7f68c75fb3
gcc/fortran/ * iresolve.c (gfc_resolve_all, gfc_resolve_any, gfc_resolve_count, gfc_resolve_cshift, gfc_resolve_dot_product, gfc_resolve_eoshift, gfc_resolve_matmul, gfc_resolve_maxloc, gfc_resolve_maxval, gfc_resolve_minloc, gfc_resolve_minval, gfc_resolve_pack, gfc_resolve_product, gfc_resolve_reshape, gfc_resolve_shape, gfc_resolve_spread, gfc_resolve_sum, gfc_resolve_transpose, gfc_resolve_unpack: Use PREFIX. libgfortran/ * intrinsics/cshift0.c, intrinsics/eoshift0.c, intrinsics/eoshift2.c, intrinsics/pack_generic.c, intrinsics/reshape_generic.c, intrinsics/spread_generic.c, intrinsics/transpose_generic.c, intrinsics/unpack_generic.c, m4/cshift1.m4, m4/dotprod.m4, m4/dotprodc.m4, m4/dotprodl.m4, m4/eoshift1.m4, m4/eoshift3.m4, m4/iforeach.m4, m4/ifunction.m4, m4/matmul.m4, m4/matmull.m4, m4/reshape.m4, m4/shape.m4, m4/transpose.m4: Use standard prefix instead of "__". * generated/*: Rebuild. From-SVN: r92075
188 lines
5.1 KiB
Plaintext
188 lines
5.1 KiB
Plaintext
`/* Implementation of the MATMUL intrinsic
|
|
Copyright 2002 Free Software Foundation, Inc.
|
|
Contributed by Paul Brook <paul@nowt.org>
|
|
|
|
This file is part of the GNU Fortran 95 runtime library (libgfor).
|
|
|
|
Libgfortran is free software; you can redistribute it and/or
|
|
modify it under the terms of the GNU Lesser General Public
|
|
License as published by the Free Software Foundation; either
|
|
version 2.1 of the License, or (at your option) any later version.
|
|
|
|
Libgfortran is distributed in the hope that it will be useful,
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
GNU Lesser General Public License for more details.
|
|
|
|
You should have received a copy of the GNU Lesser General Public
|
|
License along with libgfor; see the file COPYING.LIB. If not,
|
|
write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
|
|
Boston, MA 02111-1307, USA. */
|
|
|
|
#include "config.h"
|
|
#include <stdlib.h>
|
|
#include <assert.h>
|
|
#include "libgfortran.h"'
|
|
include(iparm.m4)dnl
|
|
|
|
/* Dimensions: retarray(x,y) a(x, count) b(count,y).
|
|
Either a or b can be rank 1. In this case x or y is 1. */
|
|
|
|
extern void matmul_`'rtype_code (rtype *, gfc_array_l4 *, gfc_array_l4 *);
|
|
export_proto(matmul_`'rtype_code);
|
|
|
|
void
|
|
matmul_`'rtype_code (rtype * retarray, gfc_array_l4 * a, gfc_array_l4 * b)
|
|
{
|
|
GFC_INTEGER_4 *abase;
|
|
GFC_INTEGER_4 *bbase;
|
|
rtype_name *dest;
|
|
index_type rxstride;
|
|
index_type rystride;
|
|
index_type xcount;
|
|
index_type ycount;
|
|
index_type xstride;
|
|
index_type ystride;
|
|
index_type x;
|
|
index_type y;
|
|
|
|
GFC_INTEGER_4 *pa;
|
|
GFC_INTEGER_4 *pb;
|
|
index_type astride;
|
|
index_type bstride;
|
|
index_type count;
|
|
index_type n;
|
|
|
|
assert (GFC_DESCRIPTOR_RANK (a) == 2
|
|
|| GFC_DESCRIPTOR_RANK (b) == 2);
|
|
|
|
if (retarray->data == NULL)
|
|
{
|
|
if (GFC_DESCRIPTOR_RANK (a) == 1)
|
|
{
|
|
retarray->dim[0].lbound = 0;
|
|
retarray->dim[0].ubound = b->dim[1].ubound - b->dim[1].lbound;
