3 * $Id: mpx.c,v 1.19 2004/04/03 03:29:40 mdw Exp $
5 * Low-level multiprecision arithmetic
7 * (c) 1999 Straylight/Edgeware
10 /*----- Licensing notice --------------------------------------------------*
12 * This file is part of Catacomb.
14 * Catacomb is free software; you can redistribute it and/or modify
15 * it under the terms of the GNU Library General Public License as
16 * published by the Free Software Foundation; either version 2 of the
17 * License, or (at your option) any later version.
19 * Catacomb is distributed in the hope that it will be useful,
20 * but WITHOUT ANY WARRANTY; without even the implied warranty of
21 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
22 * GNU Library General Public License for more details.
24 * You should have received a copy of the GNU Library General Public
25 * License along with Catacomb; if not, write to the Free
26 * Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
30 /*----- Revision history --------------------------------------------------*
33 * Revision 1.19 2004/04/03 03:29:40 mdw
34 * Fix overrun in @mpx_lsr@.
36 * Revision 1.18 2004/04/01 12:50:09 mdw
37 * Add cyclic group abstraction, with test code. Separate off exponentation
38 * functions for better static linking. Fix a buttload of bugs on the way.
39 * Generally ensure that negative exponents do inversion correctly. Add
40 * table of standard prime-field subgroups. (Binary field subgroups are
41 * currently unimplemented but easy to add if anyone ever finds a good one.)
43 * Revision 1.17 2004/03/27 00:04:46 mdw
44 * Implement efficient reduction for pleasant-looking primes.
46 * Revision 1.16 2003/05/16 09:09:24 mdw
47 * Fix @mp_lsl2c@. Turns out to be surprisingly tricky.
49 * Revision 1.15 2002/10/20 01:12:31 mdw
50 * Two's complement I/O fixes.
52 * Revision 1.14 2002/10/19 18:55:08 mdw
53 * Fix overflows in shift primitives.
55 * Revision 1.13 2002/10/19 17:56:50 mdw
56 * Fix bit operations. Test them (a bit) better.
58 * Revision 1.12 2002/10/06 22:52:50 mdw
59 * Pile of changes for supporting two's complement properly.
61 * Revision 1.11 2001/04/03 19:36:05 mdw
62 * Add some simple bitwise operations so that Perl can use them.
64 * Revision 1.10 2000/10/08 12:06:12 mdw
65 * Provide @mpx_ueq@ for rapidly testing equality of two integers.
67 * Revision 1.9 2000/06/26 07:52:50 mdw
68 * Portability fix for the bug fix.
70 * Revision 1.8 2000/06/25 12:59:02 mdw
71 * (mpx_udiv): Fix bug in quotient digit estimation.
73 * Revision 1.7 1999/12/22 15:49:07 mdw
74 * New function for division by a small integer.
76 * Revision 1.6 1999/11/20 22:43:44 mdw
77 * Integrate testing for MPX routines.
79 * Revision 1.5 1999/11/20 22:23:27 mdw
80 * Add function versions of some low-level macros with wider use.
82 * Revision 1.4 1999/11/17 18:04:09 mdw
83 * Add two's-complement functionality. Improve mpx_udiv a little by
84 * performing the multiplication of the divisor by q with the subtraction
87 * Revision 1.3 1999/11/13 01:57:31 mdw
88 * Remove stray debugging code.
90 * Revision 1.2 1999/11/13 01:50:59 mdw
91 * Multiprecision routines finished and tested.
93 * Revision 1.1 1999/09/03 08:41:12 mdw
98 /*----- Header files ------------------------------------------------------*/
105 #include <mLib/bits.h>
111 /*----- Loading and storing -----------------------------------------------*/
113 /* --- @mpx_storel@ --- *
115 * Arguments: @const mpw *v, *vl@ = base and limit of source vector
116 * @void *pp@ = pointer to octet array
117 * @size_t sz@ = size of octet array
121 * Use: Stores an MP in an octet array, least significant octet
122 * first. High-end octets are silently discarded if there
123 * isn't enough space for them.
