X-Git-Url: https://git.distorted.org.uk/u/mdw/catacomb/blobdiff_plain/7d5fa32a96bb9badc88fe1b07086c6f770c0de31..6d56743beed37f62273209341fa812bcd6aa220f:/mpx-kmul.c

diff --git a/mpx-kmul.c b/mpx-kmul.c
index dd492ad..1981a28 100644
--- a/mpx-kmul.c
+++ b/mpx-kmul.c
@@ -1,6 +1,6 @@
 /* -*-c-*-
  *
- * $Id: mpx-kmul.c,v 1.4 2000/06/17 11:42:11 mdw Exp $
+ * $Id: mpx-kmul.c,v 1.8 2002/10/09 00:36:03 mdw Exp $
  *
  * Karatsuba's multiplication algorithm
  *
@@ -30,6 +30,18 @@
 /*----- Revision history --------------------------------------------------* 
  *
  * $Log: mpx-kmul.c,v $
+ * Revision 1.8  2002/10/09 00:36:03  mdw
+ * Fix bounds on workspace for Karatsuba operations.
+ *
+ * Revision 1.7  2000/10/08 15:48:35  mdw
+ * Rename Karatsuba constants now that we have @gfx_kmul@ too.
+ *
+ * Revision 1.6  2000/10/08 12:11:01  mdw
+ * Use @mpx_ueq@ instead of @MPX_UCMP@.
+ *
+ * Revision 1.5  2000/07/29 17:04:02  mdw
+ * Remove useless header `mpscan.h'.
+ *
  * Revision 1.4  2000/06/17 11:42:11  mdw
  * Moved the Karatsuba macros into a separate file for better sharing.
  * Fixed some comments.
@@ -51,13 +63,13 @@
 #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 ---------------------------------------------------------*/
@@ -76,9 +88,9 @@
  *		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,
@@ -100,7 +112,7 @@ 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;
   }
@@ -120,7 +132,7 @@ void mpx_kmul(mpw *dv, mpw *dvl,
   /* --- 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.
    */
 
@@ -140,20 +152,17 @@ void mpx_kmul(mpw *dv, mpw *dvl,
       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);
@@ -169,7 +178,7 @@ void mpx_kmul(mpw *dv, mpw *dvl,
     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);
@@ -178,7 +187,7 @@ void mpx_kmul(mpw *dv, mpw *dvl,
       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);
@@ -195,8 +204,6 @@ void mpx_kmul(mpw *dv, mpw *dvl,
 #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));                                        \
@@ -239,11 +246,11 @@ static int umul(dstr *v)
   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);