1*b1e83836Smrg /* Implementation of the CSHIFT intrinsic.
2*b1e83836Smrg Copyright (C) 2017-2022 Free Software Foundation, Inc.
3*b1e83836Smrg Contributed by Thomas Koenig <tkoenig@gcc.gnu.org>
4*b1e83836Smrg
5*b1e83836Smrg This file is part of the GNU Fortran 95 runtime library (libgfortran).
6*b1e83836Smrg
7*b1e83836Smrg Libgfortran is free software; you can redistribute it and/or
8*b1e83836Smrg modify it under the terms of the GNU General Public
9*b1e83836Smrg License as published by the Free Software Foundation; either
10*b1e83836Smrg version 3 of the License, or (at your option) any later version.
11*b1e83836Smrg
12*b1e83836Smrg Libgfortran is distributed in the hope that it will be useful,
13*b1e83836Smrg but WITHOUT ANY WARRANTY; without even the implied warranty of
14*b1e83836Smrg MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15*b1e83836Smrg GNU General Public License for more details.
16*b1e83836Smrg
17*b1e83836Smrg Under Section 7 of GPL version 3, you are granted additional
18*b1e83836Smrg permissions described in the GCC Runtime Library Exception, version
19*b1e83836Smrg 3.1, as published by the Free Software Foundation.
20*b1e83836Smrg
21*b1e83836Smrg You should have received a copy of the GNU General Public License and
22*b1e83836Smrg a copy of the GCC Runtime Library Exception along with this program;
23*b1e83836Smrg see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
24*b1e83836Smrg <http://www.gnu.org/licenses/>. */
25*b1e83836Smrg
26*b1e83836Smrg #include "libgfortran.h"
27*b1e83836Smrg #include <string.h>
28*b1e83836Smrg
29*b1e83836Smrg #if defined (HAVE_GFC_REAL_17) && defined (HAVE_GFC_INTEGER_16)
30*b1e83836Smrg
31*b1e83836Smrg void
cshift1_16_r17(gfc_array_r17 * const restrict ret,const gfc_array_r17 * const restrict array,const gfc_array_i16 * const restrict h,const GFC_INTEGER_16 * const restrict pwhich)32*b1e83836Smrg cshift1_16_r17 (gfc_array_r17 * const restrict ret,
33*b1e83836Smrg const gfc_array_r17 * const restrict array,
34*b1e83836Smrg const gfc_array_i16 * const restrict h,
35*b1e83836Smrg const GFC_INTEGER_16 * const restrict pwhich)
36*b1e83836Smrg {
37*b1e83836Smrg /* r.* indicates the return array. */
38*b1e83836Smrg index_type rstride[GFC_MAX_DIMENSIONS];
39*b1e83836Smrg index_type rstride0;
40*b1e83836Smrg index_type roffset;
41*b1e83836Smrg GFC_REAL_17 *rptr;
42*b1e83836Smrg GFC_REAL_17 *dest;
43*b1e83836Smrg /* s.* indicates the source array. */
44*b1e83836Smrg index_type sstride[GFC_MAX_DIMENSIONS];
45*b1e83836Smrg index_type sstride0;
46*b1e83836Smrg index_type soffset;
47*b1e83836Smrg const GFC_REAL_17 *sptr;
48*b1e83836Smrg const GFC_REAL_17 *src;
49*b1e83836Smrg /* h.* indicates the shift array. */
50*b1e83836Smrg index_type hstride[GFC_MAX_DIMENSIONS];
51*b1e83836Smrg index_type hstride0;
52*b1e83836Smrg const GFC_INTEGER_16 *hptr;
53*b1e83836Smrg
54*b1e83836Smrg index_type count[GFC_MAX_DIMENSIONS];
55*b1e83836Smrg index_type extent[GFC_MAX_DIMENSIONS];
56*b1e83836Smrg index_type rs_ex[GFC_MAX_DIMENSIONS];
57*b1e83836Smrg index_type ss_ex[GFC_MAX_DIMENSIONS];
58*b1e83836Smrg index_type hs_ex[GFC_MAX_DIMENSIONS];
59*b1e83836Smrg
60*b1e83836Smrg index_type dim;
61*b1e83836Smrg index_type len;
62*b1e83836Smrg index_type n;
63*b1e83836Smrg int which;
64*b1e83836Smrg GFC_INTEGER_16 sh;
65*b1e83836Smrg
66*b1e83836Smrg /* Bounds checking etc is already done by the caller. */
67*b1e83836Smrg
68*b1e83836Smrg if (pwhich)
69*b1e83836Smrg which = *pwhich - 1;
70*b1e83836Smrg else
71*b1e83836Smrg which = 0;
72*b1e83836Smrg
73*b1e83836Smrg extent[0] = 1;
74*b1e83836Smrg count[0] = 0;
75*b1e83836Smrg n = 0;
76*b1e83836Smrg
77*b1e83836Smrg /* Initialized for avoiding compiler warnings. */
78*b1e83836Smrg roffset = 1;
79*b1e83836Smrg soffset = 1;
80*b1e83836Smrg len = 0;
81*b1e83836Smrg
82*b1e83836Smrg for (dim = 0; dim < GFC_DESCRIPTOR_RANK (array); dim++)
