S_MXN_H
#define BLIS_SCAL2RI3S_MXN_H
// scal2ri3s_mxn
static void bli_cscal2ri3s_mxn
(
const conj_t conjx,
const dim_t m,
const dim_t n,
scomplex* restrict alpha,
scomplex* restrict x, const inc_t rs_x, const inc_t cs_x,
scomplex* restrict y, const inc_t rs_y, const inc_t cs_y, const inc_t is_y
)
{
float* restrict alpha_r = ( float* )alpha; \
float* restrict alpha_i = ( float* )alpha + 1; \
float* restrict x_r = ( float* )x; \
float* restrict x_i = ( float* )x + 1; \
float* restrict y_r = ( float* )y; \
float* restrict y_i = ( float* )y + is_y; \
float* restrict y_rpi = ( float* )y + 2*is_y; \
const dim_t incx2 = 2*rs_x; \
const dim_t ldx2 = 2*cs_x; \
/* Treat the micro-panel as panel_dim x panel_len and column-stored
(unit row stride). */ \
if ( bli_is_conj( conjx ) )
{
for ( dim_t j = 0; j < n; ++j )
for ( dim_t i = 0; i < m; ++i )
{
float* restrict chi11_r = x_r + (i )*incx2 + (j )*ldx2;
float* restrict chi11_i = x_i + (i )*incx2 + (j )*ldx2;
float* restrict psi11_r = y_r + (i )*1 + (j )*cs_y;
float* restrict psi11_i = y_i + (i )*1 + (j )*cs_y;
float* restrict psi11_rpi = y_rpi + (i )*1 + (j )*cs_y;
bli_cscal2jri3s
(
*alpha_r,
*alpha_i,
*chi11_r,
*chi11_i,
*psi11_r,
*psi11_i,
*psi11_rpi
);
}
}
else /* if ( bli_is_noconj( conjx ) ) */
{
for ( dim_t j = 0; j < n; ++j )
for ( dim_t i = 0; i < m; ++i )
{
float* restrict chi11_r = x_r + (i )*incx2 + (j )*ldx2;
float* restrict chi11_i = x_i + (i )*incx2 + (j )*ldx2;
float* restrict psi11_r = y_r + (i )*1 + (j )*cs_y;
float* restrict psi11_i = y_i + (i )*1 + (j )*cs_y;
float* restrict psi11_rpi = y_rpi + (i )*1 + (j )*cs_y;
bli_cscal2ri3s
(
*alpha_r,
*alpha_i,
*chi11_r,
*chi11_i,
*psi11_r,
*psi11_i,
*psi11_rpi
);
}
}
}
static void bli_zscal2ri3s_mxn
(
const conj_t conjx,
const dim_t m,
const dim_t n,
dcomplex* restrict alpha,
dcomplex* restrict x, const inc_t rs_x, const inc_t cs_x,
dcomplex* restrict y, const inc_t rs_y, const inc_t cs_y, const inc_t is_y
)
{
double* restrict alpha_r = ( double* )alpha; \
double* restrict alpha_i = ( double* )alpha + 1; \
double* restrict x_r = ( double* )x; \
double* restrict x_i = ( double* )x + 1; \
double* restrict y_r = ( double* )y; \
double* restrict y_i = ( double* )y + is_y; \
double* restrict y_rpi = ( double* )y + 2*is_y; \
const dim_t incx2 = 2*rs_x; \
const dim_t ldx2 = 2*cs_x; \
/* Treat the micro-panel as panel_dim x panel_len and column-stored
(unit row stride). */ \
if ( bli_is_conj( conjx ) )
{
for ( dim_t j = 0; j < n; ++j )
for ( dim_t i = 0; i < m; ++i )
{
double* restrict chi11_r = x_r + (i )*incx2 + (j )*ldx2;
double* restrict chi11_i = x_i + (i )*incx2 + (j )*ldx2;
double* restrict psi11_r = y_r + (i )*1 + (j )*cs_y;
double* restrict psi11_i = y_i + (i )*1 + (j )*cs_y;
