3 * $Id: mpx.c,v 1.17 2004/03/27 00:04:46 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 1.17 2004/03/27 00:04:46 mdw
34 * Implement efficient reduction for pleasant-looking primes.
36 * Revision 1.16 2003/05/16 09:09:24 mdw
37 * Fix @mp_lsl2c@. Turns out to be surprisingly tricky.
39 * Revision 1.15 2002/10/20 01:12:31 mdw
40 * Two's complement I/O fixes.
42 * Revision 1.14 2002/10/19 18:55:08 mdw
43 * Fix overflows in shift primitives.
45 * Revision 1.13 2002/10/19 17:56:50 mdw
46 * Fix bit operations. Test them (a bit) better.
48 * Revision 1.12 2002/10/06 22:52:50 mdw
49 * Pile of changes for supporting two's complement properly.
51 * Revision 1.11 2001/04/03 19:36:05 mdw
52 * Add some simple bitwise operations so that Perl can use them.
54 * Revision 1.10 2000/10/08 12:06:12 mdw
55 * Provide @mpx_ueq@ for rapidly testing equality of two integers.
57 * Revision 1.9 2000/06/26 07:52:50 mdw
58 * Portability fix for the bug fix.
60 * Revision 1.8 2000/06/25 12:59:02 mdw
61 * (mpx_udiv): Fix bug in quotient digit estimation.
63 * Revision 1.7 1999/12/22 15:49:07 mdw
64 * New function for division by a small integer.
66 * Revision 1.6 1999/11/20 22:43:44 mdw
67 * Integrate testing for MPX routines.
69 * Revision 1.5 1999/11/20 22:23:27 mdw
70 * Add function versions of some low-level macros with wider use.
72 * Revision 1.4 1999/11/17 18:04:09 mdw
73 * Add two's-complement functionality. Improve mpx_udiv a little by
74 * performing the multiplication of the divisor by q with the subtraction
77 * Revision 1.3 1999/11/13 01:57:31 mdw
78 * Remove stray debugging code.
80 * Revision 1.2 1999/11/13 01:50:59 mdw
81 * Multiprecision routines finished and tested.
83 * Revision 1.1 1999/09/03 08:41:12 mdw
88 /*----- Header files ------------------------------------------------------*/
95 #include <mLib/bits.h>
101 /*----- Loading and storing -----------------------------------------------*/
103 /* --- @mpx_storel@ --- *
105 * Arguments: @const mpw *v, *vl@ = base and limit of source vector
106 * @void *pp@ = pointer to octet array
107 * @size_t sz@ = size of octet array
111 * Use: Stores an MP in an octet array, least significant octet
112 * first. High-end octets are silently discarded if there
113 * isn't enough space for them.
116 void mpx_storel(const mpw *v, const mpw *vl, void *pp, size_t sz)
119 octet *p = pp, *q = p + sz;
129 *p++ = U8(w | n << bits);
131 bits += MPW_BITS - 8;
141 /* --- @mpx_loadl@ --- *
143 * Arguments: @mpw *v, *vl@ = base and limit of destination vector
144 * @const void *pp@ = pointer to octet array
145 * @size_t sz@ = size of octet array
149 * Use: Loads an MP in an octet array, least significant octet
150 * first. High-end octets are ignored if there isn't enough
154 void mpx_loadl(mpw *v, mpw *vl, const void *pp, size_t sz)
158 const octet *p = pp, *q = p + sz;
167 if (bits >= MPW_BITS) {
169 w = n >> (MPW_BITS - bits + 8);
179 /* --- @mpx_storeb@ --- *
181 * Arguments: @const mpw *v, *vl@ = base and limit of source vector
182 * @void *pp@ = pointer to octet array
183 * @size_t sz@ = size of octet array
187 * Use: Stores an MP in an octet array, most significant octet
188 * first. High-end octets are silently discarded if there
189 * isn't enough space for them.