126 void mpx_storel(const mpw *v, const mpw *vl, void *pp, size_t sz)
129 octet *p = pp, *q = p + sz;
139 *p++ = U8(w | n << bits);
141 bits += MPW_BITS - 8;
151 /* --- @mpx_loadl@ --- *
153 * Arguments: @mpw *v, *vl@ = base and limit of destination vector
154 * @const void *pp@ = pointer to octet array
155 * @size_t sz@ = size of octet array
159 * Use: Loads an MP in an octet array, least significant octet
160 * first. High-end octets are ignored if there isn't enough
164 void mpx_loadl(mpw *v, mpw *vl, const void *pp, size_t sz)
168 const octet *p = pp, *q = p + sz;
177 if (bits >= MPW_BITS) {
179 w = n >> (MPW_BITS - bits + 8);
189 /* --- @mpx_storeb@ --- *
191 * Arguments: @const mpw *v, *vl@ = base and limit of source vector
192 * @void *pp@ = pointer to octet array
193 * @size_t sz@ = size of octet array
197 * Use: Stores an MP in an octet array, most significant octet
198 * first. High-end octets are silently discarded if there
199 * isn't enough space for them.
202 void mpx_storeb(const mpw *v, const mpw *vl, void *pp, size_t sz)
205 octet *p = pp, *q = p + sz;
215 *--q = U8(w | n << bits);
217 bits += MPW_BITS - 8;
227 /* --- @mpx_loadb@ --- *
229 * Arguments: @mpw *v, *vl@ = base and limit of destination vector
230 * @const void *pp@ = pointer to octet array
231 * @size_t sz@ = size of octet array
235 * Use: Loads an MP in an octet array, most significant octet
236 * first. High-end octets are ignored if there isn't enough
240 void mpx_loadb(mpw *v, mpw *vl, const void *pp, size_t sz)
244 const octet *p = pp, *q = p + sz;
253 if (bits >= MPW_BITS) {
255 w = n >> (MPW_BITS - bits + 8);
265 /* --- @mpx_storel2cn@ --- *
267 * Arguments: @const mpw *v, *vl@ = base and limit of source vector
268 * @void *pp@ = pointer to octet array
269 * @size_t sz@ = size of octet array
273 * Use: Stores a negative MP in an octet array, least significant
274 * octet first, as two's complement. High-end octets are
275 * silently discarded if there isn't enough space for them.
276 * This obviously makes the output bad.
279 void mpx_storel2cn(const mpw *v, const mpw *vl, void *pp, size_t sz)
284 octet *p = pp, *q = p + sz;
296 bits += MPW_BITS - 8;
314 /* --- @mpx_loadl2cn@ --- *
316 * Arguments: @mpw *v, *vl@ = base and limit of destination vector
317 * @const void *pp@ = pointer to octet array
318 * @size_t sz@ = size of octet array
322 * Use: Loads a negative MP in an octet array, least significant
323 * octet first, as two's complement. High-end octets are
324 * ignored if there isn't enough space for them. This probably
325 * means you made the wrong choice coming here.
328 void mpx_loadl2cn(mpw *v, mpw *vl, const void *pp, size_t sz)
333 const octet *p = pp, *q = p + sz;
343 if (bits >= MPW_BITS) {
345 w = n >> (MPW_BITS - bits + 8);
355 /* --- @mpx_storeb2cn@ --- *
357 * Arguments: @const mpw *v, *vl@ = base and limit of source vector
358 * @void *pp@ = pointer to octet array
359 * @size_t sz@ = size of octet array
363 * Use: Stores a negative MP in an octet array, most significant
364 * octet first, as two's complement. High-end octets are
365 * silently discarded if there isn't enough space for them,
366 * which probably isn't what you meant.
369 void mpx_storeb2cn(const mpw *v, const mpw *vl, void *pp, size_t sz)
374 octet *p = pp, *q = p + sz;
386 bits += MPW_BITS - 8;
398 c = c && !(b & 0xff);
404 /* --- @mpx_loadb2cn@ --- *
406 * Arguments: @mpw *v, *vl@ = base and limit of destination vector
407 * @const void *pp@ = pointer to octet array
408 * @size_t sz@ = size of octet array
412 * Use: Loads a negative MP in an octet array, most significant octet
413 * first as two's complement. High-end octets are ignored if
414 * there isn't enough space for them. This probably means you
415 * chose this function wrongly.
418 void mpx_loadb2cn(mpw *v, mpw *vl, const void *pp, size_t sz)
423 const octet *p = pp, *q = p + sz;
433 if (bits >= MPW_BITS) {
435 w = n >> (MPW_BITS - bits + 8);
445 /*----- Logical shifting --------------------------------------------------*/
447 /* --- @mpx_lsl@ --- *
449 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
450 * @const mpw *av, *avl@ = source vector base and limit
451 * @size_t n@ = number of bit positions to shift by
455 * Use: Performs a logical shift left operation on an integer.