double* restrict psi11_rpi = y_rpi + (i )*1 + (j )*cs_y;
bli_zscal2jri3s
(
*alpha_r,
*alpha_i,
*chi11_r,
*chi11_i,
*psi11_r,
*psi11_i,
*psi11_rpi
);
}
}
else /* if ( bli_is_noconj( conjx ) ) */
{
for ( dim_t j = 0; j < n; ++j )
for ( dim_t i = 0; i < m; ++i )
{
double* restrict chi11_r = x_r + (i )*incx2 + (j )*ldx2;
double* restrict chi11_i = x_i + (i )*incx2 + (j )*ldx2;
double* restrict psi11_r = y_r + (i )*1 + (j )*cs_y;
double* restrict psi11_i = y_i + (i )*1 + (j )*cs_y;
double* restrict psi11_rpi = y_rpi + (i )*1 + (j )*cs_y;
bli_zscal2ri3s
(
*alpha_r,
*alpha_i,
*chi11_r,
*chi11_i,
*psi11_r,
*psi11_i,
*psi11_rpi
);
}
}
}
#endif

283
frame/include/level0/rih/bli_scal2rihs_mxn.h Normal file
View File

@@ -0,0 +1,283 @@
/*
BLIS
An object-based framework for developing high-performance BLAS-like
libraries.
Copyright (C) 2014, The University of Texas at Austin
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are
met:
- Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
- Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in the
documentation and/or other materials provided with the distribution.
- Neither the name(s) of the copyright holder(s) nor the names of its
contributors may be used to endorse or promote products derived
from this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#ifndef BLIS_SCAL2RIHS_MXN_H
#define BLIS_SCAL2RIHS_MXN_H
// scal2rihs_mxn
static void bli_cscal2rihs_mxn
(
const pack_t schema,
const conj_t conjx,
const dim_t m,
const dim_t n,
scomplex* restrict alpha,
scomplex* restrict x, const inc_t rs_x, const inc_t cs_x,
scomplex* restrict y, const inc_t rs_y, const inc_t cs_y
)
{
scomplex* restrict x_r = x;
float* restrict y_r = ( float* )y;
if ( bli_is_ro_packed( schema ) )
{
if ( bli_is_conj( conjx ) )
{
for ( dim_t j = 0; j < n; ++j )
for ( dim_t i = 0; i < m; ++i )
{
scomplex* restrict chi11 = x_r + (i )*rs_x + (j )*cs_x;
float* restrict psi11_r = y_r + (i )*rs_y + (j )*cs_y;
bli_cscal2jros
(
*alpha,
*chi11,
*psi11_r
);
}
}
else /* if ( bli_is_noconj( conjx ) ) */
{
for ( dim_t j = 0; j < n; ++j )
for ( dim_t i = 0; i < m; ++i )
{
scomplex* restrict chi11 = x_r + (i )*rs_x + (j )*cs_x;
float* restrict psi11_r = y_r + (i )*rs_y + (j )*cs_y;
bli_cscal2ros
(
*alpha,
*chi11,
*psi11_r
);
}
}
}
else if ( bli_is_io_packed( schema ) )
{
if ( bli_is_conj( conjx ) )
{
for ( dim_t j = 0; j < n; ++j )
for ( dim_t i = 0; i < m; ++i )
{
scomplex* restrict chi11 = x_r + (i )*rs_x + (j )*cs_x;
float* restrict psi11_r = y_r + (i )*rs_y + (j )*cs_y;
bli_cscal2jios
(
*alpha,
*chi11,
*psi11_r
);
}
}
else /* if ( bli_is_noconj( conjx ) ) */
{
for ( dim_t j = 0; j < n; ++j )
for ( dim_t i = 0; i < m; ++i )
{
scomplex* restrict chi11 = x_r + (i )*rs_x + (j )*cs_x;
