Annotation of OpenXM_contrib2/asir2000/engine/nd.c, Revision 1.144
1.144 ! noro 1: /* $OpenXM: OpenXM_contrib2/asir2000/engine/nd.c,v 1.143 2006/08/09 02:48:49 noro Exp $ */
1.2 noro 2:
1.94 noro 3: #include "nd.h"
1.63 noro 4:
1.131 noro 5: int diag_period = 6;
1.61 noro 6: int (*ndl_compare_function)(UINT *a1,UINT *a2);
1.94 noro 7: int nd_dcomp;
8: NM _nm_free_list;
9: ND _nd_free_list;
10: ND_pairs _ndp_free_list;
1.32 noro 11:
1.121 noro 12: static int nd_ntrans;
1.117 noro 13: static int nd_nalg;
1.103 noro 14: #if 0
1.74 noro 15: static int ndv_alloc;
1.103 noro 16: #endif
1.87 noro 17: #if 1
1.69 noro 18: static int nd_f4_nsp=0x7fffffff;
1.87 noro 19: #else
20: static int nd_f4_nsp=50;
21: #endif
1.42 noro 22: static double nd_scale=2;
1.61 noro 23: static UINT **nd_bound;
1.42 noro 24: static struct order_spec *nd_ord;
25: static EPOS nd_epos;
1.43 noro 26: static BlockMask nd_blockmask;
1.42 noro 27: static int nd_nvar;
28: static int nd_isrlex;
29: static int nd_epw,nd_bpe,nd_wpd,nd_exporigin;
1.61 noro 30: static UINT nd_mask[32];
31: static UINT nd_mask0,nd_mask1;
1.42 noro 32:
1.20 noro 33: static NDV *nd_ps;
1.53 noro 34: static NDV *nd_ps_trace;
1.42 noro 35: static RHist *nd_psh;
36: static int nd_psn,nd_pslen;
37: static RHist *nd_red;
1.96 noro 38: static int *nd_work_vector;
39: static int **nd_matrix;
40: static int nd_matrix_len;
1.97 noro 41: static struct weight_or_block *nd_worb;
42: static int nd_worb_len;
1.42 noro 43: static int nd_found,nd_create,nd_notfirst;
44: static int nmv_adv;
1.77 noro 45: static int nd_demand;
1.1 noro 46:
1.119 noro 47: NumberField get_numberfield();
1.114 noro 48: UINT *nd_det_compute_bound(NDV **dm,int n,int j);
49: void nd_det_reconstruct(NDV **dm,int n,int j,NDV d);
1.118 noro 50: int nd_monic(int m,ND *p);
1.129 noro 51: NDV plain_vect_to_ndv_q(Q *mat,int col,UINT *s0vect);
1.114 noro 52:
1.1 noro 53: void nd_free_private_storage()
54: {
55: _nm_free_list = 0;
1.5 noro 56: _ndp_free_list = 0;
1.71 noro 57: #if 0
1.1 noro 58: GC_gcollect();
1.71 noro 59: #endif
1.1 noro 60: }
61:
62: void _NM_alloc()
63: {
64: NM p;
65: int i;
66:
1.11 noro 67: for ( i = 0; i < 1024; i++ ) {
1.61 noro 68: p = (NM)GC_malloc(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT));
1.1 noro 69: p->next = _nm_free_list; _nm_free_list = p;
70: }
71: }
72:
73: void _ND_alloc()
74: {
75: ND p;
76: int i;
77:
78: for ( i = 0; i < 1024; i++ ) {
79: p = (ND)GC_malloc(sizeof(struct oND));
80: p->body = (NM)_nd_free_list; _nd_free_list = p;
81: }
82: }
83:
84: void _NDP_alloc()
85: {
86: ND_pairs p;
87: int i;
88:
1.11 noro 89: for ( i = 0; i < 1024; i++ ) {
1.1 noro 90: p = (ND_pairs)GC_malloc(sizeof(struct oND_pairs)
1.61 noro 91: +(nd_wpd-1)*sizeof(UINT));
1.1 noro 92: p->next = _ndp_free_list; _ndp_free_list = p;
93: }
94: }
95:
1.30 noro 96: INLINE int nd_length(ND p)
1.1 noro 97: {
98: NM m;
99: int i;
100:
101: if ( !p )
102: return 0;
103: else {
104: for ( i = 0, m = BDY(p); m; m = NEXT(m), i++ );
105: return i;
106: }
107: }
108:
1.61 noro 109: INLINE int ndl_reducible(UINT *d1,UINT *d2)
1.1 noro 110: {
1.61 noro 111: UINT u1,u2;
1.1 noro 112: int i,j;
113:
1.34 noro 114: if ( TD(d1) < TD(d2) ) return 0;
1.65 noro 115: #if USE_UNROLL
1.1 noro 116: switch ( nd_bpe ) {
1.62 noro 117: case 3:
118: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
119: u1 = d1[i]; u2 = d2[i];
120: if ( (u1&0x38000000) < (u2&0x38000000) ) return 0;
121: if ( (u1& 0x7000000) < (u2& 0x7000000) ) return 0;
122: if ( (u1& 0xe00000) < (u2& 0xe00000) ) return 0;
123: if ( (u1& 0x1c0000) < (u2& 0x1c0000) ) return 0;
124: if ( (u1& 0x38000) < (u2& 0x38000) ) return 0;
125: if ( (u1& 0x7000) < (u2& 0x7000) ) return 0;
126: if ( (u1& 0xe00) < (u2& 0xe00) ) return 0;
127: if ( (u1& 0x1c0) < (u2& 0x1c0) ) return 0;
128: if ( (u1& 0x38) < (u2& 0x38) ) return 0;
129: if ( (u1& 0x7) < (u2& 0x7) ) return 0;
130: }
131: return 1;
132: break;
1.1 noro 133: case 4:
1.41 noro 134: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 135: u1 = d1[i]; u2 = d2[i];
136: if ( (u1&0xf0000000) < (u2&0xf0000000) ) return 0;
1.62 noro 137: if ( (u1& 0xf000000) < (u2& 0xf000000) ) return 0;
138: if ( (u1& 0xf00000) < (u2& 0xf00000) ) return 0;
139: if ( (u1& 0xf0000) < (u2& 0xf0000) ) return 0;
140: if ( (u1& 0xf000) < (u2& 0xf000) ) return 0;
141: if ( (u1& 0xf00) < (u2& 0xf00) ) return 0;
142: if ( (u1& 0xf0) < (u2& 0xf0) ) return 0;
143: if ( (u1& 0xf) < (u2& 0xf) ) return 0;
1.1 noro 144: }
145: return 1;
146: break;
147: case 6:
1.41 noro 148: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 149: u1 = d1[i]; u2 = d2[i];
150: if ( (u1&0x3f000000) < (u2&0x3f000000) ) return 0;
1.62 noro 151: if ( (u1& 0xfc0000) < (u2& 0xfc0000) ) return 0;
152: if ( (u1& 0x3f000) < (u2& 0x3f000) ) return 0;
153: if ( (u1& 0xfc0) < (u2& 0xfc0) ) return 0;
154: if ( (u1& 0x3f) < (u2& 0x3f) ) return 0;
1.1 noro 155: }
156: return 1;
157: break;
158: case 8:
1.41 noro 159: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 160: u1 = d1[i]; u2 = d2[i];
161: if ( (u1&0xff000000) < (u2&0xff000000) ) return 0;
1.62 noro 162: if ( (u1& 0xff0000) < (u2& 0xff0000) ) return 0;
163: if ( (u1& 0xff00) < (u2& 0xff00) ) return 0;
164: if ( (u1& 0xff) < (u2& 0xff) ) return 0;
1.1 noro 165: }
166: return 1;
167: break;
168: case 16:
1.41 noro 169: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 170: u1 = d1[i]; u2 = d2[i];
171: if ( (u1&0xffff0000) < (u2&0xffff0000) ) return 0;
1.62 noro 172: if ( (u1& 0xffff) < (u2& 0xffff) ) return 0;
1.1 noro 173: }
174: return 1;
175: break;
176: case 32:
1.41 noro 177: for ( i = nd_exporigin; i < nd_wpd; i++ )
1.1 noro 178: if ( d1[i] < d2[i] ) return 0;
179: return 1;
180: break;
181: default:
1.41 noro 182: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 183: u1 = d1[i]; u2 = d2[i];
184: for ( j = 0; j < nd_epw; j++ )
185: if ( (u1&nd_mask[j]) < (u2&nd_mask[j]) ) return 0;
186: }
187: return 1;
188: }
1.65 noro 189: #else
190: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
191: u1 = d1[i]; u2 = d2[i];
192: for ( j = 0; j < nd_epw; j++ )
193: if ( (u1&nd_mask[j]) < (u2&nd_mask[j]) ) return 0;
194: }
195: return 1;
196: #endif
1.1 noro 197: }
198:
1.61 noro 199: /*
200: * If the current order is a block order,
201: * then the last block is length 1 and contains
202: * the homo variable. Otherwise, the original
203: * order is either 0 or 2.
