=================================================================== RCS file: /home/cvs/OpenXM_contrib2/asir2000/engine/nd.c,v retrieving revision 1.52 retrieving revision 1.57 diff -u -p -r1.52 -r1.57 --- OpenXM_contrib2/asir2000/engine/nd.c 2003/08/27 02:21:16 1.52 +++ OpenXM_contrib2/asir2000/engine/nd.c 2003/09/05 05:02:53 1.57 @@ -1,4 +1,4 @@ -/* $OpenXM: OpenXM_contrib2/asir2000/engine/nd.c,v 1.51 2003/08/27 01:53:29 noro Exp $ */ +/* $OpenXM: OpenXM_contrib2/asir2000/engine/nd.c,v 1.56 2003/09/04 08:35:09 noro Exp $ */ #include "ca.h" #include "inline.h" @@ -109,7 +109,7 @@ static ND _nd_free_list; static ND_pairs _ndp_free_list; static NDV *nd_ps; -static NDV *nd_psq; +static NDV *nd_ps_trace; static RHist *nd_psh; static int nd_psn,nd_pslen; @@ -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(); @@ -225,6 +226,7 @@ NODE nd_gb_trace(int m); /* ndl functions */ int ndl_weight(unsigned int *d); int ndl_weight_mask(unsigned int *d,int i); +void ndl_set_blockweight(unsigned int *d); void ndl_dehomogenize(unsigned int *p); void ndl_reconstruct(int obpe,EPOS oepos,unsigned int *d,unsigned int *r); INLINE int ndl_reducible(unsigned int *d1,unsigned int *d2); @@ -233,15 +235,16 @@ 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); /* normal forms */ INLINE int nd_find_reducer(ND g); INLINE int nd_find_reducer_direct(ND g,NDV *ps,int len); -int nd_sp(int mod,ND_pairs p,ND *nf); -int nd_nf(int mod,ND g,int full,ND *nf); -int nd_nf_pbucket(int mod,ND g,int full,ND *nf); +int nd_sp(int mod,int trace,ND_pairs p,ND *nf); +int nd_nf(int mod,ND g,NDV *ps,int full,ND *nf); +int nd_nf_pbucket(int mod,ND g,NDV *ps,int full,ND *nf); int nd_nf_direct(int mod,ND g,BaseSet base,int full,ND *rp); /* finalizers */ @@ -271,6 +274,7 @@ void nd_reconstruct_direct(int mod,NDV *ps,int len); void nd_setup(int mod,int trace,NODE f); void nd_setup_parameters(); BlockMask nd_create_blockmask(struct order_spec *ord); +EPOS nd_create_epos(struct order_spec *ord); int nd_get_exporigin(struct order_spec *ord); /* ND functions */ @@ -281,17 +285,18 @@ ND nd_remove_head(ND p); int nd_length(ND p); void nd_append_red(unsigned int *d,int i); unsigned int *ndv_compute_bound(NDV p); -unsigned int *dp_compute_bound(DP p); ND nd_copy(ND p); ND nd_add(int mod,ND p1,ND p2); 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); @@ -536,6 +541,16 @@ void ndl_lcm(unsigned int *d1,unsigned *d2,unsigned in } } +void ndl_set_blockweight(unsigned int *d) { + int l,j; + + if ( nd_blockmask ) { + l = nd_blockmask->n; + for ( j = 0; j < l; j++ ) + d[j+1] = ndl_weight_mask(d,j); + } +} + int ndl_weight(unsigned int *d) { unsigned int t,u; @@ -661,6 +676,31 @@ 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; @@ -1038,7 +1078,7 @@ ND nd_add_q(ND p1,ND p2) } /* ret=1 : success, ret=0 : overflow */ -int nd_nf(int mod,ND g,int full,ND *rp) +int nd_nf(int mod,ND g,NDV *ps,int full,ND *rp) { ND d; NM m,mrd,tail; @@ -1068,17 +1108,16 @@ int nd_nf(int mod,ND g,int full,ND *rp) nd_free(g); nd_free(d); return 0; } + p = ps[index]; if ( mod ) { - p = nd_ps[index]; c1 = invm(HCM(p),mod); c2 = mod-HCM(g); DMAR(c1,c2,0,mod,c); CM(mul) = c; } else { - p = nd_psq[index]; igcd_cofactor(HCQ(g),HCQ(p),&gcd,&cg,&cred); 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); @@ -1106,7 +1145,7 @@ int nd_nf(int mod,ND g,int full,ND *rp) return 1; } -int nd_nf_pbucket(int mod,ND g,int full,ND *rp) +int nd_nf_pbucket(int mod,ND g,NDV *ps,int full,ND *rp) { int hindex,index; NDV p; @@ -1149,12 +1188,11 @@ int nd_nf_pbucket(int mod,ND g,int full,ND *rp) *rp = 0; return 0; } + p = ps[index]; if ( mod ) { - p = nd_ps[index]; c1 = invm(HCM(p),mod); c2 = mod-HCM(g); DMAR(c1,c2,0,mod,c); CM(mul) = c; } else { - p = nd_psq[index]; igcd_cofactor(HCQ(g),HCQ(p),&gcd,&cg,&cred); chsgnq(cg,&CQ(mul)); nd_mul_c_q(d,cred); @@ -1162,7 +1200,7 @@ int nd_nf_pbucket(int mod,ND g,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); @@ -1251,7 +1289,7 @@ 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); @@ -1301,7 +1339,7 @@ int nd_check_candidate(NODE input,NODE cand) for ( t = input; t; t = NEXT(t) ) { again: d = dptond(0,(DP)BDY(t)); - stat = nd_nf(0,d,0,&nf); + stat = nd_nf(0,d,nd_ps,0,&nf); if ( !stat ) { nd_reconstruct(0,0,0); goto again; @@ -1494,16 +1532,16 @@ again: sugar = SG(l); fprintf(asir_out,"%d",sugar); } - stat = nd_sp(m,l,&h); + stat = nd_sp(m,0,l,&h); if ( !stat ) { NEXT(l) = d; d = l; d = nd_reconstruct(m,0,d); goto again; } #if USE_GEOBUCKET - stat = m?nd_nf_pbucket(m,h,!Top,&nf):nd_nf(m,h,!Top,&nf); + stat = m?nd_nf_pbucket(m,h,nd_ps,!Top,&nf):nd_nf(m,h,nd_ps,!Top,&nf); #else - stat = nd_nf(m,h,!Top,&nf); + stat = nd_nf(m,h,nd_ps,!Top,&nf); #endif if ( !stat ) { NEXT(l) = d; d = l; @@ -1512,7 +1550,7 @@ again: } else if ( nf ) { if ( checkonly ) return 0; printf("+"); fflush(stdout); - nh = m?nd_newps(m,nf,0):nd_newps(m,0,nf); + nh = nd_newps(m,nf,0); d = update_pairs(d,g,nh); g = update_base(g,nh); FREENDP(l); @@ -1521,10 +1559,7 @@ again: FREENDP(l); } } - if ( m ) - for ( t = g; t; t = NEXT(t) ) BDY(t) = (pointer)nd_ps[(int)BDY(t)]; - else - for ( t = g; t; t = NEXT(t) ) BDY(t) = (pointer)nd_psq[(int)BDY(t)]; + for ( t = g; t; t = NEXT(t) ) BDY(t) = (pointer)nd_ps[(int)BDY(t)]; return g; } @@ -1549,16 +1584,16 @@ again: sugar = SG(l); fprintf(asir_out,"%d",sugar); } - stat = nd_sp(m,l,&h); + stat = nd_sp(m,0,l,&h); if ( !stat ) { NEXT(l) = d; d = l; d = nd_reconstruct(m,1,d); goto again; } #if USE_GEOBUCKET - stat = nd_nf_pbucket(m,h,!Top,&nf); + stat = nd_nf_pbucket(m,h,nd_ps,!Top,&nf); #else - stat = nd_nf(m,h,!Top,&nf); + stat = nd_nf(m,h,nd_ps,!Top,&nf); #endif if ( !stat ) { NEXT(l) = d; d = l; @@ -1566,8 +1601,8 @@ again: goto again; } else if ( nf ) { /* overflow does not occur */ - nd_sp(0,l,&h); - nd_nf(0,h,!Top,&nfq); + nd_sp(0,1,l,&h); + nd_nf(0,h,nd_ps_trace,!Top,&nfq); if ( nfq ) { printf("+"); fflush(stdout); nh = nd_newps(m,nf,nfq); @@ -1584,7 +1619,7 @@ again: FREENDP(l); } for ( t = g; t; t = NEXT(t) ) - BDY(t) = (pointer)nd_psq[(int)BDY(t)]; + BDY(t) = (pointer)nd_ps_trace[(int)BDY(t)]; return g; } @@ -1610,11 +1645,9 @@ 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; @@ -1666,15 +1699,19 @@ 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 ) @@ -1738,8 +1775,6 @@ 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; @@ -1899,35 +1934,31 @@ int nd_newps(int mod,ND a,ND aq) RHist r; NDV b; + if ( aq ) { + /* trace lifting */ + if ( !rem(NM(HCQ(aq)),mod) ) + return -1; + } if ( nd_psn == nd_pslen ) { nd_pslen *= 2; nd_ps = (NDV *)REALLOC((char *)nd_ps,nd_pslen*sizeof(NDV)); - nd_psq = (NDV *)REALLOC((char *)nd_psq,nd_pslen*sizeof(NDV)); + nd_ps_trace = (NDV *)REALLOC((char *)nd_ps_trace,nd_pslen*sizeof(NDV)); nd_psh = (RHist *)REALLOC((char *)nd_psh,nd_pslen*sizeof(RHist)); nd_bound = (unsigned int **) REALLOC((char *)nd_bound,nd_pslen*sizeof(unsigned int *)); } - if ( a && aq ) { - /* trace lifting */ - if ( !rem(NM(HCQ(aq)),mod) ) - return -1; - } NEWRHist(r); nd_psh[nd_psn] = r; + nd_removecont(mod,a); nd_ps[nd_psn] = ndtondv(mod,a); if ( aq ) { - nd_removecont(0,aq); - nd_psq[nd_psn] = ndtondv(0,aq); - nd_bound[nd_psn] = ndv_compute_bound(nd_psq[nd_psn]); + nd_removecont(0,aq); nd_ps_trace[nd_psn] = ndtondv(0,aq); + nd_bound[nd_psn] = ndv_compute_bound(nd_ps_trace[nd_psn]); SG(r) = SG(aq); ndl_copy(HDL(aq),DL(r)); + nd_free(a); nd_free(aq); + } else { + nd_bound[nd_psn] = ndv_compute_bound(nd_ps[nd_psn]); + SG(r) = SG(a); ndl_copy(HDL(a),DL(r)); + nd_free(a); } - if ( a ) { - nd_removecont(mod,a); - nd_ps[nd_psn] = ndtondv(mod,a); - if ( !aq ) { - nd_bound[nd_psn] = ndv_compute_bound(nd_ps[nd_psn]); - SG(r) = SG(a); ndl_copy(HDL(a),DL(r)); - } - } - nd_free(a); nd_free(aq); return nd_psn++; } @@ -1938,23 +1969,26 @@ void nd_setup(int mod,int trace,NODE f) unsigned int *d; RHist r; NDV a; + MP t; nd_found = 0; nd_notfirst = 0; nd_create = 0; nd_psn = length(f); nd_pslen = 2*nd_psn; nd_ps = (NDV *)MALLOC(nd_pslen*sizeof(NDV)); - nd_psq = (NDV *)MALLOC(nd_pslen*sizeof(NDV)); + nd_ps_trace = (NDV *)MALLOC(nd_pslen*sizeof(NDV)); nd_psh = (RHist *)MALLOC(nd_pslen*sizeof(RHist)); nd_bound = (unsigned