libgfortran/generated/matmul_l4.c

   1 /* Implementation of the MATMUL intrinsic
   2    Copyright 2002, 2005, 2006, 2007 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 "libgfortran.h"
  32 #include <stdlib.h>
  33 #include <assert.h>
  34
  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_l1 * const restrict, gfc_array_l1 * const restrict);
  43 export_proto(matmul_l4);
  44
  45 void
  46 matmul_l4 (gfc_array_l4 * const restrict retarray,
  47         gfc_array_l1 * const restrict a, gfc_array_l1 * const restrict b)
  48 {
  49   const GFC_LOGICAL_1 * restrict abase;
  50   const GFC_LOGICAL_1 * 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   int a_kind;
  61   int b_kind;
  62
  63   const GFC_LOGICAL_1 * restrict pa;
  64   const GFC_LOGICAL_1 * restrict pb;
  65   index_type astride;
  66   index_type bstride;
  67   index_type count;
  68   index_type n;
  69
  70   assert (GFC_DESCRIPTOR_RANK (a) == 2
  71           || GFC_DESCRIPTOR_RANK (b) == 2);
  72
  73   if (retarray->data == NULL)
  74     {
  75       if (GFC_DESCRIPTOR_RANK (a) == 1)
  76         {
  77           retarray->dim[0].lbound = 0;
  78           retarray->dim[0].ubound = b->dim[1].ubound - b->dim[1].lbound;
  79           retarray->dim[0].stride = 1;
  80         }
  81       else if (GFC_DESCRIPTOR_RANK (b) == 1)
  82         {
  83           retarray->dim[0].lbound = 0;
  84           retarray->dim[0].ubound = a->dim[0].ubound - a->dim[0].lbound;
  85           retarray->dim[0].stride = 1;
  86         }
  87       else
  88         {
  89           retarray->dim[0].lbound = 0;
  90           retarray->dim[0].ubound = a->dim[0].ubound - a->dim[0].lbound;
  91           retarray->dim[0].stride = 1;
  92
  93           retarray->dim[1].lbound = 0;
  94           retarray->dim[1].ubound = b->dim[1].ubound - b->dim[1].lbound;
  95           retarray->dim[1].stride = retarray->dim[0].ubound+1;
  96         }
  97
  98       retarray->data
  99         = internal_malloc_size (sizeof (GFC_LOGICAL_4) * size0 ((array_t *) retarray));
 100       retarray->offset = 0;
 101     }
 102
 103   abase = a->data;
 104   a_kind = GFC_DESCRIPTOR_SIZE (a);
 105
 106   if (a_kind == 1 || a_kind == 2 || a_kind == 4 || a_kind == 8
 107 #ifdef HAVE_GFC_LOGICAL_16
 108      || a_kind == 16
 109 #endif
 110      )
 111     abase = GFOR_POINTER_TO_L1 (abase, a_kind);
 112   else
 113     internal_error (NULL, "Funny sized logical array");
 114
 115   bbase = b->data;
 116   b_kind = GFC_DESCRIPTOR_SIZE (b);
 117
 118   if (b_kind == 1 || b_kind == 2 || b_kind == 4 || b_kind == 8
 119 #ifdef HAVE_GFC_LOGICAL_16
 120      || b_kind == 16
 121 #endif
 122      )
 123     bbase = GFOR_POINTER_TO_L1 (bbase, b_kind);
 124   else
 125     internal_error (NULL, "Funny sized logical array");
 126
 127   dest = retarray->data;
 128
 129
 130   if (GFC_DESCRIPTOR_RANK (retarray) == 1)
 131     {
 132       rxstride = retarray->dim[0].stride;
 133       rystride = rxstride;
 134     }
 135   else
 136     {
 137       rxstride = retarray->dim[0].stride;
 138       rystride = retarray->dim[1].stride;
 139     }
 140
 141   /* If we have rank 1 parameters, zero the absent stride, and set the size to
 142      one.  */
 143   if (GFC_DESCRIPTOR_RANK (a) == 1)
 144     {
 145       astride = a->dim[0].stride * a_kind;
 146       count = a->dim[0].ubound + 1 - a->dim[0].lbound;
 147       xstride = 0;
 148       rxstride = 0;
 149       xcount = 1;
 150     }
 151   else
 152     {
 153       astride = a->dim[1].stride * a_kind;
 154       count = a->dim[1].ubound + 1 - a->dim[1].lbound;
 155       xstride = a->dim[0].stride;
 156       xcount = a->dim[0].ubound + 1 - a->dim[0].lbound;
 157     }
 158   if (GFC_DESCRIPTOR_RANK (b) == 1)
 159     {
 160       bstride = b->dim[0].stride * b_kind;
 161       assert(count == b->dim[0].ubound + 1 - b->dim[0].lbound);
 162       ystride = 0;
 163       rystride = 0;
 164       ycount = 1;
 165     }
 166   else
 167     {
 168       bstride = b->dim[0].stride * b_kind;
 169       assert(count == b->dim[0].ubound + 1 - b->dim[0].lbound);
 170       ystride = b->dim[1].stride;
 171       ycount = b->dim[1].ubound + 1 - b->dim[1].lbound;
 172     }
 173
 174   for (y = 0; y < ycount; y++)
 175     {
 176       for (x = 0; x < xcount; x++)
 177         {
 178           /* Do the summation for this element.  For real and integer types
 179              this is the same as DOT_PRODUCT.  For complex types we use do
 180              a*b, not conjg(a)*b.  */
 181           pa = abase;
 182           pb = bbase;
 183           *dest = 0;
 184
 185           for (n = 0; n < count; n++)
 186             {
 187               if (*pa && *pb)
 188                 {
 189                   *dest = 1;
 190                   break;
 191                 }
 192               pa += astride;
 193               pb += bstride;
 194             }
 195
 196           dest += rxstride;
 197           abase += xstride;
 198         }
 199       abase -= xstride * xcount;
 200       bbase += ystride;
 201       dest += rystride - (rxstride * xcount);
 202     }
 203 }
 204
 205 #endif
 206