192 void mpx_storeb(const mpw *v, const mpw *vl, void *pp, size_t sz)
195 octet *p = pp, *q = p + sz;
205 *--q = U8(w | n << bits);
207 bits += MPW_BITS - 8;
217 /* --- @mpx_loadb@ --- *
219 * Arguments: @mpw *v, *vl@ = base and limit of destination vector
220 * @const void *pp@ = pointer to octet array
221 * @size_t sz@ = size of octet array
225 * Use: Loads an MP in an octet array, most significant octet
226 * first. High-end octets are ignored if there isn't enough
230 void mpx_loadb(mpw *v, mpw *vl, const void *pp, size_t sz)
234 const octet *p = pp, *q = p + sz;
243 if (bits >= MPW_BITS) {
245 w = n >> (MPW_BITS - bits + 8);
255 /* --- @mpx_storel2cn@ --- *
257 * Arguments: @const mpw *v, *vl@ = base and limit of source vector
258 * @void *pp@ = pointer to octet array
259 * @size_t sz@ = size of octet array
263 * Use: Stores a negative MP in an octet array, least significant
264 * octet first, as two's complement. High-end octets are
265 * silently discarded if there isn't enough space for them.
266 * This obviously makes the output bad.
269 void mpx_storel2cn(const mpw *v, const mpw *vl, void *pp, size_t sz)
274 octet *p = pp, *q = p + sz;
286 bits += MPW_BITS - 8;
304 /* --- @mpx_loadl2cn@ --- *
306 * Arguments: @mpw *v, *vl@ = base and limit of destination vector
307 * @const void *pp@ = pointer to octet array
308 * @size_t sz@ = size of octet array
312 * Use: Loads a negative MP in an octet array, least significant
313 * octet first, as two's complement. High-end octets are
314 * ignored if there isn't enough space for them. This probably
315 * means you made the wrong choice coming here.
318 void mpx_loadl2cn(mpw *v, mpw *vl, const void *pp, size_t sz)
323 const octet *p = pp, *q = p + sz;
333 if (bits >= MPW_BITS) {
335 w = n >> (MPW_BITS - bits + 8);
345 /* --- @mpx_storeb2cn@ --- *
347 * Arguments: @const mpw *v, *vl@ = base and limit of source vector
348 * @void *pp@ = pointer to octet array
349 * @size_t sz@ = size of octet array
353 * Use: Stores a negative MP in an octet array, most significant
354 * octet first, as two's complement. High-end octets are
355 * silently discarded if there isn't enough space for them,
356 * which probably isn't what you meant.
359 void mpx_storeb2cn(const mpw *v, const mpw *vl, void *pp, size_t sz)
364 octet *p = pp, *q = p + sz;
376 bits += MPW_BITS - 8;
388 c = c && !(b & 0xff);
394 /* --- @mpx_loadb2cn@ --- *
396 * Arguments: @mpw *v, *vl@ = base and limit of destination vector
397 * @const void *pp@ = pointer to octet array
398 * @size_t sz@ = size of octet array
402 * Use: Loads a negative MP in an octet array, most significant octet
403 * first as two's complement. High-end octets are ignored if
404 * there isn't enough space for them. This probably means you
405 * chose this function wrongly.
408 void mpx_loadb2cn(mpw *v, mpw *vl, const void *pp, size_t sz)
413 const octet *p = pp, *q = p + sz;
423 if (bits >= MPW_BITS) {
425 w = n >> (MPW_BITS - bits + 8);
435 /*----- Logical shifting --------------------------------------------------*/
437 /* --- @mpx_lsl@ --- *
439 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
440 * @const mpw *av, *avl@ = source vector base and limit
441 * @size_t n@ = number of bit positions to shift by
445 * Use: Performs a logical shift left operation on an integer.
448 void mpx_lsl(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, size_t n)
453 /* --- Trivial special case --- */
456 MPX_COPY(dv, dvl, av, avl);
458 /* --- Single bit shifting --- */
467 *dv++ = MPW((t << 1) | w);
468 w = t >> (MPW_BITS - 1);
477 /* --- Break out word and bit shifts for more sophisticated work --- */
482 /* --- Handle a shift by a multiple of the word size --- */
488 MPX_COPY(dv + nw, dvl, av, avl);
489 memset(dv, 0, MPWS(nw));
493 /* --- And finally the difficult case --- *
495 * This is a little convoluted, because I have to start from the end and
496 * work backwards to avoid overwriting the source, if they're both the same
502 size_t nr = MPW_BITS - nb;
503 size_t dvn = dvl - dv;
504 size_t avn = avl - av;
511 if (dvn > avn + nw) {
512 size_t off = avn + nw + 1;
513 MPX_ZERO(dv + off, dvl);
523 *--dvl = (t >> nr) | w;
534 /* --- @mpx_lslc@ --- *
536 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
537 * @const mpw *av, *avl@ = source vector base and limit
538 * @size_t n@ = number of bit positions to shift by
542 * Use: Performs a logical shift left operation on an integer, only
543 * it fills in the bits with ones instead of zeroes.
