libgfortran/intrinsics/pack_generic.c

   1 /* Generic implementation of the PACK intrinsic
   2    Copyright (C) 2002, 2004, 2005, 2006 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 Ligbfortran 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 <string.h>
  35 #include "libgfortran.h"
  36
  37 /* PACK is specified as follows:
  38
  39    13.14.80 PACK (ARRAY, MASK, [VECTOR])
  40
  41    Description: Pack an array into an array of rank one under the
  42    control of a mask.
  43
  44    Class: Transformational function.
  45
  46    Arguments:
  47       ARRAY   may be of any type. It shall not be scalar.
  48       MASK    shall be of type LOGICAL. It shall be conformable with ARRAY.
  49       VECTOR  (optional) shall be of the same type and type parameters
  50               as ARRAY. VECTOR shall have at least as many elements as
  51               there are true elements in MASK. If MASK is a scalar
  52               with the value true, VECTOR shall have at least as many
  53               elements as there are in ARRAY.
  54
  55    Result Characteristics: The result is an array of rank one with the
  56    same type and type parameters as ARRAY. If VECTOR is present, the
  57    result size is that of VECTOR; otherwise, the result size is the
  58    number /t/ of true elements in MASK unless MASK is scalar with the
  59    value true, in which case the result size is the size of ARRAY.
  60
  61    Result Value: Element /i/ of the result is the element of ARRAY
  62    that corresponds to the /i/th true element of MASK, taking elements
  63    in array element order, for /i/ = 1, 2, ..., /t/. If VECTOR is
  64    present and has size /n/ > /t/, element /i/ of the result has the
  65    value VECTOR(/i/), for /i/ = /t/ + 1, ..., /n/.
  66
  67    Examples: The nonzero elements of an array M with the value
  68    | 0 0 0 |
  69    | 9 0 0 | may be "gathered" by the function PACK. The result of
  70    | 0 0 7 |
  71    PACK (M, MASK = M.NE.0) is [9,7] and the result of PACK (M, M.NE.0,
  72    VECTOR = (/ 2,4,6,8,10,12 /)) is [9,7,6,8,10,12].
  73
  74 There are two variants of the PACK intrinsic: one, where MASK is
  75 array valued, and the other one where MASK is scalar.  */
  76
  77 static void
  78 pack_internal (gfc_array_char *ret, const gfc_array_char *array,
  79                const gfc_array_l4 *mask, const gfc_array_char *vector,
  80                index_type size)
  81 {
  82   /* r.* indicates the return array.  */
  83   index_type rstride0;
  84   char *rptr;
  85   /* s.* indicates the source array.  */
  86   index_type sstride[GFC_MAX_DIMENSIONS];
  87   index_type sstride0;
  88   const char *sptr;
  89   /* m.* indicates the mask array.  */
  90   index_type mstride[GFC_MAX_DIMENSIONS];
  91   index_type mstride0;
  92   const GFC_LOGICAL_4 *mptr;
  93
  94   index_type count[GFC_MAX_DIMENSIONS];
  95   index_type extent[GFC_MAX_DIMENSIONS];
  96   int zero_sized;
  97   index_type n;
  98   index_type dim;
  99   index_type nelem;
 100   index_type total;
 101
 102   dim = GFC_DESCRIPTOR_RANK (array);
 103   zero_sized = 0;
 104   for (n = 0; n < dim; n++)
 105     {
 106       count[n] = 0;
 107       extent[n] = array->dim[n].ubound + 1 - array->dim[n].lbound;
 108       if (extent[n] <= 0)
 109        zero_sized = 1;
 110       sstride[n] = array->dim[n].stride * size;
 111       mstride[n] = mask->dim[n].stride;
 112     }
 113   if (sstride[0] == 0)
 114     sstride[0] = size;
 115   if (mstride[0] == 0)
 116     mstride[0] = 1;
 117
 118   sptr = array->data;
 119   mptr = mask->data;
 120
 121   /* Use the same loop for both logical types. */
 122   if (GFC_DESCRIPTOR_SIZE (mask) != 4)
 123     {
 124       if (GFC_DESCRIPTOR_SIZE (mask) != 8)
 125         runtime_error ("Funny sized logical array");
 126       for (n = 0; n < dim; n++)
 127         mstride[n] <<= 1;
 128       mptr = GFOR_POINTER_L8_TO_L4 (mptr);
 129     }
 130
 131   if (ret->data == NULL || compile_options.bounds_check)
 132     {
 133       /* Count the elements, either for allocating memory or
 134          for bounds checking.  */
 135
 136       if (vector != NULL)
 137         {
 138           /* The return array will have as many
 139              elements as there are in VECTOR.  */
 140           total = vector->dim[0].ubound + 1 - vector->dim[0].lbound;
 141         }
 142       else
 143         {
 144           /* We have to count the true elements in MASK.  */
 145
 146           /* TODO: We could speed up pack easily in the case of only
 147              few .TRUE. entries in MASK, by keeping track of where we
 148              would be in the source array during the initial traversal
 149              of MASK, and caching the pointers to those elements. Then,
 150              supposed the number of elements is small enough, we would
 151              only have to traverse the list, and copy those elements
 152              into the result array. In the case of datatypes which fit
 153              in one of the integer types we could also cache the
 154              value instead of a pointer to it.