float* restrict psi11_r = y_r + (i )*rs_y + (j )*cs_y;
bli_cscal2ios
(
*alpha,
*chi11,
*psi11_r
);
}
}
}
else /* if ( bli_is_rpi_packed( schema ) ) */
{
if ( bli_is_conj( conjx ) )
{
for ( dim_t j = 0; j < n; ++j )
for ( dim_t i = 0; i < m; ++i )
{
scomplex* restrict chi11 = x_r + (i )*rs_x + (j )*cs_x;
float* restrict psi11_r = y_r + (i )*rs_y + (j )*cs_y;
bli_cscal2jrpis
(
*alpha,
*chi11,
*psi11_r
);
}
}
else /* if ( bli_is_noconj( conjx ) ) */
{
for ( dim_t j = 0; j < n; ++j )
for ( dim_t i = 0; i < m; ++i )
{
scomplex* restrict chi11 = x_r + (i )*rs_x + (j )*cs_x;
float* restrict psi11_r = y_r + (i )*rs_y + (j )*cs_y;
bli_cscal2rpis
(
*alpha,
*chi11,
*psi11_r
);
}
}
}
}
static void bli_zscal2rihs_mxn
(
const pack_t schema,
const conj_t conjx,
const dim_t m,
const dim_t n,
dcomplex* restrict alpha,
dcomplex* restrict x, const inc_t rs_x, const inc_t cs_x,
dcomplex* restrict y, const inc_t rs_y, const inc_t cs_y
)
{
dcomplex* restrict x_r = x;
double* restrict y_r = ( double* )y;
if ( bli_is_ro_packed( schema ) )
{
if ( bli_is_conj( conjx ) )
{
for ( dim_t j = 0; j < n; ++j )
for ( dim_t i = 0; i < m; ++i )
{
dcomplex* restrict chi11 = x_r + (i )*rs_x + (j )*cs_x;
double* restrict psi11_r = y_r + (i )*rs_y + (j )*cs_y;
bli_zscal2jros
(
*alpha,
*chi11,
*psi11_r
);
}
}
else /* if ( bli_is_noconj( conjx ) ) */
{
for ( dim_t j = 0; j < n; ++j )
for ( dim_t i = 0; i < m; ++i )
{
dcomplex* restrict chi11 = x_r + (i )*rs_x + (j )*cs_x;
double* restrict psi11_r = y_r + (i )*rs_y + (j )*cs_y;
bli_zscal2ros
(
*alpha,
*chi11,
*psi11_r
);
}
}
}
else if ( bli_is_io_packed( schema ) )
{
if ( bli_is_conj( conjx ) )
{
for ( dim_t j = 0; j < n; ++j )
for ( dim_t i = 0; i < m; ++i )
{
dcomplex* restrict chi11 = x_r + (i )*rs_x + (j )*cs_x;
double* restrict psi11_r = y_r + (i )*rs_y + (j )*cs_y;
bli_zscal2jios
(
*alpha,
*chi11,
*psi11_r
);
}
}
else /* if ( bli_is_noconj( conjx ) ) */
{
for ( dim_t j = 0; j < n; ++j )
for ( dim_t i = 0; i < m; ++i )
{
dcomplex* restrict chi11 = x_r + (i )*rs_x + (j )*cs_x;
double* restrict psi11_r = y_r + (i )*rs_y + (j )*cs_y;
bli_zscal2ios
(
*alpha,
*chi11,
*psi11_r
);
}
}
}
else /* if ( bli_is_rpi_packed( schema ) ) */
{
if ( bli_is_conj( conjx ) )
{
for ( dim_t j = 0; j < n; ++j )
for ( dim_t i = 0; i < m; ++i )
{
dcomplex* restrict chi11 = x_r + (i )*rs_x + (j )*cs_x;
double* restrict psi11_r = y_r + (i )*rs_y + (j )*cs_y;
bli_zscal2jrpis
(
*alpha,
*chi11,
*psi11_r
);
}
}
else /* if ( bli_is_noconj( conjx ) ) */
{
for ( dim_t j = 0; j < n; ++j )
for ( dim_t i = 0; i < m; ++i )
{
dcomplex* restrict chi11 = x_r + (i )*rs_x + (j )*cs_x;
double* restrict psi11_r = y_r + (i )*rs_y + (j )*cs_y;
bli_zscal2rpis
(
*alpha,
*chi11,
*psi11_r
);
}
}
}
}
#endif