3 * $Id: mpx.c,v 1.18 2004/04/01 12:50:09 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 1.18 2004/04/01 12:50:09 mdw
34 * Add cyclic group abstraction, with test code. Separate off exponentation
35 * functions for better static linking. Fix a buttload of bugs on the way.
36 * Generally ensure that negative exponents do inversion correctly. Add
37 * table of standard prime-field subgroups. (Binary field subgroups are
38 * currently unimplemented but easy to add if anyone ever finds a good one.)
40 * Revision 1.17 2004/03/27 00:04:46 mdw
41 * Implement efficient reduction for pleasant-looking primes.
43 * Revision 1.16 2003/05/16 09:09:24 mdw
44 * Fix @mp_lsl2c@. Turns out to be surprisingly tricky.
46 * Revision 1.15 2002/10/20 01:12:31 mdw
47 * Two's complement I/O fixes.
49 * Revision 1.14 2002/10/19 18:55:08 mdw
50 * Fix overflows in shift primitives.
52 * Revision 1.13 2002/10/19 17:56:50 mdw
53 * Fix bit operations. Test them (a bit) better.
55 * Revision 1.12 2002/10/06 22:52:50 mdw
56 * Pile of changes for supporting two's complement properly.
58 * Revision 1.11 2001/04/03 19:36:05 mdw
59 * Add some simple bitwise operations so that Perl can use them.
61 * Revision 1.10 2000/10/08 12:06:12 mdw
62 * Provide @mpx_ueq@ for rapidly testing equality of two integers.
64 * Revision 1.9 2000/06/26 07:52:50 mdw
65 * Portability fix for the bug fix.
67 * Revision 1.8 2000/06/25 12:59:02 mdw
68 * (mpx_udiv): Fix bug in quotient digit estimation.
70 * Revision 1.7 1999/12/22 15:49:07 mdw
71 * New function for division by a small integer.
73 * Revision 1.6 1999/11/20 22:43:44 mdw
74 * Integrate testing for MPX routines.
76 * Revision 1.5 1999/11/20 22:23:27 mdw
77 * Add function versions of some low-level macros with wider use.
79 * Revision 1.4 1999/11/17 18:04:09 mdw
80 * Add two's-complement functionality. Improve mpx_udiv a little by
81 * performing the multiplication of the divisor by q with the subtraction
84 * Revision 1.3 1999/11/13 01:57:31 mdw
85 * Remove stray debugging code.
87 * Revision 1.2 1999/11/13 01:50:59 mdw
88 * Multiprecision routines finished and tested.
90 * Revision 1.1 1999/09/03 08:41:12 mdw
95 /*----- Header files ------------------------------------------------------*/
102 #include <mLib/bits.h>
108 /*----- Loading and storing -----------------------------------------------*/
110 /* --- @mpx_storel@ --- *
112 * Arguments: @const mpw *v, *vl@ = base and limit of source vector
113 * @void *pp@ = pointer to octet array
114 * @size_t sz@ = size of octet array
118 * Use: Stores an MP in an octet array, least significant octet
119 * first. High-end octets are silently discarded if there
120 * isn't enough space for them.
123 void mpx_storel(const mpw *v, const mpw *vl, void *pp, size_t sz)
126 octet *p = pp, *q = p + sz;
136 *p++ = U8(w | n << bits);
138 bits += MPW_BITS - 8;
148 /* --- @mpx_loadl@ --- *
150 * Arguments: @mpw *v, *vl@ = base and limit of destination vector
151 * @const void *pp@ = pointer to octet array
152 * @size_t sz@ = size of octet array
156 * Use: Loads an MP in an octet array, least significant octet
157 * first. High-end octets are ignored if there isn't enough
161 void mpx_loadl(mpw *v, mpw *vl, const void *pp, size_t sz)
165 const octet *p = pp, *q = p + sz;
174 if (bits >= MPW_BITS) {
176 w = n >> (MPW_BITS - bits + 8);
186 /* --- @mpx_storeb@ --- *
188 * Arguments: @const mpw *v, *vl@ = base and limit of source vector
189 * @void *pp@ = pointer to octet array
190 * @size_t sz@ = size of octet array
194 * Use: Stores an MP in an octet array, most significant octet
195 * first. High-end octets are silently discarded if there
196 * isn't enough space for them.
