Annotation of OpenXM_contrib2/asir2000/engine/nd.c, Revision 1.139
1.139 ! noro 1: /* $OpenXM: OpenXM_contrib2/asir2000/engine/nd.c,v 1.138 2006/06/11 11:41:15 noro Exp $ */
1.2 noro 2:
1.94 noro 3: #include "nd.h"
1.63 noro 4:
1.131 noro 5: int diag_period = 6;
1.61 noro 6: int (*ndl_compare_function)(UINT *a1,UINT *a2);
1.94 noro 7: int nd_dcomp;
8: NM _nm_free_list;
9: ND _nd_free_list;
10: ND_pairs _ndp_free_list;
1.32 noro 11:
1.121 noro 12: static int nd_ntrans;
1.117 noro 13: static int nd_nalg;
1.103 noro 14: #if 0
1.74 noro 15: static int ndv_alloc;
1.103 noro 16: #endif
1.87 noro 17: #if 1
1.69 noro 18: static int nd_f4_nsp=0x7fffffff;
1.87 noro 19: #else
20: static int nd_f4_nsp=50;
21: #endif
1.42 noro 22: static double nd_scale=2;
1.61 noro 23: static UINT **nd_bound;
1.42 noro 24: static struct order_spec *nd_ord;
25: static EPOS nd_epos;
1.43 noro 26: static BlockMask nd_blockmask;
1.42 noro 27: static int nd_nvar;
28: static int nd_isrlex;
29: static int nd_epw,nd_bpe,nd_wpd,nd_exporigin;
1.61 noro 30: static UINT nd_mask[32];
31: static UINT nd_mask0,nd_mask1;
1.42 noro 32:
1.20 noro 33: static NDV *nd_ps;
1.53 noro 34: static NDV *nd_ps_trace;
1.42 noro 35: static RHist *nd_psh;
36: static int nd_psn,nd_pslen;
37: static RHist *nd_red;
1.96 noro 38: static int *nd_work_vector;
39: static int **nd_matrix;
40: static int nd_matrix_len;
1.97 noro 41: static struct weight_or_block *nd_worb;
42: static int nd_worb_len;
1.42 noro 43: static int nd_found,nd_create,nd_notfirst;
44: static int nmv_adv;
1.77 noro 45: static int nd_demand;
1.1 noro 46:
1.119 noro 47: NumberField get_numberfield();
1.114 noro 48: UINT *nd_det_compute_bound(NDV **dm,int n,int j);
49: void nd_det_reconstruct(NDV **dm,int n,int j,NDV d);
1.118 noro 50: int nd_monic(int m,ND *p);
1.129 noro 51: NDV plain_vect_to_ndv_q(Q *mat,int col,UINT *s0vect);
1.114 noro 52:
1.1 noro 53: void nd_free_private_storage()
54: {
55: _nm_free_list = 0;
1.5 noro 56: _ndp_free_list = 0;
1.71 noro 57: #if 0
1.1 noro 58: GC_gcollect();
1.71 noro 59: #endif
1.1 noro 60: }
61:
62: void _NM_alloc()
63: {
64: NM p;
65: int i;
66:
1.11 noro 67: for ( i = 0; i < 1024; i++ ) {
1.61 noro 68: p = (NM)GC_malloc(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT));
1.1 noro 69: p->next = _nm_free_list; _nm_free_list = p;
70: }
71: }
72:
73: void _ND_alloc()
74: {
75: ND p;
76: int i;
77:
78: for ( i = 0; i < 1024; i++ ) {
79: p = (ND)GC_malloc(sizeof(struct oND));
80: p->body = (NM)_nd_free_list; _nd_free_list = p;
81: }
82: }
83:
84: void _NDP_alloc()
85: {
86: ND_pairs p;
87: int i;
88:
1.11 noro 89: for ( i = 0; i < 1024; i++ ) {
1.1 noro 90: p = (ND_pairs)GC_malloc(sizeof(struct oND_pairs)
1.61 noro 91: +(nd_wpd-1)*sizeof(UINT));
1.1 noro 92: p->next = _ndp_free_list; _ndp_free_list = p;
93: }
94: }
95:
1.30 noro 96: INLINE int nd_length(ND p)
1.1 noro 97: {
98: NM m;
99: int i;
100:
101: if ( !p )
102: return 0;
103: else {
104: for ( i = 0, m = BDY(p); m; m = NEXT(m), i++ );
105: return i;
106: }
107: }
108:
1.61 noro 109: INLINE int ndl_reducible(UINT *d1,UINT *d2)
1.1 noro 110: {
1.61 noro 111: UINT u1,u2;
1.1 noro 112: int i,j;
113:
1.34 noro 114: if ( TD(d1) < TD(d2) ) return 0;
1.65 noro 115: #if USE_UNROLL
1.1 noro 116: switch ( nd_bpe ) {
1.62 noro 117: case 3:
118: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
119: u1 = d1[i]; u2 = d2[i];
120: if ( (u1&0x38000000) < (u2&0x38000000) ) return 0;
121: if ( (u1& 0x7000000) < (u2& 0x7000000) ) return 0;
122: if ( (u1& 0xe00000) < (u2& 0xe00000) ) return 0;
123: if ( (u1& 0x1c0000) < (u2& 0x1c0000) ) return 0;
124: if ( (u1& 0x38000) < (u2& 0x38000) ) return 0;
125: if ( (u1& 0x7000) < (u2& 0x7000) ) return 0;
126: if ( (u1& 0xe00) < (u2& 0xe00) ) return 0;
127: if ( (u1& 0x1c0) < (u2& 0x1c0) ) return 0;
128: if ( (u1& 0x38) < (u2& 0x38) ) return 0;
129: if ( (u1& 0x7) < (u2& 0x7) ) return 0;
130: }
131: return 1;
132: break;
1.1 noro 133: case 4:
1.41 noro 134: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 135: u1 = d1[i]; u2 = d2[i];
136: if ( (u1&0xf0000000) < (u2&0xf0000000) ) return 0;
1.62 noro 137: if ( (u1& 0xf000000) < (u2& 0xf000000) ) return 0;
138: if ( (u1& 0xf00000) < (u2& 0xf00000) ) return 0;
139: if ( (u1& 0xf0000) < (u2& 0xf0000) ) return 0;
140: if ( (u1& 0xf000) < (u2& 0xf000) ) return 0;
141: if ( (u1& 0xf00) < (u2& 0xf00) ) return 0;
142: if ( (u1& 0xf0) < (u2& 0xf0) ) return 0;
143: if ( (u1& 0xf) < (u2& 0xf) ) return 0;
1.1 noro 144: }
145: return 1;
146: break;
147: case 6:
1.41 noro 148: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 149: u1 = d1[i]; u2 = d2[i];
150: if ( (u1&0x3f000000) < (u2&0x3f000000) ) return 0;
1.62 noro 151: if ( (u1& 0xfc0000) < (u2& 0xfc0000) ) return 0;
152: if ( (u1& 0x3f000) < (u2& 0x3f000) ) return 0;
153: if ( (u1& 0xfc0) < (u2& 0xfc0) ) return 0;
154: if ( (u1& 0x3f) < (u2& 0x3f) ) return 0;
1.1 noro 155: }
156: return 1;
157: break;
158: case 8:
1.41 noro 159: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 160: u1 = d1[i]; u2 = d2[i];
161: if ( (u1&0xff000000) < (u2&0xff000000) ) return 0;
1.62 noro 162: if ( (u1& 0xff0000) < (u2& 0xff0000) ) return 0;
163: if ( (u1& 0xff00) < (u2& 0xff00) ) return 0;
164: if ( (u1& 0xff) < (u2& 0xff) ) return 0;
1.1 noro 165: }
166: return 1;
167: break;
168: case 16:
1.41 noro 169: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 170: u1 = d1[i]; u2 = d2[i];
171: if ( (u1&0xffff0000) < (u2&0xffff0000) ) return 0;
1.62 noro 172: if ( (u1& 0xffff) < (u2& 0xffff) ) return 0;
1.1 noro 173: }
174: return 1;
175: break;
176: case 32:
1.41 noro 177: for ( i = nd_exporigin; i < nd_wpd; i++ )
1.1 noro 178: if ( d1[i] < d2[i] ) return 0;
179: return 1;
180: break;
181: default:
1.41 noro 182: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 183: u1 = d1[i]; u2 = d2[i];
184: for ( j = 0; j < nd_epw; j++ )
185: if ( (u1&nd_mask[j]) < (u2&nd_mask[j]) ) return 0;
186: }
187: return 1;
188: }
1.65 noro 189: #else
190: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
191: u1 = d1[i]; u2 = d2[i];
192: for ( j = 0; j < nd_epw; j++ )
193: if ( (u1&nd_mask[j]) < (u2&nd_mask[j]) ) return 0;
194: }
195: return 1;
196: #endif
1.1 noro 197: }
198:
1.61 noro 199: /*
200: * If the current order is a block order,
201: * then the last block is length 1 and contains
202: * the homo variable. Otherwise, the original
203: * order is either 0 or 2.
204: */
205:
206: void ndl_homogenize(UINT *d,UINT *r,int obpe,EPOS oepos,int weight)
1.23 noro 207: {
1.61 noro 208: int w,i,e,n,omask0;
209:
210: omask0 = (1<<obpe)-1;
211: n = nd_nvar-1;
212: ndl_zero(r);
213: for ( i = 0; i < n; i++ ) {
214: e = GET_EXP_OLD(d,i);
215: PUT_EXP(r,i,e);
216: }
217: w = TD(d);
218: PUT_EXP(r,nd_nvar-1,weight-w);
219: TD(r) = weight;
220: if ( nd_blockmask ) ndl_weight_mask(r);
221: }
222:
223: void ndl_dehomogenize(UINT *d)
224: {
225: UINT mask;
226: UINT h;
1.31 noro 227: int i,bits;
1.23 noro 228:
1.44 noro 229: if ( nd_blockmask ) {
230: h = GET_EXP(d,nd_nvar-1);
1.45 noro 231: XOR_EXP(d,nd_nvar-1,h);
1.44 noro 232: TD(d) -= h;
233: d[nd_exporigin-1] -= h;
234: } else {
235: if ( nd_isrlex ) {
236: if ( nd_bpe == 32 ) {
237: h = d[nd_exporigin];
238: for ( i = nd_exporigin+1; i < nd_wpd; i++ )
239: d[i-1] = d[i];
240: d[i-1] = 0;
241: TD(d) -= h;
242: } else {
243: bits = nd_epw*nd_bpe;
244: mask = bits==32?0xffffffff:((1<<(nd_epw*nd_bpe))-1);
245: h = (d[nd_exporigin]>>((nd_epw-1)*nd_bpe))&nd_mask0;
246: for ( i = nd_exporigin; i < nd_wpd; i++ )
247: d[i] = ((d[i]<<nd_bpe)&mask)
248: |(i+1<nd_wpd?((d[i+1]>>((nd_epw-1)*nd_bpe))&nd_mask0):0);
249: TD(d) -= h;
250: }
1.45 noro 251: } else {
252: h = GET_EXP(d,nd_nvar-1);
253: XOR_EXP(d,nd_nvar-1,h);
254: TD(d) -= h;
255: }
1.44 noro 256: }
1.23 noro 257: }
258:
1.61 noro 259: void ndl_lcm(UINT *d1,unsigned *d2,UINT *d)
1.1 noro 260: {
1.61 noro 261: UINT t1,t2,u,u1,u2;
1.43 noro 262: int i,j,l;
1.1 noro 263:
1.65 noro 264: #if USE_UNROLL
1.1 noro 265: switch ( nd_bpe ) {
1.62 noro 266: case 3:
267: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
268: u1 = d1[i]; u2 = d2[i];
269: t1 = (u1&0x38000000); t2 = (u2&0x38000000); u = t1>t2?t1:t2;
270: t1 = (u1& 0x7000000); t2 = (u2& 0x7000000); u |= t1>t2?t1:t2;
271: t1 = (u1& 0xe00000); t2 = (u2& 0xe00000); u |= t1>t2?t1:t2;
272: t1 = (u1& 0x1c0000); t2 = (u2& 0x1c0000); u |= t1>t2?t1:t2;
273: t1 = (u1& 0x38000); t2 = (u2& 0x38000); u |= t1>t2?t1:t2;
274: t1 = (u1& 0x7000); t2 = (u2& 0x7000); u |= t1>t2?t1:t2;
275: t1 = (u1& 0xe00); t2 = (u2& 0xe00); u |= t1>t2?t1:t2;
276: t1 = (u1& 0x1c0); t2 = (u2& 0x1c0); u |= t1>t2?t1:t2;
277: t1 = (u1& 0x38); t2 = (u2& 0x38); u |= t1>t2?t1:t2;
278: t1 = (u1& 0x7); t2 = (u2& 0x7); u |= t1>t2?t1:t2;
279: d[i] = u;
280: }
281: break;
1.1 noro 282: case 4:
1.41 noro 283: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 284: u1 = d1[i]; u2 = d2[i];
285: t1 = (u1&0xf0000000); t2 = (u2&0xf0000000); u = t1>t2?t1:t2;
1.62 noro 286: t1 = (u1& 0xf000000); t2 = (u2& 0xf000000); u |= t1>t2?t1:t2;
287: t1 = (u1& 0xf00000); t2 = (u2& 0xf00000); u |= t1>t2?t1:t2;
288: t1 = (u1& 0xf0000); t2 = (u2& 0xf0000); u |= t1>t2?t1:t2;
289: t1 = (u1& 0xf000); t2 = (u2& 0xf000); u |= t1>t2?t1:t2;
290: t1 = (u1& 0xf00); t2 = (u2& 0xf00); u |= t1>t2?t1:t2;
291: t1 = (u1& 0xf0); t2 = (u2& 0xf0); u |= t1>t2?t1:t2;
292: t1 = (u1& 0xf); t2 = (u2& 0xf); u |= t1>t2?t1:t2;
1.1 noro 293: d[i] = u;
294: }
295: break;
296: case 6:
1.41 noro 297: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 298: u1 = d1[i]; u2 = d2[i];
299: t1 = (u1&0x3f000000); t2 = (u2&0x3f000000); u = t1>t2?t1:t2;
1.62 noro 300: t1 = (u1& 0xfc0000); t2 = (u2& 0xfc0000); u |= t1>t2?t1:t2;
301: t1 = (u1& 0x3f000); t2 = (u2& 0x3f000); u |= t1>t2?t1:t2;
302: t1 = (u1& 0xfc0); t2 = (u2& 0xfc0); u |= t1>t2?t1:t2;
303: t1 = (u1& 0x3f); t2 = (u2& 0x3f); u |= t1>t2?t1:t2;
1.1 noro 304: d[i] = u;
305: }
306: break;
307: case 8:
1.41 noro 308: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 309: u1 = d1[i]; u2 = d2[i];
310: t1 = (u1&0xff000000); t2 = (u2&0xff000000); u = t1>t2?t1:t2;
1.62 noro 311: t1 = (u1& 0xff0000); t2 = (u2& 0xff0000); u |= t1>t2?t1:t2;
312: t1 = (u1& 0xff00); t2 = (u2& 0xff00); u |= t1>t2?t1:t2;
313: t1 = (u1& 0xff); t2 = (u2& 0xff); u |= t1>t2?t1:t2;
1.1 noro 314: d[i] = u;
315: }
316: break;
317: case 16:
1.41 noro 318: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 319: u1 = d1[i]; u2 = d2[i];
320: t1 = (u1&0xffff0000); t2 = (u2&0xffff0000); u = t1>t2?t1:t2;
1.62 noro 321: t1 = (u1& 0xffff); t2 = (u2& 0xffff); u |= t1>t2?t1:t2;
1.1 noro 322: d[i] = u;
323: }
324: break;
325: case 32:
1.41 noro 326: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 327: u1 = d1[i]; u2 = d2[i];
328: d[i] = u1>u2?u1:u2;
329: }
330: break;
331: default:
1.41 noro 332: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 333: u1 = d1[i]; u2 = d2[i];
334: for ( j = 0, u = 0; j < nd_epw; j++ ) {
335: t1 = (u1&nd_mask[j]); t2 = (u2&nd_mask[j]); u |= t1>t2?t1:t2;
336: }
337: d[i] = u;
338: }
339: break;
340: }
1.65 noro 341: #else
342: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
343: u1 = d1[i]; u2 = d2[i];
344: for ( j = 0, u = 0; j < nd_epw; j++ ) {
345: t1 = (u1&nd_mask[j]); t2 = (u2&nd_mask[j]); u |= t1>t2?t1:t2;
346: }
347: d[i] = u;
348: }
349: #endif
1.39 noro 350: TD(d) = ndl_weight(d);
1.61 noro 351: if ( nd_blockmask ) ndl_weight_mask(d);
1.57 noro 352: }
353:
1.61 noro 354: int ndl_weight(UINT *d)
1.1 noro 355: {
1.61 noro 356: UINT t,u;
1.1 noro 357: int i,j;
358:
1.60 noro 359: if ( current_dl_weight_vector )
360: for ( i = 0, t = 0; i < nd_nvar; i++ ) {
361: u = GET_EXP(d,i);
362: t += MUL_WEIGHT(u,i);
363: }
364: else
365: for ( t = 0, i = nd_exporigin; i < nd_wpd; i++ ) {
366: u = d[i];
367: for ( j = 0; j < nd_epw; j++, u>>=nd_bpe )
368: t += (u&nd_mask0);
369: }
1.1 noro 370: return t;
371: }
372:
1.61 noro 373: void ndl_weight_mask(UINT *d)
1.43 noro 374: {
1.61 noro 375: UINT t,u;
376: UINT *mask;
377: int i,j,k,l;
1.43 noro 378:
1.61 noro 379: l = nd_blockmask->n;
380: for ( k = 0; k < l; k++ ) {
381: mask = nd_blockmask->mask[k];
382: if ( current_dl_weight_vector )
383: for ( i = 0, t = 0; i < nd_nvar; i++ ) {
384: u = GET_EXP_MASK(d,i,mask);
385: t += MUL_WEIGHT(u,i);
386: }
387: else
388: for ( t = 0, i = nd_exporigin; i < nd_wpd; i++ ) {
389: u = d[i]&mask[i];
390: for ( j = 0; j < nd_epw; j++, u>>=nd_bpe )
391: t += (u&nd_mask0);
392: }
393: d[k+1] = t;
394: }
1.43 noro 395: }
396:
1.61 noro 397: int ndl_lex_compare(UINT *d1,UINT *d2)
1.1 noro 398: {
399: int i;
400:
1.41 noro 401: d1 += nd_exporigin;
402: d2 += nd_exporigin;
403: for ( i = nd_exporigin; i < nd_wpd; i++, d1++, d2++ )
1.1 noro 404: if ( *d1 > *d2 )
1.32 noro 405: return nd_isrlex ? -1 : 1;
1.1 noro 406: else if ( *d1 < *d2 )
1.32 noro 407: return nd_isrlex ? 1 : -1;
1.1 noro 408: return 0;
409: }
410:
1.61 noro 411: int ndl_block_compare(UINT *d1,UINT *d2)
1.43 noro 412: {
413: int i,l,j,ord_o,ord_l;
414: struct order_pair *op;
1.61 noro 415: UINT t1,t2,m;
416: UINT *mask;
1.43 noro 417:
418: l = nd_blockmask->n;
419: op = nd_blockmask->order_pair;
420: for ( j = 0; j < l; j++ ) {
421: mask = nd_blockmask->mask[j];
422: ord_o = op[j].order;
423: if ( ord_o < 2 )
1.44 noro 424: if ( (t1=d1[j+1]) > (t2=d2[j+1]) ) return 1;
425: else if ( t1 < t2 ) return -1;
1.43 noro 426: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.44 noro 427: m = mask[i];
428: t1 = d1[i]&m;
429: t2 = d2[i]&m;
1.43 noro 430: if ( t1 > t2 )
431: return !ord_o ? -1 : 1;
432: else if ( t1 < t2 )
433: return !ord_o ? 1 : -1;
434: }
435: }
436: return 0;
437: }
438:
1.96 noro 439: int ndl_matrix_compare(UINT *d1,UINT *d2)
440: {
441: int i,j,s;
442: int *v;
443:
444: for ( j = 0; j < nd_nvar; j++ )
445: nd_work_vector[j] = GET_EXP(d1,j)-GET_EXP(d2,j);
446: for ( i = 0; i < nd_matrix_len; i++ ) {
447: v = nd_matrix[i];
448: for ( j = 0, s = 0; j < nd_nvar; j++ )
449: s += v[j]*nd_work_vector[j];
450: if ( s > 0 ) return 1;
451: else if ( s < 0 ) return -1;
452: }
453: return 0;
454: }
455:
1.97 noro 456: int ndl_composite_compare(UINT *d1,UINT *d2)
457: {
458: int i,j,s,start,end,len,o;
459: int *v;
460: struct sparse_weight *sw;
461:
462: for ( j = 0; j < nd_nvar; j++ )
463: nd_work_vector[j] = GET_EXP(d1,j)-GET_EXP(d2,j);
464: for ( i = 0; i < nd_worb_len; i++ ) {
465: len = nd_worb[i].length;
466: switch ( nd_worb[i].type ) {
467: case IS_DENSE_WEIGHT:
468: v = nd_worb[i].body.dense_weight;
469: for ( j = 0, s = 0; j < len; j++ )
470: s += v[j]*nd_work_vector[j];
471: if ( s > 0 ) return 1;
472: else if ( s < 0 ) return -1;
473: break;
474: case IS_SPARSE_WEIGHT:
475: sw = nd_worb[i].body.sparse_weight;
476: for ( j = 0, s = 0; j < len; j++ )
477: s += sw[j].value*nd_work_vector[sw[j].pos];
478: if ( s > 0 ) return 1;
479: else if ( s < 0 ) return -1;
480: break;
481: case IS_BLOCK:
482: o = nd_worb[i].body.block.order;
483: start = nd_worb[i].body.block.start;
484: switch ( o ) {
485: case 0:
486: end = start+len;
487: for ( j = start, s = 0; j < end; j++ )
488: s += MUL_WEIGHT(nd_work_vector[j],j);
489: if ( s > 0 ) return 1;
490: else if ( s < 0 ) return -1;
491: for ( j = end-1; j >= start; j-- )
492: if ( nd_work_vector[j] < 0 ) return 1;
493: else if ( nd_work_vector[j] > 0 ) return -1;
494: break;
495: case 1:
496: end = start+len;
497: for ( j = start, s = 0; j < end; j++ )
498: s += MUL_WEIGHT(nd_work_vector[j],j);
499: if ( s > 0 ) return 1;
500: else if ( s < 0 ) return -1;
501: for ( j = start; j < end; j++ )
502: if ( nd_work_vector[j] > 0 ) return 1;
503: else if ( nd_work_vector[j] < 0 ) return -1;
504: break;
505: case 2:
506: for ( j = start; j < end; j++ )
507: if ( nd_work_vector[j] > 0 ) return 1;
508: else if ( nd_work_vector[j] < 0 ) return -1;
509: break;
510: }
511: break;
512: }
513: }
514: return 0;
515: }
516:
1.58 noro 517: /* TDH -> WW -> TD-> RL */
518:
1.61 noro 519: int ndl_ww_lex_compare(UINT *d1,UINT *d2)
1.58 noro 520: {
521: int i,m,e1,e2;
522:
523: if ( TD(d1) > TD(d2) ) return 1;
524: else if ( TD(d1) < TD(d2) ) return -1;
525: m = nd_nvar>>1;
526: for ( i = 0, e1 = e2 = 0; i < m; i++ ) {
527: e1 += current_weyl_weight_vector[i]*(GET_EXP(d1,m+i)-GET_EXP(d1,i));
528: e2 += current_weyl_weight_vector[i]*(GET_EXP(d2,m+i)-GET_EXP(d2,i));
529: }
530: if ( e1 > e2 ) return 1;
531: else if ( e1 < e2 ) return -1;
532: return ndl_lex_compare(d1,d2);
533: }
534:
1.61 noro 535: INLINE int ndl_equal(UINT *d1,UINT *d2)
1.1 noro 536: {
537: int i;
538:
1.81 noro 539: switch ( nd_wpd ) {
540: case 2:
541: if ( TD(d2) != TD(d1) ) return 0;
542: if ( d2[1] != d1[1] ) return 0;
543: return 1;
544: break;
545: case 3:
546: if ( TD(d2) != TD(d1) ) return 0;
547: if ( d2[1] != d1[1] ) return 0;
548: if ( d2[2] != d1[2] ) return 0;
549: return 1;
550: break;
551: default:
552: for ( i = 0; i < nd_wpd; i++ )
553: if ( *d1++ != *d2++ ) return 0;
554: return 1;
555: break;
556: }
1.1 noro 557: }
558:
1.61 noro 559: INLINE void ndl_copy(UINT *d1,UINT *d2)
1.6 noro 560: {
561: int i;
562:
563: switch ( nd_wpd ) {
1.41 noro 564: case 2:
1.34 noro 565: TD(d2) = TD(d1);
566: d2[1] = d1[1];
1.6 noro 567: break;
1.41 noro 568: case 3:
1.34 noro 569: TD(d2) = TD(d1);
1.6 noro 570: d2[1] = d1[1];
1.34 noro 571: d2[2] = d1[2];
1.6 noro 572: break;
573: default:
1.41 noro 574: for ( i = 0; i < nd_wpd; i++ )
1.6 noro 575: d2[i] = d1[i];
576: break;
577: }
578: }
579:
1.61 noro 580: INLINE void ndl_zero(UINT *d)
581: {
582: int i;
583: for ( i = 0; i < nd_wpd; i++ ) d[i] = 0;
584: }
585:
586: INLINE void ndl_add(UINT *d1,UINT *d2,UINT *d)
1.1 noro 587: {
588: int i;
589:
1.43 noro 590: #if 1
1.6 noro 591: switch ( nd_wpd ) {
1.41 noro 592: case 2:
593: TD(d) = TD(d1)+TD(d2);
1.34 noro 594: d[1] = d1[1]+d2[1];
1.6 noro 595: break;
1.41 noro 596: case 3:
597: TD(d) = TD(d1)+TD(d2);
1.6 noro 598: d[1] = d1[1]+d2[1];
1.34 noro 599: d[2] = d1[2]+d2[2];
1.6 noro 600: break;
601: default:
1.43 noro 602: for ( i = 0; i < nd_wpd; i++ ) d[i] = d1[i]+d2[i];
1.6 noro 603: break;
604: }
1.43 noro 605: #else
606: for ( i = 0; i < nd_wpd; i++ ) d[i] = d1[i]+d2[i];
607: #endif
1.6 noro 608: }
609:
1.55 noro 610: /* d1 += d2 */
1.61 noro 611: INLINE void ndl_addto(UINT *d1,UINT *d2)
1.55 noro 612: {
613: int i;
614:
615: #if 1
616: switch ( nd_wpd ) {
617: case 2:
618: TD(d1) += TD(d2);
619: d1[1] += d2[1];
620: break;
621: case 3:
622: TD(d1) += TD(d2);
623: d1[1] += d2[1];
624: d1[2] += d2[2];
625: break;
626: default:
627: for ( i = 0; i < nd_wpd; i++ ) d1[i] += d2[i];
628: break;
629: }
630: #else
631: for ( i = 0; i < nd_wpd; i++ ) d1[i] += d2[i];
632: #endif
633: }
634:
1.61 noro 635: INLINE void ndl_sub(UINT *d1,UINT *d2,UINT *d)
1.6 noro 636: {
637: int i;
638:
1.43 noro 639: for ( i = 0; i < nd_wpd; i++ ) d[i] = d1[i]-d2[i];
1.1 noro 640: }
641:
1.61 noro 642: int ndl_disjoint(UINT *d1,UINT *d2)
1.1 noro 643: {
1.61 noro 644: UINT t1,t2,u,u1,u2;
1.1 noro 645: int i,j;
646:
1.65 noro 647: #if USE_UNROLL
1.1 noro 648: switch ( nd_bpe ) {
1.62 noro 649: case 3:
650: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
651: u1 = d1[i]; u2 = d2[i];
652: t1 = u1&0x38000000; t2 = u2&0x38000000; if ( t1&&t2 ) return 0;
653: t1 = u1& 0x7000000; t2 = u2& 0x7000000; if ( t1&&t2 ) return 0;
654: t1 = u1& 0xe00000; t2 = u2& 0xe00000; if ( t1&&t2 ) return 0;
655: t1 = u1& 0x1c0000; t2 = u2& 0x1c0000; if ( t1&&t2 ) return 0;
656: t1 = u1& 0x38000; t2 = u2& 0x38000; if ( t1&&t2 ) return 0;
657: t1 = u1& 0x7000; t2 = u2& 0x7000; if ( t1&&t2 ) return 0;
658: t1 = u1& 0xe00; t2 = u2& 0xe00; if ( t1&&t2 ) return 0;
659: t1 = u1& 0x1c0; t2 = u2& 0x1c0; if ( t1&&t2 ) return 0;
660: t1 = u1& 0x38; t2 = u2& 0x38; if ( t1&&t2 ) return 0;
661: t1 = u1& 0x7; t2 = u2& 0x7; if ( t1&&t2 ) return 0;
662: }
663: return 1;
664: break;
1.1 noro 665: case 4:
1.41 noro 666: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 667: u1 = d1[i]; u2 = d2[i];
668: t1 = u1&0xf0000000; t2 = u2&0xf0000000; if ( t1&&t2 ) return 0;
1.62 noro 669: t1 = u1& 0xf000000; t2 = u2& 0xf000000; if ( t1&&t2 ) return 0;
670: t1 = u1& 0xf00000; t2 = u2& 0xf00000; if ( t1&&t2 ) return 0;
671: t1 = u1& 0xf0000; t2 = u2& 0xf0000; if ( t1&&t2 ) return 0;
672: t1 = u1& 0xf000; t2 = u2& 0xf000; if ( t1&&t2 ) return 0;
673: t1 = u1& 0xf00; t2 = u2& 0xf00; if ( t1&&t2 ) return 0;
674: t1 = u1& 0xf0; t2 = u2& 0xf0; if ( t1&&t2 ) return 0;
675: t1 = u1& 0xf; t2 = u2& 0xf; if ( t1&&t2 ) return 0;
1.1 noro 676: }
677: return 1;
678: break;
679: case 6:
1.41 noro 680: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 681: u1 = d1[i]; u2 = d2[i];
682: t1 = u1&0x3f000000; t2 = u2&0x3f000000; if ( t1&&t2 ) return 0;
1.62 noro 683: t1 = u1& 0xfc0000; t2 = u2& 0xfc0000; if ( t1&&t2 ) return 0;
684: t1 = u1& 0x3f000; t2 = u2& 0x3f000; if ( t1&&t2 ) return 0;
685: t1 = u1& 0xfc0; t2 = u2& 0xfc0; if ( t1&&t2 ) return 0;
686: t1 = u1& 0x3f; t2 = u2& 0x3f; if ( t1&&t2 ) return 0;
1.1 noro 687: }
688: return 1;
689: break;
690: case 8:
1.41 noro 691: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 692: u1 = d1[i]; u2 = d2[i];
693: t1 = u1&0xff000000; t2 = u2&0xff000000; if ( t1&&t2 ) return 0;
1.62 noro 694: t1 = u1& 0xff0000; t2 = u2& 0xff0000; if ( t1&&t2 ) return 0;
695: t1 = u1& 0xff00; t2 = u2& 0xff00; if ( t1&&t2 ) return 0;
696: t1 = u1& 0xff; t2 = u2& 0xff; if ( t1&&t2 ) return 0;
1.1 noro 697: }
698: return 1;
699: break;
700: case 16:
1.41 noro 701: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 702: u1 = d1[i]; u2 = d2[i];
