xref: /netbsd-src/external/gpl3/gcc/dist/libgfortran/generated/cshift0_r17.c (revision b1e838363e3c6fc78a55519254d99869742dd33c)
1*b1e83836Smrg /* Helper function for cshift functions.
2*b1e83836Smrg    Copyright (C) 2008-2022 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 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 
30*b1e83836Smrg #if defined (HAVE_GFC_REAL_17)
31*b1e83836Smrg 
32*b1e83836Smrg void
cshift0_r17(gfc_array_r17 * ret,const gfc_array_r17 * array,ptrdiff_t shift,int which)33*b1e83836Smrg cshift0_r17 (gfc_array_r17 *ret, const gfc_array_r17 *array, ptrdiff_t shift,
34*b1e83836Smrg 		     int which)
35*b1e83836Smrg {
36*b1e83836Smrg   /* r.* indicates the return array.  */
37*b1e83836Smrg   index_type rstride[GFC_MAX_DIMENSIONS];
38*b1e83836Smrg   index_type rstride0;
39*b1e83836Smrg   index_type roffset;
40*b1e83836Smrg   GFC_REAL_17 *rptr;
41*b1e83836Smrg 
42*b1e83836Smrg   /* s.* indicates the source array.  */
43*b1e83836Smrg   index_type sstride[GFC_MAX_DIMENSIONS];
44*b1e83836Smrg   index_type sstride0;
45*b1e83836Smrg   index_type soffset;
46*b1e83836Smrg   const GFC_REAL_17 *sptr;
47*b1e83836Smrg 
48*b1e83836Smrg   index_type count[GFC_MAX_DIMENSIONS];
49*b1e83836Smrg   index_type extent[GFC_MAX_DIMENSIONS];
50*b1e83836Smrg   index_type dim;
51*b1e83836Smrg   index_type len;
52*b1e83836Smrg   index_type n;
53*b1e83836Smrg 
54*b1e83836Smrg   bool do_blocked;
55*b1e83836Smrg   index_type r_ex, a_ex;
56*b1e83836Smrg 
57*b1e83836Smrg   which = which - 1;
58*b1e83836Smrg   sstride[0] = 0;
59*b1e83836Smrg   rstride[0] = 0;
60*b1e83836Smrg 
61*b1e83836Smrg   extent[0] = 1;
62*b1e83836Smrg   count[0] = 0;
63*b1e83836Smrg   n = 0;
64*b1e83836Smrg   /* Initialized for avoiding compiler warnings.  */
65*b1e83836Smrg   roffset = 1;
66*b1e83836Smrg   soffset = 1;
67*b1e83836Smrg   len = 0;
68*b1e83836Smrg 
69*b1e83836Smrg   r_ex = 1;
70*b1e83836Smrg   a_ex = 1;
71*b1e83836Smrg 
72*b1e83836Smrg   if (which > 0)
73*b1e83836Smrg     {
74*b1e83836Smrg       /* Test if both ret and array are contiguous.  */
75*b1e83836Smrg       do_blocked = true;
76*b1e83836Smrg       dim = GFC_DESCRIPTOR_RANK (array);
77*b1e83836Smrg       for (n = 0; n < dim; n ++)
78*b1e83836Smrg 	{
79*b1e83836Smrg 	  index_type rs, as;
80*b1e83836Smrg 	  rs = GFC_DESCRIPTOR_STRIDE (ret, n);
81*b1e83836Smrg 	  if (rs != r_ex)
82*b1e83836Smrg 	    {
83*b1e83836Smrg 	      do_blocked = false;
84*b1e83836Smrg 	      break;
85*b1e83836Smrg 	    }
86*b1e83836Smrg 	  as = GFC_DESCRIPTOR_STRIDE (array, n);
87*b1e83836Smrg 	  if (as != a_ex)
88*b1e83836Smrg 	    {
89*b1e83836Smrg 	      do_blocked = false;
90*b1e83836Smrg 	      break;
91*b1e83836Smrg 	    }
92*b1e83836Smrg 	  r_ex *= GFC_DESCRIPTOR_EXTENT (ret, n);
93*b1e83836Smrg 	  a_ex *= GFC_DESCRIPTOR_EXTENT (array, n);
94*b1e83836Smrg 	}
95*b1e83836Smrg     }
96*b1e83836Smrg   else
97*b1e83836Smrg     do_blocked = false;
98*b1e83836Smrg 
99*b1e83836Smrg   n = 0;
