mdw@git.distorted.org.uk Git - catacomb/blob - math/gfx-sqr.c

   1 /* -*-c-*-
   2  *
   3  * Sqaring binary polynomials
   4  *
   5  * (c) 2000 Straylight/Edgeware
   6  */
   7
   8 /*----- Licensing notice --------------------------------------------------*
   9  *
  10  * This file is part of Catacomb.
  11  *
  12  * Catacomb is free software; you can redistribute it and/or modify
  13  * it under the terms of the GNU Library General Public License as
  14  * published by the Free Software Foundation; either version 2 of the
  15  * License, or (at your option) any later version.
  16  *
  17  * Catacomb is distributed in the hope that it will be useful,
  18  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  19  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  20  * GNU Library General Public License for more details.
  21  *
  22  * You should have received a copy of the GNU Library General Public
  23  * License along with Catacomb; if not, write to the Free
  24  * Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
  25  * MA 02111-1307, USA.
  26  */
  27
  28 /*----- Header files ------------------------------------------------------*/
  29
  30 #include "mpx.h"
  31 #include "gfx.h"
  32 #include "permute.h"
  33
  34 /*----- Main code ---------------------------------------------------------*/
  35
  36 /* --- @gfx_sqr@ --- *
  37  *
  38  * Arguments:   @mpw *dv, *dvl@ = destination vector base and limit
  39  *              @const mpw *av, *avl@ = argument vector base and limit
  40  *
  41  * Returns:     ---
  42  *
  43  * Use:         Performs squaring of binary polynomials.
  44  */
  45
  46 void gfx_sqr(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl)
  47 {
  48 #if MPW_BITS <= 16
  49 #  define REGWD 16
  50 #elif MPW_BITS <= 32
  51 #  define REGWD 32
  52 #elif MPW_BITS <= 64
  53 #  define REGWD 64
  54 #elif MPW_BITS <= 128
  55 #  define REGWD 128
  56 #else
  57 #  error "unsupported limb width: extend `gfx-sqr.c'"
  58 #endif
  59
  60 #if MPW_BITS == REGWD
  61   typedef mpw regty;
  62 #else
  63   typedef mpd regty;
  64 #endif
  65
  66   regty a, t;
  67
  68   /* --- Simple stuff --- */
  69
  70   if (dv >= dvl)
  71     return;
  72   MPX_SHRINK(av, avl);
  73
  74   /* --- The main algorithm --- *
  75    *
  76    * Our method depends on the fact that, in a field of characteristic 2, we
  77    * have that %$(a + b)^2 = a^2 + b^2$%.  Thus, to square a polynomial, it's
  78    * sufficient just to put a zero bit between each of the bits of the
  79    * original argument.
  80    */
  81
  82   while (av < avl) {
  83     a = *av++;
  84                                         /* ..., 7, 6, 5, 4, 3, 2, 1, 0 */
  85     SWIZZLE_EXCH(a, 0, 1);              /* ..., 7, 6, 5, 4, 3, 2, 0, 1 */
  86     SWIZZLE_EXCH(a, 0, 2);              /* ..., 7, 6, 5, 4, 3, 1, 0, 2 */
  87     SWIZZLE_EXCH(a, 0, 3);              /* ..., 7, 6, 5, 4, 2, 1, 0, 3 */
  88 #if MPW_BITS > 16
  89     SWIZZLE_EXCH(a, 0, 4);              /* ..., 7, 6, 5, 3, 2, 1, 0, 4 */
  90 #endif
  91 #if MPW_BITS > 32
  92     SWIZZLE_EXCH(a, 0, 5);              /* ..., 7, 6, 4, 3, 2, 1, 0, 5 */
  93 #endif
  94 #if MPW_BITS > 64
  95     SWIZZLE_EXCH(a, 0, 6);              /* ..., 7, 5, 4, 3, 2, 1, 0, 6 */
  96 #endif
  97 #if MPW_BITS > 128
  98 #  error "unsupported limb width: extend `gfx-sqr.c'"
  99 #endif
 100
 101     /* Write out the low half, which consists of the low-order bits in
 102      * even-numbered positions.