546 void mpx_lslc(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, size_t n)
551 /* --- Trivial special case --- */
554 MPX_COPY(dv, dvl, av, avl);
556 /* --- Single bit shifting --- */
565 *dv++ = MPW((t << 1) | w);
566 w = t >> (MPW_BITS - 1);
575 /* --- Break out word and bit shifts for more sophisticated work --- */
580 /* --- Handle a shift by a multiple of the word size --- */
586 MPX_COPY(dv + nw, dvl, av, avl);
587 MPX_ONE(dv, dv + nw);
591 /* --- And finally the difficult case --- *
593 * This is a little convoluted, because I have to start from the end and
594 * work backwards to avoid overwriting the source, if they're both the same
600 size_t nr = MPW_BITS - nb;
601 size_t dvn = dvl - dv;
602 size_t avn = avl - av;
609 if (dvn > avn + nw) {
610 size_t off = avn + nw + 1;
611 MPX_ZERO(dv + off, dvl);
621 *--dvl = (t >> nr) | w;
625 *--dvl = (MPW_MAX >> nr) | w;
632 /* --- @mpx_lsr@ --- *
634 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
635 * @const mpw *av, *avl@ = source vector base and limit
636 * @size_t n@ = number of bit positions to shift by
640 * Use: Performs a logical shift right operation on an integer.
643 void mpx_lsr(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, size_t n)
648 /* --- Trivial special case --- */
651 MPX_COPY(dv, dvl, av, avl);
653 /* --- Single bit shifting --- */
662 *dv++ = MPW((t << (MPW_BITS - 1)) | w);
672 /* --- Break out word and bit shifts for more sophisticated work --- */
677 /* --- Handle a shift by a multiple of the word size --- */
683 MPX_COPY(dv, dvl, av + nw, avl);
686 /* --- And finally the difficult case --- */
690 size_t nr = MPW_BITS - nb;
693 w = av < avl ? *av++ : 0;
699 *dv++ = MPW((w >> nb) | (t << nr));
703 *dv++ = MPW(w >> nb);
711 /*----- Bitwise operations ------------------------------------------------*/
713 /* --- @mpx_bitop@ --- *
715 * Arguments: @mpw *dv, *dvl@ = destination vector
716 * @const mpw *av, *avl@ = first source vector
717 * @const mpw *bv, *bvl@ = second source vector
721 * Use; Provides the dyadic boolean functions.
724 #define MPX_BITBINOP(string) \
726 void mpx_bit##string(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, \
727 const mpw *bv, const mpw *bvl) \
729 MPX_SHRINK(av, avl); \
730 MPX_SHRINK(bv, bvl); \
734 a = (av < avl) ? *av++ : 0; \
735 b = (bv < bvl) ? *bv++ : 0; \
736 *dv++ = B##string(a, b); \
740 MPX_DOBIN(MPX_BITBINOP)
742 void mpx_not(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl)
748 a = (av < avl) ? *av++ : 0;
753 /*----- Unsigned arithmetic -----------------------------------------------*/
755 /* --- @mpx_2c@ --- *
757 * Arguments: @mpw *dv, *dvl@ = destination vector
758 * @const mpw *v, *vl@ = source vector
762 * Use: Calculates the two's complement of @v@.
765 void mpx_2c(mpw *dv, mpw *dvl, const mpw *v, const mpw *vl)
768 while (dv < dvl && v < vl)
769 *dv++ = c = MPW(~*v++);
776 MPX_UADDN(dv, dvl, 1);
779 /* --- @mpx_ueq@ --- *
781 * Arguments: @const mpw *av, *avl@ = first argument vector base and limit
782 * @const mpw *bv, *bvl@ = second argument vector base and limit
784 * Returns: Nonzero if the two vectors are equal.
786 * Use: Performs an unsigned integer test for equality.
789 int mpx_ueq(const mpw *av, const mpw *avl, const mpw *bv, const mpw *bvl)
793 if (avl - av != bvl - bv)
802 /* --- @mpx_ucmp@ --- *
804 * Arguments: @const mpw *av, *avl@ = first argument vector base and limit
805 * @const mpw *bv, *bvl@ = second argument vector base and limit
807 * Returns: Less than, equal to, or greater than zero depending on
808 * whether @a@ is less than, equal to or greater than @b@,
811 * Use: Performs an unsigned integer comparison.
