3 * $Id: mpx.c,v 1.4 1999/11/17 18:04:09 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 1.4 1999/11/17 18:04:09 mdw
34 * Add two's-complement functionality. Improve mpx_udiv a little by
35 * performing the multiplication of the divisor by q with the subtraction
38 * Revision 1.3 1999/11/13 01:57:31 mdw
39 * Remove stray debugging code.
41 * Revision 1.2 1999/11/13 01:50:59 mdw
42 * Multiprecision routines finished and tested.
44 * Revision 1.1 1999/09/03 08:41:12 mdw
49 /*----- Header files ------------------------------------------------------*/
56 #include <mLib/bits.h>
61 /*----- Loading and storing -----------------------------------------------*/
63 /* --- @mpx_storel@ --- *
65 * Arguments: @const mpw *v, *vl@ = base and limit of source vector
66 * @void *pp@ = pointer to octet array
67 * @size_t sz@ = size of octet array
71 * Use: Stores an MP in an octet array, least significant octet
72 * first. High-end octets are silently discarded if there
73 * isn't enough space for them.
76 void mpx_storel(const mpw *v, const mpw *vl, void *pp, size_t sz)
79 octet *p = pp, *q = p + sz;
89 *p++ = U8(w | n << bits);
101 /* --- @mpx_loadl@ --- *
103 * Arguments: @mpw *v, *vl@ = base and limit of destination vector
104 * @const void *pp@ = pointer to octet array
105 * @size_t sz@ = size of octet array
109 * Use: Loads an MP in an octet array, least significant octet
110 * first. High-end octets are ignored if there isn't enough
114 void mpx_loadl(mpw *v, mpw *vl, const void *pp, size_t sz)
118 const octet *p = pp, *q = p + sz;
127 if (bits >= MPW_BITS) {
129 w = n >> (MPW_BITS - bits + 8);
139 /* --- @mpx_storeb@ --- *
141 * Arguments: @const mpw *v, *vl@ = base and limit of source vector
142 * @void *pp@ = pointer to octet array
143 * @size_t sz@ = size of octet array
147 * Use: Stores an MP in an octet array, most significant octet
148 * first. High-end octets are silently discarded if there
149 * isn't enough space for them.
152 void mpx_storeb(const mpw *v, const mpw *vl, void *pp, size_t sz)
155 octet *p = pp, *q = p + sz;
165 *--q = U8(w | n << bits);
167 bits += MPW_BITS - 8;
177 /* --- @mpx_loadb@ --- *
179 * Arguments: @mpw *v, *vl@ = base and limit of destination vector
180 * @const void *pp@ = pointer to octet array
181 * @size_t sz@ = size of octet array
185 * Use: Loads an MP in an octet array, most significant octet
186 * first. High-end octets are ignored if there isn't enough
190 void mpx_loadb(mpw *v, mpw *vl, const void *pp, size_t sz)
194 const octet *p = pp, *q = p + sz;
203 if (bits >= MPW_BITS) {
205 w = n >> (MPW_BITS - bits + 8);
215 /*----- Logical shifting --------------------------------------------------*/
217 /* --- @mpx_lsl@ --- *
219 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
220 * @const mpw *av, *avl@ = source vector base and limit
221 * @size_t n@ = number of bit positions to shift by
225 * Use: Performs a logical shift left operation on an integer.
228 void mpx_lsl(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, size_t n)
233 /* --- Trivial special case --- */
236 MPX_COPY(dv, dvl, av, avl);
238 /* --- Single bit shifting --- */
247 *dv++ = MPW((t << 1) | w);
248 w = t >> (MPW_BITS - 1);
257 /* --- Break out word and bit shifts for more sophisticated work --- */
262 /* --- Handle a shift by a multiple of the word size --- */
265 MPX_COPY(dv + nw, dvl, av, avl);
266 memset(dv, 0, MPWS(nw));
269 /* --- And finally the difficult case --- *
271 * This is a little convoluted, because I have to start from the end and
272 * work backwards to avoid overwriting the source, if they're both the same
278 size_t nr = MPW_BITS - nb;
279 size_t dvn = dvl - dv;
280 size_t avn = avl - av;
287 if (dvn > avn + nw) {
288 size_t off = avn + nw + 1;
289 MPX_ZERO(dv + off, dvl);
299 *--dvl = (t >> nr) | w;
310 /* --- @mpx_lsr@ --- *
312 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
313 * @const mpw *av, *avl@ = source vector base and limit
314 * @size_t n@ = number of bit positions to shift by
318 * Use: Performs a logical shift right operation on an integer.
