gcc/libgfortran/generated/matmul_i8.c
Paul Thomas a4a11197f9 re PR fortran/25049 (TRANSPOSE not allowed in initialisation expression)
2006-06-20  Paul Thomas  <pault@gcc.gnu.org>

	PR fortran/25049
	PR fortran/25050
	* check.c (non_init_transformational): New function.
	(find_substring_ref): New function to signal use of disallowed
	transformational intrinsic in an initialization expression.
	(gfc_check_all_any): Call previous if initialization expr.
	(gfc_check_count): The same.
	(gfc_check_cshift): The same.
	(gfc_check_dot_product): The same.
	(gfc_check_eoshift): The same.
	(gfc_check_minloc_maxloc): The same.
	(gfc_check_minval_maxval): The same.
	(gfc_check_gfc_check_product_sum): The same.
	(gfc_check_pack): The same.
	(gfc_check_spread): The same.
	(gfc_check_transpose): The same.
	(gfc_check_unpack): The same.

	PR fortran/18769
	*intrinsic.c (add_functions): Add gfc_simplify_transfer.
	*intrinsic.h : Add prototype for gfc_simplify_transfer.
	*simplify.c (gfc_simplify_transfer) : New function to act as
	placeholder for eventual implementation.  Emit error for now.

	PR fortran/16206
	* expr.c (find_array_element): Eliminate condition on length of
	offset. Add bounds checking. Rearrange exit. Return try and
	put gfc_constructor result as an argument.
	(find_array_section): New function.
	(find_substring_ref): New function.
	(simplify_const_ref): Add calls to previous.
	(simplify_parameter_variable): Return on NULL expr.
	(gfc_simplify_expr): Only call gfc_expand_constructor for full
	arrays.

	PR fortran/20876
	* match.c (gfc_match_forall): Add missing locus to gfc_code.

2006-06-20  Paul Thomas  <pault@gcc.gnu.org>

	PR libfortran/28005
	* m4/matmul.m4: aystride = 1 does not uniquely detect the
	presence of a temporary transpose; an array element in the
	first dimension produces the same signature.  Detect this
	using the rank of a and add specific code.
	* generated/matmul_r4.c: Regenerate.
	* generated/matmul_r8.c: Regenerate.
	* generated/matmul_r10.c: Regenerate.
	* generated/matmul_r16.c: Regenerate.
	* generated/matmul_c4.c: Regenerate.
	* generated/matmul_c8.c: Regenerate.
	* generated/matmul_c10.c: Regenerate.
	* generated/matmul_c16.c: Regenerate.
	* generated/matmul_i4.c: Regenerate.
	* generated/matmul_i8.c: Regenerate.
	* generated/matmul_i16.c: Regenerate.

2006-06-20  Paul Thomas  <pault@gcc.gnu.org>

	PR fortran/16206
	* gfortran.dg/array_initializer_1.f90: New test.

	PR fortran/28005
	* gfortran.dg/matmul_3.f90: New test.