int **)MALLOC(nd_pslen*sizeof(unsigned int *)); - for ( max = 0, i = 0, s = f; i < nd_psn; i++, s = NEXT(s) ) { - nd_bound[i] = d = dp_compute_bound((DP)BDY(s)); - for ( j = 0; j < nd_nvar; j++ ) - max = MAX(d[j],max); - } + if ( !nd_red ) nd_red = (RHist *)MALLOC(REDTAB_LEN*sizeof(RHist)); bzero(nd_red,REDTAB_LEN*sizeof(RHist)); + for ( max = 0, s = f; s; s = NEXT(s) ) + for ( t = BDY((DP)BDY(s)); t; t = NEXT(t) ) { + d = t->dl->d; + for ( j = 0; j < nd_nvar; j++ ) max = MAX(d[j],max); + } + if ( max < 2 ) nd_bpe = 2; else if ( max < 4 ) nd_bpe = 4; else if ( max < 64 ) nd_bpe = 6; @@ -1968,15 +2002,12 @@ void nd_setup(int mod,int trace,NODE f) NEWRHist(r); a = dptondv(mod,(DP)BDY(f)); ndv_removecont(mod,a); SG(r) = HTD(a); ndl_copy(HDL(a),DL(r)); - + nd_ps[i] = a; if ( trace ) { - nd_ps[i] = a; a = dptondv(0,(DP)BDY(f)); ndv_removecont(0,a); - nd_psq[i] = a; - } else { - if ( mod ) nd_ps[i] = a; - else nd_psq[i] = a; + nd_ps_trace[i] = a; } + nd_bound[i] = ndv_compute_bound(a); nd_psh[i] = r; } } @@ -2122,7 +2153,7 @@ void nd_gr_trace(LIST f,LIST v,int trace,int homo,stru void dltondl(int n,DL dl,unsigned int *r) { unsigned int *d; - int i,j,l,s,ord_l,ord_o; + int i,j,l,s,ord_l; struct order_pair *op; d = dl->d; @@ -2131,24 +2162,14 @@ void dltondl(int n,DL dl,unsigned int *r) l = nd_blockmask->n; op = nd_blockmask->order_pair; for ( j = 0, s = 0; j < l; j++ ) { - ord_o = op[j].order; ord_l = op[j].length; - if ( !ord_o ) - for ( i = 0; i < ord_l; i++ ) - PUT_EXP(r,s+ord_l-i-1,d[s+i]); - else - for ( i = 0; i < ord_l; i++ ) - PUT_EXP(r,s+i,d[s+i]); - s += ord_l; + for ( i = 0; i < ord_l; i++, s++ ) PUT_EXP(r,s,d[s]); } TD(r) = ndl_weight(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); } } @@ -2157,7 +2178,7 @@ DL ndltodl(int n,unsigned int *ndl) { DL dl; int *d; - int i,j,l,s,ord_l,ord_o; + int i,j,l,s,ord_l; struct order_pair *op; NEWDL(dl,n); @@ -2167,23 +2188,11 @@ DL ndltodl(int n,unsigned int *ndl) l = nd_blockmask->n; op = nd_blockmask->order_pair; for ( j = 0, s = 0; j < l; j++ ) { - ord_o = op[j].order; ord_l = op[j].length; - if ( !ord_o ) - for ( i = 0; i < ord_l; i++ ) - d[s+i] = GET_EXP(ndl,s+ord_l-i-1); - else - for ( i = 0; i < ord_l; i++ ) - d[s+i] = GET_EXP(ndl,s+i); - s += ord_l; + for ( i = 0; i < ord_l; i++, s++ ) d[s] = GET_EXP(ndl,s); } } 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; } @@ -2237,7 +2246,7 @@ DP ndtodp(int mod,ND p) void ndl_print(unsigned int *dl) { int n; - int i,j,l,ord_o,ord_l,s,s0; + int i,j,l,ord_l,s,s0; struct order_pair *op; n = nd_nvar; @@ -2246,20 +2255,12 @@ void ndl_print(unsigned int *dl) l = nd_blockmask->n; op = nd_blockmask->order_pair; for ( j = 0, s = s0 = 0; j < l; j++ ) { - ord_o = op[j].order; ord_l = op[j].length; - if ( !ord_o ) - for ( i = 0, s0 += ord_l; i < ord_l; i++, s++ ) - printf(s==n-1?"