814 int mpx_ucmp(const mpw *av, const mpw *avl, const mpw *bv, const mpw *bvl)
819 if (avl - av > bvl - bv)
821 else if (avl - av < bvl - bv)
823 else while (avl > av) {
824 mpw a = *--avl, b = *--bvl;
833 /* --- @mpx_uadd@ --- *
835 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
836 * @const mpw *av, *avl@ = first addend vector base and limit
837 * @const mpw *bv, *bvl@ = second addend vector base and limit
841 * Use: Performs unsigned integer addition. If the result overflows
842 * the destination vector, high-order bits are discarded. This
843 * means that two's complement addition happens more or less for
844 * free, although that's more a side-effect than anything else.
845 * The result vector may be equal to either or both source
846 * vectors, but may not otherwise overlap them.
849 void mpx_uadd(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl,
850 const mpw *bv, const mpw *bvl)
854 while (av < avl || bv < bvl) {
859 a = (av < avl) ? *av++ : 0;
860 b = (bv < bvl) ? *bv++ : 0;
861 x = (mpd)a + (mpd)b + c;
871 /* --- @mpx_uaddn@ --- *
873 * Arguments: @mpw *dv, *dvl@ = source and destination base and limit
874 * @mpw n@ = other addend
878 * Use: Adds a small integer to a multiprecision number.
881 void mpx_uaddn(mpw *dv, mpw *dvl, mpw n) { MPX_UADDN(dv, dvl, n); }
883 /* --- @mpx_uaddnlsl@ --- *
885 * Arguments: @mpw *dv, *dvl@ = destination and first argument vector
886 * @mpw a@ = second argument
887 * @unsigned o@ = offset in bits
891 * Use: Computes %$d + 2^o a$%. If the result overflows then
892 * high-order bits are discarded, as usual. We must have
893 * @0 < o < MPW_BITS@.
896 void mpx_uaddnlsl(mpw *dv, mpw *dvl, mpw a, unsigned o)
900 while (x && dv < dvl) {
907 /* --- @mpx_usub@ --- *
909 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
910 * @const mpw *av, *avl@ = first argument vector base and limit
911 * @const mpw *bv, *bvl@ = second argument vector base and limit
915 * Use: Performs unsigned integer subtraction. If the result
916 * overflows the destination vector, high-order bits are
917 * discarded. This means that two's complement subtraction
918 * happens more or less for free, althuogh that's more a side-
919 * effect than anything else. The result vector may be equal to
920 * either or both source vectors, but may not otherwise overlap
924 void mpx_usub(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl,
925 const mpw *bv, const mpw *bvl)
929 while (av < avl || bv < bvl) {
934 a = (av < avl) ? *av++ : 0;
935 b = (bv < bvl) ? *bv++ : 0;
936 x = (mpd)a - (mpd)b - c;
949 /* --- @mpx_usubn@ --- *
951 * Arguments: @mpw *dv, *dvl@ = source and destination base and limit
956 * Use: Subtracts a small integer from a multiprecision number.
959 void mpx_usubn(mpw *dv, mpw *dvl, mpw n) { MPX_USUBN(dv, dvl, n); }
961 /* --- @mpx_uaddnlsl@ --- *
963 * Arguments: @mpw *dv, *dvl@ = destination and first argument vector
964 * @mpw a@ = second argument
965 * @unsigned o@ = offset in bits
969 * Use: Computes %$d + 2^o a$%. If the result overflows then
970 * high-order bits are discarded, as usual. We must have
971 * @0 < o < MPW_BITS@.
974 void mpx_usubnlsl(mpw *dv, mpw *dvl, mpw a, unsigned o)
976 mpw b = a >> (MPW_BITS - o);
980 mpd x = (mpd)*dv - (mpd)a;
984 MPX_USUBN(dv, dvl, b);
988 /* --- @mpx_umul@ --- *
990 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
991 * @const mpw *av, *avl@ = multiplicand vector base and limit
992 * @const mpw *bv, *bvl@ = multiplier vector base and limit
996 * Use: Performs unsigned integer multiplication. If the result
997 * overflows the desination vector, high-order bits are
998 * discarded. The result vector may not overlap the argument
999 * vectors in any way.
