Annotation of OpenXM_contrib2/asir2000/engine/nd.c, Revision 1.118
1.118 ! noro 1: /* $OpenXM: OpenXM_contrib2/asir2000/engine/nd.c,v 1.117 2004/12/03 08:57:30 noro Exp $ */
1.2 noro 2:
1.94 noro 3: #include "nd.h"
1.63 noro 4:
1.61 noro 5: int (*ndl_compare_function)(UINT *a1,UINT *a2);
1.94 noro 6: int nd_dcomp;
7: NM _nm_free_list;
8: ND _nd_free_list;
9: ND_pairs _ndp_free_list;
1.32 noro 10:
1.117 noro 11: static int nd_nalg;
1.103 noro 12: #if 0
1.74 noro 13: static int ndv_alloc;
1.103 noro 14: #endif
1.87 noro 15: #if 1
1.69 noro 16: static int nd_f4_nsp=0x7fffffff;
1.87 noro 17: #else
18: static int nd_f4_nsp=50;
19: #endif
1.42 noro 20: static double nd_scale=2;
1.61 noro 21: static UINT **nd_bound;
1.42 noro 22: static struct order_spec *nd_ord;
23: static EPOS nd_epos;
1.43 noro 24: static BlockMask nd_blockmask;
1.42 noro 25: static int nd_nvar;
26: static int nd_isrlex;
27: static int nd_epw,nd_bpe,nd_wpd,nd_exporigin;
1.61 noro 28: static UINT nd_mask[32];
29: static UINT nd_mask0,nd_mask1;
1.42 noro 30:
1.20 noro 31: static NDV *nd_ps;
1.53 noro 32: static NDV *nd_ps_trace;
1.42 noro 33: static RHist *nd_psh;
34: static int nd_psn,nd_pslen;
35: static RHist *nd_red;
1.96 noro 36: static int *nd_work_vector;
37: static int **nd_matrix;
38: static int nd_matrix_len;
1.97 noro 39: static struct weight_or_block *nd_worb;
40: static int nd_worb_len;
1.42 noro 41: static int nd_found,nd_create,nd_notfirst;
42: static int nmv_adv;
1.77 noro 43: static int nd_demand;
1.1 noro 44:
1.114 noro 45: UINT *nd_det_compute_bound(NDV **dm,int n,int j);
46: void nd_det_reconstruct(NDV **dm,int n,int j,NDV d);
1.118 ! noro 47: int nd_monic(int m,ND *p);
1.114 noro 48:
1.1 noro 49: void nd_free_private_storage()
50: {
51: _nm_free_list = 0;
1.5 noro 52: _ndp_free_list = 0;
1.71 noro 53: #if 0
1.1 noro 54: GC_gcollect();
1.71 noro 55: #endif
1.1 noro 56: }
57:
58: void _NM_alloc()
59: {
60: NM p;
61: int i;
62:
1.11 noro 63: for ( i = 0; i < 1024; i++ ) {
1.61 noro 64: p = (NM)GC_malloc(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT));
1.1 noro 65: p->next = _nm_free_list; _nm_free_list = p;
66: }
67: }
68:
69: void _ND_alloc()
70: {
71: ND p;
72: int i;
73:
74: for ( i = 0; i < 1024; i++ ) {
75: p = (ND)GC_malloc(sizeof(struct oND));
76: p->body = (NM)_nd_free_list; _nd_free_list = p;
77: }
78: }
79:
80: void _NDP_alloc()
81: {
82: ND_pairs p;
83: int i;
84:
1.11 noro 85: for ( i = 0; i < 1024; i++ ) {
1.1 noro 86: p = (ND_pairs)GC_malloc(sizeof(struct oND_pairs)
1.61 noro 87: +(nd_wpd-1)*sizeof(UINT));
1.1 noro 88: p->next = _ndp_free_list; _ndp_free_list = p;
89: }
90: }
91:
1.30 noro 92: INLINE int nd_length(ND p)
1.1 noro 93: {
94: NM m;
95: int i;
96:
97: if ( !p )
98: return 0;
99: else {
100: for ( i = 0, m = BDY(p); m; m = NEXT(m), i++ );
101: return i;
102: }
103: }
104:
1.61 noro 105: INLINE int ndl_reducible(UINT *d1,UINT *d2)
1.1 noro 106: {
1.61 noro 107: UINT u1,u2;
1.1 noro 108: int i,j;
109:
1.34 noro 110: if ( TD(d1) < TD(d2) ) return 0;
1.65 noro 111: #if USE_UNROLL
1.1 noro 112: switch ( nd_bpe ) {
1.62 noro 113: case 3:
114: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
115: u1 = d1[i]; u2 = d2[i];
116: if ( (u1&0x38000000) < (u2&0x38000000) ) return 0;
117: if ( (u1& 0x7000000) < (u2& 0x7000000) ) return 0;
118: if ( (u1& 0xe00000) < (u2& 0xe00000) ) return 0;
119: if ( (u1& 0x1c0000) < (u2& 0x1c0000) ) return 0;
120: if ( (u1& 0x38000) < (u2& 0x38000) ) return 0;
121: if ( (u1& 0x7000) < (u2& 0x7000) ) return 0;
122: if ( (u1& 0xe00) < (u2& 0xe00) ) return 0;
123: if ( (u1& 0x1c0) < (u2& 0x1c0) ) return 0;
124: if ( (u1& 0x38) < (u2& 0x38) ) return 0;
125: if ( (u1& 0x7) < (u2& 0x7) ) return 0;
126: }
127: return 1;
128: break;
1.1 noro 129: case 4:
1.41 noro 130: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 131: u1 = d1[i]; u2 = d2[i];
132: if ( (u1&0xf0000000) < (u2&0xf0000000) ) return 0;
1.62 noro 133: if ( (u1& 0xf000000) < (u2& 0xf000000) ) return 0;
134: if ( (u1& 0xf00000) < (u2& 0xf00000) ) return 0;
135: if ( (u1& 0xf0000) < (u2& 0xf0000) ) return 0;
136: if ( (u1& 0xf000) < (u2& 0xf000) ) return 0;
137: if ( (u1& 0xf00) < (u2& 0xf00) ) return 0;
138: if ( (u1& 0xf0) < (u2& 0xf0) ) return 0;
139: if ( (u1& 0xf) < (u2& 0xf) ) return 0;
1.1 noro 140: }
141: return 1;
142: break;
143: case 6:
1.41 noro 144: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 145: u1 = d1[i]; u2 = d2[i];
146: if ( (u1&0x3f000000) < (u2&0x3f000000) ) return 0;
1.62 noro 147: if ( (u1& 0xfc0000) < (u2& 0xfc0000) ) return 0;
148: if ( (u1& 0x3f000) < (u2& 0x3f000) ) return 0;
149: if ( (u1& 0xfc0) < (u2& 0xfc0) ) return 0;
150: if ( (u1& 0x3f) < (u2& 0x3f) ) return 0;
1.1 noro 151: }
152: return 1;
153: break;
154: case 8:
1.41 noro 155: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 156: u1 = d1[i]; u2 = d2[i];
157: if ( (u1&0xff000000) < (u2&0xff000000) ) return 0;
1.62 noro 158: if ( (u1& 0xff0000) < (u2& 0xff0000) ) return 0;
159: if ( (u1& 0xff00) < (u2& 0xff00) ) return 0;
160: if ( (u1& 0xff) < (u2& 0xff) ) return 0;
1.1 noro 161: }
162: return 1;
163: break;
164: case 16:
1.41 noro 165: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 166: u1 = d1[i]; u2 = d2[i];
167: if ( (u1&0xffff0000) < (u2&0xffff0000) ) return 0;
1.62 noro 168: if ( (u1& 0xffff) < (u2& 0xffff) ) return 0;
1.1 noro 169: }
170: return 1;
171: break;
172: case 32:
1.41 noro 173: for ( i = nd_exporigin; i < nd_wpd; i++ )
1.1 noro 174: if ( d1[i] < d2[i] ) return 0;
175: return 1;
176: break;
177: default:
1.41 noro 178: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 179: u1 = d1[i]; u2 = d2[i];
180: for ( j = 0; j < nd_epw; j++ )
181: if ( (u1&nd_mask[j]) < (u2&nd_mask[j]) ) return 0;
182: }
183: return 1;
184: }
1.65 noro 185: #else
186: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
187: u1 = d1[i]; u2 = d2[i];
188: for ( j = 0; j < nd_epw; j++ )
189: if ( (u1&nd_mask[j]) < (u2&nd_mask[j]) ) return 0;
190: }
191: return 1;
192: #endif
1.1 noro 193: }
194:
1.61 noro 195: /*
196: * If the current order is a block order,
197: * then the last block is length 1 and contains
198: * the homo variable. Otherwise, the original
199: * order is either 0 or 2.
200: */
201:
202: void ndl_homogenize(UINT *d,UINT *r,int obpe,EPOS oepos,int weight)
1.23 noro 203: {
1.61 noro 204: int w,i,e,n,omask0;
205:
206: omask0 = (1<<obpe)-1;
207: n = nd_nvar-1;
208: ndl_zero(r);
209: for ( i = 0; i < n; i++ ) {
210: e = GET_EXP_OLD(d,i);
211: PUT_EXP(r,i,e);
212: }
213: w = TD(d);
214: PUT_EXP(r,nd_nvar-1,weight-w);
215: TD(r) = weight;
216: if ( nd_blockmask ) ndl_weight_mask(r);
217: }
218:
219: void ndl_dehomogenize(UINT *d)
220: {
221: UINT mask;
222: UINT h;
1.31 noro 223: int i,bits;
1.23 noro 224:
1.44 noro 225: if ( nd_blockmask ) {
226: h = GET_EXP(d,nd_nvar-1);
1.45 noro 227: XOR_EXP(d,nd_nvar-1,h);
1.44 noro 228: TD(d) -= h;
229: d[nd_exporigin-1] -= h;
230: } else {
231: if ( nd_isrlex ) {
232: if ( nd_bpe == 32 ) {
233: h = d[nd_exporigin];
234: for ( i = nd_exporigin+1; i < nd_wpd; i++ )
235: d[i-1] = d[i];
236: d[i-1] = 0;
237: TD(d) -= h;
238: } else {
239: bits = nd_epw*nd_bpe;
240: mask = bits==32?0xffffffff:((1<<(nd_epw*nd_bpe))-1);
241: h = (d[nd_exporigin]>>((nd_epw-1)*nd_bpe))&nd_mask0;
242: for ( i = nd_exporigin; i < nd_wpd; i++ )
243: d[i] = ((d[i]<<nd_bpe)&mask)
244: |(i+1<nd_wpd?((d[i+1]>>((nd_epw-1)*nd_bpe))&nd_mask0):0);
245: TD(d) -= h;
246: }
1.45 noro 247: } else {
248: h = GET_EXP(d,nd_nvar-1);
249: XOR_EXP(d,nd_nvar-1,h);
250: TD(d) -= h;
251: }
1.44 noro 252: }
1.23 noro 253: }
254:
1.61 noro 255: void ndl_lcm(UINT *d1,unsigned *d2,UINT *d)
1.1 noro 256: {
1.61 noro 257: UINT t1,t2,u,u1,u2;
1.43 noro 258: int i,j,l;
1.1 noro 259:
1.65 noro 260: #if USE_UNROLL
1.1 noro 261: switch ( nd_bpe ) {
1.62 noro 262: case 3:
263: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
264: u1 = d1[i]; u2 = d2[i];
265: t1 = (u1&0x38000000); t2 = (u2&0x38000000); u = t1>t2?t1:t2;
266: t1 = (u1& 0x7000000); t2 = (u2& 0x7000000); u |= t1>t2?t1:t2;
267: t1 = (u1& 0xe00000); t2 = (u2& 0xe00000); u |= t1>t2?t1:t2;
268: t1 = (u1& 0x1c0000); t2 = (u2& 0x1c0000); u |= t1>t2?t1:t2;
269: t1 = (u1& 0x38000); t2 = (u2& 0x38000); u |= t1>t2?t1:t2;
270: t1 = (u1& 0x7000); t2 = (u2& 0x7000); u |= t1>t2?t1:t2;
271: t1 = (u1& 0xe00); t2 = (u2& 0xe00); u |= t1>t2?t1:t2;
272: t1 = (u1& 0x1c0); t2 = (u2& 0x1c0); u |= t1>t2?t1:t2;
273: t1 = (u1& 0x38); t2 = (u2& 0x38); u |= t1>t2?t1:t2;
274: t1 = (u1& 0x7); t2 = (u2& 0x7); u |= t1>t2?t1:t2;
275: d[i] = u;
276: }
277: break;
1.1 noro 278: case 4:
1.41 noro 279: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 280: u1 = d1[i]; u2 = d2[i];
281: t1 = (u1&0xf0000000); t2 = (u2&0xf0000000); u = t1>t2?t1:t2;
1.62 noro 282: t1 = (u1& 0xf000000); t2 = (u2& 0xf000000); u |= t1>t2?t1:t2;
283: t1 = (u1& 0xf00000); t2 = (u2& 0xf00000); u |= t1>t2?t1:t2;
284: t1 = (u1& 0xf0000); t2 = (u2& 0xf0000); u |= t1>t2?t1:t2;
285: t1 = (u1& 0xf000); t2 = (u2& 0xf000); u |= t1>t2?t1:t2;
286: t1 = (u1& 0xf00); t2 = (u2& 0xf00); u |= t1>t2?t1:t2;
287: t1 = (u1& 0xf0); t2 = (u2& 0xf0); u |= t1>t2?t1:t2;
288: t1 = (u1& 0xf); t2 = (u2& 0xf); u |= t1>t2?t1:t2;
1.1 noro 289: d[i] = u;
290: }
291: break;
292: case 6:
1.41 noro 293: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 294: u1 = d1[i]; u2 = d2[i];
295: t1 = (u1&0x3f000000); t2 = (u2&0x3f000000); u = t1>t2?t1:t2;
1.62 noro 296: t1 = (u1& 0xfc0000); t2 = (u2& 0xfc0000); u |= t1>t2?t1:t2;
297: t1 = (u1& 0x3f000); t2 = (u2& 0x3f000); u |= t1>t2?t1:t2;
298: t1 = (u1& 0xfc0); t2 = (u2& 0xfc0); u |= t1>t2?t1:t2;
299: t1 = (u1& 0x3f); t2 = (u2& 0x3f); u |= t1>t2?t1:t2;
1.1 noro 300: d[i] = u;
301: }
302: break;
303: case 8:
1.41 noro 304: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 305: u1 = d1[i]; u2 = d2[i];
306: t1 = (u1&0xff000000); t2 = (u2&0xff000000); u = t1>t2?t1:t2;
1.62 noro 307: t1 = (u1& 0xff0000); t2 = (u2& 0xff0000); u |= t1>t2?t1:t2;
308: t1 = (u1& 0xff00); t2 = (u2& 0xff00); u |= t1>t2?t1:t2;
309: t1 = (u1& 0xff); t2 = (u2& 0xff); u |= t1>t2?t1:t2;
1.1 noro 310: d[i] = u;
311: }
312: break;
313: case 16:
1.41 noro 314: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 315: u1 = d1[i]; u2 = d2[i];
316: t1 = (u1&0xffff0000); t2 = (u2&0xffff0000); u = t1>t2?t1:t2;
1.62 noro 317: t1 = (u1& 0xffff); t2 = (u2& 0xffff); u |= t1>t2?t1:t2;
1.1 noro 318: d[i] = u;
319: }
320: break;
321: case 32:
1.41 noro 322: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 323: u1 = d1[i]; u2 = d2[i];
324: d[i] = u1>u2?u1:u2;
325: }
326: break;
327: default:
1.41 noro 328: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 329: u1 = d1[i]; u2 = d2[i];
330: for ( j = 0, u = 0; j < nd_epw; j++ ) {
331: t1 = (u1&nd_mask[j]); t2 = (u2&nd_mask[j]); u |= t1>t2?t1:t2;
332: }
333: d[i] = u;
334: }
335: break;
336: }
1.65 noro 337: #else
338: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
339: u1 = d1[i]; u2 = d2[i];
340: for ( j = 0, u = 0; j < nd_epw; j++ ) {
341: t1 = (u1&nd_mask[j]); t2 = (u2&nd_mask[j]); u |= t1>t2?t1:t2;
342: }
343: d[i] = u;
344: }
345: #endif
1.39 noro 346: TD(d) = ndl_weight(d);
1.61 noro 347: if ( nd_blockmask ) ndl_weight_mask(d);
1.57 noro 348: }
349:
1.61 noro 350: int ndl_weight(UINT *d)
1.1 noro 351: {
1.61 noro 352: UINT t,u;
1.1 noro 353: int i,j;
354:
1.60 noro 355: if ( current_dl_weight_vector )
356: for ( i = 0, t = 0; i < nd_nvar; i++ ) {
357: u = GET_EXP(d,i);
358: t += MUL_WEIGHT(u,i);
359: }
360: else
361: for ( t = 0, i = nd_exporigin; i < nd_wpd; i++ ) {
362: u = d[i];
363: for ( j = 0; j < nd_epw; j++, u>>=nd_bpe )
364: t += (u&nd_mask0);
365: }
1.1 noro 366: return t;
367: }
368:
1.61 noro 369: void ndl_weight_mask(UINT *d)
1.43 noro 370: {
1.61 noro 371: UINT t,u;
372: UINT *mask;
373: int i,j,k,l;
1.43 noro 374:
1.61 noro 375: l = nd_blockmask->n;
376: for ( k = 0; k < l; k++ ) {
377: mask = nd_blockmask->mask[k];
378: if ( current_dl_weight_vector )
379: for ( i = 0, t = 0; i < nd_nvar; i++ ) {
380: u = GET_EXP_MASK(d,i,mask);
381: t += MUL_WEIGHT(u,i);
382: }
383: else
384: for ( t = 0, i = nd_exporigin; i < nd_wpd; i++ ) {
385: u = d[i]&mask[i];
386: for ( j = 0; j < nd_epw; j++, u>>=nd_bpe )
387: t += (u&nd_mask0);
388: }
389: d[k+1] = t;
390: }
1.43 noro 391: }
392:
1.61 noro 393: int ndl_lex_compare(UINT *d1,UINT *d2)
1.1 noro 394: {
395: int i;
396:
1.41 noro 397: d1 += nd_exporigin;
398: d2 += nd_exporigin;
399: for ( i = nd_exporigin; i < nd_wpd; i++, d1++, d2++ )
1.1 noro 400: if ( *d1 > *d2 )
1.32 noro 401: return nd_isrlex ? -1 : 1;
1.1 noro 402: else if ( *d1 < *d2 )
1.32 noro 403: return nd_isrlex ? 1 : -1;
1.1 noro 404: return 0;
405: }
406:
1.61 noro 407: int ndl_block_compare(UINT *d1,UINT *d2)
1.43 noro 408: {
409: int i,l,j,ord_o,ord_l;
410: struct order_pair *op;
1.61 noro 411: UINT t1,t2,m;
412: UINT *mask;
1.43 noro 413:
414: l = nd_blockmask->n;
415: op = nd_blockmask->order_pair;
416: for ( j = 0; j < l; j++ ) {
417: mask = nd_blockmask->mask[j];
418: ord_o = op[j].order;
419: if ( ord_o < 2 )
1.44 noro 420: if ( (t1=d1[j+1]) > (t2=d2[j+1]) ) return 1;
421: else if ( t1 < t2 ) return -1;
1.43 noro 422: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.44 noro 423: m = mask[i];
424: t1 = d1[i]&m;
425: t2 = d2[i]&m;
1.43 noro 426: if ( t1 > t2 )
427: return !ord_o ? -1 : 1;
428: else if ( t1 < t2 )
429: return !ord_o ? 1 : -1;
430: }
431: }
432: return 0;
433: }
434:
1.96 noro 435: int ndl_matrix_compare(UINT *d1,UINT *d2)
436: {
437: int i,j,s;
438: int *v;
439:
440: for ( j = 0; j < nd_nvar; j++ )
441: nd_work_vector[j] = GET_EXP(d1,j)-GET_EXP(d2,j);
442: for ( i = 0; i < nd_matrix_len; i++ ) {
443: v = nd_matrix[i];
444: for ( j = 0, s = 0; j < nd_nvar; j++ )
445: s += v[j]*nd_work_vector[j];
446: if ( s > 0 ) return 1;
447: else if ( s < 0 ) return -1;
448: }
449: return 0;
450: }
451:
1.97 noro 452: int ndl_composite_compare(UINT *d1,UINT *d2)
453: {
454: int i,j,s,start,end,len,o;
455: int *v;
456: struct sparse_weight *sw;
457:
458: for ( j = 0; j < nd_nvar; j++ )
459: nd_work_vector[j] = GET_EXP(d1,j)-GET_EXP(d2,j);
460: for ( i = 0; i < nd_worb_len; i++ ) {
461: len = nd_worb[i].length;
462: switch ( nd_worb[i].type ) {
463: case IS_DENSE_WEIGHT:
464: v = nd_worb[i].body.dense_weight;
465: for ( j = 0, s = 0; j < len; j++ )
466: s += v[j]*nd_work_vector[j];
467: if ( s > 0 ) return 1;
468: else if ( s < 0 ) return -1;
469: break;
470: case IS_SPARSE_WEIGHT:
471: sw = nd_worb[i].body.sparse_weight;
472: for ( j = 0, s = 0; j < len; j++ )
473: s += sw[j].value*nd_work_vector[sw[j].pos];
474: if ( s > 0 ) return 1;
475: else if ( s < 0 ) return -1;
476: break;
477: case IS_BLOCK:
478: o = nd_worb[i].body.block.order;
479: start = nd_worb[i].body.block.start;
480: switch ( o ) {
481: case 0:
482: end = start+len;
483: for ( j = start, s = 0; j < end; j++ )
484: s += MUL_WEIGHT(nd_work_vector[j],j);
485: if ( s > 0 ) return 1;
486: else if ( s < 0 ) return -1;
487: for ( j = end-1; j >= start; j-- )
488: if ( nd_work_vector[j] < 0 ) return 1;
489: else if ( nd_work_vector[j] > 0 ) return -1;
490: break;
491: case 1:
492: end = start+len;
493: for ( j = start, s = 0; j < end; j++ )
494: s += MUL_WEIGHT(nd_work_vector[j],j);
495: if ( s > 0 ) return 1;
496: else if ( s < 0 ) return -1;
497: for ( j = start; j < end; j++ )
498: if ( nd_work_vector[j] > 0 ) return 1;
499: else if ( nd_work_vector[j] < 0 ) return -1;
500: break;
501: case 2:
502: for ( j = start; j < end; j++ )
503: if ( nd_work_vector[j] > 0 ) return 1;
504: else if ( nd_work_vector[j] < 0 ) return -1;
505: break;
506: }
507: break;
508: }
509: }
510: return 0;
511: }
512:
1.58 noro 513: /* TDH -> WW -> TD-> RL */
514:
1.61 noro 515: int ndl_ww_lex_compare(UINT *d1,UINT *d2)
1.58 noro 516: {
517: int i,m,e1,e2;
518:
519: if ( TD(d1) > TD(d2) ) return 1;
520: else if ( TD(d1) < TD(d2) ) return -1;
521: m = nd_nvar>>1;
522: for ( i = 0, e1 = e2 = 0; i < m; i++ ) {
523: e1 += current_weyl_weight_vector[i]*(GET_EXP(d1,m+i)-GET_EXP(d1,i));
524: e2 += current_weyl_weight_vector[i]*(GET_EXP(d2,m+i)-GET_EXP(d2,i));
525: }
526: if ( e1 > e2 ) return 1;
527: else if ( e1 < e2 ) return -1;
528: return ndl_lex_compare(d1,d2);
529: }
530:
1.61 noro 531: INLINE int ndl_equal(UINT *d1,UINT *d2)
1.1 noro 532: {
533: int i;
534:
1.81 noro 535: switch ( nd_wpd ) {
536: case 2:
537: if ( TD(d2) != TD(d1) ) return 0;
538: if ( d2[1] != d1[1] ) return 0;
539: return 1;
540: break;
541: case 3:
542: if ( TD(d2) != TD(d1) ) return 0;
543: if ( d2[1] != d1[1] ) return 0;
544: if ( d2[2] != d1[2] ) return 0;
545: return 1;
546: break;
547: default:
548: for ( i = 0; i < nd_wpd; i++ )
549: if ( *d1++ != *d2++ ) return 0;
550: return 1;
551: break;
552: }
1.1 noro 553: }
554:
1.61 noro 555: INLINE void ndl_copy(UINT *d1,UINT *d2)
1.6 noro 556: {
557: int i;
558:
559: switch ( nd_wpd ) {
1.41 noro 560: case 2:
1.34 noro 561: TD(d2) = TD(d1);
562: d2[1] = d1[1];
1.6 noro 563: break;
1.41 noro 564: case 3:
1.34 noro 565: TD(d2) = TD(d1);
1.6 noro 566: d2[1] = d1[1];
1.34 noro 567: d2[2] = d1[2];
1.6 noro 568: break;
569: default:
1.41 noro 570: for ( i = 0; i < nd_wpd; i++ )
1.6 noro 571: d2[i] = d1[i];
572: break;
573: }
574: }
575:
1.61 noro 576: INLINE void ndl_zero(UINT *d)
577: {
578: int i;
579: for ( i = 0; i < nd_wpd; i++ ) d[i] = 0;
580: }
581:
582: INLINE void ndl_add(UINT *d1,UINT *d2,UINT *d)
1.1 noro 583: {
584: int i;
585:
1.43 noro 586: #if 1
1.6 noro 587: switch ( nd_wpd ) {
1.41 noro 588: case 2:
589: TD(d) = TD(d1)+TD(d2);
1.34 noro 590: d[1] = d1[1]+d2[1];
1.6 noro 591: break;
1.41 noro 592: case 3:
593: TD(d) = TD(d1)+TD(d2);
1.6 noro 594: d[1] = d1[1]+d2[1];
1.34 noro 595: d[2] = d1[2]+d2[2];
1.6 noro 596: break;
597: default:
1.43 noro 598: for ( i = 0; i < nd_wpd; i++ ) d[i] = d1[i]+d2[i];
1.6 noro 599: break;
600: }
1.43 noro 601: #else
602: for ( i = 0; i < nd_wpd; i++ ) d[i] = d1[i]+d2[i];
603: #endif
1.6 noro 604: }
605:
1.55 noro 606: /* d1 += d2 */
1.61 noro 607: INLINE void ndl_addto(UINT *d1,UINT *d2)
1.55 noro 608: {
609: int i;
610:
611: #if 1
612: switch ( nd_wpd ) {
613: case 2:
614: TD(d1) += TD(d2);
615: d1[1] += d2[1];
616: break;
617: case 3:
618: TD(d1) += TD(d2);
619: d1[1] += d2[1];
620: d1[2] += d2[2];
621: break;
622: default:
623: for ( i = 0; i < nd_wpd; i++ ) d1[i] += d2[i];
624: break;
625: }
626: #else
627: for ( i = 0; i < nd_wpd; i++ ) d1[i] += d2[i];
628: #endif
629: }
630:
1.61 noro 631: INLINE void ndl_sub(UINT *d1,UINT *d2,UINT *d)
1.6 noro 632: {
633: int i;
634:
1.43 noro 635: for ( i = 0; i < nd_wpd; i++ ) d[i] = d1[i]-d2[i];
1.1 noro 636: }
637:
1.61 noro 638: int ndl_disjoint(UINT *d1,UINT *d2)
1.1 noro 639: {
1.61 noro 640: UINT t1,t2,u,u1,u2;
1.1 noro 641: int i,j;
642:
1.65 noro 643: #if USE_UNROLL
1.1 noro 644: switch ( nd_bpe ) {
1.62 noro 645: case 3:
646: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
647: u1 = d1[i]; u2 = d2[i];
648: t1 = u1&0x38000000; t2 = u2&0x38000000; if ( t1&&t2 ) return 0;
649: t1 = u1& 0x7000000; t2 = u2& 0x7000000; if ( t1&&t2 ) return 0;
650: t1 = u1& 0xe00000; t2 = u2& 0xe00000; if ( t1&&t2 ) return 0;
651: t1 = u1& 0x1c0000; t2 = u2& 0x1c0000; if ( t1&&t2 ) return 0;
652: t1 = u1& 0x38000; t2 = u2& 0x38000; if ( t1&&t2 ) return 0;