 155              This approach might be bad from the point of view of
 156              cache behavior in the case where our cache is not big
 157              enough to hold all elements that have to be copied.  */
 158
 159           const GFC_LOGICAL_4 *m = mptr;
 160
 161           total = 0;
 162           if (zero_sized)
 163             m = NULL;
 164
 165           while (m)
 166             {
 167               /* Test this element.  */
 168               if (*m)
 169                 total++;
 170
 171               /* Advance to the next element.  */
 172               m += mstride[0];
 173               count[0]++;
 174               n = 0;
 175               while (count[n] == extent[n])
 176                 {
 177                   /* When we get to the end of a dimension, reset it
 178                      and increment the next dimension.  */
 179                   count[n] = 0;
 180                   /* We could precalculate this product, but this is a
 181                      less frequently used path so probably not worth
 182                      it.  */
 183                   m -= mstride[n] * extent[n];
 184                   n++;
 185                   if (n >= dim)
 186                     {
 187                       /* Break out of the loop.  */
 188                       m = NULL;
 189                       break;
 190                     }
 191                   else
 192                     {
 193                       count[n]++;
 194                       m += mstride[n];
 195                     }
 196                 }
 197             }
 198         }
 199
 200       if (ret->data == NULL)
 201         {
 202           /* Setup the array descriptor.  */
 203           ret->dim[0].lbound = 0;
 204           ret->dim[0].ubound = total - 1;
 205           ret->dim[0].stride = 1;
 206
 207           ret->offset = 0;
 208           if (total == 0)
 209             {
 210               /* In this case, nothing remains to be done.  */
 211               ret->data = internal_malloc_size (1);
 212               return;
 213             }
 214           else
 215             ret->data = internal_malloc_size (size * total);
 216         }
 217       else
 218         {
 219           /* We come here because of range checking.  */
 220           if (total != ret->dim[0].ubound + 1 - ret->dim[0].lbound)
 221             runtime_error ("Incorrect extent in return value of"
 222                            " PACK intrinsic");
 223         }
 224     }
 225
 226   rstride0 = ret->dim[0].stride * size;
 227   if (rstride0 == 0)
 228     rstride0 = size;
 229   sstride0 = sstride[0];
 230   mstride0 = mstride[0];
 231   rptr = ret->data;
 232
 233   while (sptr && mptr)
 234     {
 235       /* Test this element.  */
 236       if (*mptr)
 237         {
 238           /* Add it.  */
 239           memcpy (rptr, sptr, size);
 240           rptr += rstride0;
 241         }
 242       /* Advance to the next element.  */
 243       sptr += sstride0;
 244       mptr += mstride0;
 245       count[0]++;
 246       n = 0;
 247       while (count[n] == extent[n])
 248         {
 249           /* When we get to the end of a dimension, reset it and increment
 250              the next dimension.  */
 251           count[n] = 0;
 252           /* We could precalculate these products, but this is a less
 253              frequently used path so probably not worth it.  */
 254           sptr -= sstride[n] * extent[n];
 255           mptr -= mstride[n] * extent[n];
 256           n++;
 257           if (n >= dim)
 258             {
 259               /* Break out of the loop.  */
 260               sptr = NULL;
 261               break;
 262             }
 263           else
 264             {
 265               count[n]++;
 266               sptr += sstride[n];
 267               mptr += mstride[n];
 268             }
 269         }
 270     }
 271
 272   /* Add any remaining elements from VECTOR.  */
 273   if (vector)
 274     {
 275       n = vector->dim[0].ubound + 1 - vector->dim[0].lbound;
 276       nelem = ((rptr - ret->data) / rstride0);
 277       if (n > nelem)
