version 1.1, 1999/12/03 07:39:07 |
version 1.45, 2007/09/16 09:08:25 |
|
|
/* $OpenXM: OpenXM/src/asir99/builtin/dp-supp.c,v 1.2 1999/11/18 05:42:01 noro Exp $ */ |
/* |
|
* Copyright (c) 1994-2000 FUJITSU LABORATORIES LIMITED |
|
* All rights reserved. |
|
* |
|
* FUJITSU LABORATORIES LIMITED ("FLL") hereby grants you a limited, |
|
* non-exclusive and royalty-free license to use, copy, modify and |
|
* redistribute, solely for non-commercial and non-profit purposes, the |
|
* computer program, "Risa/Asir" ("SOFTWARE"), subject to the terms and |
|
* conditions of this Agreement. For the avoidance of doubt, you acquire |
|
* only a limited right to use the SOFTWARE hereunder, and FLL or any |
|
* third party developer retains all rights, including but not limited to |
|
* copyrights, in and to the SOFTWARE. |
|
* |
|
* (1) FLL does not grant you a license in any way for commercial |
|
* purposes. You may use the SOFTWARE only for non-commercial and |
|
* non-profit purposes only, such as academic, research and internal |
|
* business use. |
|
* (2) The SOFTWARE is protected by the Copyright Law of Japan and |
|
* international copyright treaties. If you make copies of the SOFTWARE, |
|
* with or without modification, as permitted hereunder, you shall affix |
|
* to all such copies of the SOFTWARE the above copyright notice. |
|
* (3) An explicit reference to this SOFTWARE and its copyright owner |
|
* shall be made on your publication or presentation in any form of the |
|
* results obtained by use of the SOFTWARE. |
|
* (4) In the event that you modify the SOFTWARE, you shall notify FLL by |
|
* e-mail at risa-admin@sec.flab.fujitsu.co.jp of the detailed specification |
|
* for such modification or the source code of the modified part of the |
|
* SOFTWARE. |
|
* |
|
* THE SOFTWARE IS PROVIDED AS IS WITHOUT ANY WARRANTY OF ANY KIND. FLL |
|
* MAKES ABSOLUTELY NO WARRANTIES, EXPRESSED, IMPLIED OR STATUTORY, AND |
|
* EXPRESSLY DISCLAIMS ANY IMPLIED WARRANTY OF MERCHANTABILITY, FITNESS |
|
* FOR A PARTICULAR PURPOSE OR NONINFRINGEMENT OF THIRD PARTIES' |
|
* RIGHTS. NO FLL DEALER, AGENT, EMPLOYEES IS AUTHORIZED TO MAKE ANY |
|
* MODIFICATIONS, EXTENSIONS, OR ADDITIONS TO THIS WARRANTY. |
|
* UNDER NO CIRCUMSTANCES AND UNDER NO LEGAL THEORY, TORT, CONTRACT, |
|
* OR OTHERWISE, SHALL FLL BE LIABLE TO YOU OR ANY OTHER PERSON FOR ANY |
|
* DIRECT, INDIRECT, SPECIAL, INCIDENTAL, PUNITIVE OR CONSEQUENTIAL |
|
* DAMAGES OF ANY CHARACTER, INCLUDING, WITHOUT LIMITATION, DAMAGES |
|
* ARISING OUT OF OR RELATING TO THE SOFTWARE OR THIS AGREEMENT, DAMAGES |
|
* FOR LOSS OF GOODWILL, WORK STOPPAGE, OR LOSS OF DATA, OR FOR ANY |
|
* DAMAGES, EVEN IF FLL SHALL HAVE BEEN INFORMED OF THE POSSIBILITY OF |
|
* SUCH DAMAGES, OR FOR ANY CLAIM BY ANY OTHER PARTY. EVEN IF A PART |
|
* OF THE SOFTWARE HAS BEEN DEVELOPED BY A THIRD PARTY, THE THIRD PARTY |
|
* DEVELOPER SHALL HAVE NO LIABILITY IN CONNECTION WITH THE USE, |
|
* PERFORMANCE OR NON-PERFORMANCE OF THE SOFTWARE. |
|
* |
|
* $OpenXM: OpenXM_contrib2/asir2000/builtin/dp-supp.c,v 1.44 2007/09/15 10:17:08 noro Exp $ |
|
*/ |
#include "ca.h" |
#include "ca.h" |
#include "base.h" |
#include "base.h" |
|
#include "inline.h" |
#include "parse.h" |
#include "parse.h" |
#include "ox.h" |
#include "ox.h" |
|
|
|
#define HMAG(p) (p_mag(BDY(p)->c)) |
|
|
extern int (*cmpdl)(); |
extern int (*cmpdl)(); |
double pz_t_e,pz_t_d,pz_t_d1,pz_t_c; |
extern double pz_t_e,pz_t_d,pz_t_d1,pz_t_c; |
|
extern int dp_nelim,dp_fcoeffs; |
|
extern int NoGCD; |
|
extern int GenTrace; |
|
extern NODE TraceList; |
|
|
void Pdp_mbase(),Pdp_lnf_mod(),Pdp_nf_tab_mod(),Pdp_mdtod(); |
int show_orderspec; |
void Pdp_vtoe(), Pdp_etov(), Pdp_dtov(), Pdp_idiv(), Pdp_sep(); |
|
void Pdp_cont(), Pdp_igcdv_hist(), Pdp_idivv_hist(); |
|
|
|
struct ftab dp_supp_tab[] = { |
void print_composite_order_spec(struct order_spec *spec); |
{"dp_mbase",Pdp_mbase,1}, |
|
{"dp_lnf_mod",Pdp_lnf_mod,3}, |
|
{"dp_nf_tab_mod",Pdp_nf_tab_mod,3}, |
|
{"dp_etov",Pdp_etov,1}, |
|
{"dp_vtoe",Pdp_vtoe,1}, |
|
{"dp_dtov",Pdp_dtov,1}, |
|
{"dp_idiv",Pdp_idiv,2}, |
|
{"dp_cont",Pdp_cont,1}, |
|
{"dp_sep",Pdp_sep,2}, |
|
{"dp_mdtod",Pdp_mdtod,1}, |
|
{"dp_igcdv_hist",Pdp_igcdv_hist,1}, |
|
{"dp_idivv_hist",Pdp_idivv_hist,1}, |
|
{0,0,0} |
|
}; |
|
|
|
void Pdp_mdtod(arg,rp) |
/* |
NODE arg; |
* content reduction |
DP *rp; |
* |
|
*/ |
|
|
|
void dp_ptozp(DP p,DP *rp) |
{ |
{ |
MP m,mr,mr0; |
MP m,mr,mr0; |
DP p; |
int i,n; |
|
Q *w; |
|
Q dvr; |
P t; |
P t; |
|
|
p = (DP)ARG0(arg); |
|
if ( !p ) |
if ( !p ) |
*rp = 0; |
*rp = 0; |
else { |
else { |
|
for ( m =BDY(p), n = 0; m; m = NEXT(m), n++ ); |
|
w = (Q *)ALLOCA(n*sizeof(Q)); |
|
for ( m =BDY(p), i = 0; i < n; m = NEXT(m), i++ ) |
|
if ( NUM(m->c) ) |
|
w[i] = (Q)m->c; |
|
else |
|
ptozp(m->c,1,&w[i],&t); |
|
sortbynm(w,n); |
|
qltozl(w,n,&dvr); |
for ( mr0 = 0, m = BDY(p); m; m = NEXT(m) ) { |
for ( mr0 = 0, m = BDY(p); m; m = NEXT(m) ) { |
mptop(m->c,&t); NEXTMP(mr0,mr); mr->c = t; mr->dl = m->dl; |
NEXTMP(mr0,mr); divsp(CO,m->c,(P)dvr,&mr->c); mr->dl = m->dl; |
} |
} |
NEXT(mr) = 0; MKDP(p->nv,mr0,*rp); (*rp)->sugar = p->sugar; |
NEXT(mr) = 0; MKDP(p->nv,mr0,*rp); (*rp)->sugar = p->sugar; |
} |
} |
} |
} |
|
|
void Pdp_sep(arg,rp) |
void dp_ptozp2(DP p0,DP p1,DP *hp,DP *rp) |
NODE arg; |
|
VECT *rp; |
|
{ |
{ |
DP p,r; |
DP t,s,h,r; |
MP m,t; |
MP m,mr,mr0,m0; |
MP *w0,*w; |
|
int i,n,d,nv,sugar; |
|
VECT v; |
|
pointer *pv; |
|
|
|
p = (DP)ARG0(arg); m = BDY(p); |
addd(CO,p0,p1,&t); dp_ptozp(t,&s); |
d = QTOS((Q)ARG1(arg)); |
if ( !p0 ) { |
for ( t = m, n = 0; t; t = NEXT(t), n++ ); |
h = 0; r = s; |
if ( d > n ) |
} else if ( !p1 ) { |
d = n; |
h = s; r = 0; |
MKVECT(v,d); *rp = v; |
} else { |
pv = BDY(v); nv = p->nv; sugar = p->sugar; |
for ( mr0 = 0, m = BDY(s), m0 = BDY(p0); m0; |
w0 = (MP *)MALLOC(d*sizeof(MP)); bzero(w0,d*sizeof(MP)); |
m = NEXT(m), m0 = NEXT(m0) ) { |
w = (MP *)MALLOC(d*sizeof(MP)); bzero(w,d*sizeof(MP)); |
NEXTMP(mr0,mr); mr->c = m->c; mr->dl = m->dl; |
for ( t = BDY(p), i = 0; t; t = NEXT(t), i++, i %= d ) { |
} |
NEXTMP(w0[i],w[i]); w[i]->c = t->c; w[i]->dl = t->dl; |
NEXT(mr) = 0; MKDP(p0->nv,mr0,h); MKDP(p0->nv,m,r); |
} |
} |
for ( i = 0; i < d; i++ ) { |
if ( h ) |
NEXT(w[i]) = 0; MKDP(nv,w0[i],r); r->sugar = sugar; |
h->sugar = p0->sugar; |
pv[i] = (pointer)r; |
if ( r ) |
} |
r->sugar = p1->sugar; |
|
*hp = h; *rp = r; |
} |
} |
|
|
void Pdp_idiv(arg,rp) |
void dp_ptozp3(DP p,Q *dvr,DP *rp) |
NODE arg; |
|
DP *rp; |
|
{ |
{ |
dp_idiv((DP)ARG0(arg),(Q)ARG1(arg),rp); |
MP m,mr,mr0; |
} |
int i,n; |
|
Q *w; |
|
P t; |
|
|
void Pdp_cont(arg,rp) |
if ( !p ) { |
NODE arg; |
*rp = 0; *dvr = 0; |
Q *rp; |
}else { |
{ |
for ( m =BDY(p), n = 0; m; m = NEXT(m), n++ ); |
dp_cont((DP)ARG0(arg),rp); |
w = (Q *)ALLOCA(n*sizeof(Q)); |
} |
for ( m =BDY(p), i = 0; i < n; m = NEXT(m), i++ ) |
|
if ( NUM(m->c) ) |
void Pdp_dtov(arg,rp) |
w[i] = (Q)m->c; |
NODE arg; |
else |
VECT *rp; |
ptozp(m->c,1,&w[i],&t); |
{ |
sortbynm(w,n); |
dp_dtov((DP)ARG0(arg),rp); |
qltozl(w,n,dvr); |
} |
for ( mr0 = 0, m = BDY(p); m; m = NEXT(m) ) { |
|
NEXTMP(mr0,mr); divsp(CO,m->c,(P)(*dvr),&mr->c); mr->dl = m->dl; |
void Pdp_mbase(arg,rp) |
} |
NODE arg; |
NEXT(mr) = 0; MKDP(p->nv,mr0,*rp); (*rp)->sugar = p->sugar; |
LIST *rp; |
|
{ |
|
NODE mb; |
|
|
|
asir_assert(ARG0(arg),O_LIST,"dp_mbase"); |
|
dp_mbase(BDY((LIST)ARG0(arg)),&mb); |
|
MKLIST(*rp,mb); |
|
} |
|
|
|
void Pdp_etov(arg,rp) |
|
NODE arg; |
|
VECT *rp; |
|
{ |
|
DP dp; |
|
int n,i; |
|
int *d; |
|
VECT v; |
|
Q t; |
|
|
|
dp = (DP)ARG0(arg); |
|
asir_assert(dp,O_DP,"dp_etov"); |
|
n = dp->nv; d = BDY(dp)->dl->d; |
|
MKVECT(v,n); |
|
for ( i = 0; i < n; i++ ) { |
|
STOQ(d[i],t); v->body[i] = (pointer)t; |
|
} |
} |
*rp = v; |
|
} |
} |
|
|
void Pdp_vtoe(arg,rp) |
void dp_idiv(DP p,Q c,DP *rp) |
NODE arg; |
|
DP *rp; |
|
{ |
{ |
DP dp; |
|
DL dl; |
|
MP m; |
|
int n,i,td; |
|
int *d; |
|
VECT v; |
|
|
|
v = (VECT)ARG0(arg); |
|
asir_assert(v,O_VECT,"dp_vtoe"); |
|
n = v->len; |
|
NEWDL(dl,n); d = dl->d; |
|
for ( i = 0, td = 0; i < n; i++ ) { |
|
d[i] = QTOS((Q)(v->body[i])); td += d[i]; |
|
} |
|
dl->td = td; |
|
NEWMP(m); m->dl = dl; m->c = (P)ONE; NEXT(m) = 0; |
|
MKDP(n,m,dp); dp->sugar = td; |
|
*rp = dp; |
|
} |
|
|
|
void Pdp_lnf_mod(arg,rp) |
|
NODE arg; |
|
LIST *rp; |
|
{ |
|
DP r1,r2; |
|
NODE b,g,n; |
|
int mod; |
|
|
|
asir_assert(ARG0(arg),O_LIST,"dp_lnf_mod"); |
|
asir_assert(ARG1(arg),O_LIST,"dp_lnf_mod"); |
|
asir_assert(ARG2(arg),O_N,"dp_lnf_mod"); |
|
b = BDY((LIST)ARG0(arg)); g = BDY((LIST)ARG1(arg)); |
|
mod = QTOS((Q)ARG2(arg)); |
|
dp_lnf_mod((DP)BDY(b),(DP)BDY(NEXT(b)),g,mod,&r1,&r2); |
|
NEWNODE(n); BDY(n) = (pointer)r1; |
|
NEWNODE(NEXT(n)); BDY(NEXT(n)) = (pointer)r2; |
|
NEXT(NEXT(n)) = 0; MKLIST(*rp,n); |
|
} |
|
|
|
void Pdp_nf_tab_mod(arg,rp) |
|
NODE arg; |
|
DP *rp; |
|
{ |
|
asir_assert(ARG0(arg),O_DP,"dp_nf_tab_mod"); |
|
asir_assert(ARG1(arg),O_VECT,"dp_nf_tab_mod"); |
|
asir_assert(ARG2(arg),O_N,"dp_nf_tab_mod"); |
|
dp_nf_tab_mod((DP)ARG0(arg),(LIST *)BDY((VECT)ARG1(arg)), |
|
QTOS((Q)ARG2(arg)),rp); |
|
} |
|
|
|
void Pdp_igcdv_hist(arg,rp) |
|
NODE arg; |
|
Q *rp; |
|
{ |
|
dp_igcdv_hist((DP)ARG0(arg),rp); |
|
} |
|
|
|
void Pdp_idivv_hist(arg,rp) |
|
NODE arg; |
|
DP *rp; |
|
{ |
|
dp_idivv_hist((Q)ARG0(arg),rp); |
|
} |
|
|
|
void dp_idiv(p,c,rp) |
|
DP p; |
|
Q c; |
|
DP *rp; |
|
{ |
|
Q t; |
Q t; |
N nm,q; |
N nm,q; |
int sgn,s; |
int sgn,s; |
|
|
} |
} |
} |
} |
|
|
void dp_cont(p,rp) |
void dp_mbase(NODE hlist,NODE *mbase) |
DP p; |
|
Q *rp; |
|
{ |
{ |
VECT v; |
|
|
|
dp_dtov(p,&v); igcdv(v,rp); |
|
} |
|
|
|
void dp_dtov(dp,rp) |
|
DP dp; |
|
VECT *rp; |
|
{ |
|
MP m,t; |
|
int i,n; |
|
VECT v; |
|
pointer *p; |
|
|
|
m = BDY(dp); |
|
for ( t = m, n = 0; t; t = NEXT(t), n++ ); |
|
MKVECT(v,n); |
|
for ( i = 0, p = BDY(v), t = m; i < n; t = NEXT(t), i++ ) |
|
p[i] = (pointer)(t->c); |
|
*rp = v; |
|
} |
|
|
|
void dp_mbase(hlist,mbase) |
|
NODE hlist; |
|
NODE *mbase; |
|
{ |
|
DL *dl; |
DL *dl; |
DL d; |
DL d; |
int i,j,n,nvar,td; |
int i,j,n,nvar,td; |
|
|
while ( 1 ) { |
while ( 1 ) { |
insert_to_node(d,mbase,nvar); |
insert_to_node(d,mbase,nvar); |