653: t1 = u1& 0x7000; t2 = u2& 0x7000; if ( t1&&t2 ) return 0;
654: t1 = u1& 0xe00; t2 = u2& 0xe00; if ( t1&&t2 ) return 0;
655: t1 = u1& 0x1c0; t2 = u2& 0x1c0; if ( t1&&t2 ) return 0;
656: t1 = u1& 0x38; t2 = u2& 0x38; if ( t1&&t2 ) return 0;
657: t1 = u1& 0x7; t2 = u2& 0x7; if ( t1&&t2 ) return 0;
658: }
659: return 1;
660: break;
1.1 noro 661: case 4:
1.41 noro 662: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 663: u1 = d1[i]; u2 = d2[i];
664: t1 = u1&0xf0000000; t2 = u2&0xf0000000; if ( t1&&t2 ) return 0;
1.62 noro 665: t1 = u1& 0xf000000; t2 = u2& 0xf000000; if ( t1&&t2 ) return 0;
666: t1 = u1& 0xf00000; t2 = u2& 0xf00000; if ( t1&&t2 ) return 0;
667: t1 = u1& 0xf0000; t2 = u2& 0xf0000; if ( t1&&t2 ) return 0;
668: t1 = u1& 0xf000; t2 = u2& 0xf000; if ( t1&&t2 ) return 0;
669: t1 = u1& 0xf00; t2 = u2& 0xf00; if ( t1&&t2 ) return 0;
670: t1 = u1& 0xf0; t2 = u2& 0xf0; if ( t1&&t2 ) return 0;
671: t1 = u1& 0xf; t2 = u2& 0xf; if ( t1&&t2 ) return 0;
1.1 noro 672: }
673: return 1;
674: break;
675: case 6:
1.41 noro 676: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 677: u1 = d1[i]; u2 = d2[i];
678: t1 = u1&0x3f000000; t2 = u2&0x3f000000; if ( t1&&t2 ) return 0;
1.62 noro 679: t1 = u1& 0xfc0000; t2 = u2& 0xfc0000; if ( t1&&t2 ) return 0;
680: t1 = u1& 0x3f000; t2 = u2& 0x3f000; if ( t1&&t2 ) return 0;
681: t1 = u1& 0xfc0; t2 = u2& 0xfc0; if ( t1&&t2 ) return 0;
682: t1 = u1& 0x3f; t2 = u2& 0x3f; if ( t1&&t2 ) return 0;
1.1 noro 683: }
684: return 1;
685: break;
686: case 8:
1.41 noro 687: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 688: u1 = d1[i]; u2 = d2[i];
689: t1 = u1&0xff000000; t2 = u2&0xff000000; if ( t1&&t2 ) return 0;
1.62 noro 690: t1 = u1& 0xff0000; t2 = u2& 0xff0000; if ( t1&&t2 ) return 0;
691: t1 = u1& 0xff00; t2 = u2& 0xff00; if ( t1&&t2 ) return 0;
692: t1 = u1& 0xff; t2 = u2& 0xff; if ( t1&&t2 ) return 0;
1.1 noro 693: }
694: return 1;
695: break;
696: case 16:
1.41 noro 697: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 698: u1 = d1[i]; u2 = d2[i];
699: t1 = u1&0xffff0000; t2 = u2&0xffff0000; if ( t1&&t2 ) return 0;
1.62 noro 700: t1 = u1& 0xffff; t2 = u2& 0xffff; if ( t1&&t2 ) return 0;
1.1 noro 701: }
702: return 1;
703: break;
704: case 32:
1.41 noro 705: for ( i = nd_exporigin; i < nd_wpd; i++ )
1.1 noro 706: if ( d1[i] && d2[i] ) return 0;
707: return 1;
708: break;
709: default:
1.41 noro 710: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 711: u1 = d1[i]; u2 = d2[i];
712: for ( j = 0; j < nd_epw; j++ ) {
713: if ( (u1&nd_mask0) && (u2&nd_mask0) ) return 0;
714: u1 >>= nd_bpe; u2 >>= nd_bpe;
715: }
716: }
717: return 1;
718: break;
719: }
1.65 noro 720: #else
721: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
722: u1 = d1[i]; u2 = d2[i];
723: for ( j = 0; j < nd_epw; j++ ) {
724: if ( (u1&nd_mask0) && (u2&nd_mask0) ) return 0;
725: u1 >>= nd_bpe; u2 >>= nd_bpe;
726: }
727: }
728: return 1;
729: #endif
1.1 noro 730: }
731:
1.114 noro 732: int ndl_check_bound(UINT *d1,UINT *d2)
1.1 noro 733: {
1.61 noro 734: UINT u2;
1.5 noro 735: int i,j,ind,k;
1.1 noro 736:
1.5 noro 737: ind = 0;
1.65 noro 738: #if USE_UNROLL
1.5 noro 739: switch ( nd_bpe ) {
1.62 noro 740: case 3:
741: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
742: u2 = d2[i];
743: if ( d1[ind++]+((u2>>27)&0x7) >= 0x8 ) return 1;
744: if ( d1[ind++]+((u2>>24)&0x7) >= 0x8 ) return 1;
745: if ( d1[ind++]+((u2>>21)&0x7) >= 0x8 ) return 1;
746: if ( d1[ind++]+((u2>>18)&0x7) >= 0x8 ) return 1;
747: if ( d1[ind++]+((u2>>15)&0x7) >= 0x8 ) return 1;
748: if ( d1[ind++]+((u2>>12)&0x7) >= 0x8 ) return 1;
749: if ( d1[ind++]+((u2>>9)&0x7) >= 0x8 ) return 1;
750: if ( d1[ind++]+((u2>>6)&0x7) >= 0x8 ) return 1;
751: if ( d1[ind++]+((u2>>3)&0x7) >= 0x8 ) return 1;
752: if ( d1[ind++]+(u2&0x7) >= 0x8 ) return 1;
753: }
754: return 0;
755: break;
1.5 noro 756: case 4:
1.41 noro 757: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 758: u2 = d2[i];
759: if ( d1[ind++]+((u2>>28)&0xf) >= 0x10 ) return 1;
760: if ( d1[ind++]+((u2>>24)&0xf) >= 0x10 ) return 1;
761: if ( d1[ind++]+((u2>>20)&0xf) >= 0x10 ) return 1;
762: if ( d1[ind++]+((u2>>16)&0xf) >= 0x10 ) return 1;
763: if ( d1[ind++]+((u2>>12)&0xf) >= 0x10 ) return 1;
764: if ( d1[ind++]+((u2>>8)&0xf) >= 0x10 ) return 1;
765: if ( d1[ind++]+((u2>>4)&0xf) >= 0x10 ) return 1;
766: if ( d1[ind++]+(u2&0xf) >= 0x10 ) return 1;
767: }
768: return 0;
769: break;
770: case 6:
1.41 noro 771: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 772: u2 = d2[i];
773: if ( d1[ind++]+((u2>>24)&0x3f) >= 0x40 ) return 1;
774: if ( d1[ind++]+((u2>>18)&0x3f) >= 0x40 ) return 1;
775: if ( d1[ind++]+((u2>>12)&0x3f) >= 0x40 ) return 1;
776: if ( d1[ind++]+((u2>>6)&0x3f) >= 0x40 ) return 1;
777: if ( d1[ind++]+(u2&0x3f) >= 0x40 ) return 1;
778: }
779: return 0;
780: break;
781: case 8:
1.41 noro 782: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 783: u2 = d2[i];
784: if ( d1[ind++]+((u2>>24)&0xff) >= 0x100 ) return 1;
785: if ( d1[ind++]+((u2>>16)&0xff) >= 0x100 ) return 1;
786: if ( d1[ind++]+((u2>>8)&0xff) >= 0x100 ) return 1;
787: if ( d1[ind++]+(u2&0xff) >= 0x100 ) return 1;
788: }
789: return 0;
790: break;
791: case 16:
1.41 noro 792: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 793: u2 = d2[i];
794: if ( d1[ind++]+((u2>>16)&0xffff) > 0x10000 ) return 1;
795: if ( d1[ind++]+(u2&0xffff) > 0x10000 ) return 1;
796: }
797: return 0;
798: break;
799: case 32:
1.41 noro 800: for ( i = nd_exporigin; i < nd_wpd; i++ )
1.5 noro 801: if ( d1[i]+d2[i]<d1[i] ) return 1;
802: return 0;
803: break;
804: default:
1.41 noro 805: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 806: u2 = d2[i];
807: k = (nd_epw-1)*nd_bpe;
808: for ( j = 0; j < nd_epw; j++, k -= nd_bpe )
809: if ( d1[ind++]+((u2>>k)&nd_mask0) > nd_mask0 ) return 1;
810: }
811: return 0;
812: break;
813: }
1.65 noro 814: #else
815: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
816: u2 = d2[i];
817: k = (nd_epw-1)*nd_bpe;
818: for ( j = 0; j < nd_epw; j++, k -= nd_bpe )
819: if ( d1[ind++]+((u2>>k)&nd_mask0) > nd_mask0 ) return 1;
820: }
821: return 0;
822: #endif
1.1 noro 823: }
824:
1.114 noro 825: int ndl_check_bound2(int index,UINT *d2)
826: {
827: return ndl_check_bound(nd_bound[index],d2);
828: }
829:
1.61 noro 830: INLINE int ndl_hash_value(UINT *d)
1.1 noro 831: {
832: int i;
833: int r;
834:
1.34 noro 835: r = 0;
1.41 noro 836: for ( i = 0; i < nd_wpd; i++ )
1.1 noro 837: r = ((r<<16)+d[i])%REDTAB_LEN;
838: return r;
839: }
840:
1.63 noro 841: INLINE int ndl_find_reducer(UINT *dg)
1.1 noro 842: {
1.13 noro 843: RHist r;
1.6 noro 844: int d,k,i;
1.1 noro 845:
1.63 noro 846: d = ndl_hash_value(dg);
1.13 noro 847: for ( r = nd_red[d], k = 0; r; r = NEXT(r), k++ ) {
1.34 noro 848: if ( ndl_equal(dg,DL(r)) ) {
1.1 noro 849: if ( k > 0 ) nd_notfirst++;
850: nd_found++;
1.13 noro 851: return r->index;
1.1 noro 852: }
853: }
1.13 noro 854: if ( Reverse )
855: for ( i = nd_psn-1; i >= 0; i-- ) {
856: r = nd_psh[i];
1.34 noro 857: if ( ndl_reducible(dg,DL(r)) ) {
1.13 noro 858: nd_create++;
1.34 noro 859: nd_append_red(dg,i);
1.13 noro 860: return i;
861: }
862: }
863: else
864: for ( i = 0; i < nd_psn; i++ ) {
865: r = nd_psh[i];
1.34 noro 866: if ( ndl_reducible(dg,DL(r)) ) {
1.13 noro 867: nd_create++;
1.34 noro 868: nd_append_red(dg,i);
1.13 noro 869: return i;
870: }
1.1 noro 871: }
1.6 noro 872: return -1;
1.1 noro 873: }
874:
1.63 noro 875: ND nd_merge(ND p1,ND p2)
876: {
877: int n,c;
878: int t,can,td1,td2;
879: ND r;
880: NM m1,m2,mr0,mr,s;
881:
882: if ( !p1 ) return p2;
883: else if ( !p2 ) return p1;
884: else {
885: can = 0;
886: for ( n = NV(p1), m1 = BDY(p1), m2 = BDY(p2), mr0 = 0; m1 && m2; ) {
887: c = DL_COMPARE(DL(m1),DL(m2));
888: switch ( c ) {
889: case 0:
890: s = m1; m1 = NEXT(m1);
891: can++; NEXTNM2(mr0,mr,s);
892: s = m2; m2 = NEXT(m2); FREENM(s);
893: break;
894: case 1:
895: s = m1; m1 = NEXT(m1); NEXTNM2(mr0,mr,s);
896: break;
897: case -1:
898: s = m2; m2 = NEXT(m2); NEXTNM2(mr0,mr,s);
899: break;
900: }
901: }
902: if ( !mr0 )
903: if ( m1 ) mr0 = m1;
904: else if ( m2 ) mr0 = m2;
905: else return 0;
906: else if ( m1 ) NEXT(mr) = m1;
907: else if ( m2 ) NEXT(mr) = m2;
908: else NEXT(mr) = 0;
909: BDY(p1) = mr0;
910: SG(p1) = MAX(SG(p1),SG(p2));
911: LEN(p1) = LEN(p1)+LEN(p2)-can;
912: FREEND(p2);
913: return p1;
914: }
915: }
916:
1.31 noro 917: ND nd_add(int mod,ND p1,ND p2)
1.1 noro 918: {
919: int n,c;
1.34 noro 920: int t,can,td1,td2;
1.1 noro 921: ND r;
922: NM m1,m2,mr0,mr,s;
923:
1.34 noro 924: if ( !p1 ) return p2;
925: else if ( !p2 ) return p1;
1.71 noro 926: else if ( mod == -1 ) return nd_add_sf(p1,p2);
1.113 noro 927: else if ( !mod ) return nd_add_q(p1,p2);
1.1 noro 928: else {
1.30 noro 929: can = 0;
1.1 noro 930: for ( n = NV(p1), m1 = BDY(p1), m2 = BDY(p2), mr0 = 0; m1 && m2; ) {
1.34 noro 931: c = DL_COMPARE(DL(m1),DL(m2));
1.1 noro 932: switch ( c ) {
933: case 0:
1.19 noro 934: t = ((CM(m1))+(CM(m2))) - mod;
1.34 noro 935: if ( t < 0 ) t += mod;
1.1 noro 936: s = m1; m1 = NEXT(m1);
937: if ( t ) {
1.34 noro 938: can++; NEXTNM2(mr0,mr,s); CM(mr) = (t);
1.17 noro 939: } else {
1.34 noro 940: can += 2; FREENM(s);
1.17 noro 941: }
942: s = m2; m2 = NEXT(m2); FREENM(s);
1.95 noro 943: break;
944: case 1:
945: s = m1; m1 = NEXT(m1); NEXTNM2(mr0,mr,s);
946: break;
947: case -1:
948: s = m2; m2 = NEXT(m2); NEXTNM2(mr0,mr,s);
949: break;
950: }
951: }
952: if ( !mr0 )
953: if ( m1 ) mr0 = m1;
954: else if ( m2 ) mr0 = m2;
955: else return 0;
956: else if ( m1 ) NEXT(mr) = m1;
957: else if ( m2 ) NEXT(mr) = m2;
958: else NEXT(mr) = 0;
959: BDY(p1) = mr0;
960: SG(p1) = MAX(SG(p1),SG(p2));
961: LEN(p1) = LEN(p1)+LEN(p2)-can;
962: FREEND(p2);
963: return p1;
964: }
965: }
966:
967: /* XXX on opteron, the inlined manipulation of destructive additon of
968: * two NM seems to make gcc optimizer get confused, so the part is
969: * done in a function.
970: */
971:
1.113 noro 972: int nm_destructive_add_q(NM *m1,NM *m2,NM *mr0,NM *mr)
1.95 noro 973: {
974: NM s;
1.113 noro 975: Q t;
1.95 noro 976: int can;
977:
1.113 noro 978: addq(CQ(*m1),CQ(*m2),&t);
1.95 noro 979: s = *m1; *m1 = NEXT(*m1);
980: if ( t ) {
1.113 noro 981: can = 1; NEXTNM2(*mr0,*mr,s); CQ(*mr) = (t);
1.95 noro 982: } else {
983: can = 2; FREENM(s);
984: }
985: s = *m2; *m2 = NEXT(*m2); FREENM(s);
986: return can;
987: }
988:
1.113 noro 989: ND nd_add_q(ND p1,ND p2)
1.95 noro 990: {
991: int n,c,can;
992: ND r;
993: NM m1,m2,mr0,mr,s;
1.113 noro 994: Q t;
1.95 noro 995:
996: if ( !p1 ) return p2;
997: else if ( !p2 ) return p1;
998: else {
999: can = 0;
1000: for ( n = NV(p1), m1 = BDY(p1), m2 = BDY(p2), mr0 = 0; m1 && m2; ) {
1001: c = DL_COMPARE(DL(m1),DL(m2));
1002: switch ( c ) {
1003: case 0:
1004: #if defined(__x86_64__)
1.113 noro 1005: can += nm_destructive_add_q(&m1,&m2,&mr0,&mr);
1.95 noro 1006: #else
1.113 noro 1007: addq(CQ(m1),CQ(m2),&t);
1.95 noro 1008: s = m1; m1 = NEXT(m1);
1009: if ( t ) {
1.113 noro 1010: can++; NEXTNM2(mr0,mr,s); CQ(mr) = (t);
1.95 noro 1011: } else {
1012: can += 2; FREENM(s);
1013: }
1014: s = m2; m2 = NEXT(m2); FREENM(s);
1015: #endif
1.17 noro 1016: break;
1017: case 1:
1018: s = m1; m1 = NEXT(m1); NEXTNM2(mr0,mr,s);
1019: break;
1020: case -1:
1021: s = m2; m2 = NEXT(m2); NEXTNM2(mr0,mr,s);
1022: break;
1023: }
1024: }
1025: if ( !mr0 )
1.34 noro 1026: if ( m1 ) mr0 = m1;
1027: else if ( m2 ) mr0 = m2;
1028: else return 0;
1029: else if ( m1 ) NEXT(mr) = m1;
1030: else if ( m2 ) NEXT(mr) = m2;
1031: else NEXT(mr) = 0;
1.17 noro 1032: BDY(p1) = mr0;
1033: SG(p1) = MAX(SG(p1),SG(p2));
1.31 noro 1034: LEN(p1) = LEN(p1)+LEN(p2)-can;
1.17 noro 1035: FREEND(p2);
1036: return p1;
1037: }
1038: }
1039:
1.71 noro 1040: ND nd_add_sf(ND p1,ND p2)
1041: {
1042: int n,c,can;
1043: ND r;
1044: NM m1,m2,mr0,mr,s;
1045: int t;
1046:
1047: if ( !p1 ) return p2;
1048: else if ( !p2 ) return p1;
1049: else {
1050: can = 0;
1051: for ( n = NV(p1), m1 = BDY(p1), m2 = BDY(p2), mr0 = 0; m1 && m2; ) {
1052: c = DL_COMPARE(DL(m1),DL(m2));
1053: switch ( c ) {
1054: case 0:
1055: t = _addsf(CM(m1),CM(m2));
1056: s = m1; m1 = NEXT(m1);
1057: if ( t ) {
1058: can++; NEXTNM2(mr0,mr,s); CM(mr) = (t);
1059: } else {
1060: can += 2; FREENM(s);
1061: }
1062: s = m2; m2 = NEXT(m2); FREENM(s);
1063: break;
1064: case 1:
1065: s = m1; m1 = NEXT(m1); NEXTNM2(mr0,mr,s);
1066: break;
1067: case -1:
1068: s = m2; m2 = NEXT(m2); NEXTNM2(mr0,mr,s);
1069: break;
1070: }
1071: }
1072: if ( !mr0 )
1073: if ( m1 ) mr0 = m1;
1074: else if ( m2 ) mr0 = m2;
1075: else return 0;
1076: else if ( m1 ) NEXT(mr) = m1;
1077: else if ( m2 ) NEXT(mr) = m2;
1078: else NEXT(mr) = 0;
1079: BDY(p1) = mr0;
1080: SG(p1) = MAX(SG(p1),SG(p2));
1081: LEN(p1) = LEN(p1)+LEN(p2)-can;
1082: FREEND(p2);
1083: return p1;
1084: }
1085: }
1086:
1.1 noro 1087: /* ret=1 : success, ret=0 : overflow */
1.113 noro 1088: int nd_nf(int mod,ND g,NDV *ps,int full,NDC dn,ND *rp)
1.1 noro 1089: {
1.11 noro 1090: ND d;
1.1 noro 1091: NM m,mrd,tail;
1.7 noro 1092: NM mul;
1.10 noro 1093: int n,sugar,psugar,sugar0,stat,index;
1.30 noro 1094: int c,c1,c2,dummy;
1.17 noro 1095: RHist h;
1.11 noro 1096: NDV p,red;
1.113 noro 1097: Q cg,cred,gcd,tq,qq;
1.21 noro 1098: double hmag;
1.1 noro 1099:
1.113 noro 1100: if ( dn ) {
1101: if ( mod )
1102: dn->m = 1;
1103: else
1104: dn->z = ONE;
1105: }
1.1 noro 1106: if ( !g ) {
1107: *rp = 0;
1108: return 1;
1109: }
1.113 noro 1110: if ( !mod ) hmag = ((double)p_mag((P)HCQ(g)))*nd_scale;
1.21 noro 1111:
1.14 noro 1112: sugar0 = sugar = SG(g);
1.1 noro 1113: n = NV(g);
1.61 noro 1114: mul = (NM)ALLOCA(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT));
1.1 noro 1115: for ( d = 0; g; ) {
1.63 noro 1116: index = ndl_find_reducer(HDL(g));
1.6 noro 1117: if ( index >= 0 ) {
1.17 noro 1118: h = nd_psh[index];
1119: ndl_sub(HDL(g),DL(h),DL(mul));
1.14 noro 1120: if ( ndl_check_bound2(index,DL(mul)) ) {
1.6 noro 1121: nd_free(g); nd_free(d);
1122: return 0;
1123: }
1.77 noro 1124: if ( nd_demand )
1125: p = ndv_load(index);
1126: else
1127: p = ps[index];
1.71 noro 1128: if ( mod == -1 )
1129: CM(mul) = _mulsf(_invsf(HCM(p)),_chsgnsf(HCM(g)));
1130: else if ( mod ) {
1.19 noro 1131: c1 = invm(HCM(p),mod); c2 = mod-HCM(g);
1132: DMAR(c1,c2,0,mod,c); CM(mul) = c;
1.16 noro 1133: } else {
1.113 noro 1134: igcd_cofactor(HCQ(g),HCQ(p),&gcd,&cg,&cred);
1135: chsgnq(cg,&CQ(mul));
1136: nd_mul_c_q(d,cred); nd_mul_c_q(g,cred);
1.69 noro 1137: if ( dn ) {
1.113 noro 1138: mulq(dn->z,cred,&tq); dn->z = tq;
1.69 noro 1139: }
1.16 noro 1140: }
1.55 noro 1141: g = nd_add(mod,g,ndv_mul_nm(mod,mul,p));
1.34 noro 1142: sugar = MAX(sugar,SG(p)+TD(DL(mul)));
1.113 noro 1143: if ( !mod && hmag && g && ((double)(p_mag((P)HCQ(g))) > hmag) ) {
1144: tq = HCQ(g);
1.21 noro 1145: nd_removecont2(d,g);
1.69 noro 1146: if ( dn ) {
1.113 noro 1147: divq(tq,HCQ(g),&qq); divq(dn->z,qq,&tq); dn->z = tq;
1.69 noro 1148: }
1.113 noro 1149: hmag = ((double)p_mag((P)HCQ(g)))*nd_scale;
1.21 noro 1150: }
1.1 noro 1151: } else if ( !full ) {
1152: *rp = g;
1153: return 1;
1154: } else {
1155: m = BDY(g);
1156: if ( NEXT(m) ) {
1.34 noro 1157: BDY(g) = NEXT(m); NEXT(m) = 0; LEN(g)--;
1.1 noro 1158: } else {
1159: FREEND(g); g = 0;
1160: }
1161: if ( d ) {
1.34 noro 1162: NEXT(tail)=m; tail=m; LEN(d)++;
1.1 noro 1163: } else {
1.34 noro 1164: MKND(n,m,1,d); tail = BDY(d);
1.1 noro 1165: }
1166: }
1167: }
1.34 noro 1168: if ( d ) SG(d) = sugar;
1.1 noro 1169: *rp = d;
1170: return 1;
1171: }
1.28 noro 1172:
1.53 noro 1173: int nd_nf_pbucket(int mod,ND g,NDV *ps,int full,ND *rp)
1.25 noro 1174: {
1175: int hindex,index;
1176: NDV p;
1177: ND u,d,red;
1178: NODE l;
1.31 noro 1179: NM mul,m,mrd,tail;
1.25 noro 1180: int sugar,psugar,n,h_reducible;
1181: PGeoBucket bucket;
1182: int c,c1,c2;
1.113 noro 1183: Q cg,cred,gcd,zzz;
1.25 noro 1184: RHist h;
1.28 noro 1185: double hmag,gmag;
1.77 noro 1186: int count = 0;
1187: int hcount = 0;
1.25 noro 1188:
1189: if ( !g ) {
1190: *rp = 0;
1191: return 1;
1192: }
1193: sugar = SG(g);
1194: n = NV(g);
1.113 noro 1195: if ( !mod ) hmag = ((double)p_mag((P)HCQ(g)))*nd_scale;
1.25 noro 1196: bucket = create_pbucket();
1.31 noro 1197: add_pbucket(mod,bucket,g);
1.25 noro 1198: d = 0;
1.61 noro 1199: mul = (NM)ALLOCA(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT));
1.25 noro 1200: while ( 1 ) {
1.113 noro 1201: hindex = mod?head_pbucket(mod,bucket):head_pbucket_q(bucket);
1.25 noro 1202: if ( hindex < 0 ) {
1.77 noro 1203: if ( DP_Print > 3 ) printf("(%d %d)",count,hcount);
1.34 noro 1204: if ( d ) SG(d) = sugar;
1.25 noro 1205: *rp = d;
1206: return 1;
1207: }
1208: g = bucket->body[hindex];
1.63 noro 1209: index = ndl_find_reducer(HDL(g));
1.25 noro 1210: if ( index >= 0 ) {
1.77 noro 1211: count++;
1212: if ( !d ) hcount++;
1.25 noro 1213: h = nd_psh[index];
1214: ndl_sub(HDL(g),DL(h),DL(mul));
1215: if ( ndl_check_bound2(index,DL(mul)) ) {
1.26 noro 1216: nd_free(d);
1.25 noro 1217: free_pbucket(bucket);
1218: *rp = 0;
1219: return 0;
1220: }
1.53 noro 1221: p = ps[index];
1.71 noro 1222: if ( mod == -1 )
1223: CM(mul) = _mulsf(_invsf(HCM(p)),_chsgnsf(HCM(g)));
1224: else if ( mod ) {
1.25 noro 1225: c1 = invm(HCM(p),mod); c2 = mod-HCM(g);
1226: DMAR(c1,c2,0,mod,c); CM(mul) = c;
1227: } else {
1.113 noro 1228: igcd_cofactor(HCQ(g),HCQ(p),&gcd,&cg,&cred);
1229: chsgnq(cg,&CQ(mul));
1230: nd_mul_c_q(d,cred);
1231: mulq_pbucket(bucket,cred);
1.26 noro 1232: g = bucket->body[hindex];
1.113 noro 1233: gmag = (double)p_mag((P)HCQ(g));
1.25 noro 1234: }
1.55 noro 1235: red = ndv_mul_nm(mod,mul,p);
1.25 noro 1236: bucket->body[hindex] = nd_remove_head(g);
1237: red = nd_remove_head(red);
1.31 noro 1238: add_pbucket(mod,bucket,red);
1.34 noro 1239: psugar = SG(p)+TD(DL(mul));
1240: sugar = MAX(sugar,psugar);
1.28 noro 1241: if ( !mod && hmag && (gmag > hmag) ) {
1242: g = normalize_pbucket(mod,bucket);
1243: if ( !g ) {
1.34 noro 1244: if ( d ) SG(d) = sugar;
1.28 noro 1245: *rp = d;
1246: return 1;
1247: }
1248: nd_removecont2(d,g);
1.113 noro 1249: hmag = ((double)p_mag((P)HCQ(g)))*nd_scale;
1.31 noro 1250: add_pbucket(mod,bucket,g);
1.28 noro 1251: }
1.25 noro 1252: } else if ( !full ) {
1253: g = normalize_pbucket(mod,bucket);
1.34 noro 1254: if ( g ) SG(g) = sugar;
1.25 noro 1255: *rp = g;
1256: return 1;
1257: } else {
1258: m = BDY(g);
1259: if ( NEXT(m) ) {
1.34 noro 1260: BDY(g) = NEXT(m); NEXT(m) = 0; LEN(g)--;
1.25 noro 1261: } else {
1262: FREEND(g); g = 0;
1263: }
1264: bucket->body[hindex] = g;
1265: NEXT(m) = 0;
1266: if ( d ) {
1.34 noro 1267: NEXT(tail)=m; tail=m; LEN(d)++;
1.25 noro 1268: } else {
1.34 noro 1269: MKND(n,m,1,d); tail = BDY(d);
1.25 noro 1270: }
1271: }
1272: }
1273: }
1.27 noro 1274:
1.61 noro 1275: /* input : list of NDV, cand : list of NDV */
1.28 noro 1276:
1.61 noro 1277: int ndv_check_candidate(NODE input,int obpe,int oadv,EPOS oepos,NODE cand)
1.28 noro 1278: {
1279: int n,i,stat;
1280: ND nf,d;
1.61 noro 1281: NDV r;
1.45 noro 1282: NODE t,s;
1.113 noro 1283: union oNDC dn;
1.45 noro 1284:
1.61 noro 1285: ndv_setup(0,0,cand);
1.31 noro 1286: n = length(cand);
1.28 noro 1287:
1288: /* membercheck : list is a subset of Id(cand) ? */
1289: for ( t = input; t; t = NEXT(t) ) {