199 void mpx_storeb(const mpw *v, const mpw *vl, void *pp, size_t sz)
202 octet *p = pp, *q = p + sz;
212 *--q = U8(w | n << bits);
214 bits += MPW_BITS - 8;
224 /* --- @mpx_loadb@ --- *
226 * Arguments: @mpw *v, *vl@ = base and limit of destination vector
227 * @const void *pp@ = pointer to octet array
228 * @size_t sz@ = size of octet array
232 * Use: Loads an MP in an octet array, most significant octet
233 * first. High-end octets are ignored if there isn't enough
237 void mpx_loadb(mpw *v, mpw *vl, const void *pp, size_t sz)
241 const octet *p = pp, *q = p + sz;
250 if (bits >= MPW_BITS) {
252 w = n >> (MPW_BITS - bits + 8);
262 /* --- @mpx_storel2cn@ --- *
264 * Arguments: @const mpw *v, *vl@ = base and limit of source vector
265 * @void *pp@ = pointer to octet array
266 * @size_t sz@ = size of octet array
270 * Use: Stores a negative MP in an octet array, least significant
271 * octet first, as two's complement. High-end octets are
272 * silently discarded if there isn't enough space for them.
273 * This obviously makes the output bad.
276 void mpx_storel2cn(const mpw *v, const mpw *vl, void *pp, size_t sz)
281 octet *p = pp, *q = p + sz;
293 bits += MPW_BITS - 8;
311 /* --- @mpx_loadl2cn@ --- *
313 * Arguments: @mpw *v, *vl@ = base and limit of destination vector
314 * @const void *pp@ = pointer to octet array
315 * @size_t sz@ = size of octet array
319 * Use: Loads a negative MP in an octet array, least significant
320 * octet first, as two's complement. High-end octets are
321 * ignored if there isn't enough space for them. This probably
322 * means you made the wrong choice coming here.
325 void mpx_loadl2cn(mpw *v, mpw *vl, const void *pp, size_t sz)
330 const octet *p = pp, *q = p + sz;
340 if (bits >= MPW_BITS) {
342 w = n >> (MPW_BITS - bits + 8);
352 /* --- @mpx_storeb2cn@ --- *
354 * Arguments: @const mpw *v, *vl@ = base and limit of source vector
355 * @void *pp@ = pointer to octet array
356 * @size_t sz@ = size of octet array
360 * Use: Stores a negative MP in an octet array, most significant
361 * octet first, as two's complement. High-end octets are
362 * silently discarded if there isn't enough space for them,
363 * which probably isn't what you meant.
366 void mpx_storeb2cn(const mpw *v, const mpw *vl, void *pp, size_t sz)
371 octet *p = pp, *q = p + sz;
383 bits += MPW_BITS - 8;
395 c = c && !(b & 0xff);
401 /* --- @mpx_loadb2cn@ --- *
403 * Arguments: @mpw *v, *vl@ = base and limit of destination vector
404 * @const void *pp@ = pointer to octet array
405 * @size_t sz@ = size of octet array
409 * Use: Loads a negative MP in an octet array, most significant octet
410 * first as two's complement. High-end octets are ignored if
411 * there isn't enough space for them. This probably means you
412 * chose this function wrongly.
415 void mpx_loadb2cn(mpw *v, mpw *vl, const void *pp, size_t sz)
420 const octet *p = pp, *q = p + sz;
430 if (bits >= MPW_BITS) {
432 w = n >> (MPW_BITS - bits + 8);
442 /*----- Logical shifting --------------------------------------------------*/
444 /* --- @mpx_lsl@ --- *
446 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
447 * @const mpw *av, *avl@ = source vector base and limit
448 * @size_t n@ = number of bit positions to shift by
452 * Use: Performs a logical shift left operation on an integer.
455 void mpx_lsl(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, size_t n)
460 /* --- Trivial special case --- */
463 MPX_COPY(dv, dvl, av, avl);
465 /* --- Single bit shifting --- */
474 *dv++ = MPW((t << 1) | w);
475 w = t >> (MPW_BITS - 1);
484 /* --- Break out word and bit shifts for more sophisticated work --- */
489 /* --- Handle a shift by a multiple of the word size --- */
495 MPX_COPY(dv + nw, dvl, av, avl);
496 memset(dv, 0, MPWS(nw));
500 /* --- And finally the difficult case --- *
502 * This is a little convoluted, because I have to start from the end and
503 * work backwards to avoid overwriting the source, if they're both the same
509 size_t nr = MPW_BITS - nb;
510 size_t dvn = dvl - dv;
511 size_t avn = avl - av;
518 if (dvn > avn + nw) {
519 size_t off = avn + nw + 1;
520 MPX_ZERO(dv + off, dvl);
530 *--dvl = (t >> nr) | w;
541 /* --- @mpx_lslc@ --- *
543 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
544 * @const mpw *av, *avl@ = source vector base and limit
545 * @size_t n@ = number of bit positions to shift by
549 * Use: Performs a logical shift left operation on an integer, only
550 * it fills in the bits with ones instead of zeroes.