for ( i = nvar-1; i >= 0; ) { |
for ( i = nvar-1; i >= 0; ) { |
d->d[i]++; d->td++; |
d->d[i]++; |
|
d->td += MUL_WEIGHT(1,i); |
for ( j = 0; j < n; j++ ) { |
for ( j = 0; j < n; j++ ) { |
if ( _dl_redble(dl[j],d,nvar) ) |
if ( _dl_redble(dl[j],d,nvar) ) |
break; |
break; |
|
|
for ( j = nvar-1; j >= i; j-- ) |
for ( j = nvar-1; j >= i; j-- ) |
d->d[j] = 0; |
d->d[j] = 0; |
for ( j = 0, td = 0; j < i; j++ ) |
for ( j = 0, td = 0; j < i; j++ ) |
td += d->d[j]; |
td += MUL_WEIGHT(d->d[j],j); |
d->td = td; |
d->td = td; |
i--; |
i--; |
} else |
} else |
|
|
} |
} |
} |
} |
|
|
int _dl_redble(d1,d2,nvar) |
int _dl_redble(DL d1,DL d2,int nvar) |
DL d1,d2; |
|
int nvar; |
|
{ |
{ |
int i; |
int i; |
|
|
|
|
return 1; |
return 1; |
} |
} |
|
|
void insert_to_node(d,n,nvar) |
void insert_to_node(DL d,NODE *n,int nvar) |
DL d; |
|
NODE *n; |
|
int nvar; |
|
{ |
{ |
DL d1; |
DL d1; |
MP m; |
MP m; |
|
|
} |
} |
} |
} |
|
|
void dp_lnf_mod(p1,p2,g,mod,r1p,r2p) |
void dp_vtod(Q *c,DP p,DP *rp) |
DP p1,p2; |
|
NODE g; |
|
int mod; |
|
DP *r1p,*r2p; |
|
{ |
{ |
DP r1,r2,b1,b2,t,s; |
|
P c; |
|
MQ c1,c2; |
|
NODE l,b; |
|
int n; |
|
|
|
if ( !p1 ) { |
|
*r1p = p1; *r2p = p2; return; |
|
} |
|
n = p1->nv; |
|
for ( l = g, r1 = p1, r2 = p2; l; l = NEXT(l) ) { |
|
if ( !r1 ) { |
|
*r1p = r1; *r2p = r2; return; |
|
} |
|
b = BDY((LIST)BDY(l)); b1 = (DP)BDY(b); |
|
if ( dl_equal(n,BDY(r1)->dl,BDY(b1)->dl) ) { |
|
b2 = (DP)BDY(NEXT(b)); |
|
invmq(mod,(MQ)BDY(b1)->c,&c1); |
|
mulmq(mod,c1,(MQ)BDY(r1)->c,&c2); chsgnmp(mod,(P)c2,&c); |
|
mulmdc(CO,mod,b1,c,&t); addmd(CO,mod,r1,t,&s); r1 = s; |
|
mulmdc(CO,mod,b2,c,&t); addmd(CO,mod,r2,t,&s); r2 = s; |
|
} |
|
} |
|
*r1p = r1; *r2p = r2; |
|
} |
|
|
|
void dp_nf_tab_mod(p,tab,mod,rp) |
|
DP p; |
|
LIST *tab; |
|
int mod; |
|
DP *rp; |
|
{ |
|
DP s,t,u; |
|
MP m; |
|
DL h; |
|
int i,n; |
|
|
|
if ( !p ) { |
|
*rp = p; return; |
|
} |
|
n = p->nv; |
|
for ( s = 0, i = 0, m = BDY(p); m; m = NEXT(m) ) { |
|
h = m->dl; |
|
while ( !dl_equal(n,h,BDY((DP)BDY(BDY(tab[i])))->dl ) ) |
|
i++; |
|
mulmdc(CO,mod,(DP)BDY(NEXT(BDY(tab[i]))),m->c,&t); |
|
addmd(CO,mod,s,t,&u); s = u; |
|
} |
|
*rp = s; |
|
} |
|
|
|
void dp_ptozp(p,rp) |
|
DP p,*rp; |
|
{ |
|
MP m,mr,mr0; |
|
int i,n; |
|
Q *w; |
|
Q dvr; |
|
P t; |
|
|
|
if ( !p ) |
|
*rp = 0; |
|
else { |
|
for ( m =BDY(p), n = 0; m; m = NEXT(m), n++ ); |
|
w = (Q *)ALLOCA(n*sizeof(Q)); |
|
for ( m =BDY(p), i = 0; i < n; m = NEXT(m), i++ ) |
|
if ( NUM(m->c) ) |
|
w[i] = (Q)m->c; |
|
else |
|
ptozp(m->c,1,&w[i],&t); |
|
sortbynm(w,n); |
|
qltozl(w,n,&dvr); |
|
for ( mr0 = 0, m = BDY(p); m; m = NEXT(m) ) { |
|
NEXTMP(mr0,mr); divsp(CO,m->c,(P)dvr,&mr->c); mr->dl = m->dl; |
|
} |
|
NEXT(mr) = 0; MKDP(p->nv,mr0,*rp); (*rp)->sugar = p->sugar; |
|
} |
|
} |
|
|
|
void dp_ptozp2(p0,p1,hp,rp) |
|
DP p0,p1; |
|
DP *hp,*rp; |
|
{ |
|
DP t,s,h,r; |
|
MP m,mr,mr0,m0; |
|
|
|
addd(CO,p0,p1,&t); dp_ptozp(t,&s); |
|
if ( !p0 ) { |
|
h = 0; r = s; |
|
} else if ( !p1 ) { |
|
h = s; r = 0; |
|
} else { |
|
for ( mr0 = 0, m = BDY(s), m0 = BDY(p0); m0; |
|
m = NEXT(m), m0 = NEXT(m0) ) { |
|
NEXTMP(mr0,mr); mr->c = m->c; mr->dl = m->dl; |
|
} |
|
NEXT(mr) = 0; MKDP(p0->nv,mr0,h); MKDP(p0->nv,m,r); |
|
} |
|
if ( h ) |
|
h->sugar = p0->sugar; |
|
if ( r ) |
|
r->sugar = p1->sugar; |
|
*hp = h; *rp = r; |
|
} |
|
|
|
struct igcdv_hist { |
|
int len; |
|
DP dp; |
|
Q d,d1; |
|
Q *q,*q1; |
|
}; |
|
|
|
static struct igcdv_hist current_igcdv_hist; |
|
|
|
void dp_igcdv_hist(p,rp) |
|
DP p; |
|
Q *rp; |
|
{ |
|
VECT c; |
|
Q *a,*q,*r,*q1,*r1; |
|
int i,n,nz; |
|
Q d,d1; |
|
N qn,rn,gn; |
|
struct oVECT v; |
|
|
|
dp_dtov(p,&c); a = (Q *)c->body; |
|
n = c->len; |
|
q = (Q *)MALLOC(n*sizeof(Q)); r = (Q *)MALLOC(n*sizeof(Q)); |
|
igcdv_estimate(c,&d); |
|
for ( i = 0, nz = 0; i < n; i++ ) { |
|
divn(NM(a[i]),NM(d),&qn,&rn); |
|
NTOQ(qn,SGN(a[i]),q[i]); NTOQ(rn,SGN(a[i]),r[i]); |
|
if ( r[i] ) |
|
nz = 1; |
|
} |
|
current_igcdv_hist.len = n; |
|
current_igcdv_hist.dp = p; |
|
current_igcdv_hist.d = d; |
|
current_igcdv_hist.q = q; |
|
if ( nz ) { |
|
v.id = O_VECT; v.len = n; v.body = (pointer *)r; |
|
igcdv(&v,&d1); |
|
q1 = (Q *)MALLOC(n*sizeof(Q)); r1 = (Q *)MALLOC(n*sizeof(Q)); |
|
for ( i = 0; i < n; i++ ) { |
|
if ( r[i] ) |
|
divn(NM(r[i]),NM(d1),&qn,&rn); |
|
else { |
|
qn = rn = 0; |
|
} |
|
NTOQ(qn,SGN(r[i]),q1[i]); NTOQ(rn,SGN(r[i]),r1[i]); |
|
} |
|
gcdn(NM(d),NM(d1),&gn); NTOQ(gn,1,*rp); |
|
} else { |
|
d1 = 0; q1 = 0; *rp = d; |
|
} |
|
current_igcdv_hist.d1 = d1; |
|
current_igcdv_hist.q1 = q1; |
|
} |
|
|
|
void dp_idivv_hist(g,rp) |
|
Q g; |
|
DP *rp; |
|
{ |
|
int i,n; |
|
Q *q,*q1,*s; |
|
Q t,u,m; |
|
N qn; |
|
|
|
n = current_igcdv_hist.len; |
|
q = current_igcdv_hist.q; q1 = current_igcdv_hist.q1; |
|
divsn(NM(current_igcdv_hist.d),NM(g),&qn); NTOQ(qn,1,m); |
|
s = (Q *)MALLOC(n*sizeof(Q)); |
|
for ( i = 0; i < n; i++ ) |
|
mulq(m,q[i],&s[i]); |
|
if ( q1 ) { |
|
divsn(NM(current_igcdv_hist.d1),NM(g),&qn); NTOQ(qn,1,m); |
|
for ( i = 0; i < n; i++ ) { |
|
mulq(m,q1[i],&t); addq(t,s[i],&u); s[i] = u; |
|
} |
|
} |
|
dp_vtod(s,current_igcdv_hist.dp,rp); |
|
} |
|
|
|
void dp_vtod(c,p,rp) |
|
Q *c; |
|
DP p; |
|
DP *rp; |
|
{ |
|
MP mr0,m,mr; |
MP mr0,m,mr; |
int i; |
int i; |
|
|
|
|
} |
} |
} |
} |
|
|
#if INET |
extern int mpi_mag; |
void dp_ptozp_d_old(dist,ndist,p,rp) |
extern int PCoeffs; |
NODE dist; |
|
int ndist; |
|
DP p,*rp; |
|
{ |
|
DP d,s,u,su; |
|
MP m,mr,mr0; |
|
MP *w0,*w; |
|
Q *g; |
|
int i,n,nv,nsep; |
|
P t; |
|
NODE tn,n0,n1,n2; |
|
struct oVECT v; |
|
int id,sindex; |
|
Obj dmy; |
|
Q gcd; |
|
STRING cont,div; |
|
|
|
if ( !p ) |
void dp_ptozp_d(DP p,DP *rp) |
*rp = 0; |
|
else { |
|
for ( m = BDY(p), n = 0; m; m = NEXT(m), n++ ); |
|
nsep = ndist + 1; |
|
if ( n <= nsep ) { |
|
dp_ptozp(p,rp); return; |
|
} |
|
nv = p->nv; |
|
w0 = (MP *)MALLOC(nsep*sizeof(MP)); bzero(w0,nsep*sizeof(MP)); |
|
w = (MP *)MALLOC(nsep*sizeof(MP)); bzero(w,nsep*sizeof(MP)); |
|
g = (Q *)MALLOC(nsep*sizeof(Q)); bzero(g,nsep*sizeof(Q)); |
|
for ( m = BDY(p), i = 0; m; m = NEXT(m), i++, i %= nsep ) { |
|
NEXTMP(w0[i],w[i]); w[i]->c = m->c; w[i]->dl = m->dl; |
|
} |
|
MKSTR(cont,"reg_dp_cont"); MKSTR(div,"reg_dp_idiv"); |
|
|
|
for ( i = 0, tn = dist; i < ndist; i++, tn = NEXT(tn) ) { |
|
NEXT(w[i]) = 0; MKDP(nv,w0[i],d); d->sugar = p->sugar; |
|
MKNODE(n2,d,0); MKNODE(n1,cont,n2); MKNODE(n0,BDY(tn),n1); |
|
Pox_rpc(n0,&dmy); |
|
} |
|
NEXT(w[i]) = 0; MKDP(nv,w0[i],d); d->sugar = p->sugar; |
|
dp_cont(d,&g[i]); |
|
|
|
for ( i = 0, tn = dist; i < ndist; i++, tn = NEXT(tn) ) |
|
Pox_pop_local(tn,&g[i]); |
|
v.id = O_VECT; v.len = nsep; v.body = (pointer *)g; igcdv(&v,&gcd); |
|
|
|
for ( i = 0, tn = dist; i < ndist; i++, tn = NEXT(tn) ) { |
|
MKNODE(n2,gcd,0); MKNODE(n1,div,n2); MKNODE(n0,BDY(tn),n1); |
|
Pox_rpc(n0,&dmy); |
|
} |
|
dp_idiv(d,gcd,&s); |
|
|
|
for ( i = 0, tn = dist; i < ndist; i++, tn = NEXT(tn) ) { |
|
Pox_pop_local(tn,&u); |
|
addd(CO,s,u,&su); s = su; |
|
} |
|
*rp = s; |
|
} |
|
} |
|
|
|
void dp_ptozp_d(dist,ndist,p,rp) |
|
NODE dist; |
|
int ndist; |
|
DP p,*rp; |
|
{ |
{ |
int i,j,k,l,n,nsep; |
int i,j,k,l,n,nsep; |
MP m; |
MP m; |
|
|
VECT c,cs; |
VECT c,cs; |
VECT qi,ri; |
VECT qi,ri; |
LIST *qr; |
LIST *qr; |
int s,id; |
|
Obj dmy; |
Obj dmy; |
Q d0,d1,gcd,a,u,u1; |
Q d0,d1,gcd,a,u,u1; |
Q *q,*r; |
Q *q,*r; |
|
|
N qn,gn; |
N qn,gn; |
double get_rtime(); |
double get_rtime(); |
int blen; |
int blen; |
|
NODE dist; |
|
int ndist; |
double t0; |
double t0; |
double t_e,t_d,t_d1,t_c; |
double t_e,t_d,t_d1,t_c; |
|
extern int DP_NFStat; |
|
extern LIST Dist; |
|
void Pox_rpc(); |
|
void Pox_pop_local(); |
|
|
if ( !p ) |
if ( !p ) |
*rp = 0; |
*rp = 0; |
else { |
else { |
|
if ( PCoeffs ) { |
|
dp_ptozp(p,rp); return; |
|
} |
|
if ( !Dist || p_mag(BDY(p)->c) <= mpi_mag ) { |
|
dist = 0; ndist = 0; |
|
if ( DP_NFStat ) fprintf(asir_out,"L"); |
|
} else { |
|
dist = BDY(Dist); ndist = length(dist); |
|
if ( DP_NFStat ) fprintf(asir_out,"D"); |
|
} |
for ( m = BDY(p), n = 0; m; m = NEXT(m), n++ ); |
for ( m = BDY(p), n = 0; m; m = NEXT(m), n++ ); |
nsep = ndist + 1; |
nsep = ndist + 1; |
if ( n <= nsep ) { |
if ( n <= nsep ) { |
|
|
} |
} |
} |
} |
|
|
void dp_ptozp2_d(dist,ndist,p0,p1,hp,rp) |
void dp_ptozp2_d(DP p0,DP p1,DP *hp,DP *rp) |
NODE dist; |
|
int ndist; |
|
DP p0,p1; |
|
DP *hp,*rp; |
|
{ |
{ |
DP t,s,h,r; |
DP t,s,h,r; |
MP m,mr,mr0,m0; |
MP m,mr,mr0,m0; |
|
|
addd(CO,p0,p1,&t); dp_ptozp_d(dist,ndist,t,&s); |
addd(CO,p0,p1,&t); dp_ptozp_d(t,&s); |
if ( !p0 ) { |
if ( !p0 ) { |
h = 0; r = s; |
h = 0; r = s; |
} else if ( !p1 ) { |
} else if ( !p1 ) { |
|
|
r->sugar = p1->sugar; |
r->sugar = p1->sugar; |
*hp = h; *rp = r; |
*hp = h; *rp = r; |
} |
} |
#endif |
|
|
|
|
int have_sf_coef(P p) |
|
{ |
|
DCP dc; |
|
|
|
if ( !p ) |
|
return 0; |
|
else if ( NUM(p) ) |
|
return NID((Num)p) == N_GFS ? 1 : 0; |
|
else { |
|
for ( dc = DC(p); dc; dc = NEXT(dc) ) |
|
if ( have_sf_coef(COEF(dc)) ) |
|
return 1; |
|
return 0; |
|
} |
|
} |
|
|
|
void head_coef(P p,Num *c) |
|
{ |
|
if ( !p ) |
|
*c = 0; |
|
else if ( NUM(p) ) |
|
*c = (Num)p; |
|
else |
|
head_coef(COEF(DC(p)),c); |
|
} |
|
|
|
void dp_monic_sf(DP p,DP *rp) |
|
{ |
|
Num c; |
|
|
|
if ( !p ) |
|
*rp = 0; |
|
else { |
|
head_coef(BDY(p)->c,&c); |
|
divsdc(CO,p,(P)c,rp); |
|
} |
|
} |
|
|
|
void dp_prim(DP p,DP *rp) |
|
{ |
|
P t,g; |
|
DP p1; |
|
MP m,mr,mr0; |
|
int i,n; |
|
P *w; |
|
Q *c; |
|
Q dvr; |
|
|
|
if ( !p ) |
|
*rp = 0; |
|
else if ( dp_fcoeffs == N_GFS ) { |
|
for ( m = BDY(p); m; m = NEXT(m) ) |
|
if ( OID(m->c) == O_N ) { |
|
/* GCD of coeffs = 1 */ |
|
dp_monic_sf(p,rp); |
|
return; |
|
} else break; |
|
/* compute GCD over the finite fieid */ |
|
for ( m = BDY(p), n = 0; m; m = NEXT(m), n++ ); |
|
w = (P *)ALLOCA(n*sizeof(P)); |
|
for ( m = BDY(p), i = 0; i < n; m = NEXT(m), i++ ) |
|
w[i] = m->c; |
|
gcdsf(CO,w,n,&g); |
|
if ( NUM(g) ) |
|
dp_monic_sf(p,rp); |
|
else { |
|
for ( mr0 = 0, m = BDY(p); m; m = NEXT(m) ) { |
|
NEXTMP(mr0,mr); divsp(CO,m->c,g,&mr->c); mr->dl = m->dl; |
|
} |
|
NEXT(mr) = 0; MKDP(p->nv,mr0,p1); p1->sugar = p->sugar; |
|
dp_monic_sf(p1,rp); |
|
} |
|
return; |
|
} else if ( dp_fcoeffs ) |