From-SVN: r114802
2006-06-20 04:30:48 +00:00

285 lines
7.8 KiB
C

/* Implementation of the MATMUL intrinsic
Copyright 2002, 2005, 2006 Free Software Foundation, Inc.
Contributed by Paul Brook <paul@nowt.org>
This file is part of the GNU Fortran 95 runtime library (libgfortran).
Libgfortran is free software; you can redistribute it and/or
modify it under the terms of the GNU General Public
License as published by the Free Software Foundation; either
version 2 of the License, or (at your option) any later version.
In addition to the permissions in the GNU General Public License, the
Free Software Foundation gives you unlimited permission to link the
compiled version of this file into combinations with other programs,
and to distribute those combinations without any restriction coming
from the use of this file. (The General Public License restrictions
do apply in other respects; for example, they cover modification of
the file, and distribution when not linked into a combine
executable.)
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 General Public License for more details.
You should have received a copy of the GNU General Public
License along with libgfortran; see the file COPYING. If not,
write to the Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor,
Boston, MA 02110-1301, USA. */
#include "config.h"
#include <stdlib.h>
#include <string.h>
#include <assert.h>
#include "libgfortran.h"
#if defined (HAVE_GFC_INTEGER_8)
/* The order of loops is different in the case of plain matrix
multiplication C=MATMUL(A,B), and in the frequent special case where
the argument A is the temporary result of a TRANSPOSE intrinsic:
C=MATMUL(TRANSPOSE(A),B). Transposed temporaries are detected by
looking at their strides.
The equivalent Fortran pseudo-code is:
DIMENSION A(M,COUNT), B(COUNT,N), C(M,N)
IF (.NOT.IS_TRANSPOSED(A)) THEN
C = 0
DO J=1,N
DO K=1,COUNT
DO I=1,M
C(I,J) = C(I,J)+A(I,K)*B(K,J)
ELSE
DO J=1,N
DO I=1,M
S = 0
DO K=1,COUNT
S = S+A(I,K)+B(K,J)
C(I,J) = S
ENDIF
*/
extern void matmul_i8 (gfc_array_i8 * const restrict retarray,
gfc_array_i8 * const restrict a, gfc_array_i8 * const restrict b);
export_proto(matmul_i8);
void
matmul_i8 (gfc_array_i8 * const restrict retarray,
gfc_array_i8 * const restrict a, gfc_array_i8 * const restrict b)
{
const GFC_INTEGER_8 * restrict abase;
const GFC_INTEGER_8 * restrict bbase;
GFC_INTEGER_8 * restrict dest;
index_type rxstride, rystride, axstride, aystride, bxstride, bystride;
index_type x, y, n, count, xcount, ycount;
assert (GFC_DESCRIPTOR_RANK (a) == 2
|| GFC_DESCRIPTOR_RANK (b) == 2);
/* C[xcount,ycount] = A[xcount, count] * B[count,ycount]
Either A or B (but not both) can be rank 1:
o One-dimensional argument A is implicitly treated as a row matrix
dimensioned [1,count], so xcount=1.
o One-dimensional argument B is implicitly treated as a column matrix
dimensioned [count, 1], so ycount=1.
*/
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 (GFC_INTEGER_8) * size0 ((array_t *) retarray));
retarray->offset = 0;
}
if (GFC_DESCRIPTOR_RANK (retarray) == 1)
{
/* One-dimensional result may be addressed in the code below
either as a row or a column matrix. We want both cases to
work. */
rxstride = rystride = retarray->dim[0].stride;
}
else
{
rxstride = retarray->dim[0].stride;
rystride = retarray->dim[1].stride;
}
if (GFC_DESCRIPTOR_RANK (a) == 1)
{
/* Treat it as a a row matrix A[1,count]. */
axstride = a->dim[0].stride;
aystride = 1;
xcount = 1;
count = a->dim[0].ubound + 1 - a->dim[0].lbound;
}
else
{
axstride = a->dim[0].stride;
aystride = a->dim[1].stride;
count = a->dim[1].ubound + 1 - a->dim[1].lbound;
xcount = a->dim[0].ubound + 1 - a->dim[0].lbound;
}
assert(count == b->dim[0].ubound + 1 - b->dim[0].lbound);
if (GFC_DESCRIPTOR_RANK (b) == 1)
{
/* Treat it as a column matrix B[count,1] */
bxstride = b->dim[0].stride;
/* bystride should never be used for 1-dimensional b.
in case it is we want it to cause a segfault, rather than
an incorrect result. */
bystride = 0xDEADBEEF;
ycount = 1;
}
else
{
bxstride = b->dim[0].stride;
bystride = b->dim[1].stride;
ycount = b->dim[1].ubound + 1 - b->dim[1].lbound;
}
abase = a->data;
bbase = b->data;
dest = retarray->data;
if (rxstride == 1 && axstride == 1 && bxstride == 1)
{
const GFC_INTEGER_8 * restrict bbase_y;
GFC_INTEGER_8 * restrict dest_y;
const GFC_INTEGER_8 * restrict abase_n;
GFC_INTEGER_8 bbase_yn;
if (rystride == xcount)
memset (dest, 0, (sizeof (GFC_INTEGER_8) * xcount * ycount));
else
{
for (y = 0; y < ycount; y++)
for (x = 0; x < xcount; x++)
dest[x + y*rystride] = (GFC_INTEGER_8)0;
}
for (y = 0; y < ycount; y++)
{
bbase_y = bbase + y*bystride;
dest_y = dest + y*rystride;
for (n = 0; n < count; n++)
{
abase_n = abase + n*aystride;
bbase_yn = bbase_y[n];
for (x = 0; x < xcount; x++)
{
dest_y[x] += abase_n[x] * bbase_yn;
}
}
}
}
else if (rxstride == 1 && aystride == 1 && bxstride == 1)
{
if (GFC_DESCRIPTOR_RANK (a) != 1)
{
const GFC_INTEGER_8 *restrict abase_x;
const GFC_INTEGER_8 *restrict bbase_y;
GFC_INTEGER_8 *restrict dest_y;
GFC_INTEGER_8 s;
for (y = 0; y < ycount; y++)
{
bbase_y = &bbase[y*bystride];
dest_y = &dest[y*rystride];
for (x = 0; x < xcount; x++)
{
abase_x = &abase[x*axstride];
s = (GFC_INTEGER_8) 0;
for (n = 0; n < count; n++)
s += abase_x[n] * bbase_y[n];
dest_y[x] = s;
}
}
}
else
{
const GFC_INTEGER_8 *restrict bbase_y;
GFC_INTEGER_8 s;
for (y = 0; y < ycount; y++)
{
bbase_y = &bbase[y*bystride];
s = (GFC_INTEGER_8) 0;
for (n = 0; n < count; n++)
s += abase[n*axstride] * bbase_y[n];
dest[y*rystride] = s;
}
}
}
else if (axstride < aystride)
{
for (y = 0; y < ycount; y++)
for (x = 0; x < xcount; x++)
dest[x*rxstride + y*rystride] = (GFC_INTEGER_8)0;
for (y = 0; y < ycount; y++)
for (n = 0; n < count; n++)
for (x = 0; x < xcount; x++)
/* dest[x,y] += a[x,n] * b[n,y] */
dest[x*rxstride + y*rystride] += abase[x*axstride + n*aystride] * bbase[n*bxstride + y*bystride];
}
else
{
const GFC_INTEGER_8 *restrict abase_x;
const GFC_INTEGER_8 *restrict bbase_y;
GFC_INTEGER_8 *restrict dest_y;
GFC_INTEGER_8 s;
for (y = 0; y < ycount; y++)
{
bbase_y = &bbase[y*bystride];
dest_y = &dest[y*rystride];
for (x = 0; x < xcount; x++)
{
abase_x = &abase[x*axstride];
s = (GFC_INTEGER_8) 0;
for (n = 0; n < count; n++)
s += abase_x[n*aystride] * bbase_y[n*bxstride];
dest_y[x*rxstride] = s;
}
}
}
}
#endif