83*b1e83836Smrg {
84*b1e83836Smrg if (dim == which)
85*b1e83836Smrg {
86*b1e83836Smrg roffset = GFC_DESCRIPTOR_STRIDE(ret,dim);
87*b1e83836Smrg if (roffset == 0)
88*b1e83836Smrg roffset = 1;
89*b1e83836Smrg soffset = GFC_DESCRIPTOR_STRIDE(array,dim);
90*b1e83836Smrg if (soffset == 0)
91*b1e83836Smrg soffset = 1;
92*b1e83836Smrg len = GFC_DESCRIPTOR_EXTENT(array,dim);
93*b1e83836Smrg }
94*b1e83836Smrg else
95*b1e83836Smrg {
96*b1e83836Smrg count[n] = 0;
97*b1e83836Smrg extent[n] = GFC_DESCRIPTOR_EXTENT(array,dim);
98*b1e83836Smrg rstride[n] = GFC_DESCRIPTOR_STRIDE(ret,dim);
99*b1e83836Smrg sstride[n] = GFC_DESCRIPTOR_STRIDE(array,dim);
100*b1e83836Smrg hstride[n] = GFC_DESCRIPTOR_STRIDE(h,n);
101*b1e83836Smrg rs_ex[n] = rstride[n] * extent[n];
102*b1e83836Smrg ss_ex[n] = sstride[n] * extent[n];
103*b1e83836Smrg hs_ex[n] = hstride[n] * extent[n];
104*b1e83836Smrg n++;
105*b1e83836Smrg }
106*b1e83836Smrg }
107*b1e83836Smrg if (sstride[0] == 0)
108*b1e83836Smrg sstride[0] = 1;
109*b1e83836Smrg if (rstride[0] == 0)
110*b1e83836Smrg rstride[0] = 1;
111*b1e83836Smrg if (hstride[0] == 0)
112*b1e83836Smrg hstride[0] = 1;
113*b1e83836Smrg
114*b1e83836Smrg dim = GFC_DESCRIPTOR_RANK (array);
115*b1e83836Smrg rstride0 = rstride[0];
116*b1e83836Smrg sstride0 = sstride[0];
117*b1e83836Smrg hstride0 = hstride[0];
118*b1e83836Smrg rptr = ret->base_addr;
119*b1e83836Smrg sptr = array->base_addr;
120*b1e83836Smrg hptr = h->base_addr;
121*b1e83836Smrg
122*b1e83836Smrg while (rptr)
123*b1e83836Smrg {
124*b1e83836Smrg /* Do the shift for this dimension. */
125*b1e83836Smrg sh = *hptr;
126*b1e83836Smrg /* Normal case should be -len < sh < len; try to
127*b1e83836Smrg avoid the expensive remainder operation if possible. */
128*b1e83836Smrg if (sh < 0)
129*b1e83836Smrg sh += len;
130*b1e83836Smrg if (unlikely(sh >= len || sh < 0))
131*b1e83836Smrg {
132*b1e83836Smrg sh = sh % len;
133*b1e83836Smrg if (sh < 0)
134*b1e83836Smrg sh += len;
135*b1e83836Smrg }
136*b1e83836Smrg src = &sptr[sh * soffset];
137*b1e83836Smrg dest = rptr;
138*b1e83836Smrg if (soffset == 1 && roffset == 1)
139*b1e83836Smrg {
140*b1e83836Smrg size_t len1 = sh * sizeof (GFC_REAL_17);
141*b1e83836Smrg size_t len2 = (len - sh) * sizeof (GFC_REAL_17);
142*b1e83836Smrg memcpy (rptr, sptr + sh, len2);
143*b1e83836Smrg memcpy (rptr + (len - sh), sptr, len1);
144*b1e83836Smrg }
145*b1e83836Smrg else
146*b1e83836Smrg {
147*b1e83836Smrg for (n = 0; n < len - sh; n++)
148*b1e83836Smrg {
149*b1e83836Smrg *dest = *src;
150*b1e83836Smrg dest += roffset;
151*b1e83836Smrg src += soffset;
152*b1e83836Smrg }
153*b1e83836Smrg for (src = sptr, n = 0; n < sh; n++)
154*b1e83836Smrg {
155*b1e83836Smrg *dest = *src;
156*b1e83836Smrg dest += roffset;
157*b1e83836Smrg src += soffset;
158*b1e83836Smrg }
159*b1e83836Smrg }
160*b1e83836Smrg
161*b1e83836Smrg /* Advance to the next section. */
162*b1e83836Smrg rptr += rstride0;
163*b1e83836Smrg sptr += sstride0;
164*b1e83836Smrg hptr += hstride0;
165*b1e83836Smrg count[0]++;
166*b1e83836Smrg n = 0;
167*b1e83836Smrg while (count[n] == extent[n])
168*b1e83836Smrg {
169*b1e83836Smrg /* When we get to the end of a dimension, reset it and increment
170*b1e83836Smrg the next dimension. */
171*b1e83836Smrg count[n] = 0;
172*b1e83836Smrg rptr -= rs_ex[n];
173*b1e83836Smrg sptr -= ss_ex[n];
174*b1e83836Smrg hptr -= hs_ex[n];
175*b1e83836Smrg n++;
176*b1e83836Smrg if (n >= dim - 1)
177*b1e83836Smrg {
178*b1e83836Smrg /* Break out of the loop. */
179*b1e83836Smrg rptr = NULL;
180*b1e83836Smrg break;
181*b1e83836Smrg }
182*b1e83836Smrg else
183*b1e83836Smrg {
184*b1e83836Smrg count[n]++;
185*b1e83836Smrg rptr += rstride[n];
186*b1e83836Smrg sptr += sstride[n];
187*b1e83836Smrg hptr += hstride[n];
188*b1e83836Smrg }
189*b1e83836Smrg }
190*b1e83836Smrg }
191*b1e83836Smrg }
192*b1e83836Smrg
193*b1e83836Smrg #endif
194