553 void mpx_lslc(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, size_t n)
558 /* --- Trivial special case --- */
561 MPX_COPY(dv, dvl, av, avl);
563 /* --- Single bit shifting --- */
572 *dv++ = MPW((t << 1) | w);
573 w = t >> (MPW_BITS - 1);
582 /* --- Break out word and bit shifts for more sophisticated work --- */
587 /* --- Handle a shift by a multiple of the word size --- */
593 MPX_COPY(dv + nw, dvl, av, avl);
594 MPX_ONE(dv, dv + nw);
598 /* --- And finally the difficult case --- *
600 * This is a little convoluted, because I have to start from the end and
601 * work backwards to avoid overwriting the source, if they're both the same
607 size_t nr = MPW_BITS - nb;
608 size_t dvn = dvl - dv;
609 size_t avn = avl - av;
616 if (dvn > avn + nw) {
617 size_t off = avn + nw + 1;
618 MPX_ZERO(dv + off, dvl);
628 *--dvl = (t >> nr) | w;
632 *--dvl = (MPW_MAX >> nr) | w;
639 /* --- @mpx_lsr@ --- *
641 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
642 * @const mpw *av, *avl@ = source vector base and limit
643 * @size_t n@ = number of bit positions to shift by
647 * Use: Performs a logical shift right operation on an integer.
650 void mpx_lsr(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, size_t n)
655 /* --- Trivial special case --- */
658 MPX_COPY(dv, dvl, av, avl);
660 /* --- Single bit shifting --- */
669 *dv++ = MPW((t << (MPW_BITS - 1)) | w);
679 /* --- Break out word and bit shifts for more sophisticated work --- */
684 /* --- Handle a shift by a multiple of the word size --- */
690 MPX_COPY(dv, dvl, av + nw, avl);
693 /* --- And finally the difficult case --- */
697 size_t nr = MPW_BITS - nb;
700 w = av < avl ? *av++ : 0;
706 *dv++ = MPW((w >> nb) | (t << nr));
710 *dv++ = MPW(w >> nb);
718 /*----- Bitwise operations ------------------------------------------------*/
720 /* --- @mpx_bitop@ --- *
722 * Arguments: @mpw *dv, *dvl@ = destination vector
723 * @const mpw *av, *avl@ = first source vector
724 * @const mpw *bv, *bvl@ = second source vector
728 * Use; Provides the dyadic boolean functions.
731 #define MPX_BITBINOP(string) \
733 void mpx_bit##string(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, \
734 const mpw *bv, const mpw *bvl) \
736 MPX_SHRINK(av, avl); \
737 MPX_SHRINK(bv, bvl); \
741 a = (av < avl) ? *av++ : 0; \
742 b = (bv < bvl) ? *bv++ : 0; \
743 *dv++ = B##string(a, b); \
747 MPX_DOBIN(MPX_BITBINOP)
749 void mpx_not(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl)
755 a = (av < avl) ? *av++ : 0;
760 /*----- Unsigned arithmetic -----------------------------------------------*/
762 /* --- @mpx_2c@ --- *
764 * Arguments: @mpw *dv, *dvl@ = destination vector
765 * @const mpw *v, *vl@ = source vector
769 * Use: Calculates the two's complement of @v@.
772 void mpx_2c(mpw *dv, mpw *dvl, const mpw *v, const mpw *vl)
775 while (dv < dvl && v < vl)
776 *dv++ = c = MPW(~*v++);
783 MPX_UADDN(dv, dvl, 1);
786 /* --- @mpx_ueq@ --- *
788 * Arguments: @const mpw *av, *avl@ = first argument vector base and limit
789 * @const mpw *bv, *bvl@ = second argument vector base and limit
791 * Returns: Nonzero if the two vectors are equal.
793 * Use: Performs an unsigned integer test for equality.
796 int mpx_ueq(const mpw *av, const mpw *avl, const mpw *bv, const mpw *bvl)
800 if (avl - av != bvl - bv)
809 /* --- @mpx_ucmp@ --- *
811 * Arguments: @const mpw *av, *avl@ = first argument vector base and limit
812 * @const mpw *bv, *bvl@ = second argument vector base and limit
814 * Returns: Less than, equal to, or greater than zero depending on
815 * whether @a@ is less than, equal to or greater than @b@,
818 * Use: Performs an unsigned integer comparison.