458 void mpx_lsl(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, size_t n)
463 /* --- Trivial special case --- */
466 MPX_COPY(dv, dvl, av, avl);
468 /* --- Single bit shifting --- */
477 *dv++ = MPW((t << 1) | w);
478 w = t >> (MPW_BITS - 1);
487 /* --- Break out word and bit shifts for more sophisticated work --- */
492 /* --- Handle a shift by a multiple of the word size --- */
498 MPX_COPY(dv + nw, dvl, av, avl);
499 memset(dv, 0, MPWS(nw));
503 /* --- And finally the difficult case --- *
505 * This is a little convoluted, because I have to start from the end and
506 * work backwards to avoid overwriting the source, if they're both the same
512 size_t nr = MPW_BITS - nb;
513 size_t dvn = dvl - dv;
514 size_t avn = avl - av;
521 if (dvn > avn + nw) {
522 size_t off = avn + nw + 1;
523 MPX_ZERO(dv + off, dvl);
533 *--dvl = (t >> nr) | w;
544 /* --- @mpx_lslc@ --- *
546 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
547 * @const mpw *av, *avl@ = source vector base and limit
548 * @size_t n@ = number of bit positions to shift by
552 * Use: Performs a logical shift left operation on an integer, only
553 * it fills in the bits with ones instead of zeroes.
556 void mpx_lslc(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, size_t n)
561 /* --- Trivial special case --- */
564 MPX_COPY(dv, dvl, av, avl);
566 /* --- Single bit shifting --- */
575 *dv++ = MPW((t << 1) | w);
576 w = t >> (MPW_BITS - 1);
585 /* --- Break out word and bit shifts for more sophisticated work --- */
590 /* --- Handle a shift by a multiple of the word size --- */
596 MPX_COPY(dv + nw, dvl, av, avl);
597 MPX_ONE(dv, dv + nw);
601 /* --- And finally the difficult case --- *
603 * This is a little convoluted, because I have to start from the end and
604 * work backwards to avoid overwriting the source, if they're both the same
610 size_t nr = MPW_BITS - nb;
611 size_t dvn = dvl - dv;
612 size_t avn = avl - av;
619 if (dvn > avn + nw) {
620 size_t off = avn + nw + 1;
621 MPX_ZERO(dv + off, dvl);
631 *--dvl = (t >> nr) | w;
635 *--dvl = (MPW_MAX >> nr) | w;
642 /* --- @mpx_lsr@ --- *
644 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
645 * @const mpw *av, *avl@ = source vector base and limit
646 * @size_t n@ = number of bit positions to shift by
650 * Use: Performs a logical shift right operation on an integer.
653 void mpx_lsr(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, size_t n)
658 /* --- Trivial special case --- */
661 MPX_COPY(dv, dvl, av, avl);
663 /* --- Single bit shifting --- */
666 mpw w = av < avl ? *av++ >> 1 : 0;
672 *dv++ = MPW((t << (MPW_BITS - 1)) | w);
682 /* --- Break out word and bit shifts for more sophisticated work --- */
687 /* --- Handle a shift by a multiple of the word size --- */
693 MPX_COPY(dv, dvl, av + nw, avl);
696 /* --- And finally the difficult case --- */
700 size_t nr = MPW_BITS - nb;
703 w = av < avl ? *av++ : 0;
709 *dv++ = MPW((w >> nb) | (t << nr));
713 *dv++ = MPW(w >> nb);
721 /*----- Bitwise operations ------------------------------------------------*/
723 /* --- @mpx_bitop@ --- *
725 * Arguments: @mpw *dv, *dvl@ = destination vector
726 * @const mpw *av, *avl@ = first source vector
727 * @const mpw *bv, *bvl@ = second source vector
731 * Use; Provides the dyadic boolean functions.
734 #define MPX_BITBINOP(string) \
736 void mpx_bit##string(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, \
737 const mpw *bv, const mpw *bvl) \
739 MPX_SHRINK(av, avl); \
740 MPX_SHRINK(bv, bvl); \
744 a = (av < avl) ? *av++ : 0; \
745 b = (bv < bvl) ? *bv++ : 0; \
746 *dv++ = B##string(a, b); \
750 MPX_DOBIN(MPX_BITBINOP)
752 void mpx_not(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl)
758 a = (av < avl) ? *av++ : 0;
763 /*----- Unsigned arithmetic -----------------------------------------------*/
765 /* --- @mpx_2c@ --- *
767 * Arguments: @mpw *dv, *dvl@ = destination vector
768 * @const mpw *v, *vl@ = source vector
772 * Use: Calculates the two's complement of @v@.