|
*rp = p; |
|
else if ( NoGCD ) |
|
dp_ptozp(p,rp); |
|
else { |
|
dp_ptozp(p,&p1); p = p1; |
|
for ( m = BDY(p), n = 0; m; m = NEXT(m), n++ ); |
|
if ( n == 1 ) { |
|
m = BDY(p); |
|
NEWMP(mr); mr->dl = m->dl; mr->c = (P)ONE; NEXT(mr) = 0; |
|
MKDP(p->nv,mr,*rp); (*rp)->sugar = p->sugar; |
|
return; |
|
} |
|
w = (P *)ALLOCA(n*sizeof(P)); |
|
c = (Q *)ALLOCA(n*sizeof(Q)); |
|
for ( m =BDY(p), i = 0; i < n; m = NEXT(m), i++ ) |
|
if ( NUM(m->c) ) { |
|
c[i] = (Q)m->c; w[i] = (P)ONE; |
|
} else |
|
ptozp(m->c,1,&c[i],&w[i]); |
|
qltozl(c,n,&dvr); heu_nezgcdnpz(CO,w,n,&t); mulp(CO,t,(P)dvr,&g); |
|
if ( NUM(g) ) |
|
*rp = p; |
|
else { |
|
for ( mr0 = 0, m = BDY(p); m; m = NEXT(m) ) { |
|
NEXTMP(mr0,mr); divsp(CO,m->c,g,&mr->c); mr->dl = m->dl; |
|
} |
|
NEXT(mr) = 0; MKDP(p->nv,mr0,*rp); (*rp)->sugar = p->sugar; |
|
} |
|
} |
|
} |
|
|
|
void heu_nezgcdnpz(VL vl,P *pl,int m,P *pr) |
|
{ |
|
int i,r; |
|
P gcd,t,s1,s2,u; |
|
Q rq; |
|
DCP dc; |
|
extern int DP_Print; |
|
|
|
while ( 1 ) { |
|
for ( i = 0, s1 = 0; i < m; i++ ) { |
|
r = random(); UTOQ(r,rq); |
|
mulp(vl,pl[i],(P)rq,&t); addp(vl,s1,t,&u); s1 = u; |
|
} |
|
for ( i = 0, s2 = 0; i < m; i++ ) { |
|
r = random(); UTOQ(r,rq); |
|
mulp(vl,pl[i],(P)rq,&t); addp(vl,s2,t,&u); s2 = u; |
|
} |
|
ezgcdp(vl,s1,s2,&gcd); |
|
if ( DP_Print > 2 ) |
|
{ fprintf(asir_out,"(%d)",nmonop(gcd)); fflush(asir_out); } |
|
for ( i = 0; i < m; i++ ) { |
|
if ( !divtpz(vl,pl[i],gcd,&t) ) |
|
break; |
|
} |
|
if ( i == m ) |
|
break; |
|
} |
|
*pr = gcd; |
|
} |
|
|
|
void dp_prim_mod(DP p,int mod,DP *rp) |
|
{ |
|
P t,g; |
|
MP m,mr,mr0; |
|
|
|
if ( !p ) |
|
*rp = 0; |
|
else if ( NoGCD ) |
|
*rp = p; |
|
else { |
|
for ( m = BDY(p), g = m->c, m = NEXT(m); m; m = NEXT(m) ) { |
|
gcdprsmp(CO,mod,g,m->c,&t); g = t; |
|
} |
|
for ( mr0 = 0, m = BDY(p); m; m = NEXT(m) ) { |
|
NEXTMP(mr0,mr); divsmp(CO,mod,m->c,g,&mr->c); mr->dl = m->dl; |
|
} |
|
NEXT(mr) = 0; MKDP(p->nv,mr0,*rp); (*rp)->sugar = p->sugar; |
|
} |
|
} |
|
|
|
void dp_cont(DP p,Q *rp) |
|
{ |
|
VECT v; |
|
|
|
dp_dtov(p,&v); igcdv(v,rp); |
|
} |
|
|
|
void dp_dtov(DP dp,VECT *rp) |
|
{ |
|
MP m,t; |
|
int i,n; |
|
VECT v; |
|
pointer *p; |
|
|
|
m = BDY(dp); |
|
for ( t = m, n = 0; t; t = NEXT(t), n++ ); |
|
MKVECT(v,n); |
|
for ( i = 0, p = BDY(v), t = m; i < n; t = NEXT(t), i++ ) |
|
p[i] = (pointer)(t->c); |
|
*rp = v; |
|
} |
|
|
/* |
/* |
* Old codes |
* s-poly computation |
|
* |
*/ |
*/ |
|
|
#if 0 |
void dp_sp(DP p1,DP p2,DP *rp) |
void Pdp_sep(arg,rp) |
|
NODE arg; |
|
VECT *rp; |
|
{ |
{ |
DP p,s; |
int i,n,td; |
MP m,t,mr,mr0; |
int *w; |
int i,j,n,d,r,q,q1,nv; |
DL d1,d2,d; |
VECT w; |
MP m; |
pointer *pw; |
DP t,s1,s2,u; |
|
Q c,c1,c2; |
|
N gn,tn; |
|
|
p = (DP)ARG0(arg); m = BDY(p); |
n = p1->nv; d1 = BDY(p1)->dl; d2 = BDY(p2)->dl; |
d = QTOS((Q)ARG1(arg)); |
w = (int *)ALLOCA(n*sizeof(int)); |
for ( t = m, n = 0; t; t = NEXT(t), n++ ); |
for ( i = 0, td = 0; i < n; i++ ) { |
if ( d > n ) |
w[i] = MAX(d1->d[i],d2->d[i]); td += MUL_WEIGHT(w[i],i); |
d = n; |
} |
q = n/d; r = n%d; q1 = q+1; |
|
MKVECT(w,d); *rp = w; |
NEWDL(d,n); d->td = td - d1->td; |
t = m; pw = BDY(w); nv = p->nv; |
for ( i = 0; i < n; i++ ) |
for ( i = 0; i < r; i++ ) { |
d->d[i] = w[i] - d1->d[i]; |
for ( mr0 = 0, j = 0; j < q1; j++, t = NEXT(t) ) { |
c1 = (Q)BDY(p1)->c; c2 = (Q)BDY(p2)->c; |
NEXTMP(mr0,mr); |
if ( INT(c1) && INT(c2) ) { |
mr->c = t->c; |
gcdn(NM(c1),NM(c2),&gn); |
mr->dl = t->dl; |
if ( !UNIN(gn) ) { |
|
divsn(NM(c1),gn,&tn); NTOQ(tn,SGN(c1),c); c1 = c; |
|
divsn(NM(c2),gn,&tn); NTOQ(tn,SGN(c2),c); c2 = c; |
} |
} |
NEXT(mr) = 0; MKDP(nv,mr0,s); s->sugar = p->sugar; |
|
pw[i] = (pointer)s; |
|
} |
} |
for ( ; i < d; i++ ) { |
|
for ( mr0 = 0, j = 0; j < q; j++, t = NEXT(t) ) { |
NEWMP(m); m->dl = d; m->c = (P)c2; NEXT(m) = 0; |
NEXTMP(mr0,mr); |
MKDP(n,m,s1); s1->sugar = d->td; muld(CO,s1,p1,&t); |
mr->c = t->c; |
|
mr->dl = t->dl; |
NEWDL(d,n); d->td = td - d2->td; |
|
for ( i = 0; i < n; i++ ) |
|
d->d[i] = w[i] - d2->d[i]; |
|
NEWMP(m); m->dl = d; m->c = (P)c1; NEXT(m) = 0; |
|
MKDP(n,m,s2); s2->sugar = d->td; muld(CO,s2,p2,&u); |
|
|
|
subd(CO,t,u,rp); |
|
if ( GenTrace ) { |
|
LIST hist; |
|
NODE node; |
|
|
|
node = mknode(4,ONE,0,s1,ONE); |
|
MKLIST(hist,node); |
|
MKNODE(TraceList,hist,0); |
|
|
|
node = mknode(4,ONE,0,0,ONE); |
|
chsgnd(s2,(DP *)&ARG2(node)); |
|
MKLIST(hist,node); |
|
MKNODE(node,hist,TraceList); TraceList = node; |
|
} |
|
} |
|
|
|
void _dp_sp_dup(DP p1,DP p2,DP *rp) |
|
{ |
|
int i,n,td; |
|
int *w; |
|
DL d1,d2,d; |
|
MP m; |
|
DP t,s1,s2,u; |
|
Q c,c1,c2; |
|
N gn,tn; |
|
|
|
n = p1->nv; d1 = BDY(p1)->dl; d2 = BDY(p2)->dl; |
|
w = (int *)ALLOCA(n*sizeof(int)); |
|
for ( i = 0, td = 0; i < n; i++ ) { |
|
w[i] = MAX(d1->d[i],d2->d[i]); td += MUL_WEIGHT(w[i],i); |
|
} |
|
|
|
_NEWDL(d,n); d->td = td - d1->td; |
|
for ( i = 0; i < n; i++ ) |
|
d->d[i] = w[i] - d1->d[i]; |
|
c1 = (Q)BDY(p1)->c; c2 = (Q)BDY(p2)->c; |
|
if ( INT(c1) && INT(c2) ) { |
|
gcdn(NM(c1),NM(c2),&gn); |
|
if ( !UNIN(gn) ) { |
|
divsn(NM(c1),gn,&tn); NTOQ(tn,SGN(c1),c); c1 = c; |
|
divsn(NM(c2),gn,&tn); NTOQ(tn,SGN(c2),c); c2 = c; |
} |
} |
NEXT(mr) = 0; MKDP(nv,mr0,s); s->sugar = p->sugar; |
|
pw[i] = (pointer)s; |
|
} |
} |
|
|
|
_NEWMP(m); m->dl = d; m->c = (P)c2; NEXT(m) = 0; |
|
_MKDP(n,m,s1); s1->sugar = d->td; _muld_dup(CO,s1,p1,&t); _free_dp(s1); |
|
|
|
_NEWDL(d,n); d->td = td - d2->td; |
|
for ( i = 0; i < n; i++ ) |
|
d->d[i] = w[i] - d2->d[i]; |
|
_NEWMP(m); m->dl = d; chsgnp((P)c1,&m->c); NEXT(m) = 0; |
|
_MKDP(n,m,s2); s2->sugar = d->td; _muld_dup(CO,s2,p2,&u); _free_dp(s2); |
|
|
|
_addd_destructive(CO,t,u,rp); |
|
if ( GenTrace ) { |
|
LIST hist; |
|
NODE node; |
|
|
|
node = mknode(4,ONE,0,s1,ONE); |
|
MKLIST(hist,node); |
|
MKNODE(TraceList,hist,0); |
|
|
|
node = mknode(4,ONE,0,0,ONE); |
|
chsgnd(s2,(DP *)&ARG2(node)); |
|
MKLIST(hist,node); |
|
MKNODE(node,hist,TraceList); TraceList = node; |
|
} |
} |
} |
|
|
void dp_ptozp_mpi(dist,ndist,p,rp) |
void dp_sp_mod(DP p1,DP p2,int mod,DP *rp) |
NODE dist; |
|
int ndist; |
|
DP p,*rp; |
|
{ |
{ |
int i,j,k,l,n,nsep; |
int i,n,td; |
|
int *w; |
|
DL d1,d2,d; |
MP m; |
MP m; |
NODE tn,n0,n1,n2,n3; |
DP t,s,u; |
|
|
|
n = p1->nv; d1 = BDY(p1)->dl; d2 = BDY(p2)->dl; |
|
w = (int *)ALLOCA(n*sizeof(int)); |
|
for ( i = 0, td = 0; i < n; i++ ) { |
|
w[i] = MAX(d1->d[i],d2->d[i]); td += MUL_WEIGHT(w[i],i); |
|
} |
|
NEWDL_NOINIT(d,n); d->td = td - d1->td; |
|
for ( i = 0; i < n; i++ ) |
|
d->d[i] = w[i] - d1->d[i]; |
|
NEWMP(m); m->dl = d; m->c = (P)BDY(p2)->c; NEXT(m) = 0; |
|
MKDP(n,m,s); s->sugar = d->td; mulmd(CO,mod,p1,s,&t); |
|
NEWDL_NOINIT(d,n); d->td = td - d2->td; |
|
for ( i = 0; i < n; i++ ) |
|
d->d[i] = w[i] - d2->d[i]; |
|
NEWMP(m); m->dl = d; m->c = (P)BDY(p1)->c; NEXT(m) = 0; |
|
MKDP(n,m,s); s->sugar = d->td; mulmd(CO,mod,p2,s,&u); |
|
submd(CO,mod,t,u,rp); |
|
} |
|
|
|
void _dp_sp_mod_dup(DP p1,DP p2,int mod,DP *rp) |
|
{ |
|
int i,n,td; |
|
int *w; |
|
DL d1,d2,d; |
|
MP m; |
|
DP t,s,u; |
|
|
|
n = p1->nv; d1 = BDY(p1)->dl; d2 = BDY(p2)->dl; |
|
w = (int *)ALLOCA(n*sizeof(int)); |
|
for ( i = 0, td = 0; i < n; i++ ) { |
|
w[i] = MAX(d1->d[i],d2->d[i]); td += MUL_WEIGHT(w[i],i); |
|
} |
|
_NEWDL(d,n); d->td = td - d1->td; |
|
for ( i = 0; i < n; i++ ) |
|
d->d[i] = w[i] - d1->d[i]; |
|
_NEWMP(m); m->dl = d; m->c = BDY(p2)->c; NEXT(m) = 0; |
|
_MKDP(n,m,s); s->sugar = d->td; _mulmd_dup(mod,s,p1,&t); _free_dp(s); |
|
_NEWDL(d,n); d->td = td - d2->td; |
|
for ( i = 0; i < n; i++ ) |
|
d->d[i] = w[i] - d2->d[i]; |
|
_NEWMP(m); m->dl = d; m->c = STOI(mod - ITOS(BDY(p1)->c)); NEXT(m) = 0; |
|
_MKDP(n,m,s); s->sugar = d->td; _mulmd_dup(mod,s,p2,&u); _free_dp(s); |
|
_addmd_destructive(mod,t,u,rp); |
|
} |
|
|
|
void _dp_sp_mod(DP p1,DP p2,int mod,DP *rp) |
|
{ |
|
int i,n,td; |
|
int *w; |
|
DL d1,d2,d; |
|
MP m; |
|
DP t,s,u; |
|
|
|
n = p1->nv; d1 = BDY(p1)->dl; d2 = BDY(p2)->dl; |
|
w = (int *)ALLOCA(n*sizeof(int)); |
|
for ( i = 0, td = 0; i < n; i++ ) { |
|
w[i] = MAX(d1->d[i],d2->d[i]); td += MUL_WEIGHT(w[i],i); |
|
} |
|
NEWDL(d,n); d->td = td - d1->td; |
|
for ( i = 0; i < n; i++ ) |
|
d->d[i] = w[i] - d1->d[i]; |
|
NEWMP(m); m->dl = d; m->c = BDY(p2)->c; NEXT(m) = 0; |
|
MKDP(n,m,s); s->sugar = d->td; mulmd_dup(mod,s,p1,&t); |
|
NEWDL(d,n); d->td = td - d2->td; |
|
for ( i = 0; i < n; i++ ) |
|
d->d[i] = w[i] - d2->d[i]; |
|
NEWMP(m); m->dl = d; m->c = STOI(mod - ITOS(BDY(p1)->c)); NEXT(m) = 0; |
|
MKDP(n,m,s); s->sugar = d->td; mulmd_dup(mod,s,p2,&u); |
|
addmd_destructive(mod,t,u,rp); |
|
} |
|
|
|
/* |
|
* m-reduction |
|
* do content reduction over Z or Q(x,...) |
|
* do nothing over finite fields |
|
* |
|
*/ |
|
|
|
void dp_red(DP p0,DP p1,DP p2,DP *head,DP *rest,P *dnp,DP *multp) |
|
{ |
|
int i,n; |
|
DL d1,d2,d; |
|
MP m; |
|
DP t,s,r,h; |
|
Q c,c1,c2; |
|
N gn,tn; |
|
P g,a; |
|
P p[2]; |
|
|
|
n = p1->nv; d1 = BDY(p1)->dl; d2 = BDY(p2)->dl; |
|
NEWDL(d,n); d->td = d1->td - d2->td; |
|
for ( i = 0; i < n; i++ ) |
|
d->d[i] = d1->d[i]-d2->d[i]; |
|
c1 = (Q)BDY(p1)->c; c2 = (Q)BDY(p2)->c; |
|
if ( dp_fcoeffs == N_GFS ) { |
|
p[0] = (P)c1; p[1] = (P)c2; |
|
gcdsf(CO,p,2,&g); |
|
divsp(CO,(P)c1,g,&a); c1 = (Q)a; divsp(CO,(P)c2,g,&a); c2 = (Q)a; |
|
} else if ( dp_fcoeffs ) { |
|
/* do nothing */ |
|
} else if ( INT(c1) && INT(c2) ) { |
|
gcdn(NM(c1),NM(c2),&gn); |
|
if ( !UNIN(gn) ) { |
|
divsn(NM(c1),gn,&tn); NTOQ(tn,SGN(c1),c); c1 = c; |
|
divsn(NM(c2),gn,&tn); NTOQ(tn,SGN(c2),c); c2 = c; |
|
} |
|
} else { |
|
ezgcdpz(CO,(P)c1,(P)c2,&g); |
|
divsp(CO,(P)c1,g,&a); c1 = (Q)a; divsp(CO,(P)c2,g,&a); c2 = (Q)a; |
|
} |
|
NEWMP(m); m->dl = d; chsgnp((P)c1,&m->c); NEXT(m) = 0; MKDP(n,m,s); s->sugar = d->td; |
|
*multp = s; |
|
muld(CO,s,p2,&t); muldc(CO,p1,(P)c2,&s); addd(CO,s,t,&r); |
|
muldc(CO,p0,(P)c2,&h); |
|
*head = h; *rest = r; *dnp = (P)c2; |
|
} |
|
|
|
/* |
|
* m-reduction by a marked poly |
|
* do content reduction over Z or Q(x,...) |
|
* do nothing over finite fields |
|
* |
|
*/ |
|
|
|
|
|
void dp_red_marked(DP p0,DP p1,DP p2,DP hp2,DP *head,DP *rest,P *dnp,DP *multp) |
|
{ |
|
int i,n; |
|
DL d1,d2,d; |
|
MP m; |
|
DP t,s,r,h; |
|
Q c,c1,c2; |
|
N gn,tn; |
|
P g,a; |