703: t1 = u1&0xffff0000; t2 = u2&0xffff0000; if ( t1&&t2 ) return 0;
1.62 noro 704: t1 = u1& 0xffff; t2 = u2& 0xffff; if ( t1&&t2 ) return 0;
1.1 noro 705: }
706: return 1;
707: break;
708: case 32:
1.41 noro 709: for ( i = nd_exporigin; i < nd_wpd; i++ )
1.1 noro 710: if ( d1[i] && d2[i] ) return 0;
711: return 1;
712: break;
713: default:
1.41 noro 714: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 715: u1 = d1[i]; u2 = d2[i];
716: for ( j = 0; j < nd_epw; j++ ) {
717: if ( (u1&nd_mask0) && (u2&nd_mask0) ) return 0;
718: u1 >>= nd_bpe; u2 >>= nd_bpe;
719: }
720: }
721: return 1;
722: break;
723: }
1.65 noro 724: #else
725: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
726: u1 = d1[i]; u2 = d2[i];
727: for ( j = 0; j < nd_epw; j++ ) {
728: if ( (u1&nd_mask0) && (u2&nd_mask0) ) return 0;
729: u1 >>= nd_bpe; u2 >>= nd_bpe;
730: }
731: }
732: return 1;
733: #endif
1.1 noro 734: }
735:
1.114 noro 736: int ndl_check_bound(UINT *d1,UINT *d2)
1.1 noro 737: {
1.61 noro 738: UINT u2;
1.5 noro 739: int i,j,ind,k;
1.1 noro 740:
1.5 noro 741: ind = 0;
1.65 noro 742: #if USE_UNROLL
1.5 noro 743: switch ( nd_bpe ) {
1.62 noro 744: case 3:
745: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
746: u2 = d2[i];
747: if ( d1[ind++]+((u2>>27)&0x7) >= 0x8 ) return 1;
748: if ( d1[ind++]+((u2>>24)&0x7) >= 0x8 ) return 1;
749: if ( d1[ind++]+((u2>>21)&0x7) >= 0x8 ) return 1;
750: if ( d1[ind++]+((u2>>18)&0x7) >= 0x8 ) return 1;
751: if ( d1[ind++]+((u2>>15)&0x7) >= 0x8 ) return 1;
752: if ( d1[ind++]+((u2>>12)&0x7) >= 0x8 ) return 1;
753: if ( d1[ind++]+((u2>>9)&0x7) >= 0x8 ) return 1;
754: if ( d1[ind++]+((u2>>6)&0x7) >= 0x8 ) return 1;
755: if ( d1[ind++]+((u2>>3)&0x7) >= 0x8 ) return 1;
756: if ( d1[ind++]+(u2&0x7) >= 0x8 ) return 1;
757: }
758: return 0;
759: break;
1.5 noro 760: case 4:
1.41 noro 761: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 762: u2 = d2[i];
763: if ( d1[ind++]+((u2>>28)&0xf) >= 0x10 ) return 1;
764: if ( d1[ind++]+((u2>>24)&0xf) >= 0x10 ) return 1;
765: if ( d1[ind++]+((u2>>20)&0xf) >= 0x10 ) return 1;
766: if ( d1[ind++]+((u2>>16)&0xf) >= 0x10 ) return 1;
767: if ( d1[ind++]+((u2>>12)&0xf) >= 0x10 ) return 1;
768: if ( d1[ind++]+((u2>>8)&0xf) >= 0x10 ) return 1;
769: if ( d1[ind++]+((u2>>4)&0xf) >= 0x10 ) return 1;
770: if ( d1[ind++]+(u2&0xf) >= 0x10 ) return 1;
771: }
772: return 0;
773: break;
774: case 6:
1.41 noro 775: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 776: u2 = d2[i];
777: if ( d1[ind++]+((u2>>24)&0x3f) >= 0x40 ) return 1;
778: if ( d1[ind++]+((u2>>18)&0x3f) >= 0x40 ) return 1;
779: if ( d1[ind++]+((u2>>12)&0x3f) >= 0x40 ) return 1;
780: if ( d1[ind++]+((u2>>6)&0x3f) >= 0x40 ) return 1;
781: if ( d1[ind++]+(u2&0x3f) >= 0x40 ) return 1;
782: }
783: return 0;
784: break;
785: case 8:
1.41 noro 786: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 787: u2 = d2[i];
788: if ( d1[ind++]+((u2>>24)&0xff) >= 0x100 ) return 1;
789: if ( d1[ind++]+((u2>>16)&0xff) >= 0x100 ) return 1;
790: if ( d1[ind++]+((u2>>8)&0xff) >= 0x100 ) return 1;
791: if ( d1[ind++]+(u2&0xff) >= 0x100 ) return 1;
792: }
793: return 0;
794: break;
795: case 16:
1.41 noro 796: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 797: u2 = d2[i];
798: if ( d1[ind++]+((u2>>16)&0xffff) > 0x10000 ) return 1;
799: if ( d1[ind++]+(u2&0xffff) > 0x10000 ) return 1;
800: }
801: return 0;
802: break;
803: case 32:
1.41 noro 804: for ( i = nd_exporigin; i < nd_wpd; i++ )
1.5 noro 805: if ( d1[i]+d2[i]<d1[i] ) return 1;
806: return 0;
807: break;
808: default:
1.41 noro 809: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 810: u2 = d2[i];
811: k = (nd_epw-1)*nd_bpe;
812: for ( j = 0; j < nd_epw; j++, k -= nd_bpe )
813: if ( d1[ind++]+((u2>>k)&nd_mask0) > nd_mask0 ) return 1;
814: }
815: return 0;
816: break;
817: }
1.65 noro 818: #else
819: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
820: u2 = d2[i];
821: k = (nd_epw-1)*nd_bpe;
822: for ( j = 0; j < nd_epw; j++, k -= nd_bpe )
823: if ( d1[ind++]+((u2>>k)&nd_mask0) > nd_mask0 ) return 1;
824: }
825: return 0;
826: #endif
1.1 noro 827: }
828:
1.114 noro 829: int ndl_check_bound2(int index,UINT *d2)
830: {
831: return ndl_check_bound(nd_bound[index],d2);
832: }
833:
1.61 noro 834: INLINE int ndl_hash_value(UINT *d)
1.1 noro 835: {
836: int i;
837: int r;
838:
1.34 noro 839: r = 0;
1.41 noro 840: for ( i = 0; i < nd_wpd; i++ )
1.1 noro 841: r = ((r<<16)+d[i])%REDTAB_LEN;
842: return r;
843: }
844:
1.63 noro 845: INLINE int ndl_find_reducer(UINT *dg)
1.1 noro 846: {
1.13 noro 847: RHist r;
1.6 noro 848: int d,k,i;
1.1 noro 849:
1.63 noro 850: d = ndl_hash_value(dg);
1.13 noro 851: for ( r = nd_red[d], k = 0; r; r = NEXT(r), k++ ) {
1.34 noro 852: if ( ndl_equal(dg,DL(r)) ) {
1.1 noro 853: if ( k > 0 ) nd_notfirst++;
854: nd_found++;
1.13 noro 855: return r->index;
1.1 noro 856: }
857: }
1.13 noro 858: if ( Reverse )
859: for ( i = nd_psn-1; i >= 0; i-- ) {
860: r = nd_psh[i];
1.34 noro 861: if ( ndl_reducible(dg,DL(r)) ) {
1.13 noro 862: nd_create++;
1.34 noro 863: nd_append_red(dg,i);
1.13 noro 864: return i;
865: }
866: }
867: else
868: for ( i = 0; i < nd_psn; i++ ) {
869: r = nd_psh[i];
1.34 noro 870: if ( ndl_reducible(dg,DL(r)) ) {
1.13 noro 871: nd_create++;
1.34 noro 872: nd_append_red(dg,i);
1.13 noro 873: return i;
874: }
1.1 noro 875: }
1.6 noro 876: return -1;
1.1 noro 877: }
878:
1.63 noro 879: ND nd_merge(ND p1,ND p2)
880: {
881: int n,c;
882: int t,can,td1,td2;
883: ND r;
884: NM m1,m2,mr0,mr,s;
885:
886: if ( !p1 ) return p2;
887: else if ( !p2 ) return p1;
888: else {
889: can = 0;
890: for ( n = NV(p1), m1 = BDY(p1), m2 = BDY(p2), mr0 = 0; m1 && m2; ) {
891: c = DL_COMPARE(DL(m1),DL(m2));
892: switch ( c ) {
893: case 0:
894: s = m1; m1 = NEXT(m1);
895: can++; NEXTNM2(mr0,mr,s);
896: s = m2; m2 = NEXT(m2); FREENM(s);
897: break;
898: case 1:
899: s = m1; m1 = NEXT(m1); NEXTNM2(mr0,mr,s);
900: break;
901: case -1:
902: s = m2; m2 = NEXT(m2); NEXTNM2(mr0,mr,s);
903: break;
904: }
905: }
906: if ( !mr0 )
907: if ( m1 ) mr0 = m1;
908: else if ( m2 ) mr0 = m2;
909: else return 0;
910: else if ( m1 ) NEXT(mr) = m1;
911: else if ( m2 ) NEXT(mr) = m2;
912: else NEXT(mr) = 0;
913: BDY(p1) = mr0;
914: SG(p1) = MAX(SG(p1),SG(p2));
915: LEN(p1) = LEN(p1)+LEN(p2)-can;
916: FREEND(p2);
917: return p1;
918: }
919: }
920:
1.31 noro 921: ND nd_add(int mod,ND p1,ND p2)
1.1 noro 922: {
923: int n,c;
1.34 noro 924: int t,can,td1,td2;
1.1 noro 925: ND r;
926: NM m1,m2,mr0,mr,s;
927:
1.34 noro 928: if ( !p1 ) return p2;
929: else if ( !p2 ) return p1;
1.71 noro 930: else if ( mod == -1 ) return nd_add_sf(p1,p2);
1.113 noro 931: else if ( !mod ) return nd_add_q(p1,p2);
1.1 noro 932: else {
1.30 noro 933: can = 0;
1.1 noro 934: for ( n = NV(p1), m1 = BDY(p1), m2 = BDY(p2), mr0 = 0; m1 && m2; ) {
1.34 noro 935: c = DL_COMPARE(DL(m1),DL(m2));
1.1 noro 936: switch ( c ) {
937: case 0:
1.19 noro 938: t = ((CM(m1))+(CM(m2))) - mod;
1.34 noro 939: if ( t < 0 ) t += mod;
1.1 noro 940: s = m1; m1 = NEXT(m1);
941: if ( t ) {
1.34 noro 942: can++; NEXTNM2(mr0,mr,s); CM(mr) = (t);
1.17 noro 943: } else {
1.34 noro 944: can += 2; FREENM(s);
1.17 noro 945: }
946: s = m2; m2 = NEXT(m2); FREENM(s);
1.95 noro 947: break;
948: case 1:
949: s = m1; m1 = NEXT(m1); NEXTNM2(mr0,mr,s);
950: break;
951: case -1:
952: s = m2; m2 = NEXT(m2); NEXTNM2(mr0,mr,s);
953: break;
954: }
955: }
956: if ( !mr0 )
957: if ( m1 ) mr0 = m1;
958: else if ( m2 ) mr0 = m2;
959: else return 0;
960: else if ( m1 ) NEXT(mr) = m1;
961: else if ( m2 ) NEXT(mr) = m2;
962: else NEXT(mr) = 0;
963: BDY(p1) = mr0;
964: SG(p1) = MAX(SG(p1),SG(p2));
965: LEN(p1) = LEN(p1)+LEN(p2)-can;
966: FREEND(p2);
967: return p1;
968: }
969: }
970:
971: /* XXX on opteron, the inlined manipulation of destructive additon of
972: * two NM seems to make gcc optimizer get confused, so the part is
973: * done in a function.
974: */
975:
1.113 noro 976: int nm_destructive_add_q(NM *m1,NM *m2,NM *mr0,NM *mr)
1.95 noro 977: {
978: NM s;
1.113 noro 979: Q t;
1.95 noro 980: int can;
981:
1.113 noro 982: addq(CQ(*m1),CQ(*m2),&t);
1.95 noro 983: s = *m1; *m1 = NEXT(*m1);
984: if ( t ) {
1.113 noro 985: can = 1; NEXTNM2(*mr0,*mr,s); CQ(*mr) = (t);
1.95 noro 986: } else {
987: can = 2; FREENM(s);
988: }
989: s = *m2; *m2 = NEXT(*m2); FREENM(s);
990: return can;
991: }
992:
1.113 noro 993: ND nd_add_q(ND p1,ND p2)
1.95 noro 994: {
995: int n,c,can;
996: ND r;
997: NM m1,m2,mr0,mr,s;
1.113 noro 998: Q t;
1.95 noro 999:
1000: if ( !p1 ) return p2;
1001: else if ( !p2 ) return p1;
1002: else {
1003: can = 0;
1004: for ( n = NV(p1), m1 = BDY(p1), m2 = BDY(p2), mr0 = 0; m1 && m2; ) {
1005: c = DL_COMPARE(DL(m1),DL(m2));
1006: switch ( c ) {
1007: case 0:
1008: #if defined(__x86_64__)
1.113 noro 1009: can += nm_destructive_add_q(&m1,&m2,&mr0,&mr);
1.95 noro 1010: #else
1.113 noro 1011: addq(CQ(m1),CQ(m2),&t);
1.95 noro 1012: s = m1; m1 = NEXT(m1);
1013: if ( t ) {
1.113 noro 1014: can++; NEXTNM2(mr0,mr,s); CQ(mr) = (t);
1.95 noro 1015: } else {
1016: can += 2; FREENM(s);
1017: }
1018: s = m2; m2 = NEXT(m2); FREENM(s);
1019: #endif
1.17 noro 1020: break;
1021: case 1:
1022: s = m1; m1 = NEXT(m1); NEXTNM2(mr0,mr,s);
1023: break;
1024: case -1:
1025: s = m2; m2 = NEXT(m2); NEXTNM2(mr0,mr,s);
1026: break;
1027: }
1028: }
1029: if ( !mr0 )
1.34 noro 1030: if ( m1 ) mr0 = m1;
1031: else if ( m2 ) mr0 = m2;
1032: else return 0;
1033: else if ( m1 ) NEXT(mr) = m1;
1034: else if ( m2 ) NEXT(mr) = m2;
1035: else NEXT(mr) = 0;
1.17 noro 1036: BDY(p1) = mr0;
1037: SG(p1) = MAX(SG(p1),SG(p2));
1.31 noro 1038: LEN(p1) = LEN(p1)+LEN(p2)-can;
1.17 noro 1039: FREEND(p2);
1040: return p1;
1041: }
1042: }
1043:
1.71 noro 1044: ND nd_add_sf(ND p1,ND p2)
1045: {
1046: int n,c,can;
1047: ND r;
1048: NM m1,m2,mr0,mr,s;
1049: int t;
1050:
1051: if ( !p1 ) return p2;
1052: else if ( !p2 ) return p1;
1053: else {
1054: can = 0;
1055: for ( n = NV(p1), m1 = BDY(p1), m2 = BDY(p2), mr0 = 0; m1 && m2; ) {
1056: c = DL_COMPARE(DL(m1),DL(m2));
1057: switch ( c ) {
1058: case 0:
1059: t = _addsf(CM(m1),CM(m2));
1060: s = m1; m1 = NEXT(m1);
1061: if ( t ) {
1062: can++; NEXTNM2(mr0,mr,s); CM(mr) = (t);
1063: } else {
1064: can += 2; FREENM(s);
1065: }
1066: s = m2; m2 = NEXT(m2); FREENM(s);
1067: break;
1068: case 1:
1069: s = m1; m1 = NEXT(m1); NEXTNM2(mr0,mr,s);
1070: break;
1071: case -1:
1072: s = m2; m2 = NEXT(m2); NEXTNM2(mr0,mr,s);
1073: break;
1074: }
1075: }
1076: if ( !mr0 )
1077: if ( m1 ) mr0 = m1;
1078: else if ( m2 ) mr0 = m2;
1079: else return 0;
1080: else if ( m1 ) NEXT(mr) = m1;
1081: else if ( m2 ) NEXT(mr) = m2;
1082: else NEXT(mr) = 0;
1083: BDY(p1) = mr0;
1084: SG(p1) = MAX(SG(p1),SG(p2));
1085: LEN(p1) = LEN(p1)+LEN(p2)-can;
1086: FREEND(p2);
1087: return p1;
1088: }
1089: }
1090:
1.1 noro 1091: /* ret=1 : success, ret=0 : overflow */
1.113 noro 1092: int nd_nf(int mod,ND g,NDV *ps,int full,NDC dn,ND *rp)
1.1 noro 1093: {
1.11 noro 1094: ND d;
1.1 noro 1095: NM m,mrd,tail;
1.7 noro 1096: NM mul;
1.10 noro 1097: int n,sugar,psugar,sugar0,stat,index;
1.30 noro 1098: int c,c1,c2,dummy;
1.17 noro 1099: RHist h;
1.11 noro 1100: NDV p,red;
1.113 noro 1101: Q cg,cred,gcd,tq,qq;
1.21 noro 1102: double hmag;
1.1 noro 1103:
1.113 noro 1104: if ( dn ) {
1105: if ( mod )
1106: dn->m = 1;
1107: else
1108: dn->z = ONE;
1109: }
1.1 noro 1110: if ( !g ) {
1111: *rp = 0;
1112: return 1;
1113: }
1.113 noro 1114: if ( !mod ) hmag = ((double)p_mag((P)HCQ(g)))*nd_scale;
1.21 noro 1115:
1.14 noro 1116: sugar0 = sugar = SG(g);
1.1 noro 1117: n = NV(g);
1.61 noro 1118: mul = (NM)ALLOCA(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT));
1.1 noro 1119: for ( d = 0; g; ) {
1.63 noro 1120: index = ndl_find_reducer(HDL(g));
1.6 noro 1121: if ( index >= 0 ) {
1.17 noro 1122: h = nd_psh[index];
1123: ndl_sub(HDL(g),DL(h),DL(mul));
1.14 noro 1124: if ( ndl_check_bound2(index,DL(mul)) ) {
1.6 noro 1125: nd_free(g); nd_free(d);
1126: return 0;
1127: }
1.77 noro 1128: if ( nd_demand )
1129: p = ndv_load(index);
1130: else
1131: p = ps[index];
1.71 noro 1132: if ( mod == -1 )
1133: CM(mul) = _mulsf(_invsf(HCM(p)),_chsgnsf(HCM(g)));
1134: else if ( mod ) {
1.19 noro 1135: c1 = invm(HCM(p),mod); c2 = mod-HCM(g);
1136: DMAR(c1,c2,0,mod,c); CM(mul) = c;
1.16 noro 1137: } else {
1.113 noro 1138: igcd_cofactor(HCQ(g),HCQ(p),&gcd,&cg,&cred);
1139: chsgnq(cg,&CQ(mul));
1140: nd_mul_c_q(d,cred); nd_mul_c_q(g,cred);
1.69 noro 1141: if ( dn ) {
1.113 noro 1142: mulq(dn->z,cred,&tq); dn->z = tq;
1.69 noro 1143: }
1.16 noro 1144: }
1.55 noro 1145: g = nd_add(mod,g,ndv_mul_nm(mod,mul,p));
1.34 noro 1146: sugar = MAX(sugar,SG(p)+TD(DL(mul)));
1.113 noro 1147: if ( !mod && hmag && g && ((double)(p_mag((P)HCQ(g))) > hmag) ) {
1148: tq = HCQ(g);
1.21 noro 1149: nd_removecont2(d,g);
1.69 noro 1150: if ( dn ) {
1.113 noro 1151: divq(tq,HCQ(g),&qq); divq(dn->z,qq,&tq); dn->z = tq;
1.69 noro 1152: }
1.113 noro 1153: hmag = ((double)p_mag((P)HCQ(g)))*nd_scale;
1.21 noro 1154: }
1.1 noro 1155: } else if ( !full ) {
1156: *rp = g;
1157: return 1;
1158: } else {
1159: m = BDY(g);
1160: if ( NEXT(m) ) {
1.34 noro 1161: BDY(g) = NEXT(m); NEXT(m) = 0; LEN(g)--;
1.1 noro 1162: } else {
1163: FREEND(g); g = 0;
1164: }
1165: if ( d ) {
1.34 noro 1166: NEXT(tail)=m; tail=m; LEN(d)++;
1.1 noro 1167: } else {
1.34 noro 1168: MKND(n,m,1,d); tail = BDY(d);
1.1 noro 1169: }
1170: }
1171: }
1.34 noro 1172: if ( d ) SG(d) = sugar;
1.1 noro 1173: *rp = d;
1174: return 1;
1175: }
1.28 noro 1176:
1.53 noro 1177: int nd_nf_pbucket(int mod,ND g,NDV *ps,int full,ND *rp)
1.25 noro 1178: {
1179: int hindex,index;
1180: NDV p;
1181: ND u,d,red;
1182: NODE l;
1.31 noro 1183: NM mul,m,mrd,tail;
1.25 noro 1184: int sugar,psugar,n,h_reducible;
1185: PGeoBucket bucket;
1186: int c,c1,c2;
1.113 noro 1187: Q cg,cred,gcd,zzz;
1.25 noro 1188: RHist h;
1.28 noro 1189: double hmag,gmag;
1.77 noro 1190: int count = 0;
1191: int hcount = 0;
1.25 noro 1192:
1193: if ( !g ) {
1194: *rp = 0;
1195: return 1;
1196: }
1197: sugar = SG(g);
1198: n = NV(g);
1.113 noro 1199: if ( !mod ) hmag = ((double)p_mag((P)HCQ(g)))*nd_scale;
1.25 noro 1200: bucket = create_pbucket();
1.31 noro 1201: add_pbucket(mod,bucket,g);
1.25 noro 1202: d = 0;
1.61 noro 1203: mul = (NM)ALLOCA(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT));
1.25 noro 1204: while ( 1 ) {
1.113 noro 1205: hindex = mod?head_pbucket(mod,bucket):head_pbucket_q(bucket);
1.25 noro 1206: if ( hindex < 0 ) {
1.77 noro 1207: if ( DP_Print > 3 ) printf("(%d %d)",count,hcount);
1.34 noro 1208: if ( d ) SG(d) = sugar;
1.25 noro 1209: *rp = d;
1210: return 1;
1211: }
1212: g = bucket->body[hindex];
1.63 noro 1213: index = ndl_find_reducer(HDL(g));
1.25 noro 1214: if ( index >= 0 ) {
1.77 noro 1215: count++;
1216: if ( !d ) hcount++;
1.25 noro 1217: h = nd_psh[index];
1218: ndl_sub(HDL(g),DL(h),DL(mul));
1219: if ( ndl_check_bound2(index,DL(mul)) ) {
1.26 noro 1220: nd_free(d);
1.25 noro 1221: free_pbucket(bucket);
1222: *rp = 0;
1223: return 0;
1224: }
1.53 noro 1225: p = ps[index];
1.71 noro 1226: if ( mod == -1 )
1227: CM(mul) = _mulsf(_invsf(HCM(p)),_chsgnsf(HCM(g)));
1228: else if ( mod ) {
1.25 noro 1229: c1 = invm(HCM(p),mod); c2 = mod-HCM(g);
1230: DMAR(c1,c2,0,mod,c); CM(mul) = c;
1231: } else {
1.113 noro 1232: igcd_cofactor(HCQ(g),HCQ(p),&gcd,&cg,&cred);
1233: chsgnq(cg,&CQ(mul));
1234: nd_mul_c_q(d,cred);
1235: mulq_pbucket(bucket,cred);
1.26 noro 1236: g = bucket->body[hindex];
1.113 noro 1237: gmag = (double)p_mag((P)HCQ(g));
1.25 noro 1238: }
1.55 noro 1239: red = ndv_mul_nm(mod,mul,p);
1.25 noro 1240: bucket->body[hindex] = nd_remove_head(g);
1241: red = nd_remove_head(red);
1.31 noro 1242: add_pbucket(mod,bucket,red);
1.34 noro 1243: psugar = SG(p)+TD(DL(mul));
1244: sugar = MAX(sugar,psugar);
1.28 noro 1245: if ( !mod && hmag && (gmag > hmag) ) {
1246: g = normalize_pbucket(mod,bucket);
1247: if ( !g ) {
1.34 noro 1248: if ( d ) SG(d) = sugar;
1.28 noro 1249: *rp = d;
1250: return 1;
1251: }
1252: nd_removecont2(d,g);
1.113 noro 1253: hmag = ((double)p_mag((P)HCQ(g)))*nd_scale;
1.31 noro 1254: add_pbucket(mod,bucket,g);
1.28 noro 1255: }
1.25 noro 1256: } else if ( !full ) {
1257: g = normalize_pbucket(mod,bucket);
1.34 noro 1258: if ( g ) SG(g) = sugar;
1.25 noro 1259: *rp = g;
1260: return 1;
1261: } else {
1262: m = BDY(g);
1263: if ( NEXT(m) ) {
1.34 noro 1264: BDY(g) = NEXT(m); NEXT(m) = 0; LEN(g)--;
1.25 noro 1265: } else {
1266: FREEND(g); g = 0;
1267: }
1268: bucket->body[hindex] = g;
1269: NEXT(m) = 0;
1270: if ( d ) {
1.34 noro 1271: NEXT(tail)=m; tail=m; LEN(d)++;
1.25 noro 1272: } else {
1.34 noro 1273: MKND(n,m,1,d); tail = BDY(d);
1.25 noro 1274: }
1275: }
1276: }
1277: }
1.27 noro 1278:
1.61 noro 1279: /* input : list of NDV, cand : list of NDV */
1.28 noro 1280:
1.61 noro 1281: int ndv_check_candidate(NODE input,int obpe,int oadv,EPOS oepos,NODE cand)
1.28 noro 1282: {
1283: int n,i,stat;
1284: ND nf,d;
1.61 noro 1285: NDV r;
1.45 noro 1286: NODE t,s;
1.113 noro 1287: union oNDC dn;
1.45 noro 1288:
1.128 noro 1289: ndv_setup(0,0,cand,0);
1.31 noro 1290: n = length(cand);
1.28 noro 1291:
1292: /* membercheck : list is a subset of Id(cand) ? */
1293: for ( t = input; t; t = NEXT(t) ) {
1.45 noro 1294: again:
1.61 noro 1295: if ( nd_bpe > obpe )
1296: r = ndv_dup_realloc((NDV)BDY(t),obpe,oadv,oepos);
1297: else
1298: r = (NDV)BDY(t);
1299: d = ndvtond(0,r);
1.69 noro 1300: stat = nd_nf(0,d,nd_ps,0,0,&nf);
1.45 noro 1301: if ( !stat ) {
1.103 noro 1302: nd_reconstruct(0,0);
1.45 noro 1303: goto again;
1304: } else if ( nf ) return 0;
1.71 noro 1305: if ( DP_Print ) { printf("."); fflush(stdout); }
1.28 noro 1306: }
1.71 noro 1307: if ( DP_Print ) { printf("\n"); }
1.28 noro 1308: /* gbcheck : cand is a GB of Id(cand) ? */
1.77 noro 1309: if ( !nd_gb(0,0,1) ) return 0;
1.28 noro 1310: /* XXX */
1.23 noro 1311: return 1;
1312: }
1.1 noro 1313:
1314: ND nd_remove_head(ND p)
1315: {
1316: NM m;
1317:
1318: m = BDY(p);
1319: if ( !NEXT(m) ) {
1.34 noro 1320: FREEND(p); p = 0;
1.31 noro 1321: } else {
1.34 noro 1322: BDY(p) = NEXT(m); LEN(p)--;
1.31 noro 1323: }
1.1 noro 1324: FREENM(m);
1325: return p;
1326: }
1327:
1.69 noro 1328: ND nd_separate_head(ND p,ND *head)
1329: {
1330: NM m,m0;
1331: ND r;
1332:
1333: m = BDY(p);
1334: if ( !NEXT(m) ) {
1335: *head = p; p = 0;
1336: } else {
1337: m0 = m;
1338: BDY(p) = NEXT(m); LEN(p)--;
1339: NEXT(m0) = 0;
1340: MKND(NV(p),m0,1,r);
1341: *head = r;
1342: }
1343: return p;
1344: }
1345:
1.1 noro 1346: PGeoBucket create_pbucket()
1347: {
1348: PGeoBucket g;
1349:
1350: g = CALLOC(1,sizeof(struct oPGeoBucket));
1351: g->m = -1;
1352: return g;
1353: }
1354:
1.25 noro 1355: void free_pbucket(PGeoBucket b) {
1356: int i;
1357:
1.26 noro 1358: for ( i = 0; i <= b->m; i++ )
1.25 noro 1359: if ( b->body[i] ) {
1360: nd_free(b->body[i]);
1361: b->body[i] = 0;
1362: }
1363: GC_free(b);
1364: }
1365:
1.63 noro 1366: void add_pbucket_symbolic(PGeoBucket g,ND d)
1367: {
1368: int l,i,k,m;
1369:
1370: if ( !d )
1371: return;
1372: l = LEN(d);
1373: for ( k = 0, m = 1; l > m; k++, m <<= 1 );
1374: /* 2^(k-1) < l <= 2^k (=m) */
1375: d = nd_merge(g->body[k],d);
1376: for ( ; d && LEN(d) > m; k++, m <<= 1 ) {
1377: g->body[k] = 0;
1378: d = nd_merge(g->body[k+1],d);
1379: }
1380: g->body[k] = d;
1381: g->m = MAX(g->m,k);
1382: }
1383:
1.31 noro 1384: void add_pbucket(int mod,PGeoBucket g,ND d)
1.1 noro 1385: {
1.31 noro 1386: int l,i,k,m;
1.1 noro 1387:
1.31 noro 1388: if ( !d )
1389: return;
1390: l = LEN(d);
1.29 noro 1391: for ( k = 0, m = 1; l > m; k++, m <<= 1 );
1392: /* 2^(k-1) < l <= 2^k (=m) */
1.31 noro 1393: d = nd_add(mod,g->body[k],d);
1394: for ( ; d && LEN(d) > m; k++, m <<= 1 ) {
1.1 noro 1395: g->body[k] = 0;
1.31 noro 1396: d = nd_add(mod,g->body[k+1],d);
1.1 noro 1397: }
1398: g->body[k] = d;
1399: g->m = MAX(g->m,k);
1400: }
1401:
1.113 noro 1402: void mulq_pbucket(PGeoBucket g,Q c)
1.26 noro 1403: {
1404: int k;
1405:
1406: for ( k = 0; k <= g->m; k++ )
1.113 noro 1407: nd_mul_c_q(g->body[k],c);
1.26 noro 1408: }
1409:
1.63 noro 1410: NM remove_head_pbucket_symbolic(PGeoBucket g)
1411: {
1412: int j,i,k,c;
1413: NM head;
1414:
1415: k = g->m;
1416: j = -1;
1417: for ( i = 0; i <= k; i++ ) {
1418: if ( !g->body[i] ) continue;
1419: if ( j < 0 ) j = i;
1420: else {
1421: c = DL_COMPARE(HDL(g->body[i]),HDL(g->body[j]));
1422: if ( c > 0 )
1423: j = i;
1424: else if ( c == 0 )
1425: g->body[i] = nd_remove_head(g->body[i]);
1426: }
1427: }
1428: if ( j < 0 ) return 0;
1429: else {
1430: head = BDY(g->body[j]);
1431: if ( !NEXT(head) ) {
1432: FREEND(g->body[j]);
1433: g->body[j] = 0;
1434: } else {
1435: BDY(g->body[j]) = NEXT(head);
1436: LEN(g->body[j])--;
1437: }
1438: return head;
1439: }
1440: }
1441:
1.19 noro 1442: int head_pbucket(int mod,PGeoBucket g)
1.1 noro 1443: {
1444: int j,i,c,k,nv,sum;
1.61 noro 1445: UINT *di,*dj;
1.1 noro 1446: ND gi,gj;
1447:
1448: k = g->m;
1449: while ( 1 ) {
1450: j = -1;
1451: for ( i = 0; i <= k; i++ ) {
1452: if ( !(gi = g->body[i]) )
1453: continue;
1454: if ( j < 0 ) {
1455: j = i;
1456: gj = g->body[j];
1457: dj = HDL(gj);
1.14 noro 1458: sum = HCM(gj);
1.1 noro 1459: } else {
1.34 noro 1460: c = DL_COMPARE(HDL(gi),dj);
1.1 noro 1461: if ( c > 0 ) {
1.34 noro 1462: if ( sum ) HCM(gj) = sum;
1463: else g->body[j] = nd_remove_head(gj);
1.1 noro 1464: j = i;
1465: gj = g->body[j];
1466: dj = HDL(gj);
1.14 noro 1467: sum = HCM(gj);
1.1 noro 1468: } else if ( c == 0 ) {