1002 void mpx_umul(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl,
1003 const mpw *bv, const mpw *bvl)
1005 /* --- This is probably worthwhile on a multiply --- */
1007 MPX_SHRINK(av, avl);
1008 MPX_SHRINK(bv, bvl);
1010 /* --- Deal with a multiply by zero --- */
1017 /* --- Do the initial multiply and initialize the accumulator --- */
1019 MPX_UMULN(dv, dvl, av, avl, *bv++);
1021 /* --- Do the remaining multiply/accumulates --- */
1023 while (dv < dvl && bv < bvl) {
1026 const mpw *avv = av;
1033 x = (mpd)*dvv + (mpd)m * (mpd)*avv++ + c;
1037 MPX_UADDN(dvv, dvl, c);
1042 /* --- @mpx_umuln@ --- *
1044 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
1045 * @const mpw *av, *avl@ = multiplicand vector base and limit
1046 * @mpw m@ = multiplier
1050 * Use: Multiplies a multiprecision integer by a single-word value.
1051 * The destination and source may be equal. The destination
1052 * is completely cleared after use.
1055 void mpx_umuln(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, mpw m)
1057 MPX_UMULN(dv, dvl, av, avl, m);
1060 /* --- @mpx_umlan@ --- *
1062 * Arguments: @mpw *dv, *dvl@ = destination/accumulator base and limit
1063 * @const mpw *av, *avl@ = multiplicand vector base and limit
1064 * @mpw m@ = multiplier
1068 * Use: Multiplies a multiprecision integer by a single-word value
1069 * and adds the result to an accumulator.
1072 void mpx_umlan(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, mpw m)
1074 MPX_UMLAN(dv, dvl, av, avl, m);
1077 /* --- @mpx_usqr@ --- *
1079 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
1080 * @const mpw *av, *av@ = source vector base and limit
1084 * Use: Performs unsigned integer squaring. The result vector must
1085 * not overlap the source vector in any way.
1088 void mpx_usqr(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl)
1092 /* --- Main loop --- */
1095 const mpw *avv = av;
1100 /* --- Stop if I've run out of destination --- */
1105 /* --- Work out the square at this point in the proceedings --- */
1108 mpd x = (mpd)a * (mpd)a + *dvv;
1110 c = MPW(x >> MPW_BITS);
1113 /* --- Now fix up the rest of the vector upwards --- */
1116 while (dvv < dvl && avv < avl) {
1117 mpd x = (mpd)a * (mpd)*avv++;
1118 mpd y = ((x << 1) & MPW_MAX) + c + *dvv;
1119 c = (x >> (MPW_BITS - 1)) + (y >> MPW_BITS);
1122 while (dvv < dvl && c) {
1128 /* --- Get ready for the next round --- */
1135 /* --- @mpx_udiv@ --- *
1137 * Arguments: @mpw *qv, *qvl@ = quotient vector base and limit
1138 * @mpw *rv, *rvl@ = dividend/remainder vector base and limit
1139 * @const mpw *dv, *dvl@ = divisor vector base and limit
1140 * @mpw *sv, *svl@ = scratch workspace
1144 * Use: Performs unsigned integer division. If the result overflows
1145 * the quotient vector, high-order bits are discarded. (Clearly
1146 * the remainder vector can't overflow.) The various vectors
1147 * may not overlap in any way. Yes, I know it's a bit odd
1148 * requiring the dividend to be in the result position but it
1149 * does make some sense really. The remainder must have
1150 * headroom for at least two extra words. The scratch space
1151 * must be at least one word larger than the divisor.
1154 void mpx_udiv(mpw *qv, mpw *qvl, mpw *rv, mpw *rvl,
1155 const mpw *dv, const mpw *dvl,
1162 /* --- Initialize the quotient --- */
1166 /* --- Perform some sanity checks --- */
1168 MPX_SHRINK(dv, dvl);
1169 assert(((void)"division by zero in mpx_udiv", dv < dvl));
1171 /* --- Normalize the divisor --- *
1173 * The algorithm requires that the divisor be at least two digits long.
1174 * This is easy to fix.
1181 for (b = MPW_BITS / 2; b; b >>= 1) {
1182 if (d < (MPW_MAX >> b)) {
1191 /* --- Normalize the dividend/remainder to match --- */
1194 mpx_lsl(rv, rvl, rv, rvl, norm);
1195 mpx_lsl(sv, svl, dv, dvl, norm);
1198 MPX_SHRINK(dv, dvl);
1201 MPX_SHRINK(rv, rvl);
1205 /* --- Work out the relative scales --- */
1208 size_t rvn = rvl - rv;
1209 size_t dvn = dvl - dv;
1211 /* --- If the divisor is clearly larger, notice this --- */
1214 mpx_lsr(rv, rvl, rv, rvl, norm);
1221 /* --- Calculate the most significant quotient digit --- *
1223 * Because the divisor has its top bit set, this can only happen once. The
1224 * pointer arithmetic is a little contorted, to make sure that the
1225 * behaviour is defined.