775 void mpx_2c(mpw *dv, mpw *dvl, const mpw *v, const mpw *vl)
778 while (dv < dvl && v < vl)
779 *dv++ = c = MPW(~*v++);
786 MPX_UADDN(dv, dvl, 1);
789 /* --- @mpx_ueq@ --- *
791 * Arguments: @const mpw *av, *avl@ = first argument vector base and limit
792 * @const mpw *bv, *bvl@ = second argument vector base and limit
794 * Returns: Nonzero if the two vectors are equal.
796 * Use: Performs an unsigned integer test for equality.
799 int mpx_ueq(const mpw *av, const mpw *avl, const mpw *bv, const mpw *bvl)
803 if (avl - av != bvl - bv)
812 /* --- @mpx_ucmp@ --- *
814 * Arguments: @const mpw *av, *avl@ = first argument vector base and limit
815 * @const mpw *bv, *bvl@ = second argument vector base and limit
817 * Returns: Less than, equal to, or greater than zero depending on
818 * whether @a@ is less than, equal to or greater than @b@,
821 * Use: Performs an unsigned integer comparison.
824 int mpx_ucmp(const mpw *av, const mpw *avl, const mpw *bv, const mpw *bvl)
829 if (avl - av > bvl - bv)
831 else if (avl - av < bvl - bv)
833 else while (avl > av) {
834 mpw a = *--avl, b = *--bvl;
843 /* --- @mpx_uadd@ --- *
845 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
846 * @const mpw *av, *avl@ = first addend vector base and limit
847 * @const mpw *bv, *bvl@ = second addend vector base and limit
851 * Use: Performs unsigned integer addition. If the result overflows
852 * the destination vector, high-order bits are discarded. This
853 * means that two's complement addition happens more or less for
854 * free, although that's more a side-effect than anything else.
855 * The result vector may be equal to either or both source
856 * vectors, but may not otherwise overlap them.
859 void mpx_uadd(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl,
860 const mpw *bv, const mpw *bvl)
864 while (av < avl || bv < bvl) {
869 a = (av < avl) ? *av++ : 0;
870 b = (bv < bvl) ? *bv++ : 0;
871 x = (mpd)a + (mpd)b + c;
881 /* --- @mpx_uaddn@ --- *
883 * Arguments: @mpw *dv, *dvl@ = source and destination base and limit
884 * @mpw n@ = other addend
888 * Use: Adds a small integer to a multiprecision number.
891 void mpx_uaddn(mpw *dv, mpw *dvl, mpw n) { MPX_UADDN(dv, dvl, n); }
893 /* --- @mpx_uaddnlsl@ --- *
895 * Arguments: @mpw *dv, *dvl@ = destination and first argument vector
896 * @mpw a@ = second argument
897 * @unsigned o@ = offset in bits
901 * Use: Computes %$d + 2^o a$%. If the result overflows then
902 * high-order bits are discarded, as usual. We must have
903 * @0 < o < MPW_BITS@.
906 void mpx_uaddnlsl(mpw *dv, mpw *dvl, mpw a, unsigned o)
910 while (x && dv < dvl) {
917 /* --- @mpx_usub@ --- *
919 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
920 * @const mpw *av, *avl@ = first argument vector base and limit
921 * @const mpw *bv, *bvl@ = second argument vector base and limit
925 * Use: Performs unsigned integer subtraction. If the result
926 * overflows the destination vector, high-order bits are
927 * discarded. This means that two's complement subtraction
928 * happens more or less for free, althuogh that's more a side-
929 * effect than anything else. The result vector may be equal to
930 * either or both source vectors, but may not otherwise overlap
934 void mpx_usub(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl,
935 const mpw *bv, const mpw *bvl)
939 while (av < avl || bv < bvl) {
944 a = (av < avl) ? *av++ : 0;
945 b = (bv < bvl) ? *bv++ : 0;
946 x = (mpd)a - (mpd)b - c;
959 /* --- @mpx_usubn@ --- *
961 * Arguments: @mpw *dv, *dvl@ = source and destination base and limit
966 * Use: Subtracts a small integer from a multiprecision number.
969 void mpx_usubn(mpw *dv, mpw *dvl, mpw n) { MPX_USUBN(dv, dvl, n); }
971 /* --- @mpx_uaddnlsl@ --- *
973 * Arguments: @mpw *dv, *dvl@ = destination and first argument vector
974 * @mpw a@ = second argument
975 * @unsigned o@ = offset in bits
979 * Use: Computes %$d + 2^o a$%. If the result overflows then
980 * high-order bits are discarded, as usual. We must have
981 * @0 < o < MPW_BITS@.