1.45 noro 1290: again:
1.61 noro 1291: if ( nd_bpe > obpe )
1292: r = ndv_dup_realloc((NDV)BDY(t),obpe,oadv,oepos);
1293: else
1294: r = (NDV)BDY(t);
1295: d = ndvtond(0,r);
1.69 noro 1296: stat = nd_nf(0,d,nd_ps,0,0,&nf);
1.45 noro 1297: if ( !stat ) {
1.103 noro 1298: nd_reconstruct(0,0);
1.45 noro 1299: goto again;
1300: } else if ( nf ) return 0;
1.71 noro 1301: if ( DP_Print ) { printf("."); fflush(stdout); }
1.28 noro 1302: }
1.71 noro 1303: if ( DP_Print ) { printf("\n"); }
1.28 noro 1304: /* gbcheck : cand is a GB of Id(cand) ? */
1.77 noro 1305: if ( !nd_gb(0,0,1) ) return 0;
1.28 noro 1306: /* XXX */
1.23 noro 1307: return 1;
1308: }
1.1 noro 1309:
1310: ND nd_remove_head(ND p)
1311: {
1312: NM m;
1313:
1314: m = BDY(p);
1315: if ( !NEXT(m) ) {
1.34 noro 1316: FREEND(p); p = 0;
1.31 noro 1317: } else {
1.34 noro 1318: BDY(p) = NEXT(m); LEN(p)--;
1.31 noro 1319: }
1.1 noro 1320: FREENM(m);
1321: return p;
1322: }
1323:
1.69 noro 1324: ND nd_separate_head(ND p,ND *head)
1325: {
1326: NM m,m0;
1327: ND r;
1328:
1329: m = BDY(p);
1330: if ( !NEXT(m) ) {
1331: *head = p; p = 0;
1332: } else {
1333: m0 = m;
1334: BDY(p) = NEXT(m); LEN(p)--;
1335: NEXT(m0) = 0;
1336: MKND(NV(p),m0,1,r);
1337: *head = r;
1338: }
1339: return p;
1340: }
1341:
1.1 noro 1342: PGeoBucket create_pbucket()
1343: {
1344: PGeoBucket g;
1345:
1346: g = CALLOC(1,sizeof(struct oPGeoBucket));
1347: g->m = -1;
1348: return g;
1349: }
1350:
1.25 noro 1351: void free_pbucket(PGeoBucket b) {
1352: int i;
1353:
1.26 noro 1354: for ( i = 0; i <= b->m; i++ )
1.25 noro 1355: if ( b->body[i] ) {
1356: nd_free(b->body[i]);
1357: b->body[i] = 0;
1358: }
1359: GC_free(b);
1360: }
1361:
1.63 noro 1362: void add_pbucket_symbolic(PGeoBucket g,ND d)
1363: {
1364: int l,i,k,m;
1365:
1366: if ( !d )
1367: return;
1368: l = LEN(d);
1369: for ( k = 0, m = 1; l > m; k++, m <<= 1 );
1370: /* 2^(k-1) < l <= 2^k (=m) */
1371: d = nd_merge(g->body[k],d);
1372: for ( ; d && LEN(d) > m; k++, m <<= 1 ) {
1373: g->body[k] = 0;
1374: d = nd_merge(g->body[k+1],d);
1375: }
1376: g->body[k] = d;
1377: g->m = MAX(g->m,k);
1378: }
1379:
1.31 noro 1380: void add_pbucket(int mod,PGeoBucket g,ND d)
1.1 noro 1381: {
1.31 noro 1382: int l,i,k,m;
1.1 noro 1383:
1.31 noro 1384: if ( !d )
1385: return;
1386: l = LEN(d);
1.29 noro 1387: for ( k = 0, m = 1; l > m; k++, m <<= 1 );
1388: /* 2^(k-1) < l <= 2^k (=m) */
1.31 noro 1389: d = nd_add(mod,g->body[k],d);
1390: for ( ; d && LEN(d) > m; k++, m <<= 1 ) {
1.1 noro 1391: g->body[k] = 0;
1.31 noro 1392: d = nd_add(mod,g->body[k+1],d);
1.1 noro 1393: }
1394: g->body[k] = d;
1395: g->m = MAX(g->m,k);
1396: }
1397:
1.113 noro 1398: void mulq_pbucket(PGeoBucket g,Q c)
1.26 noro 1399: {
1400: int k;
1401:
1402: for ( k = 0; k <= g->m; k++ )
1.113 noro 1403: nd_mul_c_q(g->body[k],c);
1.26 noro 1404: }
1405:
1.63 noro 1406: NM remove_head_pbucket_symbolic(PGeoBucket g)
1407: {
1408: int j,i,k,c;
1409: NM head;
1410:
1411: k = g->m;
1412: j = -1;
1413: for ( i = 0; i <= k; i++ ) {
1414: if ( !g->body[i] ) continue;
1415: if ( j < 0 ) j = i;
1416: else {
1417: c = DL_COMPARE(HDL(g->body[i]),HDL(g->body[j]));
1418: if ( c > 0 )
1419: j = i;
1420: else if ( c == 0 )
1421: g->body[i] = nd_remove_head(g->body[i]);
1422: }
1423: }
1424: if ( j < 0 ) return 0;
1425: else {
1426: head = BDY(g->body[j]);
1427: if ( !NEXT(head) ) {
1428: FREEND(g->body[j]);
1429: g->body[j] = 0;
1430: } else {
1431: BDY(g->body[j]) = NEXT(head);
1432: LEN(g->body[j])--;
1433: }
1434: return head;
1435: }
1436: }
1437:
1.19 noro 1438: int head_pbucket(int mod,PGeoBucket g)
1.1 noro 1439: {
1440: int j,i,c,k,nv,sum;
1.61 noro 1441: UINT *di,*dj;
1.1 noro 1442: ND gi,gj;
1443:
1444: k = g->m;
1445: while ( 1 ) {
1446: j = -1;
1447: for ( i = 0; i <= k; i++ ) {
1448: if ( !(gi = g->body[i]) )
1449: continue;
1450: if ( j < 0 ) {
1451: j = i;
1452: gj = g->body[j];
1453: dj = HDL(gj);
1.14 noro 1454: sum = HCM(gj);
1.1 noro 1455: } else {
1.34 noro 1456: c = DL_COMPARE(HDL(gi),dj);
1.1 noro 1457: if ( c > 0 ) {
1.34 noro 1458: if ( sum ) HCM(gj) = sum;
1459: else g->body[j] = nd_remove_head(gj);
1.1 noro 1460: j = i;
1461: gj = g->body[j];
1462: dj = HDL(gj);
1.14 noro 1463: sum = HCM(gj);
1.1 noro 1464: } else if ( c == 0 ) {
1.71 noro 1465: if ( mod == -1 )
1466: sum = _addsf(sum,HCM(gi));
1467: else {
1468: sum = sum+HCM(gi)-mod;
1469: if ( sum < 0 ) sum += mod;
1470: }
1.1 noro 1471: g->body[i] = nd_remove_head(gi);
1472: }
1473: }
1474: }
1.34 noro 1475: if ( j < 0 ) return -1;
1.1 noro 1476: else if ( sum ) {
1.14 noro 1477: HCM(gj) = sum;
1.26 noro 1478: return j;
1.31 noro 1479: } else
1.26 noro 1480: g->body[j] = nd_remove_head(gj);
1481: }
1482: }
1483:
1.113 noro 1484: int head_pbucket_q(PGeoBucket g)
1.26 noro 1485: {
1486: int j,i,c,k,nv;
1.113 noro 1487: Q sum,t;
1.26 noro 1488: ND gi,gj;
1489:
1490: k = g->m;
1491: while ( 1 ) {
1492: j = -1;
1493: for ( i = 0; i <= k; i++ ) {
1.34 noro 1494: if ( !(gi = g->body[i]) ) continue;
1.26 noro 1495: if ( j < 0 ) {
1496: j = i;
1497: gj = g->body[j];
1.113 noro 1498: sum = HCQ(gj);
1.26 noro 1499: } else {
1500: nv = NV(gi);
1.34 noro 1501: c = DL_COMPARE(HDL(gi),HDL(gj));
1.26 noro 1502: if ( c > 0 ) {
1.113 noro 1503: if ( sum ) HCQ(gj) = sum;
1.34 noro 1504: else g->body[j] = nd_remove_head(gj);
1.26 noro 1505: j = i;
1506: gj = g->body[j];
1.113 noro 1507: sum = HCQ(gj);
1.26 noro 1508: } else if ( c == 0 ) {
1.113 noro 1509: addq(sum,HCQ(gi),&t);
1510: sum = t;
1.26 noro 1511: g->body[i] = nd_remove_head(gi);
1512: }
1513: }
1514: }
1.34 noro 1515: if ( j < 0 ) return -1;
1.26 noro 1516: else if ( sum ) {
1.113 noro 1517: HCQ(gj) = sum;
1.1 noro 1518: return j;
1.31 noro 1519: } else
1.1 noro 1520: g->body[j] = nd_remove_head(gj);
1521: }
1522: }
1523:
1.25 noro 1524: ND normalize_pbucket(int mod,PGeoBucket g)
1.1 noro 1525: {
1.31 noro 1526: int i;
1.1 noro 1527: ND r,t;
1528:
1529: r = 0;
1.28 noro 1530: for ( i = 0; i <= g->m; i++ ) {
1.31 noro 1531: r = nd_add(mod,r,g->body[i]);
1.28 noro 1532: g->body[i] = 0;
1533: }
1534: g->m = -1;
1.1 noro 1535: return r;
1536: }
1537:
1.92 noro 1538: void do_diagonalize(int sugar,int m)
1539: {
1540: int i,nh,stat;
1541: NODE r,g,t;
1542: ND h,nf,s,head;
1543: NDV nfv;
1.113 noro 1544: Q q,num,den;
1545: union oNDC dn;
1.92 noro 1546:
1547: for ( i = nd_psn-1; i >= 0 && SG(nd_psh[i]) == sugar; i-- ) {
1548: if ( nd_demand )
1549: nfv = ndv_load(i);
1550: else
1551: nfv = nd_ps[i];
1552: s = ndvtond(m,nfv);
1553: s = nd_separate_head(s,&head);
1554: nd_nf(m,s,nd_ps,1,&dn,&nf);
1555: if ( !m ) {
1.113 noro 1556: NTOQ(NM(dn.z),SGN(dn.z),num);
1557: mulq(HCQ(head),num,&q); HCQ(head) = q;
1558: if ( DN(dn.z) ) {
1559: NTOQ(DN(dn.z),1,den);
1560: nd_mul_c_q(nf,den);
1561: }
1.92 noro 1562: }
1563: nf = nd_add(m,head,nf);
1564: ndv_free(nfv);
1565: nd_removecont(m,nf);
1566: nfv = ndtondv(m,nf);
1567: nd_free(nf);
1568: nd_bound[i] = ndv_compute_bound(nfv);
1569: if ( nd_demand ) {
1570: ndv_save(nfv,i);
1571: ndv_free(nfv);
1572: } else
1573: nd_ps[i] = nfv;
1574: }
1575: }
1576:
1.27 noro 1577: /* return value = 0 => input is not a GB */
1578:
1.77 noro 1579: NODE nd_gb(int m,int ishomo,int checkonly)
1.1 noro 1580: {
1581: int i,nh,sugar,stat;
1.23 noro 1582: NODE r,g,t;
1.1 noro 1583: ND_pairs d;
1584: ND_pairs l;
1.117 noro 1585: ND h,nf,s,head,nf1;
1.63 noro 1586: NDV nfv;
1.113 noro 1587: Q q,num,den;
1588: union oNDC dn;
1.1 noro 1589:
1.23 noro 1590: g = 0; d = 0;
1591: for ( i = 0; i < nd_psn; i++ ) {
1.1 noro 1592: d = update_pairs(d,g,i);
1593: g = update_base(g,i);
1594: }
1595: sugar = 0;
1596: while ( d ) {
1597: again:
1598: l = nd_minp(d,&d);
1.14 noro 1599: if ( SG(l) != sugar ) {
1.92 noro 1600: if ( ishomo ) do_diagonalize(sugar,m);
1601:
1.14 noro 1602: sugar = SG(l);
1.71 noro 1603: if ( DP_Print ) fprintf(asir_out,"%d",sugar);
1.1 noro 1604: }
1.53 noro 1605: stat = nd_sp(m,0,l,&h);
1.1 noro 1606: if ( !stat ) {
1607: NEXT(l) = d; d = l;
1.103 noro 1608: d = nd_reconstruct(0,d);
1.1 noro 1609: goto again;
1610: }
1.41 noro 1611: #if USE_GEOBUCKET
1.69 noro 1612: stat = m?nd_nf_pbucket(m,h,nd_ps,!Top,&nf):nd_nf(m,h,nd_ps,!Top,0,&nf);
1.41 noro 1613: #else
1.69 noro 1614: stat = nd_nf(m,h,nd_ps,!Top,0,&nf);
1.41 noro 1615: #endif
1.1 noro 1616: if ( !stat ) {
1617: NEXT(l) = d; d = l;
1.103 noro 1618: d = nd_reconstruct(0,d);
1.1 noro 1619: goto again;
1620: } else if ( nf ) {
1.27 noro 1621: if ( checkonly ) return 0;
1.71 noro 1622: if ( DP_Print ) { printf("+"); fflush(stdout); }
1.63 noro 1623: nd_removecont(m,nf);
1.118 ! noro 1624: if ( !m && nd_nalg ) {
! 1625: nd_monic(0,&nf);
! 1626: nd_removecont(m,nf);
1.117 noro 1627: }
1.63 noro 1628: nfv = ndtondv(m,nf); nd_free(nf);
1.77 noro 1629: nh = ndv_newps(m,nfv,0);
1.1 noro 1630: d = update_pairs(d,g,nh);
1631: g = update_base(g,nh);
1632: FREENDP(l);
1633: } else {
1.71 noro 1634: if ( DP_Print ) { printf("."); fflush(stdout); }
1.1 noro 1635: FREENDP(l);
1636: }
1637: }
1.77 noro 1638: if ( nd_demand )
1639: for ( t = g; t; t = NEXT(t) )
1640: BDY(t) = (pointer)ndv_load((int)BDY(t));
1641: else
1642: for ( t = g; t; t = NEXT(t) )
1643: BDY(t) = (pointer)nd_ps[(int)BDY(t)];
1.1 noro 1644: return g;
1645: }
1646:
1.92 noro 1647: void do_diagonalize_trace(int sugar,int m)
1.91 noro 1648: {
1649: int i,nh,stat;
1650: NODE r,g,t;
1651: ND h,nf,nfq,s,head;
1652: NDV nfv,nfqv;
1.113 noro 1653: Q q,den,num;
1654: union oNDC dn;
1.91 noro 1655:
1656: for ( i = nd_psn-1; i >= 0 && SG(nd_psh[i]) == sugar; i-- ) {
1657: /* for nd_ps */
1658: s = ndvtond(m,nd_ps[i]);
1659: s = nd_separate_head(s,&head);
1660: nd_nf_pbucket(m,s,nd_ps,1,&nf);
1661: nf = nd_add(m,head,nf);
1662: ndv_free(nd_ps[i]);
1663: nd_ps[i] = ndtondv(m,nf);
1664: nd_free(nf);
1665:
1666: /* for nd_ps_trace */
1667: if ( nd_demand )
1668: nfv = ndv_load(i);
1669: else
1670: nfv = nd_ps_trace[i];
1671: s = ndvtond(0,nfv);
1672: s = nd_separate_head(s,&head);
1673: nd_nf(0,s,nd_ps_trace,1,&dn,&nf);
1.113 noro 1674: NTOQ(NM(dn.z),SGN(dn.z),num);
1675: mulq(HCQ(head),num,&q); HCQ(head) = q;
1676: if ( DN(dn.z) ) {
1677: NTOQ(DN(dn.z),1,den);
1678: nd_mul_c_q(nf,den);
1679: }
1.91 noro 1680: nf = nd_add(0,head,nf);
1681: ndv_free(nfv);
1682: nd_removecont(0,nf);
1683: nfv = ndtondv(0,nf);
1684: nd_free(nf);
1685: nd_bound[i] = ndv_compute_bound(nfv);
1686: if ( nd_demand ) {
1687: ndv_save(nfv,i);
1688: ndv_free(nfv);
1689: } else
1690: nd_ps_trace[i] = nfv;
1691: }
1692: }
1693:
1.118 ! noro 1694: static struct oEGT eg_invdalg;
! 1695: struct oEGT eg_le;
! 1696:
1.82 noro 1697: NODE nd_gb_trace(int m,int ishomo)
1.20 noro 1698: {
1699: int i,nh,sugar,stat;
1.23 noro 1700: NODE r,g,t;
1.20 noro 1701: ND_pairs d;
1702: ND_pairs l;
1.82 noro 1703: ND h,nf,nfq,s,head;
1.63 noro 1704: NDV nfv,nfqv;
1.113 noro 1705: Q q,den,num;
1706: union oNDC dn;
1.118 ! noro 1707: struct oEGT eg_monic,egm0,egm1;
1.20 noro 1708:
1.118 ! noro 1709: init_eg(&eg_monic);
! 1710: init_eg(&eg_invdalg);
! 1711: init_eg(&eg_le);
1.23 noro 1712: g = 0; d = 0;
1713: for ( i = 0; i < nd_psn; i++ ) {
1.20 noro 1714: d = update_pairs(d,g,i);
1715: g = update_base(g,i);
1716: }
1717: sugar = 0;
1718: while ( d ) {
1719: again:
1720: l = nd_minp(d,&d);
1721: if ( SG(l) != sugar ) {
1.93 noro 1722: if ( ishomo ) do_diagonalize_trace(sugar,m);
1.20 noro 1723: sugar = SG(l);
1.71 noro 1724: if ( DP_Print ) fprintf(asir_out,"%d",sugar);
1.20 noro 1725: }
1.53 noro 1726: stat = nd_sp(m,0,l,&h);
1.20 noro 1727: if ( !stat ) {
1728: NEXT(l) = d; d = l;
1.103 noro 1729: d = nd_reconstruct(1,d);
1.20 noro 1730: goto again;
1731: }
1.41 noro 1732: #if USE_GEOBUCKET
1.53 noro 1733: stat = nd_nf_pbucket(m,h,nd_ps,!Top,&nf);
1.41 noro 1734: #else
1.69 noro 1735: stat = nd_nf(m,h,nd_ps,!Top,0,&nf);
1.41 noro 1736: #endif
1.20 noro 1737: if ( !stat ) {
1738: NEXT(l) = d; d = l;
1.103 noro 1739: d = nd_reconstruct(1,d);
1.20 noro 1740: goto again;
1741: } else if ( nf ) {
1.77 noro 1742: if ( nd_demand ) {
1743: nfqv = ndv_load(nd_psn);
1744: nfq = ndvtond(0,nfqv);
1745: } else
1746: nfq = 0;
1747: if ( !nfq ) {
1748: if ( !nd_sp(0,1,l,&h) || !nd_nf(0,h,nd_ps_trace,!Top,0,&nfq) ) {
1749: NEXT(l) = d; d = l;
1.103 noro 1750: d = nd_reconstruct(1,d);
1.77 noro 1751: goto again;
1752: }
1.70 noro 1753: }
1.20 noro 1754: if ( nfq ) {
1.70 noro 1755: /* m|HC(nfq) => failure */
1.113 noro 1756: if ( !rem(NM(HCQ(nfq)),m) ) return 0;
1.63 noro 1757:
1.71 noro 1758: if ( DP_Print ) { printf("+"); fflush(stdout); }
1.118 ! noro 1759: if ( nd_nalg ) {
! 1760: /* m|DN(HC(nf)^(-1)) => failure */
! 1761: get_eg(&egm0);
! 1762: if ( !nd_monic(m,&nfq) ) return 0;
! 1763: get_eg(&egm1); add_eg(&eg_monic,&egm0,&egm1);
! 1764: nd_removecont(0,nfq); nfqv = ndtondv(0,nfq); nd_free(nfq);
! 1765: nfv = ndv_dup(0,nfqv); ndv_mod(m,nfv); nd_free(nf);
! 1766: } else {
! 1767: nd_removecont(0,nfq); nfqv = ndtondv(0,nfq); nd_free(nfq);
! 1768: nd_removecont(m,nf); nfv = ndtondv(m,nf); nd_free(nf);
! 1769: }
1.77 noro 1770: nh = ndv_newps(0,nfv,nfqv);
1.20 noro 1771: d = update_pairs(d,g,nh);
1772: g = update_base(g,nh);
1773: } else {
1.71 noro 1774: if ( DP_Print ) { printf("*"); fflush(stdout); }
1.20 noro 1775: }
1776: } else {
1.71 noro 1777: if ( DP_Print ) { printf("."); fflush(stdout); }
1.20 noro 1778: }
1779: FREENDP(l);
1780: }
1.77 noro 1781: if ( nd_demand )
1782: for ( t = g; t; t = NEXT(t) )
1783: BDY(t) = (pointer)ndv_load((int)BDY(t));
1784: else
1785: for ( t = g; t; t = NEXT(t) )
1786: BDY(t) = (pointer)nd_ps_trace[(int)BDY(t)];
1.118 ! noro 1787: if ( nd_nalg ) {
! 1788: print_eg("monic",&eg_monic);
! 1789: print_eg("invdalg",&eg_invdalg);
! 1790: print_eg("le",&eg_le);
! 1791: }
1.20 noro 1792: return g;
1793: }
1794:
1.23 noro 1795: int ndv_compare(NDV *p1,NDV *p2)
1796: {
1.34 noro 1797: return DL_COMPARE(HDL(*p1),HDL(*p2));
1.23 noro 1798: }
1799:
1800: int ndv_compare_rev(NDV *p1,NDV *p2)
1801: {
1.34 noro 1802: return -DL_COMPARE(HDL(*p1),HDL(*p2));
1.23 noro 1803: }
1804:
1.61 noro 1805: NODE ndv_reduceall(int m,NODE f)
1.23 noro 1806: {
1.69 noro 1807: int i,n,stat;
1808: ND nf,g,head;
1.23 noro 1809: NODE t,a0,a;
1.113 noro 1810: union oNDC dn;
1.69 noro 1811: NDV *w;
1.113 noro 1812: Q q,num,den;
1.23 noro 1813:
1.69 noro 1814: n = length(f);
1815: #if 0
1816: w = (NDV *)ALLOCA(n*sizeof(NDV));
1817: for ( i = 0, t = f; i < n; i++, t = NEXT(t) ) w[i] = (NDV)BDY(t);
1818: qsort(w,n,sizeof(NDV),
1819: (int (*)(const void *,const void *))ndv_compare);
1820: for ( t = f, i = 0; t; i++, t = NEXT(t) ) BDY(t) = (pointer)w[i];
1821: #endif
1822: ndv_setup(m,0,f);
1823: for ( i = 0; i < n; ) {
1824: g = ndvtond(m,nd_ps[i]);
1825: g = nd_separate_head(g,&head);
1826: stat = nd_nf(m,g,nd_ps,1,&dn,&nf);
1.50 noro 1827: if ( !stat )
1.103 noro 1828: nd_reconstruct(0,0);
1.69 noro 1829: else {
1.71 noro 1830: if ( DP_Print ) { printf("."); fflush(stdout); }
1.85 noro 1831: if ( !m ) {
1.113 noro 1832: NTOQ(NM(dn.z),SGN(dn.z),num);
1833: mulq(HCQ(head),num,&q); HCQ(head) = q;
1834: if ( DN(dn.z) ) {
1835: NTOQ(DN(dn.z),1,den);
1836: nd_mul_c_q(nf,den);
1837: }
1.85 noro 1838: }
1.69 noro 1839: nf = nd_add(m,head,nf);
1840: ndv_free(nd_ps[i]);
1.24 noro 1841: nd_removecont(m,nf);
1.69 noro 1842: nd_ps[i] = ndtondv(m,nf); nd_free(nf);
1843: nd_bound[i] = ndv_compute_bound(nd_ps[i]);
1.50 noro 1844: i++;
1.23 noro 1845: }
1846: }
1.71 noro 1847: if ( DP_Print ) { printf("\n"); }
1.23 noro 1848: for ( a0 = 0, i = 0; i < n; i++ ) {
1849: NEXTNODE(a0,a);
1.69 noro 1850: BDY(a) = (pointer)nd_ps[i];
1.23 noro 1851: }
1852: NEXT(a) = 0;
1853: return a0;
1854: }
1855:
1.1 noro 1856: ND_pairs update_pairs( ND_pairs d, NODE /* of index */ g, int t)
1857: {
1858: ND_pairs d1,nd,cur,head,prev,remove;
1859:
1860: if ( !g ) return d;
1861: d = crit_B(d,t);
1862: d1 = nd_newpairs(g,t);
1863: d1 = crit_M(d1);
1864: d1 = crit_F(d1);
1.55 noro 1865: if ( do_weyl )
1866: head = d1;
1867: else {
1868: prev = 0; cur = head = d1;
1869: while ( cur ) {
1870: if ( crit_2( cur->i1,cur->i2 ) ) {
1871: remove = cur;
1872: if ( !prev ) head = cur = NEXT(cur);
1873: else cur = NEXT(prev) = NEXT(cur);
1874: FREENDP(remove);
1875: } else {
1876: prev = cur; cur = NEXT(cur);
1877: }
1.1 noro 1878: }
1879: }
1880: if ( !d )
1881: return head;
1882: else {
1883: nd = d;
1.34 noro 1884: while ( NEXT(nd) ) nd = NEXT(nd);
1.1 noro 1885: NEXT(nd) = head;
1886: return d;
1887: }
1888: }
1889:
1890: ND_pairs nd_newpairs( NODE g, int t )
1891: {
1892: NODE h;
1.61 noro 1893: UINT *dl;
1.34 noro 1894: int ts,s;
1.1 noro 1895: ND_pairs r,r0;
1896:
1.20 noro 1897: dl = DL(nd_psh[t]);
1.34 noro 1898: ts = SG(nd_psh[t]) - TD(dl);
1.1 noro 1899: for ( r0 = 0, h = g; h; h = NEXT(h) ) {
1900: NEXTND_pairs(r0,r);
1901: r->i1 = (int)BDY(h);
1902: r->i2 = t;
1.20 noro 1903: ndl_lcm(DL(nd_psh[r->i1]),dl,r->lcm);
1.34 noro 1904: s = SG(nd_psh[r->i1])-TD(DL(nd_psh[r->i1]));
1905: SG(r) = MAX(s,ts) + TD(LCM(r));
1.1 noro 1906: }
1907: NEXT(r) = 0;
1908: return r0;
1909: }
1910:
1911: ND_pairs crit_B( ND_pairs d, int s )
1912: {
1913: ND_pairs cur,head,prev,remove;
1.61 noro 1914: UINT *t,*tl,*lcm;
1.1 noro 1915: int td,tdl;
1916:
1917: if ( !d ) return 0;
1.20 noro 1918: t = DL(nd_psh[s]);
1.1 noro 1919: prev = 0;
1920: head = cur = d;
1.61 noro 1921: lcm = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
1.1 noro 1922: while ( cur ) {
1923: tl = cur->lcm;
1924: if ( ndl_reducible(tl,t)
1.20 noro 1925: && (ndl_lcm(DL(nd_psh[cur->i1]),t,lcm),!ndl_equal(lcm,tl))
1926: && (ndl_lcm(DL(nd_psh[cur->i2]),t,lcm),!ndl_equal(lcm,tl)) ) {
1.1 noro 1927: remove = cur;
1928: if ( !prev ) {
1929: head = cur = NEXT(cur);
1930: } else {
1931: cur = NEXT(prev) = NEXT(cur);
1932: }
1933: FREENDP(remove);
1934: } else {
1.34 noro 1935: prev = cur; cur = NEXT(cur);
1.1 noro 1936: }
1937: }
1938: return head;
1939: }
1940:
1941: ND_pairs crit_M( ND_pairs d1 )
1942: {
1943: ND_pairs e,d2,d3,dd,p;
1.61 noro 1944: UINT *id,*jd;
1.1 noro 1945:
1946: for ( dd = 0, e = d1; e; e = d3 ) {
1947: if ( !(d2 = NEXT(e)) ) {
1948: NEXT(e) = dd;
1949: return e;
1950: }
1.34 noro 1951: id = LCM(e);
1.1 noro 1952: for ( d3 = 0; d2; d2 = p ) {
1.34 noro 1953: p = NEXT(d2);
1954: jd = LCM(d2);
1955: if ( ndl_equal(jd,id) )
1956: ;
1957: else if ( TD(jd) > TD(id) )
1.1 noro 1958: if ( ndl_reducible(jd,id) ) continue;
1959: else ;
1.34 noro 1960: else if ( ndl_reducible(id,jd) ) goto delit;
1.1 noro 1961: NEXT(d2) = d3;
1962: d3 = d2;
1963: }
1964: NEXT(e) = dd;
1965: dd = e;
1966: continue;
1967: /**/
1968: delit: NEXT(d2) = d3;
1969: d3 = d2;
1970: for ( ; p; p = d2 ) {
1971: d2 = NEXT(p);
1972: NEXT(p) = d3;
1973: d3 = p;
1974: }
1975: FREENDP(e);
1976: }
1977: return dd;
1978: }
1979:
1980: ND_pairs crit_F( ND_pairs d1 )
1981: {
1982: ND_pairs rest, head,remove;
1983: ND_pairs last, p, r, w;
1984: int s;
1985:
1986: for ( head = last = 0, p = d1; NEXT(p); ) {
1987: r = w = equivalent_pairs(p,&rest);
1.14 noro 1988: s = SG(r);
1.1 noro 1989: w = NEXT(w);
1990: while ( w ) {
1991: if ( crit_2(w->i1,w->i2) ) {
1992: r = w;
1993: w = NEXT(w);
1994: while ( w ) {
1995: remove = w;
1996: w = NEXT(w);
1997: FREENDP(remove);
1998: }
1999: break;
1.14 noro 2000: } else if ( SG(w) < s ) {
1.1 noro 2001: FREENDP(r);