204: */
205:
206: void ndl_homogenize(UINT *d,UINT *r,int obpe,EPOS oepos,int weight)
1.23 noro 207: {
1.61 noro 208: int w,i,e,n,omask0;
209:
210: omask0 = (1<<obpe)-1;
211: n = nd_nvar-1;
212: ndl_zero(r);
213: for ( i = 0; i < n; i++ ) {
214: e = GET_EXP_OLD(d,i);
215: PUT_EXP(r,i,e);
216: }
217: w = TD(d);
218: PUT_EXP(r,nd_nvar-1,weight-w);
219: TD(r) = weight;
220: if ( nd_blockmask ) ndl_weight_mask(r);
221: }
222:
223: void ndl_dehomogenize(UINT *d)
224: {
225: UINT mask;
226: UINT h;
1.31 noro 227: int i,bits;
1.23 noro 228:
1.44 noro 229: if ( nd_blockmask ) {
230: h = GET_EXP(d,nd_nvar-1);
1.45 noro 231: XOR_EXP(d,nd_nvar-1,h);
1.44 noro 232: TD(d) -= h;
233: d[nd_exporigin-1] -= h;
234: } else {
235: if ( nd_isrlex ) {
236: if ( nd_bpe == 32 ) {
237: h = d[nd_exporigin];
238: for ( i = nd_exporigin+1; i < nd_wpd; i++ )
239: d[i-1] = d[i];
240: d[i-1] = 0;
241: TD(d) -= h;
242: } else {
243: bits = nd_epw*nd_bpe;
244: mask = bits==32?0xffffffff:((1<<(nd_epw*nd_bpe))-1);
245: h = (d[nd_exporigin]>>((nd_epw-1)*nd_bpe))&nd_mask0;
246: for ( i = nd_exporigin; i < nd_wpd; i++ )
247: d[i] = ((d[i]<<nd_bpe)&mask)
248: |(i+1<nd_wpd?((d[i+1]>>((nd_epw-1)*nd_bpe))&nd_mask0):0);
249: TD(d) -= h;
250: }
1.45 noro 251: } else {
252: h = GET_EXP(d,nd_nvar-1);
253: XOR_EXP(d,nd_nvar-1,h);
254: TD(d) -= h;
255: }
1.44 noro 256: }
1.23 noro 257: }
258:
1.61 noro 259: void ndl_lcm(UINT *d1,unsigned *d2,UINT *d)
1.1 noro 260: {
1.61 noro 261: UINT t1,t2,u,u1,u2;
1.43 noro 262: int i,j,l;
1.1 noro 263:
1.65 noro 264: #if USE_UNROLL
1.1 noro 265: switch ( nd_bpe ) {
1.62 noro 266: case 3:
267: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
268: u1 = d1[i]; u2 = d2[i];
269: t1 = (u1&0x38000000); t2 = (u2&0x38000000); u = t1>t2?t1:t2;
270: t1 = (u1& 0x7000000); t2 = (u2& 0x7000000); u |= t1>t2?t1:t2;
271: t1 = (u1& 0xe00000); t2 = (u2& 0xe00000); u |= t1>t2?t1:t2;
272: t1 = (u1& 0x1c0000); t2 = (u2& 0x1c0000); u |= t1>t2?t1:t2;
273: t1 = (u1& 0x38000); t2 = (u2& 0x38000); u |= t1>t2?t1:t2;
274: t1 = (u1& 0x7000); t2 = (u2& 0x7000); u |= t1>t2?t1:t2;
275: t1 = (u1& 0xe00); t2 = (u2& 0xe00); u |= t1>t2?t1:t2;
276: t1 = (u1& 0x1c0); t2 = (u2& 0x1c0); u |= t1>t2?t1:t2;
277: t1 = (u1& 0x38); t2 = (u2& 0x38); u |= t1>t2?t1:t2;
278: t1 = (u1& 0x7); t2 = (u2& 0x7); u |= t1>t2?t1:t2;
279: d[i] = u;
280: }
281: break;
1.1 noro 282: case 4:
1.41 noro 283: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 284: u1 = d1[i]; u2 = d2[i];
285: t1 = (u1&0xf0000000); t2 = (u2&0xf0000000); u = t1>t2?t1:t2;
1.62 noro 286: t1 = (u1& 0xf000000); t2 = (u2& 0xf000000); u |= t1>t2?t1:t2;
287: t1 = (u1& 0xf00000); t2 = (u2& 0xf00000); u |= t1>t2?t1:t2;
288: t1 = (u1& 0xf0000); t2 = (u2& 0xf0000); u |= t1>t2?t1:t2;
289: t1 = (u1& 0xf000); t2 = (u2& 0xf000); u |= t1>t2?t1:t2;
290: t1 = (u1& 0xf00); t2 = (u2& 0xf00); u |= t1>t2?t1:t2;
291: t1 = (u1& 0xf0); t2 = (u2& 0xf0); u |= t1>t2?t1:t2;
292: t1 = (u1& 0xf); t2 = (u2& 0xf); u |= t1>t2?t1:t2;
1.1 noro 293: d[i] = u;
294: }
295: break;
296: case 6:
1.41 noro 297: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 298: u1 = d1[i]; u2 = d2[i];
299: t1 = (u1&0x3f000000); t2 = (u2&0x3f000000); u = t1>t2?t1:t2;
1.62 noro 300: t1 = (u1& 0xfc0000); t2 = (u2& 0xfc0000); u |= t1>t2?t1:t2;
301: t1 = (u1& 0x3f000); t2 = (u2& 0x3f000); u |= t1>t2?t1:t2;
302: t1 = (u1& 0xfc0); t2 = (u2& 0xfc0); u |= t1>t2?t1:t2;
303: t1 = (u1& 0x3f); t2 = (u2& 0x3f); u |= t1>t2?t1:t2;
1.1 noro 304: d[i] = u;
305: }
306: break;
307: case 8:
1.41 noro 308: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 309: u1 = d1[i]; u2 = d2[i];
310: t1 = (u1&0xff000000); t2 = (u2&0xff000000); u = t1>t2?t1:t2;
1.62 noro 311: t1 = (u1& 0xff0000); t2 = (u2& 0xff0000); u |= t1>t2?t1:t2;
312: t1 = (u1& 0xff00); t2 = (u2& 0xff00); u |= t1>t2?t1:t2;
313: t1 = (u1& 0xff); t2 = (u2& 0xff); u |= t1>t2?t1:t2;
1.1 noro 314: d[i] = u;
315: }
316: break;
317: case 16:
1.41 noro 318: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 319: u1 = d1[i]; u2 = d2[i];
320: t1 = (u1&0xffff0000); t2 = (u2&0xffff0000); u = t1>t2?t1:t2;
1.62 noro 321: t1 = (u1& 0xffff); t2 = (u2& 0xffff); u |= t1>t2?t1:t2;
1.1 noro 322: d[i] = u;
323: }
324: break;
325: case 32:
1.41 noro 326: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 327: u1 = d1[i]; u2 = d2[i];
328: d[i] = u1>u2?u1:u2;
329: }
330: break;
331: default:
1.41 noro 332: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 333: u1 = d1[i]; u2 = d2[i];
334: for ( j = 0, u = 0; j < nd_epw; j++ ) {
335: t1 = (u1&nd_mask[j]); t2 = (u2&nd_mask[j]); u |= t1>t2?t1:t2;
336: }
337: d[i] = u;
338: }
339: break;
340: }
1.65 noro 341: #else
342: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
343: u1 = d1[i]; u2 = d2[i];
344: for ( j = 0, u = 0; j < nd_epw; j++ ) {
345: t1 = (u1&nd_mask[j]); t2 = (u2&nd_mask[j]); u |= t1>t2?t1:t2;
346: }
347: d[i] = u;
348: }
349: #endif
1.39 noro 350: TD(d) = ndl_weight(d);
1.61 noro 351: if ( nd_blockmask ) ndl_weight_mask(d);
1.57 noro 352: }
353:
1.61 noro 354: int ndl_weight(UINT *d)
1.1 noro 355: {
1.61 noro 356: UINT t,u;
1.1 noro 357: int i,j;
358:
1.60 noro 359: if ( current_dl_weight_vector )
360: for ( i = 0, t = 0; i < nd_nvar; i++ ) {
361: u = GET_EXP(d,i);
362: t += MUL_WEIGHT(u,i);
363: }
364: else
365: for ( t = 0, i = nd_exporigin; i < nd_wpd; i++ ) {
366: u = d[i];
367: for ( j = 0; j < nd_epw; j++, u>>=nd_bpe )
368: t += (u&nd_mask0);
369: }
1.1 noro 370: return t;
371: }
372:
1.61 noro 373: void ndl_weight_mask(UINT *d)
1.43 noro 374: {
1.61 noro 375: UINT t,u;
376: UINT *mask;
377: int i,j,k,l;
1.43 noro 378:
1.61 noro 379: l = nd_blockmask->n;
380: for ( k = 0; k < l; k++ ) {
381: mask = nd_blockmask->mask[k];
382: if ( current_dl_weight_vector )
