]> git.ipfire.org Git - thirdparty/gcc.git/blob - libgfortran/m4/matmull.m4
re PR libfortran/21468 (vectorizing libfortran)
[thirdparty/gcc.git] / libgfortran / m4 / matmull.m4
1 `/* Implementation of the MATMUL intrinsic
2 Copyright 2002, 2005 Free Software Foundation, Inc.
3 Contributed by Paul Brook <paul@nowt.org>
4
5 This file is part of the GNU Fortran 95 runtime library (libgfortran).
6
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.
11
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
19 executable.)
20
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.
25
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. */
30
31 #include "config.h"
32 #include <stdlib.h>
33 #include <assert.h>
34 #include "libgfortran.h"'
35 include(iparm.m4)dnl
36
37 `#if defined (HAVE_'rtype_name`)'
38
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. */
41
42 extern void matmul_`'rtype_code (rtype * const restrict,
43 gfc_array_l4 * const restrict, gfc_array_l4 * const restrict);
44 export_proto(matmul_`'rtype_code);
45
46 void
47 matmul_`'rtype_code (rtype * const restrict retarray,
48 gfc_array_l4 * const restrict a, gfc_array_l4 * const restrict b)
49 {
50 const GFC_INTEGER_4 * restrict abase;
51 const GFC_INTEGER_4 * restrict bbase;
52 rtype_name * restrict dest;
53 index_type rxstride;
54 index_type rystride;
55 index_type xcount;
56 index_type ycount;
57 index_type xstride;
58 index_type ystride;
59 index_type x;
60 index_type y;
61
62 const GFC_INTEGER_4 * restrict pa;
63 const GFC_INTEGER_4 * restrict pb;
64 index_type astride;
65 index_type bstride;
66 index_type count;
67 index_type n;
68
69 assert (GFC_DESCRIPTOR_RANK (a) == 2
70 || GFC_DESCRIPTOR_RANK (b) == 2);
71
72 if (retarray->data == NULL)
73 {
74 if (GFC_DESCRIPTOR_RANK (a) == 1)
75 {
76 retarray->dim[0].lbound = 0;
77 retarray->dim[0].ubound = b->dim[1].ubound - b->dim[1].lbound;
78 retarray->dim[0].stride = 1;
79 }
80 else if (GFC_DESCRIPTOR_RANK (b) == 1)
81 {
82 retarray->dim[0].lbound = 0;
83 retarray->dim[0].ubound = a->dim[0].ubound - a->dim[0].lbound;
84 retarray->dim[0].stride = 1;
85 }
86 else
87 {
88 retarray->dim[0].lbound = 0;
89 retarray->dim[0].ubound = a->dim[0].ubound - a->dim[0].lbound;
90 retarray->dim[0].stride = 1;
91
92 retarray->dim[1].lbound = 0;
93 retarray->dim[1].ubound = b->dim[1].ubound - b->dim[1].lbound;
94 retarray->dim[1].stride = retarray->dim[0].ubound+1;
95 }
96
97 retarray->data
98 = internal_malloc_size (sizeof (rtype_name) * size0 ((array_t *) retarray));
99 retarray->offset = 0;
100 }
101
102 abase = a->data;
103 if (GFC_DESCRIPTOR_SIZE (a) != 4)
104 {
105 assert (GFC_DESCRIPTOR_SIZE (a) == 8);
106 abase = GFOR_POINTER_L8_TO_L4 (abase);
107 }
108 bbase = b->data;
109 if (GFC_DESCRIPTOR_SIZE (b) != 4)
110 {
111 assert (GFC_DESCRIPTOR_SIZE (b) == 8);
112 bbase = GFOR_POINTER_L8_TO_L4 (bbase);
113 }
114 dest = retarray->data;
115
116 if (retarray->dim[0].stride == 0)
117 retarray->dim[0].stride = 1;
118 if (a->dim[0].stride == 0)
119 a->dim[0].stride = 1;
120 if (b->dim[0].stride == 0)
121 b->dim[0].stride = 1;
122
123 sinclude(`matmul_asm_'rtype_code`.m4')dnl
124
125 if (GFC_DESCRIPTOR_RANK (retarray) == 1)
126 {
127 rxstride = retarray->dim[0].stride;
128 rystride = rxstride;
129 }
130 else
131 {
132 rxstride = retarray->dim[0].stride;
133 rystride = retarray->dim[1].stride;
134 }
135
136 /* If we have rank 1 parameters, zero the absent stride, and set the size to
137 one. */
138 if (GFC_DESCRIPTOR_RANK (a) == 1)
139 {
140 astride = a->dim[0].stride;
141 count = a->dim[0].ubound + 1 - a->dim[0].lbound;
142 xstride = 0;
143 rxstride = 0;
144 xcount = 1;
145 }
146 else
147 {
148 astride = a->dim[1].stride;
149 count = a->dim[1].ubound + 1 - a->dim[1].lbound;
150 xstride = a->dim[0].stride;
151 xcount = a->dim[0].ubound + 1 - a->dim[0].lbound;
152 }
153 if (GFC_DESCRIPTOR_RANK (b) == 1)
154 {
155 bstride = b->dim[0].stride;
156 assert(count == b->dim[0].ubound + 1 - b->dim[0].lbound);
157 ystride = 0;
158 rystride = 0;
159 ycount = 1;
160 }
161 else
162 {
163 bstride = b->dim[0].stride;
164 assert(count == b->dim[0].ubound + 1 - b->dim[0].lbound);
165 ystride = b->dim[1].stride;
166 ycount = b->dim[1].ubound + 1 - b->dim[1].lbound;
167 }
168
169 for (y = 0; y < ycount; y++)
170 {
171 for (x = 0; x < xcount; x++)
172 {
173 /* Do the summation for this element. For real and integer types
174 this is the same as DOT_PRODUCT. For complex types we use do
175 a*b, not conjg(a)*b. */
176 pa = abase;
177 pb = bbase;
178 *dest = 0;
179
180 for (n = 0; n < count; n++)
181 {
182 if (*pa && *pb)
183 {
184 *dest = 1;
185 break;
186 }
187 pa += astride;
188 pb += bstride;
189 }
190
191 dest += rxstride;
192 abase += xstride;
193 }
194 abase -= xstride * xcount;
195 bbase += ystride;
196 dest += rystride - (rxstride * xcount);
197 }
198 }
199
200 #endif