2002: r = w;
1.14 noro 2003: s = SG(r);
1.1 noro 2004: w = NEXT(w);
2005: } else {
2006: remove = w;
2007: w = NEXT(w);
2008: FREENDP(remove);
2009: }
2010: }
2011: if ( last ) NEXT(last) = r;
2012: else head = r;
2013: NEXT(last = r) = 0;
2014: p = rest;
2015: if ( !p ) return head;
2016: }
2017: if ( !last ) return p;
2018: NEXT(last) = p;
2019: return head;
2020: }
2021:
2022: int crit_2( int dp1, int dp2 )
2023: {
1.20 noro 2024: return ndl_disjoint(DL(nd_psh[dp1]),DL(nd_psh[dp2]));
1.1 noro 2025: }
2026:
1.40 noro 2027: ND_pairs equivalent_pairs( ND_pairs d1, ND_pairs *prest )
1.1 noro 2028: {
2029: ND_pairs w,p,r,s;
1.61 noro 2030: UINT *d;
1.1 noro 2031:
2032: w = d1;
1.34 noro 2033: d = LCM(w);
1.1 noro 2034: s = NEXT(w);
2035: NEXT(w) = 0;
2036: for ( r = 0; s; s = p ) {
2037: p = NEXT(s);
1.34 noro 2038: if ( ndl_equal(d,LCM(s)) ) {
1.39 noro 2039: NEXT(s) = w; w = s;
1.1 noro 2040: } else {
1.39 noro 2041: NEXT(s) = r; r = s;
1.1 noro 2042: }
2043: }
2044: *prest = r;
2045: return w;
2046: }
2047:
2048: NODE update_base(NODE nd,int ndp)
2049: {
1.61 noro 2050: UINT *dl, *dln;
1.1 noro 2051: NODE last, p, head;
2052:
1.20 noro 2053: dl = DL(nd_psh[ndp]);
1.1 noro 2054: for ( head = last = 0, p = nd; p; ) {
1.20 noro 2055: dln = DL(nd_psh[(int)BDY(p)]);
1.34 noro 2056: if ( ndl_reducible( dln, dl ) ) {
1.1 noro 2057: p = NEXT(p);
2058: if ( last ) NEXT(last) = p;
2059: } else {
2060: if ( !last ) head = p;
2061: p = NEXT(last = p);
2062: }
2063: }
2064: head = append_one(head,ndp);
2065: return head;
2066: }
2067:
2068: ND_pairs nd_minp( ND_pairs d, ND_pairs *prest )
2069: {
2070: ND_pairs m,ml,p,l;
1.61 noro 2071: UINT *lcm;
1.33 noro 2072: int s,td,len,tlen,c,c1;
1.1 noro 2073:
2074: if ( !(p = NEXT(m = d)) ) {
2075: *prest = p;
2076: NEXT(m) = 0;
2077: return m;
2078: }
1.14 noro 2079: s = SG(m);
1.77 noro 2080: if ( !NoSugar ) {
2081: for ( ml = 0, l = m; p; p = NEXT(l = p) )
2082: if ( (SG(p) < s)
2083: || ((SG(p) == s) && (DL_COMPARE(LCM(p),LCM(m)) < 0)) ) {
2084: ml = l; m = p; s = SG(m);
2085: }
2086: } else {
2087: for ( ml = 0, l = m; p; p = NEXT(l = p) )
2088: if ( DL_COMPARE(LCM(p),LCM(m)) < 0 ) {
2089: ml = l; m = p; s = SG(m);
2090: }
2091: }
1.1 noro 2092: if ( !ml ) *prest = NEXT(m);
2093: else {
2094: NEXT(ml) = NEXT(m);
2095: *prest = d;
2096: }
2097: NEXT(m) = 0;
2098: return m;
2099: }
2100:
1.63 noro 2101: ND_pairs nd_minsugarp( ND_pairs d, ND_pairs *prest )
2102: {
1.69 noro 2103: int msugar,i;
1.63 noro 2104: ND_pairs t,dm0,dm,dr0,dr;
2105:
2106: for ( msugar = SG(d), t = NEXT(d); t; t = NEXT(t) )
2107: if ( SG(t) < msugar ) msugar = SG(t);
2108: dm0 = 0; dr0 = 0;
1.69 noro 2109: for ( i = 0, t = d; t; t = NEXT(t) )
2110: if ( i < nd_f4_nsp && SG(t) == msugar ) {
1.63 noro 2111: if ( dm0 ) NEXT(dm) = t;
2112: else dm0 = t;
2113: dm = t;
1.69 noro 2114: i++;
1.63 noro 2115: } else {
2116: if ( dr0 ) NEXT(dr) = t;
2117: else dr0 = t;
2118: dr = t;
2119: }
2120: NEXT(dm) = 0;
2121: if ( dr0 ) NEXT(dr) = 0;
2122: *prest = dr0;
2123: return dm0;
2124: }
2125:
1.77 noro 2126: int ndv_newps(int m,NDV a,NDV aq)
1.1 noro 2127: {
1.3 noro 2128: int len;
1.13 noro 2129: RHist r;
1.20 noro 2130: NDV b;
1.3 noro 2131:
1.1 noro 2132: if ( nd_psn == nd_pslen ) {
2133: nd_pslen *= 2;
1.11 noro 2134: nd_ps = (NDV *)REALLOC((char *)nd_ps,nd_pslen*sizeof(NDV));
1.53 noro 2135: nd_ps_trace = (NDV *)REALLOC((char *)nd_ps_trace,nd_pslen*sizeof(NDV));
1.13 noro 2136: nd_psh = (RHist *)REALLOC((char *)nd_psh,nd_pslen*sizeof(RHist));
1.61 noro 2137: nd_bound = (UINT **)
2138: REALLOC((char *)nd_bound,nd_pslen*sizeof(UINT *));
1.1 noro 2139: }
1.39 noro 2140: NEWRHist(r); nd_psh[nd_psn] = r;
1.63 noro 2141: nd_ps[nd_psn] = a;
1.39 noro 2142: if ( aq ) {
1.63 noro 2143: nd_ps_trace[nd_psn] = aq;
2144: nd_bound[nd_psn] = ndv_compute_bound(aq);
1.39 noro 2145: SG(r) = SG(aq); ndl_copy(HDL(aq),DL(r));
1.53 noro 2146: } else {
1.63 noro 2147: nd_bound[nd_psn] = ndv_compute_bound(a);
1.53 noro 2148: SG(r) = SG(a); ndl_copy(HDL(a),DL(r));
1.39 noro 2149: }
1.77 noro 2150: if ( nd_demand ) {
2151: if ( aq ) {
2152: ndv_save(nd_ps_trace[nd_psn],nd_psn);
2153: nd_ps_trace[nd_psn] = 0;
2154: } else {
2155: ndv_save(nd_ps[nd_psn],nd_psn);
2156: nd_ps[nd_psn] = 0;
2157: }
2158: }
1.1 noro 2159: return nd_psn++;
2160: }
2161:
1.61 noro 2162: void ndv_setup(int mod,int trace,NODE f)
1.1 noro 2163: {
1.5 noro 2164: int i,j,td,len,max;
1.1 noro 2165: NODE s,s0,f0;
1.61 noro 2166: UINT *d;
1.13 noro 2167: RHist r;
1.69 noro 2168: NDV *w;
1.61 noro 2169: NDV a,am;
1.11 noro 2170:
2171: nd_found = 0; nd_notfirst = 0; nd_create = 0;
1.1 noro 2172:
1.69 noro 2173: for ( nd_psn = 0, s = f; s; s = NEXT(s) ) if ( BDY(s) ) nd_psn++;
2174: w = (NDV *)ALLOCA(nd_psn*sizeof(NDV));
2175: for ( i = 0, s = f; s; s = NEXT(s) ) if ( BDY(s) ) w[i++] = BDY(s);
2176: qsort(w,nd_psn,sizeof(NDV),
2177: (int (*)(const void *,const void *))ndv_compare);
2178: nd_pslen = 2*nd_psn;
1.11 noro 2179: nd_ps = (NDV *)MALLOC(nd_pslen*sizeof(NDV));
1.53 noro 2180: nd_ps_trace = (NDV *)MALLOC(nd_pslen*sizeof(NDV));
1.13 noro 2181: nd_psh = (RHist *)MALLOC(nd_pslen*sizeof(RHist));
1.61 noro 2182: nd_bound = (UINT **)MALLOC(nd_pslen*sizeof(UINT *));
1.57 noro 2183:
1.11 noro 2184: if ( !nd_red )
1.13 noro 2185: nd_red = (RHist *)MALLOC(REDTAB_LEN*sizeof(RHist));
1.71 noro 2186: for ( i = 0; i < REDTAB_LEN; i++ ) nd_red[i] = 0;
1.69 noro 2187: for ( i = 0; i < nd_psn; i++ ) {
1.39 noro 2188: if ( trace ) {
1.69 noro 2189: a = nd_ps_trace[i] = ndv_dup(0,w[i]);
1.61 noro 2190: ndv_removecont(0,a);
2191: am = nd_ps[i] = ndv_dup(mod,a);
2192: ndv_mod(mod,am);
2193: ndv_removecont(mod,am);
2194: } else {
1.69 noro 2195: a = nd_ps[i] = ndv_dup(mod,w[i]);
1.61 noro 2196: ndv_removecont(mod,a);
1.39 noro 2197: }
1.61 noro 2198: NEWRHist(r); SG(r) = HTD(a); ndl_copy(HDL(a),DL(r));
1.57 noro 2199: nd_bound[i] = ndv_compute_bound(a);
1.20 noro 2200: nd_psh[i] = r;
1.77 noro 2201: if ( nd_demand ) {
2202: if ( trace ) {
2203: ndv_save(nd_ps_trace[i],i);
2204: nd_ps_trace[i] = 0;
2205: } else {
2206: ndv_save(nd_ps[i],i);
2207: nd_ps[i] = 0;
2208: }
2209: }
1.20 noro 2210: }
2211: }
2212:
1.63 noro 2213: void nd_gr(LIST f,LIST v,int m,int f4,struct order_spec *ord,LIST *rp)
1.1 noro 2214: {
1.61 noro 2215: VL tv,fv,vv,vc;
1.1 noro 2216: NODE fd,fd0,r,r0,t,x,s,xx;
1.61 noro 2217: int e,max,nvar;
1.69 noro 2218: NDV b;
1.77 noro 2219: int ishomo;
1.1 noro 2220:
1.78 noro 2221: if ( !m && Demand ) nd_demand = 1;
2222: else nd_demand = 0;
2223:
1.103 noro 2224: #if 0
1.74 noro 2225: ndv_alloc = 0;
1.103 noro 2226: #endif
1.1 noro 2227: get_vars((Obj)f,&fv); pltovl(v,&vv);
1.74 noro 2228: for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++ );
1.88 noro 2229: switch ( ord->id ) {
2230: case 1:
2231: if ( ord->nv != nvar )
2232: error("nd_{gr,f4} : invalid order specification");
2233: break;
2234: default:
2235: break;
2236: }
1.32 noro 2237: nd_init_ord(ord);
1.61 noro 2238: for ( t = BDY(f), max = 0; t; t = NEXT(t) )
2239: for ( tv = vv; tv; tv = NEXT(tv) ) {
2240: e = getdeg(tv->v,(P)BDY(t));
2241: max = MAX(e,max);
2242: }
2243: nd_setup_parameters(nvar,max);
1.77 noro 2244: ishomo = 1;
1.1 noro 2245: for ( fd0 = 0, t = BDY(f); t; t = NEXT(t) ) {
1.61 noro 2246: b = (pointer)ptondv(CO,vv,(P)BDY(t));
1.77 noro 2247: if ( ishomo )
2248: ishomo = ishomo && ndv_ishomo(b);
1.69 noro 2249: if ( m ) ndv_mod(m,b);
1.61 noro 2250: if ( b ) { NEXTNODE(fd0,fd); BDY(fd) = (pointer)b; }
1.1 noro 2251: }
2252: if ( fd0 ) NEXT(fd) = 0;
1.61 noro 2253: ndv_setup(m,0,fd0);
1.77 noro 2254: x = f4?nd_f4(m):nd_gb(m,ishomo,0);
1.82 noro 2255: nd_demand = 0;
1.61 noro 2256: x = ndv_reducebase(x);
2257: x = ndv_reduceall(m,x);
1.23 noro 2258: for ( r0 = 0, t = x; t; t = NEXT(t) ) {
1.1 noro 2259: NEXTNODE(r0,r);
1.61 noro 2260: BDY(r) = ndvtop(m,CO,vv,BDY(t));
1.20 noro 2261: }
2262: if ( r0 ) NEXT(r) = 0;
2263: MKLIST(*rp,r0);
1.103 noro 2264: #if 0
1.74 noro 2265: fprintf(asir_out,"ndv_alloc=%d\n",ndv_alloc);
1.103 noro 2266: #endif
1.20 noro 2267: }
2268:
1.52 noro 2269: void nd_gr_trace(LIST f,LIST v,int trace,int homo,struct order_spec *ord,LIST *rp)
1.20 noro 2270: {
1.89 noro 2271: struct order_spec *ord1;
1.61 noro 2272: VL tv,fv,vv,vc;
1.27 noro 2273: NODE fd,fd0,in0,in,r,r0,t,s,cand;
1.61 noro 2274: int m,nocheck,nvar,mindex,e,max;
2275: NDV c;
2276: NMV a;
1.27 noro 2277: P p;
1.61 noro 2278: EPOS oepos;
1.82 noro 2279: int obpe,oadv,wmax,i,len,cbpe,ishomo;
1.20 noro 2280:
2281: get_vars((Obj)f,&fv); pltovl(v,&vv);
1.74 noro 2282: for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++ );
1.88 noro 2283: switch ( ord->id ) {
2284: case 1:
2285: if ( ord->nv != nvar )
2286: error("nd_gr_trace : invalid order specification");
2287: break;
2288: default:
2289: break;
2290: }
1.52 noro 2291: nocheck = 0;
2292: mindex = 0;
2293:
1.78 noro 2294: if ( Demand ) nd_demand = 1;
2295: else nd_demand = 0;
2296:
1.52 noro 2297: /* setup modulus */
2298: if ( trace < 0 ) {
2299: trace = -trace;
2300: nocheck = 1;
2301: }
2302: m = trace > 1 ? trace : get_lprime(mindex);
1.61 noro 2303: for ( t = BDY(f), max = 0; t; t = NEXT(t) )
2304: for ( tv = vv; tv; tv = NEXT(tv) ) {
2305: e = getdeg(tv->v,(P)BDY(t));
2306: max = MAX(e,max);
1.23 noro 2307: }
1.61 noro 2308: nd_init_ord(ord);
2309: nd_setup_parameters(nvar,max);
2310: obpe = nd_bpe; oadv = nmv_adv; oepos = nd_epos;
1.82 noro 2311: ishomo = 1;
1.61 noro 2312: for ( in0 = 0, fd0 = 0, t = BDY(f); t; t = NEXT(t) ) {
2313: c = ptondv(CO,vv,(P)BDY(t));
1.82 noro 2314: if ( ishomo )
2315: ishomo = ishomo && ndv_ishomo(c);
1.61 noro 2316: if ( c ) {
2317: NEXTNODE(in0,in); BDY(in) = (pointer)c;
2318: NEXTNODE(fd0,fd); BDY(fd) = (pointer)ndv_dup(0,c);
1.23 noro 2319: }
1.61 noro 2320: }
2321: if ( in0 ) NEXT(in) = 0;
2322: if ( fd0 ) NEXT(fd) = 0;
1.82 noro 2323: if ( !ishomo && homo ) {
1.61 noro 2324: for ( t = in0, wmax = 0; t; t = NEXT(t) ) {
2325: c = (NDV)BDY(t); len = LEN(c);
2326: for ( a = BDY(c), i = 0; i < len; i++, NMV_ADV(a) )
2327: wmax = MAX(TD(DL(a)),wmax);
2328: }
2329: homogenize_order(ord,nvar,&ord1);
1.89 noro 2330: nd_init_ord(ord1);
1.61 noro 2331: nd_setup_parameters(nvar+1,wmax);
2332: for ( t = fd0; t; t = NEXT(t) )
2333: ndv_homogenize((NDV)BDY(t),obpe,oadv,oepos);
1.27 noro 2334: }
1.52 noro 2335: while ( 1 ) {
1.77 noro 2336: if ( Demand )
2337: nd_demand = 1;
1.61 noro 2338: ndv_setup(m,1,fd0);
1.82 noro 2339: cand = nd_gb_trace(m,ishomo || homo);
1.52 noro 2340: if ( !cand ) {
2341: /* failure */
1.61 noro 2342: if ( trace > 1 ) { *rp = 0; return; }
2343: else m = get_lprime(++mindex);
1.52 noro 2344: continue;
2345: }
1.82 noro 2346: if ( !ishomo && homo ) {
1.27 noro 2347: /* dehomogenization */
1.61 noro 2348: for ( t = cand; t; t = NEXT(t) ) ndv_dehomogenize((NDV)BDY(t),ord);
1.45 noro 2349: nd_init_ord(ord);
1.61 noro 2350: nd_setup_parameters(nvar,0);
1.27 noro 2351: }
1.77 noro 2352: nd_demand = 0;
1.61 noro 2353: cand = ndv_reducebase(cand);
2354: cand = ndv_reduceall(0,cand);
1.78 noro 2355: cbpe = nd_bpe;
1.61 noro 2356: if ( nocheck )
2357: break;
2358: if ( ndv_check_candidate(in0,obpe,oadv,oepos,cand) )
1.52 noro 2359: /* success */
2360: break;
2361: else if ( trace > 1 ) {
2362: /* failure */
2363: *rp = 0; return;
1.61 noro 2364: } else {
1.52 noro 2365: /* try the next modulus */
2366: m = get_lprime(++mindex);
1.61 noro 2367: /* reset the parameters */
1.82 noro 2368: if ( !ishomo && homo ) {
1.89 noro 2369: nd_init_ord(ord1);
1.61 noro 2370: nd_setup_parameters(nvar+1,wmax);
2371: } else {
2372: nd_init_ord(ord);
2373: nd_setup_parameters(nvar,max);
2374: }
2375: }
1.52 noro 2376: }
1.27 noro 2377: /* dp->p */
1.61 noro 2378: nd_bpe = cbpe;
1.73 noro 2379: nd_setup_parameters(nd_nvar,0);
1.61 noro 2380: for ( r = cand; r; r = NEXT(r) ) BDY(r) = (pointer)ndvtop(0,CO,vv,BDY(r));
1.27 noro 2381: MKLIST(*rp,cand);
1.1 noro 2382: }
2383:
1.61 noro 2384: void dltondl(int n,DL dl,UINT *r)
1.1 noro 2385: {
1.61 noro 2386: UINT *d;
1.57 noro 2387: int i,j,l,s,ord_l;
1.43 noro 2388: struct order_pair *op;
1.1 noro 2389:
2390: d = dl->d;
1.41 noro 2391: for ( i = 0; i < nd_wpd; i++ ) r[i] = 0;
1.43 noro 2392: if ( nd_blockmask ) {
2393: l = nd_blockmask->n;
2394: op = nd_blockmask->order_pair;
2395: for ( j = 0, s = 0; j < l; j++ ) {
2396: ord_l = op[j].length;
1.57 noro 2397: for ( i = 0; i < ord_l; i++, s++ ) PUT_EXP(r,s,d[s]);
1.43 noro 2398: }
2399: TD(r) = ndl_weight(r);
1.61 noro 2400: ndl_weight_mask(r);
1.43 noro 2401: } else {
1.56 noro 2402: for ( i = 0; i < n; i++ ) PUT_EXP(r,i,d[i]);
1.43 noro 2403: TD(r) = ndl_weight(r);
2404: }
1.1 noro 2405: }
2406:
1.61 noro 2407: DL ndltodl(int n,UINT *ndl)
1.1 noro 2408: {
2409: DL dl;
2410: int *d;
1.57 noro 2411: int i,j,l,s,ord_l;
1.43 noro 2412: struct order_pair *op;
1.1 noro 2413:
2414: NEWDL(dl,n);
1.34 noro 2415: dl->td = TD(ndl);
1.1 noro 2416: d = dl->d;
1.43 noro 2417: if ( nd_blockmask ) {
2418: l = nd_blockmask->n;
2419: op = nd_blockmask->order_pair;
2420: for ( j = 0, s = 0; j < l; j++ ) {
2421: ord_l = op[j].length;
1.57 noro 2422: for ( i = 0; i < ord_l; i++, s++ ) d[s] = GET_EXP(ndl,s);
1.43 noro 2423: }
2424: } else {
1.56 noro 2425: for ( i = 0; i < n; i++ ) d[i] = GET_EXP(ndl,i);
1.43 noro 2426: }
1.1 noro 2427: return dl;
2428: }
2429:
1.61 noro 2430: void ndl_print(UINT *dl)
1.1 noro 2431: {
2432: int n;
1.57 noro 2433: int i,j,l,ord_l,s,s0;
1.43 noro 2434: struct order_pair *op;
1.1 noro 2435:
2436: n = nd_nvar;
2437: printf("<<");
1.43 noro 2438: if ( nd_blockmask ) {
2439: l = nd_blockmask->n;
2440: op = nd_blockmask->order_pair;
2441: for ( j = 0, s = s0 = 0; j < l; j++ ) {
2442: ord_l = op[j].length;
1.57 noro 2443: for ( i = 0; i < ord_l; i++, s++ )
2444: printf(s==n-1?"%d":"%d,",GET_EXP(dl,s));
1.43 noro 2445: }
2446: } else {
1.56 noro 2447: for ( i = 0; i < n; i++ ) printf(i==n-1?"%d":"%d,",GET_EXP(dl,i));
1.43 noro 2448: }
1.1 noro 2449: printf(">>");
2450: }
2451:
2452: void nd_print(ND p)
2453: {
2454: NM m;
2455:
2456: if ( !p )
2457: printf("0\n");
2458: else {
2459: for ( m = BDY(p); m; m = NEXT(m) ) {
1.71 noro 2460: if ( CM(m) & 0x80000000 ) printf("+@_%d*",IFTOF(CM(m)));
2461: else printf("+%d*",CM(m));
1.14 noro 2462: ndl_print(DL(m));
1.1 noro 2463: }
2464: printf("\n");
2465: }
2466: }
2467:
1.113 noro 2468: void nd_print_q(ND p)
1.16 noro 2469: {
2470: NM m;
2471:
2472: if ( !p )
2473: printf("0\n");
2474: else {
2475: for ( m = BDY(p); m; m = NEXT(m) ) {
2476: printf("+");
1.113 noro 2477: printexpr(CO,(Obj)CQ(m));
1.16 noro 2478: printf("*");
2479: ndl_print(DL(m));
2480: }
2481: printf("\n");
2482: }
2483: }
2484:
1.1 noro 2485: void ndp_print(ND_pairs d)
2486: {
2487: ND_pairs t;
2488:
1.34 noro 2489: for ( t = d; t; t = NEXT(t) ) printf("%d,%d ",t->i1,t->i2);
1.1 noro 2490: printf("\n");
2491: }
2492:
1.20 noro 2493: void nd_removecont(int mod,ND p)
1.16 noro 2494: {
2495: int i,n;
1.113 noro 2496: Q *w;
2497: Q dvr,t;
1.16 noro 2498: NM m;
1.21 noro 2499: struct oVECT v;
1.113 noro 2500: N q,r;
1.16 noro 2501:
1.71 noro 2502: if ( mod == -1 ) nd_mul_c(mod,p,_invsf(HCM(p)));
2503: else if ( mod ) nd_mul_c(mod,p,invm(HCM(p),mod));
1.20 noro 2504: else {
2505: for ( m = BDY(p), n = 0; m; m = NEXT(m), n++ );
1.113 noro 2506: w = (Q *)ALLOCA(n*sizeof(Q));
1.21 noro 2507: v.len = n;
2508: v.body = (pointer *)w;
1.113 noro 2509: for ( m = BDY(p), i = 0; i < n; m = NEXT(m), i++ ) w[i] = CQ(m);
1.21 noro 2510: removecont_array(w,n);
1.113 noro 2511: for ( m = BDY(p), i = 0; i < n; m = NEXT(m), i++ ) CQ(m) = w[i];
1.16 noro 2512: }
2513: }
2514:
1.21 noro 2515: void nd_removecont2(ND p1,ND p2)
2516: {
2517: int i,n1,n2,n;
1.113 noro 2518: Q *w;
2519: Q dvr,t;
1.21 noro 2520: NM m;
2521: struct oVECT v;
1.113 noro 2522: N q,r;
1.21 noro 2523:
2524: if ( !p1 ) {
2525: nd_removecont(0,p2); return;
2526: } else if ( !p2 ) {
2527: nd_removecont(0,p1); return;
2528: }
2529: n1 = nd_length(p1);
2530: n2 = nd_length(p2);
2531: n = n1+n2;
1.113 noro 2532: w = (Q *)ALLOCA(n*sizeof(Q));
1.21 noro 2533: v.len = n;
2534: v.body = (pointer *)w;
1.113 noro 2535: for ( m = BDY(p1), i = 0; i < n1; m = NEXT(m), i++ ) w[i] = CQ(m);
2536: for ( m = BDY(p2); i < n; m = NEXT(m), i++ ) w[i] = CQ(m);
1.21 noro 2537: removecont_array(w,n);
1.113 noro 2538: for ( m = BDY(p1), i = 0; i < n1; m = NEXT(m), i++ ) CQ(m) = w[i];
2539: for ( m = BDY(p2); i < n; m = NEXT(m), i++ ) CQ(m) = w[i];
1.21 noro 2540: }
2541:
1.20 noro 2542: void ndv_removecont(int mod,NDV p)
1.16 noro 2543: {
2544: int i,len;
1.113 noro 2545: Q *w;
2546: Q dvr,t;
1.16 noro 2547: NMV m;
2548:
1.71 noro 2549: if ( mod == -1 )
2550: ndv_mul_c(mod,p,_invsf(HCM(p)));
2551: else if ( mod )
1.20 noro 2552: ndv_mul_c(mod,p,invm(HCM(p),mod));
2553: else {
2554: len = p->len;
1.113 noro 2555: w = (Q *)ALLOCA(len*sizeof(Q));
2556: for ( m = BDY(p), i = 0; i < len; NMV_ADV(m), i++ ) w[i] = CQ(m);
2557: sortbynm(w,len);
2558: qltozl(w,len,&dvr);
2559: for ( m = BDY(p), i = 0; i < len; NMV_ADV(m), i++ ) {
2560: divq(CQ(m),dvr,&t); CQ(m) = t;
2561: }
1.16 noro 2562: }
1.21 noro 2563: }
2564:
1.61 noro 2565: void ndv_homogenize(NDV p,int obpe,int oadv,EPOS oepos)
2566: {
2567: int len,i,max;
2568: NMV m,mr0,mr,t;
2569:
2570: len = p->len;
2571: for ( m = BDY(p), i = 0, max = 0; i < len; NMV_OADV(m), i++ )
2572: max = MAX(max,TD(DL(m)));
2573: mr0 = nmv_adv>oadv?(NMV)REALLOC(BDY(p),len*nmv_adv):BDY(p);
2574: m = (NMV)((char *)mr0+(len-1)*oadv);
2575: mr = (NMV)((char *)mr0+(len-1)*nmv_adv);
2576: t = (NMV)ALLOCA(nmv_adv);
2577: for ( i = 0; i < len; i++, NMV_OPREV(m), NMV_PREV(mr) ) {
2578: ndl_homogenize(DL(m),DL(t),obpe,oepos,max);
1.113 noro 2579: CQ(mr) = CQ(m);
1.61 noro 2580: ndl_copy(DL(t),DL(mr));
2581: }
2582: NV(p)++;
2583: BDY(p) = mr0;
2584: }
2585:
1.45 noro 2586: void ndv_dehomogenize(NDV p,struct order_spec *ord)
1.23 noro 2587: {
1.45 noro 2588: int i,j,adj,len,newnvar,newwpd,newadv,newexporigin;
1.113 noro 2589: Q *w;
2590: Q dvr,t;
1.23 noro 2591: NMV m,r;
2592:
2593: len = p->len;
2594: newnvar = nd_nvar-1;
1.48 noro 2595: newexporigin = nd_get_exporigin(ord);
1.45 noro 2596: newwpd = newnvar/nd_epw+(newnvar%nd_epw?1:0)+newexporigin;
1.23 noro 2597: for ( m = BDY(p), i = 0; i < len; NMV_ADV(m), i++ )
1.34 noro 2598: ndl_dehomogenize(DL(m));
1.23 noro 2599: if ( newwpd != nd_wpd ) {
1.90 noro 2600: newadv = ROUND_FOR_ALIGN(sizeof(struct oNMV)+(newwpd-1)*sizeof(UINT));
1.61 noro 2601: for ( m = r = BDY(p), i = 0; i < len; NMV_ADV(m), NDV_NADV(r), i++ ) {
1.113 noro 2602: CQ(r) = CQ(m);
1.45 noro 2603: for ( j = 0; j < newexporigin; j++ ) DL(r)[j] = DL(m)[j];
2604: adj = nd_exporigin-newexporigin;
2605: for ( ; j < newwpd; j++ ) DL(r)[j] = DL(m)[j+adj];
1.23 noro 2606: }
2607: }
2608: NV(p)--;
2609: }
2610:
1.113 noro 2611: void removecont_array(Q *c,int n)
1.21 noro 2612: {
2613: struct oVECT v;
1.113 noro 2614: Q d0,d1,a,u,u1,gcd;
1.110 noro 2615: int i,j;
1.113 noro 2616: N qn,rn,gn;
2617: Q *q,*r;
1.21 noro 2618:
1.113 noro 2619: q = (Q *)ALLOCA(n*sizeof(Q));
2620: r = (Q *)ALLOCA(n*sizeof(Q));
1.21 noro 2621: v.id = O_VECT; v.len = n; v.body = (pointer *)c;
1.113 noro 2622: igcdv_estimate(&v,&d0);
2623: for ( i = 0; i < n; i++ ) {
2624: divn(NM(c[i]),NM(d0),&qn,&rn);