383: for ( i = 0, t = 0; i < nd_nvar; i++ ) {
384: u = GET_EXP_MASK(d,i,mask);
385: t += MUL_WEIGHT(u,i);
386: }
387: else
388: for ( t = 0, i = nd_exporigin; i < nd_wpd; i++ ) {
389: u = d[i]&mask[i];
390: for ( j = 0; j < nd_epw; j++, u>>=nd_bpe )
391: t += (u&nd_mask0);
392: }
393: d[k+1] = t;
394: }
1.43 noro 395: }
396:
1.61 noro 397: int ndl_lex_compare(UINT *d1,UINT *d2)
1.1 noro 398: {
399: int i;
400:
1.41 noro 401: d1 += nd_exporigin;
402: d2 += nd_exporigin;
403: for ( i = nd_exporigin; i < nd_wpd; i++, d1++, d2++ )
1.1 noro 404: if ( *d1 > *d2 )
1.32 noro 405: return nd_isrlex ? -1 : 1;
1.1 noro 406: else if ( *d1 < *d2 )
1.32 noro 407: return nd_isrlex ? 1 : -1;
1.1 noro 408: return 0;
409: }
410:
1.61 noro 411: int ndl_block_compare(UINT *d1,UINT *d2)
1.43 noro 412: {
413: int i,l,j,ord_o,ord_l;
414: struct order_pair *op;
1.61 noro 415: UINT t1,t2,m;
416: UINT *mask;
1.43 noro 417:
418: l = nd_blockmask->n;
419: op = nd_blockmask->order_pair;
420: for ( j = 0; j < l; j++ ) {
421: mask = nd_blockmask->mask[j];
422: ord_o = op[j].order;
423: if ( ord_o < 2 )
1.44 noro 424: if ( (t1=d1[j+1]) > (t2=d2[j+1]) ) return 1;
425: else if ( t1 < t2 ) return -1;
1.43 noro 426: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.44 noro 427: m = mask[i];
428: t1 = d1[i]&m;
429: t2 = d2[i]&m;
1.43 noro 430: if ( t1 > t2 )
431: return !ord_o ? -1 : 1;
432: else if ( t1 < t2 )
433: return !ord_o ? 1 : -1;
434: }
435: }
436: return 0;
437: }
438:
1.96 noro 439: int ndl_matrix_compare(UINT *d1,UINT *d2)
440: {
441: int i,j,s;
442: int *v;
443:
444: for ( j = 0; j < nd_nvar; j++ )
445: nd_work_vector[j] = GET_EXP(d1,j)-GET_EXP(d2,j);
446: for ( i = 0; i < nd_matrix_len; i++ ) {
447: v = nd_matrix[i];
448: for ( j = 0, s = 0; j < nd_nvar; j++ )
449: s += v[j]*nd_work_vector[j];
450: if ( s > 0 ) return 1;
451: else if ( s < 0 ) return -1;
452: }
453: return 0;
454: }
455:
1.97 noro 456: int ndl_composite_compare(UINT *d1,UINT *d2)
457: {
458: int i,j,s,start,end,len,o;
459: int *v;
460: struct sparse_weight *sw;
461:
462: for ( j = 0; j < nd_nvar; j++ )
463: nd_work_vector[j] = GET_EXP(d1,j)-GET_EXP(d2,j);
464: for ( i = 0; i < nd_worb_len; i++ ) {
465: len = nd_worb[i].length;
466: switch ( nd_worb[i].type ) {
467: case IS_DENSE_WEIGHT:
468: v = nd_worb[i].body.dense_weight;
469: for ( j = 0, s = 0; j < len; j++ )
470: s += v[j]*nd_work_vector[j];
471: if ( s > 0 ) return 1;
472: else if ( s < 0 ) return -1;
473: break;
474: case IS_SPARSE_WEIGHT:
475: sw = nd_worb[i].body.sparse_weight;
476: for ( j = 0, s = 0; j < len; j++ )
477: s += sw[j].value*nd_work_vector[sw[j].pos];
478: if ( s > 0 ) return 1;
479: else if ( s < 0 ) return -1;
480: break;
481: case IS_BLOCK:
482: o = nd_worb[i].body.block.order;
483: start = nd_worb[i].body.block.start;
484: switch ( o ) {
485: case 0:
486: end = start+len;
487: for ( j = start, s = 0; j < end; j++ )
488: s += MUL_WEIGHT(nd_work_vector[j],j);
489: if ( s > 0 ) return 1;
490: else if ( s < 0 ) return -1;
491: for ( j = end-1; j >= start; j-- )
492: if ( nd_work_vector[j] < 0 ) return 1;
493: else if ( nd_work_vector[j] > 0 ) return -1;
494: break;
495: case 1:
496: end = start+len;
497: for ( j = start, s = 0; j < end; j++ )
498: s += MUL_WEIGHT(nd_work_vector[j],j);
499: if ( s > 0 ) return 1;
500: else if ( s < 0 ) return -1;
501: for ( j = start; j < end; j++ )
502: if ( nd_work_vector[j] > 0 ) return 1;
503: else if ( nd_work_vector[j] < 0 ) return -1;
504: break;
505: case 2:
506: for ( j = start; j < end; j++ )
507: if ( nd_work_vector[j] > 0 ) return 1;
508: else if ( nd_work_vector[j] < 0 ) return -1;
509: break;
510: }
511: break;
512: }
513: }
514: return 0;
515: }
516:
1.58 noro 517: /* TDH -> WW -> TD-> RL */
518:
1.61 noro 519: int ndl_ww_lex_compare(UINT *d1,UINT *d2)
1.58 noro 520: {
521: int i,m,e1,e2;
522:
523: if ( TD(d1) > TD(d2) ) return 1;
524: else if ( TD(d1) < TD(d2) ) return -1;
525: m = nd_nvar>>1;
526: for ( i = 0, e1 = e2 = 0; i < m; i++ ) {
527: e1 += current_weyl_weight_vector[i]*(GET_EXP(d1,m+i)-GET_EXP(d1,i));
528: e2 += current_weyl_weight_vector[i]*(GET_EXP(d2,m+i)-GET_EXP(d2,i));
529: }
530: if ( e1 > e2 ) return 1;
531: else if ( e1 < e2 ) return -1;
532: return ndl_lex_compare(d1,d2);
533: }
534:
1.61 noro 535: INLINE int ndl_equal(UINT *d1,UINT *d2)
1.1 noro 536: {
537: int i;
538:
1.81 noro 539: switch ( nd_wpd ) {
540: case 2:
541: if ( TD(d2) != TD(d1) ) return 0;
542: if ( d2[1] != d1[1] ) return 0;
543: return 1;
544: break;
545: case 3:
546: if ( TD(d2) != TD(d1) ) return 0;
547: if ( d2[1] != d1[1] ) return 0;
548: if ( d2[2] != d1[2] ) return 0;
549: return 1;
550: break;
551: default:
552: for ( i = 0; i < nd_wpd; i++ )
553: if ( *d1++ != *d2++ ) return 0;
554: return 1;
555: break;
556: }
1.1 noro 557: }
558:
1.61 noro 559: INLINE void ndl_copy(UINT *d1,UINT *d2)
1.6 noro 560: {
561: int i;
562:
563: switch ( nd_wpd ) {
1.41 noro 564: case 2:
1.34 noro 565: TD(d2) = TD(d1);
566: d2[1] = d1[1];
1.6 noro 567: break;
1.41 noro 568: case 3:
1.34 noro 569: TD(d2) = TD(d1);
1.6 noro 570: d2[1] = d1[1];
1.34 noro 571: d2[2] = d1[2];
1.6 noro 572: break;
573: default:
1.41 noro 574: for ( i = 0; i < nd_wpd; i++ )
1.6 noro 575: d2[i] = d1[i];
576: break;
577: }
578: }
579:
1.61 noro 580: INLINE void ndl_zero(UINT *d)
581: {
582: int i;
583: for ( i = 0; i < nd_wpd; i++ ) d[i] = 0;
584: }
585:
586: INLINE void ndl_add(UINT *d1,UINT *d2,UINT *d)
1.1 noro 587: {
588: int i;
589:
1.43 noro 590: #if 1
1.6 noro 591: switch ( nd_wpd ) {
1.41 noro 592: case 2:
593: TD(d) = TD(d1)+TD(d2);
1.34 noro 594: d[1] = d1[1]+d2[1];
1.6 noro 595: break;
1.41 noro 596: case 3:
597: TD(d) = TD(d1)+TD(d2);
1.6 noro 598: d[1] = d1[1]+d2[1];
1.34 noro 599: d[2] = d1[2]+d2[2];
