=================================================================== RCS file: /home/cvs/OpenXM_contrib2/asir2000/engine/nd.c,v retrieving revision 1.239 retrieving revision 1.241 diff -u -p -r1.239 -r1.241 --- OpenXM_contrib2/asir2000/engine/nd.c 2017/09/14 01:34:53 1.239 +++ OpenXM_contrib2/asir2000/engine/nd.c 2017/09/16 08:02:35 1.241 @@ -1,4 +1,4 @@ -/* $OpenXM: OpenXM_contrib2/asir2000/engine/nd.c,v 1.238 2017/08/31 02:36:21 noro Exp $ */ +/* $OpenXM: OpenXM_contrib2/asir2000/engine/nd.c,v 1.240 2017/09/15 01:52:51 noro Exp $ */ #include "nd.h" @@ -65,7 +65,7 @@ static int *nd_gbblock; static NODE nd_nzlist,nd_check_splist; static int nd_splist; static int *nd_sugarweight; -static int nd_f4red,nd_rank0; +static int nd_f4red,nd_rank0,nd_last_nonzero; NumberField get_numberfield(); UINT *nd_det_compute_bound(NDV **dm,int n,int j); @@ -100,7 +100,7 @@ void nd_mul_c_lf(ND p,GZ mul); void ndv_mul_c_lf(NDV p,GZ mul); NODE nd_f4_red_main(int m,ND_pairs sp0,int nsp,UINT *s0vect,int col, NM_ind_pair *rvect,int *rhead,IndArray *imat,int nred,ND_pairs *nz); -NODE nd_f4_red_mod128_main(int m,ND_pairs sp0,int nsp,UINT *s0vect,int col, +NODE nd_f4_red_mod64_main(int m,ND_pairs sp0,int nsp,UINT *s0vect,int col, NM_ind_pair *rvect,int *rhead,IndArray *imat,int nred,ND_pairs *nz); NODE nd_f4_red_lf_main(int m,ND_pairs sp0,int nsp,int trace,UINT *s0vect,int col, NM_ind_pair *rvect,int *rhead,IndArray *imat,int nred); @@ -3228,7 +3228,7 @@ void nd_gr(LIST f,LIST v,int m,int homo,int retdp,int NODE tr,tl1,tl2,tl3,tl4,nzlist; LIST l1,l2,l3,l4,l5; int j; - Q jq,bpe; + Q jq,bpe,last_nonzero; int *perm; EPOS oepos; int obpe,oadv,ompos,cbpe; @@ -3387,7 +3387,9 @@ FINAL: if ( nd_gentrace ) { if ( f4 ) { STOQ(16,bpe); - tr = mknode(4,*rp,(!ishomo&&homo)?ONE:0,BDY(nzlist),bpe); MKLIST(*rp,tr); + STOQ(nd_last_nonzero,last_nonzero); + tr = mknode(5,*rp,(!ishomo&&homo)?ONE:0,BDY(nzlist),bpe,last_nonzero); MKLIST(*rp,tr); + } else { tl1 = reverse_node(tl1); tl2 = reverse_node(tl2); tl3 = reverse_node(tl3); @@ -3840,7 +3842,7 @@ void nd_gr_trace(LIST f,LIST v,int trace,int homo,int } get_eg(&eg1); init_eg(&eg_check); add_eg(&eg_check,&eg0,&eg1); if ( DP_Print ) - fprintf(asir_out,"check=%fsec\n",eg_check.exectime+eg_check.gctime); + fprintf(asir_out,"check=%.3fsec,",eg_check.exectime+eg_check.gctime); /* dp->p */ nd_bpe = cbpe; nd_setup_parameters(nd_nvar,0); @@ -5823,7 +5825,10 @@ int nd_to_vect(int mod,UINT *s0,int n,ND d,UINT *r) } #if defined(__GNUC__) -int nd_to_vect128(int mod,UINT *s0,int n,ND d,U128 *r) + +#define MOD128(a,c,m) ((a)=(((c)!=0||((a)>=(m)))?