 278         {
 279           sstride0 = vector->dim[0].stride * size;
 280           if (sstride0 == 0)
 281             sstride0 = size;
 282
 283           sptr = vector->data + sstride0 * nelem;
 284           n -= nelem;
 285           while (n--)
 286             {
 287               memcpy (rptr, sptr, size);
 288               rptr += rstride0;
 289               sptr += sstride0;
 290             }
 291         }
 292     }
 293 }
 294
 295 extern void pack (gfc_array_char *, const gfc_array_char *,
 296                   const gfc_array_l4 *, const gfc_array_char *);
 297 export_proto(pack);
 298
 299 void
 300 pack (gfc_array_char *ret, const gfc_array_char *array,
 301       const gfc_array_l4 *mask, const gfc_array_char *vector)
 302 {
 303   pack_internal (ret, array, mask, vector, GFC_DESCRIPTOR_SIZE (array));
 304 }
 305
 306 extern void pack_char (gfc_array_char *, GFC_INTEGER_4, const gfc_array_char *,
 307                        const gfc_array_l4 *, const gfc_array_char *,
 308                        GFC_INTEGER_4, GFC_INTEGER_4);
 309 export_proto(pack_char);
 310
 311 void
 312 pack_char (gfc_array_char *ret,
 313            GFC_INTEGER_4 ret_length __attribute__((unused)),
 314            const gfc_array_char *array, const gfc_array_l4 *mask,
 315            const gfc_array_char *vector, GFC_INTEGER_4 array_length,
 316            GFC_INTEGER_4 vector_length __attribute__((unused)))
 317 {
 318   pack_internal (ret, array, mask, vector, array_length);
 319 }
 320
 321 static void
 322 pack_s_internal (gfc_array_char *ret, const gfc_array_char *array,
 323                  const GFC_LOGICAL_4 *mask, const gfc_array_char *vector,
 324                  index_type size)
 325 {
 326   /* r.* indicates the return array.  */
 327   index_type rstride0;
 328   char *rptr;
 329   /* s.* indicates the source array.  */
 330   index_type sstride[GFC_MAX_DIMENSIONS];
 331   index_type sstride0;
 332   const char *sptr;
 333
 334   index_type count[GFC_MAX_DIMENSIONS];
 335   index_type extent[GFC_MAX_DIMENSIONS];
 336   index_type n;
 337   index_type dim;
 338   index_type ssize;
 339   index_type nelem;
 340
 341   dim = GFC_DESCRIPTOR_RANK (array);
 342   ssize = 1;
 343   for (n = 0; n < dim; n++)
 344     {
 345       count[n] = 0;
 346       extent[n] = array->dim[n].ubound + 1 - array->dim[n].lbound;
 347       sstride[n] = array->dim[n].stride * size;
 348       ssize *= extent[n];
 349     }
 350   if (sstride[0] == 0)
 351     sstride[0] = size;
 352
 353   sstride0 = sstride[0];
 354   sptr = array->data;
 355
 356   if (ret->data == NULL)
 357     {
 358       /* Allocate the memory for the result.  */
 359       int total;
 360
 361       if (vector != NULL)
 362         {
 363           /* The return array will have as many elements as there are
 364              in vector.  */
 365           total = vector->dim[0].ubound + 1 - vector->dim[0].lbound;
 366         }
 367       else
 368         {
 369           if (*mask)
 370             {
 371               /* The result array will have as many elements as the input
 372                  array.  */
 373               total = extent[0];
 374               for (n = 1; n < dim; n++)
 375                 total *= extent[n];
 376             }
 377           else
 378             /* The result array will be empty.  */
 379             total = 0;
 380         }
 381
 382       /* Setup the array descriptor.  */
 383       ret->dim[0].lbound = 0;
 384       ret->dim[0].ubound = total - 1;
 385       ret->dim[0].stride = 1;
 386       ret->offset = 0;
 387
 388       if (total == 0)
 389         {
 390           ret->data = internal_malloc_size (1);
 391           return;
 392         }
 393       else
 394         ret->data = internal_malloc_size (size * total);
 395     }
 396
 397   rstride0 = ret->dim[0].stride * size;
 398   if (rstride0 == 0)
 399     rstride0 = size;
 400   rptr = ret->data;
 401
 402   /* The remaining possibilities are now:
 403        If MASK is .TRUE., we have to copy the source array into the
 404      result array. We then have to fill it up with elements from VECTOR.