1.71 noro 1469: if ( mod == -1 )
1470: sum = _addsf(sum,HCM(gi));
1471: else {
1472: sum = sum+HCM(gi)-mod;
1473: if ( sum < 0 ) sum += mod;
1474: }
1.1 noro 1475: g->body[i] = nd_remove_head(gi);
1476: }
1477: }
1478: }
1.34 noro 1479: if ( j < 0 ) return -1;
1.1 noro 1480: else if ( sum ) {
1.14 noro 1481: HCM(gj) = sum;
1.26 noro 1482: return j;
1.31 noro 1483: } else
1.26 noro 1484: g->body[j] = nd_remove_head(gj);
1485: }
1486: }
1487:
1.113 noro 1488: int head_pbucket_q(PGeoBucket g)
1.26 noro 1489: {
1490: int j,i,c,k,nv;
1.113 noro 1491: Q sum,t;
1.26 noro 1492: ND gi,gj;
1493:
1494: k = g->m;
1495: while ( 1 ) {
1496: j = -1;
1497: for ( i = 0; i <= k; i++ ) {
1.34 noro 1498: if ( !(gi = g->body[i]) ) continue;
1.26 noro 1499: if ( j < 0 ) {
1500: j = i;
1501: gj = g->body[j];
1.113 noro 1502: sum = HCQ(gj);
1.26 noro 1503: } else {
1504: nv = NV(gi);
1.34 noro 1505: c = DL_COMPARE(HDL(gi),HDL(gj));
1.26 noro 1506: if ( c > 0 ) {
1.113 noro 1507: if ( sum ) HCQ(gj) = sum;
1.34 noro 1508: else g->body[j] = nd_remove_head(gj);
1.26 noro 1509: j = i;
1510: gj = g->body[j];
1.113 noro 1511: sum = HCQ(gj);
1.26 noro 1512: } else if ( c == 0 ) {
1.113 noro 1513: addq(sum,HCQ(gi),&t);
1514: sum = t;
1.26 noro 1515: g->body[i] = nd_remove_head(gi);
1516: }
1517: }
1518: }
1.34 noro 1519: if ( j < 0 ) return -1;
1.26 noro 1520: else if ( sum ) {
1.113 noro 1521: HCQ(gj) = sum;
1.1 noro 1522: return j;
1.31 noro 1523: } else
1.1 noro 1524: g->body[j] = nd_remove_head(gj);
1525: }
1526: }
1527:
1.25 noro 1528: ND normalize_pbucket(int mod,PGeoBucket g)
1.1 noro 1529: {
1.31 noro 1530: int i;
1.1 noro 1531: ND r,t;
1532:
1533: r = 0;
1.28 noro 1534: for ( i = 0; i <= g->m; i++ ) {
1.31 noro 1535: r = nd_add(mod,r,g->body[i]);
1.28 noro 1536: g->body[i] = 0;
1537: }
1538: g->m = -1;
1.1 noro 1539: return r;
1540: }
1541:
1.122 noro 1542: int do_diagonalize(int sugar,int m)
1.92 noro 1543: {
1544: int i,nh,stat;
1545: NODE r,g,t;
1546: ND h,nf,s,head;
1547: NDV nfv;
1.113 noro 1548: Q q,num,den;
1549: union oNDC dn;
1.92 noro 1550:
1551: for ( i = nd_psn-1; i >= 0 && SG(nd_psh[i]) == sugar; i-- ) {
1552: if ( nd_demand )
1553: nfv = ndv_load(i);
1554: else
1555: nfv = nd_ps[i];
1556: s = ndvtond(m,nfv);
1557: s = nd_separate_head(s,&head);
1.122 noro 1558: stat = nd_nf(m,s,nd_ps,1,&dn,&nf);
1559: if ( !stat ) return 0;
1.92 noro 1560: if ( !m ) {
1.113 noro 1561: NTOQ(NM(dn.z),SGN(dn.z),num);
1562: mulq(HCQ(head),num,&q); HCQ(head) = q;
1563: if ( DN(dn.z) ) {
1564: NTOQ(DN(dn.z),1,den);
1565: nd_mul_c_q(nf,den);
1566: }
1.92 noro 1567: }
1568: nf = nd_add(m,head,nf);
1569: ndv_free(nfv);
1570: nd_removecont(m,nf);
1571: nfv = ndtondv(m,nf);
1572: nd_free(nf);
1573: nd_bound[i] = ndv_compute_bound(nfv);
1574: if ( nd_demand ) {
1575: ndv_save(nfv,i);
1576: ndv_free(nfv);
1577: } else
1578: nd_ps[i] = nfv;
1579: }
1.122 noro 1580: return 1;
1.92 noro 1581: }
1582:
1.27 noro 1583: /* return value = 0 => input is not a GB */
1584:
1.77 noro 1585: NODE nd_gb(int m,int ishomo,int checkonly)
1.1 noro 1586: {
1587: int i,nh,sugar,stat;
1.23 noro 1588: NODE r,g,t;
1.1 noro 1589: ND_pairs d;
1590: ND_pairs l;
1.117 noro 1591: ND h,nf,s,head,nf1;
1.63 noro 1592: NDV nfv;
1.113 noro 1593: Q q,num,den;
1594: union oNDC dn;
1.1 noro 1595:
1.23 noro 1596: g = 0; d = 0;
1597: for ( i = 0; i < nd_psn; i++ ) {
1.1 noro 1598: d = update_pairs(d,g,i);
1599: g = update_base(g,i);
1600: }
1601: sugar = 0;
1602: while ( d ) {
1603: again:
1604: l = nd_minp(d,&d);
1.14 noro 1605: if ( SG(l) != sugar ) {
1.122 noro 1606: if ( ishomo ) {
1607: stat = do_diagonalize(sugar,m);
1608: if ( !stat ) {
1609: NEXT(l) = d; d = l;
1610: d = nd_reconstruct(0,d);
1611: goto again;
1612: }
1613: }
1.14 noro 1614: sugar = SG(l);
1.71 noro 1615: if ( DP_Print ) fprintf(asir_out,"%d",sugar);
1.1 noro 1616: }
1.53 noro 1617: stat = nd_sp(m,0,l,&h);
1.1 noro 1618: if ( !stat ) {
1619: NEXT(l) = d; d = l;
1.103 noro 1620: d = nd_reconstruct(0,d);
1.1 noro 1621: goto again;
1622: }
1.41 noro 1623: #if USE_GEOBUCKET
1.69 noro 1624: stat = m?nd_nf_pbucket(m,h,nd_ps,!Top,&nf):nd_nf(m,h,nd_ps,!Top,0,&nf);
1.41 noro 1625: #else
1.69 noro 1626: stat = nd_nf(m,h,nd_ps,!Top,0,&nf);
1.41 noro 1627: #endif
1.1 noro 1628: if ( !stat ) {
1629: NEXT(l) = d; d = l;
1.103 noro 1630: d = nd_reconstruct(0,d);
1.1 noro 1631: goto again;
1632: } else if ( nf ) {
1.27 noro 1633: if ( checkonly ) return 0;
1.71 noro 1634: if ( DP_Print ) { printf("+"); fflush(stdout); }
1.63 noro 1635: nd_removecont(m,nf);
1.118 noro 1636: if ( !m && nd_nalg ) {
1637: nd_monic(0,&nf);
1638: nd_removecont(m,nf);
1.117 noro 1639: }
1.63 noro 1640: nfv = ndtondv(m,nf); nd_free(nf);
1.77 noro 1641: nh = ndv_newps(m,nfv,0);
1.1 noro 1642: d = update_pairs(d,g,nh);
1643: g = update_base(g,nh);
1644: FREENDP(l);
1645: } else {
1.71 noro 1646: if ( DP_Print ) { printf("."); fflush(stdout); }
1.1 noro 1647: FREENDP(l);
1648: }
1649: }
1.77 noro 1650: if ( nd_demand )
1651: for ( t = g; t; t = NEXT(t) )
1652: BDY(t) = (pointer)ndv_load((int)BDY(t));
1653: else
1654: for ( t = g; t; t = NEXT(t) )
1655: BDY(t) = (pointer)nd_ps[(int)BDY(t)];
1.1 noro 1656: return g;
1657: }
1658:
1.122 noro 1659: int do_diagonalize_trace(int sugar,int m)
1.91 noro 1660: {
1661: int i,nh,stat;
1662: NODE r,g,t;
1663: ND h,nf,nfq,s,head;
1664: NDV nfv,nfqv;
1.113 noro 1665: Q q,den,num;
1666: union oNDC dn;
1.91 noro 1667:
1668: for ( i = nd_psn-1; i >= 0 && SG(nd_psh[i]) == sugar; i-- ) {
1669: /* for nd_ps */
1670: s = ndvtond(m,nd_ps[i]);
1671: s = nd_separate_head(s,&head);
1.122 noro 1672: stat = nd_nf_pbucket(m,s,nd_ps,1,&nf);
1673: if ( !stat ) return 0;
1.91 noro 1674: nf = nd_add(m,head,nf);
1675: ndv_free(nd_ps[i]);
1676: nd_ps[i] = ndtondv(m,nf);
1677: nd_free(nf);
1678:
1679: /* for nd_ps_trace */
1680: if ( nd_demand )
1681: nfv = ndv_load(i);
1682: else
1683: nfv = nd_ps_trace[i];
1684: s = ndvtond(0,nfv);
1685: s = nd_separate_head(s,&head);
1.122 noro 1686: stat = nd_nf(0,s,nd_ps_trace,1,&dn,&nf);
1687: if ( !stat ) return 0;
1.113 noro 1688: NTOQ(NM(dn.z),SGN(dn.z),num);
1689: mulq(HCQ(head),num,&q); HCQ(head) = q;
1690: if ( DN(dn.z) ) {
1691: NTOQ(DN(dn.z),1,den);
1692: nd_mul_c_q(nf,den);
1693: }
1.91 noro 1694: nf = nd_add(0,head,nf);
1695: ndv_free(nfv);
1696: nd_removecont(0,nf);
1697: nfv = ndtondv(0,nf);
1698: nd_free(nf);
1699: nd_bound[i] = ndv_compute_bound(nfv);
1700: if ( nd_demand ) {
1701: ndv_save(nfv,i);
1702: ndv_free(nfv);
1703: } else
1704: nd_ps_trace[i] = nfv;
1705: }
1.122 noro 1706: return 1;
1.91 noro 1707: }
1708:
1.118 noro 1709: static struct oEGT eg_invdalg;
1710: struct oEGT eg_le;
1711:
1.82 noro 1712: NODE nd_gb_trace(int m,int ishomo)
1.20 noro 1713: {
1714: int i,nh,sugar,stat;
1.23 noro 1715: NODE r,g,t;
1.20 noro 1716: ND_pairs d;
1717: ND_pairs l;
1.82 noro 1718: ND h,nf,nfq,s,head;
1.63 noro 1719: NDV nfv,nfqv;
1.113 noro 1720: Q q,den,num;
1721: union oNDC dn;
1.118 noro 1722: struct oEGT eg_monic,egm0,egm1;
1.130 noro 1723: int diag_count = 0;
1.20 noro 1724:
1.118 noro 1725: init_eg(&eg_monic);
1726: init_eg(&eg_invdalg);
1727: init_eg(&eg_le);
1.23 noro 1728: g = 0; d = 0;
1729: for ( i = 0; i < nd_psn; i++ ) {
1.20 noro 1730: d = update_pairs(d,g,i);
1731: g = update_base(g,i);
1732: }
1733: sugar = 0;
1734: while ( d ) {
1735: again:
1736: l = nd_minp(d,&d);
1737: if ( SG(l) != sugar ) {
1.130 noro 1738: #if 1
1.122 noro 1739: if ( ishomo ) {
1740: stat = do_diagonalize_trace(sugar,m);
1.130 noro 1741: diag_count = 0;
1.122 noro 1742: if ( !stat ) {
1743: NEXT(l) = d; d = l;
1744: d = nd_reconstruct(1,d);
1745: goto again;
1746: }
1747: }
1.130 noro 1748: #endif
1.20 noro 1749: sugar = SG(l);
1.71 noro 1750: if ( DP_Print ) fprintf(asir_out,"%d",sugar);
1.20 noro 1751: }
1.53 noro 1752: stat = nd_sp(m,0,l,&h);
1.20 noro 1753: if ( !stat ) {
1754: NEXT(l) = d; d = l;
1.103 noro 1755: d = nd_reconstruct(1,d);
1.20 noro 1756: goto again;
1757: }
1.41 noro 1758: #if USE_GEOBUCKET
1.53 noro 1759: stat = nd_nf_pbucket(m,h,nd_ps,!Top,&nf);
1.41 noro 1760: #else
1.69 noro 1761: stat = nd_nf(m,h,nd_ps,!Top,0,&nf);
1.41 noro 1762: #endif
1.20 noro 1763: if ( !stat ) {
1764: NEXT(l) = d; d = l;
1.103 noro 1765: d = nd_reconstruct(1,d);
1.20 noro 1766: goto again;
1767: } else if ( nf ) {
1.77 noro 1768: if ( nd_demand ) {
1769: nfqv = ndv_load(nd_psn);
1770: nfq = ndvtond(0,nfqv);
1771: } else
1772: nfq = 0;
1773: if ( !nfq ) {
1774: if ( !nd_sp(0,1,l,&h) || !nd_nf(0,h,nd_ps_trace,!Top,0,&nfq) ) {
1775: NEXT(l) = d; d = l;
1.103 noro 1776: d = nd_reconstruct(1,d);
1.77 noro 1777: goto again;
1778: }
1.70 noro 1779: }
1.20 noro 1780: if ( nfq ) {
1.70 noro 1781: /* m|HC(nfq) => failure */
1.113 noro 1782: if ( !rem(NM(HCQ(nfq)),m) ) return 0;
1.63 noro 1783:
1.71 noro 1784: if ( DP_Print ) { printf("+"); fflush(stdout); }
1.118 noro 1785: if ( nd_nalg ) {
1786: /* m|DN(HC(nf)^(-1)) => failure */
1787: get_eg(&egm0);
1788: if ( !nd_monic(m,&nfq) ) return 0;
1789: get_eg(&egm1); add_eg(&eg_monic,&egm0,&egm1);
1790: nd_removecont(0,nfq); nfqv = ndtondv(0,nfq); nd_free(nfq);
1791: nfv = ndv_dup(0,nfqv); ndv_mod(m,nfv); nd_free(nf);
1792: } else {
1793: nd_removecont(0,nfq); nfqv = ndtondv(0,nfq); nd_free(nfq);
1794: nd_removecont(m,nf); nfv = ndtondv(m,nf); nd_free(nf);
1795: }
1.77 noro 1796: nh = ndv_newps(0,nfv,nfqv);
1.130 noro 1797: if ( ishomo && ++diag_count == diag_period ) {
1798: diag_count = 0;
1799: stat = do_diagonalize_trace(sugar,m);
1800: if ( !stat ) {
1801: NEXT(l) = d; d = l;
1802: d = nd_reconstruct(1,d);
1803: goto again;
1804: }
1805: }
1.20 noro 1806: d = update_pairs(d,g,nh);
1807: g = update_base(g,nh);
1808: } else {
1.71 noro 1809: if ( DP_Print ) { printf("*"); fflush(stdout); }
1.20 noro 1810: }
1811: } else {
1.71 noro 1812: if ( DP_Print ) { printf("."); fflush(stdout); }
1.20 noro 1813: }
1814: FREENDP(l);
1815: }
1.77 noro 1816: if ( nd_demand )
1817: for ( t = g; t; t = NEXT(t) )
1818: BDY(t) = (pointer)ndv_load((int)BDY(t));
1819: else
1820: for ( t = g; t; t = NEXT(t) )
1821: BDY(t) = (pointer)nd_ps_trace[(int)BDY(t)];
1.118 noro 1822: if ( nd_nalg ) {
1823: print_eg("monic",&eg_monic);
1824: print_eg("invdalg",&eg_invdalg);
1825: print_eg("le",&eg_le);
1826: }
1.20 noro 1827: return g;
1828: }
1829:
1.23 noro 1830: int ndv_compare(NDV *p1,NDV *p2)
1831: {
1.34 noro 1832: return DL_COMPARE(HDL(*p1),HDL(*p2));
1.23 noro 1833: }
1834:
1835: int ndv_compare_rev(NDV *p1,NDV *p2)
1836: {
1.34 noro 1837: return -DL_COMPARE(HDL(*p1),HDL(*p2));
1.23 noro 1838: }
1839:
1.61 noro 1840: NODE ndv_reduceall(int m,NODE f)
1.23 noro 1841: {
1.69 noro 1842: int i,n,stat;
1843: ND nf,g,head;
1.23 noro 1844: NODE t,a0,a;
1.113 noro 1845: union oNDC dn;
1.69 noro 1846: NDV *w;
1.113 noro 1847: Q q,num,den;
1.23 noro 1848:
1.69 noro 1849: n = length(f);
1850: #if 0
1851: w = (NDV *)ALLOCA(n*sizeof(NDV));
1852: for ( i = 0, t = f; i < n; i++, t = NEXT(t) ) w[i] = (NDV)BDY(t);
1853: qsort(w,n,sizeof(NDV),
1854: (int (*)(const void *,const void *))ndv_compare);
1855: for ( t = f, i = 0; t; i++, t = NEXT(t) ) BDY(t) = (pointer)w[i];
1856: #endif
1.128 noro 1857: ndv_setup(m,0,f,0);
1.69 noro 1858: for ( i = 0; i < n; ) {
1859: g = ndvtond(m,nd_ps[i]);
1860: g = nd_separate_head(g,&head);
1861: stat = nd_nf(m,g,nd_ps,1,&dn,&nf);
1.50 noro 1862: if ( !stat )
1.103 noro 1863: nd_reconstruct(0,0);
1.69 noro 1864: else {
1.71 noro 1865: if ( DP_Print ) { printf("."); fflush(stdout); }
1.85 noro 1866: if ( !m ) {
1.113 noro 1867: NTOQ(NM(dn.z),SGN(dn.z),num);
1868: mulq(HCQ(head),num,&q); HCQ(head) = q;
1869: if ( DN(dn.z) ) {
1870: NTOQ(DN(dn.z),1,den);
1871: nd_mul_c_q(nf,den);
1872: }
1.85 noro 1873: }
1.69 noro 1874: nf = nd_add(m,head,nf);
1875: ndv_free(nd_ps[i]);
1.24 noro 1876: nd_removecont(m,nf);
1.69 noro 1877: nd_ps[i] = ndtondv(m,nf); nd_free(nf);
1878: nd_bound[i] = ndv_compute_bound(nd_ps[i]);
1.50 noro 1879: i++;
1.23 noro 1880: }
1881: }
1.71 noro 1882: if ( DP_Print ) { printf("\n"); }
1.23 noro 1883: for ( a0 = 0, i = 0; i < n; i++ ) {
1884: NEXTNODE(a0,a);
1.69 noro 1885: BDY(a) = (pointer)nd_ps[i];
1.23 noro 1886: }
1887: NEXT(a) = 0;
1888: return a0;
1889: }
1890:
1.1 noro 1891: ND_pairs update_pairs( ND_pairs d, NODE /* of index */ g, int t)
1892: {
1893: ND_pairs d1,nd,cur,head,prev,remove;
1894:
1895: if ( !g ) return d;
1896: d = crit_B(d,t);
1897: d1 = nd_newpairs(g,t);
1898: d1 = crit_M(d1);
1899: d1 = crit_F(d1);
1.55 noro 1900: if ( do_weyl )
1901: head = d1;
1902: else {
1903: prev = 0; cur = head = d1;
1904: while ( cur ) {
1905: if ( crit_2( cur->i1,cur->i2 ) ) {
1906: remove = cur;
1907: if ( !prev ) head = cur = NEXT(cur);
1908: else cur = NEXT(prev) = NEXT(cur);
1909: FREENDP(remove);
1910: } else {
1911: prev = cur; cur = NEXT(cur);
1912: }
1.1 noro 1913: }
1914: }
1915: if ( !d )
1916: return head;
1917: else {
1918: nd = d;
1.34 noro 1919: while ( NEXT(nd) ) nd = NEXT(nd);
1.1 noro 1920: NEXT(nd) = head;
1921: return d;
1922: }
1923: }
1924:
1925: ND_pairs nd_newpairs( NODE g, int t )
1926: {
1927: NODE h;
1.61 noro 1928: UINT *dl;
1.34 noro 1929: int ts,s;
1.1 noro 1930: ND_pairs r,r0;
1931:
1.20 noro 1932: dl = DL(nd_psh[t]);
1.34 noro 1933: ts = SG(nd_psh[t]) - TD(dl);
1.1 noro 1934: for ( r0 = 0, h = g; h; h = NEXT(h) ) {
1935: NEXTND_pairs(r0,r);
1936: r->i1 = (int)BDY(h);
1937: r->i2 = t;
1.20 noro 1938: ndl_lcm(DL(nd_psh[r->i1]),dl,r->lcm);
1.34 noro 1939: s = SG(nd_psh[r->i1])-TD(DL(nd_psh[r->i1]));
1940: SG(r) = MAX(s,ts) + TD(LCM(r));
1.1 noro 1941: }
1942: NEXT(r) = 0;
1943: return r0;
1944: }
1945:
1946: ND_pairs crit_B( ND_pairs d, int s )
1947: {
1948: ND_pairs cur,head,prev,remove;
1.61 noro 1949: UINT *t,*tl,*lcm;
1.1 noro 1950: int td,tdl;
1951:
1952: if ( !d ) return 0;
1.20 noro 1953: t = DL(nd_psh[s]);
1.1 noro 1954: prev = 0;
1955: head = cur = d;
1.61 noro 1956: lcm = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
1.1 noro 1957: while ( cur ) {
1958: tl = cur->lcm;
1959: if ( ndl_reducible(tl,t)
1.20 noro 1960: && (ndl_lcm(DL(nd_psh[cur->i1]),t,lcm),!ndl_equal(lcm,tl))
1961: && (ndl_lcm(DL(nd_psh[cur->i2]),t,lcm),!ndl_equal(lcm,tl)) ) {
1.1 noro 1962: remove = cur;
1963: if ( !prev ) {
1964: head = cur = NEXT(cur);
1965: } else {
1966: cur = NEXT(prev) = NEXT(cur);
1967: }
1968: FREENDP(remove);
1969: } else {
1.34 noro 1970: prev = cur; cur = NEXT(cur);
1.1 noro 1971: }
1972: }
1973: return head;
1974: }
1975:
1976: ND_pairs crit_M( ND_pairs d1 )
1977: {
1978: ND_pairs e,d2,d3,dd,p;
1.61 noro 1979: UINT *id,*jd;
1.1 noro 1980:
1981: for ( dd = 0, e = d1; e; e = d3 ) {
1982: if ( !(d2 = NEXT(e)) ) {
1983: NEXT(e) = dd;
1984: return e;
1985: }
1.34 noro 1986: id = LCM(e);
1.1 noro 1987: for ( d3 = 0; d2; d2 = p ) {
1.34 noro 1988: p = NEXT(d2);
1989: jd = LCM(d2);
1990: if ( ndl_equal(jd,id) )
1991: ;
1992: else if ( TD(jd) > TD(id) )
1.1 noro 1993: if ( ndl_reducible(jd,id) ) continue;
1994: else ;
1.34 noro 1995: else if ( ndl_reducible(id,jd) ) goto delit;
1.1 noro 1996: NEXT(d2) = d3;
1997: d3 = d2;
1998: }
1999: NEXT(e) = dd;
2000: dd = e;
2001: continue;
2002: /**/
2003: delit: NEXT(d2) = d3;
2004: d3 = d2;
2005: for ( ; p; p = d2 ) {
2006: d2 = NEXT(p);
2007: NEXT(p) = d3;
2008: d3 = p;
2009: }
2010: FREENDP(e);
2011: }
2012: return dd;
2013: }
2014:
2015: ND_pairs crit_F( ND_pairs d1 )
2016: {
2017: ND_pairs rest, head,remove;
2018: ND_pairs last, p, r, w;
2019: int s;
2020:
2021: for ( head = last = 0, p = d1; NEXT(p); ) {
2022: r = w = equivalent_pairs(p,&rest);
1.14 noro 2023: s = SG(r);
1.1 noro 2024: w = NEXT(w);
2025: while ( w ) {
2026: if ( crit_2(w->i1,w->i2) ) {
2027: r = w;
2028: w = NEXT(w);
2029: while ( w ) {
2030: remove = w;
2031: w = NEXT(w);
2032: FREENDP(remove);
2033: }
2034: break;
1.14 noro 2035: } else if ( SG(w) < s ) {
1.1 noro 2036: FREENDP(r);
2037: r = w;
1.14 noro 2038: s = SG(r);
1.1 noro 2039: w = NEXT(w);
2040: } else {
2041: remove = w;
2042: w = NEXT(w);
2043: FREENDP(remove);
2044: }
2045: }
2046: if ( last ) NEXT(last) = r;
2047: else head = r;
2048: NEXT(last = r) = 0;
2049: p = rest;
2050: if ( !p ) return head;
2051: }
2052: if ( !last ) return p;
2053: NEXT(last) = p;
2054: return head;
2055: }
2056:
2057: int crit_2( int dp1, int dp2 )
2058: {
1.20 noro 2059: return ndl_disjoint(DL(nd_psh[dp1]),DL(nd_psh[dp2]));
1.1 noro 2060: }
2061:
1.40 noro 2062: ND_pairs equivalent_pairs( ND_pairs d1, ND_pairs *prest )
1.1 noro 2063: {
2064: ND_pairs w,p,r,s;
1.61 noro 2065: UINT *d;
1.1 noro 2066:
2067: w = d1;
1.34 noro 2068: d = LCM(w);
1.1 noro 2069: s = NEXT(w);
2070: NEXT(w) = 0;
2071: for ( r = 0; s; s = p ) {
2072: p = NEXT(s);
1.34 noro 2073: if ( ndl_equal(d,LCM(s)) ) {
1.39 noro 2074: NEXT(s) = w; w = s;
1.1 noro 2075: } else {
1.39 noro 2076: NEXT(s) = r; r = s;
1.1 noro 2077: }
2078: }
2079: *prest = r;
2080: return w;
2081: }
2082:
2083: NODE update_base(NODE nd,int ndp)
2084: {
1.61 noro 2085: UINT *dl, *dln;
1.137 noro 2086: NODE last, p, head;
1.1 noro 2087:
1.20 noro 2088: dl = DL(nd_psh[ndp]);
1.1 noro 2089: for ( head = last = 0, p = nd; p; ) {
1.20 noro 2090: dln = DL(nd_psh[(int)BDY(p)]);
1.34 noro 2091: if ( ndl_reducible( dln, dl ) ) {
1.1 noro 2092: p = NEXT(p);
2093: if ( last ) NEXT(last) = p;
2094: } else {
2095: if ( !last ) head = p;
2096: p = NEXT(last = p);
2097: }
2098: }
2099: head = append_one(head,ndp);
2100: return head;
2101: }
2102:
2103: ND_pairs nd_minp( ND_pairs d, ND_pairs *prest )
2104: {
2105: ND_pairs m,ml,p,l;
1.61 noro 2106: UINT *lcm;
1.33 noro 2107: int s,td,len,tlen,c,c1;
1.1 noro 2108:
2109: if ( !(p = NEXT(m = d)) ) {
2110: *prest = p;
2111: NEXT(m) = 0;
2112: return m;
2113: }
1.14 noro 2114: s = SG(m);
1.77 noro 2115: if ( !NoSugar ) {
2116: for ( ml = 0, l = m; p; p = NEXT(l = p) )
2117: if ( (SG(p) < s)
2118: || ((SG(p) == s) && (DL_COMPARE(LCM(p),LCM(m)) < 0)) ) {
2119: ml = l; m = p; s = SG(m);
2120: }
2121: } else {
2122: for ( ml = 0, l = m; p; p = NEXT(l = p) )
2123: if ( DL_COMPARE(LCM(p),LCM(m)) < 0 ) {
2124: ml = l; m = p; s = SG(m);
2125: }
2126: }
1.1 noro 2127: if ( !ml ) *prest = NEXT(m);
2128: else {
2129: NEXT(ml) = NEXT(m);
2130: *prest = d;
2131: }
2132: NEXT(m) = 0;
2133: return m;
2134: }
2135:
1.63 noro 2136: ND_pairs nd_minsugarp( ND_pairs d, ND_pairs *prest )
2137: {
1.69 noro 2138: int msugar,i;
1.63 noro 2139: ND_pairs t,dm0,dm,dr0,dr;
2140:
2141: for ( msugar = SG(d), t = NEXT(d); t; t = NEXT(t) )
2142: if ( SG(t) < msugar ) msugar = SG(t);
2143: dm0 = 0; dr0 = 0;
1.69 noro 2144: for ( i = 0, t = d; t; t = NEXT(t) )
2145: if ( i < nd_f4_nsp && SG(t) == msugar ) {
1.63 noro 2146: if ( dm0 ) NEXT(dm) = t;
2147: else dm0 = t;
2148: dm = t;
1.69 noro 2149: i++;
1.63 noro 2150: } else {
2151: if ( dr0 ) NEXT(dr) = t;
2152: else dr0 = t;
2153: dr = t;
2154: }
2155: NEXT(dm) = 0;
2156: if ( dr0 ) NEXT(dr) = 0;
2157: *prest = dr0;
2158: return dm0;
2159: }
2160:
1.77 noro 2161: int ndv_newps(int m,NDV a,NDV aq)
1.1 noro 2162: {
1.3 noro 2163: int len;
1.13 noro 2164: RHist r;
1.20 noro 2165: NDV b;
1.3 noro 2166:
1.1 noro 2167: if ( nd_psn == nd_pslen ) {
2168: nd_pslen *= 2;
1.11 noro 2169: nd_ps = (NDV *)REALLOC((char *)nd_ps,nd_pslen*sizeof(NDV));
1.53 noro 2170: nd_ps_trace = (NDV *)REALLOC((char *)nd_ps_trace,nd_pslen*sizeof(NDV));
1.13 noro 2171: nd_psh = (RHist *)REALLOC((char *)nd_psh,nd_pslen*sizeof(RHist));
1.61 noro 2172: nd_bound = (UINT **)
2173: REALLOC((char *)nd_bound,nd_pslen*sizeof(UINT *));
1.1 noro 2174: }
1.39 noro 2175: NEWRHist(r); nd_psh[nd_psn] = r;
1.63 noro 2176: nd_ps[nd_psn] = a;
1.39 noro 2177: if ( aq ) {
1.63 noro 2178: nd_ps_trace[nd_psn] = aq;
2179: nd_bound[nd_psn] = ndv_compute_bound(aq);
1.39 noro 2180: SG(r) = SG(aq); ndl_copy(HDL(aq),DL(r));
1.53 noro 2181: } else {
1.63 noro 2182: nd_bound[nd_psn] = ndv_compute_bound(a);
1.53 noro 2183: SG(r) = SG(a); ndl_copy(HDL(a),DL(r));
1.39 noro 2184: }
1.77 noro 2185: if ( nd_demand ) {
2186: if ( aq ) {
2187: ndv_save(nd_ps_trace[nd_psn],nd_psn);
2188: nd_ps_trace[nd_psn] = 0;
2189: } else {
2190: ndv_save(nd_ps[nd_psn],nd_psn);
2191: nd_ps[nd_psn] = 0;
2192: }
2193: }
1.1 noro 2194: return nd_psn++;
2195: }
2196:
1.128 noro 2197: void ndv_setup(int mod,int trace,NODE f,int dont_sort)
1.1 noro 2198: {
1.5 noro 2199: int i,j,td,len,max;
1.1 noro 2200: NODE s,s0,f0;
1.61 noro 2201: UINT *d;
1.13 noro 2202: RHist r;
1.69 noro 2203: NDV *w;
1.61 noro 2204: NDV a,am;
1.11 noro 2205:
2206: nd_found = 0; nd_notfirst = 0; nd_create = 0;
1.1 noro 2207:
1.69 noro 2208: for ( nd_psn = 0, s = f; s; s = NEXT(s) ) if ( BDY(s) ) nd_psn++;
2209: w = (NDV *)ALLOCA(nd_psn*sizeof(NDV));
2210: for ( i = 0, s = f; s; s = NEXT(s) ) if ( BDY(s) ) w[i++] = BDY(s);
1.139 ! noro 2211: if ( !dont_sort ) {
! 2212: /* XXX heuristic */
! 2213: if ( !nd_ord->id && (nd_ord->ord.simple<2) )
! 2214: qsort(w,nd_psn,sizeof(NDV),
! 2215: (int (*)(const void *,const void *))ndv_compare_rev);
! 2216: else
! 2217: qsort(w,nd_psn,sizeof(NDV),
! 2218: (int (*)(const void *,const void *))ndv_compare);
! 2219: }
1.69 noro 2220: nd_pslen = 2*nd_psn;
1.11 noro 2221: nd_ps = (NDV *)MALLOC(nd_pslen*sizeof(NDV));
1.53 noro 2222: nd_ps_trace = (NDV *)MALLOC(nd_pslen*sizeof(NDV));
1.13 noro 2223: nd_psh = (RHist *)MALLOC(nd_pslen*sizeof(RHist));
1.61 noro 2224: nd_bound = (UINT **)MALLOC(nd_pslen*sizeof(UINT *));
1.57 noro 2225:
1.11 noro 2226: if ( !nd_red )
1.13 noro 2227: nd_red = (RHist *)MALLOC(REDTAB_LEN*sizeof(RHist));
1.71 noro 2228: for ( i = 0; i < REDTAB_LEN; i++ ) nd_red[i] = 0;