(((((U128)(c))<<64)+(a))%(m)):(a))) + +int nd_to_vect64(int mod,UINT *s0,int n,ND d,U64 *r) { NM m; UINT *t,*s; @@ -5833,7 +5838,7 @@ int nd_to_vect128(int mod,UINT *s0,int n,ND d,U128 *r) for ( i = 0, s = s0, m = BDY(d); m; m = NEXT(m) ) { t = DL(m); for ( ; !ndl_equal(t,s); s += nd_wpd, i++ ); - r[i] = (U128)CM(m); + r[i] = (U64)CM(m); } for ( i = 0; !r[i]; i++ ); return i; @@ -6235,11 +6240,11 @@ int ndv_reduce_vect(int m,UINT *svect,int col,IndArray } #if defined(__GNUC__) -int ndv_reduce_vect128(int m,U128 *svect,int col,IndArray *imat,NM_ind_pair *rp0,int nred) + +int ndv_reduce_vect64(int m,U64 *svect,U64 *cvect,int col,IndArray *imat,NM_ind_pair *rp0,int nred) { int i,j,k,len,pos,prev; - U64 c,c1; - U128 a; + U64 a,c,c1,c2; IndArray ivect; unsigned char *ivc; unsigned short *ivs; @@ -6249,10 +6254,14 @@ int ndv_reduce_vect128(int m,U128 *svect,int col,IndAr NODE rp; int maxrs; + for ( i = 0; i < col; i++ ) cvect[i] = 0; maxrs = 0; for ( i = 0; i < nred; i++ ) { ivect = imat[i]; - k = ivect->head; svect[k] %= m; + k = ivect->head; + a = svect[k]; c = cvect[k]; + MOD128(a,c,m); + svect[k] = a; cvect[k] = 0; if ( c = svect[k] ) { maxrs = MAX(maxrs,rp0[i]->sugar); c = m-c; redv = nd_ps[rp0[i]->index]; @@ -6263,27 +6272,41 @@ int ndv_reduce_vect128(int m,U128 *svect,int col,IndAr ivc = ivect->index.c; for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) { pos = prev+ivc[j]; c1 = CM(mr); prev = pos; - if ( c1 ) svect[pos] += c1*c; + if ( c1 ) { + c2 = svect[pos]+c1*c; + if ( c2 < svect[pos] ) cvect[pos]++; + svect[pos] = c2; + } } break; case 2: ivs = ivect->index.s; for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) { pos = prev+ivs[j]; c1 = CM(mr); prev = pos; - if ( c1 ) svect[pos] += c1*c; + if ( c1 ) { + c2 = svect[pos]+c1*c; + if ( c2 < svect[pos] ) cvect[pos]++; + svect[pos] = c2; + } } break; case 4: ivi = ivect->index.i; for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) { pos = prev+ivi[j]; c1 = CM(mr); prev = pos; - if ( c1 ) svect[pos] += c1*c; + if ( c1 ) { + c2 = svect[pos]+c1*c; + if ( c2 < svect[pos] ) cvect[pos]++; + svect[pos] = c2; + } } break; } } } - for ( i = 0; i < col; i++ ) svect[i] %= m; + for ( i = 0; i < col; i++ ) { + a = svect[i]; c = cvect[i]; MOD128(a,c,m); svect[i] = a; + } return maxrs; } #endif @@ -6546,7 +6569,7 @@ NDV vect_to_ndv(UINT *vect,int spcol,int col,int *rhea } #if defined(__GNUC__) -NDV vect128o_ndv(U128 *vect,int spcol,int col,int *rhead,UINT *s0vect) +NDV vect64_to_ndv(U64 *vect,int spcol,int col,int *rhead,UINT *s0vect) { int j,k,len; UINT *p; @@ -6813,12 +6836,13 @@ NODE nd_f4(int m,int checkonly,int **indp) } nzlist = 0; nzlist_t = nd_nzlist; - f4red = 0; + f4red = 1; + nd_last_nonzero = 0; while ( d || nzlist_t ) { get_eg(&eg0); if ( nd_nzlist ) { node = BDY((LIST)BDY(nzlist_t)); - sugar = (int)ARG0(node); + sugar = QTOS((Q)ARG0(node)); tn = BDY((LIST)ARG1(node)); if ( !tn ) { nzlist_t = NEXT(nzlist_t); @@ -6851,11 +6875,12 @@ NODE nd_f4(int m,int checkonly,int **indp) } get_eg(&eg1); init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg1); if ( DP_Print ) - fprintf(asir_out,"sugar=%d,symb=%fsec,", + fprintf(asir_out,"sugar=%d,symb=%.3fsec,", sugar,eg_f4.exectime+eg_f4.gctime); nflist = nd_f4_red(m,l,0,s0vect,col,rp0,nd_gentrace?