984 void mpx_usubnlsl(mpw *dv, mpw *dvl, mpw a, unsigned o)
986 mpw b = a >> (MPW_BITS - o);
990 mpd x = (mpd)*dv - (mpd)a;
994 MPX_USUBN(dv, dvl, b);
998 /* --- @mpx_umul@ --- *
1000 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
1001 * @const mpw *av, *avl@ = multiplicand vector base and limit
1002 * @const mpw *bv, *bvl@ = multiplier vector base and limit
1006 * Use: Performs unsigned integer multiplication. If the result
1007 * overflows the desination vector, high-order bits are
1008 * discarded. The result vector may not overlap the argument
1009 * vectors in any way.
1012 void mpx_umul(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl,
1013 const mpw *bv, const mpw *bvl)
1015 /* --- This is probably worthwhile on a multiply --- */
1017 MPX_SHRINK(av, avl);
1018 MPX_SHRINK(bv, bvl);
1020 /* --- Deal with a multiply by zero --- */
1027 /* --- Do the initial multiply and initialize the accumulator --- */
1029 MPX_UMULN(dv, dvl, av, avl, *bv++);
1031 /* --- Do the remaining multiply/accumulates --- */
1033 while (dv < dvl && bv < bvl) {
1036 const mpw *avv = av;
1043 x = (mpd)*dvv + (mpd)m * (mpd)*avv++ + c;
1047 MPX_UADDN(dvv, dvl, c);
1052 /* --- @mpx_umuln@ --- *
1054 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
1055 * @const mpw *av, *avl@ = multiplicand vector base and limit
1056 * @mpw m@ = multiplier
1060 * Use: Multiplies a multiprecision integer by a single-word value.
1061 * The destination and source may be equal. The destination
1062 * is completely cleared after use.
1065 void mpx_umuln(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, mpw m)
1067 MPX_UMULN(dv, dvl, av, avl, m);
1070 /* --- @mpx_umlan@ --- *
1072 * Arguments: @mpw *dv, *dvl@ = destination/accumulator base and limit
1073 * @const mpw *av, *avl@ = multiplicand vector base and limit
1074 * @mpw m@ = multiplier
1078 * Use: Multiplies a multiprecision integer by a single-word value
1079 * and adds the result to an accumulator.
1082 void mpx_umlan(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, mpw m)
1084 MPX_UMLAN(dv, dvl, av, avl, m);
1087 /* --- @mpx_usqr@ --- *
1089 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
1090 * @const mpw *av, *av@ = source vector base and limit
1094 * Use: Performs unsigned integer squaring. The result vector must
1095 * not overlap the source vector in any way.
1098 void mpx_usqr(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl)
1102 /* --- Main loop --- */
1105 const mpw *avv = av;
1110 /* --- Stop if I've run out of destination --- */
1115 /* --- Work out the square at this point in the proceedings --- */
1118 mpd x = (mpd)a * (mpd)a + *dvv;
1120 c = MPW(x >> MPW_BITS);
1123 /* --- Now fix up the rest of the vector upwards --- */
1126 while (dvv < dvl && avv < avl) {
1127 mpd x = (mpd)a * (mpd)*avv++;
1128 mpd y = ((x << 1) & MPW_MAX) + c + *dvv;
1129 c = (x >> (MPW_BITS - 1)) + (y >> MPW_BITS);
1132 while (dvv < dvl && c) {
1138 /* --- Get ready for the next round --- */
1145 /* --- @mpx_udiv@ --- *
1147 * Arguments: @mpw *qv, *qvl@ = quotient vector base and limit
1148 * @mpw *rv, *rvl@ = dividend/remainder vector base and limit
1149 * @const mpw *dv, *dvl@ = divisor vector base and limit
1150 * @mpw *sv, *svl@ = scratch workspace
1154 * Use: Performs unsigned integer division. If the result overflows
1155 * the quotient vector, high-order bits are discarded. (Clearly
1156 * the remainder vector can't overflow.) The various vectors
1157 * may not overlap in any way. Yes, I know it's a bit odd
1158 * requiring the dividend to be in the result position but it
1159 * does make some sense really. The remainder must have
1160 * headroom for at least two extra words. The scratch space
1161 * must be at least one word larger than the divisor.
1164 void mpx_udiv(mpw *qv, mpw *qvl, mpw *rv, mpw *rvl,
1165 const mpw *dv, const mpw *dvl,
1172 /* --- Initialize the quotient --- */
1176 /* --- Perform some sanity checks --- */
1178 MPX_SHRINK(dv, dvl);
1179 assert(((void)"division by zero in mpx_udiv", dv < dvl));
1181 /* --- Normalize the divisor --- *
1183 * The algorithm requires that the divisor be at least two digits long.
1184 * This is easy to fix.