 103      */
 104     *dv++ = MPW(a&IXMASK(0)); if (dv >= dvl) break;
 105
 106     /* Write the high half.  This is trickier, since, in general, the bits
 107      * are split across even-numbered bits in the high part of the wider
 108      * register and odd-numbered bits in the low part.  Alas, we can't
 109      * resolve this mess without special cases because shifts are broken in
 110      * C.
 111      */
 112     t = ((a >> 1)&IXMASK(0)) << (REGWD - MPW_BITS);
 113 #if MPW_BITS < REGWD
 114     t |= (a&IXMASK(0)) >> MPW_BITS;
 115 #endif
 116     *dv++ = MPW(t); if (dv >= dvl) break;
 117   }
 118
 119   /* --- Zero the rest of everything --- */
 120
 121   MPX_ZERO(dv, dvl);
 122
 123 #undef REGWD
 124 }
 125
 126 /*----- Test rig ----------------------------------------------------------*/
 127
 128 #ifdef TEST_RIG
 129
 130 #include <mLib/alloc.h>
 131 #include <mLib/dstr.h>
 132 #include <mLib/quis.h>
 133 #include <mLib/testrig.h>
 134
 135 #define ALLOC(v, vl, sz) do {                                           \
 136   size_t _sz = (sz);                                                    \
 137   mpw *_vv = xmalloc(MPWS(_sz));                                        \
 138   mpw *_vvl = _vv + _sz;                                                \
 139   (v) = _vv;                                                            \
 140   (vl) = _vvl;                                                          \
 141 } while (0)
 142
 143 #define LOAD(v, vl, d) do {                                             \
 144   const dstr *_d = (d);                                                 \
 145   mpw *_v, *_vl;                                                        \
 146   ALLOC(_v, _vl, MPW_RQ(_d->len));                                      \
 147   mpx_loadb(_v, _vl, _d->buf, _d->len);                                 \
 148   (v) = _v;                                                             \
 149   (vl) = _vl;                                                           \
 150 } while (0)
 151
 152 #define MAX(x, y) ((x) > (y) ? (x) : (y))
 153
 154 static void dumpmp(const char *msg, const mpw *v, const mpw *vl)
 155 {
 156   fputs(msg, stderr);
 157   MPX_SHRINK(v, vl);
 158   while (v < vl)
 159     fprintf(stderr, " %08lx", (unsigned long)*--vl);
 160   fputc('\n', stderr);
 161 }
 162
 163 static int vsqr(dstr *v)
 164 {
 165   mpw *a, *al;
 166   mpw *b, *bl;
 167   mpw *d, *dl;
 168   int ok = 1;
 169
 170   LOAD(a, al, &v[0]);
 171   LOAD(b, bl, &v[1]);
 172   ALLOC(d, dl, 2 * (al - a));
 173
 174   gfx_sqr(d, dl, a, al);
 175   if (!mpx_ueq(d, dl, b, bl)) {
 176     fprintf(stderr, "\n*** vsqr failed\n");
 177     dumpmp("       a", a, al);
 178     dumpmp("expected", b, bl);
 179     dumpmp("  result", d, dl);
 180     ok = 0;
 181   }
 182
 183   xfree(a); xfree(b); xfree(d);
 184   return (ok);
 185 }
 186
 187 static test_chunk defs[] = {
 188   { "sqr", vsqr, { &type_hex, &type_hex, 0 } },
 189   { 0, 0, { 0 } }
 190 };
 191
 192 int main(int argc, char *argv[])
 193 {
 194   test_run(argc, argv, defs, SRCDIR"/t/gfx");
 195   return (0);
 196 }
 197
 198 #endif
 199
 200 /*----- That's all, folks -------------------------------------------------*/