=================================================================== RCS file: /home/cvs/OpenXM_contrib2/asir2000/engine/nd.c,v retrieving revision 1.53 retrieving revision 1.56 diff -u -p -r1.53 -r1.56 --- OpenXM_contrib2/asir2000/engine/nd.c 2003/08/29 07:37:30 1.53 +++ OpenXM_contrib2/asir2000/engine/nd.c 2003/09/04 08:35:09 1.56 @@ -1,4 +1,4 @@ -/* $OpenXM: OpenXM_contrib2/asir2000/engine/nd.c,v 1.52 2003/08/27 02:21:16 noro Exp $ */ +/* $OpenXM: OpenXM_contrib2/asir2000/engine/nd.c,v 1.55 2003/09/03 07:33:35 noro Exp $ */ #include "ca.h" #include "inline.h" @@ -120,7 +120,7 @@ static int nm_adv; static int nmv_adv; static int nd_dcomp; -extern int Top,Reverse,dp_nelim; +extern int Top,Reverse,dp_nelim,do_weyl; /* fundamental macros */ #define TD(d) (d[0]) @@ -185,6 +185,7 @@ if(!(r)){NEWND_pairs(r);(c)=(r);}else{NEWND_pairs(NEXT /* macro for increasing pointer to NMV */ #define NMV_ADV(m) (m = (NMV)(((char *)m)+nmv_adv)) +#define NMV_PREV(m) (m = (NMV)(((char *)m)-nmv_adv)) /* external functions */ void GC_gcollect(); @@ -233,6 +234,7 @@ INLINE int ndl_block_compare(unsigned int *d1,unsigned INLINE int ndl_equal(unsigned int *d1,unsigned int *d2); INLINE void ndl_copy(unsigned int *d1,unsigned int *d2); INLINE void ndl_add(unsigned int *d1,unsigned int *d2,unsigned int *d); +INLINE void ndl_addto(unsigned int *d1,unsigned int *d2); INLINE void ndl_sub(unsigned int *d1,unsigned int *d2,unsigned int *d); INLINE int ndl_hash_value(unsigned int *d); @@ -288,10 +290,12 @@ ND nd_add_q(ND p1,ND p2); INLINE int nd_length(ND p); /* NDV functions */ +ND weyl_ndv_mul_nm(int mod,NM m0,NDV p); +void weyl_mul_nm_nmv(int n,int mod,NM m0,NMV m1,NM *tab,int tlen); void ndv_mul_c(int mod,NDV p,int mul); void ndv_mul_c_q(NDV p,Q mul); void ndv_realloc(NDV p,int obpe,int oadv,EPOS oepos); -ND ndv_mul_nm(int mod,NDV p,NM m0); +ND ndv_mul_nm(int mod,NM m0,NDV p); void ndv_dehomogenize(NDV p,struct order_spec *spec); void ndv_removecont(int mod,NDV p); void ndv_print(NDV p); @@ -661,6 +665,31 @@ INLINE void ndl_add(unsigned int *d1,unsigned int *d2, #endif } +/* d1 += d2 */ +INLINE void ndl_addto(unsigned int *d1,unsigned int *d2) +{ + int i; + +#if 1 + switch ( nd_wpd ) { + case 2: + TD(d1) += TD(d2); + d1[1] += d2[1]; + break; + case 3: + TD(d1) += TD(d2); + d1[1] += d2[1]; + d1[2] += d2[2]; + break; + default: + for ( i = 0; i < nd_wpd; i++ ) d1[i] += d2[i]; + break; + } +#else + for ( i = 0; i < nd_wpd; i++ ) d1[i] += d2[i]; +#endif +} + INLINE void ndl_sub(unsigned int *d1,unsigned int *d2,unsigned int *d) { int i; @@ -1077,7 +1106,7 @@ int nd_nf(int mod,ND g,NDV *ps,int full,ND *rp) chsgnq(cg,&CQ(mul)); nd_mul_c_q(d,cred); nd_mul_c_q(g,cred); } - g = nd_add(mod,g,ndv_mul_nm(mod,p,mul)); + g = nd_add(mod,g,ndv_mul_nm(mod,mul,p)); sugar = MAX(sugar,SG(p)+TD(DL(mul))); if ( !mod && hmag && g && ((double)(p_mag((P)HCQ(g))) > hmag) ) { nd_removecont2(d,g); @@ -1160,7 +1189,7 @@ int nd_nf_pbucket(int mod,ND g,NDV *ps,int full,ND *rp g = bucket->body[hindex]; gmag = (double)p_mag((P)HCQ(g)); } - red = ndv_mul_nm(mod,p,mul); + red = ndv_mul_nm(mod,mul,p); bucket->body[hindex] = nd_remove_head(g); red = nd_remove_head(red); add_pbucket(mod,bucket,red); @@ -1249,7 +1278,7 @@ int nd_nf_direct(int mod,ND g,BaseSet base,int full,ND chsgnq(cg,&CQ(mul)); nd_mul_c_q(d,cred); nd_mul_c_q(g,cred); } - g = nd_add(mod,g,ndv_mul_nm(mod,p,mul)); + g = nd_add(mod,g,ndv_mul_nm(mod,mul,p)); sugar = MAX(sugar,SG(p)+TD(DL(mul))); if ( !mod && hmag && g && ((double)(p_mag((P)HCQ(g))) > hmag) ) { nd_removecont2(d,g); @@ -1605,11 +1634,9 @@ NODE nd_reduceall(int m,NODE f) for ( n = 0, t = f; t; t = NEXT(t), n++ ); ps = (NDV *)ALLOCA(n*sizeof(NDV)); bound = (unsigned int **)ALLOCA(n*sizeof(unsigned int *)); - for ( i = 0, t = f; i < n; i++, t = NEXT(t) ) { - ps[i] = (NDV)BDY(t); - bound[i] = ndv_compute_bound(ps[i]); - } + for ( i = 0, t = f; i < n; i++, t = NEXT(t) ) ps[i] = (NDV)BDY(t); qsort(ps,n,sizeof(NDV),(int (*)(const void *,const void *))ndv_compare); + for ( i = 0; i < n; i++ ) bound[i] = ndv_compute_bound(ps[i]); base.ps = (NDV *)ALLOCA((n-1)*sizeof(NDV)); base.bound = (unsigned int **)ALLOCA((n-1)*sizeof(unsigned int *)); base.len = n-1; @@ -1661,15 +1688,19 @@ ND_pairs update_pairs( ND_pairs d, NODE /* of index */ d1 = nd_newpairs(g,t); d1 = crit_M(d1); d1 = crit_F(d1); - prev = 0; cur = head = d1; - while ( cur ) { - if ( crit_2( cur->i1,cur->i2 ) ) { - remove = cur; - if ( !prev ) head = cur = NEXT(cur); - else cur = NEXT(prev) = NEXT(cur); - FREENDP(remove); - } else { - prev = cur; cur = NEXT(cur); + if ( do_weyl ) + head = d1; + else { + prev = 0; cur = head = d1; + while ( cur ) { + if ( crit_2( cur->i1,cur->i2 ) ) { + remove = cur; + if ( !prev ) head = cur = NEXT(cur); + else cur = NEXT(prev) = NEXT(cur); + FREENDP(remove); + } else { + prev = cur; cur = NEXT(cur); + } } } if ( !d ) @@ -1733,8 +1764,6 @@ ND_pairs crit_B( ND_pairs d, int s ) return head; } -/* XXX : check is necessary */ - ND_pairs crit_M( ND_pairs d1 ) { ND_pairs e,d2,d3,dd,p; @@ -2133,10 +2162,7 @@ void dltondl(int n,DL dl,unsigned int *r) for ( j = 0; j < l; j++ ) r[j+1] = ndl_weight_mask(r,j); } else { - if ( nd_isrlex ) - for ( i = 0; i < n; i++ ) PUT_EXP(r,n-1-i,d[i]); - else - for ( i = 0; i < n; i++ ) PUT_EXP(r,i,d[i]); + for ( i = 0; i < n; i++ ) PUT_EXP(r,i,d[i]); TD(r) = ndl_weight(r); } } @@ -2166,12 +2192,7 @@ DL ndltodl(int n,unsigned int *ndl) s += ord_l; } } else { - if ( nd_isrlex ) - for ( i = 0; i < n; i++ ) - d[i] = GET_EXP(ndl,n-1-i); - else - for ( i = 0; i < n; i++ ) - d[i] = GET_EXP(ndl,i); + for ( i = 0; i < n; i++ ) d[i] = GET_EXP(ndl,i); } return dl; } @@ -2244,10 +2265,7 @@ void ndl_print(unsigned int *dl) printf(s==n-1?"