=================================================================== RCS file: /home/cvs/OpenXM_contrib2/asir2000/engine/nd.c,v retrieving revision 1.38 retrieving revision 1.110 diff -u -p -r1.38 -r1.110 --- OpenXM_contrib2/asir2000/engine/nd.c 2003/08/20 08:56:29 1.38 +++ OpenXM_contrib2/asir2000/engine/nd.c 2004/09/21 05:23:14 1.110 @@ -1,244 +1,53 @@ -/* $OpenXM: OpenXM_contrib2/asir2000/engine/nd.c,v 1.37 2003/08/20 08:38:20 noro Exp $ */ +/* $OpenXM: OpenXM_contrib2/asir2000/engine/nd.c,v 1.109 2004/09/21 04:50:15 noro Exp $ */ -#include "ca.h" -#include "inline.h" +#include "nd.h" -#if defined(__GNUC__) -#define INLINE inline -#elif defined(VISUAL) -#define INLINE __inline -#else -#define INLINE -#endif - -#define USE_GEOBUCKET 1 - -#define REDTAB_LEN 32003 - -typedef struct oPGeoBucket { - int m; - struct oND *body[32]; -} *PGeoBucket; - -typedef struct oND { - struct oNM *body; - int nv; - int len; - int sugar; -} *ND; - -typedef struct oNDV { - struct oNMV *body; - int nv; - int len; - int sugar; -} *NDV; - -typedef struct oNM { - struct oNM *next; - union { - int m; - Q z; - } c; - unsigned int dl[1]; -} *NM; - -typedef struct oNMV { - union { - int m; - Q z; - } c; - unsigned int dl[1]; -} *NMV; - -typedef struct oRHist { - struct oRHist *next; - int index; - int sugar; - unsigned int dl[1]; -} *RHist; - -typedef struct oND_pairs { - struct oND_pairs *next; - int i1,i2; - int sugar; - unsigned int lcm[1]; -} *ND_pairs; - -int (*nd_compare_function)(unsigned int *a1,unsigned int *a2); - -double nd_scale=2; -static unsigned int **nd_bound; -int nd_nvar; -int nd_isrlex; -int nd_epw,nd_bpe,nd_wpd; -unsigned int nd_mask[32]; -unsigned int nd_mask0,nd_mask1; - +int (*ndl_compare_function)(UINT *a1,UINT *a2); +int nd_dcomp; NM _nm_free_list; ND _nd_free_list; ND_pairs _ndp_free_list; +#if 0 +static int ndv_alloc; +#endif +#if 1 +static int nd_f4_nsp=0x7fffffff; +#else +static int nd_f4_nsp=50; +#endif +static double nd_scale=2; +static UINT **nd_bound; +static struct order_spec *nd_ord; +static EPOS nd_epos; +static BlockMask nd_blockmask; +static int nd_nvar; +static int nd_isrlex; +static int nd_epw,nd_bpe,nd_wpd,nd_exporigin; +static UINT nd_mask[32]; +static UINT nd_mask0,nd_mask1; + static NDV *nd_ps; -static NDV *nd_psq; -RHist *nd_psh; -int nd_psn,nd_pslen; +static NDV *nd_ps_trace; +static RHist *nd_psh; +static int nd_psn,nd_pslen; +static RHist *nd_red; +static int *nd_work_vector; +static int **nd_matrix; +static int nd_matrix_len; +static struct weight_or_block *nd_worb; +static int nd_worb_len; +static int nd_found,nd_create,nd_notfirst; +static int nmv_adv; +static int nd_demand; -RHist *nd_red; -int nd_red_len; - -int nd_found,nd_create,nd_notfirst; -int nm_adv; -int nmv_adv; -int nd_dcomp; - -extern int Top,Reverse,dp_nelim; - -#define TD(d) (d[0]) -#define HDL(d) ((d)->body->dl) -#define HTD(d) (TD(HDL(d))) -#define HCM(d) ((d)->body->c.m) -#define HCQ(d) ((d)->body->c.z) -#define CM(a) ((a)->c.m) -#define CQ(a) ((a)->c.z) -#define DL(a) ((a)->dl) -#define SG(a) ((a)->sugar) -#define LEN(a) ((a)->len) -#define LCM(a) ((a)->lcm) - -#define TD_DL_COMPARE(d1,d2)\ -(TD(d1)>TD(d2)?1:(TD(d1)next = _nm_free_list; _nm_free_list = p; } } @@ -270,7 +79,7 @@ void _NDP_alloc() for ( i = 0; i < 1024; i++ ) { p = (ND_pairs)GC_malloc(sizeof(struct oND_pairs) - +nd_wpd*sizeof(unsigned int)); + +(nd_wpd-1)*sizeof(UINT)); p->next = _ndp_free_list; _ndp_free_list = p; } } @@ -288,155 +97,230 @@ INLINE int nd_length(ND p) } } -INLINE int ndl_reducible(unsigned int *d1,unsigned int *d2) +INLINE int ndl_reducible(UINT *d1,UINT *d2) { - unsigned int u1,u2; + UINT u1,u2; int i,j; if ( TD(d1) < TD(d2) ) return 0; +#if USE_UNROLL switch ( nd_bpe ) { + case 3: + for ( i = nd_exporigin; i < nd_wpd; i++ ) { + u1 = d1[i]; u2 = d2[i]; + if ( (u1&0x38000000) < (u2&0x38000000) ) return 0; + if ( (u1& 0x7000000) < (u2& 0x7000000) ) return 0; + if ( (u1& 0xe00000) < (u2& 0xe00000) ) return 0; + if ( (u1& 0x1c0000) < (u2& 0x1c0000) ) return 0; + if ( (u1& 0x38000) < (u2& 0x38000) ) return 0; + if ( (u1& 0x7000) < (u2& 0x7000) ) return 0; + if ( (u1& 0xe00) < (u2& 0xe00) ) return 0; + if ( (u1& 0x1c0) < (u2& 0x1c0) ) return 0; + if ( (u1& 0x38) < (u2& 0x38) ) return 0; + if ( (u1& 0x7) < (u2& 0x7) ) return 0; + } + return 1; + break; case 4: - for ( i = 1; i <= nd_wpd; i++ ) { + for ( i = nd_exporigin; i < nd_wpd; i++ ) { u1 = d1[i]; u2 = d2[i]; if ( (u1&0xf0000000) < (u2&0xf0000000) ) return 0; - if ( (u1&0xf000000) < (u2&0xf000000) ) return 0; - if ( (u1&0xf00000) < (u2&0xf00000) ) return 0; - if ( (u1&0xf0000) < (u2&0xf0000) ) return 0; - if ( (u1&0xf000) < (u2&0xf000) ) return 0; - if ( (u1&0xf00) < (u2&0xf00) ) return 0; - if ( (u1&0xf0) < (u2&0xf0) ) return 0; - if ( (u1&0xf) < (u2&0xf) ) return 0; + if ( (u1& 0xf000000) < (u2& 0xf000000) ) return 0; + if ( (u1& 0xf00000) < (u2& 0xf00000) ) return 0; + if ( (u1& 0xf0000) < (u2& 0xf0000) ) return 0; + if ( (u1& 0xf000) < (u2& 0xf000) ) return 0; + if ( (u1& 0xf00) < (u2& 0xf00) ) return 0; + if ( (u1& 0xf0) < (u2& 0xf0) ) return 0; + if ( (u1& 0xf) < (u2& 0xf) ) return 0; } return 1; break; case 6: - for ( i = 1; i <= nd_wpd; i++ ) { + for ( i = nd_exporigin; i < nd_wpd; i++ ) { u1 = d1[i]; u2 = d2[i]; if ( (u1&0x3f000000) < (u2&0x3f000000) ) return 0; - if ( (u1&0xfc0000) < (u2&0xfc0000) ) return 0; - if ( (u1&0x3f000) < (u2&0x3f000) ) return 0; - if ( (u1&0xfc0) < (u2&0xfc0) ) return 0; - if ( (u1&0x3f) < (u2&0x3f) ) return 0; + if ( (u1& 0xfc0000) < (u2& 0xfc0000) ) return 0; + if ( (u1& 0x3f000) < (u2& 0x3f000) ) return 0; + if ( (u1& 0xfc0) < (u2& 0xfc0) ) return 0; + if ( (u1& 0x3f) < (u2& 0x3f) ) return 0; } return 1; break; case 8: - for ( i = 1; i <= nd_wpd; i++ ) { + for ( i = nd_exporigin; i < nd_wpd; i++ ) { u1 = d1[i]; u2 = d2[i]; if ( (u1&0xff000000) < (u2&0xff000000) ) return 0; - if ( (u1&0xff0000) < (u2&0xff0000) ) return 0; - if ( (u1&0xff00) < (u2&0xff00) ) return 0; - if ( (u1&0xff) < (u2&0xff) ) return 0; + if ( (u1& 0xff0000) < (u2& 0xff0000) ) return 0; + if ( (u1& 0xff00) < (u2& 0xff00) ) return 0; + if ( (u1& 0xff) < (u2& 0xff) ) return 0; } return 1; break; case 16: - for ( i = 1; i <= nd_wpd; i++ ) { + for ( i = nd_exporigin; i < nd_wpd; i++ ) { u1 = d1[i]; u2 = d2[i]; if ( (u1&0xffff0000) < (u2&0xffff0000) ) return 0; - if ( (u1&0xffff) < (u2&0xffff) ) return 0; + if ( (u1& 0xffff) < (u2& 0xffff) ) return 0; } return 1; break; case 32: - for ( i = 1; i <= nd_wpd; i++ ) + for ( i = nd_exporigin; i < nd_wpd; i++ ) if ( d1[i] < d2[i] ) return 0; return 1; break; default: - for ( i = 1; i <= nd_wpd; i++ ) { + for ( i = nd_exporigin; i < nd_wpd; i++ ) { u1 = d1[i]; u2 = d2[i]; for ( j = 0; j < nd_epw; j++ ) if ( (u1&nd_mask[j]) < (u2&nd_mask[j]) ) return 0; } return 1; } +#else + for ( i = nd_exporigin; i < nd_wpd; i++ ) { + u1 = d1[i]; u2 = d2[i]; + for ( j = 0; j < nd_epw; j++ ) + if ( (u1&nd_mask[j]) < (u2&nd_mask[j]) ) return 0; + } + return 1; +#endif } -void ndl_dehomogenize(unsigned int *d) +/* + * If the current order is a block order, + * then the last block is length 1 and contains + * the homo variable. Otherwise, the original + * order is either 0 or 2. + */ + +void ndl_homogenize(UINT *d,UINT *r,int obpe,EPOS oepos,int weight) { - unsigned int mask; - unsigned int h; + int w,i,e,n,omask0; + + omask0 = (1<>((nd_epw-1)*nd_bpe))&nd_mask0; + for ( i = nd_exporigin; i < nd_wpd; i++ ) + d[i] = ((d[i]<>((nd_epw-1)*nd_bpe))&nd_mask0):0); + TD(d) -= h; + } } else { - bits = nd_epw*nd_bpe; - mask = bits==32?0xffffffff:((1<<(nd_epw*nd_bpe))-1); - h = (d[1]>>((nd_epw-1)*nd_bpe))&nd_mask0; - for ( i = 1; i <= nd_wpd; i++ ) - d[i] = ((d[i]<>((nd_epw-1)*nd_bpe))&nd_mask0):0); + h = GET_EXP(d,nd_nvar-1); + XOR_EXP(d,nd_nvar-1,h); TD(d) -= h; } - } else - TD(d) -= ((d[(nd_nvar-1)/nd_epw+1]>> - ((nd_epw-((nd_nvar-1)%nd_epw)-1)*nd_bpe))&((1<t2?t1:t2; + t1 = (u1& 0x7000000); t2 = (u2& 0x7000000); u |= t1>t2?t1:t2; + t1 = (u1& 0xe00000); t2 = (u2& 0xe00000); u |= t1>t2?t1:t2; + t1 = (u1& 0x1c0000); t2 = (u2& 0x1c0000); u |= t1>t2?t1:t2; + t1 = (u1& 0x38000); t2 = (u2& 0x38000); u |= t1>t2?t1:t2; + t1 = (u1& 0x7000); t2 = (u2& 0x7000); u |= t1>t2?t1:t2; + t1 = (u1& 0xe00); t2 = (u2& 0xe00); u |= t1>t2?t1:t2; + t1 = (u1& 0x1c0); t2 = (u2& 0x1c0); u |= t1>t2?t1:t2; + t1 = (u1& 0x38); t2 = (u2& 0x38); u |= t1>t2?t1:t2; + t1 = (u1& 0x7); t2 = (u2& 0x7); u |= t1>t2?t1:t2; + d[i] = u; + } + break; case 4: - for ( i = 1; i <= nd_wpd; i++ ) { + for ( i = nd_exporigin; i < nd_wpd; i++ ) { u1 = d1[i]; u2 = d2[i]; t1 = (u1&0xf0000000); t2 = (u2&0xf0000000); u = t1>t2?t1:t2; - t1 = (u1&0xf000000); t2 = (u2&0xf000000); u |= t1>t2?t1:t2; - t1 = (u1&0xf00000); t2 = (u2&0xf00000); u |= t1>t2?t1:t2; - t1 = (u1&0xf0000); t2 = (u2&0xf0000); u |= t1>t2?t1:t2; - t1 = (u1&0xf000); t2 = (u2&0xf000); u |= t1>t2?t1:t2; - t1 = (u1&0xf00); t2 = (u2&0xf00); u |= t1>t2?t1:t2; - t1 = (u1&0xf0); t2 = (u2&0xf0); u |= t1>t2?t1:t2; - t1 = (u1&0xf); t2 = (u2&0xf); u |= t1>t2?t1:t2; + t1 = (u1& 0xf000000); t2 = (u2& 0xf000000); u |= t1>t2?t1:t2; + t1 = (u1& 0xf00000); t2 = (u2& 0xf00000); u |= t1>t2?t1:t2; + t1 = (u1& 0xf0000); t2 = (u2& 0xf0000); u |= t1>t2?t1:t2; + t1 = (u1& 0xf000); t2 = (u2& 0xf000); u |= t1>t2?t1:t2; + t1 = (u1& 0xf00); t2 = (u2& 0xf00); u |= t1>t2?t1:t2; + t1 = (u1& 0xf0); t2 = (u2& 0xf0); u |= t1>t2?t1:t2; + t1 = (u1& 0xf); t2 = (u2& 0xf); u |= t1>t2?t1:t2; d[i] = u; } break; case 6: - for ( i = 1; i <= nd_wpd; i++ ) { + for ( i = nd_exporigin; i < nd_wpd; i++ ) { u1 = d1[i]; u2 = d2[i]; t1 = (u1&0x3f000000); t2 = (u2&0x3f000000); u = t1>t2?t1:t2; - t1 = (u1&0xfc0000); t2 = (u2&0xfc0000); u |= t1>t2?t1:t2; - t1 = (u1&0x3f000); t2 = (u2&0x3f000); u |= t1>t2?t1:t2; - t1 = (u1&0xfc0); t2 = (u2&0xfc0); u |= t1>t2?t1:t2; - t1 = (u1&0x3f); t2 = (u2&0x3f); u |= t1>t2?t1:t2; + t1 = (u1& 0xfc0000); t2 = (u2& 0xfc0000); u |= t1>t2?t1:t2; + t1 = (u1& 0x3f000); t2 = (u2& 0x3f000); u |= t1>t2?t1:t2; + t1 = (u1& 0xfc0); t2 = (u2& 0xfc0); u |= t1>t2?t1:t2; + t1 = (u1& 0x3f); t2 = (u2& 0x3f); u |= t1>t2?t1:t2; d[i] = u; } break; case 8: - for ( i = 1; i <= nd_wpd; i++ ) { + for ( i = nd_exporigin; i < nd_wpd; i++ ) { u1 = d1[i]; u2 = d2[i]; t1 = (u1&0xff000000); t2 = (u2&0xff000000); u = t1>t2?t1:t2; - t1 = (u1&0xff0000); t2 = (u2&0xff0000); u |= t1>t2?t1:t2; - t1 = (u1&0xff00); t2 = (u2&0xff00); u |= t1>t2?t1:t2; - t1 = (u1&0xff); t2 = (u2&0xff); u |= t1>t2?t1:t2; + t1 = (u1& 0xff0000); t2 = (u2& 0xff0000); u |= t1>t2?t1:t2; + t1 = (u1& 0xff00); t2 = (u2& 0xff00); u |= t1>t2?t1:t2; + t1 = (u1& 0xff); t2 = (u2& 0xff); u |= t1>t2?t1:t2; d[i] = u; } break; case 16: - for ( i = 1; i <= nd_wpd; i++ ) { + for ( i = nd_exporigin; i < nd_wpd; i++ ) { u1 = d1[i]; u2 = d2[i]; t1 = (u1&0xffff0000); t2 = (u2&0xffff0000); u = t1>t2?t1:t2; - t1 = (u1&0xffff); t2 = (u2&0xffff); u |= t1>t2?t1:t2; + t1 = (u1& 0xffff); t2 = (u2& 0xffff); u |= t1>t2?t1:t2; d[i] = u; } break; case 32: - for ( i = 1; i <= nd_wpd; i++ ) { + for ( i = nd_exporigin; i < nd_wpd; i++ ) { u1 = d1[i]; u2 = d2[i]; d[i] = u1>u2?u1:u2; } break; default: - for ( i = 1; i <= nd_wpd; i++ ) { + for ( i = nd_exporigin; i < nd_wpd; i++ ) { u1 = d1[i]; u2 = d2[i]; for ( j = 0, u = 0; j < nd_epw; j++ ) { t1 = (u1&nd_mask[j]); t2 = (u2&nd_mask[j]); u |= t1>t2?t1:t2; @@ -445,27 +329,69 @@ void ndl_lcm(unsigned int *d1,unsigned *d2,unsigned in } break; } - TD(d) = ndl_td(d); +#else + for ( i = nd_exporigin; i < nd_wpd; i++ ) { + u1 = d1[i]; u2 = d2[i]; + for ( j = 0, u = 0; j < nd_epw; j++ ) { + t1 = (u1&nd_mask[j]); t2 = (u2&nd_mask[j]); u |= t1>t2?t1:t2; + } + d[i] = u; + } +#endif + TD(d) = ndl_weight(d); + if ( nd_blockmask ) ndl_weight_mask(d); } -int ndl_td(unsigned int *d) +int ndl_weight(UINT *d) { - unsigned int t,u; + UINT t,u; int i,j; - for ( t = 0, i = 1; i <= nd_wpd; i++ ) { - u = d[i]; - for ( j = 0; j < nd_epw; j++, u>>=nd_bpe ) - t += (u&nd_mask0); - } + if ( current_dl_weight_vector ) + for ( i = 0, t = 0; i < nd_nvar; i++ ) { + u = GET_EXP(d,i); + t += MUL_WEIGHT(u,i); + } + else + for ( t = 0, i = nd_exporigin; i < nd_wpd; i++ ) { + u = d[i]; + for ( j = 0; j < nd_epw; j++, u>>=nd_bpe ) + t += (u&nd_mask0); + } return t; } -INLINE int ndl_lex_compare(unsigned int *d1,unsigned int *d2) +void ndl_weight_mask(UINT *d) { + UINT t,u; + UINT *mask; + int i,j,k,l; + + l = nd_blockmask->n; + for ( k = 0; k < l; k++ ) { + mask = nd_blockmask->mask[k]; + if ( current_dl_weight_vector ) + for ( i = 0, t = 0; i < nd_nvar; i++ ) { + u = GET_EXP_MASK(d,i,mask); + t += MUL_WEIGHT(u,i); + } + else + for ( t = 0, i = nd_exporigin; i < nd_wpd; i++ ) { + u = d[i]&mask[i]; + for ( j = 0; j < nd_epw; j++, u>>=nd_bpe ) + t += (u&nd_mask0); + } + d[k+1] = t; + } +} + +int ndl_lex_compare(UINT *d1,UINT *d2) +{ int i; - for ( i = 0; i < nd_wpd; i++, d1++, d2++ ) + d1 += nd_exporigin; + d2 += nd_exporigin; + for ( i = nd_exporigin; i < nd_wpd; i++, d1++, d2++ ) if ( *d1 > *d2 ) return nd_isrlex ? -1 : 1; else if ( *d1 < *d2 ) @@ -473,122 +399,310 @@ INLINE int ndl_lex_compare(unsigned int *d1,unsigned i return 0; } -INLINE int ndl_equal(unsigned int *d1,unsigned int *d2) +int ndl_block_compare(UINT *d1,UINT *d2) { + int i,l,j,ord_o,ord_l; + struct order_pair *op; + UINT t1,t2,m; + UINT *mask; + + l = nd_blockmask->n; + op = nd_blockmask->order_pair; + for ( j = 0; j < l; j++ ) { + mask = nd_blockmask->mask[j]; + ord_o = op[j].order; + if ( ord_o < 2 ) + if ( (t1=d1[j+1]) > (t2=d2[j+1]) ) return 1; + else if ( t1 < t2 ) return -1; + for ( i = nd_exporigin; i < nd_wpd; i++ ) { + m = mask[i]; + t1 = d1[i]&m; + t2 = d2[i]&m; + if ( t1 > t2 ) + return !ord_o ? -1 : 1; + else if ( t1 < t2 ) + return !ord_o ? 