2625: NTOQ(qn,SGN(c[i])*SGN(d0),q[i]);
2626: NTOQ(rn,SGN(c[i]),r[i]);
2627: }
1.34 noro 2628: for ( i = 0; i < n; i++ ) if ( r[i] ) break;
1.21 noro 2629: if ( i < n ) {
2630: v.id = O_VECT; v.len = n; v.body = (pointer *)r;
1.113 noro 2631: igcdv(&v,&d1);
2632: gcdn(NM(d0),NM(d1),&gn); NTOQ(gn,1,gcd);
2633: divsn(NM(d0),gn,&qn); NTOQ(qn,1,a);
1.21 noro 2634: for ( i = 0; i < n; i++ ) {
1.113 noro 2635: mulq(a,q[i],&u);
2636: if ( r[i] ) {
2637: divsn(NM(r[i]),gn,&qn); NTOQ(qn,SGN(r[i]),u1);
2638: addq(u,u1,&q[i]);
2639: } else
2640: q[i] = u;
1.21 noro 2641: }
2642: }
1.34 noro 2643: for ( i = 0; i < n; i++ ) c[i] = q[i];
1.16 noro 2644: }
2645:
1.19 noro 2646: void nd_mul_c(int mod,ND p,int mul)
1.1 noro 2647: {
2648: NM m;
2649: int c,c1;
2650:
1.34 noro 2651: if ( !p ) return;
1.115 noro 2652: if ( mul == 1 ) return;
1.71 noro 2653: if ( mod == -1 )
2654: for ( m = BDY(p); m; m = NEXT(m) )
2655: CM(m) = _mulsf(CM(m),mul);
2656: else
2657: for ( m = BDY(p); m; m = NEXT(m) ) {
2658: c1 = CM(m); DMAR(c1,mul,0,mod,c); CM(m) = c;
2659: }
1.1 noro 2660: }
2661:
1.113 noro 2662: void nd_mul_c_q(ND p,Q mul)
1.16 noro 2663: {
2664: NM m;
1.113 noro 2665: Q c;
1.16 noro 2666:
1.113 noro 2667: if ( !p ) return;
1.115 noro 2668: if ( UNIQ(mul) ) return;
1.16 noro 2669: for ( m = BDY(p); m; m = NEXT(m) ) {
1.113 noro 2670: mulq(CQ(m),mul,&c); CQ(m) = c;
1.16 noro 2671: }
2672: }
2673:
1.61 noro 2674: void nd_mul_c_p(VL vl,ND p,P mul)
2675: {
2676: NM m;
2677: P c;
2678:
2679: if ( !p ) return;
2680: for ( m = BDY(p); m; m = NEXT(m) ) {
2681: mulp(vl,CP(m),mul,&c); CP(m) = c;
2682: }
2683: }
2684:
1.1 noro 2685: void nd_free(ND p)
2686: {
2687: NM t,s;
2688:
1.34 noro 2689: if ( !p ) return;
1.1 noro 2690: t = BDY(p);
2691: while ( t ) {
2692: s = NEXT(t);
2693: FREENM(t);
2694: t = s;
2695: }
2696: FREEND(p);
2697: }
2698:
1.23 noro 2699: void ndv_free(NDV p)
2700: {
2701: GC_free(BDY(p));
2702: }
2703:
1.61 noro 2704: void nd_append_red(UINT *d,int i)
1.1 noro 2705: {
1.13 noro 2706: RHist m,m0;
1.1 noro 2707: int h;
2708:
1.13 noro 2709: NEWRHist(m);
1.34 noro 2710: h = ndl_hash_value(d);
1.13 noro 2711: m->index = i;
1.14 noro 2712: ndl_copy(d,DL(m));
1.1 noro 2713: NEXT(m) = nd_red[h];
2714: nd_red[h] = m;
2715: }
2716:
1.61 noro 2717: UINT *ndv_compute_bound(NDV p)
1.1 noro 2718: {
1.61 noro 2719: UINT *d1,*d2,*t;
2720: UINT u;
1.57 noro 2721: int i,j,k,l,len,ind;
1.45 noro 2722: NMV m;
1.1 noro 2723:
2724: if ( !p )
2725: return 0;
1.61 noro 2726: d1 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
2727: d2 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
1.45 noro 2728: len = LEN(p);
2729: m = BDY(p); ndl_copy(DL(m),d1); NMV_ADV(m);
2730: for ( i = 1; i < len; i++, NMV_ADV(m) ) {
1.14 noro 2731: ndl_lcm(DL(m),d1,d2);
1.1 noro 2732: t = d1; d1 = d2; d2 = t;
2733: }
1.12 noro 2734: l = nd_nvar+31;
1.61 noro 2735: t = (UINT *)MALLOC_ATOMIC(l*sizeof(UINT));
1.57 noro 2736: for ( i = nd_exporigin, ind = 0; i < nd_wpd; i++ ) {
2737: u = d1[i];
2738: k = (nd_epw-1)*nd_bpe;
2739: for ( j = 0; j < nd_epw; j++, k -= nd_bpe, ind++ )
2740: t[ind] = (u>>k)&nd_mask0;
2741: }
2742: for ( ; ind < l; ind++ ) t[ind] = 0;
1.1 noro 2743: return t;
2744: }
2745:
1.99 noro 2746: UINT *nd_compute_bound(ND p)
2747: {
2748: UINT *d1,*d2,*t;
2749: UINT u;
2750: int i,j,k,l,len,ind;
2751: NM m;
2752:
2753: if ( !p )
2754: return 0;
2755: d1 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
2756: d2 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
2757: len = LEN(p);
2758: m = BDY(p); ndl_copy(DL(m),d1); m = NEXT(m);
2759: for ( m = NEXT(m); m; m = NEXT(m) ) {
2760: ndl_lcm(DL(m),d1,d2);
2761: t = d1; d1 = d2; d2 = t;
2762: }
2763: l = nd_nvar+31;
2764: t = (UINT *)MALLOC_ATOMIC(l*sizeof(UINT));
2765: for ( i = nd_exporigin, ind = 0; i < nd_wpd; i++ ) {
2766: u = d1[i];
2767: k = (nd_epw-1)*nd_bpe;
2768: for ( j = 0; j < nd_epw; j++, k -= nd_bpe, ind++ )
2769: t[ind] = (u>>k)&nd_mask0;
2770: }
2771: for ( ; ind < l; ind++ ) t[ind] = 0;
2772: return t;
2773: }
2774:
1.48 noro 2775: int nd_get_exporigin(struct order_spec *ord)
2776: {
1.51 noro 2777: switch ( ord->id ) {
1.96 noro 2778: case 0: case 2:
1.48 noro 2779: return 1;
1.41 noro 2780: case 1:
2781: /* block order */
1.43 noro 2782: /* d[0]:weight d[1]:w0,...,d[nd_exporigin-1]:w(n-1) */
1.48 noro 2783: return ord->ord.block.length+1;
1.96 noro 2784: case 3:
2785: error("nd_get_exporigin : composite order is not supported yet.");
1.41 noro 2786: }
1.48 noro 2787: }
2788:
1.61 noro 2789: void nd_setup_parameters(int nvar,int max) {
1.79 noro 2790: int i,j,n,elen,ord_o,ord_l,l,s,wpd;
1.57 noro 2791: struct order_pair *op;
1.48 noro 2792:
1.73 noro 2793: nd_nvar = nvar;
1.79 noro 2794: if ( max ) {
1.83 noro 2795: /* XXX */
2796: if ( do_weyl ) nd_bpe = 32;
2797: else if ( max < 2 ) nd_bpe = 1;
1.79 noro 2798: else if ( max < 4 ) nd_bpe = 2;
2799: else if ( max < 8 ) nd_bpe = 3;
2800: else if ( max < 16 ) nd_bpe = 4;
2801: else if ( max < 32 ) nd_bpe = 5;
2802: else if ( max < 64 ) nd_bpe = 6;
2803: else if ( max < 256 ) nd_bpe = 8;
2804: else if ( max < 1024 ) nd_bpe = 10;
2805: else if ( max < 65536 ) nd_bpe = 16;
2806: else nd_bpe = 32;
2807: }
1.61 noro 2808: nd_epw = (sizeof(UINT)*8)/nd_bpe;
1.48 noro 2809: elen = nd_nvar/nd_epw+(nd_nvar%nd_epw?1:0);
2810: nd_exporigin = nd_get_exporigin(nd_ord);
1.79 noro 2811: wpd = nd_exporigin+elen;
2812: if ( wpd != nd_wpd ) {
2813: nd_free_private_storage();
2814: nd_wpd = wpd;
2815: }
1.1 noro 2816: if ( nd_bpe < 32 ) {
2817: nd_mask0 = (1<<nd_bpe)-1;
2818: } else {
2819: nd_mask0 = 0xffffffff;
2820: }
2821: bzero(nd_mask,sizeof(nd_mask));
2822: nd_mask1 = 0;
2823: for ( i = 0; i < nd_epw; i++ ) {
2824: nd_mask[nd_epw-i-1] = (nd_mask0<<(i*nd_bpe));
2825: nd_mask1 |= (1<<(nd_bpe-1))<<(i*nd_bpe);
2826: }
1.90 noro 2827: nmv_adv = ROUND_FOR_ALIGN(sizeof(struct oNMV)+(nd_wpd-1)*sizeof(UINT));
1.57 noro 2828: nd_epos = nd_create_epos(nd_ord);
1.43 noro 2829: nd_blockmask = nd_create_blockmask(nd_ord);
1.96 noro 2830: nd_work_vector = (int *)REALLOC(nd_work_vector,nd_nvar*sizeof(int));
1.1 noro 2831: }
2832:
1.103 noro 2833: ND_pairs nd_reconstruct(int trace,ND_pairs d)
1.1 noro 2834: {
1.37 noro 2835: int i,obpe,oadv,h;
1.71 noro 2836: static NM prev_nm_free_list;
2837: static ND_pairs prev_ndp_free_list;
1.13 noro 2838: RHist mr0,mr;
2839: RHist r;
1.37 noro 2840: RHist *old_red;
1.71 noro 2841: ND_pairs s0,s,t;
1.43 noro 2842: EPOS oepos;
1.15 noro 2843:
1.1 noro 2844: obpe = nd_bpe;
1.11 noro 2845: oadv = nmv_adv;
1.43 noro 2846: oepos = nd_epos;
1.63 noro 2847: if ( obpe < 2 ) nd_bpe = 2;
2848: else if ( obpe < 3 ) nd_bpe = 3;
1.62 noro 2849: else if ( obpe < 4 ) nd_bpe = 4;
1.63 noro 2850: else if ( obpe < 5 ) nd_bpe = 5;
1.34 noro 2851: else if ( obpe < 6 ) nd_bpe = 6;
2852: else if ( obpe < 8 ) nd_bpe = 8;
1.63 noro 2853: else if ( obpe < 10 ) nd_bpe = 10;
1.34 noro 2854: else if ( obpe < 16 ) nd_bpe = 16;
2855: else if ( obpe < 32 ) nd_bpe = 32;
2856: else error("nd_reconstruct : exponent too large");
1.5 noro 2857:
1.73 noro 2858: nd_setup_parameters(nd_nvar,0);
1.1 noro 2859: prev_nm_free_list = _nm_free_list;
2860: prev_ndp_free_list = _ndp_free_list;
2861: _nm_free_list = 0;
2862: _ndp_free_list = 0;
1.53 noro 2863: for ( i = nd_psn-1; i >= 0; i-- ) ndv_realloc(nd_ps[i],obpe,oadv,oepos);
2864: if ( trace )
2865: for ( i = nd_psn-1; i >= 0; i-- )
2866: ndv_realloc(nd_ps_trace[i],obpe,oadv,oepos);
1.1 noro 2867: s0 = 0;
2868: for ( t = d; t; t = NEXT(t) ) {
2869: NEXTND_pairs(s0,s);
2870: s->i1 = t->i1;
2871: s->i2 = t->i2;
1.14 noro 2872: SG(s) = SG(t);
1.61 noro 2873: ndl_reconstruct(LCM(t),LCM(s),obpe,oepos);
1.1 noro 2874: }
1.37 noro 2875:
2876: old_red = (RHist *)ALLOCA(REDTAB_LEN*sizeof(RHist));
1.6 noro 2877: for ( i = 0; i < REDTAB_LEN; i++ ) {
1.37 noro 2878: old_red[i] = nd_red[i];
2879: nd_red[i] = 0;
2880: }
2881: for ( i = 0; i < REDTAB_LEN; i++ )
2882: for ( r = old_red[i]; r; r = NEXT(r) ) {
2883: NEWRHist(mr);
1.13 noro 2884: mr->index = r->index;
1.20 noro 2885: SG(mr) = SG(r);
1.61 noro 2886: ndl_reconstruct(DL(r),DL(mr),obpe,oepos);
1.37 noro 2887: h = ndl_hash_value(DL(mr));
2888: NEXT(mr) = nd_red[h];
2889: nd_red[h] = mr;
1.6 noro 2890: }
1.37 noro 2891: for ( i = 0; i < REDTAB_LEN; i++ ) old_red[i] = 0;
2892: old_red = 0;
1.11 noro 2893: for ( i = 0; i < nd_psn; i++ ) {
1.20 noro 2894: NEWRHist(r); SG(r) = SG(nd_psh[i]);
1.61 noro 2895: ndl_reconstruct(DL(nd_psh[i]),DL(r),obpe,oepos);
1.13 noro 2896: nd_psh[i] = r;
1.11 noro 2897: }
1.1 noro 2898: if ( s0 ) NEXT(s) = 0;
2899: prev_nm_free_list = 0;
2900: prev_ndp_free_list = 0;
1.71 noro 2901: #if 0
1.1 noro 2902: GC_gcollect();
1.71 noro 2903: #endif
1.1 noro 2904: return s0;
2905: }
2906:
1.61 noro 2907: void ndl_reconstruct(UINT *d,UINT *r,int obpe,EPOS oepos)
1.1 noro 2908: {
1.57 noro 2909: int n,i,ei,oepw,omask0,j,s,ord_l,l;
1.43 noro 2910: struct order_pair *op;
1.1 noro 2911:
2912: n = nd_nvar;
1.61 noro 2913: oepw = (sizeof(UINT)*8)/obpe;
1.43 noro 2914: omask0 = (1<<obpe)-1;
1.34 noro 2915: TD(r) = TD(d);
1.41 noro 2916: for ( i = nd_exporigin; i < nd_wpd; i++ ) r[i] = 0;
1.43 noro 2917: if ( nd_blockmask ) {
2918: l = nd_blockmask->n;
2919: op = nd_blockmask->order_pair;
2920: for ( i = 1; i < nd_exporigin; i++ )
2921: r[i] = d[i];
2922: for ( j = 0, s = 0; j < l; j++ ) {
2923: ord_l = op[j].length;
1.57 noro 2924: for ( i = 0; i < ord_l; i++, s++ ) {
2925: ei = GET_EXP_OLD(d,s);
2926: PUT_EXP(r,s,ei);
2927: }
1.1 noro 2928: }
1.43 noro 2929: } else {
1.56 noro 2930: for ( i = 0; i < n; i++ ) {
2931: ei = GET_EXP_OLD(d,i);
2932: PUT_EXP(r,i,ei);
2933: }
1.1 noro 2934: }
2935: }
1.3 noro 2936:
1.6 noro 2937: ND nd_copy(ND p)
2938: {
2939: NM m,mr,mr0;
1.41 noro 2940: int c,n;
1.6 noro 2941: ND r;
2942:
2943: if ( !p )
2944: return 0;
2945: else {
2946: for ( mr0 = 0, m = BDY(p); m; m = NEXT(m) ) {
2947: NEXTNM(mr0,mr);
1.14 noro 2948: CM(mr) = CM(m);
2949: ndl_copy(DL(m),DL(mr));
1.6 noro 2950: }
2951: NEXT(mr) = 0;
1.31 noro 2952: MKND(NV(p),mr0,LEN(p),r);
1.14 noro 2953: SG(r) = SG(p);
1.6 noro 2954: return r;
2955: }
2956: }
2957:
1.53 noro 2958: int nd_sp(int mod,int trace,ND_pairs p,ND *rp)
1.11 noro 2959: {
2960: NM m;
2961: NDV p1,p2;
2962: ND t1,t2;
1.61 noro 2963: UINT *lcm;
1.31 noro 2964: int td;
1.11 noro 2965:
1.77 noro 2966: if ( !mod && nd_demand ) {
2967: p1 = ndv_load(p->i1); p2 = ndv_load(p->i2);
1.53 noro 2968: } else {
1.77 noro 2969: if ( trace ) {
2970: p1 = nd_ps_trace[p->i1]; p2 = nd_ps_trace[p->i2];
2971: } else {
2972: p1 = nd_ps[p->i1]; p2 = nd_ps[p->i2];
2973: }
1.20 noro 2974: }
1.34 noro 2975: lcm = LCM(p);
1.11 noro 2976: NEWNM(m);
1.113 noro 2977: CQ(m) = HCQ(p2);
1.34 noro 2978: ndl_sub(lcm,HDL(p1),DL(m));
1.56 noro 2979: if ( ndl_check_bound2(p->i1,DL(m)) )
2980: return 0;
1.55 noro 2981: t1 = ndv_mul_nm(mod,m,p1);
1.71 noro 2982: if ( mod == -1 ) CM(m) = _chsgnsf(HCM(p1));
2983: else if ( mod ) CM(m) = mod-HCM(p1);
1.113 noro 2984: else chsgnq(HCQ(p1),&CQ(m));
1.34 noro 2985: ndl_sub(lcm,HDL(p2),DL(m));
1.14 noro 2986: if ( ndl_check_bound2(p->i2,DL(m)) ) {
1.11 noro 2987: nd_free(t1);
2988: return 0;
2989: }
1.55 noro 2990: t2 = ndv_mul_nm(mod,m,p2);
1.31 noro 2991: *rp = nd_add(mod,t1,t2);
1.11 noro 2992: FREENM(m);
2993: return 1;
2994: }
2995:
1.19 noro 2996: void ndv_mul_c(int mod,NDV p,int mul)
1.11 noro 2997: {
2998: NMV m;
2999: int c,c1,len,i;
3000:
1.34 noro 3001: if ( !p ) return;
1.14 noro 3002: len = LEN(p);
1.71 noro 3003: if ( mod == -1 )
3004: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) )
3005: CM(m) = _mulsf(CM(m),mul);
3006: else
3007: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
3008: c1 = CM(m); DMAR(c1,mul,0,mod,c); CM(m) = c;
3009: }
1.11 noro 3010: }
3011:
1.113 noro 3012: void ndv_mul_c_q(NDV p,Q mul)
1.16 noro 3013: {
3014: NMV m;
1.113 noro 3015: Q c;
1.16 noro 3016: int len,i;
3017:
1.34 noro 3018: if ( !p ) return;
1.16 noro 3019: len = LEN(p);
3020: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
1.113 noro 3021: mulq(CQ(m),mul,&c); CQ(m) = c;
1.16 noro 3022: }
3023: }
3024:
1.55 noro 3025: ND weyl_ndv_mul_nm(int mod,NM m0,NDV p) {
3026: int n2,i,j,l,n,tlen;
1.61 noro 3027: UINT *d0;
1.55 noro 3028: NM *tab,*psum;
3029: ND s,r;
3030: NM t;
3031: NMV m1;
3032:
3033: if ( !p ) return 0;
3034: n = NV(p); n2 = n>>1;
3035: d0 = DL(m0);
3036: l = LEN(p);
3037: for ( i = 0, tlen = 1; i < n2; i++ ) tlen *= (GET_EXP(d0,n2+i)+1);
3038: tab = (NM *)ALLOCA(tlen*sizeof(NM));
3039: psum = (NM *)ALLOCA(tlen*sizeof(NM));
3040: for ( i = 0; i < tlen; i++ ) psum[i] = 0;
1.56 noro 3041: m1 = (NMV)(((char *)BDY(p))+nmv_adv*(l-1));
3042: for ( i = l-1; i >= 0; i--, NMV_PREV(m1) ) {
1.55 noro 3043: /* m0(NM) * m1(NMV) => tab(NM) */
1.56 noro 3044: weyl_mul_nm_nmv(n,mod,m0,m1,tab,tlen);
1.55 noro 3045: for ( j = 0; j < tlen; j++ ) {
3046: if ( tab[j] ) {
3047: NEXT(tab[j]) = psum[j]; psum[j] = tab[j];
3048: }
3049: }
3050: }
3051: for ( i = tlen-1, r = 0; i >= 0; i-- )
3052: if ( psum[i] ) {
3053: for ( j = 0, t = psum[i]; t; t = NEXT(t), j++ );
3054: MKND(n,psum[i],j,s);
3055: r = nd_add(mod,r,s);
3056: }
1.56 noro 3057: if ( r ) SG(r) = SG(p)+TD(d0);
3058: return r;
1.55 noro 3059: }
3060:
1.56 noro 3061: /* product of monomials */
3062: /* XXX block order is not handled correctly */
3063:
1.55 noro 3064: void weyl_mul_nm_nmv(int n,int mod,NM m0,NMV m1,NM *tab,int tlen)
3065: {
1.56 noro 3066: int i,n2,j,s,curlen,homo,h,a,b,k,l,u,min;
1.61 noro 3067: UINT *d0,*d1,*d,*dt,*ctab;
1.113 noro 3068: Q *ctab_q;
3069: Q q,q1;
1.61 noro 3070: UINT c0,c1,c;
1.55 noro 3071: NM *p;
3072: NM m,t;
3073:
3074: for ( i = 0; i < tlen; i++ ) tab[i] = 0;
3075: if ( !m0 || !m1 ) return;
3076: d0 = DL(m0); d1 = DL(m1); n2 = n>>1;
3077: NEWNM(m); d = DL(m);
1.56 noro 3078: if ( mod ) {
3079: c0 = CM(m0); c1 = CM(m1); DMAR(c0,c1,0,mod,c); CM(m) = c;
3080: } else
1.113 noro 3081: mulq(CQ(m0),CQ(m1),&CQ(m));
1.55 noro 3082: for ( i = 0; i < nd_wpd; i++ ) d[i] = 0;
3083: homo = n&1 ? 1 : 0;
3084: if ( homo ) {
3085: /* offset of h-degree */
3086: h = GET_EXP(d0,n-1)+GET_EXP(d1,n-1);
3087: PUT_EXP(DL(m),n-1,h);
3088: TD(DL(m)) = h;
1.61 noro 3089: if ( nd_blockmask ) ndl_weight_mask(DL(m));
1.55 noro 3090: }
3091: tab[0] = m;
3092: NEWNM(m); d = DL(m);
1.57 noro 3093: for ( i = 0, curlen = 1; i < n2; i++ ) {
1.55 noro 3094: a = GET_EXP(d0,i); b = GET_EXP(d1,n2+i);
3095: k = GET_EXP(d0,n2+i); l = GET_EXP(d1,i);
3096: /* xi^a*(Di^k*xi^l)*Di^b */
3097: a += l; b += k;
1.56 noro 3098: s = MUL_WEIGHT(a,i)+MUL_WEIGHT(b,n2+i);
1.55 noro 3099: if ( !k || !l ) {
3100: for ( j = 0; j < curlen; j++ )
1.56 noro 3101: if ( t = tab[j] ) {
3102: dt = DL(t);
3103: PUT_EXP(dt,i,a); PUT_EXP(dt,n2+i,b); TD(dt) += s;
1.61 noro 3104: if ( nd_blockmask ) ndl_weight_mask(dt);
1.55 noro 3105: }
3106: curlen *= k+1;
3107: continue;
3108: }
3109: min = MIN(k,l);
1.56 noro 3110: if ( mod ) {
1.61 noro 3111: ctab = (UINT *)ALLOCA((min+1)*sizeof(UINT));
1.56 noro 3112: mkwcm(k,l,mod,ctab);
3113: } else {
1.113 noro 3114: ctab_q = (Q *)ALLOCA((min+1)*sizeof(Q));
3115: mkwc(k,l,ctab_q);
1.56 noro 3116: }
1.57 noro 3117: for ( j = min; j >= 0; j-- ) {
1.56 noro 3118: for ( u = 0; u < nd_wpd; u++ ) d[u] = 0;
1.55 noro 3119: PUT_EXP(d,i,a-j); PUT_EXP(d,n2+i,b-j);
1.56 noro 3120: h = MUL_WEIGHT(a-j,i)+MUL_WEIGHT(b-j,n2+i);
1.55 noro 3121: if ( homo ) {
3122: TD(d) = s;
1.56 noro 3123: PUT_EXP(d,n-1,s-h);
1.55 noro 3124: } else TD(d) = h;
1.61 noro 3125: if ( nd_blockmask ) ndl_weight_mask(d);
1.56 noro 3126: if ( mod ) c = ctab[j];
1.113 noro 3127: else q = ctab_q[j];
1.57 noro 3128: p = tab+curlen*j;
3129: if ( j == 0 ) {
3130: for ( u = 0; u < curlen; u++, p++ ) {
3131: if ( tab[u] ) {
3132: ndl_addto(DL(tab[u]),d);
3133: if ( mod ) {
3134: c0 = CM(tab[u]); DMAR(c0,c,0,mod,c1); CM(tab[u]) = c1;
3135: } else {
1.113 noro 3136: mulq(CQ(tab[u]),q,&q1); CQ(tab[u]) = q1;
1.57 noro 3137: }
3138: }
1.56 noro 3139: }
1.57 noro 3140: } else {
3141: for ( u = 0; u < curlen; u++, p++ ) {
3142: if ( tab[u] ) {
3143: NEWNM(t);
3144: ndl_add(DL(tab[u]),d,DL(t));
3145: if ( mod ) {
3146: c0 = CM(tab[u]); DMAR(c0,c,0,mod,c1); CM(t) = c1;
3147: } else
1.113 noro 3148: mulq(CQ(tab[u]),q,&CQ(t));
1.57 noro 3149: *p = t;
3150: }
1.55 noro 3151: }
3152: }
3153: }
3154: curlen *= k+1;
3155: }
3156: FREENM(m);
3157: }
3158:
1.63 noro 3159: ND ndv_mul_nm_symbolic(NM m0,NDV p)
3160: {
3161: NM mr,mr0;
3162: NMV m;
3163: UINT *d,*dt,*dm;
3164: int c,n,td,i,c1,c2,len;
1.113 noro 3165: Q q;
1.63 noro 3166: ND r;
3167:
3168: if ( !p ) return 0;
3169: else {
3170: n = NV(p); m = BDY(p);
3171: d = DL(m0);
3172: len = LEN(p);
3173: mr0 = 0;
3174: td = TD(d);
3175: c = CM(m0);
3176: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
3177: NEXTNM(mr0,mr);
3178: CM(mr) = 1;
3179: ndl_add(DL(m),d,DL(mr));
3180: }
3181: NEXT(mr) = 0;
3182: MKND(NV(p),mr0,len,r);
3183: SG(r) = SG(p) + TD(d);
3184: return r;
3185: }
3186: }
3187:
1.55 noro 3188: ND ndv_mul_nm(int mod,NM m0,NDV p)
1.9 noro 3189: {
3190: NM mr,mr0;
3191: NMV m;
1.61 noro 3192: UINT *d,*dt,*dm;
1.9 noro 3193: int c,n,td,i,c1,c2,len;
1.113 noro 3194: Q q;
1.9 noro 3195: ND r;
3196:
1.34 noro 3197: if ( !p ) return 0;
1.55 noro 3198: else if ( do_weyl )
1.71 noro 3199: if ( mod == -1 )
3200: error("ndv_mul_nm : not implemented (weyl)");
3201: else
3202: return weyl_ndv_mul_nm(mod,m0,p);
1.9 noro 3203: else {
3204: n = NV(p); m = BDY(p);
1.34 noro 3205: d = DL(m0);
1.14 noro 3206: len = LEN(p);
1.9 noro 3207: mr0 = 0;
1.34 noro 3208: td = TD(d);
1.71 noro 3209: if ( mod == -1 ) {
3210: c = CM(m0);
3211: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
3212: NEXTNM(mr0,mr);
3213: CM(mr) = _mulsf(CM(m),c);
3214: ndl_add(DL(m),d,DL(mr));
3215: }
3216: } else if ( mod ) {
1.16 noro 3217: c = CM(m0);
3218: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
3219: NEXTNM(mr0,mr);
3220: c1 = CM(m);
1.19 noro 3221: DMAR(c1,c,0,mod,c2);
1.16 noro 3222: CM(mr) = c2;
3223: ndl_add(DL(m),d,DL(mr));
3224: }
3225: } else {
1.113 noro 3226: q = CQ(m0);
1.16 noro 3227: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
3228: NEXTNM(mr0,mr);
1.113 noro 3229: mulq(CQ(m),q,&CQ(mr));
1.16 noro 3230: ndl_add(DL(m),d,DL(mr));
3231: }
1.4 noro 3232: }
1.9 noro 3233: NEXT(mr) = 0;
1.31 noro 3234: MKND(NV(p),mr0,len,r);
1.34 noro 3235: SG(r) = SG(p) + TD(d);
1.9 noro 3236: return r;
1.4 noro 3237: }
3238: }
3239:
1.104 noro 3240: ND nd_quo(int mod,PGeoBucket bucket,NDV d)
1.99 noro 3241: {
3242: NM mq0,mq;
1.102 noro 3243: NMV tm;
1.113 noro 3244: Q q;
1.104 noro 3245: int i,nv,sg,c,c1,c2,hindex;
3246: ND p,t,r;
3247: N tnm;
3248:
1.99 noro 3249: if ( !p ) return 0;
3250: else {
1.104 noro 3251: nv = NV(d);
1.99 noro 3252: mq0 = 0;
1.102 noro 3253: tm = (NMV)ALLOCA(nmv_adv);
1.104 noro 3254: while ( 1 ) {
1.113 noro 3255: hindex = mod?head_pbucket(mod,bucket):head_pbucket_q(bucket);
1.104 noro 3256: if ( hindex < 0 ) break;
3257: p = bucket->body[hindex];
1.99 noro 3258: NEXTNM(mq0,mq);
3259: ndl_sub(HDL(p),HDL(d),DL(mq));
1.102 noro 3260: ndl_copy(DL(mq),DL(tm));
3261: if ( mod ) {
3262: c1 = invm(HCM(d),mod); c2 = HCM(p);
3263: DMAR(c1,c2,0,mod,c); CM(mq) = c;
3264: CM(tm) = mod-c;
3265: } else {
1.113 noro 3266: divsn(NM(HCQ(p)),NM(HCQ(d)),&tnm);