1.69 noro 2229: for ( i = 0; i < nd_psn; i++ ) {
1.39 noro 2230: if ( trace ) {
1.69 noro 2231: a = nd_ps_trace[i] = ndv_dup(0,w[i]);
1.61 noro 2232: ndv_removecont(0,a);
2233: am = nd_ps[i] = ndv_dup(mod,a);
2234: ndv_mod(mod,am);
2235: ndv_removecont(mod,am);
2236: } else {
1.69 noro 2237: a = nd_ps[i] = ndv_dup(mod,w[i]);
1.61 noro 2238: ndv_removecont(mod,a);
1.39 noro 2239: }
1.61 noro 2240: NEWRHist(r); SG(r) = HTD(a); ndl_copy(HDL(a),DL(r));
1.57 noro 2241: nd_bound[i] = ndv_compute_bound(a);
1.20 noro 2242: nd_psh[i] = r;
1.77 noro 2243: if ( nd_demand ) {
2244: if ( trace ) {
2245: ndv_save(nd_ps_trace[i],i);
2246: nd_ps_trace[i] = 0;
2247: } else {
2248: ndv_save(nd_ps[i],i);
2249: nd_ps[i] = 0;
2250: }
2251: }
1.20 noro 2252: }
2253: }
2254:
1.119 noro 2255: struct order_spec *append_block(struct order_spec *spec,
2256: int nv,int nalg,int ord);
2257:
1.121 noro 2258: extern VECT current_dl_weight_vector_obj;
2259: static VECT prev_weight_vector_obj;
2260:
1.120 noro 2261: void preprocess_algcoef(VL vv,VL av,struct order_spec *ord,LIST f,
2262: struct order_spec **ord1p,LIST *f1p,NODE *alistp)
2263: {
1.121 noro 2264: NODE alist,t,s,r0,r,arg;
1.120 noro 2265: VL tv;
2266: P poly;
2267: DP d;
2268: Alg alpha,dp;
2269: DAlg inv,da,hc;
2270: MP m;
1.121 noro 2271: int i,nvar,nalg,n;
1.120 noro 2272: NumberField nf;
2273: LIST f1,f2;
2274: struct order_spec *current_spec;
1.121 noro 2275: VECT obj,obj0;
2276: Obj tmp;
1.120 noro 2277:
2278: for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++);
2279: for ( nalg = 0, tv = av; tv; tv = NEXT(tv), nalg++);
2280:
2281: for ( alist = 0, tv = av; tv; tv = NEXT(tv) ) {
2282: NEXTNODE(alist,t); MKV(tv->v,poly);
2283: MKAlg(poly,alpha); BDY(t) = (pointer)alpha;
2284: tv->v = tv->v->priv;
2285: }
2286: NEXT(t) = 0;
2287:
2288: /* simplification, makeing polynomials monic */
2289: setfield_dalg(alist);
2290: obj_algtodalg(f,&f1);
2291: for ( t = BDY(f); t; t = NEXT(t) ) {
2292: initd(ord); ptod(vv,vv,(P)BDY(t),&d);
2293: hc = (DAlg)BDY(d)->c;
2294: if ( NID(hc) == N_DA ) {
2295: invdalg(hc,&inv);
2296: for ( m = BDY(d); m; m = NEXT(m) ) {
2297: muldalg(inv,(DAlg)m->c,&da); m->c = (P)da;
2298: }
2299: }
2300: initd(ord); dtop(vv,vv,d,&poly); BDY(f) = (pointer)poly;
2301: }
2302: obj_dalgtoalg(f1,&f);
2303:
2304: /* append alg vars to the var list */
2305: for ( tv = vv; NEXT(tv); tv = NEXT(tv) );
2306: NEXT(tv) = av;
2307:
2308: /* append a block to ord */
2309: *ord1p = append_block(ord,nvar,nalg,2);
2310:
2311: /* create generator list */
2312: nf = get_numberfield();
2313: for ( i = nalg-1, t = BDY(f); i >= 0; i-- ) {
2314: MKAlg(nf->defpoly[i],dp);
2315: MKNODE(s,dp,t); t = s;
2316: }
2317: MKLIST(f1,t);
2318: *alistp = alist;
2319: algobjtorat(f1,f1p);
1.121 noro 2320:
2321: /* creating a new weight vector */
2322: prev_weight_vector_obj = obj0 = current_dl_weight_vector_obj;
2323: n = nvar+nalg+1;
2324: MKVECT(obj,n);
2325: if ( obj0 && obj0->len == nvar )
2326: for ( i = 0; i < nvar; i++ ) BDY(obj)[i] = BDY(obj0)[i];
2327: else
2328: for ( i = 0; i < nvar; i++ ) BDY(obj)[i] = (pointer)ONE;
2329: for ( i = 0; i < nalg; i++ ) BDY(obj)[i+nvar] = 0;
2330: BDY(obj)[n-1] = (pointer)ONE;
2331: arg = mknode(1,obj);
2332: Pdp_set_weight(arg,&tmp);
2333: }
2334:
2335: NODE postprocess_algcoef(VL av,NODE alist,NODE r)
2336: {
2337: NODE s,t,u0,u;
2338: P p;
2339: VL tv;
2340: Obj obj,tmp;
2341: NODE arg;
2342:
2343: u0 = 0;
2344: for ( t = r; t; t = NEXT(t) ) {
2345: p = (P)BDY(t);
2346: for ( tv = av, s = alist; tv; tv = NEXT(tv), s = NEXT(s) ) {
2347: substr(CO,0,(Obj)p,tv->v,(Obj)BDY(s),&obj); p = (P)obj;
2348: }
2349: if ( OID(p) == O_P || (OID(p) == O_N && NID((Num)p) != N_A) ) {
2350: NEXTNODE(u0,u);
2351: BDY(u) = (pointer)p;
2352: }
2353: }
2354: arg = mknode(1,prev_weight_vector_obj);
2355: Pdp_set_weight(arg,&tmp);
2356:
2357: return u0;
1.120 noro 2358: }
2359:
1.63 noro 2360: void nd_gr(LIST f,LIST v,int m,int f4,struct order_spec *ord,LIST *rp)
1.1 noro 2361: {
1.119 noro 2362: VL tv,fv,vv,vc,av;
2363: NODE fd,fd0,r,r0,t,x,s,xx,alist;
2364: int e,max,nvar,i;
1.69 noro 2365: NDV b;
1.119 noro 2366: int ishomo,nalg;
2367: Alg alpha,dp;
2368: P p;
2369: LIST f1,f2;
2370: Obj obj;
2371: NumberField nf;
1.120 noro 2372: struct order_spec *ord1;
1.1 noro 2373:
1.78 noro 2374: if ( !m && Demand ) nd_demand = 1;
2375: else nd_demand = 0;
2376:
1.103 noro 2377: #if 0
1.74 noro 2378: ndv_alloc = 0;
1.103 noro 2379: #endif
1.1 noro 2380: get_vars((Obj)f,&fv); pltovl(v,&vv);
1.74 noro 2381: for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++ );
1.88 noro 2382: switch ( ord->id ) {
2383: case 1:
2384: if ( ord->nv != nvar )
2385: error("nd_{gr,f4} : invalid order specification");
2386: break;
2387: default:
2388: break;
2389: }
1.119 noro 2390: nd_nalg = 0;
1.123 noro 2391: av = 0;
1.119 noro 2392: if ( !m ) {
2393: get_algtree((Obj)f,&av);
2394: for ( nalg = 0, tv = av; tv; tv = NEXT(tv), nalg++ );
1.121 noro 2395: nd_ntrans = nvar;
1.119 noro 2396: nd_nalg = nalg;
2397: /* #i -> t#i */
2398: if ( nalg ) {
1.120 noro 2399: preprocess_algcoef(vv,av,ord,f,&ord1,&f1,&alist);
2400: ord = ord1;
2401: f = f1;
1.119 noro 2402: }
1.120 noro 2403: nvar += nalg;
1.119 noro 2404: }
1.32 noro 2405: nd_init_ord(ord);
1.61 noro 2406: for ( t = BDY(f), max = 0; t; t = NEXT(t) )
2407: for ( tv = vv; tv; tv = NEXT(tv) ) {
2408: e = getdeg(tv->v,(P)BDY(t));
2409: max = MAX(e,max);
2410: }
2411: nd_setup_parameters(nvar,max);
1.77 noro 2412: ishomo = 1;
1.1 noro 2413: for ( fd0 = 0, t = BDY(f); t; t = NEXT(t) ) {
1.61 noro 2414: b = (pointer)ptondv(CO,vv,(P)BDY(t));
1.77 noro 2415: if ( ishomo )
2416: ishomo = ishomo && ndv_ishomo(b);
1.69 noro 2417: if ( m ) ndv_mod(m,b);
1.61 noro 2418: if ( b ) { NEXTNODE(fd0,fd); BDY(fd) = (pointer)b; }
1.1 noro 2419: }
2420: if ( fd0 ) NEXT(fd) = 0;
1.128 noro 2421: ndv_setup(m,0,fd0,0);
1.77 noro 2422: x = f4?nd_f4(m):nd_gb(m,ishomo,0);
1.82 noro 2423: nd_demand = 0;
1.61 noro 2424: x = ndv_reducebase(x);
2425: x = ndv_reduceall(m,x);
1.23 noro 2426: for ( r0 = 0, t = x; t; t = NEXT(t) ) {
1.1 noro 2427: NEXTNODE(r0,r);
1.61 noro 2428: BDY(r) = ndvtop(m,CO,vv,BDY(t));
1.20 noro 2429: }
2430: if ( r0 ) NEXT(r) = 0;
1.121 noro 2431: if ( nalg )
2432: r0 = postprocess_algcoef(av,alist,r0);
1.20 noro 2433: MKLIST(*rp,r0);
1.103 noro 2434: #if 0
1.74 noro 2435: fprintf(asir_out,"ndv_alloc=%d\n",ndv_alloc);
1.103 noro 2436: #endif
1.127 noro 2437: }
2438:
2439: void nd_gr_postproc(LIST f,LIST v,int m,struct order_spec *ord,int do_check,LIST *rp)
2440: {
2441: VL tv,fv,vv,vc,av;
2442: NODE fd,fd0,r,r0,t,x,s,xx,alist;
2443: int e,max,nvar,i;
2444: NDV b;
2445: int ishomo,nalg;
2446: Alg alpha,dp;
2447: P p;
2448: LIST f1,f2;
2449: Obj obj;
2450: NumberField nf;
2451: struct order_spec *ord1;
2452:
2453: get_vars((Obj)f,&fv); pltovl(v,&vv);
2454: for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++ );
2455: switch ( ord->id ) {
2456: case 1:
2457: if ( ord->nv != nvar )
2458: error("nd_check : invalid order specification");
2459: break;
2460: default:
2461: break;
2462: }
2463: nd_nalg = 0;
2464: av = 0;
2465: if ( !m ) {
2466: get_algtree((Obj)f,&av);
2467: for ( nalg = 0, tv = av; tv; tv = NEXT(tv), nalg++ );
2468: nd_ntrans = nvar;
2469: nd_nalg = nalg;
2470: /* #i -> t#i */
2471: if ( nalg ) {
2472: preprocess_algcoef(vv,av,ord,f,&ord1,&f1,&alist);
2473: ord = ord1;
2474: f = f1;
2475: }
2476: nvar += nalg;
2477: }
2478: nd_init_ord(ord);
2479: for ( t = BDY(f), max = 0; t; t = NEXT(t) )
2480: for ( tv = vv; tv; tv = NEXT(tv) ) {
2481: e = getdeg(tv->v,(P)BDY(t));
2482: max = MAX(e,max);
2483: }
2484: nd_setup_parameters(nvar,max);
2485: ishomo = 1;
2486: for ( fd0 = 0, t = BDY(f); t; t = NEXT(t) ) {
2487: b = (pointer)ptondv(CO,vv,(P)BDY(t));
2488: if ( ishomo )
2489: ishomo = ishomo && ndv_ishomo(b);
2490: if ( m ) ndv_mod(m,b);
2491: if ( b ) { NEXTNODE(fd0,fd); BDY(fd) = (pointer)b; }
2492: }
2493: if ( fd0 ) NEXT(fd) = 0;
1.128 noro 2494: ndv_setup(m,0,fd0,0);
1.127 noro 2495: for ( x = 0, i = 0; i < nd_psn; i++ )
2496: x = update_base(x,i);
2497: if ( do_check ) {
2498: x = nd_gb(m,ishomo,1);
2499: if ( !x ) {
2500: *rp = 0;
2501: return;
2502: }
2503: } else {
2504: for ( t = x; t; t = NEXT(t) )
2505: BDY(t) = (pointer)nd_ps[(int)BDY(t)];
2506: }
2507: x = ndv_reducebase(x);
2508: x = ndv_reduceall(m,x);
2509: for ( r0 = 0, t = x; t; t = NEXT(t) ) {
2510: NEXTNODE(r0,r);
2511: BDY(r) = ndvtop(m,CO,vv,BDY(t));
2512: }
2513: if ( r0 ) NEXT(r) = 0;
2514: if ( nalg )
2515: r0 = postprocess_algcoef(av,alist,r0);
2516: MKLIST(*rp,r0);
1.20 noro 2517: }
2518:
1.133 noro 2519: void nd_gr_trace(LIST f,LIST v,int trace,int homo,int f4,struct order_spec *ord,LIST *rp)
1.20 noro 2520: {
1.119 noro 2521: VL tv,fv,vv,vc,av;
2522: NODE fd,fd0,in0,in,r,r0,t,s,cand,alist;
1.61 noro 2523: int m,nocheck,nvar,mindex,e,max;
2524: NDV c;
2525: NMV a;
1.27 noro 2526: P p;
1.61 noro 2527: EPOS oepos;
1.119 noro 2528: int obpe,oadv,wmax,i,len,cbpe,ishomo,nalg;
2529: Alg alpha,dp;
2530: P poly;
2531: LIST f1,f2;
2532: Obj obj;
2533: NumberField nf;
1.120 noro 2534: struct order_spec *ord1;
1.20 noro 2535:
2536: get_vars((Obj)f,&fv); pltovl(v,&vv);
1.74 noro 2537: for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++ );
1.88 noro 2538: switch ( ord->id ) {
2539: case 1:
2540: if ( ord->nv != nvar )
2541: error("nd_gr_trace : invalid order specification");
2542: break;
2543: default:
2544: break;
2545: }
1.119 noro 2546:
2547: get_algtree((Obj)f,&av);
2548: for ( nalg = 0, tv = av; tv; tv = NEXT(tv), nalg++ );
1.121 noro 2549: nd_ntrans = nvar;
1.119 noro 2550: nd_nalg = nalg;
2551: /* #i -> t#i */
2552: if ( nalg ) {
1.120 noro 2553: preprocess_algcoef(vv,av,ord,f,&ord1,&f1,&alist);
2554: ord = ord1;
2555: f = f1;
1.119 noro 2556: }
1.120 noro 2557: nvar += nalg;
1.119 noro 2558:
1.52 noro 2559: nocheck = 0;
2560: mindex = 0;
2561:
1.78 noro 2562: if ( Demand ) nd_demand = 1;
2563: else nd_demand = 0;
2564:
1.52 noro 2565: /* setup modulus */
2566: if ( trace < 0 ) {
2567: trace = -trace;
2568: nocheck = 1;
2569: }
2570: m = trace > 1 ? trace : get_lprime(mindex);
1.61 noro 2571: for ( t = BDY(f), max = 0; t; t = NEXT(t) )
2572: for ( tv = vv; tv; tv = NEXT(tv) ) {
2573: e = getdeg(tv->v,(P)BDY(t));
2574: max = MAX(e,max);
1.23 noro 2575: }
1.61 noro 2576: nd_init_ord(ord);
2577: nd_setup_parameters(nvar,max);
2578: obpe = nd_bpe; oadv = nmv_adv; oepos = nd_epos;
1.82 noro 2579: ishomo = 1;
1.61 noro 2580: for ( in0 = 0, fd0 = 0, t = BDY(f); t; t = NEXT(t) ) {
2581: c = ptondv(CO,vv,(P)BDY(t));
1.82 noro 2582: if ( ishomo )
2583: ishomo = ishomo && ndv_ishomo(c);
1.61 noro 2584: if ( c ) {
2585: NEXTNODE(in0,in); BDY(in) = (pointer)c;
2586: NEXTNODE(fd0,fd); BDY(fd) = (pointer)ndv_dup(0,c);
1.23 noro 2587: }
1.61 noro 2588: }
2589: if ( in0 ) NEXT(in) = 0;
2590: if ( fd0 ) NEXT(fd) = 0;
1.82 noro 2591: if ( !ishomo && homo ) {
1.122 noro 2592: for ( t = in0, wmax = max; t; t = NEXT(t) ) {
1.61 noro 2593: c = (NDV)BDY(t); len = LEN(c);
2594: for ( a = BDY(c), i = 0; i < len; i++, NMV_ADV(a) )
2595: wmax = MAX(TD(DL(a)),wmax);
2596: }
2597: homogenize_order(ord,nvar,&ord1);
1.89 noro 2598: nd_init_ord(ord1);
1.61 noro 2599: nd_setup_parameters(nvar+1,wmax);
2600: for ( t = fd0; t; t = NEXT(t) )
2601: ndv_homogenize((NDV)BDY(t),obpe,oadv,oepos);
1.27 noro 2602: }
1.52 noro 2603: while ( 1 ) {
1.77 noro 2604: if ( Demand )
2605: nd_demand = 1;
1.128 noro 2606: ndv_setup(m,1,fd0,0);
1.133 noro 2607: cand = f4?nd_f4_trace(m):nd_gb_trace(m,ishomo || homo);
1.52 noro 2608: if ( !cand ) {
2609: /* failure */
1.61 noro 2610: if ( trace > 1 ) { *rp = 0; return; }
2611: else m = get_lprime(++mindex);
1.52 noro 2612: continue;
2613: }
1.82 noro 2614: if ( !ishomo && homo ) {
1.27 noro 2615: /* dehomogenization */
1.61 noro 2616: for ( t = cand; t; t = NEXT(t) ) ndv_dehomogenize((NDV)BDY(t),ord);
1.45 noro 2617: nd_init_ord(ord);
1.61 noro 2618: nd_setup_parameters(nvar,0);
1.27 noro 2619: }
1.77 noro 2620: nd_demand = 0;
1.61 noro 2621: cand = ndv_reducebase(cand);
2622: cand = ndv_reduceall(0,cand);
1.78 noro 2623: cbpe = nd_bpe;
1.61 noro 2624: if ( nocheck )
2625: break;
2626: if ( ndv_check_candidate(in0,obpe,oadv,oepos,cand) )
1.52 noro 2627: /* success */
2628: break;
2629: else if ( trace > 1 ) {
2630: /* failure */
2631: *rp = 0; return;
1.61 noro 2632: } else {
1.52 noro 2633: /* try the next modulus */
2634: m = get_lprime(++mindex);
1.61 noro 2635: /* reset the parameters */
1.82 noro 2636: if ( !ishomo && homo ) {
1.89 noro 2637: nd_init_ord(ord1);
1.61 noro 2638: nd_setup_parameters(nvar+1,wmax);
2639: } else {
2640: nd_init_ord(ord);
2641: nd_setup_parameters(nvar,max);
2642: }
2643: }
1.52 noro 2644: }
1.27 noro 2645: /* dp->p */
1.61 noro 2646: nd_bpe = cbpe;
1.73 noro 2647: nd_setup_parameters(nd_nvar,0);
1.121 noro 2648: for ( r = cand; r; r = NEXT(r) )
1.119 noro 2649: BDY(r) = (pointer)ndvtop(0,CO,vv,BDY(r));
1.121 noro 2650: if ( nalg )
2651: cand = postprocess_algcoef(av,alist,cand);
1.27 noro 2652: MKLIST(*rp,cand);
1.1 noro 2653: }
2654:
1.61 noro 2655: void dltondl(int n,DL dl,UINT *r)
1.1 noro 2656: {
1.61 noro 2657: UINT *d;
1.57 noro 2658: int i,j,l,s,ord_l;
1.43 noro 2659: struct order_pair *op;
1.1 noro 2660:
2661: d = dl->d;
1.41 noro 2662: for ( i = 0; i < nd_wpd; i++ ) r[i] = 0;
1.43 noro 2663: if ( nd_blockmask ) {
2664: l = nd_blockmask->n;
2665: op = nd_blockmask->order_pair;
2666: for ( j = 0, s = 0; j < l; j++ ) {
2667: ord_l = op[j].length;
1.57 noro 2668: for ( i = 0; i < ord_l; i++, s++ ) PUT_EXP(r,s,d[s]);
1.43 noro 2669: }
2670: TD(r) = ndl_weight(r);
1.61 noro 2671: ndl_weight_mask(r);
1.43 noro 2672: } else {
1.56 noro 2673: for ( i = 0; i < n; i++ ) PUT_EXP(r,i,d[i]);
1.43 noro 2674: TD(r) = ndl_weight(r);
2675: }
1.1 noro 2676: }
2677:
1.61 noro 2678: DL ndltodl(int n,UINT *ndl)
1.1 noro 2679: {
2680: DL dl;
2681: int *d;
1.57 noro 2682: int i,j,l,s,ord_l;
1.43 noro 2683: struct order_pair *op;
1.1 noro 2684:
2685: NEWDL(dl,n);
1.34 noro 2686: dl->td = TD(ndl);
1.1 noro 2687: d = dl->d;
1.43 noro 2688: if ( nd_blockmask ) {
2689: l = nd_blockmask->n;
2690: op = nd_blockmask->order_pair;
2691: for ( j = 0, s = 0; j < l; j++ ) {
2692: ord_l = op[j].length;
1.57 noro 2693: for ( i = 0; i < ord_l; i++, s++ ) d[s] = GET_EXP(ndl,s);
1.43 noro 2694: }
2695: } else {
1.56 noro 2696: for ( i = 0; i < n; i++ ) d[i] = GET_EXP(ndl,i);
1.43 noro 2697: }
1.1 noro 2698: return dl;
2699: }
2700:
1.61 noro 2701: void ndl_print(UINT *dl)
1.1 noro 2702: {
2703: int n;
1.57 noro 2704: int i,j,l,ord_l,s,s0;
1.43 noro 2705: struct order_pair *op;
1.1 noro 2706:
2707: n = nd_nvar;
2708: printf("<<");
1.43 noro 2709: if ( nd_blockmask ) {
2710: l = nd_blockmask->n;
2711: op = nd_blockmask->order_pair;
2712: for ( j = 0, s = s0 = 0; j < l; j++ ) {
2713: ord_l = op[j].length;
1.57 noro 2714: for ( i = 0; i < ord_l; i++, s++ )
2715: printf(s==n-1?"%d":"%d,",GET_EXP(dl,s));
1.43 noro 2716: }
2717: } else {
1.56 noro 2718: for ( i = 0; i < n; i++ ) printf(i==n-1?"%d":"%d,",GET_EXP(dl,i));
1.43 noro 2719: }
1.1 noro 2720: printf(">>");
2721: }
2722:
2723: void nd_print(ND p)
2724: {
2725: NM m;
2726:
2727: if ( !p )
2728: printf("0\n");
2729: else {
2730: for ( m = BDY(p); m; m = NEXT(m) ) {
1.71 noro 2731: if ( CM(m) & 0x80000000 ) printf("+@_%d*",IFTOF(CM(m)));
2732: else printf("+%d*",CM(m));
1.14 noro 2733: ndl_print(DL(m));
1.1 noro 2734: }
2735: printf("\n");
2736: }
2737: }
2738:
1.113 noro 2739: void nd_print_q(ND p)
1.16 noro 2740: {
2741: NM m;
2742:
2743: if ( !p )
2744: printf("0\n");
2745: else {
2746: for ( m = BDY(p); m; m = NEXT(m) ) {
2747: printf("+");
1.113 noro 2748: printexpr(CO,(Obj)CQ(m));
1.16 noro 2749: printf("*");
2750: ndl_print(DL(m));
2751: }
2752: printf("\n");
2753: }
2754: }
2755:
1.1 noro 2756: void ndp_print(ND_pairs d)
2757: {
2758: ND_pairs t;
2759:
1.34 noro 2760: for ( t = d; t; t = NEXT(t) ) printf("%d,%d ",t->i1,t->i2);
1.1 noro 2761: printf("\n");
2762: }
2763:
1.20 noro 2764: void nd_removecont(int mod,ND p)
1.16 noro 2765: {
2766: int i,n;
1.113 noro 2767: Q *w;
2768: Q dvr,t;
1.16 noro 2769: NM m;
1.21 noro 2770: struct oVECT v;
1.113 noro 2771: N q,r;
1.16 noro 2772:
1.71 noro 2773: if ( mod == -1 ) nd_mul_c(mod,p,_invsf(HCM(p)));
2774: else if ( mod ) nd_mul_c(mod,p,invm(HCM(p),mod));
1.20 noro 2775: else {
2776: for ( m = BDY(p), n = 0; m; m = NEXT(m), n++ );
1.113 noro 2777: w = (Q *)ALLOCA(n*sizeof(Q));
1.21 noro 2778: v.len = n;
2779: v.body = (pointer *)w;
1.113 noro 2780: for ( m = BDY(p), i = 0; i < n; m = NEXT(m), i++ ) w[i] = CQ(m);
1.21 noro 2781: removecont_array(w,n);
1.113 noro 2782: for ( m = BDY(p), i = 0; i < n; m = NEXT(m), i++ ) CQ(m) = w[i];
1.16 noro 2783: }
2784: }
2785:
1.21 noro 2786: void nd_removecont2(ND p1,ND p2)
2787: {
2788: int i,n1,n2,n;
1.113 noro 2789: Q *w;
2790: Q dvr,t;
1.21 noro 2791: NM m;
2792: struct oVECT v;
1.113 noro 2793: N q,r;
1.21 noro 2794:
2795: if ( !p1 ) {
2796: nd_removecont(0,p2); return;
2797: } else if ( !p2 ) {
2798: nd_removecont(0,p1); return;
2799: }
2800: n1 = nd_length(p1);
2801: n2 = nd_length(p2);
2802: n = n1+n2;
1.113 noro 2803: w = (Q *)ALLOCA(n*sizeof(Q));
1.21 noro 2804: v.len = n;
2805: v.body = (pointer *)w;
1.113 noro 2806: for ( m = BDY(p1), i = 0; i < n1; m = NEXT(m), i++ ) w[i] = CQ(m);
2807: for ( m = BDY(p2); i < n; m = NEXT(m), i++ ) w[i] = CQ(m);
1.21 noro 2808: removecont_array(w,n);
1.113 noro 2809: for ( m = BDY(p1), i = 0; i < n1; m = NEXT(m), i++ ) CQ(m) = w[i];
2810: for ( m = BDY(p2); i < n; m = NEXT(m), i++ ) CQ(m) = w[i];
1.21 noro 2811: }
2812:
1.20 noro 2813: void ndv_removecont(int mod,NDV p)
1.16 noro 2814: {
2815: int i,len;
1.113 noro 2816: Q *w;
2817: Q dvr,t;
1.16 noro 2818: NMV m;
2819:
1.71 noro 2820: if ( mod == -1 )
2821: ndv_mul_c(mod,p,_invsf(HCM(p)));
2822: else if ( mod )
1.20 noro 2823: ndv_mul_c(mod,p,invm(HCM(p),mod));
2824: else {
2825: len = p->len;
1.113 noro 2826: w = (Q *)ALLOCA(len*sizeof(Q));
2827: for ( m = BDY(p), i = 0; i < len; NMV_ADV(m), i++ ) w[i] = CQ(m);
2828: sortbynm(w,len);
2829: qltozl(w,len,&dvr);
2830: for ( m = BDY(p), i = 0; i < len; NMV_ADV(m), i++ ) {
2831: divq(CQ(m),dvr,&t); CQ(m) = t;
2832: }
1.16 noro 2833: }
1.21 noro 2834: }
2835:
1.61 noro 2836: void ndv_homogenize(NDV p,int obpe,int oadv,EPOS oepos)
2837: {
2838: int len,i,max;
2839: NMV m,mr0,mr,t;
2840:
2841: len = p->len;
2842: for ( m = BDY(p), i = 0, max = 0; i < len; NMV_OADV(m), i++ )
2843: max = MAX(max,TD(DL(m)));
2844: mr0 = nmv_adv>oadv?(NMV)REALLOC(BDY(p),len*nmv_adv):BDY(p);
2845: m = (NMV)((char *)mr0+(len-1)*oadv);
2846: mr = (NMV)((char *)mr0+(len-1)*nmv_adv);
2847: t = (NMV)ALLOCA(nmv_adv);
2848: for ( i = 0; i < len; i++, NMV_OPREV(m), NMV_PREV(mr) ) {
2849: ndl_homogenize(DL(m),DL(t),obpe,oepos,max);
1.113 noro 2850: CQ(mr) = CQ(m);
1.61 noro 2851: ndl_copy(DL(t),DL(mr));
2852: }
2853: NV(p)++;
2854: BDY(p) = mr0;
2855: }
2856:
1.45 noro 2857: void ndv_dehomogenize(NDV p,struct order_spec *ord)
1.23 noro 2858: {
1.45 noro 2859: int i,j,adj,len,newnvar,newwpd,newadv,newexporigin;
1.113 noro 2860: Q *w;
2861: Q dvr,t;
1.23 noro 2862: NMV m,r;
2863:
2864: len = p->len;
2865: newnvar = nd_nvar-1;
1.48 noro 2866: newexporigin = nd_get_exporigin(ord);
1.45 noro 2867: newwpd = newnvar/nd_epw+(newnvar%nd_epw?1:0)+newexporigin;
1.23 noro 2868: for ( m = BDY(p), i = 0; i < len; NMV_ADV(m), i++ )
1.34 noro 2869: ndl_dehomogenize(DL(m));
1.23 noro 2870: if ( newwpd != nd_wpd ) {
1.90 noro 2871: newadv = ROUND_FOR_ALIGN(sizeof(struct oNMV)+(newwpd-1)*sizeof(UINT));
1.61 noro 2872: for ( m = r = BDY(p), i = 0; i < len; NMV_ADV(m), NDV_NADV(r), i++ ) {
1.113 noro 2873: CQ(r) = CQ(m);
1.45 noro 2874: for ( j = 0; j < newexporigin; j++ ) DL(r)[j] = DL(m)[j];
2875: adj = nd_exporigin-newexporigin;
2876: for ( ; j < newwpd; j++ ) DL(r)[j] = DL(m)[j+adj];
1.23 noro 2877: }
2878: }
2879: NV(p)--;
2880: }
2881:
1.113 noro 2882: void removecont_array(Q *c,int n)
1.21 noro 2883: {
2884: struct oVECT v;
1.113 noro 2885: Q d0,d1,a,u,u1,gcd;
1.110 noro 2886: int i,j;
1.113 noro 2887: N qn,rn,gn;
2888: Q *q,*r;
1.21 noro 2889:
1.113 noro 2890: q = (Q *)ALLOCA(n*sizeof(Q));
2891: r = (Q *)ALLOCA(n*sizeof(Q));
1.21 noro 2892: v.id = O_VECT; v.len = n; v.body = (pointer *)c;
1.113 noro 2893: igcdv_estimate(&v,&d0);
2894: for ( i = 0; i < n; i++ ) {
2895: divn(NM(c[i]),NM(d0),&qn,&rn);
2896: NTOQ(qn,SGN(c[i])*SGN(d0),q[i]);
2897: NTOQ(rn,SGN(c[i]),r[i]);
2898: }
1.34 noro 2899: for ( i = 0; i < n; i++ ) if ( r[i] ) break;
1.21 noro 2900: if ( i < n ) {
2901: v.id = O_VECT; v.len = n; v.body = (pointer *)r;
1.113 noro 2902: igcdv(&v,&d1);
2903: gcdn(NM(d0),NM(d1),&gn); NTOQ(gn,1,gcd);
2904: divsn(NM(d0),gn,&qn); NTOQ(qn,1,a);
1.21 noro 2905: for ( i = 0; i < n; i++ ) {
1.113 noro 2906: mulq(a,q[i],&u);
2907: if ( r[i] ) {
2908: divsn(NM(r[i]),gn,&qn); NTOQ(qn,SGN(r[i]),u1);
2909: addq(u,u1,&q[i]);
2910: } else
2911: q[i] = u;
1.21 noro 2912: }
2913: }
1.34 noro 2914: for ( i = 0; i < n; i++ ) c[i] = q[i];
1.16 noro 2915: }
2916:
1.19 noro 2917: void nd_mul_c(int mod,ND p,int mul)
1.1 noro 2918: {
2919: NM m;
2920: int c,c1;
2921:
1.34 noro 2922: if ( !p ) return;
1.115 noro 2923: if ( mul == 1 ) return;
1.71 noro 2924: if ( mod == -1 )
2925: for ( m = BDY(p); m; m = NEXT(m) )
2926: CM(m) = _mulsf(CM(m),mul);
2927: else
2928: for ( m = BDY(p); m; m = NEXT(m) ) {
2929: c1 = CM(m); DMAR(c1,mul,0,mod,c); CM(m) = c;