|
P p[2]; |
|
|
|
n = p1->nv; d1 = BDY(p1)->dl; d2 = BDY(hp2)->dl; |
|
NEWDL(d,n); d->td = d1->td - d2->td; |
|
for ( i = 0; i < n; i++ ) |
|
d->d[i] = d1->d[i]-d2->d[i]; |
|
c1 = (Q)BDY(p1)->c; c2 = (Q)BDY(hp2)->c; |
|
if ( dp_fcoeffs == N_GFS ) { |
|
p[0] = (P)c1; p[1] = (P)c2; |
|
gcdsf(CO,p,2,&g); |
|
divsp(CO,(P)c1,g,&a); c1 = (Q)a; divsp(CO,(P)c2,g,&a); c2 = (Q)a; |
|
} else if ( dp_fcoeffs ) { |
|
/* do nothing */ |
|
} else if ( INT(c1) && INT(c2) ) { |
|
gcdn(NM(c1),NM(c2),&gn); |
|
if ( !UNIN(gn) ) { |
|
divsn(NM(c1),gn,&tn); NTOQ(tn,SGN(c1),c); c1 = c; |
|
divsn(NM(c2),gn,&tn); NTOQ(tn,SGN(c2),c); c2 = c; |
|
} |
|
} else { |
|
ezgcdpz(CO,(P)c1,(P)c2,&g); |
|
divsp(CO,(P)c1,g,&a); c1 = (Q)a; divsp(CO,(P)c2,g,&a); c2 = (Q)a; |
|
} |
|
NEWMP(m); m->dl = d; chsgnp((P)c1,&m->c); NEXT(m) = 0; MKDP(n,m,s); s->sugar = d->td; |
|
*multp = s; |
|
muld(CO,s,p2,&t); muldc(CO,p1,(P)c2,&s); addd(CO,s,t,&r); |
|
muldc(CO,p0,(P)c2,&h); |
|
*head = h; *rest = r; *dnp = (P)c2; |
|
} |
|
|
|
void dp_red_marked_mod(DP p0,DP p1,DP p2,DP hp2,int mod,DP *head,DP *rest,P *dnp) |
|
{ |
|
int i,n; |
|
DL d1,d2,d; |
|
MP m; |
|
DP t,s,r,h; |
|
P c1,c2,g,u; |
|
|
|
n = p1->nv; d1 = BDY(p1)->dl; d2 = BDY(hp2)->dl; |
|
NEWDL(d,n); d->td = d1->td - d2->td; |
|
for ( i = 0; i < n; i++ ) |
|
d->d[i] = d1->d[i]-d2->d[i]; |
|
c1 = (P)BDY(p1)->c; c2 = (P)BDY(hp2)->c; |
|
gcdprsmp(CO,mod,c1,c2,&g); |
|
divsmp(CO,mod,c1,g,&u); c1 = u; divsmp(CO,mod,c2,g,&u); c2 = u; |
|
if ( NUM(c2) ) { |
|
divsmp(CO,mod,c1,c2,&u); c1 = u; c2 = (P)ONEM; |
|
} |
|
NEWMP(m); m->dl = d; chsgnmp(mod,(P)c1,&m->c); NEXT(m) = 0; |
|
MKDP(n,m,s); s->sugar = d->td; mulmd(CO,mod,s,p2,&t); |
|
if ( NUM(c2) ) { |
|
addmd(CO,mod,p1,t,&r); h = p0; |
|
} else { |
|
mulmdc(CO,mod,p1,c2,&s); addmd(CO,mod,s,t,&r); mulmdc(CO,mod,p0,c2,&h); |
|
} |
|
*head = h; *rest = r; *dnp = c2; |
|
} |
|
|
|
/* m-reduction over a field */ |
|
|
|
void dp_red_f(DP p1,DP p2,DP *rest) |
|
{ |
|
int i,n; |
|
DL d1,d2,d; |
|
MP m; |
|
DP t,s; |
|
Obj a,b; |
|
|
|
n = p1->nv; |
|
d1 = BDY(p1)->dl; d2 = BDY(p2)->dl; |
|
|
|
NEWDL(d,n); d->td = d1->td - d2->td; |
|
for ( i = 0; i < n; i++ ) |
|
d->d[i] = d1->d[i]-d2->d[i]; |
|
|
|
NEWMP(m); m->dl = d; |
|
divr(CO,(Obj)BDY(p1)->c,(Obj)BDY(p2)->c,&a); chsgnr(a,&b); |
|
C(m) = (P)b; |
|
NEXT(m) = 0; MKDP(n,m,s); s->sugar = d->td; |
|
|
|
muld(CO,s,p2,&t); addd(CO,p1,t,rest); |
|
} |
|
|
|
void dp_red_mod(DP p0,DP p1,DP p2,int mod,DP *head,DP *rest,P *dnp) |
|
{ |
|
int i,n; |
|
DL d1,d2,d; |
|
MP m; |
|
DP t,s,r,h; |
|
P c1,c2,g,u; |
|
|
|
n = p1->nv; d1 = BDY(p1)->dl; d2 = BDY(p2)->dl; |
|
NEWDL(d,n); d->td = d1->td - d2->td; |
|
for ( i = 0; i < n; i++ ) |
|
d->d[i] = d1->d[i]-d2->d[i]; |
|
c1 = (P)BDY(p1)->c; c2 = (P)BDY(p2)->c; |
|
gcdprsmp(CO,mod,c1,c2,&g); |
|
divsmp(CO,mod,c1,g,&u); c1 = u; divsmp(CO,mod,c2,g,&u); c2 = u; |
|
if ( NUM(c2) ) { |
|
divsmp(CO,mod,c1,c2,&u); c1 = u; c2 = (P)ONEM; |
|
} |
|
NEWMP(m); m->dl = d; chsgnmp(mod,(P)c1,&m->c); NEXT(m) = 0; |
|
MKDP(n,m,s); s->sugar = d->td; mulmd(CO,mod,s,p2,&t); |
|
if ( NUM(c2) ) { |
|
addmd(CO,mod,p1,t,&r); h = p0; |
|
} else { |
|
mulmdc(CO,mod,p1,c2,&s); addmd(CO,mod,s,t,&r); mulmdc(CO,mod,p0,c2,&h); |
|
} |
|
*head = h; *rest = r; *dnp = c2; |
|
} |
|
|
|
struct oEGT eg_red_mod; |
|
|
|
void _dp_red_mod_destructive(DP p1,DP p2,int mod,DP *rp) |
|
{ |
|
int i,n; |
|
DL d1,d2,d; |
|
MP m; |
|
DP t,s; |
|
int c,c1,c2; |
|
extern int do_weyl; |
|
|
|
n = p1->nv; d1 = BDY(p1)->dl; d2 = BDY(p2)->dl; |
|
_NEWDL(d,n); d->td = d1->td - d2->td; |
|
for ( i = 0; i < n; i++ ) |
|
d->d[i] = d1->d[i]-d2->d[i]; |
|
c = invm(ITOS(BDY(p2)->c),mod); |
|
c2 = ITOS(BDY(p1)->c); |
|
DMAR(c,c2,0,mod,c1); |
|
_NEWMP(m); m->dl = d; m->c = STOI(mod-c1); NEXT(m) = 0; |
|
#if 0 |
|
_MKDP(n,m,s); s->sugar = d->td; |
|
_mulmd_dup(mod,s,p2,&t); _free_dp(s); |
|
#else |
|
if ( do_weyl ) { |
|
_MKDP(n,m,s); s->sugar = d->td; |
|
_mulmd_dup(mod,s,p2,&t); _free_dp(s); |
|
} else { |
|
_mulmdm_dup(mod,p2,m,&t); _FREEMP(m); |
|
} |
|
#endif |
|
/* get_eg(&t0); */ |
|
_addmd_destructive(mod,p1,t,rp); |
|
/* get_eg(&t1); add_eg(&eg_red_mod,&t0,&t1); */ |
|
} |
|
|
|
/* |
|
* normal form computation |
|
* |
|
*/ |
|
|
|
void dp_true_nf(NODE b,DP g,DP *ps,int full,DP *rp,P *dnp) |
|
{ |
|
DP u,p,d,s,t,dmy; |
|
NODE l; |
|
MP m,mr; |
|
int i,n; |
|
int *wb; |
|
int sugar,psugar; |
|
P dn,tdn,tdn1; |
|
|
|
dn = (P)ONE; |
|
if ( !g ) { |
|
*rp = 0; *dnp = dn; return; |
|
} |
|
for ( n = 0, l = b; l; l = NEXT(l), n++ ); |
|
wb = (int *)ALLOCA(n*sizeof(int)); |
|
for ( i = 0, l = b; i < n; l = NEXT(l), i++ ) |
|
wb[i] = QTOS((Q)BDY(l)); |
|
sugar = g->sugar; |
|
for ( d = 0; g; ) { |
|
for ( u = 0, i = 0; i < n; i++ ) { |
|
if ( dp_redble(g,p = ps[wb[i]]) ) { |
|
dp_red(d,g,p,&t,&u,&tdn,&dmy); |
|
psugar = (BDY(g)->dl->td - BDY(p)->dl->td) + p->sugar; |
|
sugar = MAX(sugar,psugar); |
|
if ( !u ) { |
|
if ( d ) |
|
d->sugar = sugar; |
|
*rp = d; *dnp = dn; return; |
|
} else { |
|
d = t; |
|
mulp(CO,dn,tdn,&tdn1); dn = tdn1; |
|
} |
|
break; |
|
} |
|
} |
|
if ( u ) |
|
g = u; |
|
else if ( !full ) { |
|
if ( g ) { |
|
MKDP(g->nv,BDY(g),t); t->sugar = sugar; g = t; |
|
} |
|
*rp = g; *dnp = dn; return; |
|
} else { |
|
m = BDY(g); NEWMP(mr); mr->dl = m->dl; mr->c = m->c; |
|
NEXT(mr) = 0; MKDP(g->nv,mr,t); t->sugar = mr->dl->td; |
|
addd(CO,d,t,&s); d = s; |
|
dp_rest(g,&t); g = t; |
|
} |
|
} |
|
if ( d ) |
|
d->sugar = sugar; |
|
*rp = d; *dnp = dn; |
|
} |
|
|
|
void dp_removecont2(DP p1,DP p2,DP *r1p,DP *r2p,Q *contp) |
|
{ |
struct oVECT v; |
struct oVECT v; |
VECT c,cs; |
int i,n1,n2,n; |
VECT qi,ri; |
MP m,m0,t; |
LIST *qr; |
Q *w; |
int id; |
Q h; |
Obj dmy; |
|
Q d0,d1,gcd,a,u,u1; |
|
Q *q,*r; |
|
STRING iqr_v; |
|
pointer *b; |
|
N qn,gn; |
|
double get_rtime(); |
|
int blen; |
|
double t0; |
|
double t_e,t_d,t_d1,t_c; |
|
|
|
|
if ( p1 ) { |
|
for ( i = 0, m = BDY(p1); m; m = NEXT(m), i++ ); |
|
n1 = i; |
|
} else |
|
n1 = 0; |
|
if ( p2 ) { |
|
for ( i = 0, m = BDY(p2); m; m = NEXT(m), i++ ); |
|
n2 = i; |
|
} else |
|
n2 = 0; |
|
n = n1+n2; |
|
if ( !n ) { |
|
*r1p = 0; *r2p = 0; *contp = ONE; return; |
|
} |
|
w = (Q *)ALLOCA(n*sizeof(Q)); |
|
v.len = n; |
|
v.body = (pointer *)w; |
|
i = 0; |
|
if ( p1 ) |
|
for ( m = BDY(p1); i < n1; m = NEXT(m), i++ ) w[i] = (Q)m->c; |
|
if ( p2 ) |
|
for ( m = BDY(p2); i < n; m = NEXT(m), i++ ) w[i] = (Q)m->c; |
|
h = w[0]; removecont_array((P *)w,n,1); divq(h,w[0],contp); |
|
i = 0; |
|
if ( p1 ) { |
|
for ( m0 = 0, t = BDY(p1); i < n1; i++, t = NEXT(t) ) { |
|
NEXTMP(m0,m); m->c = (P)w[i]; m->dl = t->dl; |
|
} |
|
NEXT(m) = 0; |
|
MKDP(p1->nv,m0,*r1p); (*r1p)->sugar = p1->sugar; |
|
} else |
|
*r1p = 0; |
|
if ( p2 ) { |
|
for ( m0 = 0, t = BDY(p2); i < n; i++, t = NEXT(t) ) { |
|
NEXTMP(m0,m); m->c = (P)w[i]; m->dl = t->dl; |
|
} |
|
NEXT(m) = 0; |
|
MKDP(p2->nv,m0,*r2p); (*r2p)->sugar = p2->sugar; |
|
} else |
|
*r2p = 0; |
|
} |
|
|
|
/* true nf by a marked GB */ |
|
|
|
void dp_true_nf_marked(NODE b,DP g,DP *ps,DP *hps,DP *rp,P *nmp,P *dnp) |
|
{ |
|
DP u,p,d,s,t,dmy,hp; |
|
NODE l; |
|
MP m,mr; |
|
int i,n,hmag; |
|
int *wb; |
|
int sugar,psugar,multiple; |
|
P nm,tnm1,dn,tdn,tdn1; |
|
Q cont; |
|
|
|
multiple = 0; |
|
hmag = multiple*HMAG(g); |
|
nm = (P)ONE; |
|
dn = (P)ONE; |
|
if ( !g ) { |
|
*rp = 0; *dnp = dn; return; |
|
} |
|
for ( n = 0, l = b; l; l = NEXT(l), n++ ); |
|
wb = (int *)ALLOCA(n*sizeof(int)); |
|
for ( i = 0, l = b; i < n; l = NEXT(l), i++ ) |
|
wb[i] = QTOS((Q)BDY(l)); |
|
sugar = g->sugar; |
|
for ( d = 0; g; ) { |
|
for ( u = 0, i = 0; i < n; i++ ) { |
|
if ( dp_redble(g,hp = hps[wb[i]]) ) { |
|
p = ps[wb[i]]; |
|
dp_red_marked(d,g,p,hp,&t,&u,&tdn,&dmy); |
|
psugar = (BDY(g)->dl->td - BDY(p)->dl->td) + p->sugar; |
|
sugar = MAX(sugar,psugar); |
|
if ( !u ) { |
|
goto last; |
|
} else { |
|
d = t; |
|
mulp(CO,dn,tdn,&tdn1); dn = tdn1; |
|
} |
|
break; |
|
} |
|
} |
|
if ( u ) { |
|
g = u; |
|
if ( multiple && ((d && HMAG(d)>hmag) || (HMAG(g)>hmag)) ) { |
|
dp_removecont2(d,g,&t,&u,&cont); d = t; g = u; |
|
mulp(CO,nm,(P)cont,&tnm1); nm = tnm1; |
|
if ( d ) |
|
hmag = multiple*HMAG(d); |
|
else |
|
hmag = multiple*HMAG(g); |
|
} |
|
} else { |
|
m = BDY(g); NEWMP(mr); mr->dl = m->dl; mr->c = m->c; |
|
NEXT(mr) = 0; MKDP(g->nv,mr,t); t->sugar = mr->dl->td; |
|
addd(CO,d,t,&s); d = s; |
|
dp_rest(g,&t); g = t; |
|
} |
|
} |
|
last: |
|
if ( d ) { |
|
dp_removecont2(d,0,&t,&u,&cont); d = t; |
|
mulp(CO,nm,(P)cont,&tnm1); nm = tnm1; |
|
d->sugar = sugar; |
|
} |
|
*rp = d; *nmp = nm; *dnp = dn; |
|
} |
|
|
|
void dp_true_nf_marked_mod(NODE b,DP g,DP *ps,DP *hps,int mod,DP *rp,P *dnp) |
|
{ |
|
DP hp,u,p,d,s,t; |
|
NODE l; |
|
MP m,mr; |
|
int i,n; |
|
int *wb; |
|
int sugar,psugar; |
|
P dn,tdn,tdn1; |
|
|
|
dn = (P)ONEM; |
|
if ( !g ) { |
|
*rp = 0; *dnp = dn; return; |
|
} |
|
for ( n = 0, l = b; l; l = NEXT(l), n++ ); |
|
wb = (int *)ALLOCA(n*sizeof(int)); |
|
for ( i = 0, l = b; i < n; l = NEXT(l), i++ ) |
|
wb[i] = QTOS((Q)BDY(l)); |
|
sugar = g->sugar; |
|
for ( d = 0; g; ) { |
|
for ( u = 0, i = 0; i < n; i++ ) { |
|
if ( dp_redble(g,hp = hps[wb[i]]) ) { |
|
p = ps[wb[i]]; |
|
dp_red_marked_mod(d,g,p,hp,mod,&t,&u,&tdn); |
|
psugar = (BDY(g)->dl->td - BDY(p)->dl->td) + p->sugar; |
|
sugar = MAX(sugar,psugar); |
|
if ( !u ) { |
|
if ( d ) |
|
d->sugar = sugar; |
|
*rp = d; *dnp = dn; return; |
|
} else { |
|
d = t; |
|
mulmp(CO,mod,dn,tdn,&tdn1); dn = tdn1; |
|
} |
|
break; |
|
} |
|
} |
|
if ( u ) |
|
g = u; |
|
else { |
|
m = BDY(g); NEWMP(mr); mr->dl = m->dl; mr->c = m->c; |
|
NEXT(mr) = 0; MKDP(g->nv,mr,t); t->sugar = mr->dl->td; |
|
addmd(CO,mod,d,t,&s); d = s; |
|
dp_rest(g,&t); g = t; |
|
} |
|
} |
|
if ( d ) |
|
d->sugar = sugar; |
|
*rp = d; *dnp = dn; |
|
} |
|
|
|
/* nf computation over Z */ |
|
|
|
void dp_nf_z(NODE b,DP g,DP *ps,int full,int multiple,DP *rp) |
|
{ |
|
DP u,p,d,s,t,dmy1; |
|
P dmy; |
|
NODE l; |
|
MP m,mr; |
|
int i,n; |
|
int *wb; |
|
int hmag; |
|
int sugar,psugar; |
|
|
|
if ( !g ) { |
|
*rp = 0; return; |
|
} |
|
for ( n = 0, l = b; l; l = NEXT(l), n++ ); |
|
wb = (int *)ALLOCA(n*sizeof(int)); |
|
for ( i = 0, l = b; i < n; l = NEXT(l), i++ ) |
|
wb[i] = QTOS((Q)BDY(l)); |