821 int mpx_ucmp(const mpw *av, const mpw *avl, const mpw *bv, const mpw *bvl)
826 if (avl - av > bvl - bv)
828 else if (avl - av < bvl - bv)
830 else while (avl > av) {
831 mpw a = *--avl, b = *--bvl;
840 /* --- @mpx_uadd@ --- *
842 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
843 * @const mpw *av, *avl@ = first addend vector base and limit
844 * @const mpw *bv, *bvl@ = second addend vector base and limit
848 * Use: Performs unsigned integer addition. If the result overflows
849 * the destination vector, high-order bits are discarded. This
850 * means that two's complement addition happens more or less for
851 * free, although that's more a side-effect than anything else.
852 * The result vector may be equal to either or both source
853 * vectors, but may not otherwise overlap them.
856 void mpx_uadd(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl,
857 const mpw *bv, const mpw *bvl)
861 while (av < avl || bv < bvl) {
866 a = (av < avl) ? *av++ : 0;
867 b = (bv < bvl) ? *bv++ : 0;
868 x = (mpd)a + (mpd)b + c;
878 /* --- @mpx_uaddn@ --- *
880 * Arguments: @mpw *dv, *dvl@ = source and destination base and limit
881 * @mpw n@ = other addend
885 * Use: Adds a small integer to a multiprecision number.
888 void mpx_uaddn(mpw *dv, mpw *dvl, mpw n) { MPX_UADDN(dv, dvl, n); }
890 /* --- @mpx_uaddnlsl@ --- *
892 * Arguments: @mpw *dv, *dvl@ = destination and first argument vector
893 * @mpw a@ = second argument
894 * @unsigned o@ = offset in bits
898 * Use: Computes %$d + 2^o a$%. If the result overflows then
899 * high-order bits are discarded, as usual. We must have
900 * @0 < o < MPW_BITS@.
903 void mpx_uaddnlsl(mpw *dv, mpw *dvl, mpw a, unsigned o)
907 while (x && dv < dvl) {
914 /* --- @mpx_usub@ --- *
916 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
917 * @const mpw *av, *avl@ = first argument vector base and limit
918 * @const mpw *bv, *bvl@ = second argument vector base and limit
922 * Use: Performs unsigned integer subtraction. If the result
923 * overflows the destination vector, high-order bits are
924 * discarded. This means that two's complement subtraction
925 * happens more or less for free, althuogh that's more a side-
926 * effect than anything else. The result vector may be equal to
927 * either or both source vectors, but may not otherwise overlap
931 void mpx_usub(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl,
932 const mpw *bv, const mpw *bvl)
936 while (av < avl || bv < bvl) {
941 a = (av < avl) ? *av++ : 0;
942 b = (bv < bvl) ? *bv++ : 0;
943 x = (mpd)a - (mpd)b - c;
956 /* --- @mpx_usubn@ --- *
958 * Arguments: @mpw *dv, *dvl@ = source and destination base and limit
963 * Use: Subtracts a small integer from a multiprecision number.
966 void mpx_usubn(mpw *dv, mpw *dvl, mpw n) { MPX_USUBN(dv, dvl, n); }
968 /* --- @mpx_uaddnlsl@ --- *
970 * Arguments: @mpw *dv, *dvl@ = destination and first argument vector
971 * @mpw a@ = second argument
972 * @unsigned o@ = offset in bits
976 * Use: Computes %$d + 2^o a$%. If the result overflows then
977 * high-order bits are discarded, as usual. We must have
978 * @0 < o < MPW_BITS@.
981 void mpx_usubnlsl(mpw *dv, mpw *dvl, mpw a, unsigned o)
983 mpw b = a >> (MPW_BITS - o);
987 mpd x = (mpd)*dv - (mpd)a;
991 MPX_USUBN(dv, dvl, b);
995 /* --- @mpx_umul@ --- *
997 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
998 * @const mpw *av, *avl@ = multiplicand vector base and limit
999 * @const mpw *bv, *bvl@ = multiplier vector base and limit
1003 * Use: Performs unsigned integer multiplication. If the result
1004 * overflows the desination vector, high-order bits are
1005 * discarded. The result vector may not overlap the argument
1006 * vectors in any way.