%d":"%d,",GET_EXP(dl,s0-i-1)); - else - for ( i = 0; i < ord_l; i++, s++ ) - printf(s==n-1?"%d":"%d,",GET_EXP(dl,s)); + for ( i = 0; i < ord_l; i++, s++ ) + 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(">>"); } @@ -2490,36 +2491,11 @@ void nd_append_red(unsigned int *d,int i) nd_red[h] = m; } -unsigned int *dp_compute_bound(DP p) -{ - unsigned int *d,*d1,*d2,*t; - MP m; - int i,l; - - if ( !p ) - return 0; - d1 = (unsigned int *)ALLOCA(nd_nvar*sizeof(unsigned int)); - d2 = (unsigned int *)ALLOCA(nd_nvar*sizeof(unsigned int)); - m = BDY(p); - d = DL(m)->d; - for ( i = 0; i < nd_nvar; i++ ) d1[i] = d[i]; - for ( m = NEXT(BDY(p)); m; m = NEXT(m) ) { - d = DL(m)->d; - for ( i = 0; i < nd_nvar; i++ ) - d2[i] = d[i] > d1[i] ? d[i] : d1[i]; - t = d1; d1 = d2; d2 = t; - } - l = (nd_nvar+31); - t = (unsigned int *)MALLOC_ATOMIC(l*sizeof(unsigned int)); - for ( i = 0; i < nd_nvar; i++ ) t[i] = d1[i]; - for ( ; i < l; i++ ) t[i] = 0; - return t; -} - unsigned int *ndv_compute_bound(NDV p) { unsigned int *d1,*d2,*t; - int i,l,len; + unsigned int u; + int i,j,k,l,len,ind; NMV m; if ( !p ) @@ -2534,8 +2510,13 @@ unsigned int *ndv_compute_bound(NDV p) } l = nd_nvar+31; t = (unsigned int *)MALLOC_ATOMIC(l*sizeof(unsigned int)); - for ( i = 0; i < nd_nvar; i++ ) t[i] = GET_EXP(d1,i); - for ( ; i < l; i++ ) t[i] = 0; + for ( i = nd_exporigin, ind = 0; i < nd_wpd; i++ ) { + u = d1[i]; + k = (nd_epw-1)*nd_bpe; + for ( j = 0; j < nd_epw; j++, k -= nd_bpe, ind++ ) + t[ind] = (u>>k)&nd_mask0; + } + for ( ; ind < l; ind++ ) t[ind] = 0; return t; } @@ -2554,18 +2535,15 @@ int nd_get_exporigin(struct order_spec *ord) } void nd_setup_parameters() { - int i,n,elen; + int i,j,n,elen,ord_o,ord_l,l,s; + struct order_pair *op; nd_epw = (sizeof(unsigned int)*8)/nd_bpe; elen = nd_nvar/nd_epw+(nd_nvar%nd_epw?1:0); 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_bpe < 32 ) { nd_mask0 = (1<= 0; i-- ) ndv_realloc(nd_ps[i],obpe,oadv,oepos); - if ( !mod || trace ) - for ( i = nd_psn-1; i >= 0; i-- ) ndv_realloc(nd_psq[i],obpe,oadv,oepos); + for ( i = nd_psn-1; i >= 0; i-- ) ndv_realloc(nd_ps[i],obpe,oadv,oepos); + if ( trace ) + for ( i = nd_psn-1; i >= 0; i-- ) + ndv_realloc(nd_ps_trace[i],obpe,oadv,oepos); s0 = 0; for ( t = d; t; t = NEXT(t) ) { NEXTND_pairs(s0,s); @@ -2682,7 +2661,7 @@ void nd_reconstruct_direct(int mod,NDV *ps,int len) void ndl_reconstruct(int obpe,EPOS oepos,unsigned int *d,unsigned int *r) { - int n,i,ei,oepw,omask0,j,s,ord_l,ord_o,l; + int n,i,ei,oepw,omask0,j,s,ord_l,l; struct order_pair *op; #define GET_EXP_OLD(d,a) (((d)[oepos[a].i]>>oepos[a].s)&omask0) #define