|
|
retarray->dim[0].stride = 1;
|
|
}
|
|
else if (GFC_DESCRIPTOR_RANK (b) == 1)
|
|
{
|
|
retarray->dim[0].lbound = 0;
|
|
retarray->dim[0].ubound = a->dim[0].ubound - a->dim[0].lbound;
|
|
retarray->dim[0].stride = 1;
|
|
}
|
|
else
|
|
{
|
|
retarray->dim[0].lbound = 0;
|
|
retarray->dim[0].ubound = a->dim[0].ubound - a->dim[0].lbound;
|
|
retarray->dim[0].stride = 1;
|
|
|
|
retarray->dim[1].lbound = 0;
|
|
retarray->dim[1].ubound = b->dim[1].ubound - b->dim[1].lbound;
|
|
retarray->dim[1].stride = retarray->dim[0].ubound+1;
|
|
}
|
|
|
|
retarray->data
|
|
= internal_malloc_size (sizeof (rtype_name) * size0 (retarray));
|
|
retarray->base = 0;
|
|
}
|
|
|
|
abase = a->data;
|
|
if (GFC_DESCRIPTOR_SIZE (a) != 4)
|
|
{
|
|
assert (GFC_DESCRIPTOR_SIZE (a) == 8);
|
|
abase = GFOR_POINTER_L8_TO_L4 (abase);
|
|
astride <<= 1;
|
|
}
|
|
bbase = b->data;
|
|
if (GFC_DESCRIPTOR_SIZE (b) != 4)
|
|
{
|
|
assert (GFC_DESCRIPTOR_SIZE (b) == 8);
|
|
bbase = GFOR_POINTER_L8_TO_L4 (bbase);
|
|
bstride <<= 1;
|
|
}
|
|
dest = retarray->data;
|
|
|
|
if (retarray->dim[0].stride == 0)
|
|
retarray->dim[0].stride = 1;
|
|
if (a->dim[0].stride == 0)
|
|
a->dim[0].stride = 1;
|
|
if (b->dim[0].stride == 0)
|
|
b->dim[0].stride = 1;
|
|
|
|
sinclude(`matmul_asm_'rtype_code`.m4')dnl
|
|
|
|
if (GFC_DESCRIPTOR_RANK (retarray) == 1)
|
|
{
|
|
rxstride = retarray->dim[0].stride;
|
|
rystride = rxstride;
|
|
}
|
|
else
|
|
{
|
|
rxstride = retarray->dim[0].stride;
|
|
rystride = retarray->dim[1].stride;
|
|
}
|
|
|
|
/* If we have rank 1 parameters, zero the absent stride, and set the size to
|
|
one. */
|
|
if (GFC_DESCRIPTOR_RANK (a) == 1)
|
|
{
|
|
astride = a->dim[0].stride;
|
|
count = a->dim[0].ubound + 1 - a->dim[0].lbound;
|
|
xstride = 0;
|
|
rxstride = 0;
|
|
xcount = 1;
|
|
}
|
|
else
|
|
{
|
|
astride = a->dim[1].stride;
|
|
count = a->dim[1].ubound + 1 - a->dim[1].lbound;
|
|
xstride = a->dim[0].stride;
|
|
xcount = a->dim[0].ubound + 1 - a->dim[0].lbound;
|
|
}
|
|
if (GFC_DESCRIPTOR_RANK (b) == 1)
|
|
{
|
|
bstride = b->dim[0].stride;
|
|
assert(count == b->dim[0].ubound + 1 - b->dim[0].lbound);
|
|
ystride = 0;
|
|
rystride = 0;
|
|
ycount = 1;
|
|
}
|
|
else
|
|
{
|
|
bstride = b->dim[0].stride;
|
|
assert(count == b->dim[0].ubound + 1 - b->dim[0].lbound);
|
|
ystride = b->dim[1].stride;
|
|
ycount = b->dim[1].ubound + 1 - b->dim[1].lbound;
|
|
}
|
|
|
|
for (y = 0; y < ycount; y++)
|
|
{
|
|
for (x = 0; x < xcount; x++)
|
|
{
|
|
/* Do the summation for this element. For real and integer types
|
|
this is the same as DOT_PRODUCT. For complex types we use do
|
|
a*b, not conjg(a)*b. */
|
|
pa = abase;
|
|
pb = bbase;
|
|
*dest = 0;
|
|
|
|
for (n = 0; n < count; n++)
|
|
{
|
|
if (*pa && *pb)
|
|
{
|
|
*dest = 1;
|
|
break;
|
|
}
|
|
pa += astride;
|
|
pb += bstride;
|
|
}
|
|
|
|
dest += rxstride;
|
|
abase += xstride;
|
|
}
|
|
abase -= xstride * xcount;
|
|
bbase += ystride;
|
|
dest += rystride - (rxstride * xcount);
|
|
}
|
|
}
|