&ll:0); if ( checkonly && nflist ) return 0; /* adding new bases */ + if ( nflist ) nd_last_nonzero = f4red; for ( r = nflist; r; r = NEXT(r) ) { nf = (NDV)BDY(r); ndv_removecont(m,nf); @@ -6872,7 +6897,7 @@ NODE nd_f4(int m,int checkonly,int **indp) g = update_base(g,nh); } if ( DP_Print ) { - fprintf(asir_out,"f4red=%d,gblen=%d,",f4red,length(g)); fflush(asir_out); + fprintf(asir_out,"f4red=%d,gblen=%d\n",f4red,length(g)); fflush(asir_out); } if ( nd_gentrace ) { for ( t = ll, tn0 = 0; t; t = NEXT(t) ) { @@ -6887,7 +6912,7 @@ NODE nd_f4(int m,int checkonly,int **indp) } if ( nd_nzlist ) nzlist_t = NEXT(nzlist_t); f4red++; - if ( nd_f4red && f4red >= nd_f4red ) break; + if ( nd_f4red && f4red > nd_f4red ) break; if ( nd_rank0 && !nflist ) break; } if ( nd_gentrace ) { @@ -6950,7 +6975,7 @@ NODE nd_f4_trace(int m,int **indp) } get_eg(&eg1); init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg1); if ( DP_Print ) - fprintf(asir_out,"sugar=%d,symb=%fsec,", + fprintf(asir_out,"sugar=%d,symb=%.3fsec,", sugar,eg_f4.exectime+eg_f4.gctime); nflist = nd_f4_red(m,l,0,s0vect,col,rp0,&l0); if ( !l0 ) continue; @@ -7164,7 +7189,7 @@ init_eg(&eg_search); get_eg(&eg2); init_eg(&eg_elim2); add_eg(&eg_elim2,&eg1,&eg2); if ( DP_Print ) { - fprintf(asir_out,"elim1=%fsec,elim2=%fsec\n", + fprintf(asir_out,"elim1=%.3fsec,elim2=%.3fsec,", eg_elim1.exectime+eg_elim1.gctime,eg_elim2.exectime+eg_elim2.gctime); fflush(asir_out); } @@ -7195,7 +7220,7 @@ init_eg(&eg_search); /* construction of index arrays */ if ( DP_Print ) { - fprintf(stderr,"%dx%d,",nsp+nred,col); + fprintf(asir_out,"%dx%d,",nsp+nred,col); } rvect = (NM_ind_pair *)MALLOC(nred*sizeof(NM_ind_pair)); s0hash = (int *)MALLOC(col*sizeof(int)); @@ -7206,13 +7231,12 @@ init_eg(&eg_search); imat[i] = nm_ind_pair_to_vect_compress(trace,s0vect,col,s0hash,rvect[i]); rhead[imat[i]->head] = 1; } -#if defined(__GNUC__) - if ( m >= (1<<15) ) - r0 = nd_f4_red_mod128_main(m,sp0,nsp,s0vect,col,rvect,rhead,imat,nred,nz); - else -#endif if ( m > 0 ) +#if defined(__GNUC__) + r0 = nd_f4_red_mod64_main(m,sp0,nsp,s0vect,col,rvect,rhead,imat,nred,nz); +#else r0 = nd_f4_red_main(m,sp0,nsp,s0vect,col,rvect,rhead,imat,nred,nz); +#endif else if ( m == -1 ) r0 = nd_f4_red_sf_main(m,sp0,nsp,s0vect,col,rvect,rhead,imat,nred,nz); else if ( m == -2 ) @@ -7223,7 +7247,9 @@ init_eg(&eg_search); #else