1009 void mpx_umul(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl,
1010 const mpw *bv, const mpw *bvl)
1012 /* --- This is probably worthwhile on a multiply --- */
1014 MPX_SHRINK(av, avl);
1015 MPX_SHRINK(bv, bvl);
1017 /* --- Deal with a multiply by zero --- */
1024 /* --- Do the initial multiply and initialize the accumulator --- */
1026 MPX_UMULN(dv, dvl, av, avl, *bv++);
1028 /* --- Do the remaining multiply/accumulates --- */
1030 while (dv < dvl && bv < bvl) {
1033 const mpw *avv = av;
1040 x = (mpd)*dvv + (mpd)m * (mpd)*avv++ + c;
1044 MPX_UADDN(dvv, dvl, c);
1049 /* --- @mpx_umuln@ --- *
1051 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
1052 * @const mpw *av, *avl@ = multiplicand vector base and limit
1053 * @mpw m@ = multiplier
1057 * Use: Multiplies a multiprecision integer by a single-word value.
1058 * The destination and source may be equal. The destination
1059 * is completely cleared after use.
1062 void mpx_umuln(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, mpw m)
1064 MPX_UMULN(dv, dvl, av, avl, m);
1067 /* --- @mpx_umlan@ --- *
1069 * Arguments: @mpw *dv, *dvl@ = destination/accumulator base and limit
1070 * @const mpw *av, *avl@ = multiplicand vector base and limit
1071 * @mpw m@ = multiplier
1075 * Use: Multiplies a multiprecision integer by a single-word value
1076 * and adds the result to an accumulator.
1079 void mpx_umlan(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, mpw m)
1081 MPX_UMLAN(dv, dvl, av, avl, m);
1084 /* --- @mpx_usqr@ --- *
1086 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
1087 * @const mpw *av, *av@ = source vector base and limit
1091 * Use: Performs unsigned integer squaring. The result vector must
1092 * not overlap the source vector in any way.
1095 void mpx_usqr(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl)
1099 /* --- Main loop --- */
1102 const mpw *avv = av;
1107 /* --- Stop if I've run out of destination --- */
1112 /* --- Work out the square at this point in the proceedings --- */
1115 mpd x = (mpd)a * (mpd)a + *dvv;
1117 c = MPW(x >> MPW_BITS);
1120 /* --- Now fix up the rest of the vector upwards --- */
1123 while (dvv < dvl && avv < avl) {
1124 mpd x = (mpd)a * (mpd)*avv++;
1125 mpd y = ((x << 1) & MPW_MAX) + c + *dvv;
1126 c = (x >> (MPW_BITS - 1)) + (y >> MPW_BITS);
1129 while (dvv < dvl && c) {
1135 /* --- Get ready for the next round --- */
1142 /* --- @mpx_udiv@ --- *
1144 * Arguments: @mpw *qv, *qvl@ = quotient vector base and limit
1145 * @mpw *rv, *rvl@ = dividend/remainder vector base and limit
1146 * @const mpw *dv, *dvl@ = divisor vector base and limit
1147 * @mpw *sv, *svl@ = scratch workspace
1151 * Use: Performs unsigned integer division. If the result overflows
1152 * the quotient vector, high-order bits are discarded. (Clearly
1153 * the remainder vector can't overflow.) The various vectors
1154 * may not overlap in any way. Yes, I know it's a bit odd
1155 * requiring the dividend to be in the result position but it
1156 * does make some sense really. The remainder must have
1157 * headroom for at least two extra words. The scratch space
1158 * must be at least one word larger than the divisor.
1161 void mpx_udiv(mpw *qv, mpw *qvl, mpw *rv, mpw *rvl,
1162 const mpw *dv, const mpw *dvl,
1169 /* --- Initialize the quotient --- */
1173 /* --- Perform some sanity checks --- */
1175 MPX_SHRINK(dv, dvl);
1176 assert(((void)"division by zero in mpx_udiv", dv < dvl));
1178 /* --- Normalize the divisor --- *
1180 * The algorithm requires that the divisor be at least two digits long.
1181 * This is easy to fix.