2930: }
1.1 noro 2931: }
2932:
1.113 noro 2933: void nd_mul_c_q(ND p,Q mul)
1.16 noro 2934: {
2935: NM m;
1.113 noro 2936: Q c;
1.16 noro 2937:
1.113 noro 2938: if ( !p ) return;
1.115 noro 2939: if ( UNIQ(mul) ) return;
1.16 noro 2940: for ( m = BDY(p); m; m = NEXT(m) ) {
1.113 noro 2941: mulq(CQ(m),mul,&c); CQ(m) = c;
1.16 noro 2942: }
2943: }
2944:
1.61 noro 2945: void nd_mul_c_p(VL vl,ND p,P mul)
2946: {
2947: NM m;
2948: P c;
2949:
2950: if ( !p ) return;
2951: for ( m = BDY(p); m; m = NEXT(m) ) {
2952: mulp(vl,CP(m),mul,&c); CP(m) = c;
2953: }
2954: }
2955:
1.1 noro 2956: void nd_free(ND p)
2957: {
2958: NM t,s;
2959:
1.34 noro 2960: if ( !p ) return;
1.1 noro 2961: t = BDY(p);
2962: while ( t ) {
2963: s = NEXT(t);
2964: FREENM(t);
2965: t = s;
2966: }
2967: FREEND(p);
2968: }
2969:
1.23 noro 2970: void ndv_free(NDV p)
2971: {
2972: GC_free(BDY(p));
2973: }
2974:
1.61 noro 2975: void nd_append_red(UINT *d,int i)
1.1 noro 2976: {
1.13 noro 2977: RHist m,m0;
1.1 noro 2978: int h;
2979:
1.13 noro 2980: NEWRHist(m);
1.34 noro 2981: h = ndl_hash_value(d);
1.13 noro 2982: m->index = i;
1.14 noro 2983: ndl_copy(d,DL(m));
1.1 noro 2984: NEXT(m) = nd_red[h];
2985: nd_red[h] = m;
2986: }
2987:
1.61 noro 2988: UINT *ndv_compute_bound(NDV p)
1.1 noro 2989: {
1.61 noro 2990: UINT *d1,*d2,*t;
2991: UINT u;
1.57 noro 2992: int i,j,k,l,len,ind;
1.45 noro 2993: NMV m;
1.1 noro 2994:
2995: if ( !p )
2996: return 0;
1.61 noro 2997: d1 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
2998: d2 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
1.45 noro 2999: len = LEN(p);
3000: m = BDY(p); ndl_copy(DL(m),d1); NMV_ADV(m);
3001: for ( i = 1; i < len; i++, NMV_ADV(m) ) {
1.14 noro 3002: ndl_lcm(DL(m),d1,d2);
1.1 noro 3003: t = d1; d1 = d2; d2 = t;
3004: }
1.12 noro 3005: l = nd_nvar+31;
1.61 noro 3006: t = (UINT *)MALLOC_ATOMIC(l*sizeof(UINT));
1.57 noro 3007: for ( i = nd_exporigin, ind = 0; i < nd_wpd; i++ ) {
3008: u = d1[i];
3009: k = (nd_epw-1)*nd_bpe;
3010: for ( j = 0; j < nd_epw; j++, k -= nd_bpe, ind++ )
3011: t[ind] = (u>>k)&nd_mask0;
3012: }
3013: for ( ; ind < l; ind++ ) t[ind] = 0;
1.1 noro 3014: return t;
3015: }
3016:
1.99 noro 3017: UINT *nd_compute_bound(ND p)
3018: {
3019: UINT *d1,*d2,*t;
3020: UINT u;
3021: int i,j,k,l,len,ind;
3022: NM m;
3023:
3024: if ( !p )
3025: return 0;
3026: d1 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
3027: d2 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
3028: len = LEN(p);
3029: m = BDY(p); ndl_copy(DL(m),d1); m = NEXT(m);
3030: for ( m = NEXT(m); m; m = NEXT(m) ) {
3031: ndl_lcm(DL(m),d1,d2);
3032: t = d1; d1 = d2; d2 = t;
3033: }
3034: l = nd_nvar+31;
3035: t = (UINT *)MALLOC_ATOMIC(l*sizeof(UINT));
3036: for ( i = nd_exporigin, ind = 0; i < nd_wpd; i++ ) {
3037: u = d1[i];
3038: k = (nd_epw-1)*nd_bpe;
3039: for ( j = 0; j < nd_epw; j++, k -= nd_bpe, ind++ )
3040: t[ind] = (u>>k)&nd_mask0;
3041: }
3042: for ( ; ind < l; ind++ ) t[ind] = 0;
3043: return t;
3044: }
3045:
1.48 noro 3046: int nd_get_exporigin(struct order_spec *ord)
3047: {
1.51 noro 3048: switch ( ord->id ) {
1.96 noro 3049: case 0: case 2:
1.48 noro 3050: return 1;
1.41 noro 3051: case 1:
3052: /* block order */
1.43 noro 3053: /* d[0]:weight d[1]:w0,...,d[nd_exporigin-1]:w(n-1) */
1.48 noro 3054: return ord->ord.block.length+1;
1.96 noro 3055: case 3:
3056: error("nd_get_exporigin : composite order is not supported yet.");
1.41 noro 3057: }
1.48 noro 3058: }
3059:
1.61 noro 3060: void nd_setup_parameters(int nvar,int max) {
1.79 noro 3061: int i,j,n,elen,ord_o,ord_l,l,s,wpd;
1.57 noro 3062: struct order_pair *op;
1.48 noro 3063:
1.73 noro 3064: nd_nvar = nvar;
1.79 noro 3065: if ( max ) {
1.83 noro 3066: /* XXX */
3067: if ( do_weyl ) nd_bpe = 32;
3068: else if ( max < 2 ) nd_bpe = 1;
1.79 noro 3069: else if ( max < 4 ) nd_bpe = 2;
3070: else if ( max < 8 ) nd_bpe = 3;
3071: else if ( max < 16 ) nd_bpe = 4;
3072: else if ( max < 32 ) nd_bpe = 5;
3073: else if ( max < 64 ) nd_bpe = 6;
3074: else if ( max < 256 ) nd_bpe = 8;
3075: else if ( max < 1024 ) nd_bpe = 10;
3076: else if ( max < 65536 ) nd_bpe = 16;
3077: else nd_bpe = 32;
3078: }
1.61 noro 3079: nd_epw = (sizeof(UINT)*8)/nd_bpe;
1.48 noro 3080: elen = nd_nvar/nd_epw+(nd_nvar%nd_epw?1:0);
3081: nd_exporigin = nd_get_exporigin(nd_ord);
1.79 noro 3082: wpd = nd_exporigin+elen;
3083: if ( wpd != nd_wpd ) {
3084: nd_free_private_storage();
3085: nd_wpd = wpd;
3086: }
1.1 noro 3087: if ( nd_bpe < 32 ) {
3088: nd_mask0 = (1<<nd_bpe)-1;
3089: } else {
3090: nd_mask0 = 0xffffffff;
3091: }
3092: bzero(nd_mask,sizeof(nd_mask));
3093: nd_mask1 = 0;
3094: for ( i = 0; i < nd_epw; i++ ) {
3095: nd_mask[nd_epw-i-1] = (nd_mask0<<(i*nd_bpe));
3096: nd_mask1 |= (1<<(nd_bpe-1))<<(i*nd_bpe);
3097: }
1.90 noro 3098: nmv_adv = ROUND_FOR_ALIGN(sizeof(struct oNMV)+(nd_wpd-1)*sizeof(UINT));
1.57 noro 3099: nd_epos = nd_create_epos(nd_ord);
1.43 noro 3100: nd_blockmask = nd_create_blockmask(nd_ord);
1.96 noro 3101: nd_work_vector = (int *)REALLOC(nd_work_vector,nd_nvar*sizeof(int));
1.1 noro 3102: }
3103:
1.103 noro 3104: ND_pairs nd_reconstruct(int trace,ND_pairs d)
1.1 noro 3105: {
1.37 noro 3106: int i,obpe,oadv,h;
1.71 noro 3107: static NM prev_nm_free_list;
3108: static ND_pairs prev_ndp_free_list;
1.13 noro 3109: RHist mr0,mr;
3110: RHist r;
1.37 noro 3111: RHist *old_red;
1.71 noro 3112: ND_pairs s0,s,t;
1.43 noro 3113: EPOS oepos;
1.15 noro 3114:
1.1 noro 3115: obpe = nd_bpe;
1.11 noro 3116: oadv = nmv_adv;
1.43 noro 3117: oepos = nd_epos;
1.63 noro 3118: if ( obpe < 2 ) nd_bpe = 2;
3119: else if ( obpe < 3 ) nd_bpe = 3;
1.62 noro 3120: else if ( obpe < 4 ) nd_bpe = 4;
1.63 noro 3121: else if ( obpe < 5 ) nd_bpe = 5;
1.34 noro 3122: else if ( obpe < 6 ) nd_bpe = 6;
3123: else if ( obpe < 8 ) nd_bpe = 8;
1.63 noro 3124: else if ( obpe < 10 ) nd_bpe = 10;
1.34 noro 3125: else if ( obpe < 16 ) nd_bpe = 16;
3126: else if ( obpe < 32 ) nd_bpe = 32;
3127: else error("nd_reconstruct : exponent too large");
1.5 noro 3128:
1.73 noro 3129: nd_setup_parameters(nd_nvar,0);
1.1 noro 3130: prev_nm_free_list = _nm_free_list;
3131: prev_ndp_free_list = _ndp_free_list;
3132: _nm_free_list = 0;
3133: _ndp_free_list = 0;
1.53 noro 3134: for ( i = nd_psn-1; i >= 0; i-- ) ndv_realloc(nd_ps[i],obpe,oadv,oepos);
3135: if ( trace )
3136: for ( i = nd_psn-1; i >= 0; i-- )
3137: ndv_realloc(nd_ps_trace[i],obpe,oadv,oepos);
1.1 noro 3138: s0 = 0;
3139: for ( t = d; t; t = NEXT(t) ) {
3140: NEXTND_pairs(s0,s);
3141: s->i1 = t->i1;
3142: s->i2 = t->i2;
1.14 noro 3143: SG(s) = SG(t);
1.61 noro 3144: ndl_reconstruct(LCM(t),LCM(s),obpe,oepos);
1.1 noro 3145: }
1.37 noro 3146:
3147: old_red = (RHist *)ALLOCA(REDTAB_LEN*sizeof(RHist));
1.6 noro 3148: for ( i = 0; i < REDTAB_LEN; i++ ) {
1.37 noro 3149: old_red[i] = nd_red[i];
3150: nd_red[i] = 0;
3151: }
3152: for ( i = 0; i < REDTAB_LEN; i++ )
3153: for ( r = old_red[i]; r; r = NEXT(r) ) {
3154: NEWRHist(mr);
1.13 noro 3155: mr->index = r->index;
1.20 noro 3156: SG(mr) = SG(r);
1.61 noro 3157: ndl_reconstruct(DL(r),DL(mr),obpe,oepos);
1.37 noro 3158: h = ndl_hash_value(DL(mr));
3159: NEXT(mr) = nd_red[h];
3160: nd_red[h] = mr;
1.6 noro 3161: }
1.37 noro 3162: for ( i = 0; i < REDTAB_LEN; i++ ) old_red[i] = 0;
3163: old_red = 0;
1.11 noro 3164: for ( i = 0; i < nd_psn; i++ ) {
1.20 noro 3165: NEWRHist(r); SG(r) = SG(nd_psh[i]);
1.61 noro 3166: ndl_reconstruct(DL(nd_psh[i]),DL(r),obpe,oepos);
1.13 noro 3167: nd_psh[i] = r;
1.11 noro 3168: }
1.1 noro 3169: if ( s0 ) NEXT(s) = 0;
3170: prev_nm_free_list = 0;
3171: prev_ndp_free_list = 0;
1.71 noro 3172: #if 0
1.1 noro 3173: GC_gcollect();
1.71 noro 3174: #endif
1.1 noro 3175: return s0;
3176: }
3177:
1.61 noro 3178: void ndl_reconstruct(UINT *d,UINT *r,int obpe,EPOS oepos)
1.1 noro 3179: {
1.57 noro 3180: int n,i,ei,oepw,omask0,j,s,ord_l,l;
1.43 noro 3181: struct order_pair *op;
1.1 noro 3182:
3183: n = nd_nvar;
1.61 noro 3184: oepw = (sizeof(UINT)*8)/obpe;
1.43 noro 3185: omask0 = (1<<obpe)-1;
1.34 noro 3186: TD(r) = TD(d);
1.41 noro 3187: for ( i = nd_exporigin; i < nd_wpd; i++ ) r[i] = 0;
1.43 noro 3188: if ( nd_blockmask ) {
3189: l = nd_blockmask->n;
3190: op = nd_blockmask->order_pair;
3191: for ( i = 1; i < nd_exporigin; i++ )
3192: r[i] = d[i];
3193: for ( j = 0, s = 0; j < l; j++ ) {
3194: ord_l = op[j].length;
1.57 noro 3195: for ( i = 0; i < ord_l; i++, s++ ) {
3196: ei = GET_EXP_OLD(d,s);
3197: PUT_EXP(r,s,ei);
3198: }
1.1 noro 3199: }
1.43 noro 3200: } else {
1.56 noro 3201: for ( i = 0; i < n; i++ ) {
3202: ei = GET_EXP_OLD(d,i);
3203: PUT_EXP(r,i,ei);
3204: }
1.1 noro 3205: }
3206: }
1.3 noro 3207:
1.6 noro 3208: ND nd_copy(ND p)
3209: {
3210: NM m,mr,mr0;
1.41 noro 3211: int c,n;
1.6 noro 3212: ND r;
3213:
3214: if ( !p )
3215: return 0;
3216: else {
3217: for ( mr0 = 0, m = BDY(p); m; m = NEXT(m) ) {
3218: NEXTNM(mr0,mr);
1.14 noro 3219: CM(mr) = CM(m);
3220: ndl_copy(DL(m),DL(mr));
1.6 noro 3221: }
3222: NEXT(mr) = 0;
1.31 noro 3223: MKND(NV(p),mr0,LEN(p),r);
1.14 noro 3224: SG(r) = SG(p);
1.6 noro 3225: return r;
3226: }
3227: }
3228:
1.53 noro 3229: int nd_sp(int mod,int trace,ND_pairs p,ND *rp)
1.11 noro 3230: {
3231: NM m;
3232: NDV p1,p2;
3233: ND t1,t2;
1.61 noro 3234: UINT *lcm;
1.31 noro 3235: int td;
1.11 noro 3236:
1.77 noro 3237: if ( !mod && nd_demand ) {
3238: p1 = ndv_load(p->i1); p2 = ndv_load(p->i2);
1.53 noro 3239: } else {
1.77 noro 3240: if ( trace ) {
3241: p1 = nd_ps_trace[p->i1]; p2 = nd_ps_trace[p->i2];
3242: } else {
3243: p1 = nd_ps[p->i1]; p2 = nd_ps[p->i2];
3244: }
1.20 noro 3245: }
1.34 noro 3246: lcm = LCM(p);
1.11 noro 3247: NEWNM(m);
1.113 noro 3248: CQ(m) = HCQ(p2);
1.34 noro 3249: ndl_sub(lcm,HDL(p1),DL(m));
1.56 noro 3250: if ( ndl_check_bound2(p->i1,DL(m)) )
3251: return 0;
1.55 noro 3252: t1 = ndv_mul_nm(mod,m,p1);
1.71 noro 3253: if ( mod == -1 ) CM(m) = _chsgnsf(HCM(p1));
3254: else if ( mod ) CM(m) = mod-HCM(p1);
1.113 noro 3255: else chsgnq(HCQ(p1),&CQ(m));
1.34 noro 3256: ndl_sub(lcm,HDL(p2),DL(m));
1.14 noro 3257: if ( ndl_check_bound2(p->i2,DL(m)) ) {
1.11 noro 3258: nd_free(t1);
3259: return 0;
3260: }
1.55 noro 3261: t2 = ndv_mul_nm(mod,m,p2);
1.31 noro 3262: *rp = nd_add(mod,t1,t2);
1.11 noro 3263: FREENM(m);
3264: return 1;
3265: }
3266:
1.19 noro 3267: void ndv_mul_c(int mod,NDV p,int mul)
1.11 noro 3268: {
3269: NMV m;
3270: int c,c1,len,i;
3271:
1.34 noro 3272: if ( !p ) return;
1.14 noro 3273: len = LEN(p);
1.71 noro 3274: if ( mod == -1 )
3275: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) )
3276: CM(m) = _mulsf(CM(m),mul);
3277: else
3278: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
3279: c1 = CM(m); DMAR(c1,mul,0,mod,c); CM(m) = c;
3280: }
1.11 noro 3281: }
3282:
1.113 noro 3283: void ndv_mul_c_q(NDV p,Q mul)
1.16 noro 3284: {
3285: NMV m;
1.113 noro 3286: Q c;
1.16 noro 3287: int len,i;
3288:
1.34 noro 3289: if ( !p ) return;
1.16 noro 3290: len = LEN(p);
3291: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
1.113 noro 3292: mulq(CQ(m),mul,&c); CQ(m) = c;
1.16 noro 3293: }
3294: }
3295:
1.55 noro 3296: ND weyl_ndv_mul_nm(int mod,NM m0,NDV p) {
3297: int n2,i,j,l,n,tlen;
1.61 noro 3298: UINT *d0;
1.55 noro 3299: NM *tab,*psum;
3300: ND s,r;
3301: NM t;
3302: NMV m1;
3303:
3304: if ( !p ) return 0;
3305: n = NV(p); n2 = n>>1;
3306: d0 = DL(m0);
3307: l = LEN(p);
3308: for ( i = 0, tlen = 1; i < n2; i++ ) tlen *= (GET_EXP(d0,n2+i)+1);
3309: tab = (NM *)ALLOCA(tlen*sizeof(NM));
3310: psum = (NM *)ALLOCA(tlen*sizeof(NM));
3311: for ( i = 0; i < tlen; i++ ) psum[i] = 0;
1.56 noro 3312: m1 = (NMV)(((char *)BDY(p))+nmv_adv*(l-1));
3313: for ( i = l-1; i >= 0; i--, NMV_PREV(m1) ) {
1.55 noro 3314: /* m0(NM) * m1(NMV) => tab(NM) */
1.56 noro 3315: weyl_mul_nm_nmv(n,mod,m0,m1,tab,tlen);
1.55 noro 3316: for ( j = 0; j < tlen; j++ ) {
3317: if ( tab[j] ) {
3318: NEXT(tab[j]) = psum[j]; psum[j] = tab[j];
3319: }
3320: }
3321: }
3322: for ( i = tlen-1, r = 0; i >= 0; i-- )
3323: if ( psum[i] ) {
3324: for ( j = 0, t = psum[i]; t; t = NEXT(t), j++ );
3325: MKND(n,psum[i],j,s);
3326: r = nd_add(mod,r,s);
3327: }
1.56 noro 3328: if ( r ) SG(r) = SG(p)+TD(d0);
3329: return r;
1.55 noro 3330: }
3331:
1.56 noro 3332: /* product of monomials */
3333: /* XXX block order is not handled correctly */
3334:
1.55 noro 3335: void weyl_mul_nm_nmv(int n,int mod,NM m0,NMV m1,NM *tab,int tlen)
3336: {
1.56 noro 3337: int i,n2,j,s,curlen,homo,h,a,b,k,l,u,min;
1.61 noro 3338: UINT *d0,*d1,*d,*dt,*ctab;
1.113 noro 3339: Q *ctab_q;
3340: Q q,q1;
1.61 noro 3341: UINT c0,c1,c;
1.55 noro 3342: NM *p;
3343: NM m,t;
3344:
3345: for ( i = 0; i < tlen; i++ ) tab[i] = 0;
3346: if ( !m0 || !m1 ) return;
3347: d0 = DL(m0); d1 = DL(m1); n2 = n>>1;
3348: NEWNM(m); d = DL(m);
1.56 noro 3349: if ( mod ) {
3350: c0 = CM(m0); c1 = CM(m1); DMAR(c0,c1,0,mod,c); CM(m) = c;
3351: } else
1.113 noro 3352: mulq(CQ(m0),CQ(m1),&CQ(m));
1.55 noro 3353: for ( i = 0; i < nd_wpd; i++ ) d[i] = 0;
3354: homo = n&1 ? 1 : 0;
3355: if ( homo ) {
3356: /* offset of h-degree */
3357: h = GET_EXP(d0,n-1)+GET_EXP(d1,n-1);
3358: PUT_EXP(DL(m),n-1,h);
3359: TD(DL(m)) = h;
1.61 noro 3360: if ( nd_blockmask ) ndl_weight_mask(DL(m));
1.55 noro 3361: }
3362: tab[0] = m;
3363: NEWNM(m); d = DL(m);
1.57 noro 3364: for ( i = 0, curlen = 1; i < n2; i++ ) {
1.55 noro 3365: a = GET_EXP(d0,i); b = GET_EXP(d1,n2+i);
3366: k = GET_EXP(d0,n2+i); l = GET_EXP(d1,i);
3367: /* xi^a*(Di^k*xi^l)*Di^b */
3368: a += l; b += k;
1.56 noro 3369: s = MUL_WEIGHT(a,i)+MUL_WEIGHT(b,n2+i);
1.55 noro 3370: if ( !k || !l ) {
3371: for ( j = 0; j < curlen; j++ )
1.56 noro 3372: if ( t = tab[j] ) {
3373: dt = DL(t);
3374: PUT_EXP(dt,i,a); PUT_EXP(dt,n2+i,b); TD(dt) += s;
1.61 noro 3375: if ( nd_blockmask ) ndl_weight_mask(dt);
1.55 noro 3376: }
3377: curlen *= k+1;
3378: continue;
3379: }
3380: min = MIN(k,l);
1.56 noro 3381: if ( mod ) {
1.61 noro 3382: ctab = (UINT *)ALLOCA((min+1)*sizeof(UINT));
1.56 noro 3383: mkwcm(k,l,mod,ctab);
3384: } else {
1.113 noro 3385: ctab_q = (Q *)ALLOCA((min+1)*sizeof(Q));
3386: mkwc(k,l,ctab_q);
1.56 noro 3387: }
1.57 noro 3388: for ( j = min; j >= 0; j-- ) {
1.56 noro 3389: for ( u = 0; u < nd_wpd; u++ ) d[u] = 0;
1.55 noro 3390: PUT_EXP(d,i,a-j); PUT_EXP(d,n2+i,b-j);
1.56 noro 3391: h = MUL_WEIGHT(a-j,i)+MUL_WEIGHT(b-j,n2+i);
1.55 noro 3392: if ( homo ) {
3393: TD(d) = s;
1.56 noro 3394: PUT_EXP(d,n-1,s-h);
1.55 noro 3395: } else TD(d) = h;
1.61 noro 3396: if ( nd_blockmask ) ndl_weight_mask(d);
1.56 noro 3397: if ( mod ) c = ctab[j];
1.113 noro 3398: else q = ctab_q[j];
1.57 noro 3399: p = tab+curlen*j;
3400: if ( j == 0 ) {
3401: for ( u = 0; u < curlen; u++, p++ ) {
3402: if ( tab[u] ) {
3403: ndl_addto(DL(tab[u]),d);
3404: if ( mod ) {
3405: c0 = CM(tab[u]); DMAR(c0,c,0,mod,c1); CM(tab[u]) = c1;
3406: } else {
1.113 noro 3407: mulq(CQ(tab[u]),q,&q1); CQ(tab[u]) = q1;
1.57 noro 3408: }
3409: }
1.56 noro 3410: }
1.57 noro 3411: } else {
3412: for ( u = 0; u < curlen; u++, p++ ) {
3413: if ( tab[u] ) {
3414: NEWNM(t);
3415: ndl_add(DL(tab[u]),d,DL(t));
3416: if ( mod ) {
3417: c0 = CM(tab[u]); DMAR(c0,c,0,mod,c1); CM(t) = c1;
3418: } else
1.113 noro 3419: mulq(CQ(tab[u]),q,&CQ(t));
1.57 noro 3420: *p = t;
3421: }
1.55 noro 3422: }
3423: }
3424: }
3425: curlen *= k+1;
3426: }
3427: FREENM(m);
3428: }
3429:
1.63 noro 3430: ND ndv_mul_nm_symbolic(NM m0,NDV p)
3431: {
3432: NM mr,mr0;
3433: NMV m;
3434: UINT *d,*dt,*dm;
3435: int c,n,td,i,c1,c2,len;
1.113 noro 3436: Q q;
1.63 noro 3437: ND r;
3438:
3439: if ( !p ) return 0;
3440: else {
3441: n = NV(p); m = BDY(p);
3442: d = DL(m0);
3443: len = LEN(p);
3444: mr0 = 0;
3445: td = TD(d);
3446: c = CM(m0);
3447: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
3448: NEXTNM(mr0,mr);
3449: CM(mr) = 1;
3450: ndl_add(DL(m),d,DL(mr));
3451: }
3452: NEXT(mr) = 0;
3453: MKND(NV(p),mr0,len,r);
3454: SG(r) = SG(p) + TD(d);
3455: return r;
3456: }
3457: }
3458:
1.55 noro 3459: ND ndv_mul_nm(int mod,NM m0,NDV p)
1.9 noro 3460: {
3461: NM mr,mr0;
3462: NMV m;
1.61 noro 3463: UINT *d,*dt,*dm;
1.9 noro 3464: int c,n,td,i,c1,c2,len;
1.113 noro 3465: Q q;
1.9 noro 3466: ND r;
3467:
1.34 noro 3468: if ( !p ) return 0;
1.55 noro 3469: else if ( do_weyl )
1.71 noro 3470: if ( mod == -1 )
3471: error("ndv_mul_nm : not implemented (weyl)");
3472: else
3473: return weyl_ndv_mul_nm(mod,m0,p);
1.9 noro 3474: else {
3475: n = NV(p); m = BDY(p);
1.34 noro 3476: d = DL(m0);
1.14 noro 3477: len = LEN(p);
1.9 noro 3478: mr0 = 0;
1.34 noro 3479: td = TD(d);
1.71 noro 3480: if ( mod == -1 ) {
3481: c = CM(m0);
3482: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
3483: NEXTNM(mr0,mr);
3484: CM(mr) = _mulsf(CM(m),c);
3485: ndl_add(DL(m),d,DL(mr));
3486: }
3487: } else if ( mod ) {
1.16 noro 3488: c = CM(m0);
3489: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
3490: NEXTNM(mr0,mr);
3491: c1 = CM(m);
1.19 noro 3492: DMAR(c1,c,0,mod,c2);
1.16 noro 3493: CM(mr) = c2;
3494: ndl_add(DL(m),d,DL(mr));
3495: }
3496: } else {
1.113 noro 3497: q = CQ(m0);
1.16 noro 3498: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
3499: NEXTNM(mr0,mr);
1.113 noro 3500: mulq(CQ(m),q,&CQ(mr));
1.16 noro 3501: ndl_add(DL(m),d,DL(mr));
3502: }
1.4 noro 3503: }
1.9 noro 3504: NEXT(mr) = 0;
1.31 noro 3505: MKND(NV(p),mr0,len,r);
1.34 noro 3506: SG(r) = SG(p) + TD(d);
1.9 noro 3507: return r;
1.4 noro 3508: }
3509: }
3510:
1.104 noro 3511: ND nd_quo(int mod,PGeoBucket bucket,NDV d)
1.99 noro 3512: {
3513: NM mq0,mq;
1.102 noro 3514: NMV tm;
1.113 noro 3515: Q q;
1.104 noro 3516: int i,nv,sg,c,c1,c2,hindex;
3517: ND p,t,r;
3518: N tnm;
3519:
1.124 noro 3520: if ( bucket->m < 0 ) return 0;
1.99 noro 3521: else {
1.104 noro 3522: nv = NV(d);
1.99 noro 3523: mq0 = 0;
1.102 noro 3524: tm = (NMV)ALLOCA(nmv_adv);
1.104 noro 3525: while ( 1 ) {
1.113 noro 3526: hindex = mod?head_pbucket(mod,bucket):head_pbucket_q(bucket);
1.104 noro 3527: if ( hindex < 0 ) break;
3528: p = bucket->body[hindex];
1.99 noro 3529: NEXTNM(mq0,mq);
3530: ndl_sub(HDL(p),HDL(d),DL(mq));
1.102 noro 3531: ndl_copy(DL(mq),DL(tm));
3532: if ( mod ) {
3533: c1 = invm(HCM(d),mod); c2 = HCM(p);
3534: DMAR(c1,c2,0,mod,c); CM(mq) = c;
3535: CM(tm) = mod-c;
3536: } else {
1.113 noro 3537: divsn(NM(HCQ(p)),NM(HCQ(d)),&tnm);
3538: NTOQ(tnm,SGN(HCQ(p))*SGN(HCQ(d)),CQ(mq));
3539: chsgnq(CQ(mq),&CQ(tm));
1.102 noro 3540: }
3541: t = ndv_mul_nmv_trunc(mod,tm,d,HDL(d));
1.104 noro 3542: bucket->body[hindex] = nd_remove_head(p);
3543: t = nd_remove_head(t);
3544: add_pbucket(mod,bucket,t);
3545: }
3546: if ( !mq0 )
3547: r = 0;
3548: else {
3549: NEXT(mq) = 0;
3550: for ( i = 0, mq = mq0; mq; mq = NEXT(mq), i++ );
3551: MKND(nv,mq0,i,r);
3552: /* XXX */
3553: SG(r) = HTD(r);
1.99 noro 3554: }
3555: return r;
3556: }
3557: }
3558:
1.43 noro 3559: void ndv_realloc(NDV p,int obpe,int oadv,EPOS oepos)
1.11 noro 3560: {
1.13 noro 3561: NMV m,mr,mr0,t;
3562: int len,i,k;
1.11 noro 3563:
1.61 noro 3564: if ( !p ) return;
3565: m = BDY(p); len = LEN(p);
3566: mr0 = nmv_adv>oadv?(NMV)REALLOC(BDY(p),len*nmv_adv):BDY(p);
3567: m = (NMV)((char *)mr0+(len-1)*oadv);
3568: mr = (NMV)((char *)mr0+(len-1)*nmv_adv);
3569: t = (NMV)ALLOCA(nmv_adv);
3570: for ( i = 0; i < len; i++, NMV_OPREV(m), NMV_PREV(mr) ) {
1.113 noro 3571: CQ(t) = CQ(m);
1.61 noro 3572: for ( k = 0; k < nd_wpd; k++ ) DL(t)[k] = 0;
3573: ndl_reconstruct(DL(m),DL(t),obpe,oepos);
1.113 noro 3574: CQ(mr) = CQ(t);
1.61 noro 3575: ndl_copy(DL(t),DL(mr));
3576: }
3577: BDY(p) = mr0;
3578: }
3579:
3580: NDV ndv_dup_realloc(NDV p,int obpe,int oadv,EPOS oepos)
3581: {
3582: NMV m,mr,mr0;
3583: int len,i;
3584: NDV r;
1.11 noro 3585:
1.61 noro 3586: if ( !p ) return 0;
3587: m = BDY(p); len = LEN(p);
3588: mr0 = mr = (NMV)MALLOC(len*nmv_adv);
3589: for ( i = 0; i < len; i++, NMV_OADV(m), NMV_ADV(mr) ) {
3590: ndl_zero(DL(mr));
3591: ndl_reconstruct(DL(m),DL(mr),obpe,oepos);
1.113 noro 3592: CQ(mr) = CQ(m);
1.11 noro 3593: }
1.61 noro 3594: MKNDV(NV(p),mr0,len,r);
3595: SG(r) = SG(p);
3596: return r;
1.11 noro 3597: }
3598:
1.61 noro 3599: /* duplicate p */
3600:
3601: NDV ndv_dup(int mod,NDV p)
1.3 noro 3602: {
3603: NDV d;
1.61 noro 3604: NMV t,m,m0;
1.3 noro 3605: int i,len;
3606:
1.34 noro 3607: if ( !p ) return 0;
1.31 noro 3608: len = LEN(p);
1.34 noro 3609: m0 = m = (NMV)(mod?MALLOC_ATOMIC(len*nmv_adv):MALLOC(len*nmv_adv));
1.61 noro 3610: for ( t = BDY(p), i = 0; i < len; i++, NMV_ADV(t), NMV_ADV(m) ) {
1.14 noro 3611: ndl_copy(DL(t),DL(m));
1.113 noro 3612: CQ(m) = CQ(t);
1.3 noro 3613: }
3614: MKNDV(NV(p),m0,len,d);
1.23 noro 3615: SG(d) = SG(p);
3616: return d;
3617: }
3618:
1.63 noro 3619: ND nd_dup(ND p)
3620: {
3621: ND d;
3622: NM t,m,m0;
3623:
3624: if ( !p ) return 0;
3625: for ( m0 = 0, t = BDY(p); t; t = NEXT(t) ) {
3626: NEXTNM(m0,m);
3627: ndl_copy(DL(t),DL(m));
1.113 noro 3628: CQ(m) = CQ(t);
1.63 noro 3629: }
3630: if ( m0 ) NEXT(m) = 0;
3631: MKND(NV(p),m0,LEN(p),d);
3632: SG(d) = SG(p);
3633: return d;
3634: }
3635:
1.61 noro 3636: /* XXX if p->len == 0 then it represents 0 */
3637:
3638: void ndv_mod(int mod,NDV p)
3639: {
3640: NMV t,d;
1.125 noro 3641: int r,s,u;
1.61 noro 3642: int i,len,dlen;
1.71 noro 3643: Obj gfs;
1.61 noro 3644:
3645: if ( !p ) return;
3646: len = LEN(p);
3647: dlen = 0;
1.71 noro 3648: if ( mod == -1 )
3649: for ( t = d = BDY(p), i = 0; i < len; i++, NMV_ADV(t) ) {
3650: simp_ff((Obj)CP(t),&gfs);
3651: r = FTOIF(CONT((GFS)gfs));
1.61 noro 3652: CM(d) = r;
3653: ndl_copy(DL(t),DL(d));
3654: NMV_ADV(d);