1191 for (b = MPW_BITS / 2; b; b >>= 1) {
1192 if (d <= (MPW_MAX >> b)) {
1201 /* --- Normalize the dividend/remainder to match --- */
1204 mpx_lsl(rv, rvl, rv, rvl, norm);
1205 mpx_lsl(sv, svl, dv, dvl, norm);
1208 MPX_SHRINK(dv, dvl);
1211 MPX_SHRINK(rv, rvl);
1215 /* --- Work out the relative scales --- */
1218 size_t rvn = rvl - rv;
1219 size_t dvn = dvl - dv;
1221 /* --- If the divisor is clearly larger, notice this --- */
1224 mpx_lsr(rv, rvl, rv, rvl, norm);
1231 /* --- Calculate the most significant quotient digit --- *
1233 * Because the divisor has its top bit set, this can only happen once. The
1234 * pointer arithmetic is a little contorted, to make sure that the
1235 * behaviour is defined.
1238 if (MPX_UCMP(rv + scale, rvl, >=, dv, dvl)) {
1239 mpx_usub(rv + scale, rvl, rv + scale, rvl, dv, dvl);
1240 if (qvl - qv > scale)
1244 /* --- Now for the main loop --- */
1253 /* --- Get an estimate for the next quotient digit --- */
1260 rh = ((mpd)r << MPW_BITS) | rr;
1266 /* --- Refine the estimate --- */
1269 mpd yh = (mpd)d * q;
1270 mpd yy = (mpd)dd * q;
1274 yh += yy >> MPW_BITS;
1277 while (yh > rh || (yh == rh && yl > rrr)) {
1286 /* --- Remove a chunk from the dividend --- */
1293 /* --- Calculate the size of the chunk --- *
1295 * This does the whole job of calculating @r >> scale - qd@.
1298 for (svv = rv + scale, dvv = dv;
1299 dvv < dvl && svv < rvl;
1301 mpd x = (mpd)*dvv * (mpd)q + mc;
1303 x = (mpd)*svv - MPW(x) - sc;
1312 mpd x = (mpd)*svv - mc - sc;
1322 /* --- Fix if the quotient was too large --- *
1324 * This doesn't seem to happen very often.
1327 if (rvl[-1] > MPW_MAX / 2) {
1328 mpx_uadd(rv + scale, rvl, rv + scale, rvl, dv, dvl);
1333 /* --- Done for another iteration --- */
1335 if (qvl - qv > scale)
1342 /* --- Now fiddle with unnormalizing and things --- */
1344 mpx_lsr(rv, rvl, rv, rvl, norm);
1347 /* --- @mpx_udivn@ --- *
1349 * Arguments: @mpw *qv, *qvl@ = storage for the quotient (may overlap
1351 * @const mpw *rv, *rvl@ = dividend
1352 * @mpw d@ = single-precision divisor
1354 * Returns: Remainder after divison.
1356 * Use: Performs a single-precision division operation.
1359 mpw mpx_udivn(mpw *qv, mpw *qvl, const mpw *rv, const mpw *rvl, mpw d)
1362 size_t ql = qvl - qv;
1368 r = (r << MPW_BITS) | rv[i];
1376 /*----- Test rig ----------------------------------------------------------*/
1380 #include <mLib/alloc.h>
1381 #include <mLib/dstr.h>
1382 #include <mLib/quis.h>
1383 #include <mLib/testrig.h>
1387 #define ALLOC(v, vl, sz) do { \
1388 size_t _sz = (sz); \
1389 mpw *_vv = xmalloc(MPWS(_sz)); \
1390 mpw *_vvl = _vv + _sz; \
1395 #define LOAD(v, vl, d) do { \
1396 const dstr *_d = (d); \
1398 ALLOC(_v, _vl, MPW_RQ(_d->len)); \
1399 mpx_loadb(_v, _vl, _d->buf, _d->len); \
1404 #define MAX(x, y) ((x) > (y) ? (x) : (y))
1406 static void dumpbits(const char *msg, const void *pp, size_t sz)
1408 const octet *p = pp;
1411 fprintf(stderr, " %02x", *p++);
1412 fputc('\n', stderr);
1415 static void dumpmp(const char *msg, const mpw *v, const mpw *vl)
1420 fprintf(stderr, " %08lx", (unsigned long)*--vl);