%d":"%d,",GET_EXP(dl,s)); } } else { - if ( nd_isrlex ) - for ( i = 0; i < n; i++ ) printf(i==n-1?"%d":"%d,",GET_EXP(dl,n-1-i)); - else - for ( i = 0; i < n; i++ ) printf(i==n-1?"%d":"%d,",GET_EXP(dl,i)); + for ( i = 0; i < n; i++ ) printf(i==n-1?"%d":"%d,",GET_EXP(dl,i)); } printf(">>"); } @@ -2550,9 +2568,16 @@ void nd_setup_parameters() { nd_exporigin = nd_get_exporigin(nd_ord); nd_wpd = nd_exporigin+elen; nd_epos = (EPOS)MALLOC_ATOMIC(nd_nvar*sizeof(struct oEPOS)); - for ( i = 0; i < nd_nvar; i++ ) { - nd_epos[i].i = nd_exporigin + i/nd_epw; - nd_epos[i].s = (nd_epw-(i%nd_epw)-1)*nd_bpe; + if ( nd_isrlex ) { + for ( i = 0; i < nd_nvar; i++ ) { + nd_epos[i].i = nd_exporigin + (nd_nvar-1-i)/nd_epw; + nd_epos[i].s = (nd_epw-((nd_nvar-1-i)%nd_epw)-1)*nd_bpe; + } + } else { + for ( i = 0; i < nd_nvar; i++ ) { + nd_epos[i].i = nd_exporigin + i/nd_epw; + nd_epos[i].s = (nd_epw-(i%nd_epw)-1)*nd_bpe; + } } if ( nd_bpe < 32 ) { nd_mask0 = (1<i1,DL(m)) ) return 0; - t1 = ndv_mul_nm(mod,p1,m); + if ( ndl_check_bound2(p->i1,DL(m)) ) + return 0; + t1 = ndv_mul_nm(mod,m,p1); if ( mod ) CM(m) = mod-HCM(p1); else chsgnq(HCQ(p1),&CQ(m)); ndl_sub(lcm,HDL(p2),DL(m)); @@ -2761,7 +2781,7 @@ int nd_sp(int mod,int trace,ND_pairs p,ND *rp) nd_free(t1); return 0; } - t2 = ndv_mul_nm(mod,p2,m); + t2 = ndv_mul_nm(mod,m,p2); *rp = nd_add(mod,t1,t2); FREENM(m); return 1; @@ -2792,8 +2812,152 @@ void ndv_mul_c_q(NDV p,Q mul) } } -ND ndv_mul_nm(int mod,NDV p,NM m0) +ND weyl_ndv_mul_nm(int mod,NM m0,NDV p) { + int n2,i,j,l,n,tlen; + unsigned int *d0; + NM *tab,*psum; + ND s,r; + NM t; + NMV m1; + + if ( !p ) return 0; + n = NV(p); n2 = n>>1; + d0 = DL(m0); + l = LEN(p); + for ( i = 0, tlen = 1; i < n2; i++ ) tlen *= (GET_EXP(d0,n2+i)+1); + tab = (NM *)ALLOCA(tlen*sizeof(NM)); + psum = (NM *)ALLOCA(tlen*sizeof(NM)); + for ( i = 0; i < tlen; i++ ) psum[i] = 0; + m1 = (NMV)(((char *)BDY(p))+nmv_adv*(l-1)); + for ( i = l-1; i >= 0; i--, NMV_PREV(m1) ) { + /* m0(NM) * m1(NMV) => tab(NM) */ + weyl_mul_nm_nmv(n,mod,m0,m1,tab,tlen); + for ( j = 0; j < tlen; j++ ) { + if ( tab[j] ) { + NEXT(tab[j]) = psum[j]; psum[j] = tab[j]; + } + } + } + for ( i = tlen-1, r = 0; i >= 0; i-- ) + if ( psum[i] ) { + for ( j = 0, t = psum[i]; t; t = NEXT(t), j++ ); + MKND(n,psum[i],j,s); + r = nd_add(mod,r,s); + } + if ( r ) SG(r) = SG(p)+TD(d0); + return r; +} + +/* product of monomials */ +/* XXX block order is not handled correctly */ + +void weyl_mul_nm_nmv(int n,int mod,NM m0,NMV m1,NM *tab,int tlen) { + int i,n2,j,s,curlen,homo,h,a,b,k,l,u,min; + unsigned int *d0,*d1,*d,*dt,*ctab; + Q *ctab_q; + Q q,q1; + unsigned int c0,c1,c; + NM *p; + NM m,t; + + for ( i = 0; i < tlen; i++ ) tab[i] = 0; + if ( !m0 || !m1 ) return; + d0 = DL(m0); d1 = DL(m1); n2 = n>>1; + NEWNM(m); d = DL(m); + if ( mod ) { + c0 = CM(m0); c1 = CM(m1); DMAR(c0,c1,0,mod,c); CM(m) = c; + } else + mulq(CQ(m0),CQ(m1),&CQ(m)); + for ( i = 0; i < nd_wpd; i++ ) d[i] = 0; + homo = n&1 ? 