3655: dlen++;
3656: }
1.71 noro 3657: else
3658: for ( t = d = BDY(p), i = 0; i < len; i++, NMV_ADV(t) ) {
1.113 noro 3659: r = rem(NM(CQ(t)),mod);
1.71 noro 3660: if ( r ) {
1.113 noro 3661: if ( SGN(CQ(t)) < 0 )
3662: r = mod-r;
1.125 noro 3663: if ( DN(CQ(t)) ) {
3664: s = rem(DN(CQ(t)),mod);
3665: if ( !s )
3666: error("ndv_mod : division by 0");
3667: s = invm(s,mod);
3668: DMAR(r,s,0,mod,u); r = u;
3669: }
1.71 noro 3670: CM(d) = r;
3671: ndl_copy(DL(t),DL(d));
3672: NMV_ADV(d);
3673: dlen++;
3674: }
3675: }
1.61 noro 3676: LEN(p) = dlen;
3677: }
3678:
3679: NDV ptondv(VL vl,VL dvl,P p)
3680: {
3681: ND nd;
3682:
3683: nd = ptond(vl,dvl,p);
3684: return ndtondv(0,nd);
3685: }
3686:
3687: ND ptond(VL vl,VL dvl,P p)
1.23 noro 3688: {
1.61 noro 3689: int n,i,j,k,e;
3690: VL tvl;
3691: V v;
3692: DCP dc;
3693: DCP *w;
3694: ND r,s,t,u;
3695: P x;
3696: int c;
3697: UINT *d;
1.23 noro 3698: NM m,m0;
1.61 noro 3699:
3700: if ( !p )
3701: return 0;
3702: else if ( NUM(p) ) {
3703: NEWNM(m);
3704: ndl_zero(DL(m));
1.113 noro 3705: CQ(m) = (Q)p;
1.61 noro 3706: NEXT(m) = 0;
3707: MKND(nd_nvar,m,1,r);
3708: SG(r) = 0;
3709: return r;
3710: } else {
3711: for ( dc = DC(p), k = 0; dc; dc = NEXT(dc), k++ );
3712: w = (DCP *)ALLOCA(k*sizeof(DCP));
3713: for ( dc = DC(p), j = 0; j < k; dc = NEXT(dc), j++ ) w[j] = dc;
3714: for ( i = 0, tvl = dvl, v = VR(p);
1.98 noro 3715: tvl && tvl->v != v; tvl = NEXT(tvl), i++ );
1.61 noro 3716: if ( !tvl ) {
3717: for ( j = k-1, s = 0, MKV(v,x); j >= 0; j-- ) {
3718: t = ptond(vl,dvl,COEF(w[j]));
3719: pwrp(vl,x,DEG(w[j]),&p);
3720: nd_mul_c_p(CO,t,p); s = nd_add(0,s,t);
3721: }
3722: return s;
3723: } else {
3724: NEWNM(m0); d = DL(m0);
3725: for ( j = k-1, s = 0; j >= 0; j-- ) {
3726: ndl_zero(d); e = QTOS(DEG(w[j])); PUT_EXP(d,i,e);
3727: TD(d) = MUL_WEIGHT(e,i);
3728: if ( nd_blockmask) ndl_weight_mask(d);
3729: t = ptond(vl,dvl,COEF(w[j]));
3730: for ( m = BDY(t); m; m = NEXT(m) )
3731: ndl_addto(DL(m),d);
3732: SG(t) += TD(d);
3733: s = nd_add(0,s,t);
3734: }
3735: FREENM(m0);
3736: return s;
3737: }
3738: }
3739: }
3740:
3741: P ndvtop(int mod,VL vl,VL dvl,NDV p)
3742: {
3743: VL tvl;
3744: int len,n,j,i,e;
3745: NMV m;
3746: Q q;
3747: P c;
3748: UINT *d;
3749: P s,r,u,t,w;
1.71 noro 3750: GFS gfs;
1.23 noro 3751:
1.34 noro 3752: if ( !p ) return 0;
1.61 noro 3753: else {
3754: len = LEN(p);
3755: n = NV(p);
3756: m = (NMV)(((char *)BDY(p))+nmv_adv*(len-1));
3757: for ( j = len-1, s = 0; j >= 0; j--, NMV_PREV(m) ) {
1.71 noro 3758: if ( mod == -1 ) {
3759: e = IFTOF(CM(m)); MKGFS(e,gfs); c = (P)gfs;
3760: } else if ( mod ) {
1.61 noro 3761: STOQ(CM(m),q); c = (P)q;
3762: } else
1.113 noro 3763: c = CP(m);
1.61 noro 3764: d = DL(m);
3765: for ( i = 0, t = c, tvl = dvl; i < n; tvl = NEXT(tvl), i++ ) {
3766: MKV(tvl->v,r); e = GET_EXP(d,i); STOQ(e,q);
3767: pwrp(vl,r,q,&u); mulp(vl,t,u,&w); t = w;
3768: }
3769: addp(vl,s,t,&u); s = u;
3770: }
3771: return s;
1.23 noro 3772: }
1.3 noro 3773: }
3774:
1.61 noro 3775: NDV ndtondv(int mod,ND p)
1.11 noro 3776: {
3777: NDV d;
1.61 noro 3778: NMV m,m0;
3779: NM t;
3780: int i,len;
1.11 noro 3781:
1.34 noro 3782: if ( !p ) return 0;
1.61 noro 3783: len = LEN(p);
1.74 noro 3784: if ( mod )
3785: m0 = m = (NMV)GC_malloc_atomic_ignore_off_page(len*nmv_adv);
3786: else
3787: m0 = m = MALLOC(len*nmv_adv);
1.103 noro 3788: #if 0
1.74 noro 3789: ndv_alloc += nmv_adv*len;
1.103 noro 3790: #endif
1.61 noro 3791: for ( t = BDY(p), i = 0; t; t = NEXT(t), i++, NMV_ADV(m) ) {
3792: ndl_copy(DL(t),DL(m));
1.113 noro 3793: CQ(m) = CQ(t);
1.11 noro 3794: }
1.61 noro 3795: MKNDV(NV(p),m0,len,d);
1.14 noro 3796: SG(d) = SG(p);
1.11 noro 3797: return d;
3798: }
3799:
1.61 noro 3800: ND ndvtond(int mod,NDV p)
1.11 noro 3801: {
1.61 noro 3802: ND d;
3803: NM m,m0;
1.11 noro 3804: NMV t;
1.61 noro 3805: int i,len;
1.11 noro 3806:
1.34 noro 3807: if ( !p ) return 0;
1.11 noro 3808: m0 = 0;
1.61 noro 3809: len = p->len;
3810: for ( t = BDY(p), i = 0; i < len; NMV_ADV(t), i++ ) {
3811: NEXTNM(m0,m);
3812: ndl_copy(DL(t),DL(m));
1.113 noro 3813: CQ(m) = CQ(t);
1.11 noro 3814: }
3815: NEXT(m) = 0;
1.61 noro 3816: MKND(NV(p),m0,len,d);
1.14 noro 3817: SG(d) = SG(p);
1.11 noro 3818: return d;
3819: }
3820:
1.3 noro 3821: void ndv_print(NDV p)
3822: {
3823: NMV m;
3824: int i,len;
3825:
1.34 noro 3826: if ( !p ) printf("0\n");
1.3 noro 3827: else {
1.14 noro 3828: len = LEN(p);
1.3 noro 3829: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
1.71 noro 3830: if ( CM(m) & 0x80000000 ) printf("+@_%d*",IFTOF(CM(m)));
3831: else printf("+%d*",CM(m));
1.16 noro 3832: ndl_print(DL(m));
3833: }
3834: printf("\n");
3835: }
3836: }
3837:
1.113 noro 3838: void ndv_print_q(NDV p)
1.16 noro 3839: {
3840: NMV m;
3841: int i,len;
3842:
1.34 noro 3843: if ( !p ) printf("0\n");
1.16 noro 3844: else {
3845: len = LEN(p);
3846: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
3847: printf("+");
1.113 noro 3848: printexpr(CO,(Obj)CQ(m));
1.16 noro 3849: printf("*");
1.14 noro 3850: ndl_print(DL(m));
1.3 noro 3851: }
3852: printf("\n");
3853: }
1.25 noro 3854: }
3855:
1.61 noro 3856: NODE ndv_reducebase(NODE x)
1.27 noro 3857: {
3858: int len,i,j;
3859: NDV *w;
3860: NODE t,t0;
3861:
3862: len = length(x);
3863: w = (NDV *)ALLOCA(len*sizeof(NDV));
3864: for ( i = 0, t = x; i < len; i++, t = NEXT(t) ) w[i] = BDY(t);
3865: for ( i = 0; i < len; i++ ) {
3866: for ( j = 0; j < i; j++ ) {
3867: if ( w[i] && w[j] )
3868: if ( ndl_reducible(HDL(w[i]),HDL(w[j])) ) w[i] = 0;
3869: else if ( ndl_reducible(HDL(w[j]),HDL(w[i])) ) w[j] = 0;
3870: }
3871: }
3872: for ( i = len-1, t0 = 0; i >= 0; i-- ) {
3873: if ( w[i] ) { NEXTNODE(t0,t); BDY(t) = (pointer)w[i]; }
3874: }
3875: NEXT(t) = 0; x = t0;
3876: return x;
1.11 noro 3877: }
1.32 noro 3878:
1.43 noro 3879: /* XXX incomplete */
3880:
1.32 noro 3881: void nd_init_ord(struct order_spec *ord)
3882: {
1.43 noro 3883: switch ( ord->id ) {
1.32 noro 3884: case 0:
1.43 noro 3885: switch ( ord->ord.simple ) {
3886: case 0:
3887: nd_dcomp = 1;
3888: nd_isrlex = 1;
3889: break;
3890: case 1:
3891: nd_dcomp = 1;
3892: nd_isrlex = 0;
3893: break;
3894: case 2:
3895: nd_dcomp = 0;
3896: nd_isrlex = 0;
1.45 noro 3897: ndl_compare_function = ndl_lex_compare;
1.58 noro 3898: break;
3899: case 11:
3900: /* XXX */
3901: nd_dcomp = 0;
3902: nd_isrlex = 1;
3903: ndl_compare_function = ndl_ww_lex_compare;
1.43 noro 3904: break;
3905: default:
3906: error("nd_gr : unsupported order");
3907: }
1.32 noro 3908: break;
3909: case 1:
1.96 noro 3910: /* block order */
1.43 noro 3911: /* XXX */
3912: nd_dcomp = -1;
1.32 noro 3913: nd_isrlex = 0;
1.45 noro 3914: ndl_compare_function = ndl_block_compare;
1.34 noro 3915: break;
1.43 noro 3916: case 2:
1.96 noro 3917: /* matrix order */
3918: /* XXX */
3919: nd_dcomp = -1;
3920: nd_isrlex = 0;
3921: nd_matrix_len = ord->ord.matrix.row;
3922: nd_matrix = ord->ord.matrix.matrix;
3923: ndl_compare_function = ndl_matrix_compare;
3924: break;
3925: case 3:
1.97 noro 3926: /* composite order */
3927: nd_dcomp = -1;
3928: nd_isrlex = 0;
3929: nd_worb_len = ord->ord.composite.length;
3930: nd_worb = ord->ord.composite.w_or_b;
3931: ndl_compare_function = ndl_composite_compare;
1.32 noro 3932: break;
3933: }
1.41 noro 3934: nd_ord = ord;
1.32 noro 3935: }
3936:
1.43 noro 3937: BlockMask nd_create_blockmask(struct order_spec *ord)
3938: {
3939: int n,i,j,s,l;
1.61 noro 3940: UINT *t;
1.43 noro 3941: BlockMask bm;
3942:
1.96 noro 3943: /* we only create mask table for block order */
3944: if ( ord->id != 1 )
1.43 noro 3945: return 0;
3946: n = ord->ord.block.length;
3947: bm = (BlockMask)MALLOC(sizeof(struct oBlockMask));
3948: bm->n = n;
3949: bm->order_pair = ord->ord.block.order_pair;
1.61 noro 3950: bm->mask = (UINT **)MALLOC(n*sizeof(UINT *));
1.43 noro 3951: for ( i = 0, s = 0; i < n; i++ ) {
1.61 noro 3952: bm->mask[i] = t = (UINT *)MALLOC_ATOMIC(nd_wpd*sizeof(UINT));
1.43 noro 3953: for ( j = 0; j < nd_wpd; j++ ) t[j] = 0;
3954: l = bm->order_pair[i].length;
3955: for ( j = 0; j < l; j++, s++ ) PUT_EXP(t,s,nd_mask0);
3956: }
3957: return bm;
1.57 noro 3958: }
3959:
3960: EPOS nd_create_epos(struct order_spec *ord)
3961: {
3962: int i,j,l,s,ord_l,ord_o;
3963: EPOS epos;
3964: struct order_pair *op;
3965:
3966: epos = (EPOS)MALLOC_ATOMIC(nd_nvar*sizeof(struct oEPOS));
3967: switch ( ord->id ) {
3968: case 0:
3969: if ( nd_isrlex ) {
3970: for ( i = 0; i < nd_nvar; i++ ) {
3971: epos[i].i = nd_exporigin + (nd_nvar-1-i)/nd_epw;
3972: epos[i].s = (nd_epw-((nd_nvar-1-i)%nd_epw)-1)*nd_bpe;
3973: }
3974: } else {
3975: for ( i = 0; i < nd_nvar; i++ ) {
3976: epos[i].i = nd_exporigin + i/nd_epw;
3977: epos[i].s = (nd_epw-(i%nd_epw)-1)*nd_bpe;
3978: }
3979: }
3980: break;
3981: case 1:
3982: /* block order */
3983: l = ord->ord.block.length;
3984: op = ord->ord.block.order_pair;
3985: for ( j = 0, s = 0; j < l; j++ ) {
3986: ord_o = op[j].order;
3987: ord_l = op[j].length;
3988: if ( !ord_o )
3989: for ( i = 0; i < ord_l; i++ ) {
3990: epos[s+i].i = nd_exporigin + (s+ord_l-i-1)/nd_epw;
3991: epos[s+i].s = (nd_epw-((s+ord_l-i-1)%nd_epw)-1)*nd_bpe;
3992: }
3993: else
3994: for ( i = 0; i < ord_l; i++ ) {
3995: epos[s+i].i = nd_exporigin + (s+i)/nd_epw;
3996: epos[s+i].s = (nd_epw-((s+i)%nd_epw)-1)*nd_bpe;
3997: }
3998: s += ord_l;
3999: }
4000: break;
4001: case 2:
1.96 noro 4002: /* matrix order */
4003: case 3:
4004: /* composite order */
4005: for ( i = 0; i < nd_nvar; i++ ) {
4006: epos[i].i = nd_exporigin + i/nd_epw;
4007: epos[i].s = (nd_epw-(i%nd_epw)-1)*nd_bpe;
4008: }
4009: break;
1.57 noro 4010: }
4011: return epos;
1.43 noro 4012: }
1.59 noro 4013:
4014: /* external interface */
4015:
4016: void nd_nf_p(P f,LIST g,LIST v,int m,struct order_spec *ord,P *rp)
4017: {
1.61 noro 4018: NODE t,in0,in;
1.59 noro 4019: ND nd,nf;
1.61 noro 4020: NDV ndv;
4021: VL vv,tv;
4022: int stat,nvar,max,e;
1.113 noro 4023: union oNDC dn;
1.59 noro 4024:
1.116 noro 4025: if ( !f ) {
4026: *rp = 0;
4027: return;
4028: }
1.59 noro 4029: pltovl(v,&vv);
1.74 noro 4030: for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++ );
1.61 noro 4031:
4032: /* get the degree bound */
4033: for ( t = BDY(g), max = 0; t; t = NEXT(t) )
4034: for ( tv = vv; tv; tv = NEXT(tv) ) {
4035: e = getdeg(tv->v,(P)BDY(t));
4036: max = MAX(e,max);
4037: }
4038: for ( tv = vv; tv; tv = NEXT(tv) ) {
4039: e = getdeg(tv->v,f);
4040: max = MAX(e,max);
4041: }
4042:
1.59 noro 4043: nd_init_ord(ord);
1.61 noro 4044: nd_setup_parameters(nvar,max);
4045:
4046: /* conversion to ndv */
4047: for ( in0 = 0, t = BDY(g); t; t = NEXT(t) ) {
4048: NEXTNODE(in0,in);
4049: BDY(in) = (pointer)ptondv(CO,vv,(P)BDY(t));
1.69 noro 4050: if ( m ) ndv_mod(m,(NDV)BDY(in));
1.61 noro 4051: }
4052: NEXTNODE(in0,in);
4053: BDY(in) = (pointer)ptondv(CO,vv,f);
1.69 noro 4054: if ( m ) ndv_mod(m,(NDV)BDY(in));
1.61 noro 4055: NEXT(in) = 0;
4056:
1.128 noro 4057: /* dont sort */
4058: ndv_setup(m,0,in0,1);
1.59 noro 4059: nd_psn--;
4060: nd_scale=2;
4061: while ( 1 ) {
4062: nd = (pointer)ndvtond(m,nd_ps[nd_psn]);
1.69 noro 4063: stat = nd_nf(m,nd,nd_ps,1,0,&nf);
1.59 noro 4064: if ( !stat ) {
4065: nd_psn++;
1.103 noro 4066: nd_reconstruct(0,0);
1.59 noro 4067: nd_psn--;
4068: } else
4069: break;
4070: }
1.61 noro 4071: *rp = ndvtop(m,CO,vv,ndtondv(m,nf));
1.63 noro 4072: }
4073:
4074: int nd_to_vect(int mod,UINT *s0,int n,ND d,UINT *r)
4075: {
4076: NM m;
4077: UINT *t,*s;
4078: int i;
4079:
4080: for ( i = 0; i < n; i++ ) r[i] = 0;
4081: for ( i = 0, s = s0, m = BDY(d); m; m = NEXT(m) ) {
4082: t = DL(m);
4083: for ( ; !ndl_equal(t,s); s += nd_wpd, i++ );
4084: r[i] = CM(m);
4085: }
4086: for ( i = 0; !r[i]; i++ );
4087: return i;
4088: }
4089:
1.113 noro 4090: int nd_to_vect_q(UINT *s0,int n,ND d,Q *r)
1.74 noro 4091: {
1.107 noro 4092: NM m;
1.74 noro 4093: UINT *t,*s;
1.107 noro 4094: int i;
1.74 noro 4095:
4096: for ( i = 0; i < n; i++ ) r[i] = 0;
1.107 noro 4097: for ( i = 0, s = s0, m = BDY(d); m; m = NEXT(m) ) {
1.74 noro 4098: t = DL(m);
4099: for ( ; !ndl_equal(t,s); s += nd_wpd, i++ );
1.113 noro 4100: r[i] = CQ(m);
1.74 noro 4101: }
4102: for ( i = 0; !r[i]; i++ );
4103: return i;
4104: }
4105:
1.129 noro 4106: Q *nm_ind_pair_to_vect(int mod,UINT *s0,int n,NM_ind_pair pair)
4107: {
4108: NM m;
4109: NMV mr;
4110: UINT *d,*t,*s;
4111: NDV p;
4112: int i,j,len;
4113: Q *r;
4114:
4115: m = pair->mul;
4116: d = DL(m);
4117: p = nd_ps[pair->index];
4118: len = LEN(p);
4119: r = (Q *)CALLOC(n,sizeof(Q));
4120: t = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
4121: for ( i = j = 0, s = s0, mr = BDY(p); j < len; j++, NMV_ADV(mr) ) {
4122: ndl_add(d,DL(mr),t);
4123: for ( ; !ndl_equal(t,s); s += nd_wpd, i++ );
4124: r[i] = CQ(mr);
4125: }
4126: return r;
4127: }
4128:
1.67 noro 4129: IndArray nm_ind_pair_to_vect_compress(int mod,UINT *s0,int n,NM_ind_pair pair)
1.64 noro 4130: {
4131: NM m;
4132: NMV mr;
4133: UINT *d,*t,*s;
4134: NDV p;
1.67 noro 4135: unsigned char *ivc;
4136: unsigned short *ivs;
1.81 noro 4137: UINT *v,*ivi,*s0v;
1.67 noro 4138: int i,j,len,prev,diff,cdiff;
4139: IndArray r;
1.64 noro 4140:
4141: m = pair->mul;
4142: d = DL(m);
4143: p = nd_ps[pair->index];
4144: len = LEN(p);
4145: t = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
1.67 noro 4146: v = (unsigned int *)ALLOCA(len*sizeof(unsigned int));
1.64 noro 4147: for ( i = j = 0, s = s0, mr = BDY(p); j < len; j++, NMV_ADV(mr) ) {
4148: ndl_add(d,DL(mr),t);
4149: for ( ; !ndl_equal(t,s); s += nd_wpd, i++ );
1.67 noro 4150: v[j] = i;
4151: }
1.81 noro 4152: r = (IndArray)MALLOC(sizeof(struct oIndArray));
1.67 noro 4153: r->head = v[0];
4154: diff = 0;
4155: for ( i = 1; i < len; i++ ) {
4156: cdiff = v[i]-v[i-1]; diff = MAX(cdiff,diff);
4157: }
4158: if ( diff < 256 ) {
4159: r->width = 1;
4160: ivc = (unsigned char *)MALLOC_ATOMIC(len*sizeof(unsigned char));
4161: r->index.c = ivc;
4162: for ( i = 1, ivc[0] = 0; i < len; i++ ) ivc[i] = v[i]-v[i-1];
4163: } else if ( diff < 65536 ) {
4164: r->width = 2;
4165: ivs = (unsigned short *)MALLOC_ATOMIC(len*sizeof(unsigned short));
4166: r->index.s = ivs;
4167: for ( i = 1, ivs[0] = 0; i < len; i++ ) ivs[i] = v[i]-v[i-1];
4168: } else {
4169: r->width = 4;
4170: ivi = (unsigned int *)MALLOC_ATOMIC(len*sizeof(unsigned int));
4171: r->index.i = ivi;
4172: for ( i = 1, ivi[0] = 0; i < len; i++ ) ivi[i] = v[i]-v[i-1];
1.64 noro 4173: }
1.67 noro 4174: return r;
1.64 noro 4175: }
4176:
1.135 noro 4177: int compress_array(Q *svect,Q *cvect,int n)
4178: {
4179: int i,j;
4180:
4181: for ( i = j = 0; i < n; i++ )
4182: if ( svect[i] ) cvect[j++] = svect[i];
4183: return j;
4184: }
4185:
4186: void expand_array(Q *svect,Q *cvect,int n)
4187: {
4188: int i,j;
4189:
4190: for ( i = j = 0; j < n; i++ )
4191: if ( svect[i] ) svect[i] = cvect[j++];
4192: }
4193:
1.133 noro 4194: int ndv_reduce_vect_q(Q *svect,int trace,int col,IndArray *imat,NM_ind_pair *rp0,int nred)
1.107 noro 4195: {
1.135 noro 4196: int i,j,k,len,pos,prev,nz;
1.113 noro 4197: Q cs,mcs,c1,c2,cr,gcd,t;
1.107 noro 4198: IndArray ivect;
4199: unsigned char *ivc;
4200: unsigned short *ivs;
4201: unsigned int *ivi;
4202: NDV redv;
4203: NMV mr;
4204: NODE rp;
4205: int maxrs;
1.133 noro 4206: double hmag;
1.135 noro 4207: Q *cvect;
1.107 noro 4208:
4209: maxrs = 0;
1.134 noro 4210: for ( i = 0; i < col && !svect[i]; i++ );
4211: if ( i == col ) return maxrs;
4212: hmag = p_mag((P)svect[i])*nd_scale;
1.135 noro 4213: cvect = (Q *)ALLOCA(col*sizeof(Q));
1.107 noro 4214: for ( i = 0; i < nred; i++ ) {
4215: ivect = imat[i];
4216: k = ivect->head;
4217: if ( svect[k] ) {
4218: maxrs = MAX(maxrs,rp0[i]->sugar);
1.133 noro 4219: redv = trace?nd_ps_trace[rp0[i]->index]:nd_ps[rp0[i]->index];
1.107 noro 4220: len = LEN(redv); mr = BDY(redv);
1.113 noro 4221: igcd_cofactor(svect[k],CQ(mr),&gcd,&cs,&cr);
4222: chsgnq(cs,&mcs);
4223: if ( !UNIQ(cr) ) {
4224: for ( j = 0; j < col; j++ ) {
4225: mulq(svect[j],cr,&c1); svect[j] = c1;
4226: }
4227: }
1.107 noro 4228: svect[k] = 0; prev = k;
4229: switch ( ivect->width ) {
4230: case 1:
4231: ivc = ivect->index.c;
4232: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
4233: pos = prev+ivc[j]; prev = pos;
1.113 noro 4234: mulq(CQ(mr),mcs,&c2); addq(svect[pos],c2,&t); svect[pos] = t;
1.107 noro 4235: }
4236: break;
4237: case 2:
4238: ivs = ivect->index.s;
4239: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
4240: pos = prev+ivs[j]; prev = pos;
1.113 noro 4241: mulq(CQ(mr),mcs,&c2); addq(svect[pos],c2,&t); svect[pos] = t;
1.107 noro 4242: }
4243: break;
4244: case 4:
4245: ivi = ivect->index.i;
4246: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
4247: pos = prev+ivi[j]; prev = pos;
1.113 noro 4248: mulq(CQ(mr),mcs,&c2); addq(svect[pos],c2,&t); svect[pos] = t;
1.107 noro 4249: }
4250: break;
4251: }
1.134 noro 4252: for ( j = k+1; j < col && !svect[j]; j++ );
4253: if ( j == col ) break;
4254: if ( hmag && ((double)p_mag((P)svect[j]) > hmag) ) {
1.135 noro 4255: nz = compress_array(svect,cvect,col);
4256: removecont_array(cvect,nz);
4257: expand_array(svect,cvect,nz);
1.134 noro 4258: hmag = ((double)p_mag((P)svect[j]))*nd_scale;
4259: }
1.107 noro 4260: }
1.133 noro 4261: }
1.135 noro 4262: nz = compress_array(svect,cvect,col);
4263: removecont_array(cvect,nz);
4264: expand_array(svect,cvect,nz);
1.133 noro 4265: if ( DP_Print ) {
4266: fprintf(asir_out,"-"); fflush(asir_out);
1.107 noro 4267: }
4268: return maxrs;
4269: }
4270:
1.76 noro 4271: int ndv_reduce_vect(int m,UINT *svect,int col,IndArray *imat,NM_ind_pair *rp0,int nred)
1.65 noro 4272: {
1.67 noro 4273: int i,j,k,len,pos,prev;
1.66 noro 4274: UINT c,c1,c2,c3,up,lo,dmy;
1.67 noro 4275: IndArray ivect;
4276: unsigned char *ivc;
4277: unsigned short *ivs;
4278: unsigned int *ivi;
1.65 noro 4279: NDV redv;
1.67 noro 4280: NMV mr;
1.65 noro 4281: NODE rp;
1.76 noro 4282: int maxrs;
1.65 noro 4283:
1.76 noro 4284: maxrs = 0;
1.74 noro 4285: for ( i = 0; i < nred; i++ ) {
1.65 noro 4286: ivect = imat[i];
1.67 noro 4287: k = ivect->head; svect[k] %= m;
1.65 noro 4288: if ( c = svect[k] ) {
1.76 noro 4289: maxrs = MAX(maxrs,rp0[i]->sugar);
1.74 noro 4290: c = m-c; redv = nd_ps[rp0[i]->index];
1.67 noro 4291: len = LEN(redv); mr = BDY(redv);
4292: svect[k] = 0; prev = k;
4293: switch ( ivect->width ) {
4294: case 1:
4295: ivc = ivect->index.c;
4296: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
4297: pos = prev+ivc[j]; c1 = CM(mr); c2 = svect[pos];
4298: prev = pos;
4299: DMA(c1,c,c2,up,lo);
4300: if ( up ) { DSAB(m,up,lo,dmy,c3); svect[pos] = c3;
4301: } else svect[pos] = lo;
4302: }
4303: break;
4304: case 2:
4305: ivs = ivect->index.s;
4306: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
4307: pos = prev+ivs[j]; c1 = CM(mr); c2 = svect[pos];
4308: prev = pos;
4309: DMA(c1,c,c2,up,lo);
4310: if ( up ) { DSAB(m,up,lo,dmy,c3); svect[pos] = c3;
4311: } else svect[pos] = lo;
4312: }
4313: break;
4314: case 4:
4315: ivi = ivect->index.i;
4316: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
4317: pos = prev+ivi[j]; c1 = CM(mr); c2 = svect[pos];
4318: prev = pos;
4319: DMA(c1,c,c2,up,lo);
4320: if ( up ) { DSAB(m,up,lo,dmy,c3); svect[pos] = c3;
4321: } else svect[pos] = lo;
4322: }
4323: break;
1.65 noro 4324: }
4325: }
4326: }
1.66 noro 4327: for ( i = 0; i < col; i++ )
4328: if ( svect[i] >= (UINT)m ) svect[i] %= m;
1.76 noro 4329: return maxrs;
1.65 noro 4330: }
4331:
1.76 noro 4332: int ndv_reduce_vect_sf(int m,UINT *svect,int col,IndArray *imat,NM_ind_pair *rp0,int nred)
1.72 noro 4333: {
4334: int i,j,k,len,pos,prev;
4335: UINT c,c1,c2,c3,up,lo,dmy;
4336: IndArray ivect;
4337: unsigned char *ivc;
4338: unsigned short *ivs;
4339: unsigned int *ivi;
4340: NDV redv;
4341: NMV mr;
4342: NODE rp;
1.76 noro 4343: int maxrs;
1.72 noro 4344:
1.76 noro 4345: maxrs = 0;
1.74 noro 4346: for ( i = 0; i < nred; i++ ) {
1.72 noro 4347: ivect = imat[i];
4348: k = ivect->head; svect[k] %= m;
4349: if ( c = svect[k] ) {
1.76 noro 4350: maxrs = MAX(maxrs,rp0[i]->sugar);
1.74 noro 4351: c = _chsgnsf(c); redv = nd_ps[rp0[i]->index];
1.72 noro 4352: len = LEN(redv); mr = BDY(redv);
4353: svect[k] = 0; prev = k;
4354: switch ( ivect->width ) {
4355: case 1:
4356: ivc = ivect->index.c;
4357: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
4358: pos = prev+ivc[j]; prev = pos;
4359: svect[pos] = _addsf(_mulsf(CM(mr),c),svect[pos]);
4360: }
4361: break;
4362: case 2:
4363: ivs = ivect->index.s;
4364: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
4365: pos = prev+ivs[j]; prev = pos;
4366: svect[pos] = _addsf(_mulsf(CM(mr),c),svect[pos]);
4367: }
4368: break;
4369: case 4:
4370: ivi = ivect->index.i;
4371: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
4372: pos = prev+ivi[j]; prev = pos;
4373: svect[pos] = _addsf(_mulsf(CM(mr),c),svect[pos]);
4374: }
4375: break;
4376: }
4377: }
4378: }
1.76 noro 4379: return maxrs;
1.72 noro 4380: }
4381:
1.65 noro 4382: NDV vect_to_ndv(UINT *vect,int spcol,int col,int *rhead,UINT *s0vect)
4383: {
4384: int j,k,len;
4385: UINT *p;
4386: UINT c;
4387: NDV r;
4388: NMV mr0,mr;
4389:
4390: for ( j = 0, len = 0; j < spcol; j++ ) if ( vect[j] ) len++;
4391: if ( !len ) return 0;
4392: else {
1.74 noro 4393: mr0 = (NMV)GC_malloc_atomic_ignore_off_page(nmv_adv*len);
1.103 noro 4394: #if 0
1.74 noro 4395: ndv_alloc += nmv_adv*len;
1.103 noro 4396: #endif
1.65 noro 4397: mr = mr0;
4398: p = s0vect;
4399: for ( j = k = 0; j < col; j++, p += nd_wpd )
4400: if ( !rhead[j] ) {
