1 `/* Implementation of the MATMUL intrinsic
2 Copyright 2002, 2005, 2006, 2007 Free Software Foundation, Inc.
3 Contributed by Paul Brook <paul@nowt.org>
5 This file is part of the GNU Fortran 95 runtime library (libgfortran).
7 Libgfortran is free software; you can redistribute it and/or
8 modify it under the terms of the GNU General Public
9 License as published by the Free Software Foundation; either
10 version 2 of the License, or (at your option) any later version.
12 In addition to the permissions in the GNU General Public License, the
13 Free Software Foundation gives you unlimited permission to link the
14 compiled version of this file into combinations with other programs,
15 and to distribute those combinations without any restriction coming
16 from the use of this file. (The General Public License restrictions
17 do apply in other respects; for example, they cover modification of
18 the file, and distribution when not linked into a combine
21 Libgfortran is distributed in the hope that it will be useful,
22 but WITHOUT ANY WARRANTY; without even the implied warranty of
23 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
24 GNU General Public License for more details.
26 You should have received a copy of the GNU General Public
27 License along with libgfortran; see the file COPYING. If not,
28 write to the Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor,
29 Boston, MA 02110-1301, USA. */
31 #include "libgfortran.h"
37 `#if defined (HAVE_'rtype_name`)
39 /* Dimensions: retarray(x,y) a(x, count) b(count,y).
40 Either a or b can be rank 1. In this case x or y is 1. */
42 extern void matmul_'rtype_code` ('rtype` * const restrict,
43 gfc_array_l1 * const restrict, gfc_array_l1 * const restrict);
44 export_proto(matmul_'rtype_code`);
47 matmul_'rtype_code` ('rtype` * const restrict retarray,
48 gfc_array_l1 * const restrict a, gfc_array_l1 * const restrict b)
50 const GFC_LOGICAL_1 * restrict abase;
51 const GFC_LOGICAL_1 * restrict bbase;
52 'rtype_name` * restrict dest;
64 const GFC_LOGICAL_1 * restrict pa;
65 const GFC_LOGICAL_1 * restrict pb;
71 assert (GFC_DESCRIPTOR_RANK (a) == 2
72 || GFC_DESCRIPTOR_RANK (b) == 2);
74 if (retarray->data == NULL)
76 if (GFC_DESCRIPTOR_RANK (a) == 1)
78 retarray->dim[0].lbound = 0;
79 retarray->dim[0].ubound = b->dim[1].ubound - b->dim[1].lbound;
80 retarray->dim[0].stride = 1;
82 else if (GFC_DESCRIPTOR_RANK (b) == 1)
84 retarray->dim[0].lbound = 0;
85 retarray->dim[0].ubound = a->dim[0].ubound - a->dim[0].lbound;
86 retarray->dim[0].stride = 1;
90 retarray->dim[0].lbound = 0;
91 retarray->dim[0].ubound = a->dim[0].ubound - a->dim[0].lbound;
92 retarray->dim[0].stride = 1;
94 retarray->dim[1].lbound = 0;
95 retarray->dim[1].ubound = b->dim[1].ubound - b->dim[1].lbound;
96 retarray->dim[1].stride = retarray->dim[0].ubound+1;
100 = internal_malloc_size (sizeof ('rtype_name`) * size0 ((array_t *) retarray));
101 retarray->offset = 0;
103 else if (compile_options.bounds_check)
105 index_type ret_extent, arg_extent;
107 if (GFC_DESCRIPTOR_RANK (a) == 1)
109 arg_extent = b->dim[1].ubound + 1 - b->dim[1].lbound;