321 void mpx_lsr(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, size_t n)
326 /* --- Trivial special case --- */
329 MPX_COPY(dv, dvl, av, avl);
331 /* --- Single bit shifting --- */
340 *dv++ = MPW((t << (MPW_BITS - 1)) | w);
350 /* --- Break out word and bit shifts for more sophisticated work --- */
355 /* --- Handle a shift by a multiple of the word size --- */
358 MPX_COPY(dv, dvl, av + nw, avl);
360 /* --- And finally the difficult case --- */
364 size_t nr = MPW_BITS - nb;
373 *dv++ = MPW((w >> nb) | (t << nr));
377 *dv++ = MPW(w >> nb);
385 /*----- Unsigned arithmetic -----------------------------------------------*/
387 /* --- @mpx_2c@ --- *
389 * Arguments: @mpw *dv, *dvl@ = destination vector
390 * @const mpw *v, *vl@ = source vector
394 * Use: Calculates the two's complement of @v@.
397 void mpx_2c(mpw *dv, mpw *dvl, const mpw *v, const mpw *vl)
400 while (dv < dvl && v < vl)
401 *dv++ = c = MPW(~*v++);
408 MPX_UADDN(dv, dvl, 1);
411 /* --- @mpx_ucmp@ --- *
413 * Arguments: @const mpw *av, *avl@ = first argument vector base and limit
414 * @const mpw *bv, *bvl@ = second argument vector base and limit
416 * Returns: Less than, equal to, or greater than zero depending on
417 * whether @a@ is less than, equal to or greater than @b@,
420 * Use: Performs an unsigned integer comparison.
423 int mpx_ucmp(const mpw *av, const mpw *avl, const mpw *bv, const mpw *bvl)
428 if (avl - av > bvl - bv)
430 else if (avl - av < bvl - bv)
432 else while (avl > av) {
433 mpw a = *--avl, b = *--bvl;
442 /* --- @mpx_uadd@ --- *
444 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
445 * @const mpw *av, *avl@ = first addend vector base and limit
446 * @const mpw *bv, *bvl@ = second addend vector base and limit
450 * Use: Performs unsigned integer addition. If the result overflows
451 * the destination vector, high-order bits are discarded. This
452 * means that two's complement addition happens more or less for
453 * free, although that's more a side-effect than anything else.
454 * The result vector may be equal to either or both source
455 * vectors, but may not otherwise overlap them.
458 void mpx_uadd(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl,
459 const mpw *bv, const mpw *bvl)
463 while (av < avl || bv < bvl) {
468 a = (av < avl) ? *av++ : 0;
469 b = (bv < bvl) ? *bv++ : 0;
470 x = (mpd)a + (mpd)b + c;
480 /* --- @mpx_usub@ --- *
482 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
483 * @const mpw *av, *avl@ = first argument vector base and limit
484 * @const mpw *bv, *bvl@ = second argument vector base and limit
488 * Use: Performs unsigned integer subtraction. If the result
489 * overflows the destination vector, high-order bits are
490 * discarded. This means that two's complement subtraction
491 * happens more or less for free, althuogh that's more a side-
492 * effect than anything else. The result vector may be equal to
493 * either or both source vectors, but may not otherwise overlap
497 void mpx_usub(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl,
498 const mpw *bv, const mpw *bvl)
502 while (av < avl || bv < bvl) {
507 a = (av < avl) ? *av++ : 0;
508 b = (bv < bvl) ? *bv++ : 0;
509 x = (mpd)a - (mpd)b - c;
522 /* --- @mpx_umul@ --- *
524 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
525 * @const mpw *av, *avl@ = multiplicand vector base and limit
526 * @const mpw *bv, *bvl@ = multiplier vector base and limit
530 * Use: Performs unsigned integer multiplication. If the result
531 * overflows the desination vector, high-order bits are
532 * discarded. The result vector may not overlap the argument
533 * vectors in any way.
536 void mpx_umul(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl,
537 const mpw *bv, const mpw *bvl)
539 /* --- This is probably worthwhile on a multiply --- */
544 /* --- Deal with a multiply by zero --- */
551 /* --- Do the initial multiply and initialize the accumulator --- */
553 MPX_UMULN(dv, dvl, av, avl, *bv++);
555 /* --- Do the remaining multiply/accumulates --- */
557 while (dv < dvl && bv < bvl) {
567 x = (mpd)*dvv + (mpd)m * (mpd)*avv++ + c;
571 MPX_UADDN(dvv, dvl, c);
576 /* --- @mpx_usqr@ --- *
578 * Arguments: @mpw *dv, *dvl@ = destination vector base and limit
579 * @const mpw *av, *av@ = source vector base and limit
583 * Use: Performs unsigned integer squaring. The result vector must
584 * not overlap the source vector in any way.