4401: if ( c = vect[k++] ) {
4402: ndl_copy(p,DL(mr)); CM(mr) = c; NMV_ADV(mr);
4403: }
4404: }
4405: MKNDV(nd_nvar,mr0,len,r);
4406: return r;
4407: }
4408: }
4409:
1.129 noro 4410: /* for preprocessed vector */
4411:
1.113 noro 4412: NDV vect_to_ndv_q(Q *vect,int spcol,int col,int *rhead,UINT *s0vect)
1.107 noro 4413: {
4414: int j,k,len;
4415: UINT *p;
1.113 noro 4416: Q c;
1.107 noro 4417: NDV r;
4418: NMV mr0,mr;
4419:
4420: for ( j = 0, len = 0; j < spcol; j++ ) if ( vect[j] ) len++;
4421: if ( !len ) return 0;
4422: else {
1.109 noro 4423: mr0 = (NMV)GC_malloc(nmv_adv*len);
1.107 noro 4424: #if 0
4425: ndv_alloc += nmv_adv*len;
4426: #endif
4427: mr = mr0;
4428: p = s0vect;
4429: for ( j = k = 0; j < col; j++, p += nd_wpd )
4430: if ( !rhead[j] ) {
4431: if ( c = vect[k++] ) {
1.113 noro 4432: if ( DN(c) )
4433: error("afo");
4434: ndl_copy(p,DL(mr)); CQ(mr) = c; NMV_ADV(mr);
1.107 noro 4435: }
4436: }
4437: MKNDV(nd_nvar,mr0,len,r);
4438: return r;
4439: }
4440: }
4441:
1.129 noro 4442: /* for plain vector */
4443:
4444: NDV plain_vect_to_ndv_q(Q *vect,int col,UINT *s0vect)
4445: {
4446: int j,k,len;
4447: UINT *p;
4448: Q c;
4449: NDV r;
4450: NMV mr0,mr;
4451:
4452: for ( j = 0, len = 0; j < col; j++ ) if ( vect[j] ) len++;
4453: if ( !len ) return 0;
4454: else {
4455: mr0 = (NMV)GC_malloc(nmv_adv*len);
4456: #if 0
4457: ndv_alloc += nmv_adv*len;
4458: #endif
4459: mr = mr0;
4460: p = s0vect;
4461: for ( j = k = 0; j < col; j++, p += nd_wpd, k++ )
4462: if ( c = vect[k] ) {
4463: if ( DN(c) )
4464: error("afo");
4465: ndl_copy(p,DL(mr)); CQ(mr) = c; NMV_ADV(mr);
4466: }
4467: MKNDV(nd_nvar,mr0,len,r);
4468: return r;
4469: }
4470: }
4471:
1.133 noro 4472: int nd_sp_f4(int m,int trace,ND_pairs l,PGeoBucket bucket)
1.65 noro 4473: {
4474: ND_pairs t;
4475: NODE sp0,sp;
4476: int stat;
4477: ND spol;
4478:
4479: for ( t = l; t; t = NEXT(t) ) {
1.133 noro 4480: stat = nd_sp(m,trace,t,&spol);
1.65 noro 4481: if ( !stat ) return 0;
4482: if ( spol ) {
4483: add_pbucket_symbolic(bucket,spol);
4484: }
4485: }
1.68 noro 4486: return 1;
1.65 noro 4487: }
4488:
1.133 noro 4489: int nd_symbolic_preproc(PGeoBucket bucket,int trace,UINT **s0vect,NODE *r)
1.65 noro 4490: {
4491: NODE rp0,rp;
4492: NM mul,head,s0,s;
1.76 noro 4493: int index,col,i,sugar;
1.65 noro 4494: RHist h;
4495: UINT *s0v,*p;
4496: NM_ind_pair pair;
4497: ND red;
1.133 noro 4498: NDV *ps;
1.65 noro 4499:
4500: s0 = 0; rp0 = 0; col = 0;
1.133 noro 4501: ps = trace?nd_ps_trace:nd_ps;
1.65 noro 4502: while ( 1 ) {
4503: head = remove_head_pbucket_symbolic(bucket);
4504: if ( !head ) break;
4505: if ( !s0 ) s0 = head;
4506: else NEXT(s) = head;
4507: s = head;
4508: index = ndl_find_reducer(DL(head));
4509: if ( index >= 0 ) {
4510: h = nd_psh[index];
4511: NEWNM(mul);
4512: ndl_sub(DL(head),DL(h),DL(mul));
4513: if ( ndl_check_bound2(index,DL(mul)) ) return 0;
1.133 noro 4514: sugar = TD(DL(mul))+SG(ps[index]);
1.76 noro 4515: MKNM_ind_pair(pair,mul,index,sugar);
1.133 noro 4516: red = ndv_mul_nm_symbolic(mul,ps[index]);
1.65 noro 4517: add_pbucket_symbolic(bucket,nd_remove_head(red));
4518: NEXTNODE(rp0,rp); BDY(rp) = (pointer)pair;
4519: }
4520: col++;
4521: }
1.72 noro 4522: if ( rp0 ) NEXT(rp) = 0;
4523: NEXT(s) = 0;
1.65 noro 4524: s0v = (UINT *)MALLOC_ATOMIC(col*nd_wpd*sizeof(UINT));
4525: for ( i = 0, p = s0v, s = s0; i < col;
4526: i++, p += nd_wpd, s = NEXT(s) ) ndl_copy(DL(s),p);
4527: *s0vect = s0v;
4528: *r = rp0;
4529: return col;
4530: }
4531:
1.69 noro 4532: NODE nd_f4(int m)
4533: {
4534: int i,nh,stat,index;
4535: NODE r,g;
4536: ND_pairs d,l,t;
4537: ND spol,red;
4538: NDV nf,redv;
4539: NM s0,s;
1.84 noro 4540: NODE rp0,srp0,nflist;
1.69 noro 4541: int nsp,nred,col,rank,len,k,j,a;
4542: UINT c;
1.74 noro 4543: UINT **spmat;
1.69 noro 4544: UINT *s0vect,*svect,*p,*v;
4545: int *colstat;
4546: IndArray *imat;
4547: int *rhead;
4548: int spcol,sprow;
4549: int sugar;
4550: PGeoBucket bucket;
4551: struct oEGT eg0,eg1,eg_f4;
4552:
1.103 noro 4553: #if 0
1.74 noro 4554: ndv_alloc = 0;
1.103 noro 4555: #endif
1.69 noro 4556: g = 0; d = 0;
4557: for ( i = 0; i < nd_psn; i++ ) {
4558: d = update_pairs(d,g,i);
4559: g = update_base(g,i);
4560: }
4561: while ( d ) {
4562: get_eg(&eg0);
4563: l = nd_minsugarp(d,&d);
4564: sugar = SG(l);
4565: bucket = create_pbucket();
1.133 noro 4566: stat = nd_sp_f4(m,0,l,bucket);
1.69 noro 4567: if ( !stat ) {
4568: for ( t = l; NEXT(t); t = NEXT(t) );
4569: NEXT(t) = d; d = l;
1.103 noro 4570: d = nd_reconstruct(0,d);
1.69 noro 4571: continue;
4572: }
1.86 noro 4573: if ( bucket->m < 0 ) continue;
1.133 noro 4574: col = nd_symbolic_preproc(bucket,0,&s0vect,&rp0);
1.69 noro 4575: if ( !col ) {
4576: for ( t = l; NEXT(t); t = NEXT(t) );
4577: NEXT(t) = d; d = l;
1.103 noro 4578: d = nd_reconstruct(0,d);
1.69 noro 4579: continue;
4580: }
4581: get_eg(&eg1); init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg1);
1.74 noro 4582: if ( DP_Print )
4583: fprintf(asir_out,"sugar=%d,symb=%fsec,",
4584: sugar,eg_f4.exectime+eg_f4.gctime);
4585: if ( 1 )
1.133 noro 4586: nflist = nd_f4_red(m,l,0,s0vect,col,rp0,0);
1.74 noro 4587: else
1.133 noro 4588: nflist = nd_f4_red_dist(m,l,s0vect,col,rp0,0);
1.69 noro 4589: /* adding new bases */
1.74 noro 4590: for ( r = nflist; r; r = NEXT(r) ) {
4591: nf = (NDV)BDY(r);
1.69 noro 4592: ndv_removecont(m,nf);
1.129 noro 4593: if ( !m && nd_nalg ) {
4594: ND nf1;
4595:
4596: nf1 = ndvtond(m,nf);
4597: nd_monic(0,&nf1);
4598: nd_removecont(m,nf1);
4599: nf = ndtondv(m,nf1);
4600: }
1.77 noro 4601: nh = ndv_newps(m,nf,0);
1.69 noro 4602: d = update_pairs(d,g,nh);
4603: g = update_base(g,nh);
4604: }
4605: }
4606: for ( r = g; r; r = NEXT(r) ) BDY(r) = (pointer)nd_ps[(int)BDY(r)];
1.103 noro 4607: #if 0
1.74 noro 4608: fprintf(asir_out,"ndv_alloc=%d\n",ndv_alloc);
1.103 noro 4609: #endif
1.69 noro 4610: return g;
4611: }
1.74 noro 4612:
1.133 noro 4613: NODE nd_f4_trace(int m)
4614: {
4615: int i,nh,stat,index;
4616: NODE r,g;
4617: ND_pairs d,l,l0,t;
4618: ND spol,red;
4619: NDV nf,redv,nfqv,nfv;
4620: NM s0,s;
4621: NODE rp0,srp0,nflist;
4622: int nsp,nred,col,rank,len,k,j,a;
4623: UINT c;
4624: UINT **spmat;
4625: UINT *s0vect,*svect,*p,*v;
4626: int *colstat;
4627: IndArray *imat;
4628: int *rhead;
4629: int spcol,sprow;
4630: int sugar;
4631: PGeoBucket bucket;
4632: struct oEGT eg0,eg1,eg_f4;
4633:
4634: g = 0; d = 0;
4635: for ( i = 0; i < nd_psn; i++ ) {
4636: d = update_pairs(d,g,i);
4637: g = update_base(g,i);
4638: }
4639: while ( d ) {
4640: get_eg(&eg0);
4641: l = nd_minsugarp(d,&d);
4642: sugar = SG(l);
4643: bucket = create_pbucket();
4644: stat = nd_sp_f4(m,0,l,bucket);
4645: if ( !stat ) {
4646: for ( t = l; NEXT(t); t = NEXT(t) );
4647: NEXT(t) = d; d = l;
4648: d = nd_reconstruct(1,d);
4649: continue;
4650: }
4651: if ( bucket->m < 0 ) continue;
4652: col = nd_symbolic_preproc(bucket,0,&s0vect,&rp0);
4653: if ( !col ) {
4654: for ( t = l; NEXT(t); t = NEXT(t) );
4655: NEXT(t) = d; d = l;
4656: d = nd_reconstruct(1,d);
4657: continue;
4658: }
4659: get_eg(&eg1); init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg1);
4660: if ( DP_Print )
4661: fprintf(asir_out,"sugar=%d,symb=%fsec,",
4662: sugar,eg_f4.exectime+eg_f4.gctime);
4663: nflist = nd_f4_red(m,l,0,s0vect,col,rp0,&l0);
4664: if ( !l0 ) continue;
4665: l = l0;
4666:
4667: /* over Q */
4668: bucket = create_pbucket();
4669: stat = nd_sp_f4(0,1,l,bucket);
4670: if ( !stat ) {
4671: for ( t = l; NEXT(t); t = NEXT(t) );
4672: NEXT(t) = d; d = l;
4673: d = nd_reconstruct(1,d);
4674: continue;
4675: }
4676: if ( bucket->m < 0 ) continue;
4677: col = nd_symbolic_preproc(bucket,1,&s0vect,&rp0);
4678: if ( !col ) {
4679: for ( t = l; NEXT(t); t = NEXT(t) );
4680: NEXT(t) = d; d = l;
4681: d = nd_reconstruct(1,d);
4682: continue;
4683: }
4684: nflist = nd_f4_red(0,l,1,s0vect,col,rp0,0);
4685: /* adding new bases */
4686: for ( r = nflist; r; r = NEXT(r) ) {
4687: nfqv = (NDV)BDY(r);
4688: ndv_removecont(0,nfqv);
4689: if ( !rem(NM(HCQ(nfqv)),m) ) return 0;
4690: if ( nd_nalg ) {
4691: ND nf1;
4692:
4693: nf1 = ndvtond(m,nfqv);
4694: nd_monic(0,&nf1);
4695: nd_removecont(0,nf1);
4696: nfqv = ndtondv(0,nf1); nd_free(nf1);
4697: }
4698: nfv = ndv_dup(0,nfqv);
4699: ndv_mod(m,nfv);
4700: ndv_removecont(m,nfv);
4701: nh = ndv_newps(0,nfv,nfqv);
4702: d = update_pairs(d,g,nh);
4703: g = update_base(g,nh);
4704: }
4705: }
4706: for ( r = g; r; r = NEXT(r) ) BDY(r) = (pointer)nd_ps_trace[(int)BDY(r)];
4707: #if 0
4708: fprintf(asir_out,"ndv_alloc=%d\n",ndv_alloc);
4709: #endif
4710: return g;
4711: }
4712:
4713: NODE nd_f4_red(int m,ND_pairs sp0,int trace,UINT *s0vect,int col,NODE rp0,ND_pairs *nz)
1.63 noro 4714: {
1.67 noro 4715: IndArray *imat;
1.106 noro 4716: int nsp,nred,i;
1.65 noro 4717: int *rhead;
1.106 noro 4718: NODE r0,rp;
1.74 noro 4719: ND_pairs sp;
4720: NM_ind_pair *rvect;
1.63 noro 4721:
1.74 noro 4722: for ( sp = sp0, nsp = 0; sp; sp = NEXT(sp), nsp++ );
1.106 noro 4723: nred = length(rp0);
1.74 noro 4724: imat = (IndArray *)ALLOCA(nred*sizeof(IndArray));
4725: rhead = (int *)ALLOCA(col*sizeof(int));
4726: for ( i = 0; i < col; i++ ) rhead[i] = 0;
1.63 noro 4727:
1.74 noro 4728: /* construction of index arrays */
4729: rvect = (NM_ind_pair *)ALLOCA(nred*sizeof(NM_ind_pair));
4730: for ( rp = rp0, i = 0; rp; i++, rp = NEXT(rp) ) {
4731: rvect[i] = (NM_ind_pair)BDY(rp);
4732: imat[i] = nm_ind_pair_to_vect_compress(m,s0vect,col,rvect[i]);
4733: rhead[imat[i]->head] = 1;
4734: }
1.107 noro 4735: if ( m )
1.133 noro 4736: r0 = nd_f4_red_main(m,sp0,nsp,s0vect,col,rvect,rhead,imat,nred,nz);
1.107 noro 4737: else
1.133 noro 4738: r0 = nd_f4_red_q_main(sp0,nsp,trace,s0vect,col,rvect,rhead,imat,nred);
1.106 noro 4739: return r0;
4740: }
1.74 noro 4741:
1.106 noro 4742: NODE nd_f4_red_main(int m,ND_pairs sp0,int nsp,UINT *s0vect,int col,
1.133 noro 4743: NM_ind_pair *rvect,int *rhead,IndArray *imat,int nred,ND_pairs *nz)
1.106 noro 4744: {
4745: int spcol,sprow,a;
4746: int i,j,k,l,rank;
4747: NODE r0,r;
4748: ND_pairs sp;
4749: ND spol;
4750: int **spmat;
4751: UINT *svect,*v;
4752: int *colstat;
4753: struct oEGT eg0,eg1,eg2,eg_f4,eg_f4_1,eg_f4_2;
4754: int maxrs;
4755: int *spsugar;
1.133 noro 4756: ND_pairs *spactive;
1.106 noro 4757:
4758: spcol = col-nred;
4759: get_eg(&eg0);
1.74 noro 4760: /* elimination (1st step) */
4761: spmat = (int **)ALLOCA(nsp*sizeof(UINT *));
4762: svect = (UINT *)ALLOCA(col*sizeof(UINT));
1.76 noro 4763: spsugar = (int *)ALLOCA(nsp*sizeof(UINT));
1.133 noro 4764: spactive = !nz?0:(ND_pairs *)ALLOCA(nsp*sizeof(ND_pairs));
1.74 noro 4765: for ( a = sprow = 0, sp = sp0; a < nsp; a++, sp = NEXT(sp) ) {
4766: nd_sp(m,0,sp,&spol);
1.75 noro 4767: if ( !spol ) continue;
1.74 noro 4768: nd_to_vect(m,s0vect,col,spol,svect);
1.76 noro 4769: if ( m == -1 )
4770: maxrs = ndv_reduce_vect_sf(m,svect,col,imat,rvect,nred);
4771: else
4772: maxrs = ndv_reduce_vect(m,svect,col,imat,rvect,nred);
1.74 noro 4773: for ( i = 0; i < col; i++ ) if ( svect[i] ) break;
4774: if ( i < col ) {
4775: spmat[sprow] = v = (UINT *)MALLOC_ATOMIC(spcol*sizeof(UINT));
4776: for ( j = k = 0; j < col; j++ )
4777: if ( !rhead[j] ) v[k++] = svect[j];
1.76 noro 4778: spsugar[sprow] = MAX(maxrs,SG(spol));
1.133 noro 4779: if ( nz )
4780: spactive[sprow] = sp;
1.74 noro 4781: sprow++;
4782: }
1.76 noro 4783: nd_free(spol);
1.74 noro 4784: }
1.80 noro 4785: get_eg(&eg1); init_eg(&eg_f4_1); add_eg(&eg_f4_1,&eg0,&eg1);
4786: if ( DP_Print ) {
4787: fprintf(asir_out,"elim1=%fsec,",eg_f4_1.exectime+eg_f4_1.gctime);
4788: fflush(asir_out);
4789: }
1.74 noro 4790: /* free index arrays */
4791: for ( i = 0; i < nred; i++ ) GC_free(imat[i]->index.c);
4792:
4793: /* elimination (2nd step) */
4794: colstat = (int *)ALLOCA(spcol*sizeof(int));
4795: if ( m == -1 )
1.76 noro 4796: rank = nd_gauss_elim_sf(spmat,spsugar,sprow,spcol,m,colstat);
1.74 noro 4797: else
1.133 noro 4798: rank = nd_gauss_elim_mod(spmat,spsugar,spactive,sprow,spcol,m,colstat);
1.74 noro 4799: r0 = 0;
4800: for ( i = 0; i < rank; i++ ) {
4801: NEXTNODE(r0,r); BDY(r) =
4802: (pointer)vect_to_ndv(spmat[i],spcol,col,rhead,s0vect);
1.76 noro 4803: SG((NDV)BDY(r)) = spsugar[i];
1.74 noro 4804: GC_free(spmat[i]);
4805: }
1.109 noro 4806: if ( r0 ) NEXT(r) = 0;
1.133 noro 4807:
1.74 noro 4808: for ( ; i < sprow; i++ ) GC_free(spmat[i]);
1.80 noro 4809: get_eg(&eg2); init_eg(&eg_f4_2); add_eg(&eg_f4_2,&eg1,&eg2);
4810: init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg2);
1.74 noro 4811: if ( DP_Print ) {
1.80 noro 4812: fprintf(asir_out,"elim2=%fsec\n",eg_f4_2.exectime+eg_f4_2.gctime);
1.74 noro 4813: fprintf(asir_out,"nsp=%d,nred=%d,spmat=(%d,%d),rank=%d ",
4814: nsp,nred,sprow,spcol,rank);
4815: fprintf(asir_out,"%fsec\n",eg_f4.exectime+eg_f4.gctime);
1.63 noro 4816: }
1.133 noro 4817: if ( nz ) {
4818: for ( i = 0; i < rank-1; i++ ) NEXT(spactive[i]) = spactive[i+1];
4819: if ( rank > 0 ) {
4820: NEXT(spactive[rank-1]) = 0;
4821: *nz = spactive[0];
4822: } else
4823: *nz = 0;
4824: }
1.74 noro 4825: return r0;
4826: }
4827:
1.133 noro 4828: #if 1
4829: NODE nd_f4_red_q_main(ND_pairs sp0,int nsp,int trace,UINT *s0vect,int col,
1.107 noro 4830: NM_ind_pair *rvect,int *rhead,IndArray *imat,int nred)
4831: {
4832: int spcol,sprow,a;
4833: int i,j,k,l,rank;
4834: NODE r0,r;
4835: ND_pairs sp;
4836: ND spol;
1.113 noro 4837: Q **spmat;
4838: Q *svect,*v;
1.107 noro 4839: int *colstat;
4840: struct oEGT eg0,eg1,eg2,eg_f4,eg_f4_1,eg_f4_2;
4841: int maxrs;
4842: int *spsugar;
1.137 noro 4843: pointer *w;
1.107 noro 4844:
4845: spcol = col-nred;
4846: get_eg(&eg0);
4847: /* elimination (1st step) */
1.113 noro 4848: spmat = (Q **)ALLOCA(nsp*sizeof(Q *));
4849: svect = (Q *)ALLOCA(col*sizeof(Q));
4850: spsugar = (int *)ALLOCA(nsp*sizeof(Q));
1.107 noro 4851: for ( a = sprow = 0, sp = sp0; a < nsp; a++, sp = NEXT(sp) ) {
1.133 noro 4852: nd_sp(0,trace,sp,&spol);
1.107 noro 4853: if ( !spol ) continue;
1.113 noro 4854: nd_to_vect_q(s0vect,col,spol,svect);
1.133 noro 4855: maxrs = ndv_reduce_vect_q(svect,trace,col,imat,rvect,nred);
1.107 noro 4856: for ( i = 0; i < col; i++ ) if ( svect[i] ) break;
4857: if ( i < col ) {
1.113 noro 4858: spmat[sprow] = v = (Q *)MALLOC(spcol*sizeof(Q));
1.107 noro 4859: for ( j = k = 0; j < col; j++ )
4860: if ( !rhead[j] ) v[k++] = svect[j];
4861: spsugar[sprow] = MAX(maxrs,SG(spol));
4862: sprow++;
4863: }
1.109 noro 4864: /* nd_free(spol); */
1.107 noro 4865: }
4866: get_eg(&eg1); init_eg(&eg_f4_1); add_eg(&eg_f4_1,&eg0,&eg1);
4867: if ( DP_Print ) {
4868: fprintf(asir_out,"elim1=%fsec,",eg_f4_1.exectime+eg_f4_1.gctime);
4869: fflush(asir_out);
4870: }
4871: /* free index arrays */
1.109 noro 4872: /* for ( i = 0; i < nred; i++ ) GC_free(imat[i]->index.c); */
1.107 noro 4873:
4874: /* elimination (2nd step) */
4875: colstat = (int *)ALLOCA(spcol*sizeof(int));
1.113 noro 4876: rank = nd_gauss_elim_q(spmat,spsugar,sprow,spcol,colstat);
1.137 noro 4877: w = (pointer *)ALLOCA(rank*sizeof(pointer));
4878: for ( i = 0; i < rank; i++ ) {
4879: w[rank-i-1] = (pointer)vect_to_ndv_q(spmat[i],spcol,col,rhead,s0vect);
4880: SG((NDV)w[rank-i-1]) = spsugar[i];
4881: /* GC_free(spmat[i]); */
4882: }
1.138 noro 4883: #if 0
1.137 noro 4884: qsort(w,rank,sizeof(NDV),
4885: (int (*)(const void *,const void *))ndv_compare);
4886: #endif
1.107 noro 4887: r0 = 0;
4888: for ( i = 0; i < rank; i++ ) {
1.137 noro 4889: NEXTNODE(r0,r); BDY(r) = w[i];
1.107 noro 4890: }
1.137 noro 4891: if ( r0 ) NEXT(r) = 0;
1.135 noro 4892:
1.109 noro 4893: /* for ( ; i < sprow; i++ ) GC_free(spmat[i]); */
1.107 noro 4894: get_eg(&eg2); init_eg(&eg_f4_2); add_eg(&eg_f4_2,&eg1,&eg2);
4895: init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg2);
4896: if ( DP_Print ) {
4897: fprintf(asir_out,"elim2=%fsec\n",eg_f4_2.exectime+eg_f4_2.gctime);
4898: fprintf(asir_out,"nsp=%d,nred=%d,spmat=(%d,%d),rank=%d ",
4899: nsp,nred,sprow,spcol,rank);
4900: fprintf(asir_out,"%fsec\n",eg_f4.exectime+eg_f4.gctime);
4901: }
4902: return r0;
4903: }
1.129 noro 4904: #else
4905: void printm(Q **mat,int row,int col)
4906: {
4907: int i,j;
4908: printf("[");
4909: for ( i = 0; i < row; i++ ) {
4910: for ( j = 0; j < col; j++ ) {
4911: printexpr(CO,mat[i][j]); printf(" ");
4912: }
4913: printf("]\n");
4914: }
4915: }
4916:
4917: NODE nd_f4_red_q_main(ND_pairs sp0,int nsp,UINT *s0vect,int col,
4918: NM_ind_pair *rvect,int *rhead,IndArray *imat,int nred)
4919: {
4920: int row,a;
4921: int i,j,rank;
4922: NODE r0,r;
4923: ND_pairs sp;
4924: ND spol;
4925: Q **mat;
4926: int *colstat;
4927: int *sugar;
4928:
4929: row = nsp+nred;
4930: /* make the matrix */
4931: mat = (Q **)ALLOCA(row*sizeof(Q *));
4932: sugar = (int *)ALLOCA(row*sizeof(int));
4933: for ( row = a = 0, sp = sp0; a < nsp; a++, sp = NEXT(sp) ) {
4934: nd_sp(0,0,sp,&spol);
4935: if ( !spol ) continue;
4936: mat[row] = (Q *)MALLOC(col*sizeof(Q));
4937: nd_to_vect_q(s0vect,col,spol,mat[row]);
4938: sugar[row] = SG(spol);
4939: row++;
4940: }
4941: for ( i = 0; i < nred; i++, row++ ) {
4942: mat[row] = nm_ind_pair_to_vect(0,s0vect,col,rvect[i]);
4943: sugar[row] = rvect[i]->sugar;
4944: }
4945: /* elimination */
4946: colstat = (int *)ALLOCA(col*sizeof(int));
4947: rank = nd_gauss_elim_q(mat,sugar,row,col,colstat);
4948: r0 = 0;
4949: for ( i = 0; i < rank; i++ ) {
4950: for ( j = 0; j < col; j++ ) if ( mat[i][j] ) break;
4951: if ( j == col ) error("nd_f4_red_q_main : cannot happen");
4952: if ( rhead[j] ) continue;
4953: NEXTNODE(r0,r); BDY(r) =
4954: (pointer)plain_vect_to_ndv_q(mat[i],col,s0vect);
4955: SG((NDV)BDY(r)) = sugar[i];
4956: }
4957: if ( r0 ) NEXT(r) = 0;
4958: printf("\n");
4959: return r0;
4960: }
4961: #endif
1.107 noro 4962:
1.74 noro 4963: FILE *nd_write,*nd_read;
4964:
4965: void nd_send_int(int a) {
4966: write_int(nd_write,&a);
4967: }
4968:
4969: void nd_send_intarray(int *p,int len) {
4970: write_intarray(nd_write,p,len);
4971: }
4972:
4973: int nd_recv_int() {
4974: int a;
4975:
4976: read_int(nd_read,&a);
4977: return a;
4978: }
4979:
4980: void nd_recv_intarray(int *p,int len) {
4981: read_intarray(nd_read,p,len);
4982: }
4983:
4984: void nd_send_ndv(NDV p) {
4985: int len,i;
4986: NMV m;
4987:
4988: if ( !p ) nd_send_int(0);
4989: else {
4990: len = LEN(p);
4991: nd_send_int(len);
4992: m = BDY(p);
4993: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
4994: nd_send_int(CM(m));
4995: nd_send_intarray(DL(m),nd_wpd);
1.68 noro 4996: }
1.74 noro 4997: }
4998: }
4999:
5000: void nd_send_nd(ND p) {
5001: int len,i;
5002: NM m;
5003:
5004: if ( !p ) nd_send_int(0);
5005: else {
5006: len = LEN(p);
5007: nd_send_int(len);
5008: m = BDY(p);
5009: for ( i = 0; i < len; i++, m = NEXT(m) ) {
5010: nd_send_int(CM(m));
5011: nd_send_intarray(DL(m),nd_wpd);
1.65 noro 5012: }
1.74 noro 5013: }
5014: }
1.65 noro 5015:
1.74 noro 5016: NDV nd_recv_ndv()
5017: {
5018: int len,i;
5019: NMV m,m0;
5020: NDV r;
1.65 noro 5021:
1.74 noro 5022: len = nd_recv_int();
5023: if ( !len ) return 0;
5024: else {
5025: m0 = m = (NMV)GC_malloc_atomic_ignore_off_page(nmv_adv*len);
1.103 noro 5026: #if 0
1.74 noro 5027: ndv_alloc += len*nmv_adv;
1.103 noro 5028: #endif
1.74 noro 5029: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
5030: CM(m) = nd_recv_int();
5031: nd_recv_intarray(DL(m),nd_wpd);
1.71 noro 5032: }
1.74 noro 5033: MKNDV(nd_nvar,m0,len,r);
5034: return r;
5035: }
5036: }
1.65 noro 5037:
1.74 noro 5038: int ox_exec_f4_red(Q proc)
5039: {
5040: Obj obj;
5041: STRING fname;
5042: NODE arg;
5043: int s;
5044: extern int ox_need_conv,ox_file_io;
5045:
5046: MKSTR(fname,"nd_exec_f4_red");
5047: arg = mknode(2,proc,fname);
5048: Pox_cmo_rpc(arg,&obj);
5049: s = get_ox_server_id(QTOS(proc));
5050: nd_write = iofp[s].out;
5051: nd_read = iofp[s].in;
5052: ox_need_conv = ox_file_io = 0;
5053: return s;
5054: }
5055:
1.133 noro 5056: NODE nd_f4_red_dist(int m,ND_pairs sp0,UINT *s0vect,int col,NODE rp0,ND_pairs *nz)
1.74 noro 5057: {
5058: int nsp,nred;
5059: int i,rank,s;
5060: NODE rp,r0,r;
5061: ND_pairs sp;
5062: NM_ind_pair pair;
5063: NMV nmv;
5064: NM nm;
5065: NDV nf;
5066: Obj proc,dmy;
5067:
5068: ox_launch_main(0,0,&proc);
5069: s = ox_exec_f4_red((Q)proc);
5070:
5071: nd_send_int(m);
5072: nd_send_int(nd_nvar);
5073: nd_send_int(nd_bpe);
5074: nd_send_int(nd_wpd);
5075: nd_send_int(nmv_adv);
5076:
1.89 noro 5077: saveobj(nd_write,dp_current_spec->obj); fflush(nd_write);
1.74 noro 5078:
5079: nd_send_int(nd_psn);
5080: for ( i = 0; i < nd_psn; i++ ) nd_send_ndv(nd_ps[i]);
5081:
5082: for ( sp = sp0, nsp = 0; sp; sp = NEXT(sp), nsp++ );
5083: nd_send_int(nsp);
5084: for ( i = 0, sp = sp0; i < nsp; i++, sp = NEXT(sp) ) {
5085: nd_send_int(sp->i1); nd_send_int(sp->i2);
5086: }
5087:
5088: nd_send_int(col); nd_send_intarray(s0vect,col*nd_wpd);
5089:
5090: nred = length(rp0); nd_send_int(nred);
5091: for ( i = 0, rp = rp0; i < nred; i++, rp = NEXT(rp) ) {
5092: pair = (NM_ind_pair)BDY(rp);
5093: nd_send_int(pair->index);
5094: nd_send_intarray(pair->mul->dl,nd_wpd);
5095: }
5096: fflush(nd_write);
5097: rank = nd_recv_int();
5098: fprintf(asir_out,"rank=%d\n",rank);
5099: r0 = 0;
5100: for ( i = 0; i < rank; i++ ) {
5101: nf = nd_recv_ndv();
5102: NEXTNODE(r0,r); BDY(r) = (pointer)nf;
5103: }
5104: Pox_shutdown(mknode(1,proc),&dmy);
5105: return r0;
5106: }
5107:
5108: /* server side */
5109:
5110: void nd_exec_f4_red_dist()
5111: {
5112: int m,i,nsp,col,s0size,nred,spcol,j,k;
5113: NM_ind_pair *rp0;
5114: NDV nf;