1228 if (MPX_UCMP(rv + scale, rvl, >=, dv, dvl)) {
1229 mpx_usub(rv + scale, rvl, rv + scale, rvl, dv, dvl);
1230 if (qvl - qv > scale)
1234 /* --- Now for the main loop --- */
1243 /* --- Get an estimate for the next quotient digit --- */
1250 rh = ((mpd)r << MPW_BITS) | rr;
1256 /* --- Refine the estimate --- */
1259 mpd yh = (mpd)d * q;
1260 mpd yy = (mpd)dd * q;
1264 yh += yy >> MPW_BITS;
1267 while (yh > rh || (yh == rh && yl > rrr)) {
1276 /* --- Remove a chunk from the dividend --- */
1283 /* --- Calculate the size of the chunk --- *
1285 * This does the whole job of calculating @r >> scale - qd@.
1288 for (svv = rv + scale, dvv = dv;
1289 dvv < dvl && svv < rvl;
1291 mpd x = (mpd)*dvv * (mpd)q + mc;
1293 x = (mpd)*svv - MPW(x) - sc;
1302 mpd x = (mpd)*svv - mc - sc;
1312 /* --- Fix if the quotient was too large --- *
1314 * This doesn't seem to happen very often.
1317 if (rvl[-1] > MPW_MAX / 2) {
1318 mpx_uadd(rv + scale, rvl, rv + scale, rvl, dv, dvl);
1323 /* --- Done for another iteration --- */
1325 if (qvl - qv > scale)
1332 /* --- Now fiddle with unnormalizing and things --- */
1334 mpx_lsr(rv, rvl, rv, rvl, norm);
1337 /* --- @mpx_udivn@ --- *
1339 * Arguments: @mpw *qv, *qvl@ = storage for the quotient (may overlap
1341 * @const mpw *rv, *rvl@ = dividend
1342 * @mpw d@ = single-precision divisor
1344 * Returns: Remainder after divison.
1346 * Use: Performs a single-precision division operation.
1349 mpw mpx_udivn(mpw *qv, mpw *qvl, const mpw *rv, const mpw *rvl, mpw d)
1352 size_t ql = qvl - qv;
1358 r = (r << MPW_BITS) | rv[i];
1366 /*----- Test rig ----------------------------------------------------------*/
1370 #include <mLib/alloc.h>
1371 #include <mLib/dstr.h>
1372 #include <mLib/quis.h>
1373 #include <mLib/testrig.h>
1377 #define ALLOC(v, vl, sz) do { \
1378 size_t _sz = (sz); \
1379 mpw *_vv = xmalloc(MPWS(_sz)); \
1380 mpw *_vvl = _vv + _sz; \
1385 #define LOAD(v, vl, d) do { \
1386 const dstr *_d = (d); \
1388 ALLOC(_v, _vl, MPW_RQ(_d->len)); \
1389 mpx_loadb(_v, _vl, _d->buf, _d->len); \
1394 #define MAX(x, y) ((x) > (y) ? (x) : (y))
1396 static void dumpbits(const char *msg, const void *pp, size_t sz)
1398 const octet *p = pp;
1401 fprintf(stderr, " %02x", *p++);
1402 fputc('\n', stderr);
1405 static void dumpmp(const char *msg, const mpw *v, const mpw *vl)
1410 fprintf(stderr, " %08lx", (unsigned long)*--vl);
1411 fputc('\n', stderr);
1414 static int chkscan(const mpw *v, const mpw *vl,
1415 const void *pp, size_t sz, int step)
1418 const octet *p = pp;
1422 mpscan_initx(&mps, v, vl);
1427 for (i = 0; i < 8 && MPSCAN_STEP(&mps); i++) {
1428 if (MPSCAN_BIT(&mps) != (x & 1)) {
1430 "\n*** error, step %i, bit %u, expected %u, found %u\n",
1431 step, bit, x & 1, MPSCAN_BIT(&mps));
1443 static int loadstore(dstr *v)
1446 size_t sz = MPW_RQ(v->len) * 2, diff;
