/* -*-c-*-
*
- * $Id: mpx-kmul.c,v 1.4 2000/06/17 11:42:11 mdw Exp $
+ * $Id: mpx-kmul.c,v 1.10 2004/04/08 01:36:15 mdw Exp $
*
* Karatsuba's multiplication algorithm
*
* MA 02111-1307, USA.
*/
-/*----- Revision history --------------------------------------------------*
- *
- * $Log: mpx-kmul.c,v $
- * Revision 1.4 2000/06/17 11:42:11 mdw
- * Moved the Karatsuba macros into a separate file for better sharing.
- * Fixed some comments.
- *
- * Revision 1.3 1999/12/13 15:35:01 mdw
- * Simplify and improve.
- *
- * Revision 1.2 1999/12/11 10:58:02 mdw
- * Remove tweakable comments.
- *
- * Revision 1.1 1999/12/10 23:23:51 mdw
- * Karatsuba-Ofman multiplication algorithm.
- *
- */
-
/*----- Header files ------------------------------------------------------*/
#include <assert.h>
#include <stdio.h>
#include "mpx.h"
-#include "mpx-kmac.h"
+#include "karatsuba.h"
/*----- Tweakables --------------------------------------------------------*/
#ifdef TEST_RIG
-# undef KARATSUBA_CUTOFF
-# define KARATSUBA_CUTOFF 2
+# undef MPK_THRESH
+# define MPK_THRESH 4 /* Smallest possible correct value */
#endif
/*----- Main code ---------------------------------------------------------*/
* multiplication (e.g., @mpx_umul@) on large numbers, although
* more expensive on small ones.
*
- * The destination must be twice as large as the larger
- * argument. The scratch space must be twice as large as the
- * larger argument, plus the magic number @KARATSUBA_SLOP@.
+ * The destination must be three times as large as the larger
+ * argument. The scratch space must be five times as large as
+ * the larger argument.
*/
void mpx_kmul(mpw *dv, mpw *dvl,
MPX_SHRINK(av, avl);
MPX_SHRINK(bv, bvl);
- if (avl - av <= KARATSUBA_CUTOFF || bvl - bv <= KARATSUBA_CUTOFF) {
+ if (avl - av <= MPK_THRESH || bvl - bv <= MPK_THRESH) {
mpx_umul(dv, dvl, av, avl, bv, bvl);
return;
}
/* --- First things --- *
*
* Sort out where to break the factors in half. I'll choose the midpoint
- * of the largest one, since this minimizes the amount of work I have to do
+ * of the larger one, since this minimizes the amount of work I have to do
* most effectively.
*/
avm = avl;
}
- assert(((void)"Destination too small for Karatsuba multiply",
- dvl - dv >= 4 * m));
- assert(((void)"Not enough workspace for Karatsuba multiply",
- svl - sv >= 4 * m));
-
/* --- Sort out the middle term --- */
{
mpw *bsv = sv + m + 1, *ssv = bsv + m + 1;
mpw *rdv = dv + m, *rdvl = rdv + 2 * (m + 2);
+ assert(rdvl <= dvl);
+ assert(ssv <= svl);
UADD2(sv, bsv, av, avm, avm, avl);
UADD2(bsv, ssv, bv, bvm, bvm, bvl);
- if (m > KARATSUBA_CUTOFF)
+ if (m > MPK_THRESH)
mpx_kmul(rdv, rdvl, sv, bsv, bsv, ssv, ssv, svl);
else
mpx_umul(rdv, rdvl, sv, bsv, bsv, ssv);
if (avl == avm || bvl == bvm)
MPX_ZERO(rdv + m + 1, dvl);
else {
- if (m > KARATSUBA_CUTOFF)
+ if (m > MPK_THRESH)
mpx_kmul(sv, ssv, avm, avl, bvm, bvl, ssv, svl);
else
mpx_umul(sv, ssv, avm, avl, bvm, bvl);
USUB(tdv, sv, svn);
}
- if (m > KARATSUBA_CUTOFF)
+ if (m > MPK_THRESH)
mpx_kmul(sv, ssv, av, avm, bv, bvm, ssv, svl);
else
mpx_umul(sv, ssv, av, avm, bv, bvm);
#include <mLib/alloc.h>
#include <mLib/testrig.h>
-#include "mpscan.h"
-
#define ALLOC(v, vl, sz) do { \
size_t _sz = (sz); \
mpw *_vv = xmalloc(MPWS(_sz)); \
LOAD(b, bl, &v[1]);
LOAD(c, cl, &v[2]);
m = MAX(al - a, bl - b) + 1;
- ALLOC(d, dl, 2 * m);
- ALLOC(s, sl, 2 * m + 32);
+ ALLOC(d, dl, 3 * m);
+ ALLOC(s, sl, 5 * m);
mpx_kmul(d, dl, a, al, b, bl, s, sl);
- if (MPX_UCMP(d, dl, !=, c, cl)) {
+ if (!mpx_ueq(d, dl, c, cl)) {
fprintf(stderr, "\n*** umul failed\n");
dumpmp(" a", a, al);
dumpmp(" b", b, bl);