1.6 noro 600: break;
601: default:
1.43 noro 602: for ( i = 0; i < nd_wpd; i++ ) d[i] = d1[i]+d2[i];
1.6 noro 603: break;
604: }
1.43 noro 605: #else
606: for ( i = 0; i < nd_wpd; i++ ) d[i] = d1[i]+d2[i];
607: #endif
1.6 noro 608: }
609:
1.55 noro 610: /* d1 += d2 */
1.61 noro 611: INLINE void ndl_addto(UINT *d1,UINT *d2)
1.55 noro 612: {
613: int i;
614:
615: #if 1
616: switch ( nd_wpd ) {
617: case 2:
618: TD(d1) += TD(d2);
619: d1[1] += d2[1];
620: break;
621: case 3:
622: TD(d1) += TD(d2);
623: d1[1] += d2[1];
624: d1[2] += d2[2];
625: break;
626: default:
627: for ( i = 0; i < nd_wpd; i++ ) d1[i] += d2[i];
628: break;
629: }
630: #else
631: for ( i = 0; i < nd_wpd; i++ ) d1[i] += d2[i];
632: #endif
633: }
634:
1.61 noro 635: INLINE void ndl_sub(UINT *d1,UINT *d2,UINT *d)
1.6 noro 636: {
637: int i;
638:
1.43 noro 639: for ( i = 0; i < nd_wpd; i++ ) d[i] = d1[i]-d2[i];
1.1 noro 640: }
641:
1.61 noro 642: int ndl_disjoint(UINT *d1,UINT *d2)
1.1 noro 643: {
1.61 noro 644: UINT t1,t2,u,u1,u2;
1.1 noro 645: int i,j;
646:
1.65 noro 647: #if USE_UNROLL
1.1 noro 648: switch ( nd_bpe ) {
1.62 noro 649: case 3:
650: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
651: u1 = d1[i]; u2 = d2[i];
652: t1 = u1&0x38000000; t2 = u2&0x38000000; if ( t1&&t2 ) return 0;
653: t1 = u1& 0x7000000; t2 = u2& 0x7000000; if ( t1&&t2 ) return 0;
654: t1 = u1& 0xe00000; t2 = u2& 0xe00000; if ( t1&&t2 ) return 0;
655: t1 = u1& 0x1c0000; t2 = u2& 0x1c0000; if ( t1&&t2 ) return 0;
656: t1 = u1& 0x38000; t2 = u2& 0x38000; if ( t1&&t2 ) return 0;
657: t1 = u1& 0x7000; t2 = u2& 0x7000; if ( t1&&t2 ) return 0;
658: t1 = u1& 0xe00; t2 = u2& 0xe00; if ( t1&&t2 ) return 0;
659: t1 = u1& 0x1c0; t2 = u2& 0x1c0; if ( t1&&t2 ) return 0;
660: t1 = u1& 0x38; t2 = u2& 0x38; if ( t1&&t2 ) return 0;
661: t1 = u1& 0x7; t2 = u2& 0x7; if ( t1&&t2 ) return 0;
662: }
663: return 1;
664: break;
1.1 noro 665: case 4:
1.41 noro 666: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 667: u1 = d1[i]; u2 = d2[i];
668: t1 = u1&0xf0000000; t2 = u2&0xf0000000; if ( t1&&t2 ) return 0;
1.62 noro 669: t1 = u1& 0xf000000; t2 = u2& 0xf000000; if ( t1&&t2 ) return 0;
670: t1 = u1& 0xf00000; t2 = u2& 0xf00000; if ( t1&&t2 ) return 0;
671: t1 = u1& 0xf0000; t2 = u2& 0xf0000; if ( t1&&t2 ) return 0;
672: t1 = u1& 0xf000; t2 = u2& 0xf000; if ( t1&&t2 ) return 0;
673: t1 = u1& 0xf00; t2 = u2& 0xf00; if ( t1&&t2 ) return 0;
674: t1 = u1& 0xf0; t2 = u2& 0xf0; if ( t1&&t2 ) return 0;
675: t1 = u1& 0xf; t2 = u2& 0xf; if ( t1&&t2 ) return 0;
1.1 noro 676: }
677: return 1;
678: break;
679: case 6:
1.41 noro 680: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 681: u1 = d1[i]; u2 = d2[i];
682: t1 = u1&0x3f000000; t2 = u2&0x3f000000; if ( t1&&t2 ) return 0;
1.62 noro 683: t1 = u1& 0xfc0000; t2 = u2& 0xfc0000; if ( t1&&t2 ) return 0;
684: t1 = u1& 0x3f000; t2 = u2& 0x3f000; if ( t1&&t2 ) return 0;
685: t1 = u1& 0xfc0; t2 = u2& 0xfc0; if ( t1&&t2 ) return 0;
686: t1 = u1& 0x3f; t2 = u2& 0x3f; if ( t1&&t2 ) return 0;
1.1 noro 687: }
688: return 1;
689: break;
690: case 8:
1.41 noro 691: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 692: u1 = d1[i]; u2 = d2[i];
693: t1 = u1&0xff000000; t2 = u2&0xff000000; if ( t1&&t2 ) return 0;
1.62 noro 694: t1 = u1& 0xff0000; t2 = u2& 0xff0000; if ( t1&&t2 ) return 0;
695: t1 = u1& 0xff00; t2 = u2& 0xff00; if ( t1&&t2 ) return 0;
696: t1 = u1& 0xff; t2 = u2& 0xff; if ( t1&&t2 ) return 0;
1.1 noro 697: }
698: return 1;
699: break;
700: case 16:
1.41 noro 701: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 702: u1 = d1[i]; u2 = d2[i];
703: t1 = u1&0xffff0000; t2 = u2&0xffff0000; if ( t1&&t2 ) return 0;
1.62 noro 704: t1 = u1& 0xffff; t2 = u2& 0xffff; if ( t1&&t2 ) return 0;
1.1 noro 705: }
706: return 1;
707: break;
708: case 32:
1.41 noro 709: for ( i = nd_exporigin; i < nd_wpd; i++ )
1.1 noro 710: if ( d1[i] && d2[i] ) return 0;
711: return 1;
712: break;
713: default:
1.41 noro 714: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 715: u1 = d1[i]; u2 = d2[i];
716: for ( j = 0; j < nd_epw; j++ ) {
717: if ( (u1&nd_mask0) && (u2&nd_mask0) ) return 0;
718: u1 >>= nd_bpe; u2 >>= nd_bpe;
719: }
720: }
721: return 1;
722: break;
723: }
1.65 noro 724: #else
725: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
726: u1 = d1[i]; u2 = d2[i];
727: for ( j = 0; j < nd_epw; j++ ) {
728: if ( (u1&nd_mask0) && (u2&nd_mask0) ) return 0;
729: u1 >>= nd_bpe; u2 >>= nd_bpe;
730: }
731: }
732: return 1;
733: #endif
1.1 noro 734: }
735:
1.114 noro 736: int ndl_check_bound(UINT *d1,UINT *d2)
1.1 noro 737: {
1.61 noro 738: UINT u2;
1.5 noro 739: int i,j,ind,k;
1.1 noro 740:
1.5 noro 741: ind = 0;
1.65 noro 742: #if USE_UNROLL
1.5 noro 743: switch ( nd_bpe ) {
1.62 noro 744: case 3:
745: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
746: u2 = d2[i];
747: if ( d1[ind++]+((u2>>27)&0x7) >= 0x8 ) return 1;
748: if ( d1[ind++]+((u2>>24)&0x7) >= 0x8 ) return 1;
749: if ( d1[ind++]+((u2>>21)&0x7) >= 0x8 ) return 1;
750: if ( d1[ind++]+((u2>>18)&0x7) >= 0x8 ) return 1;
751: if ( d1[ind++]+((u2>>15)&0x7) >= 0x8 ) return 1;
752: if ( d1[ind++]+((u2>>12)&0x7) >= 0x8 ) return 1;
753: if ( d1[ind++]+((u2>>9)&0x7) >= 0x8 ) return 1;
754: if ( d1[ind++]+((u2>>6)&0x7) >= 0x8 ) return 1;
755: if ( d1[ind++]+((u2>>3)&0x7) >= 0x8 ) return 1;
756: if ( d1[ind++]+(u2&0x7) >= 0x8 ) return 1;
757: }
758: return 0;
759: break;
1.5 noro 760: case 4:
1.41 noro 761: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 762: u2 = d2[i];
763: if ( d1[ind++]+((u2>>28)&0xf) >= 0x10 ) return 1;
764: if ( d1[ind++]+((u2>>24)&0xf) >= 0x10 ) return 1;
765: if ( d1[ind++]+((u2>>20)&0xf) >= 0x10 ) return 1;
766: if ( d1[ind++]+((u2>>16)&0xf) >= 0x10 ) return 1;