|
|
|
hmag = multiple*HMAG(g); |
|
sugar = g->sugar; |
|
|
|
for ( d = 0; g; ) { |
|
for ( u = 0, i = 0; i < n; i++ ) { |
|
if ( dp_redble(g,p = ps[wb[i]]) ) { |
|
dp_red(d,g,p,&t,&u,&dmy,&dmy1); |
|
psugar = (BDY(g)->dl->td - BDY(p)->dl->td) + p->sugar; |
|
sugar = MAX(sugar,psugar); |
|
if ( !u ) { |
|
if ( d ) |
|
d->sugar = sugar; |
|
*rp = d; return; |
|
} |
|
d = t; |
|
break; |
|
} |
|
} |
|
if ( u ) { |
|
g = u; |
|
if ( d ) { |
|
if ( multiple && HMAG(d) > hmag ) { |
|
dp_ptozp2(d,g,&t,&u); d = t; g = u; |
|
hmag = multiple*HMAG(d); |
|
} |
|
} else { |
|
if ( multiple && HMAG(g) > hmag ) { |
|
dp_ptozp(g,&t); g = t; |
|
hmag = multiple*HMAG(g); |
|
} |
|
} |
|
} |
|
else if ( !full ) { |
|
if ( g ) { |
|
MKDP(g->nv,BDY(g),t); t->sugar = sugar; g = t; |
|
} |
|
*rp = g; return; |
|
} else { |
|
m = BDY(g); NEWMP(mr); mr->dl = m->dl; mr->c = m->c; |
|
NEXT(mr) = 0; MKDP(g->nv,mr,t); t->sugar = mr->dl->td; |
|
addd(CO,d,t,&s); d = s; |
|
dp_rest(g,&t); g = t; |
|
|
|
} |
|
} |
|
if ( d ) |
|
d->sugar = sugar; |
|
*rp = d; |
|
} |
|
|
|
/* nf computation over a field */ |
|
|
|
void dp_nf_f(NODE b,DP g,DP *ps,int full,DP *rp) |
|
{ |
|
DP u,p,d,s,t; |
|
NODE l; |
|
MP m,mr; |
|
int i,n; |
|
int *wb; |
|
int sugar,psugar; |
|
|
|
if ( !g ) { |
|
*rp = 0; return; |
|
} |
|
for ( n = 0, l = b; l; l = NEXT(l), n++ ); |
|
wb = (int *)ALLOCA(n*sizeof(int)); |
|
for ( i = 0, l = b; i < n; l = NEXT(l), i++ ) |
|
wb[i] = QTOS((Q)BDY(l)); |
|
|
|
sugar = g->sugar; |
|
for ( d = 0; g; ) { |
|
for ( u = 0, i = 0; i < n; i++ ) { |
|
if ( dp_redble(g,p = ps[wb[i]]) ) { |
|
dp_red_f(g,p,&u); |
|
psugar = (BDY(g)->dl->td - BDY(p)->dl->td) + p->sugar; |
|
sugar = MAX(sugar,psugar); |
|
if ( !u ) { |
|
if ( d ) |
|
d->sugar = sugar; |
|
*rp = d; return; |
|
} |
|
break; |
|
} |
|
} |
|
if ( u ) |
|
g = u; |
|
else if ( !full ) { |
|
if ( g ) { |
|
MKDP(g->nv,BDY(g),t); t->sugar = sugar; g = t; |
|
} |
|
*rp = g; return; |
|
} else { |
|
m = BDY(g); NEWMP(mr); mr->dl = m->dl; mr->c = m->c; |
|
NEXT(mr) = 0; MKDP(g->nv,mr,t); t->sugar = mr->dl->td; |
|
addd(CO,d,t,&s); d = s; |
|
dp_rest(g,&t); g = t; |
|
} |
|
} |
|
if ( d ) |
|
d->sugar = sugar; |
|
*rp = d; |
|
} |
|
|
|
/* nf computation over GF(mod) (only for internal use) */ |
|
|
|
void dp_nf_mod(NODE b,DP g,DP *ps,int mod,int full,DP *rp) |
|
{ |
|
DP u,p,d,s,t; |
|
P dmy; |
|
NODE l; |
|
MP m,mr; |
|
int sugar,psugar; |
|
|
|
if ( !g ) { |
|
*rp = 0; return; |
|
} |
|
sugar = g->sugar; |
|
for ( d = 0; g; ) { |
|
for ( u = 0, l = b; l; l = NEXT(l) ) { |
|
if ( dp_redble(g,p = ps[(int)BDY(l)]) ) { |
|
dp_red_mod(d,g,p,mod,&t,&u,&dmy); |
|
psugar = (BDY(g)->dl->td - BDY(p)->dl->td) + p->sugar; |
|
sugar = MAX(sugar,psugar); |
|
if ( !u ) { |
|
if ( d ) |
|
d->sugar = sugar; |
|
*rp = d; return; |
|
} |
|
d = t; |
|
break; |
|
} |
|
} |
|
if ( u ) |
|
g = u; |
|
else if ( !full ) { |
|
if ( g ) { |
|
MKDP(g->nv,BDY(g),t); t->sugar = sugar; g = t; |
|
} |
|
*rp = g; return; |
|
} else { |
|
m = BDY(g); NEWMP(mr); mr->dl = m->dl; mr->c = m->c; |
|
NEXT(mr) = 0; MKDP(g->nv,mr,t); t->sugar = mr->dl->td; |
|
addmd(CO,mod,d,t,&s); d = s; |
|
dp_rest(g,&t); g = t; |
|
} |
|
} |
|
if ( d ) |
|
d->sugar = sugar; |
|
*rp = d; |
|
} |
|
|
|
void dp_true_nf_mod(NODE b,DP g,DP *ps,int mod,int full,DP *rp,P *dnp) |
|
{ |
|
DP u,p,d,s,t; |
|
NODE l; |
|
MP m,mr; |
|
int i,n; |
|
int *wb; |
|
int sugar,psugar; |
|
P dn,tdn,tdn1; |
|
|
|
dn = (P)ONEM; |
|
if ( !g ) { |
|
*rp = 0; *dnp = dn; return; |
|
} |
|
for ( n = 0, l = b; l; l = NEXT(l), n++ ); |
|
wb = (int *)ALLOCA(n*sizeof(int)); |
|
for ( i = 0, l = b; i < n; l = NEXT(l), i++ ) |
|
wb[i] = QTOS((Q)BDY(l)); |
|
sugar = g->sugar; |
|
for ( d = 0; g; ) { |
|
for ( u = 0, i = 0; i < n; i++ ) { |
|
if ( dp_redble(g,p = ps[wb[i]]) ) { |
|
dp_red_mod(d,g,p,mod,&t,&u,&tdn); |
|
psugar = (BDY(g)->dl->td - BDY(p)->dl->td) + p->sugar; |
|
sugar = MAX(sugar,psugar); |
|
if ( !u ) { |
|
if ( d ) |
|
d->sugar = sugar; |
|
*rp = d; *dnp = dn; return; |
|
} else { |
|
d = t; |
|
mulmp(CO,mod,dn,tdn,&tdn1); dn = tdn1; |
|
} |
|
break; |
|
} |
|
} |
|
if ( u ) |
|
g = u; |
|
else if ( !full ) { |
|
if ( g ) { |
|
MKDP(g->nv,BDY(g),t); t->sugar = sugar; g = t; |
|
} |
|
*rp = g; *dnp = dn; return; |
|
} else { |
|
m = BDY(g); NEWMP(mr); mr->dl = m->dl; mr->c = m->c; |
|
NEXT(mr) = 0; MKDP(g->nv,mr,t); t->sugar = mr->dl->td; |
|
addmd(CO,mod,d,t,&s); d = s; |
|
dp_rest(g,&t); g = t; |
|
} |
|
} |
|
if ( d ) |
|
d->sugar = sugar; |
|
*rp = d; *dnp = dn; |
|
} |
|
|
|
void _dp_nf_mod_destructive(NODE b,DP g,DP *ps,int mod,int full,DP *rp) |
|
{ |
|
DP u,p,d; |
|
NODE l; |
|
MP m,mrd; |
|
int sugar,psugar,n,h_reducible; |
|
|
|
if ( !g ) { |
|
*rp = 0; return; |
|
} |
|
sugar = g->sugar; |
|
n = g->nv; |
|
for ( d = 0; g; ) { |
|
for ( h_reducible = 0, l = b; l; l = NEXT(l) ) { |
|
if ( dp_redble(g,p = ps[(int)BDY(l)]) ) { |
|
h_reducible = 1; |
|
psugar = (BDY(g)->dl->td - BDY(p)->dl->td) + p->sugar; |
|
_dp_red_mod_destructive(g,p,mod,&u); g = u; |
|
sugar = MAX(sugar,psugar); |
|
if ( !g ) { |
|
if ( d ) |
|
d->sugar = sugar; |
|
_dptodp(d,rp); _free_dp(d); return; |
|
} |
|
break; |
|
} |
|
} |
|
if ( !h_reducible ) { |
|
/* head term is not reducible */ |
|
if ( !full ) { |
|
if ( g ) |
|
g->sugar = sugar; |
|
_dptodp(g,rp); _free_dp(g); return; |
|
} else { |
|
m = BDY(g); |
|
if ( NEXT(m) ) { |
|
BDY(g) = NEXT(m); NEXT(m) = 0; |
|
} else { |
|
_FREEDP(g); g = 0; |
|
} |
|
if ( d ) { |
|
for ( mrd = BDY(d); NEXT(mrd); mrd = NEXT(mrd) ); |
|
NEXT(mrd) = m; |
|
} else { |
|
_MKDP(n,m,d); |
|
} |
|
} |
|
} |
|
} |
|
if ( d ) |
|
d->sugar = sugar; |
|
_dptodp(d,rp); _free_dp(d); |
|
} |
|
|
|
/* reduction by linear base over a field */ |
|
|
|
void dp_lnf_f(DP p1,DP p2,NODE g,DP *r1p,DP *r2p) |
|
{ |
|
DP r1,r2,b1,b2,t,s; |
|
Obj c,c1,c2; |
|
NODE l,b; |
|
int n; |
|
|
|
if ( !p1 ) { |
|
*r1p = p1; *r2p = p2; return; |
|
} |
|
n = p1->nv; |
|
for ( l = g, r1 = p1, r2 = p2; l; l = NEXT(l) ) { |
|
if ( !r1 ) { |
|
*r1p = r1; *r2p = r2; return; |
|
} |
|
b = BDY((LIST)BDY(l)); b1 = (DP)BDY(b); |
|
if ( dl_equal(n,BDY(r1)->dl,BDY(b1)->dl) ) { |
|
b2 = (DP)BDY(NEXT(b)); |
|
divr(CO,(Obj)ONE,(Obj)BDY(b1)->c,&c1); |
|
mulr(CO,c1,(Obj)BDY(r1)->c,&c2); chsgnr(c2,&c); |
|
muldc(CO,b1,(P)c,&t); addd(CO,r1,t,&s); r1 = s; |
|
muldc(CO,b2,(P)c,&t); addd(CO,r2,t,&s); r2 = s; |
|
} |
|
} |
|
*r1p = r1; *r2p = r2; |
|
} |
|
|
|
/* reduction by linear base over GF(mod) */ |
|
|
|
void dp_lnf_mod(DP p1,DP p2,NODE g,int mod,DP *r1p,DP *r2p) |
|
{ |
|
DP r1,r2,b1,b2,t,s; |
|
P c; |
|
MQ c1,c2; |
|
NODE l,b; |
|
int n; |
|
|
|
if ( !p1 ) { |
|
*r1p = p1; *r2p = p2; return; |
|
} |
|
n = p1->nv; |
|
for ( l = g, r1 = p1, r2 = p2; l; l = NEXT(l) ) { |
|
if ( !r1 ) { |
|
*r1p = r1; *r2p = r2; return; |
|
} |
|
b = BDY((LIST)BDY(l)); b1 = (DP)BDY(b); |
|
if ( dl_equal(n,BDY(r1)->dl,BDY(b1)->dl) ) { |
|
b2 = (DP)BDY(NEXT(b)); |
|
invmq(mod,(MQ)BDY(b1)->c,&c1); |
|
mulmq(mod,c1,(MQ)BDY(r1)->c,&c2); chsgnmp(mod,(P)c2,&c); |
|
mulmdc(CO,mod,b1,c,&t); addmd(CO,mod,r1,t,&s); r1 = s; |
|
mulmdc(CO,mod,b2,c,&t); addmd(CO,mod,r2,t,&s); r2 = s; |
|
} |
|
} |
|
*r1p = r1; *r2p = r2; |
|
} |
|
|
|
void dp_nf_tab_mod(DP p,LIST *tab,int mod,DP *rp) |
|
{ |
|
DP s,t,u; |
|
MP m; |
|
DL h; |
|
int i,n; |
|
|
|
if ( !p ) { |
|
*rp = p; return; |
|
} |
|
n = p->nv; |
|
for ( s = 0, i = 0, m = BDY(p); m; m = NEXT(m) ) { |
|
h = m->dl; |
|
while ( !dl_equal(n,h,BDY((DP)BDY(BDY(tab[i])))->dl ) ) |
|
i++; |
|
mulmdc(CO,mod,(DP)BDY(NEXT(BDY(tab[i]))),m->c,&t); |
|
addmd(CO,mod,s,t,&u); s = u; |
|
} |
|
*rp = s; |
|
} |
|
|
|
void dp_nf_tab_f(DP p,LIST *tab,DP *rp) |
|
{ |
|
DP s,t,u; |
|
MP m; |
|
DL h; |
|
int i,n; |
|
|
|
if ( !p ) { |
|
*rp = p; return; |
|
} |
|
n = p->nv; |
|
for ( s = 0, i = 0, m = BDY(p); m; m = NEXT(m) ) { |
|
h = m->dl; |
|
while ( !dl_equal(n,h,BDY((DP)BDY(BDY(tab[i])))->dl ) ) |
|
i++; |
|
muldc(CO,(DP)BDY(NEXT(BDY(tab[i]))),m->c,&t); |
|
addd(CO,s,t,&u); s = u; |
|
} |
|
*rp = s; |
|
} |
|
|
|
/* |
|
* setting flags |
|
* call create_order_spec with vl=0 to set old type order. |
|
* |
|
*/ |
|
|
|
int create_order_spec(VL vl,Obj obj,struct order_spec **specp) |
|
{ |
|
int i,j,n,s,row,col,ret; |
|
struct order_spec *spec; |
|
struct order_pair *l; |
|
NODE node,t,tn; |
|
MAT m; |
|
pointer **b; |
|
int **w; |
|
|
|
if ( vl && obj && OID(obj) == O_LIST ) { |
|
ret = create_composite_order_spec(vl,(LIST)obj,specp); |
|
if ( show_orderspec ) |
|
print_composite_order_spec(*specp); |
|
return ret; |
|
} |
|
|
|
*specp = spec = (struct order_spec *)MALLOC(sizeof(struct order_spec)); |
|
if ( !obj || NUM(obj) ) { |
|
spec->id = 0; spec->obj = obj; |
|
spec->ord.simple = QTOS((Q)obj); |
|
return 1; |
|
} else if ( OID(obj) == O_LIST ) { |
|
node = BDY((LIST)obj); |
|
for ( n = 0, t = node; t; t = NEXT(t), n++ ); |
|
l = (struct order_pair *)MALLOC_ATOMIC(n*sizeof(struct order_pair)); |
|
for ( i = 0, t = node, s = 0; i < n; t = NEXT(t), i++ ) { |
|
tn = BDY((LIST)BDY(t)); l[i].order = QTOS((Q)BDY(tn)); |
|
tn = NEXT(tn); l[i].length = QTOS((Q)BDY(tn)); |
|
s += l[i].length; |
|
} |
|
spec->id = 1; spec->obj = obj; |
|
spec->ord.block.order_pair = l; |
|
spec->ord.block.length = n; spec->nv = s; |
|
return 1; |
|
} else if ( OID(obj) == O_MAT ) { |
|
m = (MAT)obj; row = m->row; col = m->col; b = BDY(m); |
|
w = almat(row,col); |
|
for ( i = 0; i < row; i++ ) |
|
for ( j = 0; j < col; j++ ) |
|
w[i][j] = QTOS((Q)b[i][j]); |
|
spec->id = 2; spec->obj = obj; |
|
spec->nv = col; spec->ord.matrix.row = row; |
|
spec->ord.matrix.matrix = w; |
|
return 1; |
|
} else |
|
return 0; |
|
} |
|
|
|
void print_composite_order_spec(struct order_spec *spec) |
|
{ |
|
int nv,n,len,i,j,k,start; |
|
struct weight_or_block *worb; |
|
|
|
nv = spec->nv; |
|
n = spec->ord.composite.length; |
|
worb = spec->ord.composite.w_or_b; |
|
for ( i = 0; i < n; i++, worb++ ) { |
|
len = worb->length; |
|
printf("[ "); |
|
switch ( worb->type ) { |
|
case IS_DENSE_WEIGHT: |
|
for ( j = 0; j < len; j++ ) |
|
printf("%d ",worb->body.dense_weight[j]); |
|
for ( ; j < nv; j++ ) |
|
printf("0 "); |
|
break; |
|
case IS_SPARSE_WEIGHT: |
|
for ( j = 0, k = 0; j < nv; j++ ) |