3267: NTOQ(tnm,SGN(HCQ(p))*SGN(HCQ(d)),CQ(mq));
3268: chsgnq(CQ(mq),&CQ(tm));
1.102 noro 3269: }
3270: t = ndv_mul_nmv_trunc(mod,tm,d,HDL(d));
1.104 noro 3271: bucket->body[hindex] = nd_remove_head(p);
3272: t = nd_remove_head(t);
3273: add_pbucket(mod,bucket,t);
3274: }
3275: if ( !mq0 )
3276: r = 0;
3277: else {
3278: NEXT(mq) = 0;
3279: for ( i = 0, mq = mq0; mq; mq = NEXT(mq), i++ );
3280: MKND(nv,mq0,i,r);
3281: /* XXX */
3282: SG(r) = HTD(r);
1.99 noro 3283: }
3284: return r;
3285: }
3286: }
3287:
1.43 noro 3288: void ndv_realloc(NDV p,int obpe,int oadv,EPOS oepos)
1.11 noro 3289: {
1.13 noro 3290: NMV m,mr,mr0,t;
3291: int len,i,k;
1.11 noro 3292:
1.61 noro 3293: if ( !p ) return;
3294: m = BDY(p); len = LEN(p);
3295: mr0 = nmv_adv>oadv?(NMV)REALLOC(BDY(p),len*nmv_adv):BDY(p);
3296: m = (NMV)((char *)mr0+(len-1)*oadv);
3297: mr = (NMV)((char *)mr0+(len-1)*nmv_adv);
3298: t = (NMV)ALLOCA(nmv_adv);
3299: for ( i = 0; i < len; i++, NMV_OPREV(m), NMV_PREV(mr) ) {
1.113 noro 3300: CQ(t) = CQ(m);
1.61 noro 3301: for ( k = 0; k < nd_wpd; k++ ) DL(t)[k] = 0;
3302: ndl_reconstruct(DL(m),DL(t),obpe,oepos);
1.113 noro 3303: CQ(mr) = CQ(t);
1.61 noro 3304: ndl_copy(DL(t),DL(mr));
3305: }
3306: BDY(p) = mr0;
3307: }
3308:
3309: NDV ndv_dup_realloc(NDV p,int obpe,int oadv,EPOS oepos)
3310: {
3311: NMV m,mr,mr0;
3312: int len,i;
3313: NDV r;
1.11 noro 3314:
1.61 noro 3315: if ( !p ) return 0;
3316: m = BDY(p); len = LEN(p);
3317: mr0 = mr = (NMV)MALLOC(len*nmv_adv);
3318: for ( i = 0; i < len; i++, NMV_OADV(m), NMV_ADV(mr) ) {
3319: ndl_zero(DL(mr));
3320: ndl_reconstruct(DL(m),DL(mr),obpe,oepos);
1.113 noro 3321: CQ(mr) = CQ(m);
1.11 noro 3322: }
1.61 noro 3323: MKNDV(NV(p),mr0,len,r);
3324: SG(r) = SG(p);
3325: return r;
1.11 noro 3326: }
3327:
1.61 noro 3328: /* duplicate p */
3329:
3330: NDV ndv_dup(int mod,NDV p)
1.3 noro 3331: {
3332: NDV d;
1.61 noro 3333: NMV t,m,m0;
1.3 noro 3334: int i,len;
3335:
1.34 noro 3336: if ( !p ) return 0;
1.31 noro 3337: len = LEN(p);
1.34 noro 3338: m0 = m = (NMV)(mod?MALLOC_ATOMIC(len*nmv_adv):MALLOC(len*nmv_adv));
1.61 noro 3339: for ( t = BDY(p), i = 0; i < len; i++, NMV_ADV(t), NMV_ADV(m) ) {
1.14 noro 3340: ndl_copy(DL(t),DL(m));
1.113 noro 3341: CQ(m) = CQ(t);
1.3 noro 3342: }
3343: MKNDV(NV(p),m0,len,d);
1.23 noro 3344: SG(d) = SG(p);
3345: return d;
3346: }
3347:
1.63 noro 3348: ND nd_dup(ND p)
3349: {
3350: ND d;
3351: NM t,m,m0;
3352:
3353: if ( !p ) return 0;
3354: for ( m0 = 0, t = BDY(p); t; t = NEXT(t) ) {
3355: NEXTNM(m0,m);
3356: ndl_copy(DL(t),DL(m));
1.113 noro 3357: CQ(m) = CQ(t);
1.63 noro 3358: }
3359: if ( m0 ) NEXT(m) = 0;
3360: MKND(NV(p),m0,LEN(p),d);
3361: SG(d) = SG(p);
3362: return d;
3363: }
3364:
1.61 noro 3365: /* XXX if p->len == 0 then it represents 0 */
3366:
3367: void ndv_mod(int mod,NDV p)
3368: {
3369: NMV t,d;
3370: int r;
3371: int i,len,dlen;
1.71 noro 3372: Obj gfs;
1.61 noro 3373:
3374: if ( !p ) return;
3375: len = LEN(p);
3376: dlen = 0;
1.71 noro 3377: if ( mod == -1 )
3378: for ( t = d = BDY(p), i = 0; i < len; i++, NMV_ADV(t) ) {
3379: simp_ff((Obj)CP(t),&gfs);
3380: r = FTOIF(CONT((GFS)gfs));
1.61 noro 3381: CM(d) = r;
3382: ndl_copy(DL(t),DL(d));
3383: NMV_ADV(d);
3384: dlen++;
3385: }
1.71 noro 3386: else
3387: for ( t = d = BDY(p), i = 0; i < len; i++, NMV_ADV(t) ) {
1.113 noro 3388: r = rem(NM(CQ(t)),mod);
1.71 noro 3389: if ( r ) {
1.113 noro 3390: if ( SGN(CQ(t)) < 0 )
3391: r = mod-r;
1.71 noro 3392: CM(d) = r;
3393: ndl_copy(DL(t),DL(d));
3394: NMV_ADV(d);
3395: dlen++;
3396: }
3397: }
1.61 noro 3398: LEN(p) = dlen;
3399: }
3400:
3401: NDV ptondv(VL vl,VL dvl,P p)
3402: {
3403: ND nd;
3404:
3405: nd = ptond(vl,dvl,p);
3406: return ndtondv(0,nd);
3407: }
3408:
3409: ND ptond(VL vl,VL dvl,P p)
1.23 noro 3410: {
1.61 noro 3411: int n,i,j,k,e;
3412: VL tvl;
3413: V v;
3414: DCP dc;
3415: DCP *w;
3416: ND r,s,t,u;
3417: P x;
3418: int c;
3419: UINT *d;
1.23 noro 3420: NM m,m0;
1.61 noro 3421:
3422: if ( !p )
3423: return 0;
3424: else if ( NUM(p) ) {
3425: NEWNM(m);
3426: ndl_zero(DL(m));
1.113 noro 3427: CQ(m) = (Q)p;
1.61 noro 3428: NEXT(m) = 0;
3429: MKND(nd_nvar,m,1,r);
3430: SG(r) = 0;
3431: return r;
3432: } else {
3433: for ( dc = DC(p), k = 0; dc; dc = NEXT(dc), k++ );
3434: w = (DCP *)ALLOCA(k*sizeof(DCP));
3435: for ( dc = DC(p), j = 0; j < k; dc = NEXT(dc), j++ ) w[j] = dc;
3436: for ( i = 0, tvl = dvl, v = VR(p);
1.98 noro 3437: tvl && tvl->v != v; tvl = NEXT(tvl), i++ );
1.61 noro 3438: if ( !tvl ) {
3439: for ( j = k-1, s = 0, MKV(v,x); j >= 0; j-- ) {
3440: t = ptond(vl,dvl,COEF(w[j]));
3441: pwrp(vl,x,DEG(w[j]),&p);
3442: nd_mul_c_p(CO,t,p); s = nd_add(0,s,t);
3443: }
3444: return s;
3445: } else {
3446: NEWNM(m0); d = DL(m0);
3447: for ( j = k-1, s = 0; j >= 0; j-- ) {
3448: ndl_zero(d); e = QTOS(DEG(w[j])); PUT_EXP(d,i,e);
3449: TD(d) = MUL_WEIGHT(e,i);
3450: if ( nd_blockmask) ndl_weight_mask(d);
3451: t = ptond(vl,dvl,COEF(w[j]));
3452: for ( m = BDY(t); m; m = NEXT(m) )
3453: ndl_addto(DL(m),d);
3454: SG(t) += TD(d);
3455: s = nd_add(0,s,t);
3456: }
3457: FREENM(m0);
3458: return s;
3459: }
3460: }
3461: }
3462:
3463: P ndvtop(int mod,VL vl,VL dvl,NDV p)
3464: {
3465: VL tvl;
3466: int len,n,j,i,e;
3467: NMV m;
3468: Q q;
3469: P c;
3470: UINT *d;
3471: P s,r,u,t,w;
1.71 noro 3472: GFS gfs;
1.23 noro 3473:
1.34 noro 3474: if ( !p ) return 0;
1.61 noro 3475: else {
3476: len = LEN(p);
3477: n = NV(p);
3478: m = (NMV)(((char *)BDY(p))+nmv_adv*(len-1));
3479: for ( j = len-1, s = 0; j >= 0; j--, NMV_PREV(m) ) {
1.71 noro 3480: if ( mod == -1 ) {
3481: e = IFTOF(CM(m)); MKGFS(e,gfs); c = (P)gfs;
3482: } else if ( mod ) {
1.61 noro 3483: STOQ(CM(m),q); c = (P)q;
3484: } else
1.113 noro 3485: c = CP(m);
1.61 noro 3486: d = DL(m);
3487: for ( i = 0, t = c, tvl = dvl; i < n; tvl = NEXT(tvl), i++ ) {
3488: MKV(tvl->v,r); e = GET_EXP(d,i); STOQ(e,q);
3489: pwrp(vl,r,q,&u); mulp(vl,t,u,&w); t = w;
3490: }
3491: addp(vl,s,t,&u); s = u;
3492: }
3493: return s;
1.23 noro 3494: }
1.3 noro 3495: }
3496:
1.61 noro 3497: NDV ndtondv(int mod,ND p)
1.11 noro 3498: {
3499: NDV d;
1.61 noro 3500: NMV m,m0;
3501: NM t;
3502: int i,len;
1.11 noro 3503:
1.34 noro 3504: if ( !p ) return 0;
1.61 noro 3505: len = LEN(p);
1.74 noro 3506: if ( mod )
3507: m0 = m = (NMV)GC_malloc_atomic_ignore_off_page(len*nmv_adv);
3508: else
3509: m0 = m = MALLOC(len*nmv_adv);
1.103 noro 3510: #if 0
1.74 noro 3511: ndv_alloc += nmv_adv*len;
1.103 noro 3512: #endif
1.61 noro 3513: for ( t = BDY(p), i = 0; t; t = NEXT(t), i++, NMV_ADV(m) ) {
3514: ndl_copy(DL(t),DL(m));
1.113 noro 3515: CQ(m) = CQ(t);
1.11 noro 3516: }
1.61 noro 3517: MKNDV(NV(p),m0,len,d);
1.14 noro 3518: SG(d) = SG(p);
1.11 noro 3519: return d;
3520: }
3521:
1.61 noro 3522: ND ndvtond(int mod,NDV p)
1.11 noro 3523: {
1.61 noro 3524: ND d;
3525: NM m,m0;
1.11 noro 3526: NMV t;
1.61 noro 3527: int i,len;
1.11 noro 3528:
1.34 noro 3529: if ( !p ) return 0;
1.11 noro 3530: m0 = 0;
1.61 noro 3531: len = p->len;
3532: for ( t = BDY(p), i = 0; i < len; NMV_ADV(t), i++ ) {
3533: NEXTNM(m0,m);
3534: ndl_copy(DL(t),DL(m));
1.113 noro 3535: CQ(m) = CQ(t);
1.11 noro 3536: }
3537: NEXT(m) = 0;
1.61 noro 3538: MKND(NV(p),m0,len,d);
1.14 noro 3539: SG(d) = SG(p);
1.11 noro 3540: return d;
3541: }
3542:
1.3 noro 3543: void ndv_print(NDV p)
3544: {
3545: NMV m;
3546: int i,len;
3547:
1.34 noro 3548: if ( !p ) printf("0\n");
1.3 noro 3549: else {
1.14 noro 3550: len = LEN(p);
1.3 noro 3551: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
1.71 noro 3552: if ( CM(m) & 0x80000000 ) printf("+@_%d*",IFTOF(CM(m)));
3553: else printf("+%d*",CM(m));
1.16 noro 3554: ndl_print(DL(m));
3555: }
3556: printf("\n");
3557: }
3558: }
3559:
1.113 noro 3560: void ndv_print_q(NDV p)
1.16 noro 3561: {
3562: NMV m;
3563: int i,len;
3564:
1.34 noro 3565: if ( !p ) printf("0\n");
1.16 noro 3566: else {
3567: len = LEN(p);
3568: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
3569: printf("+");
1.113 noro 3570: printexpr(CO,(Obj)CQ(m));
1.16 noro 3571: printf("*");
1.14 noro 3572: ndl_print(DL(m));
1.3 noro 3573: }
3574: printf("\n");
3575: }
1.25 noro 3576: }
3577:
1.61 noro 3578: NODE ndv_reducebase(NODE x)
1.27 noro 3579: {
3580: int len,i,j;
3581: NDV *w;
3582: NODE t,t0;
3583:
3584: len = length(x);
3585: w = (NDV *)ALLOCA(len*sizeof(NDV));
3586: for ( i = 0, t = x; i < len; i++, t = NEXT(t) ) w[i] = BDY(t);
3587: for ( i = 0; i < len; i++ ) {
3588: for ( j = 0; j < i; j++ ) {
3589: if ( w[i] && w[j] )
3590: if ( ndl_reducible(HDL(w[i]),HDL(w[j])) ) w[i] = 0;
3591: else if ( ndl_reducible(HDL(w[j]),HDL(w[i])) ) w[j] = 0;
3592: }
3593: }
3594: for ( i = len-1, t0 = 0; i >= 0; i-- ) {
3595: if ( w[i] ) { NEXTNODE(t0,t); BDY(t) = (pointer)w[i]; }
3596: }
3597: NEXT(t) = 0; x = t0;
3598: return x;
1.11 noro 3599: }
1.32 noro 3600:
1.43 noro 3601: /* XXX incomplete */
3602:
1.32 noro 3603: void nd_init_ord(struct order_spec *ord)
3604: {
1.43 noro 3605: switch ( ord->id ) {
1.32 noro 3606: case 0:
1.43 noro 3607: switch ( ord->ord.simple ) {
3608: case 0:
3609: nd_dcomp = 1;
3610: nd_isrlex = 1;
3611: break;
3612: case 1:
3613: nd_dcomp = 1;
3614: nd_isrlex = 0;
3615: break;
3616: case 2:
3617: nd_dcomp = 0;
3618: nd_isrlex = 0;
1.45 noro 3619: ndl_compare_function = ndl_lex_compare;
1.58 noro 3620: break;
3621: case 11:
3622: /* XXX */
3623: nd_dcomp = 0;
3624: nd_isrlex = 1;
3625: ndl_compare_function = ndl_ww_lex_compare;
1.43 noro 3626: break;
3627: default:
3628: error("nd_gr : unsupported order");
3629: }
1.32 noro 3630: break;
3631: case 1:
1.96 noro 3632: /* block order */
1.43 noro 3633: /* XXX */
3634: nd_dcomp = -1;
1.32 noro 3635: nd_isrlex = 0;
1.45 noro 3636: ndl_compare_function = ndl_block_compare;
1.34 noro 3637: break;
1.43 noro 3638: case 2:
1.96 noro 3639: /* matrix order */
3640: /* XXX */
3641: nd_dcomp = -1;
3642: nd_isrlex = 0;
3643: nd_matrix_len = ord->ord.matrix.row;
3644: nd_matrix = ord->ord.matrix.matrix;
3645: ndl_compare_function = ndl_matrix_compare;
3646: break;
3647: case 3:
1.97 noro 3648: /* composite order */
3649: nd_dcomp = -1;
3650: nd_isrlex = 0;
3651: nd_worb_len = ord->ord.composite.length;
3652: nd_worb = ord->ord.composite.w_or_b;
3653: ndl_compare_function = ndl_composite_compare;
1.32 noro 3654: break;
3655: }
1.41 noro 3656: nd_ord = ord;
1.32 noro 3657: }
3658:
1.43 noro 3659: BlockMask nd_create_blockmask(struct order_spec *ord)
3660: {
3661: int n,i,j,s,l;
1.61 noro 3662: UINT *t;
1.43 noro 3663: BlockMask bm;
3664:
1.96 noro 3665: /* we only create mask table for block order */
3666: if ( ord->id != 1 )
1.43 noro 3667: return 0;
3668: n = ord->ord.block.length;
3669: bm = (BlockMask)MALLOC(sizeof(struct oBlockMask));
3670: bm->n = n;
3671: bm->order_pair = ord->ord.block.order_pair;
1.61 noro 3672: bm->mask = (UINT **)MALLOC(n*sizeof(UINT *));
1.43 noro 3673: for ( i = 0, s = 0; i < n; i++ ) {
1.61 noro 3674: bm->mask[i] = t = (UINT *)MALLOC_ATOMIC(nd_wpd*sizeof(UINT));
1.43 noro 3675: for ( j = 0; j < nd_wpd; j++ ) t[j] = 0;
3676: l = bm->order_pair[i].length;
3677: for ( j = 0; j < l; j++, s++ ) PUT_EXP(t,s,nd_mask0);
3678: }
3679: return bm;
1.57 noro 3680: }
3681:
3682: EPOS nd_create_epos(struct order_spec *ord)
3683: {
3684: int i,j,l,s,ord_l,ord_o;
3685: EPOS epos;
3686: struct order_pair *op;
3687:
3688: epos = (EPOS)MALLOC_ATOMIC(nd_nvar*sizeof(struct oEPOS));
3689: switch ( ord->id ) {
3690: case 0:
3691: if ( nd_isrlex ) {
3692: for ( i = 0; i < nd_nvar; i++ ) {
3693: epos[i].i = nd_exporigin + (nd_nvar-1-i)/nd_epw;
3694: epos[i].s = (nd_epw-((nd_nvar-1-i)%nd_epw)-1)*nd_bpe;
3695: }
3696: } else {
3697: for ( i = 0; i < nd_nvar; i++ ) {
3698: epos[i].i = nd_exporigin + i/nd_epw;
3699: epos[i].s = (nd_epw-(i%nd_epw)-1)*nd_bpe;
3700: }
3701: }
3702: break;
3703: case 1:
3704: /* block order */
3705: l = ord->ord.block.length;
3706: op = ord->ord.block.order_pair;
3707: for ( j = 0, s = 0; j < l; j++ ) {
3708: ord_o = op[j].order;
3709: ord_l = op[j].length;
3710: if ( !ord_o )
3711: for ( i = 0; i < ord_l; i++ ) {
3712: epos[s+i].i = nd_exporigin + (s+ord_l-i-1)/nd_epw;
3713: epos[s+i].s = (nd_epw-((s+ord_l-i-1)%nd_epw)-1)*nd_bpe;
3714: }
3715: else
3716: for ( i = 0; i < ord_l; i++ ) {
3717: epos[s+i].i = nd_exporigin + (s+i)/nd_epw;
3718: epos[s+i].s = (nd_epw-((s+i)%nd_epw)-1)*nd_bpe;
3719: }
3720: s += ord_l;
3721: }
3722: break;
3723: case 2:
1.96 noro 3724: /* matrix order */
3725: case 3:
3726: /* composite order */
3727: for ( i = 0; i < nd_nvar; i++ ) {
3728: epos[i].i = nd_exporigin + i/nd_epw;
3729: epos[i].s = (nd_epw-(i%nd_epw)-1)*nd_bpe;
3730: }
3731: break;
1.57 noro 3732: }
3733: return epos;
1.43 noro 3734: }
1.59 noro 3735:
3736: /* external interface */
3737:
3738: void nd_nf_p(P f,LIST g,LIST v,int m,struct order_spec *ord,P *rp)
3739: {
1.61 noro 3740: NODE t,in0,in;
1.59 noro 3741: ND nd,nf;
1.61 noro 3742: NDV ndv;
3743: VL vv,tv;
3744: int stat,nvar,max,e;
1.113 noro 3745: union oNDC dn;
1.59 noro 3746:
1.116 noro 3747: if ( !f ) {
3748: *rp = 0;
3749: return;
3750: }
1.59 noro 3751: pltovl(v,&vv);
1.74 noro 3752: for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++ );
1.61 noro 3753:
3754: /* get the degree bound */
3755: for ( t = BDY(g), max = 0; t; t = NEXT(t) )
3756: for ( tv = vv; tv; tv = NEXT(tv) ) {
3757: e = getdeg(tv->v,(P)BDY(t));
3758: max = MAX(e,max);
3759: }
3760: for ( tv = vv; tv; tv = NEXT(tv) ) {
3761: e = getdeg(tv->v,f);
3762: max = MAX(e,max);
3763: }
3764:
1.59 noro 3765: nd_init_ord(ord);
1.61 noro 3766: nd_setup_parameters(nvar,max);
3767:
3768: /* conversion to ndv */
3769: for ( in0 = 0, t = BDY(g); t; t = NEXT(t) ) {
3770: NEXTNODE(in0,in);
3771: BDY(in) = (pointer)ptondv(CO,vv,(P)BDY(t));
1.69 noro 3772: if ( m ) ndv_mod(m,(NDV)BDY(in));
1.61 noro 3773: }
3774: NEXTNODE(in0,in);
3775: BDY(in) = (pointer)ptondv(CO,vv,f);
1.69 noro 3776: if ( m ) ndv_mod(m,(NDV)BDY(in));
1.61 noro 3777: NEXT(in) = 0;
3778:
3779: ndv_setup(m,0,in0);
1.59 noro 3780: nd_psn--;
3781: nd_scale=2;
3782: while ( 1 ) {
3783: nd = (pointer)ndvtond(m,nd_ps[nd_psn]);
1.69 noro 3784: stat = nd_nf(m,nd,nd_ps,1,0,&nf);
1.59 noro 3785: if ( !stat ) {
3786: nd_psn++;
1.103 noro 3787: nd_reconstruct(0,0);
1.59 noro 3788: nd_psn--;
3789: } else
3790: break;
3791: }
1.61 noro 3792: *rp = ndvtop(m,CO,vv,ndtondv(m,nf));
1.63 noro 3793: }
3794:
3795: int nd_to_vect(int mod,UINT *s0,int n,ND d,UINT *r)
3796: {
3797: NM m;
3798: UINT *t,*s;
3799: int i;
3800:
3801: for ( i = 0; i < n; i++ ) r[i] = 0;
3802: for ( i = 0, s = s0, m = BDY(d); m; m = NEXT(m) ) {
3803: t = DL(m);
3804: for ( ; !ndl_equal(t,s); s += nd_wpd, i++ );
3805: r[i] = CM(m);
3806: }
3807: for ( i = 0; !r[i]; i++ );
3808: return i;
3809: }
3810:
1.113 noro 3811: int nd_to_vect_q(UINT *s0,int n,ND d,Q *r)
1.74 noro 3812: {
1.107 noro 3813: NM m;
1.74 noro 3814: UINT *t,*s;
1.107 noro 3815: int i;
1.74 noro 3816:
3817: for ( i = 0; i < n; i++ ) r[i] = 0;
1.107 noro 3818: for ( i = 0, s = s0, m = BDY(d); m; m = NEXT(m) ) {
1.74 noro 3819: t = DL(m);
3820: for ( ; !ndl_equal(t,s); s += nd_wpd, i++ );
1.113 noro 3821: r[i] = CQ(m);
1.74 noro 3822: }
3823: for ( i = 0; !r[i]; i++ );
3824: return i;
3825: }
3826:
1.67 noro 3827: IndArray nm_ind_pair_to_vect_compress(int mod,UINT *s0,int n,NM_ind_pair pair)
1.64 noro 3828: {
3829: NM m;
3830: NMV mr;
3831: UINT *d,*t,*s;
3832: NDV p;
1.67 noro 3833: unsigned char *ivc;
3834: unsigned short *ivs;
1.81 noro 3835: UINT *v,*ivi,*s0v;
1.67 noro 3836: int i,j,len,prev,diff,cdiff;
3837: IndArray r;
1.64 noro 3838:
3839: m = pair->mul;
3840: d = DL(m);
3841: p = nd_ps[pair->index];
3842: len = LEN(p);
3843: t = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
1.67 noro 3844: v = (unsigned int *)ALLOCA(len*sizeof(unsigned int));
1.64 noro 3845: for ( i = j = 0, s = s0, mr = BDY(p); j < len; j++, NMV_ADV(mr) ) {
3846: ndl_add(d,DL(mr),t);
3847: for ( ; !ndl_equal(t,s); s += nd_wpd, i++ );
1.67 noro 3848: v[j] = i;
3849: }
1.81 noro 3850: r = (IndArray)MALLOC(sizeof(struct oIndArray));
1.67 noro 3851: r->head = v[0];
3852: diff = 0;
3853: for ( i = 1; i < len; i++ ) {
3854: cdiff = v[i]-v[i-1]; diff = MAX(cdiff,diff);
3855: }
3856: if ( diff < 256 ) {
3857: r->width = 1;
3858: ivc = (unsigned char *)MALLOC_ATOMIC(len*sizeof(unsigned char));
3859: r->index.c = ivc;
3860: for ( i = 1, ivc[0] = 0; i < len; i++ ) ivc[i] = v[i]-v[i-1];
3861: } else if ( diff < 65536 ) {
3862: r->width = 2;
3863: ivs = (unsigned short *)MALLOC_ATOMIC(len*sizeof(unsigned short));
3864: r->index.s = ivs;
3865: for ( i = 1, ivs[0] = 0; i < len; i++ ) ivs[i] = v[i]-v[i-1];
3866: } else {
3867: r->width = 4;
3868: ivi = (unsigned int *)MALLOC_ATOMIC(len*sizeof(unsigned int));
3869: r->index.i = ivi;
3870: for ( i = 1, ivi[0] = 0; i < len; i++ ) ivi[i] = v[i]-v[i-1];
1.64 noro 3871: }
1.67 noro 3872: return r;
1.64 noro 3873: }
3874:
1.65 noro 3875:
1.113 noro 3876: int ndv_reduce_vect_q(Q *svect,int col,IndArray *imat,NM_ind_pair *rp0,int nred)
1.107 noro 3877: {
3878: int i,j,k,len,pos,prev;
1.113 noro 3879: Q cs,mcs,c1,c2,cr,gcd,t;
1.107 noro 3880: IndArray ivect;
3881: unsigned char *ivc;
3882: unsigned short *ivs;
3883: unsigned int *ivi;
3884: NDV redv;
3885: NMV mr;
3886: NODE rp;
3887: int maxrs;
3888:
3889: maxrs = 0;
3890: for ( i = 0; i < nred; i++ ) {
3891: ivect = imat[i];
3892: k = ivect->head;
3893: if ( svect[k] ) {
3894: maxrs = MAX(maxrs,rp0[i]->sugar);
3895: redv = nd_ps[rp0[i]->index];
3896: len = LEN(redv); mr = BDY(redv);
1.113 noro 3897: igcd_cofactor(svect[k],CQ(mr),&gcd,&cs,&cr);
3898: chsgnq(cs,&mcs);
3899: if ( !UNIQ(cr) ) {
3900: for ( j = 0; j < col; j++ ) {
3901: mulq(svect[j],cr,&c1); svect[j] = c1;
3902: }
3903: }
1.107 noro 3904: svect[k] = 0; prev = k;
3905: switch ( ivect->width ) {
3906: case 1:
3907: ivc = ivect->index.c;
3908: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3909: pos = prev+ivc[j]; prev = pos;
1.113 noro 3910: mulq(CQ(mr),mcs,&c2); addq(svect[pos],c2,&t); svect[pos] = t;
1.107 noro 3911: }
3912: break;
3913: case 2:
3914: ivs = ivect->index.s;
3915: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3916: pos = prev+ivs[j]; prev = pos;
1.113 noro 3917: mulq(CQ(mr),mcs,&c2); addq(svect[pos],c2,&t); svect[pos] = t;
1.107 noro 3918: }
3919: break;
3920: case 4:
3921: ivi = ivect->index.i;
3922: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3923: pos = prev+ivi[j]; prev = pos;
1.113 noro 3924: mulq(CQ(mr),mcs,&c2); addq(svect[pos],c2,&t); svect[pos] = t;
1.107 noro 3925: }
3926: break;
3927: }
3928: }
3929: }
3930: return maxrs;
3931: }
3932:
1.76 noro 3933: int ndv_reduce_vect(int m,UINT *svect,int col,IndArray *imat,NM_ind_pair *rp0,int nred)
1.65 noro 3934: {
1.67 noro 3935: int i,j,k,len,pos,prev;
1.66 noro 3936: UINT c,c1,c2,c3,up,lo,dmy;
1.67 noro 3937: IndArray ivect;
3938: unsigned char *ivc;
3939: unsigned short *ivs;
3940: unsigned int *ivi;
1.65 noro 3941: NDV redv;
1.67 noro 3942: NMV mr;
1.65 noro 3943: NODE rp;
1.76 noro 3944: int maxrs;
1.65 noro 3945:
1.76 noro 3946: maxrs = 0;
1.74 noro 3947: for ( i = 0; i < nred; i++ ) {
1.65 noro 3948: ivect = imat[i];
1.67 noro 3949: k = ivect->head; svect[k] %= m;
1.65 noro 3950: if ( c = svect[k] ) {
1.76 noro 3951: maxrs = MAX(maxrs,rp0[i]->sugar);