r0 = nd_f4_red_gz_main(sp0,nsp,trace,s0vect,col,rvect,rhead,imat,nred); #endif +#if 0 if ( DP_Print ) print_eg("search",&eg_search); +#endif return r0; } @@ -7274,7 +7300,7 @@ NODE nd_f4_red_main(int m,ND_pairs sp0,int nsp,UINT *s } get_eg(&eg1); init_eg(&eg_f4_1); add_eg(&eg_f4_1,&eg0,&eg1); if ( DP_Print ) { - fprintf(asir_out,"elim1=%fsec,",eg_f4_1.exectime+eg_f4_1.gctime); + fprintf(asir_out,"elim1=%.3fsec,",eg_f4_1.exectime+eg_f4_1.gctime); fflush(asir_out); } /* free index arrays */ @@ -7299,10 +7325,10 @@ NODE nd_f4_red_main(int m,ND_pairs sp0,int nsp,UINT *s get_eg(&eg2); init_eg(&eg_f4_2); add_eg(&eg_f4_2,&eg1,&eg2); init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg2); if ( DP_Print ) { - fprintf(asir_out,"elim2=%fsec\n",eg_f4_2.exectime+eg_f4_2.gctime); - fprintf(asir_out,"nsp=%d,nred=%d,spmat=(%d,%d),rank=%d ", + fprintf(asir_out,"elim2=%.3fsec,",eg_f4_2.exectime+eg_f4_2.gctime); + fprintf(asir_out,"nsp=%d,nred=%d,spmat=(%d,%d),rank=%d ", nsp,nred,sprow,spcol,rank); - fprintf(asir_out,"%fsec\n",eg_f4.exectime+eg_f4.gctime); + fprintf(asir_out,"%.3fsec,",eg_f4.exectime+eg_f4.gctime); } if ( nz ) { for ( i = 0; i < rank-1; i++ ) NEXT(spactive[i]) = spactive[i+1]; @@ -7318,7 +7344,7 @@ NODE nd_f4_red_main(int m,ND_pairs sp0,int nsp,UINT *s #if defined(__GNUC__) /* for Fp, 2^15== md ) a %= md; - *pk = ((U64)a*inv)%md; - } - for ( i = rank+1; i < row; i++ ) { - t = mat[i]; - if ( a = (t[j]%md) ) { - sugar[i] = MAX(sugar[i],s); - red_by_vect128(md,t+j,pivot+j,(int)(md-a),col-j); - } - } - rank++; + cmat = (UINT **)MALLOC(row*sizeof(UINT *)); + for ( i = 0; i < row; i++ ) { + cmat[i] = MALLOC_ATOMIC(col*sizeof(UINT)); + bzero(cmat[i],col*sizeof(UINT)); + } + + for ( rank = 0, j = 0; j < col; j++ ) { + for ( i = rank; i < row; i++ ) { + a = mat[i][j]; c = cmat[i][j]; + MOD128(a,c,md); + mat[i][j] = a; cmat[i][j] = 0; } - for ( j = col-1, l = rank-1; j >= 0; j-- ) - if ( colstat[j] ) { - pivot = mat[l]; - for ( k = j; k < col; k++ ) - if ( pivot[k] >= md ) pivot[k] %= md; - s = sugar[l]; - for ( i = 0; i < l; i++ ) { - t = mat[i]; - t[j] %= md; - if ( a = (t[j]%md) ) { - sugar[i] = MAX(sugar[i],s); - red_by_vect128(md,t+j,pivot+j,(int)(md-a),col-j); - } - } - l--; + for ( i = rank; i < row; i++ ) + if ( mat[i][j] ) + break; + if ( i == row ) { + colstat[j] = 0; + continue; + } else + colstat[j] = 1; + if ( i != rank ) { + t = mat[i]; mat[i] = mat[rank]; mat[rank] = t; + ct = cmat[i]; cmat[i] = cmat[rank]; cmat[rank] = ct; + s = sugar[i]; sugar[i] = sugar[rank]; sugar[rank] = s; + if ( spactive ) { + pair = spactive[i]; spactive[i] = spactive[rank]; + spactive[rank] = pair; + } + } + /* column j is normalized */ + s = sugar[rank]; + inv = invm((UINT)mat[rank][j],md); + /* normalize