|
if ( j == worb->body.sparse_weight[k].pos ) |
|
printf("%d ",worb->body.sparse_weight[k++].value); |
|
else |
|
printf("0 "); |
|
break; |
|
case IS_BLOCK: |
|
start = worb->body.block.start; |
|
for ( j = 0; j < start; j++ ) printf("0 "); |
|
switch ( worb->body.block.order ) { |
|
case 0: |
|
for ( k = 0; k < len; k++, j++ ) printf("R "); |
|
break; |
|
case 1: |
|
for ( k = 0; k < len; k++, j++ ) printf("G "); |
|
break; |
|
case 2: |
|
for ( k = 0; k < len; k++, j++ ) printf("L "); |
|
break; |
|
} |
|
for ( ; j < nv; j++ ) printf("0 "); |
|
break; |
|
} |
|
printf("]\n"); |
|
} |
|
} |
|
|
|
struct order_spec *append_block(struct order_spec *spec, |
|
int nv,int nalg,int ord) |
|
{ |
|
MAT m,mat; |
|
int i,j,row,col,n; |
|
Q **b,**wp; |
|
int **w; |
|
NODE t,s,s0; |
|
struct order_pair *l,*l0; |
|
int n0,nv0; |
|
LIST list0,list1,list; |
|
Q oq,nq; |
|
struct order_spec *r; |
|
|
|
r = (struct order_spec *)MALLOC(sizeof(struct order_spec)); |
|
switch ( spec->id ) { |
|
case 0: |
|
STOQ(spec->ord.simple,oq); STOQ(nv,nq); |
|
t = mknode(2,oq,nq); MKLIST(list0,t); |
|
STOQ(ord,oq); STOQ(nalg,nq); |
|
t = mknode(2,oq,nq); MKLIST(list1,t); |
|
t = mknode(2,list0,list1); MKLIST(list,t); |
|
l = (struct order_pair *)MALLOC_ATOMIC(2*sizeof(struct order_pair)); |
|
l[0].order = spec->ord.simple; l[0].length = nv; |
|
l[1].order = ord; l[1].length = nalg; |
|
r->id = 1; r->obj = (Obj)list; |
|
r->ord.block.order_pair = l; |
|
r->ord.block.length = 2; |
|
r->nv = nv+nalg; |
|
break; |
|
case 1: |
|
if ( spec->nv != nv ) |
|
error("append_block : number of variables mismatch"); |
|
l0 = spec->ord.block.order_pair; |
|
n0 = spec->ord.block.length; |
|
nv0 = spec->nv; |
|
list0 = (LIST)spec->obj; |
|
n = n0+1; |
|
l = (struct order_pair *)MALLOC_ATOMIC(n*sizeof(struct order_pair)); |
|
for ( i = 0; i < n0; i++ ) |
|
l[i] = l0[i]; |
|
l[i].order = ord; l[i].length = nalg; |
|
for ( t = BDY(list0), s0 = 0; t; t = NEXT(t) ) { |
|
NEXTNODE(s0,s); BDY(s) = BDY(t); |
|
} |
|
STOQ(ord,oq); STOQ(nalg,nq); |
|
t = mknode(2,oq,nq); MKLIST(list,t); |
|
NEXTNODE(s0,s); BDY(s) = (pointer)list; NEXT(s) = 0; |
|
MKLIST(list,s0); |
|
r->id = 1; r->obj = (Obj)list; |
|
r->ord.block.order_pair = l; |
|
r->ord.block.length = n; |
|
r->nv = nv+nalg; |
|
break; |
|
case 2: |
|
if ( spec->nv != nv ) |
|
error("append_block : number of variables mismatch"); |
|
m = (MAT)spec->obj; |
|
row = m->row; col = m->col; b = (Q **)BDY(m); |
|
w = almat(row+nalg,col+nalg); |
|
MKMAT(mat,row+nalg,col+nalg); wp = (Q **)BDY(mat); |
|
for ( i = 0; i < row; i++ ) |
|
for ( j = 0; j < col; j++ ) { |
|
w[i][j] = QTOS(b[i][j]); |
|
wp[i][j] = b[i][j]; |
|
} |
|
for ( i = 0; i < nalg; i++ ) { |
|
w[i+row][i+col] = 1; |
|
wp[i+row][i+col] = ONE; |
|
} |
|
r->id = 2; r->obj = (Obj)mat; |
|
r->nv = col+nalg; r->ord.matrix.row = row+nalg; |
|
r->ord.matrix.matrix = w; |
|
break; |
|
case 3: |
|
default: |
|
/* XXX */ |
|
error("append_block : not implemented yet"); |
|
} |
|
return r; |
|
} |
|
|
|
int comp_sw(struct sparse_weight *a, struct sparse_weight *b) |
|
{ |
|
if ( a->pos > b->pos ) return 1; |
|
else if ( a->pos < b->pos ) return -1; |
|
else return 0; |
|
} |
|
|
|
/* order = [w_or_b, w_or_b, ... ] */ |
|
/* w_or_b = w or b */ |
|
/* w = [1,2,...] or [x,1,y,2,...] */ |
|
/* b = [@lex,x,y,...,z] etc */ |
|
|
|
int create_composite_order_spec(VL vl,LIST order,struct order_spec **specp) |
|
{ |
|
NODE wb,t,p; |
|
struct order_spec *spec; |
|
VL tvl; |
|
int n,i,j,k,l,start,end,len,w; |
|
int *dw; |
|
struct sparse_weight *sw; |
|
struct weight_or_block *w_or_b; |
|
Obj a0; |
|
NODE a; |
|
V v,sv,ev; |
|
SYMBOL sym; |
|
int *top; |
|
|
|
/* l = number of vars in vl */ |
|
for ( l = 0, tvl = vl; tvl; tvl = NEXT(tvl), l++ ); |
|
/* n = number of primitives in order */ |
|
wb = BDY(order); |
|
n = length(wb); |
|
*specp = spec = (struct order_spec *)MALLOC(sizeof(struct order_spec)); |
|
spec->id = 3; |
|
spec->obj = (Obj)order; |
|
spec->nv = l; |
|
spec->ord.composite.length = n; |
|
w_or_b = spec->ord.composite.w_or_b = (struct weight_or_block *) |
|
MALLOC(sizeof(struct weight_or_block)*(n+1)); |
|
|
|
/* top : register the top variable in each w_or_b specification */ |
|
top = (int *)ALLOCA(l*sizeof(int)); |
|
for ( i = 0; i < l; i++ ) top[i] = 0; |
|
|
|
for ( t = wb, i = 0; t; t = NEXT(t), i++ ) { |
|
if ( !BDY(t) || OID((Obj)BDY(t)) != O_LIST ) |
|
error("a list of lists must be specified for the key \"order\""); |
|
a = BDY((LIST)BDY(t)); |
|
len = length(a); |
|
a0 = (Obj)BDY(a); |
|
if ( !a0 || OID(a0) == O_N ) { |
|
/* a is a dense weight vector */ |
|
dw = (int *)MALLOC(sizeof(int)*len); |
|
for ( j = 0, p = a; j < len; p = NEXT(p), j++ ) { |
|
if ( !INT((Q)BDY(p)) ) |
|
error("a dense weight vector must be specified as a list of integers"); |
|
dw[j] = QTOS((Q)BDY(p)); |
|
} |
|
w_or_b[i].type = IS_DENSE_WEIGHT; |
|
w_or_b[i].length = len; |
|
w_or_b[i].body.dense_weight = dw; |
|
|
|
/* find the top */ |
|
for ( k = 0; k < len && !dw[k]; k++ ); |
|
if ( k < len ) top[k] = 1; |
|
|
|
} else if ( OID(a0) == O_P ) { |
|
/* a is a sparse weight vector */ |
|
len >>= 1; |
|
sw = (struct sparse_weight *) |
|
MALLOC(sizeof(struct sparse_weight)*len); |
|
for ( j = 0, p = a; j < len; j++ ) { |
|
if ( !BDY(p) || OID((P)BDY(p)) != O_P ) |
|
error("a sparse weight vector must be specified as [var1,weight1,...]"); |
|
v = VR((P)BDY(p)); p = NEXT(p); |
|
for ( tvl = vl, k = 0; tvl && tvl->v != v; |
|
k++, tvl = NEXT(tvl) ); |
|
if ( !tvl ) |
|
error("invalid variable name in a sparse weight vector"); |
|
sw[j].pos = k; |
|
if ( !INT((Q)BDY(p)) ) |
|
error("a sparse weight vector must be specified as [var1,weight1,...]"); |
|
sw[j].value = QTOS((Q)BDY(p)); p = NEXT(p); |
|
} |
|
qsort(sw,len,sizeof(struct sparse_weight), |
|
(int (*)(const void *,const void *))comp_sw); |
|
w_or_b[i].type = IS_SPARSE_WEIGHT; |
|
w_or_b[i].length = len; |
|
w_or_b[i].body.sparse_weight = sw; |
|
|
|
/* find the top */ |
|
for ( k = 0; k < len && !sw[k].value; k++ ); |
|
if ( k < len ) top[sw[k].pos] = 1; |
|
} else if ( OID(a0) == O_RANGE ) { |
|
/* [range(v1,v2),w] */ |
|
sv = VR((P)(((RANGE)a0)->start)); |
|
ev = VR((P)(((RANGE)a0)->end)); |
|
for ( tvl = vl, start = 0; tvl && tvl->v != sv; start++, tvl = NEXT(tvl) ); |
|
if ( !tvl ) |
|
error("invalid range"); |
|
for ( end = start; tvl && tvl->v != ev; end++, tvl = NEXT(tvl) ); |
|
if ( !tvl ) |
|
error("invalid range"); |
|
len = end-start+1; |
|
sw = (struct sparse_weight *) |
|
MALLOC(sizeof(struct sparse_weight)*len); |
|
w = QTOS((Q)BDY(NEXT(a))); |
|
for ( tvl = vl, k = 0; k < start; k++, tvl = NEXT(tvl) ); |
|
for ( j = 0 ; k <= end; k++, tvl = NEXT(tvl), j++ ) { |
|
sw[j].pos = k; |
|
sw[j].value = w; |
|
} |
|
w_or_b[i].type = IS_SPARSE_WEIGHT; |
|
w_or_b[i].length = len; |
|
w_or_b[i].body.sparse_weight = sw; |
|
|
|
/* register the top */ |
|
if ( w ) top[start] = 1; |
|
} else if ( OID(a0) == O_SYMBOL ) { |
|
/* a is a block */ |
|
sym = (SYMBOL)a0; a = NEXT(a); len--; |
|
if ( OID((Obj)BDY(a)) == O_RANGE ) { |
|
sv = VR((P)(((RANGE)BDY(a))->start)); |
|
ev = VR((P)(((RANGE)BDY(a))->end)); |
|
for ( tvl = vl, start = 0; tvl && tvl->v != sv; start++, tvl = NEXT(tvl) ); |
|
if ( !tvl ) |
|
error("invalid range"); |
|
for ( end = start; tvl && tvl->v != ev; end++, tvl = NEXT(tvl) ); |
|
if ( !tvl ) |
|
error("invalid range"); |
|
len = end-start+1; |
|
} else { |
|
for ( start = 0, tvl = vl; tvl->v != VR((P)BDY(a)); |
|
tvl = NEXT(tvl), start++ ); |
|
for ( p = NEXT(a), tvl = NEXT(tvl); p; |
|
p = NEXT(p), tvl = NEXT(tvl) ) { |
|
if ( !BDY(p) || OID((P)BDY(p)) != O_P ) |
|
error("a block must be specified as [ordsymbol,var1,var2,...]"); |
|
if ( tvl->v != VR((P)BDY(p)) ) break; |
|
} |
|
if ( p ) |
|
error("a block must be contiguous in the variable list"); |
|
} |
|
w_or_b[i].type = IS_BLOCK; |
|
w_or_b[i].length = len; |
|
w_or_b[i].body.block.start = start; |
|
if ( !strcmp(sym->name,"@grlex") ) |
|
w_or_b[i].body.block.order = 0; |
|
else if ( !strcmp(sym->name,"@glex") ) |
|
w_or_b[i].body.block.order = 1; |
|
else if ( !strcmp(sym->name,"@lex") ) |
|
w_or_b[i].body.block.order = 2; |
|
else |
|
error("invalid ordername"); |
|
/* register the tops */ |
|
for ( j = 0, k = start; j < len; j++, k++ ) |
|
top[k] = 1; |
|
} |
|
} |
|
for ( k = 0; k < l && top[k]; k++ ); |
|
if ( k < l ) { |
|
/* incomplete order specification; add @grlex */ |
|
w_or_b[n].type = IS_BLOCK; |
|
w_or_b[n].length = l; |
|
w_or_b[n].body.block.start = 0; |
|
w_or_b[n].body.block.order = 0; |
|
spec->ord.composite.length = n+1; |
|
} |
|
} |
|
|
|
/* module order spec */ |
|
|
|
void create_modorder_spec(int id,LIST shift,struct modorder_spec **s) |
|
{ |
|
struct modorder_spec *spec; |
|
NODE n,t; |
|
LIST list; |
|
int *ds; |
|
int i,l; |
|
Q q; |
|
|
|
*s = spec = (struct modorder_spec *)MALLOC(sizeof(struct modorder_spec)); |
|
spec->id = id; |
|
if ( shift ) { |
|
n = BDY(shift); |
|
spec->len = l = length(n); |
|
spec->degree_shift = ds = (int *)MALLOC_ATOMIC(l*sizeof(int)); |
|
for ( t = n, i = 0; t; t = NEXT(t), i++ ) |
|
ds[i] = QTOS((Q)BDY(t)); |
|
} else { |
|
spec->len = 0; |
|
spec->degree_shift = 0; |
|
} |
|
STOQ(id,q); |
|
n = mknode(2,q,shift); |
|
MKLIST(list,n); |
|
spec->obj = (Obj)list; |
|
} |
|
|
|
/* |
|
* converters |
|
* |
|
*/ |
|
|
|
void dp_homo(DP p,DP *rp) |
|
{ |
|
MP m,mr,mr0; |
|
int i,n,nv,td; |
|
DL dl,dlh; |
|
|
if ( !p ) |
if ( !p ) |
*rp = 0; |
*rp = 0; |
else { |
else { |
for ( m = BDY(p), n = 0; m; m = NEXT(m), n++ ); |
n = p->nv; nv = n + 1; |
nsep = ndist + 1; |
m = BDY(p); td = sugard(m); |
if ( n <= nsep ) { |
for ( mr0 = 0; m; m = NEXT(m) ) { |
dp_ptozp(p,rp); return; |
NEXTMP(mr0,mr); mr->c = m->c; |
|
dl = m->dl; |
|
mr->dl = dlh = (DL)MALLOC_ATOMIC((nv+1)*sizeof(int)); |
|
dlh->td = td; |
|
for ( i = 0; i < n; i++ ) |
|
dlh->d[i] = dl->d[i]; |
|
dlh->d[n] = td - dl->td; |
} |
} |
t0 = get_rtime(); |
NEXT(mr) = 0; MKDP(nv,mr0,*rp); (*rp)->sugar = p->sugar; |
dp_dtov(p,&c); |
} |
igcdv_estimate(c,&d0); |
} |
t_e = get_rtime()-t0; |
|
t0 = get_rtime(); |
void dp_dehomo(DP p,DP *rp) |
dp_dtov(p,&c); |
{ |
sepvect(c,nsep,&cs); |
MP m,mr,mr0; |