 405        If MASK is .FALSE., we have to copy VECTOR into the result
 406      array. If VECTOR were not present we would have already returned.  */
 407
 408   if (*mask && ssize != 0)
 409     {
 410       while (sptr)
 411         {
 412           /* Add this element.  */
 413           memcpy (rptr, sptr, size);
 414           rptr += rstride0;
 415
 416           /* Advance to the next element.  */
 417           sptr += sstride0;
 418           count[0]++;
 419           n = 0;
 420           while (count[n] == extent[n])
 421             {
 422               /* When we get to the end of a dimension, reset it and
 423                  increment the next dimension.  */
 424               count[n] = 0;
 425               /* We could precalculate these products, but this is a
 426                  less frequently used path so probably not worth it.  */
 427               sptr -= sstride[n] * extent[n];
 428               n++;
 429               if (n >= dim)
 430                 {
 431                   /* Break out of the loop.  */
 432                   sptr = NULL;
 433                   break;
 434                 }
 435               else
 436                 {
 437                   count[n]++;
 438                   sptr += sstride[n];
 439                 }
 440             }
 441         }
 442     }
 443
 444   /* Add any remaining elements from VECTOR.  */
 445   if (vector)
 446     {
 447       n = vector->dim[0].ubound + 1 - vector->dim[0].lbound;
 448       nelem = ((rptr - ret->data) / rstride0);
 449       if (n > nelem)
 450         {
 451           sstride0 = vector->dim[0].stride * size;
 452           if (sstride0 == 0)
 453             sstride0 = size;
 454
 455           sptr = vector->data + sstride0 * nelem;
 456           n -= nelem;
 457           while (n--)
 458             {
 459               memcpy (rptr, sptr, size);
 460               rptr += rstride0;
 461               sptr += sstride0;
 462             }
 463         }
 464     }
 465 }
 466
 467 extern void pack_s (gfc_array_char *ret, const gfc_array_char *array,
 468                     const GFC_LOGICAL_4 *, const gfc_array_char *);
 469 export_proto(pack_s);
 470
 471 void
 472 pack_s (gfc_array_char *ret, const gfc_array_char *array,
 473         const GFC_LOGICAL_4 *mask, const gfc_array_char *vector)
 474 {
 475   pack_s_internal (ret, array, mask, vector, GFC_DESCRIPTOR_SIZE (array));
 476 }
 477
 478 extern void pack_s_char (gfc_array_char *ret, GFC_INTEGER_4,
 479                          const gfc_array_char *array, const GFC_LOGICAL_4 *,
 480                          const gfc_array_char *, GFC_INTEGER_4,
 481                          GFC_INTEGER_4);
 482 export_proto(pack_s_char);
 483
 484 void
 485 pack_s_char (gfc_array_char *ret,
 486              GFC_INTEGER_4 ret_length __attribute__((unused)),
 487              const gfc_array_char *array, const GFC_LOGICAL_4 *mask,
 488              const gfc_array_char *vector, GFC_INTEGER_4 array_length,
 489              GFC_INTEGER_4 vector_length __attribute__((unused)))
 490 {
 491   pack_s_internal (ret, array, mask, vector, array_length);
 492 }