1188 for (b = MPW_BITS / 2; b; b >>= 1) {
1189 if (d <= (MPW_MAX >> b)) {
1198 /* --- Normalize the dividend/remainder to match --- */
1201 mpx_lsl(rv, rvl, rv, rvl, norm);
1202 mpx_lsl(sv, svl, dv, dvl, norm);
1205 MPX_SHRINK(dv, dvl);
1208 MPX_SHRINK(rv, rvl);
1212 /* --- Work out the relative scales --- */
1215 size_t rvn = rvl - rv;
1216 size_t dvn = dvl - dv;
1218 /* --- If the divisor is clearly larger, notice this --- */
1221 mpx_lsr(rv, rvl, rv, rvl, norm);
1228 /* --- Calculate the most significant quotient digit --- *
1230 * Because the divisor has its top bit set, this can only happen once. The
1231 * pointer arithmetic is a little contorted, to make sure that the
1232 * behaviour is defined.
1235 if (MPX_UCMP(rv + scale, rvl, >=, dv, dvl)) {
1236 mpx_usub(rv + scale, rvl, rv + scale, rvl, dv, dvl);
1237 if (qvl - qv > scale)
1241 /* --- Now for the main loop --- */
1250 /* --- Get an estimate for the next quotient digit --- */
1257 rh = ((mpd)r << MPW_BITS) | rr;
1263 /* --- Refine the estimate --- */
1266 mpd yh = (mpd)d * q;
1267 mpd yy = (mpd)dd * q;
1271 yh += yy >> MPW_BITS;
1274 while (yh > rh || (yh == rh && yl > rrr)) {
1283 /* --- Remove a chunk from the dividend --- */
1290 /* --- Calculate the size of the chunk --- *
1292 * This does the whole job of calculating @r >> scale - qd@.
1295 for (svv = rv + scale, dvv = dv;
1296 dvv < dvl && svv < rvl;
1298 mpd x = (mpd)*dvv * (mpd)q + mc;
1300 x = (mpd)*svv - MPW(x) - sc;
1309 mpd x = (mpd)*svv - mc - sc;
1319 /* --- Fix if the quotient was too large --- *
1321 * This doesn't seem to happen very often.
1324 if (rvl[-1] > MPW_MAX / 2) {
1325 mpx_uadd(rv + scale, rvl, rv + scale, rvl, dv, dvl);
1330 /* --- Done for another iteration --- */
1332 if (qvl - qv > scale)
1339 /* --- Now fiddle with unnormalizing and things --- */
1341 mpx_lsr(rv, rvl, rv, rvl, norm);
1344 /* --- @mpx_udivn@ --- *
1346 * Arguments: @mpw *qv, *qvl@ = storage for the quotient (may overlap
1348 * @const mpw *rv, *rvl@ = dividend
1349 * @mpw d@ = single-precision divisor
1351 * Returns: Remainder after divison.
1353 * Use: Performs a single-precision division operation.
1356 mpw mpx_udivn(mpw *qv, mpw *qvl, const mpw *rv, const mpw *rvl, mpw d)
1359 size_t ql = qvl - qv;
1365 r = (r << MPW_BITS) | rv[i];
1373 /*----- Test rig ----------------------------------------------------------*/
1377 #include <mLib/alloc.h>
1378 #include <mLib/dstr.h>
1379 #include <mLib/quis.h>
1380 #include <mLib/testrig.h>
1384 #define ALLOC(v, vl, sz) do { \
1385 size_t _sz = (sz); \
1386 mpw *_vv = xmalloc(MPWS(_sz)); \
1387 mpw *_vvl = _vv + _sz; \
1392 #define LOAD(v, vl, d) do { \
1393 const dstr *_d = (d); \
1395 ALLOC(_v, _vl, MPW_RQ(_d->len)); \
1396 mpx_loadb(_v, _vl, _d->buf, _d->len); \
1401 #define MAX(x, y) ((x) > (y) ? (x) : (y))
1403 static void dumpbits(const char *msg, const void *pp, size_t sz)
1405 const octet *p = pp;
1408 fprintf(stderr, " %02x", *p++);
1409 fputc('\n', stderr);
1412 static void dumpmp(const char *msg, const mpw *v, const mpw *vl)
1417 fprintf(stderr, " %08lx", (unsigned long)*--vl);