110 ret_extent = retarray->dim[0].ubound + 1 - retarray->dim[0].lbound;
111 if (arg_extent != ret_extent)
112 runtime_error ("Incorrect extent in return array in"
113 " MATMUL intrinsic: is %ld, should be %ld",
114 (long int) ret_extent, (long int) arg_extent);
116 else if (GFC_DESCRIPTOR_RANK (b) == 1)
118 arg_extent = a->dim[0].ubound + 1 - a->dim[0].lbound;
119 ret_extent = retarray->dim[0].ubound + 1 - retarray->dim[0].lbound;
120 if (arg_extent != ret_extent)
121 runtime_error ("Incorrect extent in return array in"
122 " MATMUL intrinsic: is %ld, should be %ld",
123 (long int) ret_extent, (long int) arg_extent);
127 arg_extent = a->dim[0].ubound + 1 - a->dim[0].lbound;
128 ret_extent = retarray->dim[0].ubound + 1 - retarray->dim[0].lbound;
129 if (arg_extent != ret_extent)
130 runtime_error ("Incorrect extent in return array in"
131 " MATMUL intrinsic for dimension 1:"
132 " is %ld, should be %ld",
133 (long int) ret_extent, (long int) arg_extent);
135 arg_extent = b->dim[1].ubound + 1 - b->dim[1].lbound;
136 ret_extent = retarray->dim[1].ubound + 1 - retarray->dim[1].lbound;
137 if (arg_extent != ret_extent)
138 runtime_error ("Incorrect extent in return array in"
139 " MATMUL intrinsic for dimension 2:"
140 " is %ld, should be %ld",
141 (long int) ret_extent, (long int) arg_extent);
146 a_kind = GFC_DESCRIPTOR_SIZE (a);
148 if (a_kind == 1 || a_kind == 2 || a_kind == 4 || a_kind == 8
149 #ifdef HAVE_GFC_LOGICAL_16
153 abase = GFOR_POINTER_TO_L1 (abase, a_kind);
155 internal_error (NULL, "Funny sized logical array");
158 b_kind = GFC_DESCRIPTOR_SIZE (b);
160 if (b_kind == 1 || b_kind == 2 || b_kind == 4 || b_kind == 8
161 #ifdef HAVE_GFC_LOGICAL_16
165 bbase = GFOR_POINTER_TO_L1 (bbase, b_kind);
167 internal_error (NULL, "Funny sized logical array");
169 dest = retarray->data;
171 sinclude(`matmul_asm_'rtype_code`.m4')dnl
173 if (GFC_DESCRIPTOR_RANK (retarray) == 1)
175 rxstride = retarray->dim[0].stride;
180 rxstride = retarray->dim[0].stride;
181 rystride = retarray->dim[1].stride;
184 /* If we have rank 1 parameters, zero the absent stride, and set the size to
186 if (GFC_DESCRIPTOR_RANK (a) == 1)
188 astride = a->dim[0].stride * a_kind;
189 count = a->dim[0].ubound + 1 - a->dim[0].lbound;
196 astride = a->dim[1].stride * a_kind;
197 count = a->dim[1].ubound + 1 - a->dim[1].lbound;
198 xstride = a->dim[0].stride * a_kind;
199 xcount = a->dim[0].ubound + 1 - a->dim[0].lbound;
201 if (GFC_DESCRIPTOR_RANK (b) == 1)
203 bstride = b->dim[0].stride * b_kind;
204 assert(count == b->dim[0].ubound + 1 - b->dim[0].lbound);
211 bstride = b->dim[0].stride * b_kind;
212 assert(count == b->dim[0].ubound + 1 - b->dim[0].lbound);
213 ystride = b->dim[1].stride * b_kind;
214 ycount = b->dim[1].ubound + 1 - b->dim[1].lbound;
217 for (y = 0; y < ycount; y++)
219 for (x = 0; x < xcount; x++)
221 /* Do the summation for this element. For real and integer types
222 this is the same as DOT_PRODUCT. For complex types we use do
223 a*b, not conjg(a)*b. */
228 for (n = 0; n < count; n++)
242 abase -= xstride * xcount;
244 dest += rystride - (rxstride * xcount);