767: if ( d1[ind++]+((u2>>12)&0xf) >= 0x10 ) return 1;
768: if ( d1[ind++]+((u2>>8)&0xf) >= 0x10 ) return 1;
769: if ( d1[ind++]+((u2>>4)&0xf) >= 0x10 ) return 1;
770: if ( d1[ind++]+(u2&0xf) >= 0x10 ) return 1;
771: }
772: return 0;
773: break;
774: case 6:
1.41 noro 775: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 776: u2 = d2[i];
777: if ( d1[ind++]+((u2>>24)&0x3f) >= 0x40 ) return 1;
778: if ( d1[ind++]+((u2>>18)&0x3f) >= 0x40 ) return 1;
779: if ( d1[ind++]+((u2>>12)&0x3f) >= 0x40 ) return 1;
780: if ( d1[ind++]+((u2>>6)&0x3f) >= 0x40 ) return 1;
781: if ( d1[ind++]+(u2&0x3f) >= 0x40 ) return 1;
782: }
783: return 0;
784: break;
785: case 8:
1.41 noro 786: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 787: u2 = d2[i];
788: if ( d1[ind++]+((u2>>24)&0xff) >= 0x100 ) return 1;
789: if ( d1[ind++]+((u2>>16)&0xff) >= 0x100 ) return 1;
790: if ( d1[ind++]+((u2>>8)&0xff) >= 0x100 ) return 1;
791: if ( d1[ind++]+(u2&0xff) >= 0x100 ) return 1;
792: }
793: return 0;
794: break;
795: case 16:
1.41 noro 796: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 797: u2 = d2[i];
798: if ( d1[ind++]+((u2>>16)&0xffff) > 0x10000 ) return 1;
799: if ( d1[ind++]+(u2&0xffff) > 0x10000 ) return 1;
800: }
801: return 0;
802: break;
803: case 32:
1.41 noro 804: for ( i = nd_exporigin; i < nd_wpd; i++ )
1.5 noro 805: if ( d1[i]+d2[i]<d1[i] ) return 1;
806: return 0;
807: break;
808: default:
1.41 noro 809: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 810: u2 = d2[i];
811: k = (nd_epw-1)*nd_bpe;
812: for ( j = 0; j < nd_epw; j++, k -= nd_bpe )
813: if ( d1[ind++]+((u2>>k)&nd_mask0) > nd_mask0 ) return 1;
814: }
815: return 0;
816: break;
817: }
1.65 noro 818: #else
819: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
820: u2 = d2[i];
821: k = (nd_epw-1)*nd_bpe;
822: for ( j = 0; j < nd_epw; j++, k -= nd_bpe )
823: if ( d1[ind++]+((u2>>k)&nd_mask0) > nd_mask0 ) return 1;
824: }
825: return 0;
826: #endif
1.1 noro 827: }
828:
1.114 noro 829: int ndl_check_bound2(int index,UINT *d2)
830: {
831: return ndl_check_bound(nd_bound[index],d2);
832: }
833:
1.61 noro 834: INLINE int ndl_hash_value(UINT *d)
1.1 noro 835: {
836: int i;
837: int r;
838:
1.34 noro 839: r = 0;
1.41 noro 840: for ( i = 0; i < nd_wpd; i++ )
1.1 noro 841: r = ((r<<16)+d[i])%REDTAB_LEN;
842: return r;
843: }
844:
1.63 noro 845: INLINE int ndl_find_reducer(UINT *dg)
1.1 noro 846: {
1.13 noro 847: RHist r;
1.6 noro 848: int d,k,i;
1.1 noro 849:
1.63 noro 850: d = ndl_hash_value(dg);
1.13 noro 851: for ( r = nd_red[d], k = 0; r; r = NEXT(r), k++ ) {
1.34 noro 852: if ( ndl_equal(dg,DL(r)) ) {
1.1 noro 853: if ( k > 0 ) nd_notfirst++;
854: nd_found++;
1.13 noro 855: return r->index;
1.1 noro 856: }
857: }
1.13 noro 858: if ( Reverse )
859: for ( i = nd_psn-1; i >= 0; i-- ) {
860: r = nd_psh[i];
1.34 noro 861: if ( ndl_reducible(dg,DL(r)) ) {
1.13 noro 862: nd_create++;
1.34 noro 863: nd_append_red(dg,i);
1.13 noro 864: return i;
865: }
866: }
867: else
868: for ( i = 0; i < nd_psn; i++ ) {
869: r = nd_psh[i];
1.34 noro 870: if ( ndl_reducible(dg,DL(r)) ) {
1.13 noro 871: nd_create++;
1.34 noro 872: nd_append_red(dg,i);
1.13 noro 873: return i;
874: }
1.1 noro 875: }
1.6 noro 876: return -1;
1.1 noro 877: }
878:
1.63 noro 879: ND nd_merge(ND p1,ND p2)
880: {
881: int n,c;
882: int t,can,td1,td2;
883: ND r;
884: NM m1,m2,mr0,mr,s;
885:
886: if ( !p1 ) return p2;
887: else if ( !p2 ) return p1;
888: else {
889: can = 0;
890: for ( n = NV(p1), m1 = BDY(p1), m2 = BDY(p2), mr0 = 0; m1 && m2; ) {
891: c = DL_COMPARE(DL(m1),DL(m2));
892: switch ( c ) {
893: case 0:
894: s = m1; m1 = NEXT(m1);
895: can++; NEXTNM2(mr0,mr,s);
896: s = m2; m2 = NEXT(m2); FREENM(s);
897: break;
898: case 1:
899: s = m1; m1 = NEXT(m1); NEXTNM2(mr0,mr,s);
900: break;
901: case -1:
902: s = m2; m2 = NEXT(m2); NEXTNM2(mr0,mr,s);
903: break;
904: }
905: }
906: if ( !mr0 )
907: if ( m1 ) mr0 = m1;
908: else if ( m2 ) mr0 = m2;
909: else return 0;
910: else if ( m1 ) NEXT(mr) = m1;
911: else if ( m2 ) NEXT(mr) = m2;
912: else NEXT(mr) = 0;
913: BDY(p1) = mr0;
914: SG(p1) = MAX(SG(p1),SG(p2));
915: LEN(p1) = LEN(p1)+LEN(p2)-can;
916: FREEND(p2);
917: return p1;
918: }
919: }
920:
1.31 noro 921: ND nd_add(int mod,ND p1,ND p2)
1.1 noro 922: {
923: int n,c;
1.34 noro 924: int t,can,td1,td2;
1.1 noro 925: ND r;
926: NM m1,m2,mr0,mr,s;
927:
1.34 noro 928: if ( !p1 ) return p2;
929: else if ( !p2 ) return p1;
1.71 noro 930: else if ( mod == -1 ) return nd_add_sf(p1,p2);
1.113 noro 931: else if ( !mod ) return nd_add_q(p1,p2);
1.1 noro 932: else {
1.30 noro 933: can = 0;
1.1 noro 934: for ( n = NV(p1), m1 = BDY(p1), m2 = BDY(p2), mr0 = 0; m1 && m2; ) {
1.34 noro 935: c = DL_COMPARE(DL(m1),DL(m2));
1.1 noro 936: switch ( c ) {
937: case 0:
1.19 noro 938: t = ((CM(m1))+(CM(m2))) - mod;
1.34 noro 939: if ( t < 0 ) t += mod;
1.1 noro 940: s = m1; m1 = NEXT(m1);
941: if ( t ) {
1.34 noro 942: can++; NEXTNM2(mr0,mr,s); CM(mr) = (t);
1.17 noro 943: } else {
1.34 noro 944: can += 2; FREENM(s);
1.17 noro 945: }
946: s = m2; m2 = NEXT(m2); FREENM(s);
1.95 noro 947: break;
948: case 1:
949: s = m1; m1 = NEXT(m1); NEXTNM2(mr0,mr,s);
950: break;
951: case -1:
952: s = m2; m2 = NEXT(m2); NEXTNM2(mr0,mr,s);
953: break;
954: }
955: }
956: if ( !mr0 )
957: if ( m1 ) mr0 = m1;
958: else if ( m2 ) mr0 = m2;
959: else return 0;
960: else if ( m1 ) NEXT(mr) = m1;
961: else if ( m2 ) NEXT(mr) = m2;
962: else NEXT(mr) = 0;
963: BDY(p1) = mr0;
964: SG(p1) = MAX(SG(p1),SG(p2));
965: LEN(p1) = LEN(p1)+LEN(p2)-can;
966: FREEND(p2);
967: return p1;
968: }
969: }
970:
971: /* XXX on opteron, the inlined manipulation of destructive additon of
972: * two NM seems to make gcc optimizer get confused, so the part is
973: * done in a function.