1.74 noro 3952: c = m-c; redv = nd_ps[rp0[i]->index];
1.67 noro 3953: len = LEN(redv); mr = BDY(redv);
3954: svect[k] = 0; prev = k;
3955: switch ( ivect->width ) {
3956: case 1:
3957: ivc = ivect->index.c;
3958: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3959: pos = prev+ivc[j]; c1 = CM(mr); c2 = svect[pos];
3960: prev = pos;
3961: DMA(c1,c,c2,up,lo);
3962: if ( up ) { DSAB(m,up,lo,dmy,c3); svect[pos] = c3;
3963: } else svect[pos] = lo;
3964: }
3965: break;
3966: case 2:
3967: ivs = ivect->index.s;
3968: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3969: pos = prev+ivs[j]; c1 = CM(mr); c2 = svect[pos];
3970: prev = pos;
3971: DMA(c1,c,c2,up,lo);
3972: if ( up ) { DSAB(m,up,lo,dmy,c3); svect[pos] = c3;
3973: } else svect[pos] = lo;
3974: }
3975: break;
3976: case 4:
3977: ivi = ivect->index.i;
3978: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3979: pos = prev+ivi[j]; c1 = CM(mr); c2 = svect[pos];
3980: prev = pos;
3981: DMA(c1,c,c2,up,lo);
3982: if ( up ) { DSAB(m,up,lo,dmy,c3); svect[pos] = c3;
3983: } else svect[pos] = lo;
3984: }
3985: break;
1.65 noro 3986: }
3987: }
3988: }
1.66 noro 3989: for ( i = 0; i < col; i++ )
3990: if ( svect[i] >= (UINT)m ) svect[i] %= m;
1.76 noro 3991: return maxrs;
1.65 noro 3992: }
3993:
1.76 noro 3994: int ndv_reduce_vect_sf(int m,UINT *svect,int col,IndArray *imat,NM_ind_pair *rp0,int nred)
1.72 noro 3995: {
3996: int i,j,k,len,pos,prev;
3997: UINT c,c1,c2,c3,up,lo,dmy;
3998: IndArray ivect;
3999: unsigned char *ivc;
4000: unsigned short *ivs;
4001: unsigned int *ivi;
4002: NDV redv;
4003: NMV mr;
4004: NODE rp;
1.76 noro 4005: int maxrs;
1.72 noro 4006:
1.76 noro 4007: maxrs = 0;
1.74 noro 4008: for ( i = 0; i < nred; i++ ) {
1.72 noro 4009: ivect = imat[i];
4010: k = ivect->head; svect[k] %= m;
4011: if ( c = svect[k] ) {
1.76 noro 4012: maxrs = MAX(maxrs,rp0[i]->sugar);
1.74 noro 4013: c = _chsgnsf(c); redv = nd_ps[rp0[i]->index];
1.72 noro 4014: len = LEN(redv); mr = BDY(redv);
4015: svect[k] = 0; prev = k;
4016: switch ( ivect->width ) {
4017: case 1:
4018: ivc = ivect->index.c;
4019: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
4020: pos = prev+ivc[j]; prev = pos;
4021: svect[pos] = _addsf(_mulsf(CM(mr),c),svect[pos]);
4022: }
4023: break;
4024: case 2:
4025: ivs = ivect->index.s;
4026: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
4027: pos = prev+ivs[j]; prev = pos;
4028: svect[pos] = _addsf(_mulsf(CM(mr),c),svect[pos]);
4029: }
4030: break;
4031: case 4:
4032: ivi = ivect->index.i;
4033: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
4034: pos = prev+ivi[j]; prev = pos;
4035: svect[pos] = _addsf(_mulsf(CM(mr),c),svect[pos]);
4036: }
4037: break;
4038: }
4039: }
4040: }
1.76 noro 4041: return maxrs;
1.72 noro 4042: }
4043:
1.65 noro 4044: NDV vect_to_ndv(UINT *vect,int spcol,int col,int *rhead,UINT *s0vect)
4045: {
4046: int j,k,len;
4047: UINT *p;
4048: UINT c;
4049: NDV r;
4050: NMV mr0,mr;
4051:
4052: for ( j = 0, len = 0; j < spcol; j++ ) if ( vect[j] ) len++;
4053: if ( !len ) return 0;
4054: else {
1.74 noro 4055: mr0 = (NMV)GC_malloc_atomic_ignore_off_page(nmv_adv*len);
1.103 noro 4056: #if 0
1.74 noro 4057: ndv_alloc += nmv_adv*len;
1.103 noro 4058: #endif
1.65 noro 4059: mr = mr0;
4060: p = s0vect;
4061: for ( j = k = 0; j < col; j++, p += nd_wpd )
4062: if ( !rhead[j] ) {
4063: if ( c = vect[k++] ) {
4064: ndl_copy(p,DL(mr)); CM(mr) = c; NMV_ADV(mr);
4065: }
4066: }
4067: MKNDV(nd_nvar,mr0,len,r);
4068: return r;
4069: }
4070: }
4071:
1.113 noro 4072: NDV vect_to_ndv_q(Q *vect,int spcol,int col,int *rhead,UINT *s0vect)
1.107 noro 4073: {
4074: int j,k,len;
4075: UINT *p;
1.113 noro 4076: Q c;
1.107 noro 4077: NDV r;
4078: NMV mr0,mr;
4079:
4080: for ( j = 0, len = 0; j < spcol; j++ ) if ( vect[j] ) len++;
4081: if ( !len ) return 0;
4082: else {
1.109 noro 4083: mr0 = (NMV)GC_malloc(nmv_adv*len);
1.107 noro 4084: #if 0
4085: ndv_alloc += nmv_adv*len;
4086: #endif
4087: mr = mr0;
4088: p = s0vect;
4089: for ( j = k = 0; j < col; j++, p += nd_wpd )
4090: if ( !rhead[j] ) {
4091: if ( c = vect[k++] ) {
1.113 noro 4092: if ( DN(c) )
4093: error("afo");
4094: ndl_copy(p,DL(mr)); CQ(mr) = c; NMV_ADV(mr);
1.107 noro 4095: }
4096: }
4097: MKNDV(nd_nvar,mr0,len,r);
4098: return r;
4099: }
4100: }
4101:
1.74 noro 4102: int nd_sp_f4(int m,ND_pairs l,PGeoBucket bucket)
1.65 noro 4103: {
4104: ND_pairs t;
4105: NODE sp0,sp;
4106: int stat;
4107: ND spol;
4108:
4109: for ( t = l; t; t = NEXT(t) ) {
4110: stat = nd_sp(m,0,t,&spol);
4111: if ( !stat ) return 0;
4112: if ( spol ) {
4113: add_pbucket_symbolic(bucket,spol);
4114: }
4115: }
1.68 noro 4116: return 1;
1.65 noro 4117: }
4118:
4119: int nd_symbolic_preproc(PGeoBucket bucket,UINT **s0vect,NODE *r)
4120: {
4121: NODE rp0,rp;
4122: NM mul,head,s0,s;
1.76 noro 4123: int index,col,i,sugar;
1.65 noro 4124: RHist h;
4125: UINT *s0v,*p;
4126: NM_ind_pair pair;
4127: ND red;
4128:
4129: s0 = 0; rp0 = 0; col = 0;
4130: while ( 1 ) {
4131: head = remove_head_pbucket_symbolic(bucket);
4132: if ( !head ) break;
4133: if ( !s0 ) s0 = head;
4134: else NEXT(s) = head;
4135: s = head;
4136: index = ndl_find_reducer(DL(head));
4137: if ( index >= 0 ) {
4138: h = nd_psh[index];
4139: NEWNM(mul);
4140: ndl_sub(DL(head),DL(h),DL(mul));
4141: if ( ndl_check_bound2(index,DL(mul)) ) return 0;
1.76 noro 4142: sugar = TD(DL(mul))+SG(nd_ps[index]);
4143: MKNM_ind_pair(pair,mul,index,sugar);
1.65 noro 4144: red = ndv_mul_nm_symbolic(mul,nd_ps[index]);
4145: add_pbucket_symbolic(bucket,nd_remove_head(red));
4146: NEXTNODE(rp0,rp); BDY(rp) = (pointer)pair;
4147: }
4148: col++;
4149: }
1.72 noro 4150: if ( rp0 ) NEXT(rp) = 0;
4151: NEXT(s) = 0;
1.65 noro 4152: s0v = (UINT *)MALLOC_ATOMIC(col*nd_wpd*sizeof(UINT));
4153: for ( i = 0, p = s0v, s = s0; i < col;
4154: i++, p += nd_wpd, s = NEXT(s) ) ndl_copy(DL(s),p);
4155: *s0vect = s0v;
4156: *r = rp0;
4157: return col;
4158: }
4159:
1.69 noro 4160: NODE nd_f4(int m)
4161: {
4162: int i,nh,stat,index;
4163: NODE r,g;
4164: ND_pairs d,l,t;
4165: ND spol,red;
4166: NDV nf,redv;
4167: NM s0,s;
1.84 noro 4168: NODE rp0,srp0,nflist;
1.69 noro 4169: int nsp,nred,col,rank,len,k,j,a;
4170: UINT c;
1.74 noro 4171: UINT **spmat;
1.69 noro 4172: UINT *s0vect,*svect,*p,*v;
4173: int *colstat;
4174: IndArray *imat;
4175: int *rhead;
4176: int spcol,sprow;
4177: int sugar;
4178: PGeoBucket bucket;
4179: struct oEGT eg0,eg1,eg_f4;
4180:
1.103 noro 4181: #if 0
1.74 noro 4182: ndv_alloc = 0;
1.103 noro 4183: #endif
1.69 noro 4184: g = 0; d = 0;
4185: for ( i = 0; i < nd_psn; i++ ) {
4186: d = update_pairs(d,g,i);
4187: g = update_base(g,i);
4188: }
4189: while ( d ) {
4190: get_eg(&eg0);
4191: l = nd_minsugarp(d,&d);
4192: sugar = SG(l);
4193: bucket = create_pbucket();
1.74 noro 4194: stat = nd_sp_f4(m,l,bucket);
1.69 noro 4195: if ( !stat ) {
4196: for ( t = l; NEXT(t); t = NEXT(t) );
4197: NEXT(t) = d; d = l;
1.103 noro 4198: d = nd_reconstruct(0,d);
1.69 noro 4199: continue;
4200: }
1.86 noro 4201: if ( bucket->m < 0 ) continue;
1.69 noro 4202: col = nd_symbolic_preproc(bucket,&s0vect,&rp0);
4203: if ( !col ) {
4204: for ( t = l; NEXT(t); t = NEXT(t) );
4205: NEXT(t) = d; d = l;
1.103 noro 4206: d = nd_reconstruct(0,d);
1.69 noro 4207: continue;
4208: }
4209: get_eg(&eg1); init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg1);
1.74 noro 4210: if ( DP_Print )
4211: fprintf(asir_out,"sugar=%d,symb=%fsec,",
4212: sugar,eg_f4.exectime+eg_f4.gctime);
4213: if ( 1 )
4214: nflist = nd_f4_red(m,l,s0vect,col,rp0);
4215: else
4216: nflist = nd_f4_red_dist(m,l,s0vect,col,rp0);
1.69 noro 4217: /* adding new bases */
1.74 noro 4218: for ( r = nflist; r; r = NEXT(r) ) {
4219: nf = (NDV)BDY(r);
1.69 noro 4220: ndv_removecont(m,nf);
1.77 noro 4221: nh = ndv_newps(m,nf,0);
1.69 noro 4222: d = update_pairs(d,g,nh);
4223: g = update_base(g,nh);
4224: }
4225: }
4226: for ( r = g; r; r = NEXT(r) ) BDY(r) = (pointer)nd_ps[(int)BDY(r)];
1.103 noro 4227: #if 0
1.74 noro 4228: fprintf(asir_out,"ndv_alloc=%d\n",ndv_alloc);
1.103 noro 4229: #endif
1.69 noro 4230: return g;
4231: }
1.74 noro 4232:
4233: NODE nd_f4_red(int m,ND_pairs sp0,UINT *s0vect,int col,NODE rp0)
1.63 noro 4234: {
1.67 noro 4235: IndArray *imat;
1.106 noro 4236: int nsp,nred,i;
1.65 noro 4237: int *rhead;
1.106 noro 4238: NODE r0,rp;
1.74 noro 4239: ND_pairs sp;
4240: NM_ind_pair *rvect;
1.63 noro 4241:
1.74 noro 4242: for ( sp = sp0, nsp = 0; sp; sp = NEXT(sp), nsp++ );
1.106 noro 4243: nred = length(rp0);
1.74 noro 4244: imat = (IndArray *)ALLOCA(nred*sizeof(IndArray));
4245: rhead = (int *)ALLOCA(col*sizeof(int));
4246: for ( i = 0; i < col; i++ ) rhead[i] = 0;
1.63 noro 4247:
1.74 noro 4248: /* construction of index arrays */
4249: rvect = (NM_ind_pair *)ALLOCA(nred*sizeof(NM_ind_pair));
4250: for ( rp = rp0, i = 0; rp; i++, rp = NEXT(rp) ) {
4251: rvect[i] = (NM_ind_pair)BDY(rp);
4252: imat[i] = nm_ind_pair_to_vect_compress(m,s0vect,col,rvect[i]);
4253: rhead[imat[i]->head] = 1;
4254: }
1.107 noro 4255: if ( m )
4256: r0 = nd_f4_red_main(m,sp0,nsp,s0vect,col,rvect,rhead,imat,nred);
4257: else
1.113 noro 4258: r0 = nd_f4_red_q_main(sp0,nsp,s0vect,col,rvect,rhead,imat,nred);
1.106 noro 4259: return r0;
4260: }
1.74 noro 4261:
1.106 noro 4262: NODE nd_f4_red_main(int m,ND_pairs sp0,int nsp,UINT *s0vect,int col,
4263: NM_ind_pair *rvect,int *rhead,IndArray *imat,int nred)
4264: {
4265: int spcol,sprow,a;
4266: int i,j,k,l,rank;
4267: NODE r0,r;
4268: ND_pairs sp;
4269: ND spol;
4270: int **spmat;
4271: UINT *svect,*v;
4272: int *colstat;
4273: struct oEGT eg0,eg1,eg2,eg_f4,eg_f4_1,eg_f4_2;
4274: int maxrs;
4275: int *spsugar;
4276:
4277: spcol = col-nred;
4278: get_eg(&eg0);
1.74 noro 4279: /* elimination (1st step) */
4280: spmat = (int **)ALLOCA(nsp*sizeof(UINT *));
4281: svect = (UINT *)ALLOCA(col*sizeof(UINT));
1.76 noro 4282: spsugar = (int *)ALLOCA(nsp*sizeof(UINT));
1.74 noro 4283: for ( a = sprow = 0, sp = sp0; a < nsp; a++, sp = NEXT(sp) ) {
4284: nd_sp(m,0,sp,&spol);
1.75 noro 4285: if ( !spol ) continue;
1.74 noro 4286: nd_to_vect(m,s0vect,col,spol,svect);
1.76 noro 4287: if ( m == -1 )
4288: maxrs = ndv_reduce_vect_sf(m,svect,col,imat,rvect,nred);
4289: else
4290: maxrs = ndv_reduce_vect(m,svect,col,imat,rvect,nred);
1.74 noro 4291: for ( i = 0; i < col; i++ ) if ( svect[i] ) break;
4292: if ( i < col ) {
4293: spmat[sprow] = v = (UINT *)MALLOC_ATOMIC(spcol*sizeof(UINT));
4294: for ( j = k = 0; j < col; j++ )
4295: if ( !rhead[j] ) v[k++] = svect[j];
1.76 noro 4296: spsugar[sprow] = MAX(maxrs,SG(spol));
1.74 noro 4297: sprow++;
4298: }
1.76 noro 4299: nd_free(spol);
1.74 noro 4300: }
1.80 noro 4301: get_eg(&eg1); init_eg(&eg_f4_1); add_eg(&eg_f4_1,&eg0,&eg1);
4302: if ( DP_Print ) {
4303: fprintf(asir_out,"elim1=%fsec,",eg_f4_1.exectime+eg_f4_1.gctime);
4304: fflush(asir_out);
4305: }
1.74 noro 4306: /* free index arrays */
4307: for ( i = 0; i < nred; i++ ) GC_free(imat[i]->index.c);
4308:
4309: /* elimination (2nd step) */
4310: colstat = (int *)ALLOCA(spcol*sizeof(int));
4311: if ( m == -1 )
1.76 noro 4312: rank = nd_gauss_elim_sf(spmat,spsugar,sprow,spcol,m,colstat);
1.74 noro 4313: else
1.76 noro 4314: rank = nd_gauss_elim_mod(spmat,spsugar,sprow,spcol,m,colstat);
1.74 noro 4315: r0 = 0;
4316: for ( i = 0; i < rank; i++ ) {
4317: NEXTNODE(r0,r); BDY(r) =
4318: (pointer)vect_to_ndv(spmat[i],spcol,col,rhead,s0vect);
1.76 noro 4319: SG((NDV)BDY(r)) = spsugar[i];
1.74 noro 4320: GC_free(spmat[i]);
4321: }
1.109 noro 4322: if ( r0 ) NEXT(r) = 0;
1.74 noro 4323: for ( ; i < sprow; i++ ) GC_free(spmat[i]);
1.80 noro 4324: get_eg(&eg2); init_eg(&eg_f4_2); add_eg(&eg_f4_2,&eg1,&eg2);
4325: init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg2);
1.74 noro 4326: if ( DP_Print ) {
1.80 noro 4327: fprintf(asir_out,"elim2=%fsec\n",eg_f4_2.exectime+eg_f4_2.gctime);
1.74 noro 4328: fprintf(asir_out,"nsp=%d,nred=%d,spmat=(%d,%d),rank=%d ",
4329: nsp,nred,sprow,spcol,rank);
4330: fprintf(asir_out,"%fsec\n",eg_f4.exectime+eg_f4.gctime);
1.63 noro 4331: }
1.74 noro 4332: return r0;
4333: }
4334:
1.113 noro 4335: NODE nd_f4_red_q_main(ND_pairs sp0,int nsp,UINT *s0vect,int col,
1.107 noro 4336: NM_ind_pair *rvect,int *rhead,IndArray *imat,int nred)
4337: {
4338: int spcol,sprow,a;
4339: int i,j,k,l,rank;
4340: NODE r0,r;
4341: ND_pairs sp;
4342: ND spol;
1.113 noro 4343: Q **spmat;
4344: Q *svect,*v;
1.107 noro 4345: int *colstat;
4346: struct oEGT eg0,eg1,eg2,eg_f4,eg_f4_1,eg_f4_2;
4347: int maxrs;
4348: int *spsugar;
4349:
4350: spcol = col-nred;
4351: get_eg(&eg0);
4352: /* elimination (1st step) */
1.113 noro 4353: spmat = (Q **)ALLOCA(nsp*sizeof(Q *));
4354: svect = (Q *)ALLOCA(col*sizeof(Q));
4355: spsugar = (int *)ALLOCA(nsp*sizeof(Q));
1.107 noro 4356: for ( a = sprow = 0, sp = sp0; a < nsp; a++, sp = NEXT(sp) ) {
4357: nd_sp(0,0,sp,&spol);
4358: if ( !spol ) continue;
1.113 noro 4359: nd_to_vect_q(s0vect,col,spol,svect);
4360: maxrs = ndv_reduce_vect_q(svect,col,imat,rvect,nred);
1.107 noro 4361: for ( i = 0; i < col; i++ ) if ( svect[i] ) break;
4362: if ( i < col ) {
1.113 noro 4363: spmat[sprow] = v = (Q *)MALLOC(spcol*sizeof(Q));
1.107 noro 4364: for ( j = k = 0; j < col; j++ )
4365: if ( !rhead[j] ) v[k++] = svect[j];
4366: spsugar[sprow] = MAX(maxrs,SG(spol));
4367: sprow++;
4368: }
1.109 noro 4369: /* nd_free(spol); */
1.107 noro 4370: }
4371: get_eg(&eg1); init_eg(&eg_f4_1); add_eg(&eg_f4_1,&eg0,&eg1);
4372: if ( DP_Print ) {
4373: fprintf(asir_out,"elim1=%fsec,",eg_f4_1.exectime+eg_f4_1.gctime);
4374: fflush(asir_out);
4375: }
4376: /* free index arrays */
1.109 noro 4377: /* for ( i = 0; i < nred; i++ ) GC_free(imat[i]->index.c); */
1.107 noro 4378:
4379: /* elimination (2nd step) */
4380: colstat = (int *)ALLOCA(spcol*sizeof(int));
1.113 noro 4381: rank = nd_gauss_elim_q(spmat,spsugar,sprow,spcol,colstat);
1.107 noro 4382: r0 = 0;
4383: for ( i = 0; i < rank; i++ ) {
4384: NEXTNODE(r0,r); BDY(r) =
1.113 noro 4385: (pointer)vect_to_ndv_q(spmat[i],spcol,col,rhead,s0vect);
1.107 noro 4386: SG((NDV)BDY(r)) = spsugar[i];
1.109 noro 4387: /* GC_free(spmat[i]); */
1.107 noro 4388: }
1.109 noro 4389: if ( r0 ) NEXT(r) = 0;
4390:
4391: /* for ( ; i < sprow; i++ ) GC_free(spmat[i]); */
1.107 noro 4392: get_eg(&eg2); init_eg(&eg_f4_2); add_eg(&eg_f4_2,&eg1,&eg2);
4393: init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg2);
4394: if ( DP_Print ) {
4395: fprintf(asir_out,"elim2=%fsec\n",eg_f4_2.exectime+eg_f4_2.gctime);
4396: fprintf(asir_out,"nsp=%d,nred=%d,spmat=(%d,%d),rank=%d ",
4397: nsp,nred,sprow,spcol,rank);
4398: fprintf(asir_out,"%fsec\n",eg_f4.exectime+eg_f4.gctime);
4399: }
4400: return r0;
4401: }
4402:
1.74 noro 4403: FILE *nd_write,*nd_read;
4404:
4405: void nd_send_int(int a) {
4406: write_int(nd_write,&a);
4407: }
4408:
4409: void nd_send_intarray(int *p,int len) {
4410: write_intarray(nd_write,p,len);
4411: }
4412:
4413: int nd_recv_int() {
4414: int a;
4415:
4416: read_int(nd_read,&a);
4417: return a;
4418: }
4419:
4420: void nd_recv_intarray(int *p,int len) {
4421: read_intarray(nd_read,p,len);
4422: }
4423:
4424: void nd_send_ndv(NDV p) {
4425: int len,i;
4426: NMV m;
4427:
4428: if ( !p ) nd_send_int(0);
4429: else {
4430: len = LEN(p);
4431: nd_send_int(len);
4432: m = BDY(p);
4433: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
4434: nd_send_int(CM(m));
4435: nd_send_intarray(DL(m),nd_wpd);
1.68 noro 4436: }
1.74 noro 4437: }
4438: }
4439:
4440: void nd_send_nd(ND p) {
4441: int len,i;
4442: NM m;
4443:
4444: if ( !p ) nd_send_int(0);
4445: else {
4446: len = LEN(p);
4447: nd_send_int(len);
4448: m = BDY(p);
4449: for ( i = 0; i < len; i++, m = NEXT(m) ) {
4450: nd_send_int(CM(m));
4451: nd_send_intarray(DL(m),nd_wpd);
1.65 noro 4452: }
1.74 noro 4453: }
4454: }
1.65 noro 4455:
1.74 noro 4456: NDV nd_recv_ndv()
4457: {
4458: int len,i;
4459: NMV m,m0;
4460: NDV r;
1.65 noro 4461:
1.74 noro 4462: len = nd_recv_int();
4463: if ( !len ) return 0;
4464: else {
4465: m0 = m = (NMV)GC_malloc_atomic_ignore_off_page(nmv_adv*len);
1.103 noro 4466: #if 0
1.74 noro 4467: ndv_alloc += len*nmv_adv;
1.103 noro 4468: #endif
1.74 noro 4469: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
4470: CM(m) = nd_recv_int();
4471: nd_recv_intarray(DL(m),nd_wpd);
1.71 noro 4472: }
1.74 noro 4473: MKNDV(nd_nvar,m0,len,r);
4474: return r;
4475: }
4476: }
1.65 noro 4477:
1.74 noro 4478: int ox_exec_f4_red(Q proc)
4479: {
4480: Obj obj;
4481: STRING fname;
4482: NODE arg;
4483: int s;
4484: extern int ox_need_conv,ox_file_io;
4485:
4486: MKSTR(fname,"nd_exec_f4_red");
4487: arg = mknode(2,proc,fname);
4488: Pox_cmo_rpc(arg,&obj);
4489: s = get_ox_server_id(QTOS(proc));
4490: nd_write = iofp[s].out;
4491: nd_read = iofp[s].in;
4492: ox_need_conv = ox_file_io = 0;
4493: return s;
4494: }
4495:
4496: NODE nd_f4_red_dist(int m,ND_pairs sp0,UINT *s0vect,int col,NODE rp0)
4497: {
4498: int nsp,nred;
4499: int i,rank,s;
4500: NODE rp,r0,r;
4501: ND_pairs sp;
4502: NM_ind_pair pair;
4503: NMV nmv;
4504: NM nm;
4505: NDV nf;
4506: Obj proc,dmy;
4507:
4508: ox_launch_main(0,0,&proc);
4509: s = ox_exec_f4_red((Q)proc);
4510:
4511: nd_send_int(m);
4512: nd_send_int(nd_nvar);
4513: nd_send_int(nd_bpe);
4514: nd_send_int(nd_wpd);
4515: nd_send_int(nmv_adv);
4516:
1.89 noro 4517: saveobj(nd_write,dp_current_spec->obj); fflush(nd_write);
1.74 noro 4518:
4519: nd_send_int(nd_psn);
4520: for ( i = 0; i < nd_psn; i++ ) nd_send_ndv(nd_ps[i]);
4521:
4522: for ( sp = sp0, nsp = 0; sp; sp = NEXT(sp), nsp++ );
4523: nd_send_int(nsp);
4524: for ( i = 0, sp = sp0; i < nsp; i++, sp = NEXT(sp) ) {
4525: nd_send_int(sp->i1); nd_send_int(sp->i2);
4526: }
4527:
4528: nd_send_int(col); nd_send_intarray(s0vect,col*nd_wpd);
4529:
4530: nred = length(rp0); nd_send_int(nred);
4531: for ( i = 0, rp = rp0; i < nred; i++, rp = NEXT(rp) ) {
4532: pair = (NM_ind_pair)BDY(rp);
4533: nd_send_int(pair->index);
4534: nd_send_intarray(pair->mul->dl,nd_wpd);
4535: }
4536: fflush(nd_write);
4537: rank = nd_recv_int();
4538: fprintf(asir_out,"rank=%d\n",rank);
4539: r0 = 0;
4540: for ( i = 0; i < rank; i++ ) {
4541: nf = nd_recv_ndv();
4542: NEXTNODE(r0,r); BDY(r) = (pointer)nf;
4543: }
4544: Pox_shutdown(mknode(1,proc),&dmy);
4545: return r0;
4546: }
4547:
4548: /* server side */
4549:
4550: void nd_exec_f4_red_dist()
4551: {
4552: int m,i,nsp,col,s0size,nred,spcol,j,k;
4553: NM_ind_pair *rp0;
4554: NDV nf;
4555: UINT *s0vect;
4556: IndArray *imat;
4557: int *rhead;
4558: int **spmat;
4559: UINT *svect,*v;
4560: ND_pairs *sp0;
4561: int *colstat;
4562: int a,sprow,rank;
1.89 noro 4563: struct order_spec *ord;
1.74 noro 4564: Obj ordspec;
4565: ND spol;
1.76 noro 4566: int maxrs;
4567: int *spsugar;
1.74 noro 4568:
4569: nd_read = iofp[0].in;
4570: nd_write = iofp[0].out;
4571: m = nd_recv_int();
4572: nd_nvar = nd_recv_int();
4573: nd_bpe = nd_recv_int();
4574: nd_wpd = nd_recv_int();
4575: nmv_adv = nd_recv_int();
4576:
4577: loadobj(nd_read,&ordspec);
1.89 noro 4578: create_order_spec(0,ordspec,&ord);
4579: nd_init_ord(ord);
1.74 noro 4580: nd_setup_parameters(nd_nvar,0);
4581:
4582: nd_psn = nd_recv_int();
4583: nd_ps = (NDV *)MALLOC(nd_psn*sizeof(NDV));
4584: nd_bound = (UINT **)MALLOC(nd_psn*sizeof(UINT *));
4585: for ( i = 0; i < nd_psn; i++ ) {
4586: nd_ps[i] = nd_recv_ndv();