PUT_EXP_OLD(r,a,e) ((r)[oepos[a].i] |= ((e)<i1]; p2 = nd_ps[p->i2]; + if ( trace ) { + p1 = nd_ps_trace[p->i1]; p2 = nd_ps_trace[p->i2]; } else { - p1 = nd_psq[p->i1]; p2 = nd_psq[p->i2]; + p1 = nd_ps[p->i1]; p2 = nd_ps[p->i2]; } lcm = LCM(p); NEWNM(m); CQ(m) = HCQ(p2); ndl_sub(lcm,HDL(p1),DL(m)); - if ( ndl_check_bound2(p->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)); @@ -2773,7 +2739,7 @@ int nd_sp(int mod,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; @@ -2804,8 +2770,142 @@ 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; + if ( nd_blockmask ) ndl_set_blockweight(DL(m)); + } + tab[0] = m; + NEWNM(m); d = DL(m); + for ( i = 0, curlen = 1; 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; + if ( nd_blockmask ) ndl_set_blockweight(dt); + } + 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); + } + for ( j = min; j >= 0; 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; + if ( nd_blockmask ) ndl_set_blockweight(d); + if ( mod ) c = ctab[j]; + else q = ctab_q[j]; + p = tab+curlen*j; + if ( j == 0 ) { + 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; + } + } + } + } else { + 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; + } + } + } + } + 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; @@ -2814,6 +2914,8 @@ 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); @@ -2850,7 +2952,6 @@ 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); @@ -3075,4 +3176,51 @@ BlockMask nd_create_blockmask(struct order_spec *ord) for ( j = 0; j < l; j++, s++ ) PUT_EXP(t,s,nd_mask0); } return bm; +} + +EPOS nd_create_epos(struct order_spec *ord) +{ + int i,j,l,s,ord_l,ord_o; + EPOS epos; + struct order_pair *op; + + epos = (EPOS)MALLOC_ATOMIC(nd_nvar*sizeof(struct oEPOS)); + switch ( ord->id ) { + case 0: + if ( nd_isrlex ) { + for ( i = 0; i < nd_nvar; i++ ) { + epos[i].i = nd_exporigin + (nd_nvar-1-i)/nd_epw; + epos[i].s = (nd_epw-((nd_nvar-1-i)%nd_epw)-1)*nd_bpe; + } + } else { + for ( i = 0; i < nd_nvar; i++ ) { + epos[i].i = nd_exporigin + i/nd_epw; + epos[i].s = (nd_epw-(i%nd_epw)-1)*nd_bpe; + } + } + break; + case 1: + /* block order */ + l = ord->ord.block.length; + op = ord->ord.block.order_pair; + for ( j = 0, s = 0; j < l; j++ ) { + ord_o = op[j].order; + ord_l = op[j].length; + if ( !ord_o ) + for ( i = 0; i < ord_l; i++ ) { + epos[s+i].i = nd_exporigin + (s+ord_l-i-1)/nd_epw; + epos[s+i].s = (nd_epw-((s+ord_l-i-1)%nd_epw)-1)*nd_bpe; + } + else + for ( i = 0; i < ord_l; i++ ) { + epos[s+i].i = nd_exporigin + (s+i)/nd_epw; + epos[s+i].s = (nd_epw-((s+i)%nd_epw)-1)*nd_bpe; + } + s += ord_l; + } + break; + case 2: + error("nd_create_epos : matrix order is not supported yet."); + } + return epos; }