974: */
975:
1.113 noro 976: int nm_destructive_add_q(NM *m1,NM *m2,NM *mr0,NM *mr)
1.95 noro 977: {
978: NM s;
1.113 noro 979: Q t;
1.95 noro 980: int can;
981:
1.113 noro 982: addq(CQ(*m1),CQ(*m2),&t);
1.95 noro 983: s = *m1; *m1 = NEXT(*m1);
984: if ( t ) {
1.113 noro 985: can = 1; NEXTNM2(*mr0,*mr,s); CQ(*mr) = (t);
1.95 noro 986: } else {
987: can = 2; FREENM(s);
988: }
989: s = *m2; *m2 = NEXT(*m2); FREENM(s);
990: return can;
991: }
992:
1.113 noro 993: ND nd_add_q(ND p1,ND p2)
1.95 noro 994: {
995: int n,c,can;
996: ND r;
997: NM m1,m2,mr0,mr,s;
1.113 noro 998: Q t;
1.95 noro 999:
1000: if ( !p1 ) return p2;
1001: else if ( !p2 ) return p1;
1002: else {
1003: can = 0;
1004: for ( n = NV(p1), m1 = BDY(p1), m2 = BDY(p2), mr0 = 0; m1 && m2; ) {
1005: c = DL_COMPARE(DL(m1),DL(m2));
1006: switch ( c ) {
1007: case 0:
1008: #if defined(__x86_64__)
1.113 noro 1009: can += nm_destructive_add_q(&m1,&m2,&mr0,&mr);
1.95 noro 1010: #else
1.113 noro 1011: addq(CQ(m1),CQ(m2),&t);
1.95 noro 1012: s = m1; m1 = NEXT(m1);
1013: if ( t ) {
1.113 noro 1014: can++; NEXTNM2(mr0,mr,s); CQ(mr) = (t);
1.95 noro 1015: } else {
1016: can += 2; FREENM(s);
1017: }
1018: s = m2; m2 = NEXT(m2); FREENM(s);
1019: #endif
1.17 noro 1020: break;
1021: case 1:
1022: s = m1; m1 = NEXT(m1); NEXTNM2(mr0,mr,s);
1023: break;
1024: case -1:
1025: s = m2; m2 = NEXT(m2); NEXTNM2(mr0,mr,s);
1026: break;
1027: }
1028: }
1029: if ( !mr0 )
1.34 noro 1030: if ( m1 ) mr0 = m1;
1031: else if ( m2 ) mr0 = m2;
1032: else return 0;
1033: else if ( m1 ) NEXT(mr) = m1;
1034: else if ( m2 ) NEXT(mr) = m2;
1035: else NEXT(mr) = 0;
1.17 noro 1036: BDY(p1) = mr0;
1037: SG(p1) = MAX(SG(p1),SG(p2));
1.31 noro 1038: LEN(p1) = LEN(p1)+LEN(p2)-can;
1.17 noro 1039: FREEND(p2);
1040: return p1;
1041: }
1042: }
1043:
1.71 noro 1044: ND nd_add_sf(ND p1,ND p2)
1045: {
1046: int n,c,can;
1047: ND r;
1048: NM m1,m2,mr0,mr,s;
1049: int t;
1050:
1051: if ( !p1 ) return p2;
1052: else if ( !p2 ) return p1;
1053: else {
1054: can = 0;
1055: for ( n = NV(p1), m1 = BDY(p1), m2 = BDY(p2), mr0 = 0; m1 && m2; ) {
1056: c = DL_COMPARE(DL(m1),DL(m2));
1057: switch ( c ) {
1058: case 0:
1059: t = _addsf(CM(m1),CM(m2));
1060: s = m1; m1 = NEXT(m1);
1061: if ( t ) {
1062: can++; NEXTNM2(mr0,mr,s); CM(mr) = (t);
1063: } else {
1064: can += 2; FREENM(s);
1065: }
1066: s = m2; m2 = NEXT(m2); FREENM(s);
1067: break;
1068: case 1:
1069: s = m1; m1 = NEXT(m1); NEXTNM2(mr0,mr,s);
1070: break;
1071: case -1:
1072: s = m2; m2 = NEXT(m2); NEXTNM2(mr0,mr,s);
1073: break;
1074: }
1075: }
1076: if ( !mr0 )
1077: if ( m1 ) mr0 = m1;
1078: else if ( m2 ) mr0 = m2;
1079: else return 0;
1080: else if ( m1 ) NEXT(mr) = m1;
1081: else if ( m2 ) NEXT(mr) = m2;
1082: else NEXT(mr) = 0;
1083: BDY(p1) = mr0;
1084: SG(p1) = MAX(SG(p1),SG(p2));
1085: LEN(p1) = LEN(p1)+LEN(p2)-can;
1086: FREEND(p2);
1087: return p1;
1088: }
1089: }
1090:
1.1 noro 1091: /* ret=1 : success, ret=0 : overflow */
1.113 noro 1092: int nd_nf(int mod,ND g,NDV *ps,int full,NDC dn,ND *rp)
1.1 noro 1093: {
1.11 noro 1094: ND d;
1.1 noro 1095: NM m,mrd,tail;
1.7 noro 1096: NM mul;
1.10 noro 1097: int n,sugar,psugar,sugar0,stat,index;
1.30 noro 1098: int c,c1,c2,dummy;
1.17 noro 1099: RHist h;
1.11 noro 1100: NDV p,red;
1.113 noro 1101: Q cg,cred,gcd,tq,qq;
1.21 noro 1102: double hmag;
1.1 noro 1103:
1.113 noro 1104: if ( dn ) {
1105: if ( mod )
1106: dn->m = 1;
1107: else
1108: dn->z = ONE;
1109: }
1.1 noro 1110: if ( !g ) {
1111: *rp = 0;
1112: return 1;
1113: }
1.113 noro 1114: if ( !mod ) hmag = ((double)p_mag((P)HCQ(g)))*nd_scale;
1.21 noro 1115:
1.14 noro 1116: sugar0 = sugar = SG(g);
1.1 noro 1117: n = NV(g);
1.61 noro 1118: mul = (NM)ALLOCA(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT));
1.1 noro 1119: for ( d = 0; g; ) {
1.63 noro 1120: index = ndl_find_reducer(HDL(g));
1.6 noro 1121: if ( index >= 0 ) {
1.17 noro 1122: h = nd_psh[index];
1123: ndl_sub(HDL(g),DL(h),DL(mul));
1.14 noro 1124: if ( ndl_check_bound2(index,DL(mul)) ) {
1.6 noro 1125: nd_free(g); nd_free(d);
1126: return 0;
1127: }
1.77 noro 1128: if ( nd_demand )
1129: p = ndv_load(index);
1130: else
1131: p = ps[index];
1.71 noro 1132: if ( mod == -1 )
1133: CM(mul) = _mulsf(_invsf(HCM(p)),_chsgnsf(HCM(g)));
1134: else if ( mod ) {
1.19 noro 1135: c1 = invm(HCM(p),mod); c2 = mod-HCM(g);
1136: DMAR(c1,c2,0,mod,c); CM(mul) = c;
1.16 noro 1137: } else {
1.113 noro 1138: igcd_cofactor(HCQ(g),HCQ(p),&gcd,&cg,&cred);
1139: chsgnq(cg,&CQ(mul));
1140: nd_mul_c_q(d,cred); nd_mul_c_q(g,cred);
1.69 noro 1141: if ( dn ) {
1.113 noro 1142: mulq(dn->z,cred,&tq); dn->z = tq;
1.69 noro 1143: }
1.16 noro 1144: }
1.55 noro 1145: g = nd_add(mod,g,ndv_mul_nm(mod,mul,p));
1.34 noro 1146: sugar = MAX(sugar,SG(p)+TD(DL(mul)));