100*b1e83836Smrg 
101*b1e83836Smrg   if (do_blocked)
102*b1e83836Smrg     {
103*b1e83836Smrg       /* For contiguous arrays, use the relationship that
104*b1e83836Smrg 
105*b1e83836Smrg          dimension(n1,n2,n3) :: a, b
106*b1e83836Smrg 	 b = cshift(a,sh,3)
107*b1e83836Smrg 
108*b1e83836Smrg          can be dealt with as if
109*b1e83836Smrg 
110*b1e83836Smrg 	 dimension(n1*n2*n3) :: an, bn
111*b1e83836Smrg 	 bn = cshift(a,sh*n1*n2,1)
112*b1e83836Smrg 
113*b1e83836Smrg 	 we can used a more blocked algorithm for dim>1.  */
114*b1e83836Smrg       sstride[0] = 1;
115*b1e83836Smrg       rstride[0] = 1;
116*b1e83836Smrg       roffset = 1;
117*b1e83836Smrg       soffset = 1;
118*b1e83836Smrg       len = GFC_DESCRIPTOR_STRIDE(array, which)
119*b1e83836Smrg 	* GFC_DESCRIPTOR_EXTENT(array, which);
120*b1e83836Smrg       shift *= GFC_DESCRIPTOR_STRIDE(array, which);
121*b1e83836Smrg       for (dim = which + 1; dim < GFC_DESCRIPTOR_RANK (array); dim++)
122*b1e83836Smrg 	{
123*b1e83836Smrg 	  count[n] = 0;
124*b1e83836Smrg 	  extent[n] = GFC_DESCRIPTOR_EXTENT(array,dim);
125*b1e83836Smrg 	  rstride[n] = GFC_DESCRIPTOR_STRIDE(ret,dim);
126*b1e83836Smrg 	  sstride[n] = GFC_DESCRIPTOR_STRIDE(array,dim);
127*b1e83836Smrg 	  n++;
128*b1e83836Smrg 	}
129*b1e83836Smrg       dim = GFC_DESCRIPTOR_RANK (array) - which;
130*b1e83836Smrg     }
131*b1e83836Smrg   else
132*b1e83836Smrg     {
133*b1e83836Smrg       for (dim = 0; dim < GFC_DESCRIPTOR_RANK (array); dim++)
134*b1e83836Smrg 	{
135*b1e83836Smrg 	  if (dim == which)
136*b1e83836Smrg 	    {
137*b1e83836Smrg 	      roffset = GFC_DESCRIPTOR_STRIDE(ret,dim);
138*b1e83836Smrg 	      if (roffset == 0)
139*b1e83836Smrg 		roffset = 1;
140*b1e83836Smrg 	      soffset = GFC_DESCRIPTOR_STRIDE(array,dim);
141*b1e83836Smrg 	      if (soffset == 0)
142*b1e83836Smrg 		soffset = 1;
143*b1e83836Smrg 	      len = GFC_DESCRIPTOR_EXTENT(array,dim);
144*b1e83836Smrg 	    }
145*b1e83836Smrg 	  else
146*b1e83836Smrg 	    {
147*b1e83836Smrg 	      count[n] = 0;
148*b1e83836Smrg 	      extent[n] = GFC_DESCRIPTOR_EXTENT(array,dim);
149*b1e83836Smrg 	      rstride[n] = GFC_DESCRIPTOR_STRIDE(ret,dim);
150*b1e83836Smrg 	      sstride[n] = GFC_DESCRIPTOR_STRIDE(array,dim);
151*b1e83836Smrg 	      n++;
152*b1e83836Smrg 	    }
153*b1e83836Smrg 	}
154*b1e83836Smrg       if (sstride[0] == 0)
155*b1e83836Smrg 	sstride[0] = 1;
156*b1e83836Smrg       if (rstride[0] == 0)
157*b1e83836Smrg 	rstride[0] = 1;
158*b1e83836Smrg 
159*b1e83836Smrg       dim = GFC_DESCRIPTOR_RANK (array);
160*b1e83836Smrg     }
161*b1e83836Smrg 
162*b1e83836Smrg   rstride0 = rstride[0];
163*b1e83836Smrg   sstride0 = sstride[0];
164*b1e83836Smrg   rptr = ret->base_addr;
165*b1e83836Smrg   sptr = array->base_addr;
166*b1e83836Smrg 
167*b1e83836Smrg   /* Avoid the costly modulo for trivially in-bound shifts.  */
168*b1e83836Smrg   if (shift < 0 || shift >= len)
169*b1e83836Smrg     {
170*b1e83836Smrg       shift = len == 0 ? 0 : shift % (ptrdiff_t)len;