MKSTR(iqr_v,"iqr"); |
int i,n,nv; |
qr = (LIST *)CALLOC(nsep,sizeof(LIST)); |
DL dl,dlh; |
q = (Q *)CALLOC(n,sizeof(Q)); |
|
r = (Q *)CALLOC(n,sizeof(Q)); |
if ( !p ) |
for ( i = 0, tn = dist, b = BDY(cs); i < ndist; |
*rp = 0; |
i++, tn = NEXT(tn) ) { |
else { |
div_vect_mpi(QTOS((Q)BDY(tn)),b[i],d0); |
n = p->nv; nv = n - 1; |
|
m = BDY(p); |
|
for ( mr0 = 0; m; m = NEXT(m) ) { |
|
NEXTMP(mr0,mr); mr->c = m->c; |
|
dlh = m->dl; |
|
mr->dl = dl = (DL)MALLOC_ATOMIC((nv+1)*sizeof(int)); |
|
dl->td = dlh->td - dlh->d[nv]; |
|
for ( i = 0; i < nv; i++ ) |
|
dl->d[i] = dlh->d[i]; |
} |
} |
iqrv(b[i],d0,&qr[i]); |
NEXT(mr) = 0; MKDP(nv,mr0,*rp); (*rp)->sugar = p->sugar; |
dp_dtov(p,&c); |
} |
for ( i = j = 0, tn = dist; i < ndist; i++, tn = NEXT(tn) ) { |
} |
get_div_vect_mpi(QTOS((Q)BDY(tn)),&qr[i]); |
|
} |
void dp_mod(DP p,int mod,NODE subst,DP *rp) |
t_d = get_rtime()-t0; |
{ |
t_d1 = t_d/n; |
MP m,mr,mr0; |
t0 = get_rtime(); |
P t,s,s1; |
for ( i = j = 0; i < nsep; i++ ) { |
V v; |
tn = BDY(qr[i]); qi = (VECT)BDY(tn); ri = (VECT)BDY(NEXT(tn)); |
NODE tn; |
for ( k = 0, l = qi->len; k < l; k++, j++ ) { |
|
q[j] = (Q)BDY(qi)[k]; r[j] = (Q)BDY(ri)[k]; |
if ( !p ) |
|
*rp = 0; |
|
else { |
|
for ( mr0 = 0, m = BDY(p); m; m = NEXT(m) ) { |
|
for ( tn = subst, s = m->c; tn; tn = NEXT(tn) ) { |
|
v = VR((P)BDY(tn)); tn = NEXT(tn); |
|
substp(CO,s,v,(P)BDY(tn),&s1); s = s1; |
} |
} |
} |
ptomp(mod,s,&t); |
v.id = O_VECT; v.len = n; v.body = (pointer *)r; igcdv(&v,&d1); |
if ( t ) { |
if ( d1 ) { |
NEXTMP(mr0,mr); mr->c = t; mr->dl = m->dl; |
gcdn(NM(d0),NM(d1),&gn); NTOQ(gn,1,gcd); |
|
divsn(NM(d0),gn,&qn); NTOQ(qn,1,a); |
|
for ( i = 0; i < n; i++ ) { |
|
mulq(a,q[i],&u); |
|
if ( r[i] ) { |
|
divsn(NM(r[i]),gn,&qn); NTOQ(qn,SGN(r[i]),u1); |
|
addq(u,u1,&q[i]); |
|
} else |
|
q[i] = u; |
|
} |
} |
|
} |
|
if ( mr0 ) { |
|
NEXT(mr) = 0; MKDP(p->nv,mr0,*rp); (*rp)->sugar = p->sugar; |
} else |
} else |
gcd = d0; |
*rp = 0; |
dp_vtod(q,p,rp); |
|
t_c = get_rtime()-t0; |
|
blen=p_mag(gcd); |
|
pz_t_e += t_e; pz_t_d += t_d; pz_t_d1 += t_d1; pz_t_c += t_c; |
|
if ( 0 ) |
|
fprintf(stderr,"(%d,%d)",p_mag(d0)-blen,blen); |
|
} |
} |
} |
} |
|
|
void dp_ptozp_d(dist,ndist,p,rp) |
void dp_rat(DP p,DP *rp) |
NODE dist; |
|
int ndist; |
|
DP p,*rp; |
|
{ |
{ |
DP d,s,u,su; |
|
MP m,mr,mr0; |
MP m,mr,mr0; |
MP *w0,*w; |
|
Q *g; |
if ( !p ) |
int i,n,nv,nsep; |
*rp = 0; |
P t; |
else { |
NODE tn,n0,n1,n2; |
for ( mr0 = 0, m = BDY(p); m; m = NEXT(m) ) { |
|
NEXTMP(mr0,mr); mptop(m->c,&mr->c); mr->dl = m->dl; |
|
} |
|
if ( mr0 ) { |
|
NEXT(mr) = 0; MKDP(p->nv,mr0,*rp); (*rp)->sugar = p->sugar; |
|
} else |
|
*rp = 0; |
|
} |
|
} |
|
|
|
|
|
void homogenize_order(struct order_spec *old,int n,struct order_spec **newp) |
|
{ |
|
struct order_pair *l; |
|
int length,nv,row,i,j; |
|
int **newm,**oldm; |
|
struct order_spec *new; |
|
int onv,nnv,nlen,olen,owlen; |
|
struct weight_or_block *owb,*nwb; |
|
|
|
*newp = new = (struct order_spec *)MALLOC(sizeof(struct order_spec)); |
|
switch ( old->id ) { |
|
case 0: |
|
switch ( old->ord.simple ) { |
|
case 0: |
|
new->id = 0; new->ord.simple = 0; break; |
|
case 1: |
|
l = (struct order_pair *) |
|
MALLOC_ATOMIC(2*sizeof(struct order_pair)); |
|
l[0].length = n; l[0].order = 1; |
|
l[1].length = 1; l[1].order = 2; |
|
new->id = 1; |
|
new->ord.block.order_pair = l; |
|
new->ord.block.length = 2; new->nv = n+1; |
|
break; |
|
case 2: |
|
new->id = 0; new->ord.simple = 1; break; |
|
case 3: case 4: case 5: |
|
new->id = 0; new->ord.simple = old->ord.simple+3; |
|
dp_nelim = n-1; break; |
|
case 6: case 7: case 8: case 9: |
|
new->id = 0; new->ord.simple = old->ord.simple; break; |
|
default: |
|
error("homogenize_order : invalid input"); |
|
} |
|
break; |
|
case 1: |
|
length = old->ord.block.length; |
|
l = (struct order_pair *) |
|
MALLOC_ATOMIC((length+1)*sizeof(struct order_pair)); |
|
bcopy((char *)old->ord.block.order_pair,(char *)l,length*sizeof(struct order_pair)); |
|
l[length].order = 2; l[length].length = 1; |
|
new->id = 1; new->nv = n+1; |
|
new->ord.block.order_pair = l; |
|
new->ord.block.length = length+1; |
|
break; |
|
case 2: |
|
nv = old->nv; row = old->ord.matrix.row; |
|
oldm = old->ord.matrix.matrix; newm = almat(row+1,nv+1); |
|
for ( i = 0; i <= nv; i++ ) |
|
newm[0][i] = 1; |
|
for ( i = 0; i < row; i++ ) { |
|
for ( j = 0; j < nv; j++ ) |
|
newm[i+1][j] = oldm[i][j]; |
|
newm[i+1][j] = 0; |
|
} |
|
new->id = 2; new->nv = nv+1; |
|
new->ord.matrix.row = row+1; new->ord.matrix.matrix = newm; |
|
break; |
|
case 3: |
|
onv = old->nv; |
|
nnv = onv+1; |
|
olen = old->ord.composite.length; |
|
nlen = olen+1; |
|
owb = old->ord.composite.w_or_b; |
|
nwb = (struct weight_or_block *) |
|
MALLOC(nlen*sizeof(struct weight_or_block)); |
|
for ( i = 0; i < olen; i++ ) { |
|
nwb[i].type = owb[i].type; |
|
switch ( owb[i].type ) { |
|
case IS_DENSE_WEIGHT: |
|
owlen = owb[i].length; |
|
nwb[i].length = owlen+1; |
|
nwb[i].body.dense_weight = (int *)MALLOC((owlen+1)*sizeof(int)); |
|
for ( j = 0; j < owlen; j++ ) |
|
nwb[i].body.dense_weight[j] = owb[i].body.dense_weight[j]; |
|
nwb[i].body.dense_weight[owlen] = 0; |
|
break; |
|
case IS_SPARSE_WEIGHT: |
|
nwb[i].length = owb[i].length; |
|
nwb[i].body.sparse_weight = owb[i].body.sparse_weight; |
|
break; |
|
case IS_BLOCK: |
|
nwb[i].length = owb[i].length; |
|
nwb[i].body.block = owb[i].body.block; |
|
break; |
|
} |
|
} |
|
nwb[i].type = IS_SPARSE_WEIGHT; |
|
nwb[i].body.sparse_weight = |
|
(struct sparse_weight *)MALLOC(sizeof(struct sparse_weight)); |
|
nwb[i].body.sparse_weight[0].pos = onv; |
|
nwb[i].body.sparse_weight[0].value = 1; |
|
new->id = 3; |
|
new->nv = nnv; |
|
new->ord.composite.length = nlen; |
|
new->ord.composite.w_or_b = nwb; |
|
print_composite_order_spec(new); |
|
break; |
|
default: |
|
error("homogenize_order : invalid input"); |
|
} |
|
} |
|
|
|
void qltozl(Q *w,int n,Q *dvr) |
|
{ |
|
N nm,dn; |
|
N g,l1,l2,l3; |
|
Q c,d; |
|
int i; |
struct oVECT v; |
struct oVECT v; |
int sindex,id; |
|
Obj dmy; |
|
Q gcd; |
|
STRING gcd_h,div_h; |
|
|
|
|
for ( i = 0; i < n; i++ ) |
|
if ( w[i] && !INT(w[i]) ) |
|
break; |
|
if ( i == n ) { |
|
v.id = O_VECT; v.len = n; v.body = (pointer *)w; |
|
igcdv(&v,dvr); return; |
|
} |
|
c = w[0]; nm = NM(c); dn = INT(c) ? ONEN : DN(c); |
|
for ( i = 1; i < n; i++ ) { |
|
c = w[i]; l1 = INT(c) ? ONEN : DN(c); |
|
gcdn(nm,NM(c),&g); nm = g; |
|
gcdn(dn,l1,&l2); muln(dn,l1,&l3); divsn(l3,l2,&dn); |
|
} |
|
if ( UNIN(dn) ) |
|
NTOQ(nm,1,d); |
|
else |
|
NDTOQ(nm,dn,1,d); |
|
*dvr = d; |
|
} |
|
|
|
int comp_nm(Q *a,Q *b) |
|
{ |
|
return cmpn((*a)?NM(*a):0,(*b)?NM(*b):0); |
|
} |
|
|
|
void sortbynm(Q *w,int n) |
|
{ |
|
qsort(w,n,sizeof(Q),(int (*)(const void *,const void *))comp_nm); |
|
} |
|
|
|
|
|
/* |
|
* simple operations |
|
* |
|
*/ |
|
|
|
int dp_redble(DP p1,DP p2) |
|
{ |
|
int i,n; |
|
DL d1,d2; |
|
|
|
d1 = BDY(p1)->dl; d2 = BDY(p2)->dl; |
|
if ( d1->td < d2->td ) |
|
return 0; |
|
else { |
|
for ( i = 0, n = p1->nv; i < n; i++ ) |
|
if ( d1->d[i] < d2->d[i] ) |
|
return 0; |
|
return 1; |
|
} |
|
} |
|
|
|
void dp_subd(DP p1,DP p2,DP *rp) |
|
{ |
|
int i,n; |
|
DL d1,d2,d; |
|
MP m; |
|
DP s; |
|
|
|
n = p1->nv; d1 = BDY(p1)->dl; d2 = BDY(p2)->dl; |
|
NEWDL(d,n); d->td = d1->td - d2->td; |
|
for ( i = 0; i < n; i++ ) |
|
d->d[i] = d1->d[i]-d2->d[i]; |
|
NEWMP(m); m->dl = d; m->c = (P)ONE; NEXT(m) = 0; |
|
MKDP(n,m,s); s->sugar = d->td; |
|
*rp = s; |
|
} |
|
|
|
void dltod(DL d,int n,DP *rp) |
|
{ |
|
MP m; |
|
DP s; |
|
|
|
NEWMP(m); m->dl = d; m->c = (P)ONE; NEXT(m) = 0; |
|
MKDP(n,m,s); s->sugar = d->td; |
|
*rp = s; |
|
} |
|
|
|
void dp_hm(DP p,DP *rp) |
|
{ |
|
MP m,mr; |
|
|
if ( !p ) |
if ( !p ) |
*rp = 0; |
*rp = 0; |
else { |
else { |
for ( m = BDY(p), n = 0; m; m = NEXT(m), n++ ); |
m = BDY(p); |
nsep = ndist + 1; |
NEWMP(mr); mr->dl = m->dl; mr->c = m->c; NEXT(mr) = 0; |
if ( n <= nsep ) { |
MKDP(p->nv,mr,*rp); (*rp)->sugar = mr->dl->td; /* XXX */ |
dp_ptozp(p,rp); return; |
} |
|
} |
|
|
|
void dp_ht(DP p,DP *rp) |
|
{ |
|
MP m,mr; |
|
|
|
if ( !p ) |
|
*rp = 0; |
|
else { |
|
m = BDY(p); |
|
NEWMP(mr); mr->dl = m->dl; mr->c = (P)ONE; NEXT(mr) = 0; |
|
MKDP(p->nv,mr,*rp); (*rp)->sugar = mr->dl->td; /* XXX */ |
|
} |
|
} |
|
|
|
void dp_rest(DP p,DP *rp) |
|
{ |
|
MP m; |
|
|
|
m = BDY(p); |
|
if ( !NEXT(m) ) |
|
*rp = 0; |
|
else { |
|
MKDP(p->nv,NEXT(m),*rp); |
|
if ( *rp ) |
|
(*rp)->sugar = p->sugar; |
|
} |
|
} |
|
|
|
DL lcm_of_DL(int nv,DL dl1,DL dl2,DL dl) |
|
{ |
|
register int i, *d1, *d2, *d, td; |
|
|
|
if ( !dl ) NEWDL(dl,nv); |
|
d = dl->d, d1 = dl1->d, d2 = dl2->d; |
|
for ( td = 0, i = 0; i < nv; d1++, d2++, d++, i++ ) { |
|
*d = *d1 > *d2 ? *d1 : *d2; |
|
td += MUL_WEIGHT(*d,i); |
|
} |
|
dl->td = td; |
|
return dl; |
|
} |
|
|
|
int dl_equal(int nv,DL dl1,DL dl2) |
|
{ |
|
register int *d1, *d2, n; |
|
|
|
if ( dl1->td != dl2->td ) return 0; |
|
for ( d1 = dl1->d, d2 = dl2->d, n = nv; --n >= 0; d1++, d2++ ) |
|
if ( *d1 != *d2 ) return 0; |
|
return 1; |
|
} |
|
|
|
int dp_nt(DP p) |
|
{ |
|
int i; |
|
MP m; |
|
|
|
if ( !p ) |
|
return 0; |
|
else { |
|
for ( i = 0, m = BDY(p); m; m = NEXT(m), i++ ); |
|
return i; |
|
} |
|
} |
|
|
|
int dp_homogeneous(DP p) |
|
{ |
|
MP m; |
|
int d; |
|
|
|
if ( !p ) |
|
return 1; |
|
else { |
|
m = BDY(p); |
|
d = m->dl->td; |
|
m = NEXT(m); |
|
for ( ; m; m = NEXT(m) ) { |
|
if ( m->dl->td != d ) |
|
return 0; |
} |
} |
nv = p->nv; |
return 1; |
w0 = (MP *)MALLOC(nsep*sizeof(MP)); bzero(w0,nsep*sizeof(MP)); |
} |
w = (MP *)MALLOC(nsep*sizeof(MP)); bzero(w,nsep*sizeof(MP)); |
} |
g = (Q *)MALLOC(nsep*sizeof(Q)); bzero(g,nsep*sizeof(Q)); |
|
for ( m = BDY(p), i = 0; m; m = NEXT(m), i++, i %= nsep ) { |
void _print_mp(int nv,MP m) |
NEXTMP(w0[i],w[i]); w[i]->c = m->c; w[i]->dl = m->dl; |
{ |
|
int i; |
|
|
|
if ( !m ) |
|
return; |
|
for ( ; m; m = NEXT(m) ) { |
|
fprintf(stderr,"%d<",ITOS(C(m))); |
|
for ( i = 0; i < nv; i++ ) { |
|
fprintf(stderr,"%d",m->dl->d[i]); |
|
if ( i != nv-1 ) |
|
fprintf(stderr," "); |
} |
} |
MKSTR(gcd_h,"dp_igcdv_hist"); MKSTR(div_h,"dp_idivv_hist"); |
fprintf(stderr,">",C(m)); |
|
} |
|
fprintf(stderr,"\n"); |
|
} |
|
|