1.113 noro 1147: if ( !mod && hmag && g && ((double)(p_mag((P)HCQ(g))) > hmag) ) {
1148: tq = HCQ(g);
1.21 noro 1149: nd_removecont2(d,g);
1.69 noro 1150: if ( dn ) {
1.113 noro 1151: divq(tq,HCQ(g),&qq); divq(dn->z,qq,&tq); dn->z = tq;
1.69 noro 1152: }
1.113 noro 1153: hmag = ((double)p_mag((P)HCQ(g)))*nd_scale;
1.21 noro 1154: }
1.1 noro 1155: } else if ( !full ) {
1156: *rp = g;
1157: return 1;
1158: } else {
1159: m = BDY(g);
1160: if ( NEXT(m) ) {
1.34 noro 1161: BDY(g) = NEXT(m); NEXT(m) = 0; LEN(g)--;
1.1 noro 1162: } else {
1163: FREEND(g); g = 0;
1164: }
1165: if ( d ) {
1.34 noro 1166: NEXT(tail)=m; tail=m; LEN(d)++;
1.1 noro 1167: } else {
1.34 noro 1168: MKND(n,m,1,d); tail = BDY(d);
1.1 noro 1169: }
1170: }
1171: }
1.34 noro 1172: if ( d ) SG(d) = sugar;
1.1 noro 1173: *rp = d;
1174: return 1;
1175: }
1.28 noro 1176:
1.53 noro 1177: int nd_nf_pbucket(int mod,ND g,NDV *ps,int full,ND *rp)
1.25 noro 1178: {
1179: int hindex,index;
1180: NDV p;
1181: ND u,d,red;
1182: NODE l;
1.31 noro 1183: NM mul,m,mrd,tail;
1.25 noro 1184: int sugar,psugar,n,h_reducible;
1185: PGeoBucket bucket;
1186: int c,c1,c2;
1.113 noro 1187: Q cg,cred,gcd,zzz;
1.25 noro 1188: RHist h;
1.28 noro 1189: double hmag,gmag;
1.77 noro 1190: int count = 0;
1191: int hcount = 0;
1.25 noro 1192:
1193: if ( !g ) {
1194: *rp = 0;
1195: return 1;
1196: }
1197: sugar = SG(g);
1198: n = NV(g);
1.113 noro 1199: if ( !mod ) hmag = ((double)p_mag((P)HCQ(g)))*nd_scale;
1.25 noro 1200: bucket = create_pbucket();
1.31 noro 1201: add_pbucket(mod,bucket,g);
1.25 noro 1202: d = 0;
1.61 noro 1203: mul = (NM)ALLOCA(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT));
1.25 noro 1204: while ( 1 ) {
1.113 noro 1205: hindex = mod?head_pbucket(mod,bucket):head_pbucket_q(bucket);
1.25 noro 1206: if ( hindex < 0 ) {
1.77 noro 1207: if ( DP_Print > 3 ) printf("(%d %d)",count,hcount);
1.34 noro 1208: if ( d ) SG(d) = sugar;
1.25 noro 1209: *rp = d;
1210: return 1;
1211: }
1212: g = bucket->body[hindex];
1.63 noro 1213: index = ndl_find_reducer(HDL(g));
1.25 noro 1214: if ( index >= 0 ) {
1.77 noro 1215: count++;
1216: if ( !d ) hcount++;
1.25 noro 1217: h = nd_psh[index];
1218: ndl_sub(HDL(g),DL(h),DL(mul));
1219: if ( ndl_check_bound2(index,DL(mul)) ) {
1.26 noro 1220: nd_free(d);
1.25 noro 1221: free_pbucket(bucket);
1222: *rp = 0;
1223: return 0;
1224: }
1.53 noro 1225: p = ps[index];
1.71 noro 1226: if ( mod == -1 )
1227: CM(mul) = _mulsf(_invsf(HCM(p)),_chsgnsf(HCM(g)));
1228: else if ( mod ) {
1.25 noro 1229: c1 = invm(HCM(p),mod); c2 = mod-HCM(g);
1230: DMAR(c1,c2,0,mod,c); CM(mul) = c;
1231: } else {
1.113 noro 1232: igcd_cofactor(HCQ(g),HCQ(p),&gcd,&cg,&cred);
1233: chsgnq(cg,&CQ(mul));
1234: nd_mul_c_q(d,cred);
1235: mulq_pbucket(bucket,cred);
1.26 noro 1236: g = bucket->body[hindex];
1.113 noro 1237: gmag = (double)p_mag((P)HCQ(g));
1.25 noro 1238: }
1.55 noro 1239: red = ndv_mul_nm(mod,mul,p);
1.25 noro 1240: bucket->body[hindex] = nd_remove_head(g);
1241: red = nd_remove_head(red);
1.31 noro 1242: add_pbucket(mod,bucket,red);
1.34 noro 1243: psugar = SG(p)+TD(DL(mul));
1244: sugar = MAX(sugar,psugar);
1.28 noro 1245: if ( !mod && hmag && (gmag > hmag) ) {
1246: g = normalize_pbucket(mod,bucket);
1247: if ( !g ) {
1.34 noro 1248: if ( d ) SG(d) = sugar;
1.28 noro 1249: *rp = d;
1250: return 1;
1251: }
1252: nd_removecont2(d,g);
1.113 noro 1253: hmag = ((double)p_mag((P)HCQ(g)))*nd_scale;
1.31 noro 1254: add_pbucket(mod,bucket,g);
1.28 noro 1255: }
1.25 noro 1256: } else if ( !full ) {
1257: g = normalize_pbucket(mod,bucket);
1.34 noro 1258: if ( g ) SG(g) = sugar;
1.25 noro 1259: *rp = g;
1260: return 1;
1261: } else {
1262: m = BDY(g);
1263: if ( NEXT(m) ) {
1.34 noro 1264: BDY(g) = NEXT(m); NEXT(m) = 0; LEN(g)--;
1.25 noro 1265: } else {
1266: FREEND(g); g = 0;
1267: }
1268: bucket->body[hindex] = g;
1269: NEXT(m) = 0;
1270: if ( d ) {
1.34 noro 1271: NEXT(tail)=m; tail=m; LEN(d)++;
1.25 noro 1272: } else {
1.34 noro 1273: MKND(n,m,1,d); tail = BDY(d);
1.25 noro 1274: }
1275: }
1276: }
1277: }
1.27 noro 1278:
1.61 noro 1279: /* input : list of NDV, cand : list of NDV */
1.28 noro 1280:
1.61 noro 1281: int ndv_check_candidate(NODE input,int obpe,int oadv,EPOS oepos,NODE cand)
1.28 noro 1282: {
1283: int n,i,stat;
1284: ND nf,d;
1.61 noro 1285: NDV r;
1.45 noro 1286: NODE t,s;
1.113 noro 1287: union oNDC dn;
1.45 noro 1288:
1.128 noro 1289: ndv_setup(0,0,cand,0);
1.31 noro 1290: n = length(cand);
1.28 noro 1291:
1292: /* membercheck : list is a subset of Id(cand) ? */
1293: for ( t = input; t; t = NEXT(t) ) {
1.45 noro 1294: again:
1.61 noro 1295: if ( nd_bpe > obpe )
1296: r = ndv_dup_realloc((NDV)BDY(t),obpe,oadv,oepos);
1297: else
1298: r = (NDV)BDY(t);
1299: d = ndvtond(0,r);
1.69 noro 1300: stat = nd_nf(0,d,nd_ps,0,0,&nf);