1450 dstr_ensure(&d, v->len);
1451 m = xmalloc(MPWS(sz));
1453 for (diff = 0; diff < sz; diff += 5) {
1458 mpx_loadl(m, ml, v->buf, v->len);
1459 if (!chkscan(m, ml, v->buf, v->len, +1))
1461 MPX_OCTETS(oct, m, ml);
1462 mpx_storel(m, ml, d.buf, d.sz);
1463 if (memcmp(d.buf, v->buf, oct) != 0) {
1464 dumpbits("\n*** storel failed", d.buf, d.sz);
1468 mpx_loadb(m, ml, v->buf, v->len);
1469 if (!chkscan(m, ml, v->buf + v->len - 1, v->len, -1))
1471 MPX_OCTETS(oct, m, ml);
1472 mpx_storeb(m, ml, d.buf, d.sz);
1473 if (memcmp(d.buf + d.sz - oct, v->buf + v->len - oct, oct) != 0) {
1474 dumpbits("\n*** storeb failed", d.buf, d.sz);
1480 dumpbits("input data", v->buf, v->len);
1487 static int twocl(dstr *v)
1494 sz = v[0].len; if (v[1].len > sz) sz = v[1].len;
1495 dstr_ensure(&d, sz);
1498 m = xmalloc(MPWS(sz));
1501 mpx_loadl(m, ml, v[0].buf, v[0].len);
1502 mpx_storel2cn(m, ml, d.buf, v[1].len);
1503 if (memcmp(d.buf, v[1].buf, v[1].len)) {
1504 dumpbits("\n*** storel2cn failed", d.buf, v[1].len);
1508 mpx_loadl2cn(m, ml, v[1].buf, v[1].len);
1509 mpx_storel(m, ml, d.buf, v[0].len);
1510 if (memcmp(d.buf, v[0].buf, v[0].len)) {
1511 dumpbits("\n*** loadl2cn failed", d.buf, v[0].len);
1516 dumpbits("pos", v[0].buf, v[0].len);
1517 dumpbits("neg", v[1].buf, v[1].len);
1526 static int twocb(dstr *v)
1533 sz = v[0].len; if (v[1].len > sz) sz = v[1].len;
1534 dstr_ensure(&d, sz);
1537 m = xmalloc(MPWS(sz));
1540 mpx_loadb(m, ml, v[0].buf, v[0].len);
1541 mpx_storeb2cn(m, ml, d.buf, v[1].len);
1542 if (memcmp(d.buf, v[1].buf, v[1].len)) {
1543 dumpbits("\n*** storeb2cn failed", d.buf, v[1].len);
1547 mpx_loadb2cn(m, ml, v[1].buf, v[1].len);
1548 mpx_storeb(m, ml, d.buf, v[0].len);
1549 if (memcmp(d.buf, v[0].buf, v[0].len)) {
1550 dumpbits("\n*** loadb2cn failed", d.buf, v[0].len);
1555 dumpbits("pos", v[0].buf, v[0].len);
1556 dumpbits("neg", v[1].buf, v[1].len);
1565 static int lsl(dstr *v)
1568 int n = *(int *)v[1].buf;
1575 ALLOC(d, dl, al - a + (n + MPW_BITS - 1) / MPW_BITS);
1577 mpx_lsl(d, dl, a, al, n);
1578 if (!mpx_ueq(d, dl, c, cl)) {
1579 fprintf(stderr, "\n*** lsl(%i) failed\n", n);
1580 dumpmp(" a", a, al);
1581 dumpmp("expected", c, cl);
1582 dumpmp(" result", d, dl);
1586 free(a); free(c); free(d);
1590 static int lslc(dstr *v)
1593 int n = *(int *)v[1].buf;
1600 ALLOC(d, dl, al - a + (n + MPW_BITS - 1) / MPW_BITS);
1602 mpx_lslc(d, dl, a, al, n);
1603 if (!mpx_ueq(d, dl, c, cl)) {
1604 fprintf(stderr, "\n*** lslc(%i) failed\n", n);
1605 dumpmp(" a", a, al);
1606 dumpmp("expected", c, cl);
1607 dumpmp(" result", d, dl);
1611 free(a); free(c); free(d);
1615 static int lsr(dstr *v)
1618 int n = *(int *)v[1].buf;
1625 ALLOC(d, dl, al - a + (n + MPW_BITS - 1) / MPW_BITS + 1);
1627 mpx_lsr(d, dl, a, al, n);
1628 if (!mpx_ueq(d, dl, c, cl)) {
1629 fprintf(stderr, "\n*** lsr(%i) failed\n", n);