1 : -1; + } + } + return 0; +} + +int ndl_matrix_compare(UINT *d1,UINT *d2) +{ + int i,j,s; + int *v; + + for ( j = 0; j < nd_nvar; j++ ) + nd_work_vector[j] = GET_EXP(d1,j)-GET_EXP(d2,j); + for ( i = 0; i < nd_matrix_len; i++ ) { + v = nd_matrix[i]; + for ( j = 0, s = 0; j < nd_nvar; j++ ) + s += v[j]*nd_work_vector[j]; + if ( s > 0 ) return 1; + else if ( s < 0 ) return -1; + } + return 0; +} + +int ndl_composite_compare(UINT *d1,UINT *d2) +{ + int i,j,s,start,end,len,o; + int *v; + struct sparse_weight *sw; + + for ( j = 0; j < nd_nvar; j++ ) + nd_work_vector[j] = GET_EXP(d1,j)-GET_EXP(d2,j); + for ( i = 0; i < nd_worb_len; i++ ) { + len = nd_worb[i].length; + switch ( nd_worb[i].type ) { + case IS_DENSE_WEIGHT: + v = nd_worb[i].body.dense_weight; + for ( j = 0, s = 0; j < len; j++ ) + s += v[j]*nd_work_vector[j]; + if ( s > 0 ) return 1; + else if ( s < 0 ) return -1; + break; + case IS_SPARSE_WEIGHT: + sw = nd_worb[i].body.sparse_weight; + for ( j = 0, s = 0; j < len; j++ ) + s += sw[j].value*nd_work_vector[sw[j].pos]; + if ( s > 0 ) return 1; + else if ( s < 0 ) return -1; + break; + case IS_BLOCK: + o = nd_worb[i].body.block.order; + start = nd_worb[i].body.block.start; + switch ( o ) { + case 0: + end = start+len; + for ( j = start, s = 0; j < end; j++ ) + s += MUL_WEIGHT(nd_work_vector[j],j); + if ( s > 0 ) return 1; + else if ( s < 0 ) return -1; + for ( j = end-1; j >= start; j-- ) + if ( nd_work_vector[j] < 0 ) return 1; + else if ( nd_work_vector[j] > 0 ) return -1; + break; + case 1: + end = start+len; + for ( j = start, s = 0; j < end; j++ ) + s += MUL_WEIGHT(nd_work_vector[j],j); + if ( s > 0 ) return 1; + else if ( s < 0 ) return -1; + for ( j = start; j < end; j++ ) + if ( nd_work_vector[j] > 0 ) return 1; + else if ( nd_work_vector[j] < 0 ) return -1; + break; + case 2: + for ( j = start; j < end; j++ ) + if ( nd_work_vector[j] > 0 ) return 1; + else if ( nd_work_vector[j] < 0 ) return -1; + break; + } + break; + } + } + return 0; +} + +/* TDH -> WW -> TD-> RL */ + +int ndl_ww_lex_compare(UINT *d1,UINT *d2) +{ + int i,m,e1,e2; + + if ( TD(d1) > TD(d2) ) return 1; + else if ( TD(d1) < TD(d2) ) return -1; + m = nd_nvar>>1; + for ( i = 0, e1 = e2 = 0; i < m; i++ ) { + e1 += current_weyl_weight_vector[i]*(GET_EXP(d1,m+i)-GET_EXP(d1,i)); + e2 += current_weyl_weight_vector[i]*(GET_EXP(d2,m+i)-GET_EXP(d2,i)); + } + if ( e1 > e2 ) return 1; + else if ( e1 < e2 ) return -1; + return ndl_lex_compare(d1,d2); +} + +INLINE int ndl_equal(UINT *d1,UINT *d2) +{ int i; - if ( *d1++ != *d2++ ) return 0; - for ( i = 1; i <= nd_wpd; i++ ) - if ( *d1++ != *d2++ ) - return 0; - return 1; + switch ( nd_wpd ) { + case 2: + if ( TD(d2) != TD(d1) ) return 0; + if ( d2[1] != d1[1] ) return 0; + return 1; + break; + case 3: + if ( TD(d2) != TD(d1) ) return 0; + if ( d2[1] != d1[1] ) return 0; + if ( d2[2] != d1[2] ) return 0; + return 1; + break; + default: + for ( i = 0; i < nd_wpd; i++ ) + if ( *d1++ != *d2++ ) return 0; + return 1; + break; + } } -INLINE void ndl_copy(unsigned int *d1,unsigned int *d2) +INLINE void ndl_copy(UINT *d1,UINT *d2) { int i; switch ( nd_wpd ) { - case 1: + case 2: TD(d2) = TD(d1); d2[1] = d1[1]; break; - case 2: + case 3: TD(d2) = TD(d1); d2[1] = d1[1]; d2[2] = d1[2]; break; default: - for ( i = 0; i <= nd_wpd; i++ ) + for ( i = 0; i < nd_wpd; i++ ) d2[i] = d1[i]; break; } } -INLINE void ndl_add(unsigned int *d1,unsigned int *d2,unsigned int *d) +INLINE void ndl_zero(UINT *d) { int i; + for ( i = 0; i < nd_wpd; i++ ) d[i] = 0; +} - TD(d) = TD(d1)+TD(d2); +INLINE void ndl_add(UINT *d1,UINT *d2,UINT *d) +{ + int i; + +#if 1 switch ( nd_wpd ) { - case 1: + case 2: + TD(d) = TD(d1)+TD(d2); d[1] = d1[1]+d2[1]; break; - case 2: + case 3: + TD(d) = TD(d1)+TD(d2); d[1] = d1[1]+d2[1]; d[2] = d1[2]+d2[2]; break; default: - for ( i = 1; i <= nd_wpd; i++ ) - d[i] = d1[i]+d2[i]; + for ( i = 0; i < nd_wpd; i++ ) d[i] = d1[i]+d2[i]; break; } +#else + for ( i = 0; i < nd_wpd; i++ ) d[i] = d1[i]+d2[i]; +#endif } -INLINE void ndl_sub(unsigned int *d1,unsigned int *d2,unsigned int *d) +/* d1 += d2 */ +INLINE void ndl_addto(UINT *d1,UINT *d2) { int i; - for ( i = 0; i <= nd_wpd; i++ ) - d[i] = d1[i]-d2[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 } -int ndl_disjoint(unsigned int *d1,unsigned int *d2) +INLINE void ndl_sub(UINT *d1,UINT *d2,UINT *d) { - unsigned int t1,t2,u,u1,u2; + int i; + + for ( i = 0; i < nd_wpd; i++ ) d[i] = d1[i]-d2[i]; +} + +int ndl_disjoint(UINT *d1,UINT *d2) +{ + UINT t1,t2,u,u1,u2; int i,j; +#if USE_UNROLL switch ( nd_bpe ) { + case 3: + for ( i = nd_exporigin; i < nd_wpd; i++ ) { + u1 = d1[i]; u2 = d2[i]; + t1 = u1&0x38000000; t2 = u2&0x38000000; if ( t1&&t2 ) return 0; + t1 = u1& 0x7000000; t2 = u2& 0x7000000; if ( t1&&t2 ) return 0; + t1 = u1& 0xe00000; t2 = u2& 0xe00000; if ( t1&&t2 ) return 0; + t1 = u1& 0x1c0000; t2 = u2& 0x1c0000; if ( t1&&t2 ) return 0; + t1 = u1& 0x38000; t2 = u2& 0x38000; if ( t1&&t2 ) return 0; + t1 = u1& 0x7000; t2 = u2& 0x7000; if ( t1&&t2 ) return 0; + t1 = u1& 0xe00; t2 = u2& 0xe00; if ( t1&&t2 ) return 0; + t1 = u1& 0x1c0; t2 = u2& 0x1c0; if ( t1&&t2 ) return 0; + t1 = u1& 0x38; t2 = u2& 0x38; if ( t1&&t2 ) return 0; + t1 = u1& 0x7; t2 = u2& 0x7; if ( t1&&t2 ) return 0; + } + return 1; + break; case 4: - for ( i = 1; i <= nd_wpd; i++ ) { + for ( i = nd_exporigin; i < nd_wpd; i++ ) { u1 = d1[i]; u2 = d2[i]; t1 = u1&0xf0000000; t2 = u2&0xf0000000; if ( t1&&t2 ) return 0; - t1 = u1&0xf000000; t2 = u2&0xf000000; if ( t1&&t2 ) return 0; - t1 = u1&0xf00000; t2 = u2&0xf00000; if ( t1&&t2 ) return 0; - t1 = u1&0xf0000; t2 = u2&0xf0000; if ( t1&&t2 ) return 0; - t1 = u1&0xf000; t2 = u2&0xf000; if ( t1&&t2 ) return 0; - t1 = u1&0xf00; t2 = u2&0xf00; if ( t1&&t2 ) return 0; - t1 = u1&0xf0; t2 = u2&0xf0; if ( t1&&t2 ) return 0; - t1 = u1&0xf; t2 = u2&0xf; if ( t1&&t2 ) return 0; + t1 = u1& 0xf000000; t2 = u2& 0xf000000; if ( t1&&t2 ) return 0; + t1 = u1& 0xf00000; t2 = u2& 0xf00000; if ( t1&&t2 ) return 0; + t1 = u1& 0xf0000; t2 = u2& 0xf0000; if ( t1&&t2 ) return 0; + t1 = u1& 0xf000; t2 = u2& 0xf000; if ( t1&&t2 ) return 0; + t1 = u1& 0xf00; t2 = u2& 0xf00; if ( t1&&t2 ) return 0; + t1 = u1& 0xf0; t2 = u2& 0xf0; if ( t1&&t2 ) return 0; + t1 = u1& 0xf; t2 = u2& 0xf; if ( t1&&t2 ) return 0; } return 1; break; case 6: - for ( i = 1; i <= nd_wpd; i++ ) { + for ( i = nd_exporigin; i < nd_wpd; i++ ) { u1 = d1[i]; u2 = d2[i]; t1 = u1&0x3f000000; t2 = u2&0x3f000000; if ( t1&&t2 ) return 0; - t1 = u1&0xfc0000; t2 = u2&0xfc0000; if ( t1&&t2 ) return 0; - t1 = u1&0x3f000; t2 = u2&0x3f000; if ( t1&&t2 ) return 0; - t1 = u1&0xfc0; t2 = u2&0xfc0; if ( t1&&t2 ) return 0; - t1 = u1&0x3f; t2 = u2&0x3f; if ( t1&&t2 ) return 0; + t1 = u1& 0xfc0000; t2 = u2& 0xfc0000; if ( t1&&t2 ) return 0; + t1 = u1& 0x3f000; t2 = u2& 0x3f000; if ( t1&&t2 ) return 0; + t1 = u1& 0xfc0; t2 = u2& 0xfc0; if ( t1&&t2 ) return 0; + t1 = u1& 0x3f; t2 = u2& 0x3f; if ( t1&&t2 ) return 0; } return 1; break; case 8: - for ( i = 1; i <= nd_wpd; i++ ) { + for ( i = nd_exporigin; i < nd_wpd; i++ ) { u1 = d1[i]; u2 = d2[i]; t1 = u1&0xff000000; t2 = u2&0xff000000; if ( t1&&t2 ) return 0; - t1 = u1&0xff0000; t2 = u2&0xff0000; if ( t1&&t2 ) return 0; - t1 = u1&0xff00; t2 = u2&0xff00; if ( t1&&t2 ) return 0; - t1 = u1&0xff; t2 = u2&0xff; if ( t1&&t2 ) return 0; + t1 = u1& 0xff0000; t2 = u2& 0xff0000; if ( t1&&t2 ) return 0; + t1 = u1& 0xff00; t2 = u2& 0xff00; if ( t1&&t2 ) return 0; + t1 = u1& 0xff; t2 = u2& 0xff; if ( t1&&t2 ) return 0; } return 1; break; case 16: - for ( i = 1; i <= nd_wpd; i++ ) { + for ( i = nd_exporigin; i < nd_wpd; i++ ) { u1 = d1[i]; u2 = d2[i]; t1 = u1&0xffff0000; t2 = u2&0xffff0000; if ( t1&&t2 ) return 0; - t1 = u1&0xffff; t2 = u2&0xffff; if ( t1&&t2 ) return 0; + t1 = u1& 0xffff; t2 = u2& 0xffff; if ( t1&&t2 ) return 0; } return 1; break; case 32: - for ( i = 1; i <= nd_wpd; i++ ) + for ( i = nd_exporigin; i < nd_wpd; i++ ) if ( d1[i] && d2[i] ) return 0; return 1; break; default: - for ( i = 1; i <= nd_wpd; i++ ) { + for ( i = nd_exporigin; i < nd_wpd; i++ ) { u1 = d1[i]; u2 = d2[i]; for ( j = 0; j < nd_epw; j++ ) { if ( (u1&nd_mask0) && (u2&nd_mask0) ) return 0; @@ -598,19 +712,46 @@ int ndl_disjoint(unsigned int *d1,unsigned int *d2) return 1; break; } +#else + for ( i = nd_exporigin; i < nd_wpd; i++ ) { + u1 = d1[i]; u2 = d2[i]; + for ( j = 0; j < nd_epw; j++ ) { + if ( (u1&nd_mask0) && (u2&nd_mask0) ) return 0; + u1 >>= nd_bpe; u2 >>= nd_bpe; + } + } + return 1; +#endif } -int ndl_check_bound2(int index,unsigned int *d2) +int ndl_check_bound2(int index,UINT *d2) { - unsigned int u2; - unsigned int *d1; + UINT u2; + UINT *d1; int i,j,ind,k; d1 = nd_bound[index]; ind = 0; +#if USE_UNROLL switch ( nd_bpe ) { + case 3: + for ( i = nd_exporigin; i < nd_wpd; i++ ) { + u2 = d2[i]; + if ( d1[ind++]+((u2>>27)&0x7) >= 0x8 ) return 1; + if ( d1[ind++]+((u2>>24)&0x7) >= 0x8 ) return 1; + if ( d1[ind++]+((u2>>21)&0x7) >= 0x8 ) return 1; + if ( d1[ind++]+((u2>>18)&0x7) >= 0x8 ) return 1; + if ( d1[ind++]+((u2>>15)&0x7) >= 0x8 ) return 1; + if ( d1[ind++]+((u2>>12)&0x7) >= 0x8 ) return 1; + if ( d1[ind++]+((u2>>9)&0x7) >= 0x8 ) return 1; + if ( d1[ind++]+((u2>>6)&0x7) >= 0x8 ) return 1; + if ( d1[ind++]+((u2>>3)&0x7) >= 0x8 ) return 1; + if ( d1[ind++]+(u2&0x7) >= 0x8 ) return 1; + } + return 0; + break; case 4: - for ( i = 1; i <= nd_wpd; i++ ) { + for ( i = nd_exporigin; i < nd_wpd; i++ ) { u2 = d2[i]; if ( d1[ind++]+((u2>>28)&0xf) >= 0x10 ) return 1; if ( d1[ind++]+((u2>>24)&0xf) >= 0x10 ) return 1; @@ -624,7 +765,7 @@ int ndl_check_bound2(int index,unsigned int *d2) return 0; break; case 6: - for ( i = 1; i <= nd_wpd; i++ ) { + for ( i = nd_exporigin; i < nd_wpd; i++ ) { u2 = d2[i]; if ( d1[ind++]+((u2>>24)&0x3f) >= 0x40 ) return 1; if ( d1[ind++]+((u2>>18)&0x3f) >= 0x40 ) return 1; @@ -635,7 +776,7 @@ int ndl_check_bound2(int index,unsigned int *d2) return 0; break; case 8: - for ( i = 1; i <= nd_wpd; i++ ) { + for ( i = nd_exporigin; i < nd_wpd; i++ ) { u2 = d2[i]; if ( d1[ind++]+((u2>>24)&0xff) >= 0x100 ) return 1; if ( d1[ind++]+((u2>>16)&0xff) >= 0x100 ) return 1; @@ -645,7 +786,7 @@ int ndl_check_bound2(int index,unsigned int *d2) return 0; break; case 16: - for ( i = 1; i <= nd_wpd; i++ ) { + for ( i = nd_exporigin; i < nd_wpd; i++ ) { u2 = d2[i]; if ( d1[ind++]+((u2>>16)&0xffff) > 0x10000 ) return 1; if ( d1[ind++]+(u2&0xffff) > 0x10000 ) return 1; @@ -653,12 +794,12 @@ int ndl_check_bound2(int index,unsigned int *d2) return 0; break; case 32: - for ( i = 1; i <= nd_wpd; i++ ) + for ( i = nd_exporigin; i < nd_wpd; i++ ) if ( d1[i]+d2[i]>28)&0xf)+((u2>>28)&0xf) >= 0x10 ) return 1; - if ( ((u1>>24)&0xf)+((u2>>24)&0xf) >= 0x10 ) return 1; - if ( ((u1>>20)&0xf)+((u2>>20)&0xf) >= 0x10 ) return 1; - if ( ((u1>>16)&0xf)+((u2>>16)&0xf) >= 0x10 ) return 1; - if ( ((u1>>12)&0xf)+((u2>>12)&0xf) >= 0x10 ) return 1; - if ( ((u1>>8)&0xf)+((u2>>8)&0xf) >= 0x10 ) return 1; - if ( ((u1>>4)&0xf)+((u2>>4)&0xf) >= 0x10 ) return 1; - if ( (u1&0xf)+(u2&0xf) >= 0x10 ) return 1; - } - return 0; - break; - case 6: - for ( i = 1; i <= nd_wpd; i++ ) { - u1 = d1[i]; u2 = d2[i]; - if ( ((u1>>24)&0x3f)+((u2>>24)&0x3f) >= 0x40 ) return 1; - if ( ((u1>>18)&0x3f)+((u2>>18)&0x3f) >= 0x40 ) return 1; - if ( ((u1>>12)&0x3f)+((u2>>12)&0x3f) >= 0x40 ) return 1; - if ( ((u1>>6)&0x3f)+((u2>>6)&0x3f) >= 0x40 ) return 1; - if ( (u1&0x3f)+(u2&0x3f) >= 0x40 ) return 1; - } - return 0; - break; - case 8: - for ( i = 1; i <= nd_wpd; i++ ) { - u1 = d1[i]; u2 = d2[i]; - if ( ((u1>>24)&0xff)+((u2>>24)&0xff) >= 0x100 ) return 1; - if ( ((u1>>16)&0xff)+((u2>>16)&0xff) >= 0x100 ) return 1; - if ( ((u1>>8)&0xff)+((u2>>8)&0xff) >= 0x100 ) return 1; - if ( (u1&0xff)+(u2&0xff) >= 0x100 ) return 1; - } - return 0; - break; - case 16: - for ( i = 1; i <= nd_wpd; i++ ) { - u1 = d1[i]; u2 = d2[i]; - if ( ((u1>>16)&0xffff)+((u2>>16)&0xffff) > 0x10000 ) return 1; - if ( (u2&0xffff)+(u2&0xffff) > 0x10000 ) return 1; - } - return 0; - break; - case 32: - for ( i = 1; i <= nd_wpd; i++ ) - if ( d1[i]+d2[i]>k)&nd_mask0)+((u2>>k)&nd_mask0) > nd_mask0 ) - return 1; - } - return 0; - break; +#else + for ( i = nd_exporigin; i < nd_wpd; i++ ) { + u2 = d2[i]; + k = (nd_epw-1)*nd_bpe; + for ( j = 0; j < nd_epw; j++, k -= nd_bpe ) + if ( d1[ind++]+((u2>>k)&nd_mask0) > nd_mask0 ) return 1; } + return 0; +#endif } -INLINE int ndl_hash_value(unsigned int *d) +INLINE int ndl_hash_value(UINT *d) { int i; int r; r = 0; - for ( i = 0; i <= nd_wpd; i++ ) + for ( i = 0; i < nd_wpd; i++ ) r = ((r<<16)+d[i])%REDTAB_LEN; return r; } -INLINE int nd_find_reducer(ND g) +INLINE int ndl_find_reducer(UINT *dg) { RHist r; - unsigned int *dg; int d,k,i; - dg = HDL(g); - d = ndl_hash_value(HDL(g)); + d = ndl_hash_value(dg); for ( r = nd_red[d], k = 0; r; r = NEXT(r), k++ ) { if ( ndl_equal(dg,DL(r)) ) { if ( k > 0 ) nd_notfirst++; @@ -783,29 +864,46 @@ INLINE int nd_find_reducer(ND g) return -1; } -INLINE int nd_find_reducer_direct(ND g,NDV *ps,int len) +ND nd_merge(ND p1,ND p2) { - NDV r; - RHist s; - int d,k,i; + int n,c; + int t,can,td1,td2; + ND r; + NM m1,m2,mr0,mr,s; - if ( Reverse ) - for ( i = len-1; i >= 0; i-- ) { - r = ps[i]; - if ( ndl_reducible(HDL(g),HDL(r)) ) { - nd_append_red(HDL(g),i); - return i; + if ( !p1 ) return p2; + else if ( !p2 ) return p1; + else { + can = 0; + for ( n = NV(p1), m1 = BDY(p1), m2 = BDY(p2), mr0 = 0; m1 && m2; ) { + c = DL_COMPARE(DL(m1),DL(m2)); + switch ( c ) { + case 0: + s = m1; m1 = NEXT(m1); + can++; NEXTNM2(mr0,mr,s); + s = m2; m2 = NEXT(m2); FREENM(s); + break; + case 1: + s = m1; m1 = NEXT(m1); NEXTNM2(mr0,mr,s); + break; + case -1: + s = m2; m2 = NEXT(m2); NEXTNM2(mr0,mr,s); + break; } } - else - for ( i = 0; i < len; i++ ) { - r = ps[i]; - if ( ndl_reducible(HDL(g),HDL(r)) ) { - nd_append_red(HDL(g),i); - return i; - } - } - return -1; + if ( !mr0 ) + if ( m1 ) mr0 = m1; + else if ( m2 ) mr0 = m2; + else return 0; + else if ( m1 ) NEXT(mr) = m1; + else if ( m2 ) NEXT(mr) = m2; + else NEXT(mr) = 0; + BDY(p1) = mr0; + SG(p1) = MAX(SG(p1),SG(p2)); + LEN(p1) = LEN(p1)+LEN(p2)-can; + FREEND(p2); + return p1; + } } ND nd_add(int mod,ND p1,ND p2) @@ -817,6 +915,7 @@ ND nd_add(int mod,ND p1,ND p2) if ( !p1 ) return p2; else if ( !p2 ) return p1; + else if ( mod == -1 ) return nd_add_sf(p1,p2); else if ( !mod ) return nd_add_q(p1,p2); else { can = 0; @@ -857,6 +956,28 @@ ND nd_add(int mod,ND p1,ND p2) } } +/* XXX on opteron, the inlined manipulation of destructive additon of + * two NM seems to make gcc optimizer get confused, so the part is + * done in a function. + */ + +int nm_destructive_add_q(NM *m1,NM *m2,NM *mr0,NM *mr) +{ + NM s; + Q t; + int can; + + addq(CQ(*m1),CQ(*m2),&t); + s = *m1; *m1 = NEXT(*m1); + if ( t ) { + can = 1; NEXTNM2(*mr0,*mr,s); CQ(*mr) = (t); + } else { + can = 2; FREENM(s); + } + s = *m2; *m2 = NEXT(*m2); FREENM(s); + return can; +} + ND nd_add_q(ND p1,ND p2) { int n,c,can; @@ -872,6 +993,9 @@ ND nd_add_q(ND p1,ND p2) c = DL_COMPARE(DL(m1),DL(m2)); switch ( c ) { case 0: +#if defined(__x86_64__) + can += nm_destructive_add_q(&m1,&m2,&mr0,&mr); +#else addq(CQ(m1),CQ(m2),&t); s = m1; m1 = NEXT(m1); if ( t ) { @@ -880,6 +1004,7 @@ ND nd_add_q(ND p1,ND p2) can += 2; FREENM(s); } s = m2; m2 = NEXT(m2); FREENM(s); +#endif break; case 1: s = m1; m1 = NEXT(m1); NEXTNM2(mr0,mr,s); @@ -904,8 +1029,55 @@ ND nd_add_q(ND p1,ND p2) } } +ND nd_add_sf(ND p1,ND p2) +{ + int n,c,can; + ND r; + NM m1,m2,mr0,mr,s; + int t; + + if ( !p1 ) return p2; + else if ( !p2 ) return p1; + else { + can = 0; + for ( n = NV(p1), m1 = BDY(p1), m2 = BDY(p2), mr0 = 0; m1 && m2; ) { + c = DL_COMPARE(DL(m1),DL(m2)); + switch ( c ) { + case 0: + t = _addsf(CM(m1),CM(m2)); + s = m1; m1 = NEXT(m1); + if ( t ) { + can++; NEXTNM2(mr0,mr,s); CM(mr) = (t); + } else { + can += 2; FREENM(s); + } + s = m2; m2 = NEXT(m2); FREENM(s); + break; + case 1: + s = m1; m1 = NEXT(m1); NEXTNM2(mr0,mr,s); + break; + case -1: + s = m2; m2 = NEXT(m2); NEXTNM2(mr0,mr,s); + break; + } + } + if ( !mr0 ) + if ( m1 ) mr0 = m1; + else if ( m2 ) mr0 = m2; + else return 0; + else if ( m1 ) NEXT(mr) = m1; + else if ( m2 ) NEXT(mr) = m2; + else NEXT(mr) = 0; + BDY(p1) = mr0; + SG(p1) = MAX(SG(p1),SG(p2)); + LEN(p1) = LEN(p1)+LEN(p2)-can; + FREEND(p2); + return p1; + } +} + /* 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,NDC dn,ND *rp) { ND d; NM m,mrd,tail; @@ -914,9 +1086,15 @@ int nd_nf(int mod,ND g,int full,ND *rp) int c,c1,c2,dummy; RHist h; NDV p,red; - Q cg,cred,gcd; + Q cg,cred,gcd,tq,qq; double hmag; + if ( dn ) { + if ( mod ) + dn->m = 1; + else + dn->z = ONE; + } if ( !g ) { *rp = 0; return 1; @@ -925,9 +1103,9 @@ int nd_nf(int mod,ND g,int full,ND *rp) sugar0 = sugar = SG(g); n = NV(g); - mul = (NM)ALLOCA(sizeof(struct oNM)+nd_wpd*sizeof(unsigned int)); + mul = (NM)ALLOCA(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT)); for ( d = 0; g; ) { - index = nd_find_reducer(g); + index = ndl_find_reducer(HDL(g)); if ( index >= 0 ) { h = nd_psh[index]; ndl_sub(HDL(g),DL(h),DL(mul)); @@ -935,20 +1113,31 @@ int nd_nf(int mod,ND g,int full,ND *rp) nd_free(g); nd_free(d); return 0; } - if ( mod ) { - p = nd_ps[index]; + if ( nd_demand ) + p = ndv_load(index); + else + p = ps[index]; + if ( mod == -1 ) + CM(mul) = _mulsf(_invsf(HCM(p)),_chsgnsf(HCM(g))); + else if ( mod ) { 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); + if ( dn ) { + mulq(dn->z,cred,&tq); dn->z = tq; + } } - 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) ) { + tq = HCQ(g); nd_removecont2(d,g); + if ( dn ) { + divq(tq,HCQ(g),&qq); divq(dn->z,qq,&tq); dn->z = tq; + } hmag = ((double)p_mag((P)HCQ(g)))*nd_scale; } } else if ( !full ) { @@ -973,7 +1162,7 @@ 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; @@ -986,6 +1175,8 @@ int nd_nf_pbucket(int mod,ND g,int full,ND *rp) Q cg,cred,gcd,zzz; RHist h; double hmag,gmag; + int count = 0; + int hcount = 0; if ( !g ) { *rp = 0; @@ -997,17 +1188,20 @@ int nd_nf_pbucket(int mod,ND g,int full,ND *rp) bucket = create_pbucket(); add_pbucket(mod,bucket,g); d = 0; - mul = (NM)ALLOCA(sizeof(struct oNM)+nd_wpd*sizeof(unsigned int)); + mul = (NM)ALLOCA(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT)); while ( 1 ) { hindex = mod?head_pbucket(mod,bucket):head_pbucket_q(bucket); if ( hindex < 0 ) { + if ( DP_Print > 3 ) printf("(%d %d)",count,hcount); if ( d ) SG(d) = sugar; *rp = d; return 1; } g = bucket->body[hindex]; - index = nd_find_reducer(g); + index = ndl_find_reducer(HDL(g)); if ( index >= 0 ) { + count++; + if ( !d ) hcount++; h = nd_psh[index]; ndl_sub(HDL(g),DL(h),DL(mul)); if ( ndl_check_bound2(index,DL(mul)) ) { @@ -1016,12 +1210,13 @@ int nd_nf_pbucket(int mod,ND g,int full,ND *rp) *rp = 0; return 0; } - if ( mod ) { - p = nd_ps[index]; + p = ps[index]; + if ( mod == -1 ) + CM(mul) = _mulsf(_invsf(HCM(p)),_chsgnsf(HCM(g))); + else if ( mod ) { 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); @@ -1029,7 +1224,7 @@ 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); @@ -1069,198 +1264,37 @@ int nd_nf_pbucket(int mod,ND g,int full,ND *rp) } } -int nd_nf_direct(int mod,ND g,NDV *ps,int len,int full,ND *rp) -{ - ND d; - NM m,mrd,tail; - NM mul; - int n,sugar,psugar,sugar0,stat,index; - int c,c1,c2; - RHist h; - NDV p,red; - Q cg,cred,gcd; - double hmag; +/* input : list of NDV, cand : list of NDV */ - if ( !g ) { - *rp = 0; - return 1; - } -#if 0 - if ( !mod ) - hmag = ((double)p_mag((P)HCQ(g)))*nd_scale; -#else - /* XXX */ - hmag = 0; -#endif - - sugar0 = sugar = SG(g); - n = NV(g); - mul = (NM)ALLOCA(sizeof(struct oNM)+nd_wpd*sizeof(unsigned int)); - for ( d = 0; g; ) { - index = nd_find_reducer_direct(g,ps,len); - if ( index >= 0 ) { - p = ps[index]; - ndl_sub(HDL(g),HDL(p),DL(mul)); - if ( ndl_check_bound2_direct(HDL(p),DL(mul)) ) { - nd_free(g); nd_free(d); - return 0; - } - if ( mod ) { - c1 = invm(HCM(p),mod); c2 = mod-HCM(g); - DMAR(c1,c2,0,mod,c); CM(mul) = c; - } else { - 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)); - sugar = MAX(sugar,SG(p)+TD(DL(mul))); - if ( !mod && hmag && g && ((double)(p_mag((P)HCQ(g))) > hmag) ) { - nd_removecont2(d,g); - hmag = ((double)p_mag((P)HCQ(g)))*nd_scale; - } - } else if ( !full ) { - *rp = g; - return 1; - } else { - m = BDY(g); - if ( NEXT(m) ) { - BDY(g) = NEXT(m); NEXT(m) = 0; LEN(g)--; - } else { - FREEND(g); g = 0; - } - if ( d ) { - NEXT(tail)=m; tail=m; LEN(d)++; - } else { - MKND(n,m,1,d); tail = BDY(d); - } - } - } - if ( d ) SG(d) = sugar; - *rp = d; - return 1; -} - -int nd_nf_direct_pbucket(int mod,ND g,NDV *ps,int len,int full,ND *rp) +int ndv_check_candidate(NODE input,int obpe,int oadv,EPOS oepos,NODE cand) { - int hindex,index; - NDV p; - ND u,d,red; - NODE l; - NM mul,m,mrd,tail; - int sugar,psugar,n,h_reducible; - PGeoBucket bucket; - int c,c1,c2; - Q cg,cred,gcd,zzz; - RHist h; - double hmag,gmag; - - if ( !g ) { - *rp = 0; - return 1; - } - sugar = SG(g); - n = NV(g); -#if 0 - if ( !mod ) - hmag = ((double)p_mag((P)HCQ(g)))*nd_scale; -#else - /* XXX */ - hmag = 0; -#endif - bucket = create_pbucket(); - add_pbucket(mod,bucket,g); - d = 0; - mul = (NM)ALLOCA(sizeof(struct oNM)+nd_wpd*sizeof(unsigned int)); - while ( 1 ) { - hindex = mod?head_pbucket(mod,bucket):head_pbucket_q(bucket); - if ( hindex < 0 ) { - if ( d ) - SG(d) = sugar; - *rp = d; - return 1; - } - g = bucket->body[hindex]; - index = nd_find_reducer_direct(g,ps,len); - if ( index >= 0 ) { - p = ps[index]; - ndl_sub(HDL(g),HDL(p),DL(mul)); - if ( ndl_check_bound2_direct(HDL(p),DL(mul)) ) { - nd_free(d); - free_pbucket(bucket); - *rp = 0; - return 0; - } - if ( mod ) { - c1 = invm(HCM(p),mod); c2 = mod-HCM(g); - DMAR(c1,c2,0,mod,c); CM(mul) = c; - } else { - igcd_cofactor(HCQ(g),HCQ(p),&gcd,&cg,&cred); - chsgnq(cg,&CQ(mul)); - nd_mul_c_q(d,cred); - mulq_pbucket(bucket,cred); - g = bucket->body[hindex]; - gmag = (double)p_mag((P)HCQ(g)); - } - red = ndv_mul_nm(mod,p,mul); - bucket->body[hindex] = nd_remove_head(g); - red = nd_remove_head(red); - add_pbucket(mod,bucket,red); - sugar = MAX(sugar,SG(p)+TD(DL(mul))); - if ( !mod && hmag && (gmag > hmag) ) { - g = normalize_pbucket(mod,bucket); - if ( !g ) { - if ( d ) SG(d) = sugar; - *rp = d; - return 1; - } - nd_removecont2(d,g); - hmag = ((double)p_mag((P)HCQ(g)))*nd_scale; - add_pbucket(mod,bucket,g); - } - } else if ( !full ) { - g = normalize_pbucket(mod,bucket); - if ( g ) SG(g) = sugar; - *rp = g; - return 1; - } else { - m = BDY(g); - if ( NEXT(m) ) { - BDY(g) = NEXT(m); NEXT(m) = 0; LEN(g)--; - } else { - FREEND(g); g = 0; - } - bucket->body[hindex] = g; - NEXT(m) = 0; - if ( d ) { - NEXT(tail)=m; tail=m; LEN(d)++; - } else { - MKND(n,m,1,d); tail = BDY(d); - } - } - } -} - -/* input : list of DP, cand : list of DP */ - -int nd_check_candidate(NODE input,NODE cand) -{ int n,i,stat; ND nf,d; - NODE t; + NDV r; + NODE t,s; + union oNDC dn; - nd_setup(0,cand); + ndv_setup(0,0,cand); n = length(cand); /* membercheck : list is a subset of Id(cand) ? */ for ( t = input; t; t = NEXT(t) ) { - d = dptond(0,(DP)BDY(t)); - stat = nd_nf_direct(0,d,nd_psq,n,0,&nf); - if ( !stat ) nd_reconstruct_direct(0,nd_psq,n); - else if ( nf ) return 0; +again: + if ( nd_bpe > obpe ) + r = ndv_dup_realloc((NDV)BDY(t),obpe,oadv,oepos); + else + r = (NDV)BDY(t); + d = ndvtond(0,r); + stat = nd_nf(0,d,nd_ps,0,0,&nf); + if ( !stat ) { + nd_reconstruct(0,0); + goto again; + } else if ( nf ) return 0; + if ( DP_Print ) { printf("."); fflush(stdout); } } + if ( DP_Print ) { printf("\n"); } /* gbcheck : cand is a GB of Id(cand) ? */ - if ( !nd_gb(0,1) ) return 0; + if ( !nd_gb(0,0,1) ) return 0; /* XXX */ return 1; } @@ -1279,6 +1313,24 @@ ND nd_remove_head(ND p) return p; } +ND nd_separate_head(ND p,ND *head) +{ + NM m,m0; + ND r; + + m = BDY(p); + if ( !NEXT(m) ) { + *head = p; p = 0; + } else { + m0 = m; + BDY(p) = NEXT(m); LEN(p)--; + NEXT(m0) = 0; + MKND(NV(p),m0,1,r); + *head = r; + } + return p; +} + PGeoBucket create_pbucket() { PGeoBucket g; @@ -1299,6 +1351,24 @@ void free_pbucket(PGeoBucket b) { GC_free(b); } +void add_pbucket_symbolic(PGeoBucket g,ND d) +{ + int l,i,k,m; + + if ( !d ) + return; + l = LEN(d); + for ( k = 0, m = 1; l > m; k++, m <<= 1 ); + /* 2^(k-1) < l <= 2^k (=m) */ + d = nd_merge(g->body[k],d); + for ( ; d && LEN(d) > m; k++, m <<= 1 ) { + g->body[k] = 0; + d = nd_merge(g->body[k+1],d); + } + g->body[k] = d; + g->m = MAX(g->m,k); +} + void add_pbucket(int mod,PGeoBucket g,ND d) { int l,i,k,m; @@ -1325,10 +1395,42 @@ void mulq_pbucket(PGeoBucket g,Q c) nd_mul_c_q(g->body[k],c); } +NM remove_head_pbucket_symbolic(PGeoBucket g) +{ + int j,i,k,c; + NM head; + + k = g->m; + j = -1; + for ( i = 0; i <= k; i++ ) { + if ( !g->body[i] ) continue; + if ( j < 0 ) j = i; + else { + c = DL_COMPARE(HDL(g->body[i]),HDL(g->body[j])); + if ( c > 0 ) + j = i; + else if ( c == 0 ) + g->body[i] = nd_remove_head(g->body[i]); + } + } + if ( j < 0 ) return 0; + else { + head = BDY(g->body[j]); + if ( !NEXT(head) ) { + FREEND(g->body[j]); + g->body[j] = 0; + } else { + BDY(g->body[j]) = NEXT(head); + LEN(g->body[j])--; + } + return head; + } +} + int head_pbucket(int mod,PGeoBucket g) { int j,i,c,k,nv,sum; - unsigned int *di,*dj; + UINT *di,*dj; ND gi,gj; k = g->m; @@ -1352,8 +1454,12 @@ int head_pbucket(int mod,PGeoBucket g) dj = HDL(gj); sum = HCM(gj); } else if ( c == 0 ) { - sum = sum+HCM(gi)-mod; - if ( sum < 0 ) sum += mod; + if ( mod == -1 ) + sum = _addsf(sum,HCM(gi)); + else { + sum = sum+HCM(gi)-mod; + if ( sum < 0 ) sum += mod; + } g->body[i] = nd_remove_head(gi); } } @@ -1421,15 +1527,57 @@ ND normalize_pbucket(int mod,PGeoBucket g) return r; } +void do_diagonalize(int sugar,int m) +{ + int i,nh,stat; + NODE r,g,t; + ND h,nf,s,head; + NDV nfv; + Q q,num,den; + union oNDC dn; + + for ( i = nd_psn-1; i >= 0 && SG(nd_psh[i]) == sugar; i-- ) { + if ( nd_demand ) + nfv = ndv_load(i); + else + nfv = nd_ps[i]; + s = ndvtond(m,nfv); + s = nd_separate_head(s,&head); + nd_nf(m,s,nd_ps,1,&dn,&nf); + if ( !m ) { + NTOQ(NM(dn.z),SGN(dn.z),num); + mulq(HCQ(head),num,&q); HCQ(head) = q; + if ( DN(dn.z) ) { + NTOQ(DN(dn.z),1,den); + nd_mul_c_q(nf,den); + } + } + nf = nd_add(m,head,nf); + ndv_free(nfv); + nd_removecont(m,nf); + nfv = ndtondv(m,nf); + nd_free(nf); + nd_bound[i] = ndv_compute_bound(nfv); + if ( nd_demand ) { + ndv_save(nfv,i); + ndv_free(nfv); + } else + nd_ps[i] = nfv; + } +} + /* return value = 0 => input is not a GB */ -NODE nd_gb(int m,int checkonly) +NODE nd_gb(int m,int ishomo,int checkonly) { int i,nh,sugar,stat; NODE r,g,t; ND_pairs d; ND_pairs l; - ND h,nf; + ND h,nf,s,head; + NDV nfv; + Q q,num,den; + union oNDC dn; g = 0; d = 0; for ( i = 0; i < nd_psn; i++ ) { @@ -1441,46 +1589,106 @@ NODE nd_gb(int m,int checkonly) again: l = nd_minp(d,&d); if ( SG(l) != sugar ) { + if ( ishomo ) do_diagonalize(sugar,m); + sugar = SG(l); - fprintf(asir_out,"%d",sugar); + if ( DP_Print ) 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); + d = nd_reconstruct(0,d); goto again; } - stat = m?nd_nf_pbucket(m,h,!Top,&nf):nd_nf(m,h,!Top,&nf); +#if USE_GEOBUCKET + stat = m?nd_nf_pbucket(m,h,nd_ps,!Top,&nf):nd_nf(m,h,nd_ps,!Top,0,&nf); +#else + stat = nd_nf(m,h,nd_ps,!Top,0,&nf); +#endif if ( !stat ) { NEXT(l) = d; d = l; - d = nd_reconstruct(m,0,d); + d = nd_reconstruct(0,d); goto again; } else if ( nf ) { if ( checkonly ) return 0; - printf("+"); fflush(stdout); - nh = nd_newps(m,nf); + if ( DP_Print ) { printf("+"); fflush(stdout); } + nd_removecont(m,nf); + nfv = ndtondv(m,nf); nd_free(nf); + nh = ndv_newps(m,nfv,0); d = update_pairs(d,g,nh); g = update_base(g,nh); FREENDP(l); } else { - printf("."); fflush(stdout); + if ( DP_Print ) { printf("."); fflush(stdout); } FREENDP(l); } } - if ( m ) - for ( t = g; t; t = NEXT(t) ) BDY(t) = (pointer)nd_ps[(int)BDY(t)]; + if ( nd_demand ) + for ( t = g; t; t = NEXT(t) ) + BDY(t) = (pointer)ndv_load((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; } -NODE nd_gb_trace(int m) +void do_diagonalize_trace(int sugar,int m) { + int i,nh,stat; + NODE r,g,t; + ND h,nf,nfq,s,head; + NDV nfv,nfqv; + Q q,den,num; + union oNDC dn; + + for ( i = nd_psn-1; i >= 0 && SG(nd_psh[i]) == sugar; i-- ) { + /* for nd_ps */ + s = ndvtond(m,nd_ps[i]); + s = nd_separate_head(s,&head); + nd_nf_pbucket(m,s,nd_ps,1,&nf); + nf = nd_add(m,head,nf); + ndv_free(nd_ps[i]); + nd_ps[i] = ndtondv(m,nf); + nd_free(nf); + + /* for nd_ps_trace */ + if ( nd_demand ) + nfv = ndv_load(i); + else + nfv = nd_ps_trace[i]; + s = ndvtond(0,nfv); + s = nd_separate_head(s,&head); + nd_nf(0,s,nd_ps_trace,1,&dn,&nf); + NTOQ(NM(dn.z),SGN(dn.z),num); + mulq(HCQ(head),num,&q); HCQ(head) = q; + if ( DN(dn.z) ) { + NTOQ(DN(dn.z),1,den); + nd_mul_c_q(nf,den); + } + nf = nd_add(0,head,nf); + ndv_free(nfv); + nd_removecont(0,nf); + nfv = ndtondv(0,nf); + nd_free(nf); + nd_bound[i] = ndv_compute_bound(nfv); + if ( nd_demand ) { + ndv_save(nfv,i); + ndv_free(nfv); + } else + nd_ps_trace[i] = nfv; + } +} + +NODE nd_gb_trace(int m,int ishomo) +{ int i,nh,sugar,stat; NODE r,g,t; ND_pairs d; ND_pairs l; - ND h,nf,nfq; + ND h,nf,nfq,s,head; + NDV nfv,nfqv; + Q q,den,num; + union oNDC dn; g = 0; d = 0; for ( i = 0; i < nd_psn; i++ ) { @@ -1492,41 +1700,62 @@ NODE nd_gb_trace(int m) again: l = nd_minp(d,&d); if ( SG(l) != sugar ) { + if ( ishomo ) do_diagonalize_trace(sugar,m); sugar = SG(l); - fprintf(asir_out,"%d",sugar); + if ( DP_Print ) 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); + d = nd_reconstruct(1,d); goto again; } - stat = nd_nf_pbucket(m,h,!Top,&nf); +#if USE_GEOBUCKET + stat = nd_nf_pbucket(m,h,nd_ps,!Top,&nf); +#else + stat = nd_nf(m,h,nd_ps,!Top,0,&nf); +#endif if ( !stat ) { NEXT(l) = d; d = l; - d = nd_reconstruct(m,1,d); + d = nd_reconstruct(1,d); goto again; } else if ( nf ) { - /* overflow does not occur */ - nd_sp(0,l,&h); - nd_nf(0,h,!Top,&nfq); + if ( nd_demand ) { + nfqv = ndv_load(nd_psn); + nfq = ndvtond(0,nfqv); + } else + nfq = 0; + if ( !nfq ) { + if ( !nd_sp(0,1,l,&h) || !nd_nf(0,h,nd_ps_trace,!Top,0,&nfq) ) { + NEXT(l) = d; d = l; + d = nd_reconstruct(1,d); + goto again; + } + } if ( nfq ) { - printf("+"); fflush(stdout); - nh = nd_newps_trace(m,nf,nfq); - /* failure; m|HC(nfq) */ - if ( nh < 0 ) return 0; + /* m|HC(nfq) => failure */ + if ( !rem(NM(HCQ(nfq)),m) ) return 0; + + if ( DP_Print ) { printf("+"); fflush(stdout); } + nd_removecont(m,nf); nfv = ndtondv(m,nf); nd_free(nf); + nd_removecont(0,nfq); nfqv = ndtondv(0,nfq); nd_free(nfq); + nh = ndv_newps(0,nfv,nfqv); d = update_pairs(d,g,nh); g = update_base(g,nh); } else { - printf("*"); fflush(stdout); + if ( DP_Print ) { printf("*"); fflush(stdout); } } } else { - printf("."); fflush(stdout); + if ( DP_Print ) { printf("."); fflush(stdout); } } FREENDP(l); } - for ( t = g; t; t = NEXT(t) ) - BDY(t) = (pointer)nd_psq[(int)BDY(t)]; + if ( nd_demand ) + for ( t = g; t; t = NEXT(t) ) + BDY(t) = (pointer)ndv_load((int)BDY(t)); + else + for ( t = g; t; t = NEXT(t) ) + BDY(t) = (pointer)nd_ps_trace[(int)BDY(t)]; return g; } @@ -1540,42 +1769,52 @@ int ndv_compare_rev(NDV *p1,NDV *p2) return -DL_COMPARE(HDL(*p1),HDL(*p2)); } -NODE nd_reduceall(int m,NODE f) +NODE ndv_reduceall(int m,NODE f) { - int i,j,n,stat; - NDV *w,*ps; - ND nf,g; + int i,n,stat; + ND nf,g,head; NODE t,a0,a; + union oNDC dn; + NDV *w; + Q q,num,den; - for ( n = 0, t = f; t; t = NEXT(t), n++ ); - ps = (NDV *)ALLOCA(n*sizeof(NDV)); - 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); - w = (NDV *)ALLOCA((n-1)*sizeof(NDV)); - for ( i = 0; i < n; i++ ) { - for ( j = 0; j < i; j++ ) w[j] = (NDV)ps[j]; - for ( j = i+1; j < n; j++ ) w[j-1] = ps[j]; - g = ndvtond(m,ps[i]); - stat = nd_nf_direct(m,g,w,n-1,1,&nf); + n = length(f); +#if 0 + w = (NDV *)ALLOCA(n*sizeof(NDV)); + for ( i = 0, t = f; i < n; i++, t = NEXT(t) ) w[i] = (NDV)BDY(t); + qsort(w,n,sizeof(NDV), + (int (*)(const void *,const void *))ndv_compare); + for ( t = f, i = 0; t; i++, t = NEXT(t) ) BDY(t) = (pointer)w[i]; +#endif + ndv_setup(m,0,f); + for ( i = 0; i < n; ) { + g = ndvtond(m,nd_ps[i]); + g = nd_separate_head(g,&head); + stat = nd_nf(m,g,nd_ps,1,&dn,&nf); if ( !stat ) - nd_reconstruct_direct(m,ps,n); - else if ( !nf ) { - printf("."); fflush(stdout); - ndv_free(ps[i]); - for ( j = i+1; j < n; j++ ) ps[j-1] = ps[j]; - n--; - } else { - printf("."); fflush(stdout); - ndv_free(ps[i]); + nd_reconstruct(0,0); + else { + if ( DP_Print ) { printf("."); fflush(stdout); } + if ( !m ) { + NTOQ(NM(dn.z),SGN(dn.z),num); + mulq(HCQ(head),num,&q); HCQ(head) = q; + if ( DN(dn.z) ) { + NTOQ(DN(dn.z),1,den); + nd_mul_c_q(nf,den); + } + } + nf = nd_add(m,head,nf); + ndv_free(nd_ps[i]); nd_removecont(m,nf); - ps[i] = ndtondv(m,nf); - nd_free(nf); + nd_ps[i] = ndtondv(m,nf); nd_free(nf); + nd_bound[i] = ndv_compute_bound(nd_ps[i]); + i++; } } + if ( DP_Print ) { printf("\n"); } for ( a0 = 0, i = 0; i < n; i++ ) { NEXTNODE(a0,a); - BDY(a) = (pointer)ps[i]; + BDY(a) = (pointer)nd_ps[i]; } NEXT(a) = 0; return a0; @@ -1590,15 +1829,19 @@ 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 ) @@ -1614,7 +1857,7 @@ ND_pairs update_pairs( ND_pairs d, NODE /* of index */ ND_pairs nd_newpairs( NODE g, int t ) { NODE h; - unsigned int *dl; + UINT *dl; int ts,s; ND_pairs r,r0; @@ -1635,14 +1878,14 @@ ND_pairs nd_newpairs( NODE g, int t ) ND_pairs crit_B( ND_pairs d, int s ) { ND_pairs cur,head,prev,remove; - unsigned int *t,*tl,*lcm; + UINT *t,*tl,*lcm; int td,tdl; if ( !d ) return 0; t = DL(nd_psh[s]); prev = 0; head = cur = d; - lcm = (unsigned int *)ALLOCA((nd_wpd+1)*sizeof(unsigned int)); + lcm = (UINT *)ALLOCA(nd_wpd*sizeof(UINT)); while ( cur ) { tl = cur->lcm; if ( ndl_reducible(tl,t) @@ -1662,12 +1905,10 @@ 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; - unsigned int *id,*jd; + UINT *id,*jd; for ( dd = 0, e = d1; e; e = d3 ) { if ( !(d2 = NEXT(e)) ) { @@ -1750,10 +1991,10 @@ int crit_2( int dp1, int dp2 ) return ndl_disjoint(DL(nd_psh[dp1]),DL(nd_psh[dp2])); } -static ND_pairs equivalent_pairs( ND_pairs d1, ND_pairs *prest ) +ND_pairs equivalent_pairs( ND_pairs d1, ND_pairs *prest ) { ND_pairs w,p,r,s; - unsigned int *d; + UINT *d; w = d1; d = LCM(w); @@ -1762,11 +2003,9 @@ static ND_pairs equivalent_pairs( ND_pairs d1, ND_pair for ( r = 0; s; s = p ) { p = NEXT(s); if ( ndl_equal(d,LCM(s)) ) { - NEXT(s) = w; - w = s; + NEXT(s) = w; w = s; } else { - NEXT(s) = r; - r = s; + NEXT(s) = r; r = s; } } *prest = r; @@ -1775,7 +2014,7 @@ static ND_pairs equivalent_pairs( ND_pairs d1, ND_pair NODE update_base(NODE nd,int ndp) { - unsigned int *dl, *dln; + UINT *dl, *dln; NODE last, p, head; dl = DL(nd_psh[ndp]); @@ -1796,7 +2035,7 @@ NODE update_base(NODE nd,int ndp) ND_pairs nd_minp( ND_pairs d, ND_pairs *prest ) { ND_pairs m,ml,p,l; - unsigned int *lcm; + UINT *lcm; int s,td,len,tlen,c,c1; if ( !(p = NEXT(m = d)) ) { @@ -1805,13 +2044,18 @@ ND_pairs nd_minp( ND_pairs d, ND_pairs *prest ) return m; } s = SG(m); - for ( ml = 0, l = m; p; p = NEXT(l = p) ) - if ( (SG(p) < s) - || ((SG(p) == s) && (DL_COMPARE(LCM(p),LCM(m)) < 0)) ) { - ml = l; - m = p; - s = SG(m); - } + if ( !NoSugar ) { + for ( ml = 0, l = m; p; p = NEXT(l = p) ) + if ( (SG(p) < s) + || ((SG(p) == s) && (DL_COMPARE(LCM(p),LCM(m)) < 0)) ) { + ml = l; m = p; s = SG(m); + } + } else { + for ( ml = 0, l = m; p; p = NEXT(l = p) ) + if ( DL_COMPARE(LCM(p),LCM(m)) < 0 ) { + ml = l; m = p; s = SG(m); + } + } if ( !ml ) *prest = NEXT(m); else { NEXT(ml) = NEXT(m); @@ -1821,35 +2065,32 @@ ND_pairs nd_minp( ND_pairs d, ND_pairs *prest ) return m; } -int nd_newps(int mod,ND a) +ND_pairs nd_minsugarp( ND_pairs d, ND_pairs *prest ) { - int len; - RHist r; - NDV b; + int msugar,i; + ND_pairs t,dm0,dm,dr0,dr; - 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_psh = (RHist *)REALLOC((char *)nd_psh,nd_pslen*sizeof(RHist)); - nd_bound = (unsigned int **) - REALLOC((char *)nd_bound,nd_pslen*sizeof(unsigned int *)); - } - nd_removecont(mod,a); - nd_bound[nd_psn] = nd_compute_bound(a); - NEWRHist(r); SG(r) = SG(a); ndl_copy(HDL(a),DL(r)); - nd_psh[nd_psn] = r; - b = ndtondv(mod,a); - len = LEN(b); - if ( mod ) - nd_ps[nd_psn] = b; - else - nd_psq[nd_psn] = b; - nd_free(a); - return nd_psn++; + for ( msugar = SG(d), t = NEXT(d); t; t = NEXT(t) ) + if ( SG(t) < msugar ) msugar = SG(t); + dm0 = 0; dr0 = 0; + for ( i = 0, t = d; t; t = NEXT(t) ) + if ( i < nd_f4_nsp && SG(t) == msugar ) { + if ( dm0 ) NEXT(dm) = t; + else dm0 = t; + dm = t; + i++; + } else { + if ( dr0 ) NEXT(dr) = t; + else dr0 = t; + dr = t; + } + NEXT(dm) = 0; + if ( dr0 ) NEXT(dr) = 0; + *prest = dr0; + return dm0; } -int nd_newps_trace(int mod,ND nf,ND nfq) +int ndv_newps(int m,NDV a,NDV aq) { int len; RHist r; @@ -1858,318 +2099,320 @@ int nd_newps_trace(int mod,ND nf,ND nfq) 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 *)); + nd_bound = (UINT **) + REALLOC((char *)nd_bound,nd_pslen*sizeof(UINT *)); } - if ( !rem(NM(HCQ(nfq)),mod) ) return -1; - nd_removecont(mod,nf); - nd_ps[nd_psn] = ndtondv(mod,nf); - - nd_removecont(0,nfq); - nd_psq[nd_psn] = ndtondv(0,nfq); - - nd_bound[nd_psn] = nd_compute_bound(nfq); - NEWRHist(r); SG(r) = SG(nf); ndl_copy(HDL(nf),DL(r)); - nd_psh[nd_psn] = r; - nd_free(nf); nd_free(nfq); - return nd_psn++; -} - -void nd_setup(int mod,NODE f) -{ - int i,j,td,len,max; - NODE s,s0,f0; - unsigned int *d; - RHist r; - NDV a; - - 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_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); + NEWRHist(r); nd_psh[nd_psn] = r; + nd_ps[nd_psn] = a; + if ( aq ) { + nd_ps_trace[nd_psn] = aq; + nd_bound[nd_psn] = ndv_compute_bound(aq); + SG(r) = SG(aq); ndl_copy(HDL(aq),DL(r)); + } else { + nd_bound[nd_psn] = ndv_compute_bound(a); + SG(r) = SG(a); ndl_copy(HDL(a),DL(r)); } - if ( !nd_red ) - nd_red = (RHist *)MALLOC(REDTAB_LEN*sizeof(RHist)); - bzero(nd_red,REDTAB_LEN*sizeof(RHist)); - - if ( max < 2 ) nd_bpe = 2; - else if ( max < 4 ) nd_bpe = 4; - else if ( max < 64 ) nd_bpe = 6; - else if ( max < 256 ) nd_bpe = 8; - else if ( max < 65536 ) nd_bpe = 16; - else nd_bpe = 32; - - nd_setup_parameters(); - nd_free_private_storage(); - len = 0; - for ( i = 0; i < nd_psn; i++, f = NEXT(f) ) { - NEWRHist(r); - a = dptondv(mod,(DP)BDY(f)); - ndv_removecont(mod,a); - len = MAX(len,LEN(a)); - SG(r) = HTD(a); ndl_copy(HDL(a),DL(r)); - if ( mod ) nd_ps[i] = a; - else nd_psq[i] = a; - nd_psh[i] = r; + if ( nd_demand ) { + if ( aq ) { + ndv_save(nd_ps_trace[nd_psn],nd_psn); + nd_ps_trace[nd_psn] = 0; + } else { + ndv_save(nd_ps[nd_psn],nd_psn); + nd_ps[nd_psn] = 0; + } } + return nd_psn++; } -void nd_setup_trace(int mod,NODE f) +void ndv_setup(int mod,int trace,NODE f) { int i,j,td,len,max; NODE s,s0,f0; - unsigned int *d; + UINT *d; RHist r; - NDV a; + NDV *w; + NDV a,am; nd_found = 0; nd_notfirst = 0; nd_create = 0; - nd_psn = length(f); nd_pslen = 2*nd_psn; + for ( nd_psn = 0, s = f; s; s = NEXT(s) ) if ( BDY(s) ) nd_psn++; + w = (NDV *)ALLOCA(nd_psn*sizeof(NDV)); + for ( i = 0, s = f; s; s = NEXT(s) ) if ( BDY(s) ) w[i++] = BDY(s); + qsort(w,nd_psn,sizeof(NDV), + (int (*)(const void *,const void *))ndv_compare); + 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); - } + nd_bound = (UINT **)MALLOC(nd_pslen*sizeof(UINT *)); + if ( !nd_red ) nd_red = (RHist *)MALLOC(REDTAB_LEN*sizeof(RHist)); - bzero(nd_red,REDTAB_LEN*sizeof(RHist)); - - if ( max < 2 ) nd_bpe = 2; - else if ( max < 4 ) nd_bpe = 4; - else if ( max < 64 ) nd_bpe = 6; - else if ( max < 256 ) nd_bpe = 8; - else if ( max < 65536 ) nd_bpe = 16; - else nd_bpe = 32; - - nd_setup_parameters(); - nd_free_private_storage(); - len = 0; - for ( i = 0; i < nd_psn; i++, f = NEXT(f) ) { - a = dptondv(mod,(DP)BDY(f)); ndv_removecont(mod,a); nd_ps[i] = a; - a = dptondv(0,(DP)BDY(f)); ndv_removecont(0,a); nd_psq[i] = a; - NEWRHist(r); - len = MAX(len,LEN(a)); - SG(r) = HTD(a); ndl_copy(HDL(a),DL(r)); + for ( i = 0; i < REDTAB_LEN; i++ ) nd_red[i] = 0; + for ( i = 0; i < nd_psn; i++ ) { + if ( trace ) { + a = nd_ps_trace[i] = ndv_dup(0,w[i]); + ndv_removecont(0,a); + am = nd_ps[i] = ndv_dup(mod,a); + ndv_mod(mod,am); + ndv_removecont(mod,am); + } else { + a = nd_ps[i] = ndv_dup(mod,w[i]); + ndv_removecont(mod,a); + } + NEWRHist(r); SG(r) = HTD(a); ndl_copy(HDL(a),DL(r)); + nd_bound[i] = ndv_compute_bound(a); nd_psh[i] = r; + if ( nd_demand ) { + if ( trace ) { + ndv_save(nd_ps_trace[i],i); + nd_ps_trace[i] = 0; + } else { + ndv_save(nd_ps[i],i); + nd_ps[i] = 0; + } + } } } -void nd_gr(LIST f,LIST v,int m,struct order_spec *ord,LIST *rp) +void nd_gr(LIST f,LIST v,int m,int f4,struct order_spec *ord,LIST *rp) { - struct order_spec ord1; - VL fv,vv,vc; + VL tv,fv,vv,vc; NODE fd,fd0,r,r0,t,x,s,xx; - DP a,b,c; + int e,max,nvar; + NDV b; + int ishomo; + if ( !m && Demand ) nd_demand = 1; + else nd_demand = 0; + +#if 0 + ndv_alloc = 0; +#endif get_vars((Obj)f,&fv); pltovl(v,&vv); - nd_nvar = length(vv); + for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++ ); + switch ( ord->id ) { + case 1: + if ( ord->nv != nvar ) + error("nd_{gr,f4} : invalid order specification"); + break; + default: + break; + } nd_init_ord(ord); - /* XXX for DP */ - initd(ord); + for ( t = BDY(f), max = 0; t; t = NEXT(t) ) + for ( tv = vv; tv; tv = NEXT(tv) ) { + e = getdeg(tv->v,(P)BDY(t)); + max = MAX(e,max); + } + nd_setup_parameters(nvar,max); + ishomo = 1; for ( fd0 = 0, t = BDY(f); t; t = NEXT(t) ) { - ptod(CO,vv,(P)BDY(t),&b); - NEXTNODE(fd0,fd); BDY(fd) = (pointer)b; + b = (pointer)ptondv(CO,vv,(P)BDY(t)); + if ( ishomo ) + ishomo = ishomo && ndv_ishomo(b); + if ( m ) ndv_mod(m,b); + if ( b ) { NEXTNODE(fd0,fd); BDY(fd) = (pointer)b; } } if ( fd0 ) NEXT(fd) = 0; - nd_setup(m,fd0); - x = nd_gb(m,0); - fprintf(asir_out,"found=%d,notfirst=%d,create=%d\n", - nd_found,nd_notfirst,nd_create); - x = nd_reduceall(m,x); + ndv_setup(m,0,fd0); + x = f4?nd_f4(m):nd_gb(m,ishomo,0); + nd_demand = 0; + x = ndv_reducebase(x); + x = ndv_reduceall(m,x); for ( r0 = 0, t = x; t; t = NEXT(t) ) { NEXTNODE(r0,r); - if ( m ) { - a = ndvtodp(m,BDY(t)); - _dtop_mod(CO,vv,a,(P *)&BDY(r)); - } else { - a = ndvtodp(m,BDY(t)); - dtop(CO,vv,a,(P *)&BDY(r)); - } + BDY(r) = ndvtop(m,CO,vv,BDY(t)); } if ( r0 ) NEXT(r) = 0; MKLIST(*rp,r0); +#if 0 + fprintf(asir_out,"ndv_alloc=%d\n",ndv_alloc); +#endif } -void nd_gr_trace(LIST f,LIST v,int m,int homo,struct order_spec *ord,LIST *rp) +void nd_gr_trace(LIST f,LIST v,int trace,int homo,struct order_spec *ord,LIST *rp) { - struct order_spec ord1; - VL fv,vv,vc; + struct order_spec *ord1; + VL tv,fv,vv,vc; NODE fd,fd0,in0,in,r,r0,t,s,cand; - DP a,b,c,h; + int m,nocheck,nvar,mindex,e,max; + NDV c; + NMV a; P p; + EPOS oepos; + int obpe,oadv,wmax,i,len,cbpe,ishomo; get_vars((Obj)f,&fv); pltovl(v,&vv); - nd_nvar = length(vv); - initd(ord); - if ( homo ) { - homogenize_order(ord,nd_nvar,&ord1); - for ( fd0 = 0, in0 = 0, t = BDY(f); t; t = NEXT(t) ) { - ptod(CO,vv,(P)BDY(t),&c); - if ( c ) { - dp_homo(c,&h); NEXTNODE(fd0,fd); BDY(fd) = (pointer)h; - NEXTNODE(in0,in); BDY(in) = (pointer)c; - } + for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++ ); + switch ( ord->id ) { + case 1: + if ( ord->nv != nvar ) + error("nd_gr_trace : invalid order specification"); + break; + default: + break; + } + nocheck = 0; + mindex = 0; + + if ( Demand ) nd_demand = 1; + else nd_demand = 0; + + /* setup modulus */ + if ( trace < 0 ) { + trace = -trace; + nocheck = 1; + } + m = trace > 1 ? trace : get_lprime(mindex); + for ( t = BDY(f), max = 0; t; t = NEXT(t) ) + for ( tv = vv; tv; tv = NEXT(tv) ) { + e = getdeg(tv->v,(P)BDY(t)); + max = MAX(e,max); } - if ( fd0 ) NEXT(fd) = 0; - if ( in0 ) NEXT(in) = 0; - nd_init_ord(&ord1); - initd(&ord1); - nd_nvar++; - } else { - for ( fd0 = 0, t = BDY(f); t; t = NEXT(t) ) { - ptod(CO,vv,(P)BDY(t),&c); - if ( c ) { - NEXTNODE(fd0,fd); BDY(fd) = (pointer)c; - } + nd_init_ord(ord); + nd_setup_parameters(nvar,max); + obpe = nd_bpe; oadv = nmv_adv; oepos = nd_epos; + ishomo = 1; + for ( in0 = 0, fd0 = 0, t = BDY(f); t; t = NEXT(t) ) { + c = ptondv(CO,vv,(P)BDY(t)); + if ( ishomo ) + ishomo = ishomo && ndv_ishomo(c); + if ( c ) { + NEXTNODE(in0,in); BDY(in) = (pointer)c; + NEXTNODE(fd0,fd); BDY(fd) = (pointer)ndv_dup(0,c); } - if ( fd0 ) NEXT(fd) = 0; - in0 = fd0; - nd_init_ord(ord); } - do { - nd_setup_trace(m,fd0); - cand = nd_gb_trace(m); - if ( !cand ) continue; - if ( homo ) { + if ( in0 ) NEXT(in) = 0; + if ( fd0 ) NEXT(fd) = 0; + if ( !ishomo && homo ) { + for ( t = in0, wmax = 0; t; t = NEXT(t) ) { + c = (NDV)BDY(t); len = LEN(c); + for ( a = BDY(c), i = 0; i < len; i++, NMV_ADV(a) ) + wmax = MAX(TD(DL(a)),wmax); + } + homogenize_order(ord,nvar,&ord1); + nd_init_ord(ord1); + nd_setup_parameters(nvar+1,wmax); + for ( t = fd0; t; t = NEXT(t) ) + ndv_homogenize((NDV)BDY(t),obpe,oadv,oepos); + } + while ( 1 ) { + if ( Demand ) + nd_demand = 1; + ndv_setup(m,1,fd0); + cand = nd_gb_trace(m,ishomo || homo); + if ( !cand ) { + /* failure */ + if ( trace > 1 ) { *rp = 0; return; } + else m = get_lprime(++mindex); + continue; + } + if ( !ishomo && homo ) { /* dehomogenization */ - for ( t = cand; t; t = NEXT(t) ) - ndv_dehomogenize((NDV)BDY(t)); - nd_nvar--; - nd_setup_parameters(); - initd(ord); - cand = nd_reducebase(cand); + for ( t = cand; t; t = NEXT(t) ) ndv_dehomogenize((NDV)BDY(t),ord); + nd_init_ord(ord); + nd_setup_parameters(nvar,0); } - fprintf(asir_out,"found=%d,notfirst=%d,create=%d\n", - nd_found,nd_notfirst,nd_create); - cand = nd_reduceall(0,cand); - initd(ord); - for ( r0 = 0; cand; cand = NEXT(cand) ) { - NEXTNODE(r0,r); - BDY(r) = (pointer)ndvtodp(0,(NDV)BDY(cand)); + nd_demand = 0; + cand = ndv_reducebase(cand); + cand = ndv_reduceall(0,cand); + cbpe = nd_bpe; + if ( nocheck ) + break; + if ( ndv_check_candidate(in0,obpe,oadv,oepos,cand) ) + /* success */ + break; + else if ( trace > 1 ) { + /* failure */ + *rp = 0; return; + } else { + /* try the next modulus */ + m = get_lprime(++mindex); + /* reset the parameters */ + if ( !ishomo && homo ) { + nd_init_ord(ord1); + nd_setup_parameters(nvar+1,wmax); + } else { + nd_init_ord(ord); + nd_setup_parameters(nvar,max); + } } - if ( r0 ) NEXT(r) = 0; - cand = r0; - } while ( !nd_check_candidate(in0,cand) ); - /* dp->p */ - for ( r = cand; r; r = NEXT(r) ) { - dtop(CO,vv,BDY(r),&p); - BDY(r) = (pointer)p; } + /* dp->p */ + nd_bpe = cbpe; + nd_setup_parameters(nd_nvar,0); + for ( r = cand; r; r = NEXT(r) ) BDY(r) = (pointer)ndvtop(0,CO,vv,BDY(r)); MKLIST(*rp,cand); } -void dltondl(int n,DL dl,unsigned int *r) +void dltondl(int n,DL dl,UINT *r) { - unsigned int *d; - int i; + UINT *d; + int i,j,l,s,ord_l; + struct order_pair *op; d = dl->d; - for ( i = 0; i <= nd_wpd; i++ ) r[i] = 0; - if ( nd_isrlex ) - for ( i = 0; i < n; i++ ) - r[(n-1-i)/nd_epw+1] |= (d[i]<<((nd_epw-((n-1-i)%nd_epw)-1)*nd_bpe)); - else - for ( i = 0; i < n; i++ ) - r[i/nd_epw+1] |= d[i]<<((nd_epw-(i%nd_epw)-1)*nd_bpe); - TD(r) = ndl_td(r); + for ( i = 0; i < nd_wpd; i++ ) r[i] = 0; + if ( nd_blockmask ) { + l = nd_blockmask->n; + op = nd_blockmask->order_pair; + for ( j = 0, s = 0; j < l; j++ ) { + ord_l = op[j].length; + for ( i = 0; i < ord_l; i++, s++ ) PUT_EXP(r,s,d[s]); + } + TD(r) = ndl_weight(r); + ndl_weight_mask(r); + } else { + for ( i = 0; i < n; i++ ) PUT_EXP(r,i,d[i]); + TD(r) = ndl_weight(r); + } } -DL ndltodl(int n,unsigned int *ndl) +DL ndltodl(int n,UINT *ndl) { DL dl; int *d; - int i; + int i,j,l,s,ord_l; + struct order_pair *op; NEWDL(dl,n); dl->td = TD(ndl); d = dl->d; - if ( nd_isrlex ) - for ( i = 0; i < n; i++ ) - d[i] = (ndl[(n-1-i)/nd_epw+1]>>((nd_epw-((n-1-i)%nd_epw)-1)*nd_bpe)) - &((1<>((nd_epw-(i%nd_epw)-1)*nd_bpe)) - &((1<n; + op = nd_blockmask->order_pair; + for ( j = 0, s = 0; j < l; j++ ) { + ord_l = op[j].length; + for ( i = 0; i < ord_l; i++, s++ ) d[s] = GET_EXP(ndl,s); + } + } else { + for ( i = 0; i < n; i++ ) d[i] = GET_EXP(ndl,i); + } return dl; } -ND dptond(int mod,DP p) +void ndl_print(UINT *dl) { - ND d; - NM m0,m; - MP t; - int n,l; - - if ( !p ) - return 0; - n = NV(p); - m0 = 0; - for ( t = BDY(p), l = 0; t; t = NEXT(t), l++ ) { - NEXTNM(m0,m); - if ( mod ) CM(m) = ITOS(C(t)); - else CQ(m) = (Q)C(t); - dltondl(n,DL(t),DL(m)); - } - NEXT(m) = 0; - MKND(n,m0,l,d); - SG(d) = SG(p); - return d; -} - -DP ndtodp(int mod,ND p) -{ - DP d; - MP m0,m; - NM t; int n; + int i,j,l,ord_l,s,s0; + struct order_pair *op; - if ( !p ) - return 0; - n = NV(p); - m0 = 0; - for ( t = BDY(p); t; t = NEXT(t) ) { - NEXTMP(m0,m); - if ( mod ) C(m) = STOI(CM(t)); - else C(m) = (P)CQ(t); - DL(m) = ndltodl(n,DL(t)); - } - NEXT(m) = 0; - MKDP(n,m0,d); - SG(d) = SG(p); - return d; -} - -void ndl_print(unsigned int *dl) -{ - int n; - int i; - n = nd_nvar; printf("<<"); - if ( nd_isrlex ) - for ( i = 0; i < n; i++ ) - printf(i==n-1?"%d":"%d,", - (dl[(n-1-i)/nd_epw+1]>>((nd_epw-((n-1-i)%nd_epw)-1)*nd_bpe)) - &((1<>((nd_epw-(i%nd_epw)-1)*nd_bpe)) - &((1<n; + op = nd_blockmask->order_pair; + for ( j = 0, s = s0 = 0; j < l; j++ ) { + ord_l = op[j].length; + for ( i = 0; i < ord_l; i++, s++ ) + printf(s==n-1?"%d":"%d,",GET_EXP(dl,s)); + } + } else { + for ( i = 0; i < n; i++ ) printf(i==n-1?"%d":"%d,",GET_EXP(dl,i)); + } printf(">>"); } @@ -2181,7 +2424,8 @@ void nd_print(ND p) printf("0\n"); else { for ( m = BDY(p); m; m = NEXT(m) ) { - printf("+%d*",CM(m)); + if ( CM(m) & 0x80000000 ) printf("+@_%d*",IFTOF(CM(m))); + else printf("+%d*",CM(m)); ndl_print(DL(m)); } printf("\n"); @@ -2197,7 +2441,7 @@ void nd_print_q(ND p) else { for ( m = BDY(p); m; m = NEXT(m) ) { printf("+"); - printexpr(CO,CQ(m)); + printexpr(CO,(Obj)CQ(m)); printf("*"); ndl_print(DL(m)); } @@ -2222,7 +2466,8 @@ void nd_removecont(int mod,ND p) struct oVECT v; N q,r; - if ( mod ) nd_mul_c(mod,p,invm(HCM(p),mod)); + if ( mod == -1 ) nd_mul_c(mod,p,_invsf(HCM(p))); + else if ( mod ) nd_mul_c(mod,p,invm(HCM(p),mod)); else { for ( m = BDY(p), n = 0; m; m = NEXT(m), n++ ); w = (Q *)ALLOCA(n*sizeof(Q)); @@ -2268,7 +2513,9 @@ void ndv_removecont(int mod,NDV p) Q dvr,t; NMV m; - if ( mod ) + if ( mod == -1 ) + ndv_mul_c(mod,p,_invsf(HCM(p))); + else if ( mod ) ndv_mul_c(mod,p,invm(HCM(p),mod)); else { len = p->len; @@ -2282,25 +2529,47 @@ void ndv_removecont(int mod,NDV p) } } -void ndv_dehomogenize(NDV p) +void ndv_homogenize(NDV p,int obpe,int oadv,EPOS oepos) { - int i,len,newnvar,newwpd,newadv; + int len,i,max; + NMV m,mr0,mr,t; + + len = p->len; + for ( m = BDY(p), i = 0, max = 0; i < len; NMV_OADV(m), i++ ) + max = MAX(max,TD(DL(m))); + mr0 = nmv_adv>oadv?(NMV)REALLOC(BDY(p),len*nmv_adv):BDY(p); + m = (NMV)((char *)mr0+(len-1)*oadv); + mr = (NMV)((char *)mr0+(len-1)*nmv_adv); + t = (NMV)ALLOCA(nmv_adv); + for ( i = 0; i < len; i++, NMV_OPREV(m), NMV_PREV(mr) ) { + ndl_homogenize(DL(m),DL(t),obpe,oepos,max); + CQ(mr) = CQ(m); + ndl_copy(DL(t),DL(mr)); + } + NV(p)++; + BDY(p) = mr0; +} + +void ndv_dehomogenize(NDV p,struct order_spec *ord) +{ + int i,j,adj,len,newnvar,newwpd,newadv,newexporigin; Q *w; Q dvr,t; NMV m,r; - unsigned int *d; -#define NEWADV(m) (m = (NMV)(((char *)m)+newadv)) len = p->len; newnvar = nd_nvar-1; - newwpd = newnvar/nd_epw+(newnvar%nd_epw?1:0); + newexporigin = nd_get_exporigin(ord); + newwpd = newnvar/nd_epw+(newnvar%nd_epw?1:0)+newexporigin; for ( m = BDY(p), i = 0; i < len; NMV_ADV(m), i++ ) ndl_dehomogenize(DL(m)); if ( newwpd != nd_wpd ) { - d = (unsigned int *)ALLOCA((newwpd+1)*sizeof(unsigned int)); - newadv = sizeof(struct oNMV)+newwpd*sizeof(unsigned int); - for ( m = r = BDY(p), i = 0; i < len; NMV_ADV(m), NEWADV(r), i++ ) { - CQ(r) = CQ(m); ndl_copy(DL(m),d); ndl_copy(d,DL(r)); + newadv = ROUND_FOR_ALIGN(sizeof(struct oNMV)+(newwpd-1)*sizeof(UINT)); + for ( m = r = BDY(p), i = 0; i < len; NMV_ADV(m), NDV_NADV(r), i++ ) { + CQ(r) = CQ(m); + for ( j = 0; j < newexporigin; j++ ) DL(r)[j] = DL(m)[j]; + adj = nd_exporigin-newexporigin; + for ( ; j < newwpd; j++ ) DL(r)[j] = DL(m)[j+adj]; } } NV(p)--; @@ -2310,10 +2579,16 @@ void removecont_array(Q *c,int n) { struct oVECT v; Q d0,d1,a,u,u1,gcd; - int i; + int i,j; N qn,rn,gn; - Q *q,*r; + Q *t,*q,*r; + t = (Q *)ALLOCA(n*sizeof(Q)); + for ( i = j = 0; i < n; i++ ) + if ( c[i] ) + t[j++] = c[i]; + n = j; + c = t; q = (Q *)ALLOCA(n*sizeof(Q)); r = (Q *)ALLOCA(n*sizeof(Q)); v.id = O_VECT; v.len = n; v.body = (pointer *)c; @@ -2347,11 +2622,13 @@ void nd_mul_c(int mod,ND p,int mul) int c,c1; if ( !p ) return; - for ( m = BDY(p); m; m = NEXT(m) ) { - c1 = CM(m); - DMAR(c1,mul,0,mod,c); - CM(m) = c; - } + if ( mod == -1 ) + for ( m = BDY(p); m; m = NEXT(m) ) + CM(m) = _mulsf(CM(m),mul); + else + for ( m = BDY(p); m; m = NEXT(m) ) { + c1 = CM(m); DMAR(c1,mul,0,mod,c); CM(m) = c; + } } void nd_mul_c_q(ND p,Q mul) @@ -2365,6 +2642,17 @@ void nd_mul_c_q(ND p,Q mul) } } +void nd_mul_c_p(VL vl,ND p,P mul) +{ + NM m; + P c; + + if ( !p ) return; + for ( m = BDY(p); m; m = NEXT(m) ) { + mulp(vl,CP(m),mul,&c); CP(m) = c; + } +} + void nd_free(ND p) { NM t,s; @@ -2384,7 +2672,7 @@ void ndv_free(NDV p) GC_free(BDY(p)); } -void nd_append_red(unsigned int *d,int i) +void nd_append_red(UINT *d,int i) { RHist m,m0; int h; @@ -2397,61 +2685,105 @@ void nd_append_red(unsigned int *d,int i) nd_red[h] = m; } -unsigned int *dp_compute_bound(DP p) +UINT *ndv_compute_bound(NDV p) { - unsigned int *d,*d1,*d2,*t; - MP m; - int i,l; + UINT *d1,*d2,*t; + UINT u; + int i,j,k,l,len,ind; + NMV m; 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]; + d1 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT)); + d2 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT)); + len = LEN(p); + m = BDY(p); ndl_copy(DL(m),d1); NMV_ADV(m); + for ( i = 1; i < len; i++, NMV_ADV(m) ) { + ndl_lcm(DL(m),d1,d2); 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; + l = nd_nvar+31; + t = (UINT *)MALLOC_ATOMIC(l*sizeof(UINT)); + 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; } -unsigned int *nd_compute_bound(ND p) +UINT *nd_compute_bound(ND p) { - unsigned int *d1,*d2,*t; - int i,l; + UINT *d1,*d2,*t; + UINT u; + int i,j,k,l,len,ind; NM m; if ( !p ) return 0; - d1 = (unsigned int *)ALLOCA(nd_wpd*sizeof(unsigned int)); - d2 = (unsigned int *)ALLOCA(nd_wpd*sizeof(unsigned int)); - ndl_copy(HDL(p),d1); - for ( m = NEXT(BDY(p)); m; m = NEXT(m) ) { + d1 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT)); + d2 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT)); + len = LEN(p); + m = BDY(p); ndl_copy(DL(m),d1); m = NEXT(m); + for ( m = NEXT(m); m; m = NEXT(m) ) { ndl_lcm(DL(m),d1,d2); t = d1; d1 = d2; d2 = t; } l = nd_nvar+31; - t = (unsigned int *)MALLOC_ATOMIC(l*sizeof(unsigned int)); - for ( i = 0; i < l; i++ ) t[i] = 0; - for ( i = 0; i < nd_nvar; i++ ) - t[i] = (d1[i/nd_epw+1]>>((nd_epw-(i%nd_epw)-1)*nd_bpe))&nd_mask0; - for ( ; i < l; i++ ) t[i] = 0; + t = (UINT *)MALLOC_ATOMIC(l*sizeof(UINT)); + 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; } -void nd_setup_parameters() { - int i; +int nd_get_exporigin(struct order_spec *ord) +{ + switch ( ord->id ) { + case 0: case 2: + return 1; + case 1: + /* block order */ + /* d[0]:weight d[1]:w0,...,d[nd_exporigin-1]:w(n-1) */ + return ord->ord.block.length+1; + case 3: + error("nd_get_exporigin : composite order is not supported yet."); + } +} - nd_epw = (sizeof(unsigned int)*8)/nd_bpe; - nd_wpd = nd_nvar/nd_epw+(nd_nvar%nd_epw?1:0); +void nd_setup_parameters(int nvar,int max) { + int i,j,n,elen,ord_o,ord_l,l,s,wpd; + struct order_pair *op; + + nd_nvar = nvar; + if ( max ) { + /* XXX */ + if ( do_weyl ) nd_bpe = 32; + else if ( max < 2 ) nd_bpe = 1; + else if ( max < 4 ) nd_bpe = 2; + else if ( max < 8 ) nd_bpe = 3; + else if ( max < 16 ) nd_bpe = 4; + else if ( max < 32 ) nd_bpe = 5; + else if ( max < 64 ) nd_bpe = 6; + else if ( max < 256 ) nd_bpe = 8; + else if ( max < 1024 ) nd_bpe = 10; + else if ( max < 65536 ) nd_bpe = 16; + else nd_bpe = 32; + } + nd_epw = (sizeof(UINT)*8)/nd_bpe; + elen = nd_nvar/nd_epw+(nd_nvar%nd_epw?1:0); + nd_exporigin = nd_get_exporigin(nd_ord); + wpd = nd_exporigin+elen; + if ( wpd != nd_wpd ) { + nd_free_private_storage(); + nd_wpd = wpd; + } if ( nd_bpe < 32 ) { nd_mask0 = (1< realloc nd_ps and pd_psq */ - -ND_pairs nd_reconstruct(int mod,int trace,ND_pairs d) +ND_pairs nd_reconstruct(int trace,ND_pairs d) { int i,obpe,oadv,h; - NM prev_nm_free_list; + static NM prev_nm_free_list; + static ND_pairs prev_ndp_free_list; RHist mr0,mr; RHist r; RHist *old_red; - ND_pairs s0,s,t,prev_ndp_free_list; + ND_pairs s0,s,t; + EPOS oepos; obpe = nd_bpe; oadv = nmv_adv; - if ( obpe < 4 ) nd_bpe = 4; + oepos = nd_epos; + if ( obpe < 2 ) nd_bpe = 2; + else if ( obpe < 3 ) nd_bpe = 3; + else if ( obpe < 4 ) nd_bpe = 4; + else if ( obpe < 5 ) nd_bpe = 5; else if ( obpe < 6 ) nd_bpe = 6; else if ( obpe < 8 ) nd_bpe = 8; + else if ( obpe < 10 ) nd_bpe = 10; else if ( obpe < 16 ) nd_bpe = 16; else if ( obpe < 32 ) nd_bpe = 32; else error("nd_reconstruct : exponent too large"); - nd_setup_parameters(); + nd_setup_parameters(nd_nvar,0); prev_nm_free_list = _nm_free_list; prev_ndp_free_list = _ndp_free_list; _nm_free_list = 0; _ndp_free_list = 0; - if ( mod != 0 ) - for ( i = nd_psn-1; i >= 0; i-- ) ndv_realloc(nd_ps[i],obpe,oadv); - if ( !mod || trace ) - for ( i = nd_psn-1; i >= 0; i-- ) ndv_realloc(nd_psq[i],obpe,oadv); + 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); s->i1 = t->i1; s->i2 = t->i2; SG(s) = SG(t); - ndl_dup(obpe,LCM(t),LCM(s)); + ndl_reconstruct(LCM(t),LCM(s),obpe,oepos); } old_red = (RHist *)ALLOCA(REDTAB_LEN*sizeof(RHist)); @@ -2515,7 +2854,7 @@ ND_pairs nd_reconstruct(int mod,int trace,ND_pairs d) NEWRHist(mr); mr->index = r->index; SG(mr) = SG(r); - ndl_dup(obpe,DL(r),DL(mr)); + ndl_reconstruct(DL(r),DL(mr),obpe,oepos); h = ndl_hash_value(DL(mr)); NEXT(mr) = nd_red[h]; nd_red[h] = mr; @@ -2524,112 +2863,57 @@ ND_pairs nd_reconstruct(int mod,int trace,ND_pairs d) old_red = 0; for ( i = 0; i < nd_psn; i++ ) { NEWRHist(r); SG(r) = SG(nd_psh[i]); - ndl_dup(obpe,DL(nd_psh[i]),DL(r)); + ndl_reconstruct(DL(nd_psh[i]),DL(r),obpe,oepos); nd_psh[i] = r; } if ( s0 ) NEXT(s) = 0; prev_nm_free_list = 0; prev_ndp_free_list = 0; +#if 0 GC_gcollect(); +#endif return s0; } -void nd_reconstruct_direct(int mod,NDV *ps,int len) +void ndl_reconstruct(UINT *d,UINT *r,int obpe,EPOS oepos) { - int i,obpe,oadv,h; - NM prev_nm_free_list; - RHist mr0,mr; - RHist r; - RHist *old_red; - ND_pairs s0,s,t,prev_ndp_free_list; + int n,i,ei,oepw,omask0,j,s,ord_l,l; + struct order_pair *op; - obpe = nd_bpe; - oadv = nmv_adv; - if ( obpe < 4 ) nd_bpe = 4; - else if ( obpe < 6 ) nd_bpe = 6; - else if ( obpe < 8 ) nd_bpe = 8; - else if ( obpe < 16 ) nd_bpe = 16; - else if ( obpe < 32 ) nd_bpe = 32; - else error("nd_reconstruct_direct : exponent too large"); - - nd_setup_parameters(); - prev_nm_free_list = _nm_free_list; - prev_ndp_free_list = _ndp_free_list; - _nm_free_list = 0; _ndp_free_list = 0; - if ( mod != 0 ) - for ( i = len-1; i >= 0; i-- ) ndv_realloc(ps[i],obpe,oadv); - old_red = (RHist *)ALLOCA(REDTAB_LEN*sizeof(RHist)); - for ( i = 0; i < REDTAB_LEN; i++ ) { - old_red[i] = nd_red[i]; - nd_red[i] = 0; - } - for ( i = 0; i < REDTAB_LEN; i++ ) - for ( r = old_red[i]; r; r = NEXT(r) ) { - NEWRHist(mr); - mr->index = r->index; - SG(mr) = SG(r); - ndl_dup(obpe,DL(r),DL(mr)); - h = ndl_hash_value(DL(mr)); - NEXT(mr) = nd_red[h]; - nd_red[h] = mr; - } - for ( i = 0; i < REDTAB_LEN; i++ ) old_red[i] = 0; - old_red = 0; - prev_nm_free_list = 0; - prev_ndp_free_list = 0; - GC_gcollect(); -} - -void ndl_dup(int obpe,unsigned int *d,unsigned int *r) -{ - int n,i,ei,oepw,cepw,cbpe; - n = nd_nvar; - oepw = (sizeof(unsigned int)*8)/obpe; - cepw = nd_epw; - cbpe = nd_bpe; + oepw = (sizeof(UINT)*8)/obpe; + omask0 = (1<>((oepw-((n-1-i)%oepw)-1)*obpe)) - &((1<n; + op = nd_blockmask->order_pair; + for ( i = 1; i < nd_exporigin; i++ ) + r[i] = d[i]; + for ( j = 0, s = 0; j < l; j++ ) { + ord_l = op[j].length; + for ( i = 0; i < ord_l; i++, s++ ) { + ei = GET_EXP_OLD(d,s); + PUT_EXP(r,s,ei); + } } - else + } else { for ( i = 0; i < n; i++ ) { - ei = (d[i/oepw+1]>>((oepw-(i%oepw)-1)*obpe)) - &((1<i1]; p2 = nd_ps[p->i2]; + if ( !mod && nd_demand ) { + p1 = ndv_load(p->i1); p2 = ndv_load(p->i2); } else { - p1 = nd_psq[p->i1]; p2 = nd_psq[p->i2]; + if ( trace ) { + p1 = nd_ps_trace[p->i1]; p2 = nd_ps_trace[p->i2]; + } else { + 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 ( mod ) CM(m) = mod-HCM(p1); + if ( ndl_check_bound2(p->i1,DL(m)) ) + return 0; + t1 = ndv_mul_nm(mod,m,p1); + if ( mod == -1 ) CM(m) = _chsgnsf(HCM(p1)); + else if ( mod ) CM(m) = mod-HCM(p1); else chsgnq(HCQ(p1),&CQ(m)); ndl_sub(lcm,HDL(p2),DL(m)); if ( ndl_check_bound2(p->i2,DL(m)) ) { 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; @@ -2681,9 +2971,13 @@ void ndv_mul_c(int mod,NDV p,int mul) if ( !p ) return; len = LEN(p); - for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) { - c1 = CM(m); DMAR(c1,mul,0,mod,c); CM(m) = c; - } + if ( mod == -1 ) + for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) + CM(m) = _mulsf(CM(m),mul); + else + for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) { + c1 = CM(m); DMAR(c1,mul,0,mod,c); CM(m) = c; + } } void ndv_mul_c_q(NDV p,Q mul) @@ -2699,11 +2993,145 @@ 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; + UINT *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; + UINT *d0,*d1,*d,*dt,*ctab; + Q *ctab_q; + Q q,q1; + UINT 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_weight_mask(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_weight_mask(dt); + } + curlen *= k+1; + continue; + } + min = MIN(k,l); + if ( mod ) { + ctab = (UINT *)ALLOCA((min+1)*sizeof(UINT)); + 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_weight_mask(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_symbolic(NM m0,NDV p) +{ NM mr,mr0; NMV m; - unsigned int *d,*dt,*dm; + UINT *d,*dt,*dm; int c,n,td,i,c1,c2,len; Q q; ND r; @@ -2715,10 +3143,51 @@ ND ndv_mul_nm(int mod,NDV p,NM m0) len = LEN(p); mr0 = 0; td = TD(d); - if ( mod ) { + c = CM(m0); + for ( i = 0; i < len; i++, NMV_ADV(m) ) { + NEXTNM(mr0,mr); + CM(mr) = 1; + ndl_add(DL(m),d,DL(mr)); + } + NEXT(mr) = 0; + MKND(NV(p),mr0,len,r); + SG(r) = SG(p) + TD(d); + return r; + } +} + +ND ndv_mul_nm(int mod,NM m0,NDV p) +{ + NM mr,mr0; + NMV m; + UINT *d,*dt,*dm; + int c,n,td,i,c1,c2,len; + Q q; + ND r; + + if ( !p ) return 0; + else if ( do_weyl ) + if ( mod == -1 ) + error("ndv_mul_nm : not implemented (weyl)"); + else + return weyl_ndv_mul_nm(mod,m0,p); + else { + n = NV(p); m = BDY(p); + d = DL(m0); + len = LEN(p); + mr0 = 0; + td = TD(d); + if ( mod == -1 ) { c = CM(m0); for ( i = 0; i < len; i++, NMV_ADV(m) ) { NEXTNM(mr0,mr); + CM(mr) = _mulsf(CM(m),c); + ndl_add(DL(m),d,DL(mr)); + } + } else if ( mod ) { + c = CM(m0); + for ( i = 0; i < len; i++, NMV_ADV(m) ) { + NEXTNM(mr0,mr); c1 = CM(m); DMAR(c1,c,0,mod,c2); CM(mr) = c2; @@ -2739,42 +3208,106 @@ ND ndv_mul_nm(int mod,NDV p,NM m0) } } -void ndv_realloc(NDV p,int obpe,int oadv) +ND nd_quo(int mod,PGeoBucket bucket,NDV d) { + NM mq0,mq; + NMV tm; + Q q; + int i,nv,sg,c,c1,c2,hindex; + ND p,t,r; + N tnm; + + if ( !p ) return 0; + else { + nv = NV(d); + mq0 = 0; + tm = (NMV)ALLOCA(nmv_adv); + while ( 1 ) { + hindex = mod?head_pbucket(mod,bucket):head_pbucket_q(bucket); + if ( hindex < 0 ) break; + p = bucket->body[hindex]; + NEXTNM(mq0,mq); + ndl_sub(HDL(p),HDL(d),DL(mq)); + ndl_copy(DL(mq),DL(tm)); + if ( mod ) { + c1 = invm(HCM(d),mod); c2 = HCM(p); + DMAR(c1,c2,0,mod,c); CM(mq) = c; + CM(tm) = mod-c; + } else { + divsn(NM(HCQ(p)),NM(HCQ(d)),&tnm); + NTOQ(tnm,SGN(HCQ(p))*SGN(HCQ(d)),CQ(mq)); + chsgnq(CQ(mq),&CQ(tm)); + } + t = ndv_mul_nmv_trunc(mod,tm,d,HDL(d)); + bucket->body[hindex] = nd_remove_head(p); + t = nd_remove_head(t); + add_pbucket(mod,bucket,t); + } + if ( !mq0 ) + r = 0; + else { + NEXT(mq) = 0; + for ( i = 0, mq = mq0; mq; mq = NEXT(mq), i++ ); + MKND(nv,mq0,i,r); + /* XXX */ + SG(r) = HTD(r); + } + return r; + } +} + +void ndv_realloc(NDV p,int obpe,int oadv,EPOS oepos) +{ NMV m,mr,mr0,t; 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 ) return; + m = BDY(p); len = LEN(p); + mr0 = nmv_adv>oadv?(NMV)REALLOC(BDY(p),len*nmv_adv):BDY(p); + m = (NMV)((char *)mr0+(len-1)*oadv); + mr = (NMV)((char *)mr0+(len-1)*nmv_adv); + t = (NMV)ALLOCA(nmv_adv); + for ( i = 0; i < len; i++, NMV_OPREV(m), NMV_PREV(mr) ) { + CQ(t) = CQ(m); + for ( k = 0; k < nd_wpd; k++ ) DL(t)[k] = 0; + ndl_reconstruct(DL(m),DL(t),obpe,oepos); + CQ(mr) = CQ(t); + ndl_copy(DL(t),DL(mr)); + } + BDY(p) = mr0; +} - if ( p ) { - m = BDY(p); len = LEN(p); - mr0 = nmv_adv>oadv?(NMV)REALLOC(BDY(p),len*nmv_adv):BDY(p); - m = (NMV)((char *)mr0+(len-1)*oadv); - mr = (NMV)((char *)mr0+(len-1)*nmv_adv); - t = (NMV)ALLOCA(nmv_adv); - for ( i = 0; i < len; i++, NMV_OPREV(m), NMV_PREV(mr) ) { - CQ(t) = CQ(m); - for ( k = 0; k < nd_wpd; k++ ) DL(t)[k] = 0; - ndl_dup(obpe,DL(m),DL(t)); - CQ(mr) = CQ(t); - ndl_copy(DL(t),DL(mr)); - } - BDY(p) = mr0; +NDV ndv_dup_realloc(NDV p,int obpe,int oadv,EPOS oepos) +{ + NMV m,mr,mr0; + int len,i; + NDV r; + + if ( !p ) return 0; + m = BDY(p); len = LEN(p); + mr0 = mr = (NMV)MALLOC(len*nmv_adv); + for ( i = 0; i < len; i++, NMV_OADV(m), NMV_ADV(mr) ) { + ndl_zero(DL(mr)); + ndl_reconstruct(DL(m),DL(mr),obpe,oepos); + CQ(mr) = CQ(m); } + MKNDV(NV(p),mr0,len,r); + SG(r) = SG(p); + return r; } -NDV ndtondv(int mod,ND p) +/* duplicate p */ + +NDV ndv_dup(int mod,NDV p) { NDV d; - NMV m,m0; - NM t; + NMV t,m,m0; int i,len; if ( !p ) return 0; len = LEN(p); m0 = m = (NMV)(mod?MALLOC_ATOMIC(len*nmv_adv):MALLOC(len*nmv_adv)); - for ( t = BDY(p), i = 0; t; t = NEXT(t), i++, NMV_ADV(m) ) { + for ( t = BDY(p), i = 0; i < len; i++, NMV_ADV(t), NMV_ADV(m) ) { ndl_copy(DL(t),DL(m)); CQ(m) = CQ(t); } @@ -2783,72 +3316,197 @@ NDV ndtondv(int mod,ND p) return d; } -ND ndvtond(int mod,NDV p) +ND nd_dup(ND p) { ND d; - NM m,m0; - NMV t; - int i,len; + NM t,m,m0; if ( !p ) return 0; - m0 = 0; - len = p->len; - for ( t = BDY(p), i = 0; i < len; NMV_ADV(t), i++ ) { + for ( m0 = 0, t = BDY(p); t; t = NEXT(t) ) { NEXTNM(m0,m); ndl_copy(DL(t),DL(m)); CQ(m) = CQ(t); } - NEXT(m) = 0; - MKND(NV(p),m0,len,d); + if ( m0 ) NEXT(m) = 0; + MKND(NV(p),m0,LEN(p),d); SG(d) = SG(p); return d; } -NDV dptondv(int mod,DP p) +/* XXX if p->len == 0 then it represents 0 */ + +void ndv_mod(int mod,NDV p) { + NMV t,d; + int r; + int i,len,dlen; + Obj gfs; + + if ( !p ) return; + len = LEN(p); + dlen = 0; + if ( mod == -1 ) + for ( t = d = BDY(p), i = 0; i < len; i++, NMV_ADV(t) ) { + simp_ff((Obj)CP(t),&gfs); + r = FTOIF(CONT((GFS)gfs)); + CM(d) = r; + ndl_copy(DL(t),DL(d)); + NMV_ADV(d); + dlen++; + } + else + for ( t = d = BDY(p), i = 0; i < len; i++, NMV_ADV(t) ) { + r = rem(NM(CQ(t)),mod); + if ( r ) { + if ( SGN(CQ(t)) < 0 ) + r = mod-r; + CM(d) = r; + ndl_copy(DL(t),DL(d)); + NMV_ADV(d); + dlen++; + } + } + LEN(p) = dlen; +} + +NDV ptondv(VL vl,VL dvl,P p) +{ + ND nd; + + nd = ptond(vl,dvl,p); + return ndtondv(0,nd); +} + +ND ptond(VL vl,VL dvl,P p) +{ + int n,i,j,k,e; + VL tvl; + V v; + DCP dc; + DCP *w; + ND r,s,t,u; + P x; + int c; + UINT *d; + NM m,m0; + + if ( !p ) + return 0; + else if ( NUM(p) ) { + NEWNM(m); + ndl_zero(DL(m)); + CQ(m) = (Q)p; + NEXT(m) = 0; + MKND(nd_nvar,m,1,r); + SG(r) = 0; + return r; + } else { + for ( dc = DC(p), k = 0; dc; dc = NEXT(dc), k++ ); + w = (DCP *)ALLOCA(k*sizeof(DCP)); + for ( dc = DC(p), j = 0; j < k; dc = NEXT(dc), j++ ) w[j] = dc; + for ( i = 0, tvl = dvl, v = VR(p); + tvl && tvl->v != v; tvl = NEXT(tvl), i++ ); + if ( !tvl ) { + for ( j = k-1, s = 0, MKV(v,x); j >= 0; j-- ) { + t = ptond(vl,dvl,COEF(w[j])); + pwrp(vl,x,DEG(w[j]),&p); + nd_mul_c_p(CO,t,p); s = nd_add(0,s,t); + } + return s; + } else { + NEWNM(m0); d = DL(m0); + for ( j = k-1, s = 0; j >= 0; j-- ) { + ndl_zero(d); e = QTOS(DEG(w[j])); PUT_EXP(d,i,e); + TD(d) = MUL_WEIGHT(e,i); + if ( nd_blockmask) ndl_weight_mask(d); + t = ptond(vl,dvl,COEF(w[j])); + for ( m = BDY(t); m; m = NEXT(m) ) + ndl_addto(DL(m),d); + SG(t) += TD(d); + s = nd_add(0,s,t); + } + FREENM(m0); + return s; + } + } +} + +P ndvtop(int mod,VL vl,VL dvl,NDV p) +{ + VL tvl; + int len,n,j,i,e; + NMV m; + Q q; + P c; + UINT *d; + P s,r,u,t,w; + GFS gfs; + + if ( !p ) return 0; + else { + len = LEN(p); + n = NV(p); + m = (NMV)(((char *)BDY(p))+nmv_adv*(len-1)); + for ( j = len-1, s = 0; j >= 0; j--, NMV_PREV(m) ) { + if ( mod == -1 ) { + e = IFTOF(CM(m)); MKGFS(e,gfs); c = (P)gfs; + } else if ( mod ) { + STOQ(CM(m),q); c = (P)q; + } else + c = CP(m); + d = DL(m); + for ( i = 0, t = c, tvl = dvl; i < n; tvl = NEXT(tvl), i++ ) { + MKV(tvl->v,r); e = GET_EXP(d,i); STOQ(e,q); + pwrp(vl,r,q,&u); mulp(vl,t,u,&w); t = w; + } + addp(vl,s,t,&u); s = u; + } + return s; + } +} + +NDV ndtondv(int mod,ND p) +{ NDV d; - NMV m0,m; - MP t; - DP q; - int l,i,n; + NMV m,m0; + NM t; + int i,len; if ( !p ) return 0; - for ( t = BDY(p), l = 0; t; t = NEXT(t), l++ ); - if ( mod ) { - _dp_mod(p,mod,0,&q); p = q; - m0 = m = (NMV)MALLOC_ATOMIC(l*nmv_adv); - } else - m0 = m = (NMV)MALLOC(l*nmv_adv); - n = NV(p); - for ( t = BDY(p), i = 0; i < l; i++, t = NEXT(t), NMV_ADV(m) ) { - if ( mod ) CM(m) = ITOS(C(t)); - else CQ(m) = (Q)C(t); - dltondl(n,DL(t),DL(m)); + len = LEN(p); + if ( mod ) + m0 = m = (NMV)GC_malloc_atomic_ignore_off_page(len*nmv_adv); + else + m0 = m = MALLOC(len*nmv_adv); +#if 0 + ndv_alloc += nmv_adv*len; +#endif + for ( t = BDY(p), i = 0; t; t = NEXT(t), i++, NMV_ADV(m) ) { + ndl_copy(DL(t),DL(m)); + CQ(m) = CQ(t); } - MKNDV(n,m0,l,d); + MKNDV(NV(p),m0,len,d); SG(d) = SG(p); return d; } -DP ndvtodp(int mod,NDV p) +ND ndvtond(int mod,NDV p) { - DP d; - MP m0,m; + ND d; + NM m,m0; NMV t; - int len,i,n; + int i,len; if ( !p ) return 0; m0 = 0; - len = LEN(p); - n = NV(p); - for ( t = BDY(p), i = 0; i < len; i++, NMV_ADV(t) ) { - NEXTMP(m0,m); - if ( mod ) C(m) = STOI(CM(t)); - else C(m) = (P)CQ(t); - DL(m) = ndltodl(n,DL(t)); + len = p->len; + for ( t = BDY(p), i = 0; i < len; NMV_ADV(t), i++ ) { + NEXTNM(m0,m); + ndl_copy(DL(t),DL(m)); + CQ(m) = CQ(t); } NEXT(m) = 0; - MKDP(NV(p),m0,d); + MKND(NV(p),m0,len,d); SG(d) = SG(p); return d; } @@ -2862,7 +3520,8 @@ void ndv_print(NDV p) else { len = LEN(p); for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) { - printf("+%d*",CM(m)); + if ( CM(m) & 0x80000000 ) printf("+@_%d*",IFTOF(CM(m))); + else printf("+%d*",CM(m)); ndl_print(DL(m)); } printf("\n"); @@ -2879,7 +3538,7 @@ void ndv_print_q(NDV p) len = LEN(p); for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) { printf("+"); - printexpr(CO,CQ(m)); + printexpr(CO,(Obj)CQ(m)); printf("*"); ndl_print(DL(m)); } @@ -2887,7 +3546,7 @@ void ndv_print_q(NDV p) } } -NODE nd_reducebase(NODE x) +NODE ndv_reducebase(NODE x) { int len,i,j; NDV *w; @@ -2910,27 +3569,1466 @@ NODE nd_reducebase(NODE x) return x; } +/* XXX incomplete */ + void nd_init_ord(struct order_spec *ord) { - if ( ord->id ) - error("nd_gr : unsupported order"); - nd_dcomp = 0; - switch ( ord->ord.simple ) { + switch ( ord->id ) { case 0: - nd_dcomp = 1; - nd_isrlex = 1; + switch ( ord->ord.simple ) { + case 0: + nd_dcomp = 1; + nd_isrlex = 1; + break; + case 1: + nd_dcomp = 1; + nd_isrlex = 0; + break; + case 2: + nd_dcomp = 0; + nd_isrlex = 0; + ndl_compare_function = ndl_lex_compare; + break; + case 11: + /* XXX */ + nd_dcomp = 0; + nd_isrlex = 1; + ndl_compare_function = ndl_ww_lex_compare; + break; + default: + error("nd_gr : unsupported order"); + } break; case 1: - nd_dcomp = 1; + /* block order */ + /* XXX */ + nd_dcomp = -1; nd_isrlex = 0; + ndl_compare_function = ndl_block_compare; break; - case 9: - /* td1->td->rlex */ - nd_dcomp = -dp_nelim; - nd_isrlex = 1; + case 2: + /* matrix order */ + /* XXX */ + nd_dcomp = -1; + nd_isrlex = 0; + nd_matrix_len = ord->ord.matrix.row; + nd_matrix = ord->ord.matrix.matrix; + ndl_compare_function = ndl_matrix_compare; break; - default: - error("nd_gr : unsupported order"); + case 3: + /* composite order */ + nd_dcomp = -1; + nd_isrlex = 0; + nd_worb_len = ord->ord.composite.length; + nd_worb = ord->ord.composite.w_or_b; + ndl_compare_function = ndl_composite_compare; + break; } + nd_ord = ord; } +BlockMask nd_create_blockmask(struct order_spec *ord) +{ + int n,i,j,s,l; + UINT *t; + BlockMask bm; + + /* we only create mask table for block order */ + if ( ord->id != 1 ) + return 0; + n = ord->ord.block.length; + bm = (BlockMask)MALLOC(sizeof(struct oBlockMask)); + bm->n = n; + bm->order_pair = ord->ord.block.order_pair; + bm->mask = (UINT **)MALLOC(n*sizeof(UINT *)); + for ( i = 0, s = 0; i < n; i++ ) { + bm->mask[i] = t = (UINT *)MALLOC_ATOMIC(nd_wpd*sizeof(UINT)); + for ( j = 0; j < nd_wpd; j++ ) t[j] = 0; + l = bm->order_pair[i].length; + 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: + /* matrix order */ + case 3: + /* composite order */ + 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; + } + return epos; +} + +/* external interface */ + +void nd_nf_p(P f,LIST g,LIST v,int m,struct order_spec *ord,P *rp) +{ + NODE t,in0,in; + ND nd,nf; + NDV ndv; + VL vv,tv; + int stat,nvar,max,e; + union oNDC dn; + + pltovl(v,&vv); + for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++ ); + + /* get the degree bound */ + for ( t = BDY(g), max = 0; t; t = NEXT(t) ) + for ( tv = vv; tv; tv = NEXT(tv) ) { + e = getdeg(tv->v,(P)BDY(t)); + max = MAX(e,max); + } + for ( tv = vv; tv; tv = NEXT(tv) ) { + e = getdeg(tv->v,f); + max = MAX(e,max); + } + + nd_init_ord(ord); + nd_setup_parameters(nvar,max); + + /* conversion to ndv */ + for ( in0 = 0, t = BDY(g); t; t = NEXT(t) ) { + NEXTNODE(in0,in); + BDY(in) = (pointer)ptondv(CO,vv,(P)BDY(t)); + if ( m ) ndv_mod(m,(NDV)BDY(in)); + } + NEXTNODE(in0,in); + BDY(in) = (pointer)ptondv(CO,vv,f); + if ( m ) ndv_mod(m,(NDV)BDY(in)); + NEXT(in) = 0; + + ndv_setup(m,0,in0); + nd_psn--; + nd_scale=2; + while ( 1 ) { + nd = (pointer)ndvtond(m,nd_ps[nd_psn]); + stat = nd_nf(m,nd,nd_ps,1,0,&nf); + if ( !stat ) { + nd_psn++; + nd_reconstruct(0,0); + nd_psn--; + } else + break; + } + *rp = ndvtop(m,CO,vv,ndtondv(m,nf)); +} + +int nd_to_vect(int mod,UINT *s0,int n,ND d,UINT *r) +{ + NM m; + UINT *t,*s; + int i; + + for ( i = 0; i < n; i++ ) r[i] = 0; + 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] = CM(m); + } + for ( i = 0; !r[i]; i++ ); + return i; +} + +int nd_to_vect_q(UINT *s0,int n,ND d,Q *r) +{ + NM m; + UINT *t,*s; + int i; + + for ( i = 0; i < n; i++ ) r[i] = 0; + 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] = CQ(m); + } + for ( i = 0; !r[i]; i++ ); + return i; +} + +IndArray nm_ind_pair_to_vect_compress(int mod,UINT *s0,int n,NM_ind_pair pair) +{ + NM m; + NMV mr; + UINT *d,*t,*s; + NDV p; + unsigned char *ivc; + unsigned short *ivs; + UINT *v,*ivi,*s0v; + int i,j,len,prev,diff,cdiff; + IndArray r; + + m = pair->mul; + d = DL(m); + p = nd_ps[pair->index]; + len = LEN(p); + t = (UINT *)ALLOCA(nd_wpd*sizeof(UINT)); + v = (unsigned int *)ALLOCA(len*sizeof(unsigned int)); + for ( i = j = 0, s = s0, mr = BDY(p); j < len; j++, NMV_ADV(mr) ) { + ndl_add(d,DL(mr),t); + for ( ; !ndl_equal(t,s); s += nd_wpd, i++ ); + v[j] = i; + } + r = (IndArray)MALLOC(sizeof(struct oIndArray)); + r->head = v[0]; + diff = 0; + for ( i = 1; i < len; i++ ) { + cdiff = v[i]-v[i-1]; diff = MAX(cdiff,diff); + } + if ( diff < 256 ) { + r->width = 1; + ivc = (unsigned char *)MALLOC_ATOMIC(len*sizeof(unsigned char)); + r->index.c = ivc; + for ( i = 1, ivc[0] = 0; i < len; i++ ) ivc[i] = v[i]-v[i-1]; + } else if ( diff < 65536 ) { + r->width = 2; + ivs = (unsigned short *)MALLOC_ATOMIC(len*sizeof(unsigned short)); + r->index.s = ivs; + for ( i = 1, ivs[0] = 0; i < len; i++ ) ivs[i] = v[i]-v[i-1]; + } else { + r->width = 4; + ivi = (unsigned int *)MALLOC_ATOMIC(len*sizeof(unsigned int)); + r->index.i = ivi; + for ( i = 1, ivi[0] = 0; i < len; i++ ) ivi[i] = v[i]-v[i-1]; + } + return r; +} + + +int ndv_reduce_vect_q(Q *svect,int col,IndArray *imat,NM_ind_pair *rp0,int nred) +{ + int i,j,k,len,pos,prev; + Q cs,mcs,c1,c2,cr,gcd,t; + IndArray ivect; + unsigned char *ivc; + unsigned short *ivs; + unsigned int *ivi; + NDV redv; + NMV mr; + NODE rp; + int maxrs; + + maxrs = 0; + for ( i = 0; i < nred; i++ ) { + ivect = imat[i]; + k = ivect->head; + if ( svect[k] ) { + maxrs = MAX(maxrs,rp0[i]->sugar); + redv = nd_ps[rp0[i]->index]; + len = LEN(redv); mr = BDY(redv); + igcd_cofactor(svect[k],CQ(mr),&gcd,&cs,&cr); + chsgnq(cs,&mcs); + if ( !UNIQ(cr) ) { + for ( j = 0; j < col; j++ ) { + mulq(svect[j],cr,&c1); svect[j] = c1; + } + } + svect[k] = 0; prev = k; + switch ( ivect->width ) { + case 1: + ivc = ivect->index.c; + for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) { + pos = prev+ivc[j]; prev = pos; + mulq(CQ(mr),mcs,&c2); addq(svect[pos],c2,&t); svect[pos] = t; + } + break; + case 2: + ivs = ivect->index.s; + for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) { + pos = prev+ivs[j]; prev = pos; + mulq(CQ(mr),mcs,&c2); addq(svect[pos],c2,&t); svect[pos] = t; + } + break; + case 4: + ivi = ivect->index.i; + for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) { + pos = prev+ivi[j]; prev = pos; + mulq(CQ(mr),mcs,&c2); addq(svect[pos],c2,&t); svect[pos] = t; + } + break; + } + } + } + return maxrs; +} + +int ndv_reduce_vect(int m,UINT *svect,int col,IndArray *imat,NM_ind_pair *rp0,int nred) +{ + int i,j,k,len,pos,prev; + UINT c,c1,c2,c3,up,lo,dmy; + IndArray ivect; + unsigned char *ivc; + unsigned short *ivs; + unsigned int *ivi; + NDV redv; + NMV mr; + NODE rp; + int maxrs; + + maxrs = 0; + for ( i = 0; i < nred; i++ ) { + ivect = imat[i]; + k = ivect->head; svect[k] %= m; + if ( c = svect[k] ) { + maxrs = MAX(maxrs,rp0[i]->sugar); + c = m-c; redv = nd_ps[rp0[i]->index]; + len = LEN(redv); mr = BDY(redv); + svect[k] = 0; prev = k; + switch ( ivect->width ) { + case 1: + ivc = ivect->index.c; + for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) { + pos = prev+ivc[j]; c1 = CM(mr); c2 = svect[pos]; + prev = pos; + DMA(c1,c,c2,up,lo); + if ( up ) { DSAB(m,up,lo,dmy,c3); svect[pos] = c3; + } else svect[pos] = lo; + } + 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); c2 = svect[pos]; + prev = pos; + DMA(c1,c,c2,up,lo); + if ( up ) { DSAB(m,up,lo,dmy,c3); svect[pos] = c3; + } else svect[pos] = lo; + } + 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); c2 = svect[pos]; + prev = pos; + DMA(c1,c,c2,up,lo); + if ( up ) { DSAB(m,up,lo,dmy,c3); svect[pos] = c3; + } else svect[pos] = lo; + } + break; + } + } + } + for ( i = 0; i < col; i++ ) + if ( svect[i] >= (UINT)m ) svect[i] %= m; + return maxrs; +} + +int ndv_reduce_vect_sf(int m,UINT *svect,int col,IndArray *imat,NM_ind_pair *rp0,int nred) +{ + int i,j,k,len,pos,prev; + UINT c,c1,c2,c3,up,lo,dmy; + IndArray ivect; + unsigned char *ivc; + unsigned short *ivs; + unsigned int *ivi; + NDV redv; + NMV mr; + NODE rp; + int maxrs; + + maxrs = 0; + for ( i = 0; i < nred; i++ ) { + ivect = imat[i]; + k = ivect->head; svect[k] %= m; + if ( c = svect[k] ) { + maxrs = MAX(maxrs,rp0[i]->sugar); + c = _chsgnsf(c); redv = nd_ps[rp0[i]->index]; + len = LEN(redv); mr = BDY(redv); + svect[k] = 0; prev = k; + switch ( ivect->width ) { + case 1: + ivc = ivect->index.c; + for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) { + pos = prev+ivc[j]; prev = pos; + svect[pos] = _addsf(_mulsf(CM(mr),c),svect[pos]); + } + break; + case 2: + ivs = ivect->index.s; + for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) { + pos = prev+ivs[j]; prev = pos; + svect[pos] = _addsf(_mulsf(CM(mr),c),svect[pos]); + } + break; + case 4: + ivi = ivect->index.i; + for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) { + pos = prev+ivi[j]; prev = pos; + svect[pos] = _addsf(_mulsf(CM(mr),c),svect[pos]); + } + break; + } + } + } + return maxrs; +} + +NDV vect_to_ndv(UINT *vect,int spcol,int col,int *rhead,UINT *s0vect) +{ + int j,k,len; + UINT *p; + UINT c; + NDV r; + NMV mr0,mr; + + for ( j = 0, len = 0; j < spcol; j++ ) if ( vect[j] ) len++; + if ( !len ) return 0; + else { + mr0 = (NMV)GC_malloc_atomic_ignore_off_page(nmv_adv*len); +#if 0 + ndv_alloc += nmv_adv*len; +#endif + mr = mr0; + p = s0vect; + for ( j = k = 0; j < col; j++, p += nd_wpd ) + if ( !rhead[j] ) { + if ( c = vect[k++] ) { + ndl_copy(p,DL(mr)); CM(mr) = c; NMV_ADV(mr); + } + } + MKNDV(nd_nvar,mr0,len,r); + return r; + } +} + +NDV vect_to_ndv_q(Q *vect,int spcol,int col,int *rhead,UINT *s0vect) +{ + int j,k,len; + UINT *p; + Q c; + NDV r; + NMV mr0,mr; + + for ( j = 0, len = 0; j < spcol; j++ ) if ( vect[j] ) len++; + if ( !len ) return 0; + else { + mr0 = (NMV)GC_malloc(nmv_adv*len); +#if 0 + ndv_alloc += nmv_adv*len; +#endif + mr = mr0; + p = s0vect; + for ( j = k = 0; j < col; j++, p += nd_wpd ) + if ( !rhead[j] ) { + if ( c = vect[k++] ) { + if ( DN(c) ) + error("afo"); + ndl_copy(p,DL(mr)); CQ(mr) = c; NMV_ADV(mr); + } + } + MKNDV(nd_nvar,mr0,len,r); + return r; + } +} + +int nd_sp_f4(int m,ND_pairs l,PGeoBucket bucket) +{ + ND_pairs t; + NODE sp0,sp; + int stat; + ND spol; + + for ( t = l; t; t = NEXT(t) ) { + stat = nd_sp(m,0,t,&spol); + if ( !stat ) return 0; + if ( spol ) { + add_pbucket_symbolic(bucket,spol); + } + } + return 1; +} + +int nd_symbolic_preproc(PGeoBucket bucket,UINT **s0vect,NODE *r) +{ + NODE rp0,rp; + NM mul,head,s0,s; + int index,col,i,sugar; + RHist h; + UINT *s0v,*p; + NM_ind_pair pair; + ND red; + + s0 = 0; rp0 = 0; col = 0; + while ( 1 ) { + head = remove_head_pbucket_symbolic(bucket); + if ( !head ) break; + if ( !s0 ) s0 = head; + else NEXT(s) = head; + s = head; + index = ndl_find_reducer(DL(head)); + if ( index >= 0 ) { + h = nd_psh[index]; + NEWNM(mul); + ndl_sub(DL(head),DL(h),DL(mul)); + if ( ndl_check_bound2(index,DL(mul)) ) return 0; + sugar = TD(DL(mul))+SG(nd_ps[index]); + MKNM_ind_pair(pair,mul,index,sugar); + red = ndv_mul_nm_symbolic(mul,nd_ps[index]); + add_pbucket_symbolic(bucket,nd_remove_head(red)); + NEXTNODE(rp0,rp); BDY(rp) = (pointer)pair; + } + col++; + } + if ( rp0 ) NEXT(rp) = 0; + NEXT(s) = 0; + s0v = (UINT *)MALLOC_ATOMIC(col*nd_wpd*sizeof(UINT)); + for ( i = 0, p = s0v, s = s0; i < col; + i++, p += nd_wpd, s = NEXT(s) ) ndl_copy(DL(s),p); + *s0vect = s0v; + *r = rp0; + return col; +} + +NODE nd_f4(int m) +{ + int i,nh,stat,index; + NODE r,g; + ND_pairs d,l,t; + ND spol,red; + NDV nf,redv; + NM s0,s; + NODE rp0,srp0,nflist; + int nsp,nred,col,rank,len,k,j,a; + UINT c; + UINT **spmat; + UINT *s0vect,*svect,*p,*v; + int *colstat; + IndArray *imat; + int *rhead; + int spcol,sprow; + int sugar; + PGeoBucket bucket; + struct oEGT eg0,eg1,eg_f4; + +#if 0 + ndv_alloc = 0; +#endif + g = 0; d = 0; + for ( i = 0; i < nd_psn; i++ ) { + d = update_pairs(d,g,i); + g = update_base(g,i); + } + while ( d ) { + get_eg(&eg0); + l = nd_minsugarp(d,&d); + sugar = SG(l); + bucket = create_pbucket(); + stat = nd_sp_f4(m,l,bucket); + if ( !stat ) { + for ( t = l; NEXT(t); t = NEXT(t) ); + NEXT(t) = d; d = l; + d = nd_reconstruct(0,d); + continue; + } + if ( bucket->m < 0 ) continue; + col = nd_symbolic_preproc(bucket,&s0vect,&rp0); + if ( !col ) { + for ( t = l; NEXT(t); t = NEXT(t) ); + NEXT(t) = d; d = l; + d = nd_reconstruct(0,d); + continue; + } + get_eg(&eg1); init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg1); + if ( DP_Print ) + fprintf(asir_out,"sugar=%d,symb=%fsec,", + sugar,eg_f4.exectime+eg_f4.gctime); + if ( 1 ) + nflist = nd_f4_red(m,l,s0vect,col,rp0); + else + nflist = nd_f4_red_dist(m,l,s0vect,col,rp0); + /* adding new bases */ + for ( r = nflist; r; r = NEXT(r) ) { + nf = (NDV)BDY(r); + ndv_removecont(m,nf); + nh = ndv_newps(m,nf,0); + d = update_pairs(d,g,nh); + g = update_base(g,nh); + } + } + for ( r = g; r; r = NEXT(r) ) BDY(r) = (pointer)nd_ps[(int)BDY(r)]; +#if 0 + fprintf(asir_out,"ndv_alloc=%d\n",ndv_alloc); +#endif + return g; +} + +NODE nd_f4_red(int m,ND_pairs sp0,UINT *s0vect,int col,NODE rp0) +{ + IndArray *imat; + int nsp,nred,i; + int *rhead; + NODE r0,rp; + ND_pairs sp; + NM_ind_pair *rvect; + + for ( sp = sp0, nsp = 0; sp; sp = NEXT(sp), nsp++ ); + nred = length(rp0); + imat = (IndArray *)ALLOCA(nred*sizeof(IndArray)); + rhead = (int *)ALLOCA(col*sizeof(int)); + for ( i = 0; i < col; i++ ) rhead[i] = 0; + + /* construction of index arrays */ + rvect = (NM_ind_pair *)ALLOCA(nred*sizeof(NM_ind_pair)); + for ( rp = rp0, i = 0; rp; i++, rp = NEXT(rp) ) { + rvect[i] = (NM_ind_pair)BDY(rp); + imat[i] = nm_ind_pair_to_vect_compress(m,s0vect,col,rvect[i]); + rhead[imat[i]->head] = 1; + } + if ( m ) + r0 = nd_f4_red_main(m,sp0,nsp,s0vect,col,rvect,rhead,imat,nred); + else + r0 = nd_f4_red_q_main(sp0,nsp,s0vect,col,rvect,rhead,imat,nred); + return r0; +} + +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) +{ + int spcol,sprow,a; + int i,j,k,l,rank; + NODE r0,r; + ND_pairs sp; + ND spol; + int **spmat; + UINT *svect,*v; + int *colstat; + struct oEGT eg0,eg1,eg2,eg_f4,eg_f4_1,eg_f4_2; + int maxrs; + int *spsugar; + + spcol = col-nred; + get_eg(&eg0); + /* elimination (1st step) */ + spmat = (int **)ALLOCA(nsp*sizeof(UINT *)); + svect = (UINT *)ALLOCA(col*sizeof(UINT)); + spsugar = (int *)ALLOCA(nsp*sizeof(UINT)); + for ( a = sprow = 0, sp = sp0; a < nsp; a++, sp = NEXT(sp) ) { + nd_sp(m,0,sp,&spol); + if ( !spol ) continue; + nd_to_vect(m,s0vect,col,spol,svect); + if ( m == -1 ) + maxrs = ndv_reduce_vect_sf(m,svect,col,imat,rvect,nred); + else + maxrs = ndv_reduce_vect(m,svect,col,imat,rvect,nred); + for ( i = 0; i < col; i++ ) if ( svect[i] ) break; + if ( i < col ) { + spmat[sprow] = v = (UINT *)MALLOC_ATOMIC(spcol*sizeof(UINT)); + for ( j = k = 0; j < col; j++ ) + if ( !rhead[j] ) v[k++] = svect[j]; + spsugar[sprow] = MAX(maxrs,SG(spol)); + sprow++; + } + nd_free(spol); + } + 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); + fflush(asir_out); + } + /* free index arrays */ + for ( i = 0; i < nred; i++ ) GC_free(imat[i]->index.c); + + /* elimination (2nd step) */ + colstat = (int *)ALLOCA(spcol*sizeof(int)); + if ( m == -1 ) + rank = nd_gauss_elim_sf(spmat,spsugar,sprow,spcol,m,colstat); + else + rank = nd_gauss_elim_mod(spmat,spsugar,sprow,spcol,m,colstat); + r0 = 0; + for ( i = 0; i < rank; i++ ) { + NEXTNODE(r0,r); BDY(r) = + (pointer)vect_to_ndv(spmat[i],spcol,col,rhead,s0vect); + SG((NDV)BDY(r)) = spsugar[i]; + GC_free(spmat[i]); + } + if ( r0 ) NEXT(r) = 0; + for ( ; i < sprow; i++ ) GC_free(spmat[i]); + 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 ", + nsp,nred,sprow,spcol,rank); + fprintf(asir_out,"%fsec\n",eg_f4.exectime+eg_f4.gctime); + } + return r0; +} + +NODE nd_f4_red_q_main(ND_pairs sp0,int nsp,UINT *s0vect,int col, + NM_ind_pair *rvect,int *rhead,IndArray *imat,int nred) +{ + int spcol,sprow,a; + int i,j,k,l,rank; + NODE r0,r; + ND_pairs sp; + ND spol; + Q **spmat; + Q *svect,*v; + int *colstat; + struct oEGT eg0,eg1,eg2,eg_f4,eg_f4_1,eg_f4_2; + int maxrs; + int *spsugar; + + spcol = col-nred; + get_eg(&eg0); + /* elimination (1st step) */ + spmat = (Q **)ALLOCA(nsp*sizeof(Q *)); + svect = (Q *)ALLOCA(col*sizeof(Q)); + spsugar = (int *)ALLOCA(nsp*sizeof(Q)); + for ( a = sprow = 0, sp = sp0; a < nsp; a++, sp = NEXT(sp) ) { + nd_sp(0,0,sp,&spol); + if ( !spol ) continue; + nd_to_vect_q(s0vect,col,spol,svect); + maxrs = ndv_reduce_vect_q(svect,col,imat,rvect,nred); + for ( i = 0; i < col; i++ ) if ( svect[i] ) break; + if ( i < col ) { + spmat[sprow] = v = (Q *)MALLOC(spcol*sizeof(Q)); + for ( j = k = 0; j < col; j++ ) + if ( !rhead[j] ) v[k++] = svect[j]; + spsugar[sprow] = MAX(maxrs,SG(spol)); + sprow++; + } +/* nd_free(spol); */ + } + 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); + fflush(asir_out); + } + /* free index arrays */ +/* for ( i = 0; i < nred; i++ ) GC_free(imat[i]->index.c); */ + + /* elimination (2nd step) */ + colstat = (int *)ALLOCA(spcol*sizeof(int)); + rank = nd_gauss_elim_q(spmat,spsugar,sprow,spcol,colstat); + r0 = 0; + for ( i = 0; i < rank; i++ ) { + NEXTNODE(r0,r); BDY(r) = + (pointer)vect_to_ndv_q(spmat[i],spcol,col,rhead,s0vect); + SG((NDV)BDY(r)) = spsugar[i]; +/* GC_free(spmat[i]); */ + } + if ( r0 ) NEXT(r) = 0; + +/* for ( ; i < sprow; i++ ) GC_free(spmat[i]); */ + 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 ", + nsp,nred,sprow,spcol,rank); + fprintf(asir_out,"%fsec\n",eg_f4.exectime+eg_f4.gctime); + } + return r0; +} + +FILE *nd_write,*nd_read; + +void nd_send_int(int a) { + write_int(nd_write,&a); +} + +void nd_send_intarray(int *p,int len) { + write_intarray(nd_write,p,len); +} + +int nd_recv_int() { + int a; + + read_int(nd_read,&a); + return a; +} + +void nd_recv_intarray(int *p,int len) { + read_intarray(nd_read,p,len); +} + +void nd_send_ndv(NDV p) { + int len,i; + NMV m; + + if ( !p ) nd_send_int(0); + else { + len = LEN(p); + nd_send_int(len); + m = BDY(p); + for ( i = 0; i < len; i++, NMV_ADV(m) ) { + nd_send_int(CM(m)); + nd_send_intarray(DL(m),nd_wpd); + } + } +} + +void nd_send_nd(ND p) { + int len,i; + NM m; + + if ( !p ) nd_send_int(0); + else { + len = LEN(p); + nd_send_int(len); + m = BDY(p); + for ( i = 0; i < len; i++, m = NEXT(m) ) { + nd_send_int(CM(m)); + nd_send_intarray(DL(m),nd_wpd); + } + } +} + +NDV nd_recv_ndv() +{ + int len,i; + NMV m,m0; + NDV r; + + len = nd_recv_int(); + if ( !len ) return 0; + else { + m0 = m = (NMV)GC_malloc_atomic_ignore_off_page(nmv_adv*len); +#if 0 + ndv_alloc += len*nmv_adv; +#endif + for ( i = 0; i < len; i++, NMV_ADV(m) ) { + CM(m) = nd_recv_int(); + nd_recv_intarray(DL(m),nd_wpd); + } + MKNDV(nd_nvar,m0,len,r); + return r; + } +} + +int ox_exec_f4_red(Q proc) +{ + Obj obj; + STRING fname; + NODE arg; + int s; + extern int ox_need_conv,ox_file_io; + + MKSTR(fname,"nd_exec_f4_red"); + arg = mknode(2,proc,fname); + Pox_cmo_rpc(arg,&obj); + s = get_ox_server_id(QTOS(proc)); + nd_write = iofp[s].out; + nd_read = iofp[s].in; + ox_need_conv = ox_file_io = 0; + return s; +} + +NODE nd_f4_red_dist(int m,ND_pairs sp0,UINT *s0vect,int col,NODE rp0) +{ + int nsp,nred; + int i,rank,s; + NODE rp,r0,r; + ND_pairs sp; + NM_ind_pair pair; + NMV nmv; + NM nm; + NDV nf; + Obj proc,dmy; + + ox_launch_main(0,0,&proc); + s = ox_exec_f4_red((Q)proc); + + nd_send_int(m); + nd_send_int(nd_nvar); + nd_send_int(nd_bpe); + nd_send_int(nd_wpd); + nd_send_int(nmv_adv); + + saveobj(nd_write,dp_current_spec->obj); fflush(nd_write); + + nd_send_int(nd_psn); + for ( i = 0; i < nd_psn; i++ ) nd_send_ndv(nd_ps[i]); + + for ( sp = sp0, nsp = 0; sp; sp = NEXT(sp), nsp++ ); + nd_send_int(nsp); + for ( i = 0, sp = sp0; i < nsp; i++, sp = NEXT(sp) ) { + nd_send_int(sp->i1); nd_send_int(sp->i2); + } + + nd_send_int(col); nd_send_intarray(s0vect,col*nd_wpd); + + nred = length(rp0); nd_send_int(nred); + for ( i = 0, rp = rp0; i < nred; i++, rp = NEXT(rp) ) { + pair = (NM_ind_pair)BDY(rp); + nd_send_int(pair->index); + nd_send_intarray(pair->mul->dl,nd_wpd); + } + fflush(nd_write); + rank = nd_recv_int(); + fprintf(asir_out,"rank=%d\n",rank); + r0 = 0; + for ( i = 0; i < rank; i++ ) { + nf = nd_recv_ndv(); + NEXTNODE(r0,r); BDY(r) = (pointer)nf; + } + Pox_shutdown(mknode(1,proc),&dmy); + return r0; +} + +/* server side */ + +void nd_exec_f4_red_dist() +{ + int m,i,nsp,col,s0size,nred,spcol,j,k; + NM_ind_pair *rp0; + NDV nf; + UINT *s0vect; + IndArray *imat; + int *rhead; + int **spmat; + UINT *svect,*v; + ND_pairs *sp0; + int *colstat; + int a,sprow,rank; + struct order_spec *ord; + Obj ordspec; + ND spol; + int maxrs; + int *spsugar; + + nd_read = iofp[0].in; + nd_write = iofp[0].out; + m = nd_recv_int(); + nd_nvar = nd_recv_int(); + nd_bpe = nd_recv_int(); + nd_wpd = nd_recv_int(); + nmv_adv = nd_recv_int(); + + loadobj(nd_read,&ordspec); + create_order_spec(0,ordspec,&ord); + nd_init_ord(ord); + nd_setup_parameters(nd_nvar,0); + + nd_psn = nd_recv_int(); + nd_ps = (NDV *)MALLOC(nd_psn*sizeof(NDV)); + nd_bound = (UINT **)MALLOC(nd_psn*sizeof(UINT *)); + for ( i = 0; i < nd_psn; i++ ) { + nd_ps[i] = nd_recv_ndv(); + nd_bound[i] = ndv_compute_bound(nd_ps[i]); + } + + nsp = nd_recv_int(); + sp0 = (ND_pairs *)MALLOC(nsp*sizeof(ND_pairs)); + for ( i = 0; i < nsp; i++ ) { + NEWND_pairs(sp0[i]); + sp0[i]->i1 = nd_recv_int(); sp0[i]->i2 = nd_recv_int(); + ndl_lcm(HDL(nd_ps[sp0[i]->i1]),HDL(nd_ps[sp0[i]->i2]),LCM(sp0[i])); + } + + col = nd_recv_int(); + s0size = col*nd_wpd; + s0vect = (UINT *)MALLOC(s0size*sizeof(UINT)); + nd_recv_intarray(s0vect,s0size); + + nred = nd_recv_int(); + rp0 = (NM_ind_pair *)MALLOC(nred*sizeof(NM_ind_pair)); + for ( i = 0; i < nred; i++ ) { + rp0[i] = (NM_ind_pair)MALLOC(sizeof(struct oNM_ind_pair)); + rp0[i]->index = nd_recv_int(); + rp0[i]->mul = (NM)MALLOC(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT)); + nd_recv_intarray(rp0[i]->mul->dl,nd_wpd); + } + + spcol = col-nred; + imat = (IndArray *)MALLOC(nred*sizeof(IndArray)); + rhead = (int *)MALLOC(col*sizeof(int)); + for ( i = 0; i < col; i++ ) rhead[i] = 0; + + /* construction of index arrays */ + for ( i = 0; i < nred; i++ ) { + imat[i] = nm_ind_pair_to_vect_compress(m,s0vect,col,rp0[i]); + rhead[imat[i]->head] = 1; + } + + /* elimination (1st step) */ + spmat = (int **)MALLOC(nsp*sizeof(UINT *)); + svect = (UINT *)MALLOC(col*sizeof(UINT)); + spsugar = (int *)ALLOCA(nsp*sizeof(UINT)); + for ( a = sprow = 0; a < nsp; a++ ) { + nd_sp(m,0,sp0[a],&spol); + if ( !spol ) continue; + nd_to_vect(m,s0vect,col,spol,svect); + if ( m == -1 ) + maxrs = ndv_reduce_vect_sf(m,svect,col,imat,rp0,nred); + else + maxrs = ndv_reduce_vect(m,svect,col,imat,rp0,nred); + for ( i = 0; i < col; i++ ) if ( svect[i] ) break; + if ( i < col ) { + spmat[sprow] = v = (UINT *)MALLOC(spcol*sizeof(UINT)); + for ( j = k = 0; j < col; j++ ) + if ( !rhead[j] ) v[k++] = svect[j]; + spsugar[sprow] = MAX(maxrs,SG(spol)); + sprow++; + } + nd_free(spol); + } + /* elimination (2nd step) */ + colstat = (int *)ALLOCA(spcol*sizeof(int)); + if ( m == -1 ) + rank = nd_gauss_elim_sf(spmat,spsugar,sprow,spcol,m,colstat); + else + rank = nd_gauss_elim_mod(spmat,spsugar,sprow,spcol,m,colstat); + nd_send_int(rank); + for ( i = 0; i < rank; i++ ) { + nf = vect_to_ndv(spmat[i],spcol,col,rhead,s0vect); + nd_send_ndv(nf); + } + fflush(nd_write); +} + +int nd_gauss_elim_q(Q **mat0,int *sugar,int row,int col,int *colstat) +{ + int mod,i,j,t,c,rank,rank0,inv; + int *ci,*ri; + Q dn; + MAT m,nm; + int **wmat; + + /* XXX */ + mod = 99999989; + wmat = (int **)ALLOCA(row*sizeof(int *)); + for ( i = 0; i < row; i++ ) { + wmat[i] = (int *)ALLOCA(col*sizeof(int)); + for ( j = 0; j < col; j++ ) { + if ( mat0[i][j] ) { + t = rem(NM(mat0[i][j]),mod); + if ( SGN(mat0[i][j]) < 0 ) t = mod-t; + wmat[i][j] = t; + } else + wmat[i][j] = 0; + } + } + rank0 = nd_gauss_elim_mod(wmat,sugar,row,col,mod,colstat); + NEWMAT(m); m->row = row; m->col = col; m->body = (pointer **)mat0; + rank = generic_gauss_elim(m,&nm,&dn,&ri,&ci); + if ( rank != rank0 ) + error("afo"); + for ( i = 0; i < row; i++ ) + for ( j = 0; j < col; j++ ) + mat0[i][j] = 0; + c = col-rank; + for ( i = 0; i < rank; i++ ) { + mat0[i][ri[i]] = dn; + for ( j = 0; j < c; j++ ) + mat0[i][ci[j]] = (Q)BDY(nm)[i][j]; + } + inv = invm(rem(NM(dn),mod),mod); + if ( SGN(dn) < 0 ) inv = mod-inv; + for ( i = 0; i < row; i++ ) + for ( j = 0; j < col; j++ ) { + if ( mat0[i][j] ) { + t = rem(NM(mat0[i][j]),mod); + if ( SGN(mat0[i][j]) < 0 ) t = mod-t; + } else + t = 0; + c = dmar(t,inv,0,mod); + if ( wmat[i][j] != c ) + error("afo"); + } + return rank; +} + +int nd_gauss_elim_mod(int **mat0,int *sugar,int row,int col,int md,int *colstat) +{ + int i,j,k,l,inv,a,rank,s; + unsigned int *t,*pivot,*pk; + unsigned int **mat; + + mat = (unsigned int **)mat0; + for ( rank = 0, j = 0; j < col; j++ ) { + for ( i = rank; i < row; i++ ) + mat[i][j] %= md; + 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; + s = sugar[i]; sugar[i] = sugar[rank]; sugar[rank] = s; + } + pivot = mat[rank]; + s = sugar[rank]; + inv = invm(pivot[j],md); + for ( k = j, pk = pivot+k; k < col; k++, pk++ ) + if ( *pk ) { + if ( *pk >= (unsigned int)md ) + *pk %= md; + DMAR(*pk,inv,0,md,*pk) + } + for ( i = rank+1; i < row; i++ ) { + t = mat[i]; + if ( a = t[j] ) { + sugar[i] = MAX(sugar[i],s); + red_by_vect(md,t+j,pivot+j,md-a,col-j); + } + } + rank++; + } + for ( j = col-1, l = rank-1; j >= 0; j-- ) + if ( colstat[j] ) { + pivot = mat[l]; + s = sugar[l]; + for ( i = 0; i < l; i++ ) { + t = mat[i]; + t[j] %= md; + if ( a = t[j] ) { + sugar[i] = MAX(sugar[i],s); + red_by_vect(md,t+j,pivot+j,md-a,col-j); + } + } + l--; + } + for ( j = 0, l = 0; l < rank; j++ ) + if ( colstat[j] ) { + t = mat[l]; + for ( k = j; k < col; k++ ) + if ( t[k] >= (unsigned int)md ) + t[k] %= md; + l++; + } + return rank; +} + +int nd_gauss_elim_sf(int **mat0,int *sugar,int row,int col,int md,int *colstat) +{ + int i,j,k,l,inv,a,rank,s; + unsigned int *t,*pivot,*pk; + unsigned int **mat; + + mat = (unsigned int **)mat0; + for ( rank = 0, j = 0; j < col; j++ ) { + 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; + s = sugar[i]; sugar[i] = sugar[rank]; sugar[rank] = s; + } + pivot = mat[rank]; + s = sugar[rank]; + inv = _invsf(pivot[j]); + for ( k = j, pk = pivot+k; k < col; k++, pk++ ) + if ( *pk ) + *pk = _mulsf(*pk,inv); + for ( i = rank+1; i < row; i++ ) { + t = mat[i]; + if ( a = t[j] ) { + sugar[i] = MAX(sugar[i],s); + red_by_vect_sf(md,t+j,pivot+j,_chsgnsf(a),col-j); + } + } + rank++; + } + for ( j = col-1, l = rank-1; j >= 0; j-- ) + if ( colstat[j] ) { + pivot = mat[l]; + s = sugar[l]; + for ( i = 0; i < l; i++ ) { + t = mat[i]; + if ( a = t[j] ) { + sugar[i] = MAX(sugar[i],s); + red_by_vect_sf(md,t+j,pivot+j,_chsgnsf(a),col-j); + } + } + l--; + } + return rank; +} + +int ndv_ishomo(NDV p) +{ + NMV m; + int len,h; + + if ( !p ) return 1; + len = LEN(p); + m = BDY(p); + h = TD(DL(m)); + NMV_ADV(m); + for ( len--; len; len--, NMV_ADV(m) ) + if ( TD(DL(m)) != h ) return 0; + return 1; +} + +void ndv_save(NDV p,int index) +{ + FILE *s; + char name[BUFSIZ]; + short id; + int nv,sugar,len,n,i,td,e,j; + NMV m; + unsigned int *dl; + + sprintf(name,"%s/%d",Demand,index); + s = fopen(name,"w"); + savevl(s,0); + if ( !p ) { + saveobj(s,0); + return; + } + id = O_DP; + nv = NV(p); + sugar = SG(p); + len = LEN(p); + write_short(s,&id); write_int(s,&nv); write_int(s,&sugar); + write_int(s,&len); + + for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) { + saveobj(s,(Obj)CQ(m)); + dl = DL(m); + td = TD(dl); + write_int(s,&td); + for ( j = 0; j < nv; j++ ) { + e = GET_EXP(dl,j); + write_int(s,&e); + } + } + fclose(s); +} + +NDV ndv_load(int index) +{ + FILE *s; + char name[BUFSIZ]; + short id; + int nv,sugar,len,n,i,td,e,j; + NDV d; + NMV m0,m; + unsigned int *dl; + Obj obj; + + sprintf(name,"%s/%d",Demand,index); + s = fopen(name,"r"); + if ( !s ) return 0; + + skipvl(s); + read_short(s,&id); + if ( !id ) return 0; + read_int(s,&nv); + read_int(s,&sugar); + read_int(s,&len); + + m0 = m = MALLOC(len*nmv_adv); + for ( i = 0; i < len; i++, NMV_ADV(m) ) { + loadobj(s,&obj); CQ(m) = (Q)obj; + dl = DL(m); + ndl_zero(dl); + read_int(s,&td); TD(dl) = td; + for ( j = 0; j < nv; j++ ) { + read_int(s,&e); + PUT_EXP(dl,j,e); + } + if ( nd_blockmask ) ndl_weight_mask(dl); + } + fclose(s); + MKNDV(nv,m0,len,d); + SG(d) = sugar; + return d; +} + +void nd_det(int mod,MAT f,P *rp) +{ + VL fv,tv; + int n,i,j,max,e,nvar,sgn,k0,l0,len0,len,k,l,a; + pointer **m; + Q mone; + NDV **dm; + NDV *t,*mi,*mj; + NDV d,s,mij,mjj; + ND u; + NMV nmv; + PGeoBucket bucket; + struct order_spec *ord; + + create_order_spec(0,0,&ord); + nd_init_ord(ord); + get_vars((Obj)f,&fv); + if ( f->row != f->col ) + error("nd_det : non-square matrix"); + n = f->row; + for ( nvar = 0, tv = fv; tv; tv = NEXT(tv), nvar++ ); + m = f->body; + for ( i = 0, max = 0; i < n; i++ ) + for ( j = 0; j < n; j++ ) + for ( tv = fv; tv; tv = NEXT(tv) ) { + e = getdeg(tv->v,(P)m[i][j]); + max = MAX(e,max); + } + nd_setup_parameters(nvar,1024); + dm = (NDV **)almat_pointer(n,n); + for ( i = 0, max = 0; i < n; i++ ) + for ( j = 0; j < n; j++ ) { + dm[i][j] = ptondv(CO,fv,m[i][j]); + if ( mod ) ndv_mod(mod,dm[i][j]); + if ( dm[i][j] && !LEN(dm[i][j]) ) dm[i][j] = 0; + } + d = ptondv(CO,fv,(P)ONE); + if ( mod ) ndv_mod(mod,d); + chsgnq(ONE,&mone); + for ( j = 0, sgn = 1; j < n; j++ ) { + if ( DP_Print ) fprintf(stderr,"j=%d\n",j); + for ( i = j; i < n && !dm[i][j]; i++ ); + if ( i == n ) { + *rp = 0; + return; + } + k0 = i; l0 = j; len0 = LEN(dm[k0][l0]); + for ( k = j; k < n; k++ ) + for ( l = j; l < n; l++ ) + if ( dm[k][l] && LEN(dm[k][l]) < len0 ) { + k0 = k; l0 = l; len0 = LEN(dm[k][l]); + } + if ( k0 != j ) { + t = dm[j]; dm[j] = dm[k0]; dm[k0] = t; + sgn = -sgn; + } + if ( l0 != j ) { + for ( k = j; k < n; k++ ) { + s = dm[k][j]; dm[k][j] = dm[k][l0]; dm[k][l0] = s; + } + sgn = -sgn; + } + for ( i = j+1, mj = dm[j], mjj = mj[j]; i < n; i++ ) { + if ( DP_Print ) fprintf(stderr," 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); + bucket = create_pbucket(); + if ( mi[k] ) { + nmv = BDY(mjj); len = LEN(mjj); + for ( a = 0; a < len; a++, NMV_ADV(nmv) ) { + u = ndv_mul_nmv_trunc(mod,nmv,mi[k],DL(BDY(d))); + add_pbucket(mod,bucket,u); + } + } + if ( mj[k] && mij ) { + nmv = BDY(mij); len = LEN(mij); + for ( a = 0; a < len; a++, NMV_ADV(nmv) ) { + u = ndv_mul_nmv_trunc(mod,nmv,mj[k],DL(BDY(d))); + add_pbucket(mod,bucket,u); + } + } + u = nd_quo(mod,bucket,d); + mi[k] = ndtondv(mod,u); + } + if ( DP_Print ) fprintf(stderr,"\n",k); + } + d = mjj; + } + if ( sgn < 0 ) + if ( mod ) + ndv_mul_c(mod,d,mod-1); + else + ndv_mul_c_q(d,mone); + *rp = ndvtop(mod,CO,fv,d); +} + +ND ndv_mul_nmv_trunc(int mod,NMV m0,NDV p,UINT *d) +{ + NM mr,mr0; + NM tnm; + NMV m; + UINT *d0,*dt,*dm; + int c,n,td,i,c1,c2,len; + Q q; + ND r; + + if ( !p ) return 0; + else { + n = NV(p); m = BDY(p); len = LEN(p); + d0 = DL(m0); + td = TD(d); + mr0 = 0; + NEWNM(tnm); + if ( mod ) { + c = CM(m0); + for ( i = 0; i < len; i++, NMV_ADV(m) ) { + ndl_add(DL(m),d0,DL(tnm)); + if ( ndl_reducible(DL(tnm),d) ) { + NEXTNM(mr0,mr); + c1 = CM(m); DMAR(c1,c,0,mod,c2); CM(mr) = c2; + ndl_copy(DL(tnm),DL(mr)); + } + } + } else { + q = CQ(m0); + for ( i = 0; i < len; i++, NMV_ADV(m) ) { + ndl_add(DL(m),d0,DL(tnm)); + if ( ndl_reducible(DL(tnm),d) ) { + NEXTNM(mr0,mr); + mulq(CQ(m),q,&CQ(mr)); + ndl_copy(DL(tnm),DL(mr)); + } + } + } + if ( !mr0 ) + return 0; + else { + NEXT(mr) = 0; + for ( len = 0, mr = mr0; mr; mr = NEXT(mr), len++ ); + MKND(NV(p),mr0,len,r); + SG(r) = SG(p) + TD(d0); + return r; + } + } +}