libgfortran/generated/matmul_l4.c

   1 /* Implementation of the MATMUL intrinsic
   2    Copyright 2002, 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 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
  36 #if defined (HAVE_GFC_LOGICAL_4)
  37
  38 /* Dimensions: retarray(x,y) a(x, count) b(count,y).
  39    Either a or b can be rank 1.  In this case x or y is 1.  */
  40
  41 extern void matmul_l4 (gfc_array_l4 * const restrict,
  42         gfc_array_l4 * const restrict, gfc_array_l4 * const restrict);
  43 export_proto(matmul_l4);
  44
  45 void
  46 matmul_l4 (gfc_array_l4 * const restrict retarray,
  47         gfc_array_l4 * const restrict a, gfc_array_l4 * const restrict b)
  48 {
  49   const GFC_INTEGER_4 * restrict abase;
  50   const GFC_INTEGER_4 * restrict bbase;
  51   GFC_LOGICAL_4 * restrict dest;
  52   index_type rxstride;
  53   index_type rystride;
  54   index_type xcount;
  55   index_type ycount;
  56   index_type xstride;
  57   index_type ystride;
  58   index_type x;
  59   index_type y;
  60
  61   const GFC_INTEGER_4 * restrict pa;
  62   const GFC_INTEGER_4 * restrict pb;
  63   index_type astride;
  64   index_type bstride;
  65   index_type count;
  66   index_type n;
  67
  68   assert (GFC_DESCRIPTOR_RANK (a) == 2
  69           || GFC_DESCRIPTOR_RANK (b) == 2);
  70
  71   if (retarray->data == NULL)
  72     {
  73       if (GFC_DESCRIPTOR_RANK (a) == 1)
  74         {
  75           retarray->dim[0].lbound = 0;
  76           retarray->dim[0].ubound = b->dim[1].ubound - b->dim[1].lbound;
  77           retarray->dim[0].stride = 1;
  78         }
  79       else if (GFC_DESCRIPTOR_RANK (b) == 1)
  80         {
  81           retarray->dim[0].lbound = 0;
  82           retarray->dim[0].ubound = a->dim[0].ubound - a->dim[0].lbound;
  83           retarray->dim[0].stride = 1;
  84         }
  85       else
  86         {
  87           retarray->dim[0].lbound = 0;
  88           retarray->dim[0].ubound = a->dim[0].ubound - a->dim[0].lbound;
  89           retarray->dim[0].stride = 1;
  90
  91           retarray->dim[1].lbound = 0;
  92           retarray->dim[1].ubound = b->dim[1].ubound - b->dim[1].lbound;
  93           retarray->dim[1].stride = retarray->dim[0].ubound+1;
  94         }
  95
  96       retarray->data
  97         = internal_malloc_size (sizeof (GFC_LOGICAL_4) * size0 ((array_t *) retarray));
  98       retarray->offset = 0;
  99     }
 100
 101   abase = a->data;
 102   if (GFC_DESCRIPTOR_SIZE (a) != 4)
 103     {
 104       assert (GFC_DESCRIPTOR_SIZE (a) == 8);
 105       abase = GFOR_POINTER_L8_TO_L4 (abase);
 106     }
 107   bbase = b->data;
 108   if (GFC_DESCRIPTOR_SIZE (b) != 4)
 109     {
 110       assert (GFC_DESCRIPTOR_SIZE (b) == 8);
 111       bbase = GFOR_POINTER_L8_TO_L4 (bbase);
 112     }
 113   dest = retarray->data;
 114
 115
 116   if (GFC_DESCRIPTOR_RANK (retarray) == 1)
 117     {
 118       rxstride = retarray->dim[0].stride;
 119       rystride = rxstride;
 120     }
 121   else
 122     {
 123       rxstride = retarray->dim[0].stride;
 124       rystride = retarray->dim[1].stride;
 125     }
 126
 127   /* If we have rank 1 parameters, zero the absent stride, and set the size to
 128      one.  */
 129   if (GFC_DESCRIPTOR_RANK (a) == 1)
 130     {
 131       astride = a->dim[0].stride;
 132       count = a->dim[0].ubound + 1 - a->dim[0].lbound;
 133       xstride = 0;
 134       rxstride = 0;
 135       xcount = 1;
 136     }
 137   else
 138     {
 139       astride = a->dim[1].stride;
 140       count = a->dim[1].ubound + 1 - a->dim[1].lbound;
 141       xstride = a->dim[0].stride;
 142       xcount = a->dim[0].ubound + 1 - a->dim[0].lbound;
 143     }
 144   if (GFC_DESCRIPTOR_RANK (b) == 1)
 145     {
 146       bstride = b->dim[0].stride;
 147       assert(count == b->dim[0].ubound + 1 - b->dim[0].lbound);
 148       ystride = 0;
 149       rystride = 0;
 150       ycount = 1;
 151     }
 152   else
 153     {
 154       bstride = b->dim[0].stride;
 155       assert(count == b->dim[0].ubound + 1 - b->dim[0].lbound);
 156       ystride = b->dim[1].stride;
 157       ycount = b->dim[1].ubound + 1 - b->dim[1].lbound;
 158     }
 159
 160   for (y = 0; y < ycount; y++)
 161     {
 162       for (x = 0; x < xcount; x++)
 163         {
 164           /* Do the summation for this element.  For real and integer types
 165              this is the same as DOT_PRODUCT.  For complex types we use do
 166              a*b, not conjg(a)*b.  */
 167           pa = abase;
 168           pb = bbase;
 169           *dest = 0;
 170
 171           for (n = 0; n < count; n++)
 172             {
 173               if (*pa && *pb)
 174                 {
 175                   *dest = 1;
 176                   break;
 177                 }
 178               pa += astride;
 179               pb += bstride;
 180             }
 181
 182           dest += rxstride;
 183           abase += xstride;
 184         }
 185       abase -= xstride * xcount;
 186       bbase += ystride;
 187       dest += rystride - (rxstride * xcount);
 188     }
 189 }
 190
 191 #endif