587 void mpx_usqr(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl)
591 /* --- Main loop --- */
599 /* --- Stop if I've run out of destination --- */
604 /* --- Work out the square at this point in the proceedings --- */
607 mpd x = (mpd)a * (mpd)a + *dvv;
609 c = MPW(x >> MPW_BITS);
612 /* --- Now fix up the rest of the vector upwards --- */
615 while (dvv < dvl && avv < avl) {
616 mpd x = (mpd)a * (mpd)*avv++;
617 mpd y = ((x << 1) & MPW_MAX) + c + *dvv;
618 c = (x >> (MPW_BITS - 1)) + (y >> MPW_BITS);
621 while (dvv < dvl && c) {
627 /* --- Get ready for the next round --- */
634 /* --- @mpx_udiv@ --- *
636 * Arguments: @mpw *qv, *qvl@ = quotient vector base and limit
637 * @mpw *rv, *rvl@ = dividend/remainder vector base and limit
638 * @const mpw *dv, *dvl@ = divisor vector base and limit
639 * @mpw *sv, *svl@ = scratch workspace
643 * Use: Performs unsigned integer division. If the result overflows
644 * the quotient vector, high-order bits are discarded. (Clearly
645 * the remainder vector can't overflow.) The various vectors
646 * may not overlap in any way. Yes, I know it's a bit odd
647 * requiring the dividend to be in the result position but it
648 * does make some sense really. The remainder must have
649 * headroom for at least two extra words. The scratch space
650 * must be at least one word larger than the divisor.
653 void mpx_udiv(mpw *qv, mpw *qvl, mpw *rv, mpw *rvl,
654 const mpw *dv, const mpw *dvl,
661 /* --- Initialize the quotient --- */
665 /* --- Perform some sanity checks --- */
668 assert(((void)"division by zero in mpx_udiv", dv < dvl));
670 /* --- Normalize the divisor --- *
672 * The algorithm requires that the divisor be at least two digits long.
673 * This is easy to fix.
680 for (b = MPW_BITS / 2; b; b >>= 1) {
681 if (d < (MPW_MAX >> b)) {
690 /* --- Normalize the dividend/remainder to match --- */
693 mpx_lsl(rv, rvl, rv, rvl, norm);
694 mpx_lsl(sv, svl, dv, dvl, norm);
704 /* --- Work out the relative scales --- */
707 size_t rvn = rvl - rv;
708 size_t dvn = dvl - dv;
710 /* --- If the divisor is clearly larger, notice this --- */
713 mpx_lsr(rv, rvl, rv, rvl, norm);
720 /* --- Calculate the most significant quotient digit --- *
722 * Because the divisor has its top bit set, this can only happen once. The
723 * pointer arithmetic is a little contorted, to make sure that the
724 * behaviour is defined.
727 if (MPX_UCMP(rv + scale, rvl, >=, dv, dvl)) {
728 mpx_usub(rv + scale, rvl, rv + scale, rvl, dv, dvl);
729 if (qvl - qv > scale)
733 /* --- Now for the main loop --- */
742 /* --- Get an estimate for the next quotient digit --- */
749 rh = ((mpd)r << MPW_BITS) | rr;
755 /* --- Refine the estimate --- */
759 mpd yl = (mpd)dd * q;
762 yh += yl >> MPW_BITS;
766 while (yh > rh || (yh == rh && yl > rrr)) {
777 /* --- Remove a chunk from the dividend --- */
784 /* --- Calculate the size of the chunk --- *
786 * This does the whole job of calculating @r >> scale - qd@.
789 for (svv = rv + scale, dvv = dv;
790 dvv < dvl && svv < rvl;
792 mpd x = (mpd)*dvv * (mpd)q + mc;
794 x = (mpd)*svv - MPW(x) - sc;
803 mpd x = (mpd)*svv - mc - sc;
813 /* --- Fix if the quotient was too large --- *
815 * This doesn't seem to happen very often.
818 if (rvl[-1] > MPW_MAX / 2) {
819 mpx_uadd(rv + scale, rvl, rv + scale, rvl, dv, dvl);
824 /* --- Done for another iteration --- */
826 if (qvl - qv > scale)
833 /* --- Now fiddle with unnormalizing and things --- */
835 mpx_lsr(rv, rvl, rv, rvl, norm);
838 /*----- That's all, folks -------------------------------------------------*/