1421 fputc('\n', stderr);
1424 static int chkscan(const mpw *v, const mpw *vl,
1425 const void *pp, size_t sz, int step)
1428 const octet *p = pp;
1432 mpscan_initx(&mps, v, vl);
1437 for (i = 0; i < 8 && MPSCAN_STEP(&mps); i++) {
1438 if (MPSCAN_BIT(&mps) != (x & 1)) {
1440 "\n*** error, step %i, bit %u, expected %u, found %u\n",
1441 step, bit, x & 1, MPSCAN_BIT(&mps));
1453 static int loadstore(dstr *v)
1456 size_t sz = MPW_RQ(v->len) * 2, diff;
1460 dstr_ensure(&d, v->len);
1461 m = xmalloc(MPWS(sz));
1463 for (diff = 0; diff < sz; diff += 5) {
1468 mpx_loadl(m, ml, v->buf, v->len);
1469 if (!chkscan(m, ml, v->buf, v->len, +1))
1471 MPX_OCTETS(oct, m, ml);
1472 mpx_storel(m, ml, d.buf, d.sz);
1473 if (memcmp(d.buf, v->buf, oct) != 0) {
1474 dumpbits("\n*** storel failed", d.buf, d.sz);
1478 mpx_loadb(m, ml, v->buf, v->len);
1479 if (!chkscan(m, ml, v->buf + v->len - 1, v->len, -1))
1481 MPX_OCTETS(oct, m, ml);
1482 mpx_storeb(m, ml, d.buf, d.sz);
1483 if (memcmp(d.buf + d.sz - oct, v->buf + v->len - oct, oct) != 0) {
1484 dumpbits("\n*** storeb failed", d.buf, d.sz);
1490 dumpbits("input data", v->buf, v->len);
1497 static int twocl(dstr *v)
1504 sz = v[0].len; if (v[1].len > sz) sz = v[1].len;
1505 dstr_ensure(&d, sz);
1508 m = xmalloc(MPWS(sz));
1511 mpx_loadl(m, ml, v[0].buf, v[0].len);
1512 mpx_storel2cn(m, ml, d.buf, v[1].len);
1513 if (memcmp(d.buf, v[1].buf, v[1].len)) {
1514 dumpbits("\n*** storel2cn failed", d.buf, v[1].len);
1518 mpx_loadl2cn(m, ml, v[1].buf, v[1].len);
1519 mpx_storel(m, ml, d.buf, v[0].len);
1520 if (memcmp(d.buf, v[0].buf, v[0].len)) {
1521 dumpbits("\n*** loadl2cn failed", d.buf, v[0].len);
1526 dumpbits("pos", v[0].buf, v[0].len);
1527 dumpbits("neg", v[1].buf, v[1].len);
1536 static int twocb(dstr *v)
1543 sz = v[0].len; if (v[1].len > sz) sz = v[1].len;
1544 dstr_ensure(&d, sz);
1547 m = xmalloc(MPWS(sz));
1550 mpx_loadb(m, ml, v[0].buf, v[0].len);
1551 mpx_storeb2cn(m, ml, d.buf, v[1].len);
1552 if (memcmp(d.buf, v[1].buf, v[1].len)) {
1553 dumpbits("\n*** storeb2cn failed", d.buf, v[1].len);
1557 mpx_loadb2cn(m, ml, v[1].buf, v[1].len);
1558 mpx_storeb(m, ml, d.buf, v[0].len);
1559 if (memcmp(d.buf, v[0].buf, v[0].len)) {
1560 dumpbits("\n*** loadb2cn failed", d.buf, v[0].len);
1565 dumpbits("pos", v[0].buf, v[0].len);
1566 dumpbits("neg", v[1].buf, v[1].len);
1575 static int lsl(dstr *v)
1578 int n = *(int *)v[1].buf;
1585 ALLOC(d, dl, al - a + (n + MPW_BITS - 1) / MPW_BITS);
1587 mpx_lsl(d, dl, a, al, n);
1588 if (!mpx_ueq(d, dl, c, cl)) {
1589 fprintf(stderr, "\n*** lsl(%i) failed\n", n);
1590 dumpmp(" a", a, al);
1591 dumpmp("expected", c, cl);
1592 dumpmp(" result", d, dl);
1596 free(a); free(c); free(d);
1600 static int lslc(dstr *v)
1603 int n = *(int *)v[1].buf;
1610 ALLOC(d, dl, al - a + (n + MPW_BITS - 1) / MPW_BITS);
1612 mpx_lslc(d, dl, a, al, n);
1613 if (!mpx_ueq(d, dl, c, cl)) {
1614 fprintf(stderr, "\n*** lslc(%i) failed\n", n);
1615 dumpmp(" a", a, al);
1616 dumpmp("expected", c, cl);
1617 dumpmp(" result", d, dl);
1621 free(a); free(c); free(d);