1630 dumpmp(" a", a, al);
1631 dumpmp("expected", c, cl);
1632 dumpmp(" result", d, dl);
1636 free(a); free(c); free(d);
1640 static int uadd(dstr *v)
1651 ALLOC(d, dl, MAX(al - a, bl - b) + 1);
1653 mpx_uadd(d, dl, a, al, b, bl);
1654 if (!mpx_ueq(d, dl, c, cl)) {
1655 fprintf(stderr, "\n*** uadd failed\n");
1656 dumpmp(" a", a, al);
1657 dumpmp(" b", b, bl);
1658 dumpmp("expected", c, cl);
1659 dumpmp(" result", d, dl);
1663 free(a); free(b); free(c); free(d);
1667 static int usub(dstr *v)
1678 ALLOC(d, dl, al - a);
1680 mpx_usub(d, dl, a, al, b, bl);
1681 if (!mpx_ueq(d, dl, c, cl)) {
1682 fprintf(stderr, "\n*** usub failed\n");
1683 dumpmp(" a", a, al);
1684 dumpmp(" b", b, bl);
1685 dumpmp("expected", c, cl);
1686 dumpmp(" result", d, dl);
1690 free(a); free(b); free(c); free(d);
1694 static int umul(dstr *v)
1705 ALLOC(d, dl, (al - a) + (bl - b));
1707 mpx_umul(d, dl, a, al, b, bl);
1708 if (!mpx_ueq(d, dl, c, cl)) {
1709 fprintf(stderr, "\n*** umul failed\n");
1710 dumpmp(" a", a, al);
1711 dumpmp(" b", b, bl);
1712 dumpmp("expected", c, cl);
1713 dumpmp(" result", d, dl);
1717 free(a); free(b); free(c); free(d);
1721 static int usqr(dstr *v)
1730 ALLOC(d, dl, 2 * (al - a));
1732 mpx_usqr(d, dl, a, al);
1733 if (!mpx_ueq(d, dl, c, cl)) {
1734 fprintf(stderr, "\n*** usqr failed\n");
1735 dumpmp(" a", a, al);
1736 dumpmp("expected", c, cl);
1737 dumpmp(" result", d, dl);
1741 free(a); free(c); free(d);
1745 static int udiv(dstr *v)
1755 ALLOC(a, al, MPW_RQ(v[0].len) + 2); mpx_loadb(a, al, v[0].buf, v[0].len);
1759 ALLOC(qq, qql, al - a);
1760 ALLOC(s, sl, (bl - b) + 1);
1762 mpx_udiv(qq, qql, a, al, b, bl, s, sl);
1763 if (!mpx_ueq(qq, qql, q, ql) ||
1764 !mpx_ueq(a, al, r, rl)) {
1765 fprintf(stderr, "\n*** udiv failed\n");
1766 dumpmp(" divisor", b, bl);
1767 dumpmp("expect r", r, rl);
1768 dumpmp("result r", a, al);
1769 dumpmp("expect q", q, ql);
1770 dumpmp("result q", qq, qql);
1774 free(a); free(b); free(r); free(q); free(s); free(qq);
1778 static test_chunk defs[] = {
1779 { "load-store", loadstore, { &type_hex, 0 } },
1780 { "2cl", twocl, { &type_hex, &type_hex, } },
1781 { "2cb", twocb, { &type_hex, &type_hex, } },
1782 { "lsl", lsl, { &type_hex, &type_int, &type_hex, 0 } },
1783 { "lslc", lslc, { &type_hex, &type_int, &type_hex, 0 } },
1784 { "lsr", lsr, { &type_hex, &type_int, &type_hex, 0 } },
1785 { "uadd", uadd, { &type_hex, &type_hex, &type_hex, 0 } },
1786 { "usub", usub, { &type_hex, &type_hex, &type_hex, 0 } },
1787 { "umul", umul, { &type_hex, &type_hex, &type_hex, 0 } },
1788 { "usqr", usqr, { &type_hex, &type_hex, 0 } },
1789 { "udiv", udiv, { &type_hex, &type_hex, &type_hex, &type_hex, 0 } },
1793 int main(int argc, char *argv[])
1795 test_run(argc, argv, defs, SRCDIR"/tests/mpx");
1801 /*----- That's all, folks -------------------------------------------------*/