5115: UINT *s0vect;
5116: IndArray *imat;
5117: int *rhead;
5118: int **spmat;
5119: UINT *svect,*v;
5120: ND_pairs *sp0;
5121: int *colstat;
5122: int a,sprow,rank;
1.89 noro 5123: struct order_spec *ord;
1.74 noro 5124: Obj ordspec;
5125: ND spol;
1.76 noro 5126: int maxrs;
5127: int *spsugar;
1.74 noro 5128:
5129: nd_read = iofp[0].in;
5130: nd_write = iofp[0].out;
5131: m = nd_recv_int();
5132: nd_nvar = nd_recv_int();
5133: nd_bpe = nd_recv_int();
5134: nd_wpd = nd_recv_int();
5135: nmv_adv = nd_recv_int();
5136:
5137: loadobj(nd_read,&ordspec);
1.89 noro 5138: create_order_spec(0,ordspec,&ord);
5139: nd_init_ord(ord);
1.74 noro 5140: nd_setup_parameters(nd_nvar,0);
5141:
5142: nd_psn = nd_recv_int();
5143: nd_ps = (NDV *)MALLOC(nd_psn*sizeof(NDV));
5144: nd_bound = (UINT **)MALLOC(nd_psn*sizeof(UINT *));
5145: for ( i = 0; i < nd_psn; i++ ) {
5146: nd_ps[i] = nd_recv_ndv();
5147: nd_bound[i] = ndv_compute_bound(nd_ps[i]);
5148: }
5149:
5150: nsp = nd_recv_int();
5151: sp0 = (ND_pairs *)MALLOC(nsp*sizeof(ND_pairs));
5152: for ( i = 0; i < nsp; i++ ) {
5153: NEWND_pairs(sp0[i]);
5154: sp0[i]->i1 = nd_recv_int(); sp0[i]->i2 = nd_recv_int();
5155: ndl_lcm(HDL(nd_ps[sp0[i]->i1]),HDL(nd_ps[sp0[i]->i2]),LCM(sp0[i]));
5156: }
5157:
5158: col = nd_recv_int();
5159: s0size = col*nd_wpd;
5160: s0vect = (UINT *)MALLOC(s0size*sizeof(UINT));
5161: nd_recv_intarray(s0vect,s0size);
5162:
5163: nred = nd_recv_int();
5164: rp0 = (NM_ind_pair *)MALLOC(nred*sizeof(NM_ind_pair));
5165: for ( i = 0; i < nred; i++ ) {
5166: rp0[i] = (NM_ind_pair)MALLOC(sizeof(struct oNM_ind_pair));
5167: rp0[i]->index = nd_recv_int();
5168: rp0[i]->mul = (NM)MALLOC(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT));
5169: nd_recv_intarray(rp0[i]->mul->dl,nd_wpd);
5170: }
5171:
5172: spcol = col-nred;
5173: imat = (IndArray *)MALLOC(nred*sizeof(IndArray));
5174: rhead = (int *)MALLOC(col*sizeof(int));
5175: for ( i = 0; i < col; i++ ) rhead[i] = 0;
5176:
5177: /* construction of index arrays */
5178: for ( i = 0; i < nred; i++ ) {
5179: imat[i] = nm_ind_pair_to_vect_compress(m,s0vect,col,rp0[i]);
5180: rhead[imat[i]->head] = 1;
5181: }
5182:
5183: /* elimination (1st step) */
5184: spmat = (int **)MALLOC(nsp*sizeof(UINT *));
5185: svect = (UINT *)MALLOC(col*sizeof(UINT));
1.76 noro 5186: spsugar = (int *)ALLOCA(nsp*sizeof(UINT));
1.74 noro 5187: for ( a = sprow = 0; a < nsp; a++ ) {
5188: nd_sp(m,0,sp0[a],&spol);
1.75 noro 5189: if ( !spol ) continue;
1.74 noro 5190: nd_to_vect(m,s0vect,col,spol,svect);
1.76 noro 5191: if ( m == -1 )
5192: maxrs = ndv_reduce_vect_sf(m,svect,col,imat,rp0,nred);
5193: else
5194: maxrs = ndv_reduce_vect(m,svect,col,imat,rp0,nred);
1.74 noro 5195: for ( i = 0; i < col; i++ ) if ( svect[i] ) break;
5196: if ( i < col ) {
5197: spmat[sprow] = v = (UINT *)MALLOC(spcol*sizeof(UINT));
5198: for ( j = k = 0; j < col; j++ )
5199: if ( !rhead[j] ) v[k++] = svect[j];
1.76 noro 5200: spsugar[sprow] = MAX(maxrs,SG(spol));
1.74 noro 5201: sprow++;
5202: }
1.76 noro 5203: nd_free(spol);
1.74 noro 5204: }
5205: /* elimination (2nd step) */
5206: colstat = (int *)ALLOCA(spcol*sizeof(int));
5207: if ( m == -1 )
1.76 noro 5208: rank = nd_gauss_elim_sf(spmat,spsugar,sprow,spcol,m,colstat);
1.74 noro 5209: else
1.133 noro 5210: rank = nd_gauss_elim_mod(spmat,spsugar,0,sprow,spcol,m,colstat);
1.74 noro 5211: nd_send_int(rank);
5212: for ( i = 0; i < rank; i++ ) {
5213: nf = vect_to_ndv(spmat[i],spcol,col,rhead,s0vect);
5214: nd_send_ndv(nf);
1.63 noro 5215: }
1.74 noro 5216: fflush(nd_write);
1.107 noro 5217: }
5218:
1.113 noro 5219: int nd_gauss_elim_q(Q **mat0,int *sugar,int row,int col,int *colstat)
1.107 noro 5220: {
1.109 noro 5221: int mod,i,j,t,c,rank,rank0,inv;
5222: int *ci,*ri;
1.113 noro 5223: Q dn;
1.109 noro 5224: MAT m,nm;
1.108 noro 5225: int **wmat;
5226:
5227: /* XXX */
5228: mod = 99999989;
5229: wmat = (int **)ALLOCA(row*sizeof(int *));
5230: for ( i = 0; i < row; i++ ) {
5231: wmat[i] = (int *)ALLOCA(col*sizeof(int));
5232: for ( j = 0; j < col; j++ ) {
1.113 noro 5233: if ( mat0[i][j] ) {
5234: t = rem(NM(mat0[i][j]),mod);
5235: if ( SGN(mat0[i][j]) < 0 ) t = mod-t;
5236: wmat[i][j] = t;
5237: } else
1.108 noro 5238: wmat[i][j] = 0;
5239: }
5240: }
1.133 noro 5241: rank0 = nd_gauss_elim_mod(wmat,sugar,0,row,col,mod,colstat);
1.109 noro 5242: NEWMAT(m); m->row = row; m->col = col; m->body = (pointer **)mat0;
1.113 noro 5243: rank = generic_gauss_elim(m,&nm,&dn,&ri,&ci);
1.109 noro 5244: if ( rank != rank0 )
5245: error("afo");
5246: for ( i = 0; i < row; i++ )
5247: for ( j = 0; j < col; j++ )
5248: mat0[i][j] = 0;
5249: c = col-rank;
5250: for ( i = 0; i < rank; i++ ) {
1.113 noro 5251: mat0[i][ri[i]] = dn;
1.109 noro 5252: for ( j = 0; j < c; j++ )
1.113 noro 5253: mat0[i][ci[j]] = (Q)BDY(nm)[i][j];
1.109 noro 5254: }
1.113 noro 5255: inv = invm(rem(NM(dn),mod),mod);
5256: if ( SGN(dn) < 0 ) inv = mod-inv;
1.109 noro 5257: for ( i = 0; i < row; i++ )
5258: for ( j = 0; j < col; j++ ) {
1.113 noro 5259: if ( mat0[i][j] ) {
5260: t = rem(NM(mat0[i][j]),mod);
5261: if ( SGN(mat0[i][j]) < 0 ) t = mod-t;
5262: } else
1.109 noro 5263: t = 0;
5264: c = dmar(t,inv,0,mod);
5265: if ( wmat[i][j] != c )
5266: error("afo");
5267: }
5268: return rank;
1.76 noro 5269: }
5270:
1.133 noro 5271: int nd_gauss_elim_mod(int **mat0,int *sugar,ND_pairs *spactive,int row,int col,int md,int *colstat)
1.76 noro 5272: {
5273: int i,j,k,l,inv,a,rank,s;
5274: unsigned int *t,*pivot,*pk;
5275: unsigned int **mat;
1.133 noro 5276: ND_pairs pair;
1.76 noro 5277:
5278: mat = (unsigned int **)mat0;
5279: for ( rank = 0, j = 0; j < col; j++ ) {
5280: for ( i = rank; i < row; i++ )
5281: mat[i][j] %= md;
5282: for ( i = rank; i < row; i++ )
5283: if ( mat[i][j] )
5284: break;
5285: if ( i == row ) {
5286: colstat[j] = 0;
5287: continue;
5288: } else
5289: colstat[j] = 1;
5290: if ( i != rank ) {
5291: t = mat[i]; mat[i] = mat[rank]; mat[rank] = t;
5292: s = sugar[i]; sugar[i] = sugar[rank]; sugar[rank] = s;
1.133 noro 5293: if ( spactive ) {
5294: pair = spactive[i]; spactive[i] = spactive[rank];
5295: spactive[rank] = pair;
5296: }
1.76 noro 5297: }
5298: pivot = mat[rank];
5299: s = sugar[rank];
5300: inv = invm(pivot[j],md);
5301: for ( k = j, pk = pivot+k; k < col; k++, pk++ )
5302: if ( *pk ) {
5303: if ( *pk >= (unsigned int)md )
5304: *pk %= md;
5305: DMAR(*pk,inv,0,md,*pk)
5306: }
5307: for ( i = rank+1; i < row; i++ ) {
5308: t = mat[i];
5309: if ( a = t[j] ) {
5310: sugar[i] = MAX(sugar[i],s);
5311: red_by_vect(md,t+j,pivot+j,md-a,col-j);
5312: }
5313: }
5314: rank++;
5315: }
5316: for ( j = col-1, l = rank-1; j >= 0; j-- )
5317: if ( colstat[j] ) {
5318: pivot = mat[l];
5319: s = sugar[l];
5320: for ( i = 0; i < l; i++ ) {
5321: t = mat[i];
5322: t[j] %= md;
5323: if ( a = t[j] ) {
5324: sugar[i] = MAX(sugar[i],s);
5325: red_by_vect(md,t+j,pivot+j,md-a,col-j);
5326: }
5327: }
5328: l--;
5329: }
5330: for ( j = 0, l = 0; l < rank; j++ )
5331: if ( colstat[j] ) {
5332: t = mat[l];
5333: for ( k = j; k < col; k++ )
5334: if ( t[k] >= (unsigned int)md )
5335: t[k] %= md;
5336: l++;
5337: }
5338: return rank;
5339: }
5340:
5341: int nd_gauss_elim_sf(int **mat0,int *sugar,int row,int col,int md,int *colstat)
5342: {
5343: int i,j,k,l,inv,a,rank,s;
5344: unsigned int *t,*pivot,*pk;
5345: unsigned int **mat;
5346:
5347: mat = (unsigned int **)mat0;
5348: for ( rank = 0, j = 0; j < col; j++ ) {
5349: for ( i = rank; i < row; i++ )
5350: if ( mat[i][j] )
5351: break;
5352: if ( i == row ) {
5353: colstat[j] = 0;
5354: continue;
5355: } else
5356: colstat[j] = 1;
5357: if ( i != rank ) {
5358: t = mat[i]; mat[i] = mat[rank]; mat[rank] = t;
5359: s = sugar[i]; sugar[i] = sugar[rank]; sugar[rank] = s;
5360: }
5361: pivot = mat[rank];
5362: s = sugar[rank];
5363: inv = _invsf(pivot[j]);
5364: for ( k = j, pk = pivot+k; k < col; k++, pk++ )
5365: if ( *pk )
5366: *pk = _mulsf(*pk,inv);
5367: for ( i = rank+1; i < row; i++ ) {
5368: t = mat[i];
5369: if ( a = t[j] ) {
5370: sugar[i] = MAX(sugar[i],s);
5371: red_by_vect_sf(md,t+j,pivot+j,_chsgnsf(a),col-j);
5372: }
5373: }
5374: rank++;
5375: }
5376: for ( j = col-1, l = rank-1; j >= 0; j-- )
5377: if ( colstat[j] ) {
5378: pivot = mat[l];
5379: s = sugar[l];
5380: for ( i = 0; i < l; i++ ) {
5381: t = mat[i];
5382: if ( a = t[j] ) {
5383: sugar[i] = MAX(sugar[i],s);
5384: red_by_vect_sf(md,t+j,pivot+j,_chsgnsf(a),col-j);
5385: }
5386: }
5387: l--;
5388: }
5389: return rank;
1.77 noro 5390: }
5391:
5392: int ndv_ishomo(NDV p)
5393: {
5394: NMV m;
5395: int len,h;
5396:
5397: if ( !p ) return 1;
5398: len = LEN(p);
5399: m = BDY(p);
5400: h = TD(DL(m));
5401: NMV_ADV(m);
5402: for ( len--; len; len--, NMV_ADV(m) )
5403: if ( TD(DL(m)) != h ) return 0;
5404: return 1;
5405: }
5406:
5407: void ndv_save(NDV p,int index)
5408: {
5409: FILE *s;
5410: char name[BUFSIZ];
5411: short id;
5412: int nv,sugar,len,n,i,td,e,j;
5413: NMV m;
5414: unsigned int *dl;
5415:
5416: sprintf(name,"%s/%d",Demand,index);
5417: s = fopen(name,"w");
5418: savevl(s,0);
5419: if ( !p ) {
5420: saveobj(s,0);
5421: return;
5422: }
5423: id = O_DP;
5424: nv = NV(p);
5425: sugar = SG(p);
5426: len = LEN(p);
5427: write_short(s,&id); write_int(s,&nv); write_int(s,&sugar);
5428: write_int(s,&len);
5429:
5430: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
1.113 noro 5431: saveobj(s,(Obj)CQ(m));
1.77 noro 5432: dl = DL(m);
5433: td = TD(dl);
5434: write_int(s,&td);
5435: for ( j = 0; j < nv; j++ ) {
5436: e = GET_EXP(dl,j);
5437: write_int(s,&e);
5438: }
5439: }
5440: fclose(s);
5441: }
5442:
5443: NDV ndv_load(int index)
5444: {
5445: FILE *s;
5446: char name[BUFSIZ];
5447: short id;
5448: int nv,sugar,len,n,i,td,e,j;
5449: NDV d;
5450: NMV m0,m;
5451: unsigned int *dl;
5452: Obj obj;
5453:
5454: sprintf(name,"%s/%d",Demand,index);
5455: s = fopen(name,"r");
5456: if ( !s ) return 0;
5457:
5458: skipvl(s);
5459: read_short(s,&id);
5460: if ( !id ) return 0;
5461: read_int(s,&nv);
5462: read_int(s,&sugar);
5463: read_int(s,&len);
5464:
5465: m0 = m = MALLOC(len*nmv_adv);
5466: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
1.113 noro 5467: loadobj(s,&obj); CQ(m) = (Q)obj;
1.77 noro 5468: dl = DL(m);
5469: ndl_zero(dl);
5470: read_int(s,&td); TD(dl) = td;
5471: for ( j = 0; j < nv; j++ ) {
5472: read_int(s,&e);
5473: PUT_EXP(dl,j,e);
5474: }
5475: if ( nd_blockmask ) ndl_weight_mask(dl);
5476: }
5477: fclose(s);
5478: MKNDV(nv,m0,len,d);
5479: SG(d) = sugar;
5480: return d;
1.99 noro 5481: }
5482:
1.102 noro 5483: void nd_det(int mod,MAT f,P *rp)
1.99 noro 5484: {
5485: VL fv,tv;
1.102 noro 5486: int n,i,j,max,e,nvar,sgn,k0,l0,len0,len,k,l,a;
1.99 noro 5487: pointer **m;
1.113 noro 5488: Q mone;
1.125 noro 5489: P **w;
1.126 noro 5490: P mp,r;
1.102 noro 5491: NDV **dm;
5492: NDV *t,*mi,*mj;
5493: NDV d,s,mij,mjj;
5494: ND u;
5495: NMV nmv;
1.114 noro 5496: UINT *bound;
1.99 noro 5497: PGeoBucket bucket;
5498: struct order_spec *ord;
1.126 noro 5499: Q dq,dt,ds;
5500: N gn,qn,dn0,nm,dn;
1.99 noro 5501:
5502: create_order_spec(0,0,&ord);
5503: nd_init_ord(ord);
5504: get_vars((Obj)f,&fv);
5505: if ( f->row != f->col )
5506: error("nd_det : non-square matrix");
5507: n = f->row;
1.125 noro 5508: m = f->body;
1.99 noro 5509: for ( nvar = 0, tv = fv; tv; tv = NEXT(tv), nvar++ );
1.125 noro 5510:
5511: if ( !nvar ) {
5512: if ( !mod )
5513: detp(CO,(P **)m,n,rp);
5514: else {
5515: w = (P **)almat_pointer(n,n);
5516: for ( i = 0; i < n; i++ )
5517: for ( j = 0; j < n; j++ )
5518: ptomp(mod,(P)m[i][j],&w[i][j]);
5519: detmp(CO,mod,w,n,&mp);
5520: mptop(mp,rp);
5521: }
5522: return;
5523: }
1.126 noro 5524:
5525: if ( !mod ) {
5526: w = (P **)almat_pointer(n,n);
5527: dq = ONE;
5528: for ( i = 0; i < n; i++ ) {
5529: dn0 = ONEN;
5530: for ( j = 0; j < n; j++ ) {
5531: if ( !m[i][j] ) continue;
5532: lgp(m[i][j],&nm,&dn);
5533: gcdn(dn0,dn,&gn); divsn(dn0,gn,&qn); muln(qn,dn,&dn0);
5534: }
5535: if ( !UNIN(dn0) ) {
5536: NTOQ(dn0,1,ds);
5537: for ( j = 0; j < n; j++ )
5538: mulp(CO,(P)m[i][j],(P)ds,&w[i][j]);
5539: mulq(dq,ds,&dt); dq = dt;
5540: } else
5541: for ( j = 0; j < n; j++ )
5542: w[i][j] = (P)m[i][j];
5543: }
5544: m = (pointer **)w;
5545: }
1.125 noro 5546:
1.99 noro 5547: for ( i = 0, max = 0; i < n; i++ )
5548: for ( j = 0; j < n; j++ )
5549: for ( tv = fv; tv; tv = NEXT(tv) ) {
5550: e = getdeg(tv->v,(P)m[i][j]);
5551: max = MAX(e,max);
5552: }
1.114 noro 5553: nd_setup_parameters(nvar,max);
1.102 noro 5554: dm = (NDV **)almat_pointer(n,n);
1.99 noro 5555: for ( i = 0, max = 0; i < n; i++ )
1.102 noro 5556: for ( j = 0; j < n; j++ ) {
5557: dm[i][j] = ptondv(CO,fv,m[i][j]);
5558: if ( mod ) ndv_mod(mod,dm[i][j]);
5559: if ( dm[i][j] && !LEN(dm[i][j]) ) dm[i][j] = 0;
5560: }
5561: d = ptondv(CO,fv,(P)ONE);
5562: if ( mod ) ndv_mod(mod,d);
1.113 noro 5563: chsgnq(ONE,&mone);
1.99 noro 5564: for ( j = 0, sgn = 1; j < n; j++ ) {
1.125 noro 5565: if ( DP_Print ) fprintf(stderr,".",j);
1.99 noro 5566: for ( i = j; i < n && !dm[i][j]; i++ );
5567: if ( i == n ) {
5568: *rp = 0;
5569: return;
5570: }
1.102 noro 5571: k0 = i; l0 = j; len0 = LEN(dm[k0][l0]);
1.99 noro 5572: for ( k = j; k < n; k++ )
5573: for ( l = j; l < n; l++ )
1.102 noro 5574: if ( dm[k][l] && LEN(dm[k][l]) < len0 ) {
5575: k0 = k; l0 = l; len0 = LEN(dm[k][l]);
1.99 noro 5576: }
5577: if ( k0 != j ) {
5578: t = dm[j]; dm[j] = dm[k0]; dm[k0] = t;
5579: sgn = -sgn;
5580: }
5581: if ( l0 != j ) {
5582: for ( k = j; k < n; k++ ) {
5583: s = dm[k][j]; dm[k][j] = dm[k][l0]; dm[k][l0] = s;
5584: }
5585: sgn = -sgn;
5586: }
1.114 noro 5587: bound = nd_det_compute_bound(dm,n,j);
5588: if ( ndl_check_bound(bound,bound) )
5589: nd_det_reconstruct(dm,n,j,d);
5590:
1.99 noro 5591: for ( i = j+1, mj = dm[j], mjj = mj[j]; i < n; i++ ) {
1.114 noro 5592: /* if ( DP_Print ) fprintf(stderr," i=%d\n ",i); */
1.99 noro 5593: mi = dm[i]; mij = mi[j];
1.102 noro 5594: if ( mod )
5595: ndv_mul_c(mod,mij,mod-1);
5596: else
1.113 noro 5597: ndv_mul_c_q(mij,mone);
1.99 noro 5598: for ( k = j+1; k < n; k++ ) {
1.114 noro 5599: /* if ( DP_Print ) fprintf(stderr,"k=%d ",k); */
1.99 noro 5600: bucket = create_pbucket();
1.104 noro 5601: if ( mi[k] ) {
1.102 noro 5602: nmv = BDY(mjj); len = LEN(mjj);
5603: for ( a = 0; a < len; a++, NMV_ADV(nmv) ) {
5604: u = ndv_mul_nmv_trunc(mod,nmv,mi[k],DL(BDY(d)));
5605: add_pbucket(mod,bucket,u);
1.99 noro 5606: }
1.104 noro 5607: }
1.99 noro 5608: if ( mj[k] && mij ) {
1.102 noro 5609: nmv = BDY(mij); len = LEN(mij);
5610: for ( a = 0; a < len; a++, NMV_ADV(nmv) ) {
5611: u = ndv_mul_nmv_trunc(mod,nmv,mj[k],DL(BDY(d)));
5612: add_pbucket(mod,bucket,u);
1.99 noro 5613: }
5614: }
1.104 noro 5615: u = nd_quo(mod,bucket,d);
1.102 noro 5616: mi[k] = ndtondv(mod,u);
1.99 noro 5617: }
1.114 noro 5618: /* if ( DP_Print ) fprintf(stderr,"\n",k); */
1.99 noro 5619: }
5620: d = mjj;
5621: }
1.125 noro 5622: if ( DP_Print ) fprintf(stderr,"\n",k);
1.100 noro 5623: if ( sgn < 0 )
1.102 noro 5624: if ( mod )
5625: ndv_mul_c(mod,d,mod-1);
5626: else
1.113 noro 5627: ndv_mul_c_q(d,mone);
1.126 noro 5628: r = ndvtop(mod,CO,fv,d);
5629: if ( !mod && !UNIQ(dq) )
5630: divsp(CO,r,(P)dq,rp);
5631: else
5632: *rp = r;
1.99 noro 5633: }
5634:
1.102 noro 5635: ND ndv_mul_nmv_trunc(int mod,NMV m0,NDV p,UINT *d)
1.99 noro 5636: {
5637: NM mr,mr0;
1.102 noro 5638: NM tnm;
5639: NMV m;
1.99 noro 5640: UINT *d0,*dt,*dm;
5641: int c,n,td,i,c1,c2,len;
1.113 noro 5642: Q q;
1.99 noro 5643: ND r;
5644:
5645: if ( !p ) return 0;
5646: else {
1.102 noro 5647: n = NV(p); m = BDY(p); len = LEN(p);
1.99 noro 5648: d0 = DL(m0);
1.102 noro 5649: td = TD(d);
1.99 noro 5650: mr0 = 0;
5651: NEWNM(tnm);
1.102 noro 5652: if ( mod ) {
5653: c = CM(m0);
5654: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
5655: ndl_add(DL(m),d0,DL(tnm));
5656: if ( ndl_reducible(DL(tnm),d) ) {
5657: NEXTNM(mr0,mr);
5658: c1 = CM(m); DMAR(c1,c,0,mod,c2); CM(mr) = c2;
1.104 noro 5659: ndl_copy(DL(tnm),DL(mr));
1.102 noro 5660: }
5661: }
5662: } else {
1.113 noro 5663: q = CQ(m0);
1.102 noro 5664: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
5665: ndl_add(DL(m),d0,DL(tnm));
5666: if ( ndl_reducible(DL(tnm),d) ) {
5667: NEXTNM(mr0,mr);
1.113 noro 5668: mulq(CQ(m),q,&CQ(mr));
1.104 noro 5669: ndl_copy(DL(tnm),DL(mr));
1.102 noro 5670: }
1.99 noro 5671: }
5672: }
5673: if ( !mr0 )
5674: return 0;
5675: else {
5676: NEXT(mr) = 0;
1.105 noro 5677: for ( len = 0, mr = mr0; mr; mr = NEXT(mr), len++ );
1.99 noro 5678: MKND(NV(p),mr0,len,r);
5679: SG(r) = SG(p) + TD(d0);
5680: return r;
5681: }
5682: }
1.114 noro 5683: }
5684:
5685: void nd_det_reconstruct(NDV **dm,int n,int j,NDV d)
5686: {
5687: int i,obpe,oadv,h,k,l;
5688: static NM prev_nm_free_list;
5689: EPOS oepos;
5690:
5691: obpe = nd_bpe;
5692: oadv = nmv_adv;
5693: oepos = nd_epos;
5694: if ( obpe < 2 ) nd_bpe = 2;
5695: else if ( obpe < 3 ) nd_bpe = 3;
5696: else if ( obpe < 4 ) nd_bpe = 4;
5697: else if ( obpe < 5 ) nd_bpe = 5;
5698: else if ( obpe < 6 ) nd_bpe = 6;
5699: else if ( obpe < 8 ) nd_bpe = 8;
5700: else if ( obpe < 10 ) nd_bpe = 10;
5701: else if ( obpe < 16 ) nd_bpe = 16;
5702: else if ( obpe < 32 ) nd_bpe = 32;
5703: else error("nd_det_reconstruct : exponent too large");
5704:
5705: nd_setup_parameters(nd_nvar,0);
5706: prev_nm_free_list = _nm_free_list;
5707: _nm_free_list = 0;
5708: for ( k = j; k < n; k++ )
5709: for (l = j; l < n; l++ )
5710: ndv_realloc(dm[k][l],obpe,oadv,oepos);
5711: ndv_realloc(d,obpe,oadv,oepos);
5712: prev_nm_free_list = 0;
5713: #if 0
5714: GC_gcollect();
5715: #endif
5716: }
5717:
5718: UINT *nd_det_compute_bound(NDV **dm,int n,int j)
5719: {
5720: UINT *d0,*d1,*d,*t,*r;
5721: int k,l;
5722:
5723: d0 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
5724: d1 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
5725: for ( k = 0; k < nd_wpd; k++ ) d0[k] = 0;
5726: for ( k = j; k < n; k++ )
5727: for ( l = j; l < n; l++ )
5728: if ( dm[k][l] ) {
5729: d = ndv_compute_bound(dm[k][l]);
5730: ndl_lcm(d,d0,d1);
5731: t = d1; d1 = d0; d0 = t;
5732: }
5733: r = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
5734: for ( k = 0; k < nd_wpd; k++ ) r[k] = d0[k];
5735: return r;
1.117 noro 5736: }
5737:
5738: DL nd_separate_d(UINT *d,UINT *trans)
5739: {
1.121 noro 5740: int n,td,i,e,j;
1.117 noro 5741: DL a;
5742:
5743: ndl_zero(trans);
5744: td = 0;
1.121 noro 5745: for ( i = 0; i < nd_ntrans; i++ ) {
5746: e = GET_EXP(d,i);
5747: PUT_EXP(trans,i,e);
5748: td += MUL_WEIGHT(e,i);
5749: }
5750: if ( nd_ntrans+nd_nalg < nd_nvar ) {
5751: /* homogenized */
5752: i = nd_nvar-1;
1.117 noro 5753: e = GET_EXP(d,i);
5754: PUT_EXP(trans,i,e);
5755: td += MUL_WEIGHT(e,i);
5756: }
5757: TD(trans) = td;
5758: if ( nd_blockmask) ndl_weight_mask(trans);
5759: NEWDL(a,nd_nalg);
5760: td = 0;
1.121 noro 5761: for ( i = 0; i < nd_nalg; i++ ) {
5762: j = nd_ntrans+i;
5763: e = GET_EXP(d,j);
5764: a->d[i] = e;
1.117 noro 5765: td += e;
5766: }
5767: a->td = td;
5768: return a;
5769: }
5770:
1.118 noro 5771: int nd_monic(int mod,ND *p)
1.117 noro 5772: {
5773: UINT *trans,*t;
5774: DL alg;
5775: MP mp0,mp;
1.118 noro 5776: NM m,m0,m1,ma0,ma,mb,mr0,mr;
5777: ND r;
1.117 noro 5778: DL dl;
5779: DP nm;
5780: NDV ndv;
1.118 noro 5781: DAlg inv,cd;
1.117 noro 5782: ND s,c;
1.118 noro 5783: Q l,mul;
5784: N ln;
5785: int n,ntrans,i,e,td,is_lc,len;
5786: NumberField nf;
5787: struct oEGT eg0,eg1;
5788:
5789: if ( !(nf = get_numberfield()) )
5790: error("nd_monic : current_numberfield is not set");
1.117 noro 5791:
1.118 noro 5792: /* Q coef -> DAlg coef */
5793: NEWNM(ma0); ma = ma0;
5794: m = BDY(*p);
5795: is_lc = 1;
5796: while ( 1 ) {
5797: NEWMP(mp0); mp = mp0;
5798: mp->c = (P)CQ(m);
5799: mp->dl = nd_separate_d(DL(m),DL(ma));
5800: NEWNM(mb);
5801: for ( m = NEXT(m); m; m = NEXT(m) ) {
5802: alg = nd_separate_d(DL(m),DL(mb));
5803: if ( !ndl_equal(DL(ma),DL(mb)) )
5804: break;
5805: NEXTMP(mp0,mp); mp->c = (P)CQ(m); mp->dl = alg;
5806: }
5807: NEXT(mp) = 0;
5808: MKDP(nd_nalg,mp0,nm);
5809: MKDAlg(nm,ONE,cd);
5810: if ( is_lc == 1 ) {
5811: /* if the lc is a rational number, we have nothing to do */
5812: if ( !mp0->dl->td )
5813: return 1;
5814:
5815: get_eg(&eg0);
5816: invdalg(cd,&inv);
5817: get_eg(&eg1); add_eg(&eg_invdalg,&eg0,&eg1);
5818: /* check the validity of inv */
5819: if ( mod && !rem(NM(inv->dn),mod) )
5820: return 0;
5821: CA(ma) = nf->one;
5822: is_lc = 0;
5823: ln = ONEN;
5824: } else {
5825: muldalg(cd,inv,&CA(ma));
5826: lcmn(ln,NM(CA(ma)->dn),&ln);
5827: }
5828: if ( m ) {
5829: NEXT(ma) = mb; ma = mb;
5830: } else {
5831: NEXT(ma) = 0;
5832: break;
5833: }
5834: }
5835: /* l = lcm(denoms) */
5836: NTOQ(ln,1,l);
5837: for ( mr0 = 0, m = ma0; m; m = NEXT(m) ) {
5838: divq(l,CA(m)->dn,&mul);
5839: for ( mp = BDY(CA(m)->nm); mp; mp = NEXT(mp) ) {
5840: NEXTNM(mr0,mr);
5841: mulq((Q)mp->c,mul,&CQ(mr));
5842: dl = mp->dl;
5843: td = TD(DL(m));
5844: ndl_copy(DL(m),DL(mr));
1.121 noro 5845: for ( i = 0; i < nd_nalg; i++ ) {
5846: e = dl->d[i];
5847: PUT_EXP(DL(mr),i+nd_ntrans,e);
5848: td += MUL_WEIGHT(e,i+nd_ntrans);
1.118 noro 5849: }
5850: TD(DL(mr)) = td;
5851: if ( nd_blockmask) ndl_weight_mask(DL(mr));
5852: }
1.117 noro 5853: }
1.118 noro 5854: NEXT(mr) = 0;
5855: for ( len = 0, mr = mr0; mr; mr = NEXT(mr), len++ );
5856: MKND(NV(*p),mr0,len,r);
5857: /* XXX */
5858: SG(r) = SG(*p);
5859: nd_free(*p);
5860: *p = r;
5861: return 1;
1.59 noro 5862: }
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>