4587: nd_bound[i] = ndv_compute_bound(nd_ps[i]);
4588: }
4589:
4590: nsp = nd_recv_int();
4591: sp0 = (ND_pairs *)MALLOC(nsp*sizeof(ND_pairs));
4592: for ( i = 0; i < nsp; i++ ) {
4593: NEWND_pairs(sp0[i]);
4594: sp0[i]->i1 = nd_recv_int(); sp0[i]->i2 = nd_recv_int();
4595: ndl_lcm(HDL(nd_ps[sp0[i]->i1]),HDL(nd_ps[sp0[i]->i2]),LCM(sp0[i]));
4596: }
4597:
4598: col = nd_recv_int();
4599: s0size = col*nd_wpd;
4600: s0vect = (UINT *)MALLOC(s0size*sizeof(UINT));
4601: nd_recv_intarray(s0vect,s0size);
4602:
4603: nred = nd_recv_int();
4604: rp0 = (NM_ind_pair *)MALLOC(nred*sizeof(NM_ind_pair));
4605: for ( i = 0; i < nred; i++ ) {
4606: rp0[i] = (NM_ind_pair)MALLOC(sizeof(struct oNM_ind_pair));
4607: rp0[i]->index = nd_recv_int();
4608: rp0[i]->mul = (NM)MALLOC(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT));
4609: nd_recv_intarray(rp0[i]->mul->dl,nd_wpd);
4610: }
4611:
4612: spcol = col-nred;
4613: imat = (IndArray *)MALLOC(nred*sizeof(IndArray));
4614: rhead = (int *)MALLOC(col*sizeof(int));
4615: for ( i = 0; i < col; i++ ) rhead[i] = 0;
4616:
4617: /* construction of index arrays */
4618: for ( i = 0; i < nred; i++ ) {
4619: imat[i] = nm_ind_pair_to_vect_compress(m,s0vect,col,rp0[i]);
4620: rhead[imat[i]->head] = 1;
4621: }
4622:
4623: /* elimination (1st step) */
4624: spmat = (int **)MALLOC(nsp*sizeof(UINT *));
4625: svect = (UINT *)MALLOC(col*sizeof(UINT));
1.76 noro 4626: spsugar = (int *)ALLOCA(nsp*sizeof(UINT));
1.74 noro 4627: for ( a = sprow = 0; a < nsp; a++ ) {
4628: nd_sp(m,0,sp0[a],&spol);
1.75 noro 4629: if ( !spol ) continue;
1.74 noro 4630: nd_to_vect(m,s0vect,col,spol,svect);
1.76 noro 4631: if ( m == -1 )
4632: maxrs = ndv_reduce_vect_sf(m,svect,col,imat,rp0,nred);
4633: else
4634: maxrs = ndv_reduce_vect(m,svect,col,imat,rp0,nred);
1.74 noro 4635: for ( i = 0; i < col; i++ ) if ( svect[i] ) break;
4636: if ( i < col ) {
4637: spmat[sprow] = v = (UINT *)MALLOC(spcol*sizeof(UINT));
4638: for ( j = k = 0; j < col; j++ )
4639: if ( !rhead[j] ) v[k++] = svect[j];
1.76 noro 4640: spsugar[sprow] = MAX(maxrs,SG(spol));
1.74 noro 4641: sprow++;
4642: }
1.76 noro 4643: nd_free(spol);
1.74 noro 4644: }
4645: /* elimination (2nd step) */
4646: colstat = (int *)ALLOCA(spcol*sizeof(int));
4647: if ( m == -1 )
1.76 noro 4648: rank = nd_gauss_elim_sf(spmat,spsugar,sprow,spcol,m,colstat);
1.74 noro 4649: else
1.76 noro 4650: rank = nd_gauss_elim_mod(spmat,spsugar,sprow,spcol,m,colstat);
1.74 noro 4651: nd_send_int(rank);
4652: for ( i = 0; i < rank; i++ ) {
4653: nf = vect_to_ndv(spmat[i],spcol,col,rhead,s0vect);
4654: nd_send_ndv(nf);
1.63 noro 4655: }
1.74 noro 4656: fflush(nd_write);
1.107 noro 4657: }
4658:
1.113 noro 4659: int nd_gauss_elim_q(Q **mat0,int *sugar,int row,int col,int *colstat)
1.107 noro 4660: {
1.109 noro 4661: int mod,i,j,t,c,rank,rank0,inv;
4662: int *ci,*ri;
1.113 noro 4663: Q dn;
1.109 noro 4664: MAT m,nm;
1.108 noro 4665: int **wmat;
4666:
4667: /* XXX */
4668: mod = 99999989;
4669: wmat = (int **)ALLOCA(row*sizeof(int *));
4670: for ( i = 0; i < row; i++ ) {
4671: wmat[i] = (int *)ALLOCA(col*sizeof(int));
4672: for ( j = 0; j < col; j++ ) {
1.113 noro 4673: if ( mat0[i][j] ) {
4674: t = rem(NM(mat0[i][j]),mod);
4675: if ( SGN(mat0[i][j]) < 0 ) t = mod-t;
4676: wmat[i][j] = t;
4677: } else
1.108 noro 4678: wmat[i][j] = 0;
4679: }
4680: }
1.109 noro 4681: rank0 = nd_gauss_elim_mod(wmat,sugar,row,col,mod,colstat);
4682: NEWMAT(m); m->row = row; m->col = col; m->body = (pointer **)mat0;
1.113 noro 4683: rank = generic_gauss_elim(m,&nm,&dn,&ri,&ci);
1.109 noro 4684: if ( rank != rank0 )
4685: error("afo");
4686: for ( i = 0; i < row; i++ )
4687: for ( j = 0; j < col; j++ )
4688: mat0[i][j] = 0;
4689: c = col-rank;
4690: for ( i = 0; i < rank; i++ ) {
1.113 noro 4691: mat0[i][ri[i]] = dn;
1.109 noro 4692: for ( j = 0; j < c; j++ )
1.113 noro 4693: mat0[i][ci[j]] = (Q)BDY(nm)[i][j];
1.109 noro 4694: }
1.113 noro 4695: inv = invm(rem(NM(dn),mod),mod);
4696: if ( SGN(dn) < 0 ) inv = mod-inv;
1.109 noro 4697: for ( i = 0; i < row; i++ )
4698: for ( j = 0; j < col; j++ ) {
1.113 noro 4699: if ( mat0[i][j] ) {
4700: t = rem(NM(mat0[i][j]),mod);
4701: if ( SGN(mat0[i][j]) < 0 ) t = mod-t;
4702: } else
1.109 noro 4703: t = 0;
4704: c = dmar(t,inv,0,mod);
4705: if ( wmat[i][j] != c )
4706: error("afo");
4707: }
4708: return rank;
1.76 noro 4709: }
4710:
4711: int nd_gauss_elim_mod(int **mat0,int *sugar,int row,int col,int md,int *colstat)
4712: {
4713: int i,j,k,l,inv,a,rank,s;
4714: unsigned int *t,*pivot,*pk;
4715: unsigned int **mat;
4716:
4717: mat = (unsigned int **)mat0;
4718: for ( rank = 0, j = 0; j < col; j++ ) {
4719: for ( i = rank; i < row; i++ )
4720: mat[i][j] %= md;
4721: for ( i = rank; i < row; i++ )
4722: if ( mat[i][j] )
4723: break;
4724: if ( i == row ) {
4725: colstat[j] = 0;
4726: continue;
4727: } else
4728: colstat[j] = 1;
4729: if ( i != rank ) {
4730: t = mat[i]; mat[i] = mat[rank]; mat[rank] = t;
4731: s = sugar[i]; sugar[i] = sugar[rank]; sugar[rank] = s;
4732: }
4733: pivot = mat[rank];
4734: s = sugar[rank];
4735: inv = invm(pivot[j],md);
4736: for ( k = j, pk = pivot+k; k < col; k++, pk++ )
4737: if ( *pk ) {
4738: if ( *pk >= (unsigned int)md )
4739: *pk %= md;
4740: DMAR(*pk,inv,0,md,*pk)
4741: }
4742: for ( i = rank+1; i < row; i++ ) {
4743: t = mat[i];
4744: if ( a = t[j] ) {
4745: sugar[i] = MAX(sugar[i],s);
4746: red_by_vect(md,t+j,pivot+j,md-a,col-j);
4747: }
4748: }
4749: rank++;
4750: }
4751: for ( j = col-1, l = rank-1; j >= 0; j-- )
4752: if ( colstat[j] ) {
4753: pivot = mat[l];
4754: s = sugar[l];
4755: for ( i = 0; i < l; i++ ) {
4756: t = mat[i];
4757: t[j] %= md;
4758: if ( a = t[j] ) {
4759: sugar[i] = MAX(sugar[i],s);
4760: red_by_vect(md,t+j,pivot+j,md-a,col-j);
4761: }
4762: }
4763: l--;
4764: }
4765: for ( j = 0, l = 0; l < rank; j++ )
4766: if ( colstat[j] ) {
4767: t = mat[l];
4768: for ( k = j; k < col; k++ )
4769: if ( t[k] >= (unsigned int)md )
4770: t[k] %= md;
4771: l++;
4772: }
4773: return rank;
4774: }
4775:
4776: int nd_gauss_elim_sf(int **mat0,int *sugar,int row,int col,int md,int *colstat)
4777: {
4778: int i,j,k,l,inv,a,rank,s;
4779: unsigned int *t,*pivot,*pk;
4780: unsigned int **mat;
4781:
4782: mat = (unsigned int **)mat0;
4783: for ( rank = 0, j = 0; j < col; j++ ) {
4784: for ( i = rank; i < row; i++ )
4785: if ( mat[i][j] )
4786: break;
4787: if ( i == row ) {
4788: colstat[j] = 0;
4789: continue;
4790: } else
4791: colstat[j] = 1;
4792: if ( i != rank ) {
4793: t = mat[i]; mat[i] = mat[rank]; mat[rank] = t;
4794: s = sugar[i]; sugar[i] = sugar[rank]; sugar[rank] = s;
4795: }
4796: pivot = mat[rank];
4797: s = sugar[rank];
4798: inv = _invsf(pivot[j]);
4799: for ( k = j, pk = pivot+k; k < col; k++, pk++ )
4800: if ( *pk )
4801: *pk = _mulsf(*pk,inv);
4802: for ( i = rank+1; i < row; i++ ) {
4803: t = mat[i];
4804: if ( a = t[j] ) {
4805: sugar[i] = MAX(sugar[i],s);
4806: red_by_vect_sf(md,t+j,pivot+j,_chsgnsf(a),col-j);
4807: }
4808: }
4809: rank++;
4810: }
4811: for ( j = col-1, l = rank-1; j >= 0; j-- )
4812: if ( colstat[j] ) {
4813: pivot = mat[l];
4814: s = sugar[l];
4815: for ( i = 0; i < l; i++ ) {
4816: t = mat[i];
4817: if ( a = t[j] ) {
4818: sugar[i] = MAX(sugar[i],s);
4819: red_by_vect_sf(md,t+j,pivot+j,_chsgnsf(a),col-j);
4820: }
4821: }
4822: l--;
4823: }
4824: return rank;
1.77 noro 4825: }
4826:
4827: int ndv_ishomo(NDV p)
4828: {
4829: NMV m;
4830: int len,h;
4831:
4832: if ( !p ) return 1;
4833: len = LEN(p);
4834: m = BDY(p);
4835: h = TD(DL(m));
4836: NMV_ADV(m);
4837: for ( len--; len; len--, NMV_ADV(m) )
4838: if ( TD(DL(m)) != h ) return 0;
4839: return 1;
4840: }
4841:
4842: void ndv_save(NDV p,int index)
4843: {
4844: FILE *s;
4845: char name[BUFSIZ];
4846: short id;
4847: int nv,sugar,len,n,i,td,e,j;
4848: NMV m;
4849: unsigned int *dl;
4850:
4851: sprintf(name,"%s/%d",Demand,index);
4852: s = fopen(name,"w");
4853: savevl(s,0);
4854: if ( !p ) {
4855: saveobj(s,0);
4856: return;
4857: }
4858: id = O_DP;
4859: nv = NV(p);
4860: sugar = SG(p);
4861: len = LEN(p);
4862: write_short(s,&id); write_int(s,&nv); write_int(s,&sugar);
4863: write_int(s,&len);
4864:
4865: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
1.113 noro 4866: saveobj(s,(Obj)CQ(m));
1.77 noro 4867: dl = DL(m);
4868: td = TD(dl);
4869: write_int(s,&td);
4870: for ( j = 0; j < nv; j++ ) {
4871: e = GET_EXP(dl,j);
4872: write_int(s,&e);
4873: }
4874: }
4875: fclose(s);
4876: }
4877:
4878: NDV ndv_load(int index)
4879: {
4880: FILE *s;
4881: char name[BUFSIZ];
4882: short id;
4883: int nv,sugar,len,n,i,td,e,j;
4884: NDV d;
4885: NMV m0,m;
4886: unsigned int *dl;
4887: Obj obj;
4888:
4889: sprintf(name,"%s/%d",Demand,index);
4890: s = fopen(name,"r");
4891: if ( !s ) return 0;
4892:
4893: skipvl(s);
4894: read_short(s,&id);
4895: if ( !id ) return 0;
4896: read_int(s,&nv);
4897: read_int(s,&sugar);
4898: read_int(s,&len);
4899:
4900: m0 = m = MALLOC(len*nmv_adv);
4901: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
1.113 noro 4902: loadobj(s,&obj); CQ(m) = (Q)obj;
1.77 noro 4903: dl = DL(m);
4904: ndl_zero(dl);
4905: read_int(s,&td); TD(dl) = td;
4906: for ( j = 0; j < nv; j++ ) {
4907: read_int(s,&e);
4908: PUT_EXP(dl,j,e);
4909: }
4910: if ( nd_blockmask ) ndl_weight_mask(dl);
4911: }
4912: fclose(s);
4913: MKNDV(nv,m0,len,d);
4914: SG(d) = sugar;
4915: return d;
1.99 noro 4916: }
4917:
1.102 noro 4918: void nd_det(int mod,MAT f,P *rp)
1.99 noro 4919: {
4920: VL fv,tv;
1.102 noro 4921: int n,i,j,max,e,nvar,sgn,k0,l0,len0,len,k,l,a;
1.99 noro 4922: pointer **m;
1.113 noro 4923: Q mone;
1.102 noro 4924: NDV **dm;
4925: NDV *t,*mi,*mj;
4926: NDV d,s,mij,mjj;
4927: ND u;
4928: NMV nmv;
1.114 noro 4929: UINT *bound;
1.99 noro 4930: PGeoBucket bucket;
4931: struct order_spec *ord;
4932:
4933: create_order_spec(0,0,&ord);
4934: nd_init_ord(ord);
4935: get_vars((Obj)f,&fv);
4936: if ( f->row != f->col )
4937: error("nd_det : non-square matrix");
4938: n = f->row;
4939: for ( nvar = 0, tv = fv; tv; tv = NEXT(tv), nvar++ );
4940: m = f->body;
4941: for ( i = 0, max = 0; i < n; i++ )
4942: for ( j = 0; j < n; j++ )
4943: for ( tv = fv; tv; tv = NEXT(tv) ) {
4944: e = getdeg(tv->v,(P)m[i][j]);
4945: max = MAX(e,max);
4946: }
1.114 noro 4947: nd_setup_parameters(nvar,max);
1.102 noro 4948: dm = (NDV **)almat_pointer(n,n);
1.99 noro 4949: for ( i = 0, max = 0; i < n; i++ )
1.102 noro 4950: for ( j = 0; j < n; j++ ) {
4951: dm[i][j] = ptondv(CO,fv,m[i][j]);
4952: if ( mod ) ndv_mod(mod,dm[i][j]);
4953: if ( dm[i][j] && !LEN(dm[i][j]) ) dm[i][j] = 0;
4954: }
4955: d = ptondv(CO,fv,(P)ONE);
4956: if ( mod ) ndv_mod(mod,d);
1.113 noro 4957: chsgnq(ONE,&mone);
1.99 noro 4958: for ( j = 0, sgn = 1; j < n; j++ ) {
1.104 noro 4959: if ( DP_Print ) fprintf(stderr,"j=%d\n",j);
1.99 noro 4960: for ( i = j; i < n && !dm[i][j]; i++ );
4961: if ( i == n ) {
4962: *rp = 0;
4963: return;
4964: }
1.102 noro 4965: k0 = i; l0 = j; len0 = LEN(dm[k0][l0]);
1.99 noro 4966: for ( k = j; k < n; k++ )
4967: for ( l = j; l < n; l++ )
1.102 noro 4968: if ( dm[k][l] && LEN(dm[k][l]) < len0 ) {
4969: k0 = k; l0 = l; len0 = LEN(dm[k][l]);
1.99 noro 4970: }
4971: if ( k0 != j ) {
4972: t = dm[j]; dm[j] = dm[k0]; dm[k0] = t;
4973: sgn = -sgn;
4974: }
4975: if ( l0 != j ) {
4976: for ( k = j; k < n; k++ ) {
4977: s = dm[k][j]; dm[k][j] = dm[k][l0]; dm[k][l0] = s;
4978: }
4979: sgn = -sgn;
4980: }
1.114 noro 4981: bound = nd_det_compute_bound(dm,n,j);
4982: if ( ndl_check_bound(bound,bound) )
4983: nd_det_reconstruct(dm,n,j,d);
4984:
1.99 noro 4985: for ( i = j+1, mj = dm[j], mjj = mj[j]; i < n; i++ ) {
1.114 noro 4986: /* if ( DP_Print ) fprintf(stderr," i=%d\n ",i); */
1.99 noro 4987: mi = dm[i]; mij = mi[j];
1.102 noro 4988: if ( mod )
4989: ndv_mul_c(mod,mij,mod-1);
4990: else
1.113 noro 4991: ndv_mul_c_q(mij,mone);
1.99 noro 4992: for ( k = j+1; k < n; k++ ) {
1.114 noro 4993: /* if ( DP_Print ) fprintf(stderr,"k=%d ",k); */
1.99 noro 4994: bucket = create_pbucket();
1.104 noro 4995: if ( mi[k] ) {
1.102 noro 4996: nmv = BDY(mjj); len = LEN(mjj);
4997: for ( a = 0; a < len; a++, NMV_ADV(nmv) ) {
4998: u = ndv_mul_nmv_trunc(mod,nmv,mi[k],DL(BDY(d)));
4999: add_pbucket(mod,bucket,u);
1.99 noro 5000: }
1.104 noro 5001: }
1.99 noro 5002: if ( mj[k] && mij ) {
1.102 noro 5003: nmv = BDY(mij); len = LEN(mij);
5004: for ( a = 0; a < len; a++, NMV_ADV(nmv) ) {
5005: u = ndv_mul_nmv_trunc(mod,nmv,mj[k],DL(BDY(d)));
5006: add_pbucket(mod,bucket,u);
1.99 noro 5007: }
5008: }
1.104 noro 5009: u = nd_quo(mod,bucket,d);
1.102 noro 5010: mi[k] = ndtondv(mod,u);
1.99 noro 5011: }
1.114 noro 5012: /* if ( DP_Print ) fprintf(stderr,"\n",k); */
1.99 noro 5013: }
5014: d = mjj;
5015: }
1.100 noro 5016: if ( sgn < 0 )
1.102 noro 5017: if ( mod )
5018: ndv_mul_c(mod,d,mod-1);
5019: else
1.113 noro 5020: ndv_mul_c_q(d,mone);
1.102 noro 5021: *rp = ndvtop(mod,CO,fv,d);
1.99 noro 5022: }
5023:
1.102 noro 5024: ND ndv_mul_nmv_trunc(int mod,NMV m0,NDV p,UINT *d)
1.99 noro 5025: {
5026: NM mr,mr0;
1.102 noro 5027: NM tnm;
5028: NMV m;
1.99 noro 5029: UINT *d0,*dt,*dm;
5030: int c,n,td,i,c1,c2,len;
1.113 noro 5031: Q q;
1.99 noro 5032: ND r;
5033:
5034: if ( !p ) return 0;
5035: else {
1.102 noro 5036: n = NV(p); m = BDY(p); len = LEN(p);
1.99 noro 5037: d0 = DL(m0);
1.102 noro 5038: td = TD(d);
1.99 noro 5039: mr0 = 0;
5040: NEWNM(tnm);
1.102 noro 5041: if ( mod ) {
5042: c = CM(m0);
5043: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
5044: ndl_add(DL(m),d0,DL(tnm));
5045: if ( ndl_reducible(DL(tnm),d) ) {
5046: NEXTNM(mr0,mr);
5047: c1 = CM(m); DMAR(c1,c,0,mod,c2); CM(mr) = c2;
1.104 noro 5048: ndl_copy(DL(tnm),DL(mr));
1.102 noro 5049: }
5050: }
5051: } else {
1.113 noro 5052: q = CQ(m0);
1.102 noro 5053: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
5054: ndl_add(DL(m),d0,DL(tnm));
5055: if ( ndl_reducible(DL(tnm),d) ) {
5056: NEXTNM(mr0,mr);
1.113 noro 5057: mulq(CQ(m),q,&CQ(mr));
1.104 noro 5058: ndl_copy(DL(tnm),DL(mr));
1.102 noro 5059: }
1.99 noro 5060: }
5061: }
5062: if ( !mr0 )
5063: return 0;
5064: else {
5065: NEXT(mr) = 0;
1.105 noro 5066: for ( len = 0, mr = mr0; mr; mr = NEXT(mr), len++ );
1.99 noro 5067: MKND(NV(p),mr0,len,r);
5068: SG(r) = SG(p) + TD(d0);
5069: return r;
5070: }
5071: }
1.114 noro 5072: }
5073:
5074: void nd_det_reconstruct(NDV **dm,int n,int j,NDV d)
5075: {
5076: int i,obpe,oadv,h,k,l;
5077: static NM prev_nm_free_list;
5078: EPOS oepos;
5079:
5080: obpe = nd_bpe;
5081: oadv = nmv_adv;
5082: oepos = nd_epos;
5083: if ( obpe < 2 ) nd_bpe = 2;
5084: else if ( obpe < 3 ) nd_bpe = 3;
5085: else if ( obpe < 4 ) nd_bpe = 4;
5086: else if ( obpe < 5 ) nd_bpe = 5;
5087: else if ( obpe < 6 ) nd_bpe = 6;
5088: else if ( obpe < 8 ) nd_bpe = 8;
5089: else if ( obpe < 10 ) nd_bpe = 10;
5090: else if ( obpe < 16 ) nd_bpe = 16;
5091: else if ( obpe < 32 ) nd_bpe = 32;
5092: else error("nd_det_reconstruct : exponent too large");
5093:
5094: nd_setup_parameters(nd_nvar,0);
5095: prev_nm_free_list = _nm_free_list;
5096: _nm_free_list = 0;
5097: for ( k = j; k < n; k++ )
5098: for (l = j; l < n; l++ )
5099: ndv_realloc(dm[k][l],obpe,oadv,oepos);
5100: ndv_realloc(d,obpe,oadv,oepos);
5101: prev_nm_free_list = 0;
5102: #if 0
5103: GC_gcollect();
5104: #endif
5105: }
5106:
5107: UINT *nd_det_compute_bound(NDV **dm,int n,int j)
5108: {
5109: UINT *d0,*d1,*d,*t,*r;
5110: int k,l;
5111:
5112: d0 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
5113: d1 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
5114: for ( k = 0; k < nd_wpd; k++ ) d0[k] = 0;
5115: for ( k = j; k < n; k++ )
5116: for ( l = j; l < n; l++ )
5117: if ( dm[k][l] ) {
5118: d = ndv_compute_bound(dm[k][l]);
5119: ndl_lcm(d,d0,d1);
5120: t = d1; d1 = d0; d0 = t;
5121: }
5122: r = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
5123: for ( k = 0; k < nd_wpd; k++ ) r[k] = d0[k];
5124: return r;
1.117 noro 5125: }
5126:
5127: DL nd_separate_d(UINT *d,UINT *trans)
5128: {
5129: int n,ntrans,td,i,e;
5130: DL a;
5131:
5132: n = nd_nvar; ntrans = n-nd_nalg;
5133: ndl_zero(trans);
5134: td = 0;
5135: for ( i = 0; i < ntrans; i++ ) {
5136: e = GET_EXP(d,i);
5137: PUT_EXP(trans,i,e);
5138: td += MUL_WEIGHT(e,i);
5139: }
5140: TD(trans) = td;
5141: if ( nd_blockmask) ndl_weight_mask(trans);
5142: NEWDL(a,nd_nalg);
5143: td = 0;
5144: for ( ; i < n; i++ ) {
5145: e = GET_EXP(d,i);
5146: a->d[i-ntrans] = e;
5147: td += e;
5148: }
5149: a->td = td;
5150: return a;
5151: }
5152:
1.118 ! noro 5153: NumberField get_numberfield();
! 5154:
! 5155: int nd_monic(int mod,ND *p)
1.117 noro 5156: {
5157: UINT *trans,*t;
5158: DL alg;
5159: MP mp0,mp;
1.118 ! noro 5160: NM m,m0,m1,ma0,ma,mb,mr0,mr;
! 5161: ND r;
1.117 noro 5162: DL dl;
5163: DP nm;
5164: NDV ndv;
1.118 ! noro 5165: DAlg inv,cd;
1.117 noro 5166: ND s,c;
1.118 ! noro 5167: Q l,mul;
! 5168: N ln;
! 5169: int n,ntrans,i,e,td,is_lc,len;
! 5170: NumberField nf;
! 5171: struct oEGT eg0,eg1;
! 5172:
! 5173: if ( !(nf = get_numberfield()) )
! 5174: error("nd_monic : current_numberfield is not set");
1.117 noro 5175:
5176: n = nd_nvar; ntrans = n-nd_nalg;
1.118 ! noro 5177: /* Q coef -> DAlg coef */
! 5178: NEWNM(ma0); ma = ma0;
! 5179: m = BDY(*p);
! 5180: is_lc = 1;
! 5181: while ( 1 ) {
! 5182: NEWMP(mp0); mp = mp0;
! 5183: mp->c = (P)CQ(m);
! 5184: mp->dl = nd_separate_d(DL(m),DL(ma));
! 5185: NEWNM(mb);
! 5186: for ( m = NEXT(m); m; m = NEXT(m) ) {
! 5187: alg = nd_separate_d(DL(m),DL(mb));
! 5188: if ( !ndl_equal(DL(ma),DL(mb)) )
! 5189: break;
! 5190: NEXTMP(mp0,mp); mp->c = (P)CQ(m); mp->dl = alg;
! 5191: }
! 5192: NEXT(mp) = 0;
! 5193: MKDP(nd_nalg,mp0,nm);
! 5194: MKDAlg(nm,ONE,cd);
! 5195: if ( is_lc == 1 ) {
! 5196: /* if the lc is a rational number, we have nothing to do */
! 5197: if ( !mp0->dl->td )
! 5198: return 1;
! 5199:
! 5200: get_eg(&eg0);
! 5201: invdalg(cd,&inv);
! 5202: get_eg(&eg1); add_eg(&eg_invdalg,&eg0,&eg1);
! 5203: /* check the validity of inv */
! 5204: if ( mod && !rem(NM(inv->dn),mod) )
! 5205: return 0;
! 5206: CA(ma) = nf->one;
! 5207: is_lc = 0;
! 5208: ln = ONEN;
! 5209: } else {
! 5210: muldalg(cd,inv,&CA(ma));
! 5211: lcmn(ln,NM(CA(ma)->dn),&ln);
! 5212: }
! 5213: if ( m ) {
! 5214: NEXT(ma) = mb; ma = mb;
! 5215: } else {
! 5216: NEXT(ma) = 0;
! 5217: break;
! 5218: }
! 5219: }
! 5220: /* l = lcm(denoms) */
! 5221: NTOQ(ln,1,l);
! 5222: for ( mr0 = 0, m = ma0; m; m = NEXT(m) ) {
! 5223: divq(l,CA(m)->dn,&mul);
! 5224: for ( mp = BDY(CA(m)->nm); mp; mp = NEXT(mp) ) {
! 5225: NEXTNM(mr0,mr);
! 5226: mulq((Q)mp->c,mul,&CQ(mr));
! 5227: dl = mp->dl;
! 5228: td = TD(DL(m));
! 5229: ndl_copy(DL(m),DL(mr));
! 5230: for ( i = ntrans; i < n; i++ ) {
! 5231: e = dl->d[i-ntrans];
! 5232: PUT_EXP(DL(mr),i,e);
! 5233: td += MUL_WEIGHT(e,i);
! 5234: }
! 5235: TD(DL(mr)) = td;
! 5236: if ( nd_blockmask) ndl_weight_mask(DL(mr));
! 5237: }
1.117 noro 5238: }
1.118 ! noro 5239: NEXT(mr) = 0;
! 5240: for ( len = 0, mr = mr0; mr; mr = NEXT(mr), len++ );
! 5241: MKND(NV(*p),mr0,len,r);
! 5242: /* XXX */
! 5243: SG(r) = SG(*p);
! 5244: nd_free(*p);
! 5245: *p = r;
! 5246: return 1;
! 5247: }
! 5248:
! 5249: void nd_set_nalg(int nalg)
! 5250: {
! 5251: nd_nalg = nalg;
1.59 noro 5252: }
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>