171*b1e83836Smrg       if (shift < 0)
172*b1e83836Smrg 	shift += len;
173*b1e83836Smrg     }
174*b1e83836Smrg 
175*b1e83836Smrg   while (rptr)
176*b1e83836Smrg     {
177*b1e83836Smrg       /* Do the shift for this dimension.  */
178*b1e83836Smrg 
179*b1e83836Smrg       /* If elements are contiguous, perform the operation
180*b1e83836Smrg 	 in two block moves.  */
181*b1e83836Smrg       if (soffset == 1 && roffset == 1)
182*b1e83836Smrg 	{
183*b1e83836Smrg 	  size_t len1 = shift * sizeof (GFC_REAL_17);
184*b1e83836Smrg 	  size_t len2 = (len - shift) * sizeof (GFC_REAL_17);
185*b1e83836Smrg 	  memcpy (rptr, sptr + shift, len2);
186*b1e83836Smrg 	  memcpy (rptr + (len - shift), sptr, len1);
187*b1e83836Smrg 	}
188*b1e83836Smrg       else
189*b1e83836Smrg 	{
190*b1e83836Smrg 	  /* Otherwise, we will have to perform the copy one element at
191*b1e83836Smrg 	     a time.  */
192*b1e83836Smrg 	  GFC_REAL_17 *dest = rptr;
193*b1e83836Smrg 	  const GFC_REAL_17 *src = &sptr[shift * soffset];
194*b1e83836Smrg 
195*b1e83836Smrg 	  for (n = 0; n < len - shift; n++)
196*b1e83836Smrg 	    {
197*b1e83836Smrg 	      *dest = *src;
198*b1e83836Smrg 	      dest += roffset;
199*b1e83836Smrg 	      src += soffset;
200*b1e83836Smrg 	    }
201*b1e83836Smrg 	  for (src = sptr, n = 0; n < shift; n++)
202*b1e83836Smrg 	    {
203*b1e83836Smrg 	      *dest = *src;
204*b1e83836Smrg 	      dest += roffset;
205*b1e83836Smrg 	      src += soffset;
206*b1e83836Smrg 	    }
207*b1e83836Smrg 	}
208*b1e83836Smrg 
209*b1e83836Smrg       /* Advance to the next section.  */
210*b1e83836Smrg       rptr += rstride0;
211*b1e83836Smrg       sptr += sstride0;
212*b1e83836Smrg       count[0]++;
213*b1e83836Smrg       n = 0;
214*b1e83836Smrg       while (count[n] == extent[n])
215*b1e83836Smrg         {
216*b1e83836Smrg           /* When we get to the end of a dimension, reset it and increment
217*b1e83836Smrg              the next dimension.  */
218*b1e83836Smrg           count[n] = 0;
219*b1e83836Smrg           /* We could precalculate these products, but this is a less
220*b1e83836Smrg              frequently used path so probably not worth it.  */
221*b1e83836Smrg           rptr -= rstride[n] * extent[n];
222*b1e83836Smrg           sptr -= sstride[n] * extent[n];
223*b1e83836Smrg           n++;
224*b1e83836Smrg           if (n >= dim - 1)
225*b1e83836Smrg             {
226*b1e83836Smrg               /* Break out of the loop.  */
227*b1e83836Smrg               rptr = NULL;
228*b1e83836Smrg               break;
229*b1e83836Smrg             }
230*b1e83836Smrg           else
231*b1e83836Smrg             {
232*b1e83836Smrg               count[n]++;
233*b1e83836Smrg               rptr += rstride[n];
234*b1e83836Smrg               sptr += sstride[n];
235*b1e83836Smrg             }
236*b1e83836Smrg         }
237*b1e83836Smrg     }
238*b1e83836Smrg 
239*b1e83836Smrg   return;
240*b1e83836Smrg }
241*b1e83836Smrg 
242*b1e83836Smrg #endif
243