1418 fputc('\n', stderr);
1421 static int chkscan(const mpw *v, const mpw *vl,
1422 const void *pp, size_t sz, int step)
1425 const octet *p = pp;
1429 mpscan_initx(&mps, v, vl);
1434 for (i = 0; i < 8 && MPSCAN_STEP(&mps); i++) {
1435 if (MPSCAN_BIT(&mps) != (x & 1)) {
1437 "\n*** error, step %i, bit %u, expected %u, found %u\n",
1438 step, bit, x & 1, MPSCAN_BIT(&mps));
1450 static int loadstore(dstr *v)
1453 size_t sz = MPW_RQ(v->len) * 2, diff;
1457 dstr_ensure(&d, v->len);
1458 m = xmalloc(MPWS(sz));
1460 for (diff = 0; diff < sz; diff += 5) {
1465 mpx_loadl(m, ml, v->buf, v->len);
1466 if (!chkscan(m, ml, v->buf, v->len, +1))
1468 MPX_OCTETS(oct, m, ml);
1469 mpx_storel(m, ml, d.buf, d.sz);
1470 if (memcmp(d.buf, v->buf, oct) != 0) {
1471 dumpbits("\n*** storel failed", d.buf, d.sz);
1475 mpx_loadb(m, ml, v->buf, v->len);
1476 if (!chkscan(m, ml, v->buf + v->len - 1, v->len, -1))
1478 MPX_OCTETS(oct, m, ml);
1479 mpx_storeb(m, ml, d.buf, d.sz);
1480 if (memcmp(d.buf + d.sz - oct, v->buf + v->len - oct, oct) != 0) {
1481 dumpbits("\n*** storeb failed", d.buf, d.sz);
1487 dumpbits("input data", v->buf, v->len);
1494 static int twocl(dstr *v)
1501 sz = v[0].len; if (v[1].len > sz) sz = v[1].len;
1502 dstr_ensure(&d, sz);
1505 m = xmalloc(MPWS(sz));
1508 mpx_loadl(m, ml, v[0].buf, v[0].len);
1509 mpx_storel2cn(m, ml, d.buf, v[1].len);
1510 if (memcmp(d.buf, v[1].buf, v[1].len)) {
1511 dumpbits("\n*** storel2cn failed", d.buf, v[1].len);
1515 mpx_loadl2cn(m, ml, v[1].buf, v[1].len);
1516 mpx_storel(m, ml, d.buf, v[0].len);
1517 if (memcmp(d.buf, v[0].buf, v[0].len)) {
1518 dumpbits("\n*** loadl2cn failed", d.buf, v[0].len);
1523 dumpbits("pos", v[0].buf, v[0].len);
1524 dumpbits("neg", v[1].buf, v[1].len);
1533 static int twocb(dstr *v)
1540 sz = v[0].len; if (v[1].len > sz) sz = v[1].len;
1541 dstr_ensure(&d, sz);
1544 m = xmalloc(MPWS(sz));
1547 mpx_loadb(m, ml, v[0].buf, v[0].len);
1548 mpx_storeb2cn(m, ml, d.buf, v[1].len);
1549 if (memcmp(d.buf, v[1].buf, v[1].len)) {
1550 dumpbits("\n*** storeb2cn failed", d.buf, v[1].len);
1554 mpx_loadb2cn(m, ml, v[1].buf, v[1].len);
1555 mpx_storeb(m, ml, d.buf, v[0].len);
1556 if (memcmp(d.buf, v[0].buf, v[0].len)) {
1557 dumpbits("\n*** loadb2cn failed", d.buf, v[0].len);
1562 dumpbits("pos", v[0].buf, v[0].len);
1563 dumpbits("neg", v[1].buf, v[1].len);
1572 static int lsl(dstr *v)
1575 int n = *(int *)v[1].buf;
1582 ALLOC(d, dl, al - a + (n + MPW_BITS - 1) / MPW_BITS);
1584 mpx_lsl(d, dl, a, al, n);
1585 if (!mpx_ueq(d, dl, c, cl)) {
1586 fprintf(stderr, "\n*** lsl(%i) failed\n", n);
1587 dumpmp(" a", a, al);
1588 dumpmp("expected", c, cl);
1589 dumpmp(" result", d, dl);
1593 free(a); free(c); free(d);
1597 static int lslc(dstr *v)
1600 int n = *(int *)v[1].buf;
1607 ALLOC(d, dl, al - a + (n + MPW_BITS - 1) / MPW_BITS);
1609 mpx_lslc(d, dl, a, al, n);
1610 if (!mpx_ueq(d, dl, c, cl)) {
1611 fprintf(stderr, "\n*** lslc(%i) failed\n", n);
1612 dumpmp(" a", a, al);
1613 dumpmp("expected", c, cl);
1614 dumpmp(" result", d, dl);
1618 free(a); free(c); free(d);