1 : 0; + if ( homo ) { + /* offset of h-degree */ + h = GET_EXP(d0,n-1)+GET_EXP(d1,n-1); + PUT_EXP(DL(m),n-1,h); + TD(DL(m)) = h; + /* XXX other weights */ + } + tab[0] = m; + curlen = 1; + NEWNM(m); d = DL(m); + for ( i = 0; i < n2; i++ ) { + a = GET_EXP(d0,i); b = GET_EXP(d1,n2+i); + k = GET_EXP(d0,n2+i); l = GET_EXP(d1,i); + /* xi^a*(Di^k*xi^l)*Di^b */ + a += l; b += k; + s = MUL_WEIGHT(a,i)+MUL_WEIGHT(b,n2+i); + if ( !k || !l ) { + for ( j = 0; j < curlen; j++ ) + if ( t = tab[j] ) { + dt = DL(t); + PUT_EXP(dt,i,a); PUT_EXP(dt,n2+i,b); TD(dt) += s; + /* XXX other weights */ + } + curlen *= k+1; + continue; + } + min = MIN(k,l); + if ( mod ) { + ctab = (unsigned int *)ALLOCA((min+1)*sizeof(unsigned int)); + mkwcm(k,l,mod,ctab); + } else { + ctab_q = (Q *)ALLOCA((min+1)*sizeof(Q)); + mkwc(k,l,ctab_q); + } + p = tab+curlen; + for ( j = 1; j <= min; j++ ) { + for ( u = 0; u < nd_wpd; u++ ) d[u] = 0; + PUT_EXP(d,i,a-j); PUT_EXP(d,n2+i,b-j); + h = MUL_WEIGHT(a-j,i)+MUL_WEIGHT(b-j,n2+i); + if ( homo ) { + TD(d) = s; + PUT_EXP(d,n-1,s-h); + } else TD(d) = h; + /* XXX other weights */ + if ( mod ) c = ctab[j]; + else q = ctab_q[j]; + for ( u = 0; u < curlen; u++, p++ ) { + if ( tab[u] ) { + NEWNM(t); + ndl_add(DL(tab[u]),d,DL(t)); + if ( mod ) { + c0 = CM(tab[u]); DMAR(c0,c,0,mod,c1); CM(t) = c1; + } else + mulq(CQ(tab[u]),q,&CQ(t)); + *p = t; + } + } + } + /* destructive for j = 0 */ + for ( u = 0; u < nd_wpd; u++ ) d[u] = 0; + PUT_EXP(d,i,a); PUT_EXP(d,n2+i,b); + h = MUL_WEIGHT(a,i)+MUL_WEIGHT(b,n2+i); + if ( homo ) { + TD(d) = s; + PUT_EXP(d,n-1,s-h); + } else TD(d) = h; + /* XXX other weights */ + if ( mod ) c = ctab[0]; + else q = ctab_q[0]; + p = tab; + for ( u = 0; u < curlen; u++, p++ ) { + if ( tab[u] ) { + ndl_addto(DL(tab[u]),d); + if ( mod ) { + c0 = CM(tab[u]); DMAR(c0,c,0,mod,c1); CM(tab[u]) = c1; + } else { + mulq(CQ(tab[u]),q,&q1); CQ(tab[u]) = q1; + } + } + } + curlen *= k+1; + } + FREENM(m); +} + +ND ndv_mul_nm(int mod,NM m0,NDV p) +{ NM mr,mr0; NMV m; unsigned int *d,*dt,*dm; @@ -2802,6 +2966,8 @@ ND ndv_mul_nm(int mod,NDV p,NM m0) ND r; if ( !p ) return 0; + else if ( do_weyl ) + return weyl_ndv_mul_nm(mod,m0,p); else { n = NV(p); m = BDY(p); d = DL(m0); @@ -2838,7 +3004,6 @@ void ndv_realloc(NDV p,int obpe,int oadv,EPOS oepos) int len,i,k; #define NMV_OPREV(m) (m = (NMV)(((char *)m)-oadv)) -#define NMV_PREV(m) (m = (NMV)(((char *)m)-nmv_adv)) if ( p ) { m = BDY(p); len = LEN(p);