for ( i = 0, tn = dist; i < ndist; i++, tn = NEXT(tn) ) { |
static int cmp_mp_nvar; |
NEXT(w[i]) = 0; MKDP(nv,w0[i],d); d->sugar = p->sugar; |
|
MKNODE(n2,d,0); MKNODE(n1,gcd_h,n2); MKNODE(n0,BDY(tn),n1); |
int comp_mp(MP *a,MP *b) |
Pox_rpc(n0,&dmy); |
{ |
|
return -(*cmpdl)(cmp_mp_nvar,(*a)->dl,(*b)->dl); |
|
} |
|
|
|
void dp_sort(DP p,DP *rp) |
|
{ |
|
MP t,mp,mp0; |
|
int i,n; |
|
DP r; |
|
MP *w; |
|
|
|
if ( !p ) { |
|
*rp = 0; |
|
return; |
|
} |
|
for ( t = BDY(p), n = 0; t; t = NEXT(t), n++ ); |
|
w = (MP *)ALLOCA(n*sizeof(MP)); |
|
for ( t = BDY(p), i = 0; i < n; t = NEXT(t), i++ ) |
|
w[i] = t; |
|
cmp_mp_nvar = NV(p); |
|
qsort(w,n,sizeof(MP),(int (*)(const void *,const void *))comp_mp); |
|
mp0 = 0; |
|
for ( i = n-1; i >= 0; i-- ) { |
|
NEWMP(mp); mp->dl = w[i]->dl; C(mp) = C(w[i]); |
|
NEXT(mp) = mp0; mp0 = mp; |
|
} |
|
MKDP(p->nv,mp0,r); |
|
r->sugar = p->sugar; |
|
*rp = r; |
|
} |
|
|
|
DP extract_initial_term_from_dp(DP p,int *weight,int n); |
|
LIST extract_initial_term(LIST f,int *weight,int n); |
|
|
|
DP extract_initial_term_from_dp(DP p,int *weight,int n) |
|
{ |
|
int w,t,i,top; |
|
MP m,r0,r; |
|
DP dp; |
|
|
|
if ( !p ) return 0; |
|
top = 1; |
|
for ( m = BDY(p); m; m = NEXT(m) ) { |
|
for ( i = 0, t = 0; i < n; i++ ) |
|
t += weight[i]*m->dl->d[i]; |
|
if ( top || t > w ) { |
|
r0 = 0; |
|
w = t; |
|
top = 0; |
} |
} |
NEXT(w[i]) = 0; MKDP(nv,w0[i],d); d->sugar = p->sugar; |
if ( t == w ) { |
dp_igcdv_hist(d,&g[i]); |
NEXTMP(r0,r); |
|
r->dl = m->dl; |
|
r->c = m->c; |
|
} |
|
} |
|
NEXT(r) = 0; |
|
MKDP(p->nv,r0,dp); |
|
return dp; |
|
} |
|
|
for ( i = 0, tn = dist; i < ndist; i++, tn = NEXT(tn) ) |
LIST extract_initial_term(LIST f,int *weight,int n) |
Pox_pop_local(tn,&g[i]); |
{ |
v.id = O_VECT; v.len = nsep; v.body = (pointer *)g; igcdv(&v,&gcd); |
NODE nd,r0,r; |
|
Obj p; |
|
LIST l; |
|
|
for ( i = 0, tn = dist; i < ndist; i++, tn = NEXT(tn) ) { |
nd = BDY(f); |
MKNODE(n2,gcd,0); MKNODE(n1,div_h,n2); MKNODE(n0,BDY(tn),n1); |
for ( r0 = 0; nd; nd = NEXT(nd) ) { |
Pox_rpc(n0,&dmy); |
NEXTNODE(r0,r); |
|
p = (Obj)BDY(nd); |
|
BDY(r) = (pointer)extract_initial_term_from_dp((DP)p,weight,n); |
|
} |
|
if ( r0 ) NEXT(r) = 0; |
|
MKLIST(l,r0); |
|
return l; |
|
} |
|
|
|
LIST dp_initial_term(LIST f,struct order_spec *ord) |
|
{ |
|
int n,l,i; |
|
struct weight_or_block *worb; |
|
int *weight; |
|
|
|
switch ( ord->id ) { |
|
case 2: /* matrix order */ |
|
/* extract the first row */ |
|
n = ord->nv; |
|
weight = ord->ord.matrix.matrix[0]; |
|
return extract_initial_term(f,weight,n); |
|
case 3: /* composite order */ |
|
/* the first w_or_b */ |
|
worb = ord->ord.composite.w_or_b; |
|
switch ( worb->type ) { |
|
case IS_DENSE_WEIGHT: |
|
n = worb->length; |
|
weight = worb->body.dense_weight; |
|
return extract_initial_term(f,weight,n); |
|
case IS_SPARSE_WEIGHT: |
|
n = ord->nv; |
|
weight = (int *)ALLOCA(n*sizeof(int)); |
|
for ( i = 0; i < n; i++ ) weight[i] = 0; |
|
l = worb->length; |
|
for ( i = 0; i < l; i++ ) |
|
weight[worb->body.sparse_weight[i].pos] |
|
= worb->body.sparse_weight[i].value; |
|
return extract_initial_term(f,weight,n); |
|
default: |
|
error("dp_initial_term : unsupported order"); |
|
} |
|
default: |
|
error("dp_initial_term : unsupported order"); |
|
} |
|
} |
|
|
|
int highest_order_dp(DP p,int *weight,int n); |
|
LIST highest_order(LIST f,int *weight,int n); |
|
|
|
int highest_order_dp(DP p,int *weight,int n) |
|
{ |
|
int w,t,i,top; |
|
MP m; |
|
|
|
if ( !p ) return -1; |
|
top = 1; |
|
for ( m = BDY(p); m; m = NEXT(m) ) { |
|
for ( i = 0, t = 0; i < n; i++ ) |
|
t += weight[i]*m->dl->d[i]; |
|
if ( top || t > w ) { |
|
w = t; |
|
top = 0; |
} |
} |
dp_idivv_hist(gcd,&s); |
} |
|
return w; |
|
} |
|
|
for ( i = 0, tn = dist; i < ndist; i++, tn = NEXT(tn) ) { |
LIST highest_order(LIST f,int *weight,int n) |
Pox_pop_local(tn,&u); |
{ |
addd(CO,s,u,&su); s = su; |
int h; |
|
NODE nd,r0,r; |
|
Obj p; |
|
LIST l; |
|
Q q; |
|
|
|
nd = BDY(f); |
|
for ( r0 = 0; nd; nd = NEXT(nd) ) { |
|
NEXTNODE(r0,r); |
|
p = (Obj)BDY(nd); |
|
h = highest_order_dp((DP)p,weight,n); |
|
STOQ(h,q); |
|
BDY(r) = (pointer)q; |
|
} |
|
if ( r0 ) NEXT(r) = 0; |
|
MKLIST(l,r0); |
|
return l; |
|
} |
|
|
|
LIST dp_order(LIST f,struct order_spec *ord) |
|
{ |
|
int n,l,i; |
|
struct weight_or_block *worb; |
|
int *weight; |
|
|
|
switch ( ord->id ) { |
|
case 2: /* matrix order */ |
|
/* extract the first row */ |
|
n = ord->nv; |
|
weight = ord->ord.matrix.matrix[0]; |
|
return highest_order(f,weight,n); |
|
case 3: /* composite order */ |
|
/* the first w_or_b */ |
|
worb = ord->ord.composite.w_or_b; |
|
switch ( worb->type ) { |
|
case IS_DENSE_WEIGHT: |
|
n = worb->length; |
|
weight = worb->body.dense_weight; |
|
return highest_order(f,weight,n); |
|
case IS_SPARSE_WEIGHT: |
|
n = ord->nv; |
|
weight = (int *)ALLOCA(n*sizeof(int)); |
|
for ( i = 0; i < n; i++ ) weight[i] = 0; |
|
l = worb->length; |
|
for ( i = 0; i < l; i++ ) |
|
weight[worb->body.sparse_weight[i].pos] |
|
= worb->body.sparse_weight[i].value; |
|
return highest_order(f,weight,n); |
|
default: |
|
error("dp_initial_term : unsupported order"); |
|
} |
|
default: |
|
error("dp_initial_term : unsupported order"); |
|
} |
|
} |
|
|
|
int dpv_ht(DPV p,DP *h) |
|
{ |
|
int len,max,maxi,i,t; |
|
DP *e; |
|
MP m,mr; |
|
|
|
len = p->len; |
|
e = p->body; |
|
max = -1; |
|
maxi = -1; |
|
for ( i = 0; i < len; i++ ) |
|
if ( e[i] && (t = BDY(e[i])->dl->td) > max ) { |
|
max = t; |
|
maxi = i; |
} |
} |
*rp = s; |
if ( max < 0 ) { |
|
*h = 0; |
|
return -1; |
|
} else { |
|
m = BDY(e[maxi]); |
|
NEWMP(mr); mr->dl = m->dl; mr->c = (P)ONE; NEXT(mr) = 0; |
|
MKDP(e[maxi]->nv,mr,*h); (*h)->sugar = mr->dl->td; /* XXX */ |
|
return maxi; |
} |
} |
} |
} |
|
|
|
/* return 1 if 0 <_w1 v && v <_w2 0 */ |
|
|
|
int in_c12(int n,int *v,int row1,int **w1,int row2, int **w2) |
|
{ |
|
int t1,t2; |
|
|
|
t1 = compare_zero(n,v,row1,w1); |
|
t2 = compare_zero(n,v,row2,w2); |
|
if ( t1 > 0 && t2 < 0 ) return 1; |
|
else return 0; |
|
} |
|
|
|
/* 0 < u => 1, 0 > u => -1 */ |
|
|
|
int compare_zero(int n,int *u,int row,int **w) |
|
{ |
|
int i,j,t; |
|
int *wi; |
|
|
|
for ( i = 0; i < row; i++ ) { |
|
wi = w[i]; |
|
for ( j = 0, t = 0; j < n; j++ ) t += u[j]*wi[j]; |
|
if ( t > 0 ) return 1; |
|
else if ( t < 0 ) return -1; |
|
} |
|
return 0; |
|
} |
|
|
|
/* functions for generic groebner walk */ |
|
/* u=0 means u=-infty */ |
|
|
|
int compare_facet_preorder(int n,int *u,int *v, |
|
int row1,int **w1,int row2,int **w2) |
|
{ |
|
int i,j,s,t,tu,tv; |
|
int *w2i,*uv; |
|
|
|
if ( !u ) return 1; |
|
uv = W_ALLOC(n); |
|
for ( i = 0; i < row2; i++ ) { |
|
w2i = w2[i]; |
|
for ( j = 0, tu = tv = 0; j < n; j++ ) |
|
if ( s = w2i[j] ) { |
|
tu += s*u[j]; tv += s*v[j]; |
|
} |
|
for ( j = 0; j < n; j++ ) uv[j] = u[j]*tv-v[j]*tu; |
|
t = compare_zero(n,uv,row1,w1); |
|
if ( t > 0 ) return 1; |
|
else if ( t < 0 ) return 0; |
|
} |
|
return 1; |
|
} |
|
|
|
/* return 0 if last_w = infty */ |
|
|
|
NODE compute_last_w(NODE g,NODE gh,int n,int **w, |
|
int row1,int **w1,int row2,int **w2) |
|
{ |
|
DP d; |
|
MP f,m0,m; |
|
int *wt,*v,*h; |
|
NODE t,s,n0,tn,n1,r0,r; |
|
int i; |
|
|
|
wt = W_ALLOC(n); |
|
n0 = 0; |
|
for ( t = g, s = gh; t; t = NEXT(t), s = NEXT(s) ) { |
|
f = BDY((DP)BDY(t)); |
|
h = BDY((DP)BDY(s))->dl->d; |
|
for ( ; f; f = NEXT(f) ) { |
|
for ( i = 0; i < n; i++ ) wt[i] = h[i]-f->dl->d[i]; |
|
for ( i = 0; i < n && !wt[i]; i++ ); |
|
if ( i == n ) continue; |
|
|
|
if ( in_c12(n,wt,row1,w1,row2,w2) && |
|
compare_facet_preorder(n,*w,wt,row1,w1,row2,w2) ) { |
|
v = (int *)MALLOC_ATOMIC(n*sizeof(int)); |
|
for ( i = 0; i < n; i++ ) v[i] = wt[i]; |
|
MKNODE(n1,v,n0); n0 = n1; |
|
} |
|
} |
|
} |
|
if ( !n0 ) return 0; |
|
for ( t = n0; t; t = NEXT(t) ) { |
|
v = (int *)BDY(t); |
|
for ( s = n0; s; s = NEXT(s) ) |
|
if ( !compare_facet_preorder(n,v,(int *)BDY(s),row1,w1,row2,w2) ) |
|
break; |
|
if ( !s ) { |
|
*w = v; |
|
break; |
|
} |
|
} |
|
if ( !t ) |
|
error("compute_last_w : cannot happen"); |
|
r0 = 0; |
|
for ( t = g, s = gh; t; t = NEXT(t), s = NEXT(s) ) { |
|
f = BDY((DP)BDY(t)); |
|
h = BDY((DP)BDY(s))->dl->d; |
|
for ( m0 = 0; f; f = NEXT(f) ) { |
|
for ( i = 0; i < n; i++ ) wt[i] = h[i]-f->dl->d[i]; |
|
for ( i = 0; i < n && !wt[i]; i++ ); |
|
if ( i == n || |
|
(compare_facet_preorder(n,wt,*w,row1,w1,row2,w2) |
|
&& compare_facet_preorder(n,*w,wt,row1,w1,row2,w2)) ) { |
|
NEXTMP(m0,m); m->c = f->c; m->dl = f->dl; |
|
} |
|
} |
|
NEXT(m) = 0; |
|
MKDP(((DP)BDY(t))->nv,m0,d); d->sugar = ((DP)BDY(t))->sugar; |
|
NEXTNODE(r0,r); BDY(r) = (pointer)d; |
|
} |
|
NEXT(r) = 0; |
|
return r0; |
|
} |
|
|
|
/* compute a sufficient set of d(f)=u-v */ |
|
|
|
NODE compute_essential_df(DP *g,DP *gh,int ng) |
|
{ |
|
int nv,i,j,k,t,lj; |
|
NODE r,r1,ri,rt,r0; |
|
MP m; |
|
MP *mj; |
|
DL di,hj,dl,dlt; |
|
int *d,*dt; |
|
LIST l; |
|
Q q; |
|
|
|
nv = g[0]->nv; |
|
r = 0; |
|
for ( j = 0; j < ng; j++ ) { |
|
for ( m = BDY(g[j]), lj = 0; m; m = NEXT(m), lj++ ); |
|
mj = (MP *)ALLOCA(lj*sizeof(MP)); |
|
for ( m = BDY(g[j]), k = 0; m; m = NEXT(m), k++ ) |
|
mj[k] = m; |
|
for ( i = 0; i < lj; i++ ) { |
|
for ( di = mj[i]->dl, k = i+1; k < lj; k++ ) |
|
if ( _dl_redble(di,mj[k]->dl,nv) ) break; |
|
if ( k < lj ) mj[i] = 0; |
|
} |
|
hj = BDY(gh[j])->dl; |
|
_NEWDL(dl,nv); d = dl->d; |
|
r0 = r; |
|
for ( i = 0; i < lj; i++ ) { |
|
if ( mj[i] && !dl_equal(nv,di=mj[i]->dl,hj) ) { |
|
for ( k = 0, t = 0; k < nv; k++ ) { |
|
d[k] = hj->d[k]-di->d[k]; |
|
t += d[k]; |
|
} |
|
dl->td = t; |
|
#if 1 |
|
for ( rt = r0; rt; rt = NEXT(rt) ) { |
|
dlt = (DL)BDY(rt); |
|
if ( dlt->td != dl->td ) continue; |
|
for ( dt = dlt->d, k = 0; k < nv; k++ ) |
|
if ( d[k] != dt[k] ) break; |
|
if ( k == nv ) break; |
|
} |
|
#else |
|
rt = 0; |
#endif |
#endif |
|
if ( !rt ) { |
|
MKNODE(r1,dl,r); r = r1; |
|
_NEWDL(dl,nv); d = dl->d; |
|
} |
|
} |
|
} |
|
} |
|
for ( rt = r; rt; rt = NEXT(rt) ) { |
|
dl = (DL)BDY(rt); d = dl->d; |
|
ri = 0; |
|
for ( k = nv-1; k >= 0; k-- ) { |
|
STOQ(d[k],q); |
|
MKNODE(r1,q,ri); ri = r1; |
|
} |
|
MKNODE(r1,0,ri); MKLIST(l,r1); |
|
BDY(rt) = (pointer)l; |
|
} |
|
return r; |
|
} |