1622 static int lsr(dstr *v)
1625 int n = *(int *)v[1].buf;
1632 ALLOC(d, dl, al - a + (n + MPW_BITS - 1) / MPW_BITS + 1);
1634 mpx_lsr(d, dl, a, al, n);
1635 if (!mpx_ueq(d, dl, c, cl)) {
1636 fprintf(stderr, "\n*** lsr(%i) failed\n", n);
1637 dumpmp(" a", a, al);
1638 dumpmp("expected", c, cl);
1639 dumpmp(" result", d, dl);
1643 free(a); free(c); free(d);
1647 static int uadd(dstr *v)
1658 ALLOC(d, dl, MAX(al - a, bl - b) + 1);
1660 mpx_uadd(d, dl, a, al, b, bl);
1661 if (!mpx_ueq(d, dl, c, cl)) {
1662 fprintf(stderr, "\n*** uadd failed\n");
1663 dumpmp(" a", a, al);
1664 dumpmp(" b", b, bl);
1665 dumpmp("expected", c, cl);
1666 dumpmp(" result", d, dl);
1670 free(a); free(b); free(c); free(d);
1674 static int usub(dstr *v)
1685 ALLOC(d, dl, al - a);
1687 mpx_usub(d, dl, a, al, b, bl);
1688 if (!mpx_ueq(d, dl, c, cl)) {
1689 fprintf(stderr, "\n*** usub failed\n");
1690 dumpmp(" a", a, al);
1691 dumpmp(" b", b, bl);
1692 dumpmp("expected", c, cl);
1693 dumpmp(" result", d, dl);
1697 free(a); free(b); free(c); free(d);
1701 static int umul(dstr *v)
1712 ALLOC(d, dl, (al - a) + (bl - b));
1714 mpx_umul(d, dl, a, al, b, bl);
1715 if (!mpx_ueq(d, dl, c, cl)) {
1716 fprintf(stderr, "\n*** umul failed\n");
1717 dumpmp(" a", a, al);
1718 dumpmp(" b", b, bl);
1719 dumpmp("expected", c, cl);
1720 dumpmp(" result", d, dl);
1724 free(a); free(b); free(c); free(d);
1728 static int usqr(dstr *v)
1737 ALLOC(d, dl, 2 * (al - a));
1739 mpx_usqr(d, dl, a, al);
1740 if (!mpx_ueq(d, dl, c, cl)) {
1741 fprintf(stderr, "\n*** usqr failed\n");
1742 dumpmp(" a", a, al);
1743 dumpmp("expected", c, cl);
1744 dumpmp(" result", d, dl);
1748 free(a); free(c); free(d);
1752 static int udiv(dstr *v)
1762 ALLOC(a, al, MPW_RQ(v[0].len) + 2); mpx_loadb(a, al, v[0].buf, v[0].len);
1766 ALLOC(qq, qql, al - a);
1767 ALLOC(s, sl, (bl - b) + 1);
1769 mpx_udiv(qq, qql, a, al, b, bl, s, sl);
1770 if (!mpx_ueq(qq, qql, q, ql) ||
1771 !mpx_ueq(a, al, r, rl)) {
1772 fprintf(stderr, "\n*** udiv failed\n");
1773 dumpmp(" divisor", b, bl);
1774 dumpmp("expect r", r, rl);
1775 dumpmp("result r", a, al);
1776 dumpmp("expect q", q, ql);
1777 dumpmp("result q", qq, qql);
1781 free(a); free(b); free(r); free(q); free(s); free(qq);
1785 static test_chunk defs[] = {
1786 { "load-store", loadstore, { &type_hex, 0 } },
1787 { "2cl", twocl, { &type_hex, &type_hex, } },
1788 { "2cb", twocb, { &type_hex, &type_hex, } },
1789 { "lsl", lsl, { &type_hex, &type_int, &type_hex, 0 } },
1790 { "lslc", lslc, { &type_hex, &type_int, &type_hex, 0 } },
1791 { "lsr", lsr, { &type_hex, &type_int, &type_hex, 0 } },
1792 { "uadd", uadd, { &type_hex, &type_hex, &type_hex, 0 } },
1793 { "usub", usub, { &type_hex, &type_hex, &type_hex, 0 } },
1794 { "umul", umul, { &type_hex, &type_hex, &type_hex, 0 } },
1795 { "usqr", usqr, { &type_hex, &type_hex, 0 } },
1796 { "udiv", udiv, { &type_hex, &type_hex, &type_hex, &type_hex, 0 } },
1800 int main(int argc, char *argv[])
1802 test_run(argc, argv, defs, SRCDIR"/tests/mpx");
1808 /*----- That's all, folks -------------------------------------------------*/