pivot row */ + for ( k = j, pk = mat[rank]+j, ck = cmat[rank]+j; k < col; k++, pk++, ck++ ) { + a = *pk; c = *ck; MOD128(a,c,md); *pk = (a*inv)%md; *ck = 0; + } + for ( i = rank+1; i < row; i++ ) { + if ( a = mat[i][j] ) { + sugar[i] = MAX(sugar[i],s); + red_by_vect64(md,mat[i]+j,cmat[i]+j,mat[rank]+j,(int)(md-a),col-j); + } + } + rank++; + } + for ( j = col-1, l = rank-1; j >= 0; j-- ) + if ( colstat[j] ) { + for ( k = j, pk = mat[l]+j, ck = cmat[l]+j; k < col; k++, pk++, ck++ ) { + a = *pk; c = *ck; MOD128(a,c,md); *pk = a; *ck = 0; + } + s = sugar[l]; + for ( i = 0; i < l; i++ ) { + a = mat[i][j]; c = cmat[i][j]; MOD128(a,c,md); mat[i][j] = a; cmat[i][j] = 0; + if ( a ) { + sugar[i] = MAX(sugar[i],s); + red_by_vect64(md,mat[i]+j,cmat[i]+j,mat[l]+j,(int)(md-a),col-j); } - return rank; + } + l--; + } + for ( i = 0; i < row; i++ ) GCFREE(cmat[i]); + GCFREE(cmat); + return rank; } #endif @@ -8330,7 +8371,7 @@ void nd_det(int mod,MAT f,P *rp) chsgnq(ONE,&mone); for ( j = 0, sgn = 1; j < n; j++ ) { if ( DP_Print ) { - fprintf(stderr,".",j); + fprintf(asir_out,".",j); } for ( i = j; i < n && !dm[i][j]; i++ ); if ( i == n ) { @@ -8360,14 +8401,14 @@ void nd_det(int mod,MAT f,P *rp) nd_det_reconstruct(dm,n,j,d); for ( i = j+1, mj = dm[j], mjj = mj[j]; i < n; i++ ) { -/* if ( DP_Print ) fprintf(stderr," i=%d\n ",i); */ +/* if ( DP_Print ) fprintf(asir_out," i=%d\n ",i); */ mi = dm[i]; mij = mi[j]; if ( mod ) ndv_mul_c(mod,mij,mod-1); else ndv_mul_c_q(mij,mone); for ( k = j+1; k < n; k++ ) { -/* if ( DP_Print ) fprintf(stderr,"k=%d ",k); */ +/* if ( DP_Print ) fprintf(asir_out,"k=%d ",k); */ bucket = create_pbucket(); if ( mi[k] ) { nmv = BDY(mjj); len = LEN(mjj); @@ -8386,12 +8427,12 @@ void nd_det(int mod,MAT f,P *rp) u = nd_quo(mod,bucket,d); mi[k] = ndtondv(mod,u); } -/* if ( DP_Print ) fprintf(stderr,"\n",k); */ +/* if ( DP_Print ) fprintf(asir_out,"\n",k); */ } d = mjj; } if ( DP_Print ) { - fprintf(stderr,"\n",k); + fprintf(asir_out,"\n",k); } if ( sgn < 0 ) if ( mod ) @@ -9190,7 +9231,7 @@ void nd_f4_lf_trace(LIST f,LIST v,int trace,int homo,s } get_eg(&eg1); init_eg(&eg_check); add_eg(&eg_check,&eg0,&eg1); if ( DP_Print ) - fprintf(asir_out,"check=%fsec\n",eg_check.exectime+eg_check.gctime); + fprintf(asir_out,"check=%.3fsec\n",eg_check.exectime+eg_check.gctime); /* dp->p */ nd_bpe = cbpe; nd_setup_parameters(nd_nvar,0); @@ -9250,7 +9291,7 @@ NODE nd_f4_lf_trace_main(int m,int **indp) } get_eg(&eg1); init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg1); if ( DP_Print ) - fprintf(asir_out,"sugar=%d,symb=%fsec,", + fprintf(asir_out,"sugar=%d,symb=%.3fsec,", sugar,eg_f4.exectime+eg_f4.gctime); nflist = nd_f4_red(m,l,0,s0vect,col,rp0,&l0); if ( !l0 ) continue;