1625 static int lsr(dstr *v)
1628 int n = *(int *)v[1].buf;
1635 ALLOC(d, dl, al - a + (n + MPW_BITS - 1) / MPW_BITS + 1);
1637 mpx_lsr(d, dl, a, al, n);
1638 if (!mpx_ueq(d, dl, c, cl)) {
1639 fprintf(stderr, "\n*** lsr(%i) failed\n", n);
1640 dumpmp(" a", a, al);
1641 dumpmp("expected", c, cl);
1642 dumpmp(" result", d, dl);
1646 free(a); free(c); free(d);
1650 static int uadd(dstr *v)
1661 ALLOC(d, dl, MAX(al - a, bl - b) + 1);
1663 mpx_uadd(d, dl, a, al, b, bl);
1664 if (!mpx_ueq(d, dl, c, cl)) {
1665 fprintf(stderr, "\n*** uadd failed\n");
1666 dumpmp(" a", a, al);
1667 dumpmp(" b", b, bl);
1668 dumpmp("expected", c, cl);
1669 dumpmp(" result", d, dl);
1673 free(a); free(b); free(c); free(d);
1677 static int usub(dstr *v)
1688 ALLOC(d, dl, al - a);
1690 mpx_usub(d, dl, a, al, b, bl);
1691 if (!mpx_ueq(d, dl, c, cl)) {
1692 fprintf(stderr, "\n*** usub failed\n");
1693 dumpmp(" a", a, al);
1694 dumpmp(" b", b, bl);
1695 dumpmp("expected", c, cl);
1696 dumpmp(" result", d, dl);
1700 free(a); free(b); free(c); free(d);
1704 static int umul(dstr *v)
1715 ALLOC(d, dl, (al - a) + (bl - b));
1717 mpx_umul(d, dl, a, al, b, bl);
1718 if (!mpx_ueq(d, dl, c, cl)) {
1719 fprintf(stderr, "\n*** umul failed\n");
1720 dumpmp(" a", a, al);
1721 dumpmp(" b", b, bl);
1722 dumpmp("expected", c, cl);
1723 dumpmp(" result", d, dl);
1727 free(a); free(b); free(c); free(d);
1731 static int usqr(dstr *v)
1740 ALLOC(d, dl, 2 * (al - a));
1742 mpx_usqr(d, dl, a, al);
1743 if (!mpx_ueq(d, dl, c, cl)) {
1744 fprintf(stderr, "\n*** usqr failed\n");
1745 dumpmp(" a", a, al);
1746 dumpmp("expected", c, cl);
1747 dumpmp(" result", d, dl);
1751 free(a); free(c); free(d);
1755 static int udiv(dstr *v)
1765 ALLOC(a, al, MPW_RQ(v[0].len) + 2); mpx_loadb(a, al, v[0].buf, v[0].len);
1769 ALLOC(qq, qql, al - a);
1770 ALLOC(s, sl, (bl - b) + 1);
1772 mpx_udiv(qq, qql, a, al, b, bl, s, sl);
1773 if (!mpx_ueq(qq, qql, q, ql) ||
1774 !mpx_ueq(a, al, r, rl)) {
1775 fprintf(stderr, "\n*** udiv failed\n");
1776 dumpmp(" divisor", b, bl);
1777 dumpmp("expect r", r, rl);
1778 dumpmp("result r", a, al);
1779 dumpmp("expect q", q, ql);
1780 dumpmp("result q", qq, qql);
1784 free(a); free(b); free(r); free(q); free(s); free(qq);
1788 static test_chunk defs[] = {
1789 { "load-store", loadstore, { &type_hex, 0 } },
1790 { "2cl", twocl, { &type_hex, &type_hex, } },
1791 { "2cb", twocb, { &type_hex, &type_hex, } },
1792 { "lsl", lsl, { &type_hex, &type_int, &type_hex, 0 } },
1793 { "lslc", lslc, { &type_hex, &type_int, &type_hex, 0 } },
1794 { "lsr", lsr, { &type_hex, &type_int, &type_hex, 0 } },
1795 { "uadd", uadd, { &type_hex, &type_hex, &type_hex, 0 } },
1796 { "usub", usub, { &type_hex, &type_hex, &type_hex, 0 } },
1797 { "umul", umul, { &type_hex, &type_hex, &type_hex, 0 } },
1798 { "usqr", usqr, { &type_hex, &type_hex, 0 } },
1799 { "udiv", udiv, { &type_hex, &type_hex, &type_hex, &type_hex, 0 } },
1803 int main(int argc, char *argv[])
1805 test_run(argc, argv, defs, SRCDIR"/tests/mpx");
1811 /*----- That's all, folks -------------------------------------------------*/