1.45 noro 1301: if ( !stat ) {
1.103 noro 1302: nd_reconstruct(0,0);
1.45 noro 1303: goto again;
1304: } else if ( nf ) return 0;
1.71 noro 1305: if ( DP_Print ) { printf("."); fflush(stdout); }
1.28 noro 1306: }
1.71 noro 1307: if ( DP_Print ) { printf("\n"); }
1.28 noro 1308: /* gbcheck : cand is a GB of Id(cand) ? */
1.77 noro 1309: if ( !nd_gb(0,0,1) ) return 0;
1.28 noro 1310: /* XXX */
1.23 noro 1311: return 1;
1312: }
1.1 noro 1313:
1314: ND nd_remove_head(ND p)
1315: {
1316: NM m;
1317:
1318: m = BDY(p);
1319: if ( !NEXT(m) ) {
1.34 noro 1320: FREEND(p); p = 0;
1.31 noro 1321: } else {
1.34 noro 1322: BDY(p) = NEXT(m); LEN(p)--;
1.31 noro 1323: }
1.1 noro 1324: FREENM(m);
1325: return p;
1326: }
1327:
1.69 noro 1328: ND nd_separate_head(ND p,ND *head)
1329: {
1330: NM m,m0;
1331: ND r;
1332:
1333: m = BDY(p);
1334: if ( !NEXT(m) ) {
1335: *head = p; p = 0;
1336: } else {
1337: m0 = m;
1338: BDY(p) = NEXT(m); LEN(p)--;
1339: NEXT(m0) = 0;
1340: MKND(NV(p),m0,1,r);
1341: *head = r;
1342: }
1343: return p;
1344: }
1345:
1.1 noro 1346: PGeoBucket create_pbucket()
1347: {
1348: PGeoBucket g;
1349:
1350: g = CALLOC(1,sizeof(struct oPGeoBucket));
1351: g->m = -1;
1352: return g;
1353: }
1354:
1.25 noro 1355: void free_pbucket(PGeoBucket b) {
1356: int i;
1357:
1.26 noro 1358: for ( i = 0; i <= b->m; i++ )
1.25 noro 1359: if ( b->body[i] ) {
1360: nd_free(b->body[i]);
1361: b->body[i] = 0;
1362: }
1363: GC_free(b);
1364: }
1365:
1.63 noro 1366: void add_pbucket_symbolic(PGeoBucket g,ND d)
1367: {
1368: int l,i,k,m;
1369:
1370: if ( !d )
1371: return;
1372: l = LEN(d);
1373: for ( k = 0, m = 1; l > m; k++, m <<= 1 );
1374: /* 2^(k-1) < l <= 2^k (=m) */
1375: d = nd_merge(g->body[k],d);
1376: for ( ; d && LEN(d) > m; k++, m <<= 1 ) {
1377: g->body[k] = 0;
1378: d = nd_merge(g->body[k+1],d);
1379: }
1380: g->body[k] = d;
1381: g->m = MAX(g->m,k);
1382: }
1383:
1.31 noro 1384: void add_pbucket(int mod,PGeoBucket g,ND d)
1.1 noro 1385: {
1.31 noro 1386: int l,i,k,m;
1.1 noro 1387:
1.31 noro 1388: if ( !d )
1389: return;
1390: l = LEN(d);
1.29 noro 1391: for ( k = 0, m = 1; l > m; k++, m <<= 1 );
1392: /* 2^(k-1) < l <= 2^k (=m) */
1.31 noro 1393: d = nd_add(mod,g->body[k],d);
1394: for ( ; d && LEN(d) > m; k++, m <<= 1 ) {
1.1 noro 1395: g->body[k] = 0;
1.31 noro 1396: d = nd_add(mod,g->body[k+1],d);
1.1 noro 1397: }
1398: g->body[k] = d;
1399: g->m = MAX(g->m,k);
1400: }
1401:
1.113 noro 1402: void mulq_pbucket(PGeoBucket g,Q c)
1.26 noro 1403: {
1404: int k;
1405:
1406: for ( k = 0; k <= g->m; k++ )
1.113 noro 1407: nd_mul_c_q(g->body[k],c);
1.26 noro 1408: }
1409:
1.63 noro 1410: NM remove_head_pbucket_symbolic(PGeoBucket g)
1411: {
1412: int j,i,k,c;
1413: NM head;
1414:
1415: k = g->m;
1416: j = -1;
1417: for ( i = 0; i <= k; i++ ) {
1418: if ( !g->body[i] ) continue;
1419: if ( j < 0 ) j = i;
1420: else {
1421: c = DL_COMPARE(HDL(g->body[i]),HDL(g->body[j]));
1422: if ( c > 0 )
1423: j = i;
1424: else if ( c == 0 )
1425: g->body[i] = nd_remove_head(g->body[i]);
1426: }
1427: }
1428: if ( j < 0 ) return 0;
1429: else {
1430: head = BDY(g->body[j]);
1431: if ( !NEXT(head) ) {
1432: FREEND(g->body[j]);
1433: g->body[j] = 0;
1434: } else {
1435: BDY(g->body[j]) = NEXT(head);
1436: LEN(g->body[j])--;
1437: }
1438: return head;
1439: }
1440: }
1441:
1.19 noro 1442: int head_pbucket(int mod,PGeoBucket g)
1.1 noro 1443: {
1444: int j,i,c,k,nv,sum;
1.61 noro 1445: UINT *di,*dj;
1.1 noro 1446: ND gi,gj;
1447:
1448: k = g->m;
1449: while ( 1 ) {
1450: j = -1;
1451: for ( i = 0; i <= k; i++ ) {
1452: if ( !(gi = g->body[i]) )
1453: continue;
1454: if ( j < 0 ) {
1455: j = i;
1456: gj = g->body[j];
1457: dj = HDL(gj);
1.14 noro 1458: sum = HCM(gj);
1.1 noro 1459: } else {
1.34 noro 1460: c = DL_COMPARE(HDL(gi),dj);
1.1 noro 1461: if ( c > 0 ) {
1.34 noro 1462: if ( sum ) HCM(gj) = sum;
1463: else g->body[j] = nd_remove_head(gj);
1.1 noro 1464: j = i;
1465: gj = g->body[j];
1466: dj = HDL(gj);
1.14 noro 1467: sum = HCM(gj);
1.1 noro 1468: } else if ( c == 0 ) {
1.71 noro 1469: if ( mod == -1 )
1470: sum = _addsf(sum,HCM(gi));
1471: else {
1472: sum = sum+HCM(gi)-mod;
1473: if ( sum < 0 ) sum += mod;
1474: }
1.1 noro 1475: g->body[i] = nd_remove_head(gi);
1476: }
1477: }
1478: }
1.34 noro 1479: if ( j < 0 ) return -1;
1.1 noro 1480: else if ( sum ) {
1.14 noro 1481: HCM(gj) = sum;
1.26 noro 1482: return j;
1.31 noro 1483: } else
1.26 noro 1484: g->body[j] = nd_remove_head(gj);
1485: }
1486: }
1487:
1.113 noro 1488: int head_pbucket_q(PGeoBucket g)
1.26 noro 1489: {
1490: int j,i,c,k,nv;
1.113 noro 1491: Q sum,t;
1.26 noro 1492: ND gi,gj;
1493:
1494: k = g->m;
1495: while ( 1 ) {
1496: j = -1;
1497: for ( i = 0; i <= k; i++ ) {
1.34 noro 1498: if ( !(gi = g->body[i]) ) continue;
1.26 noro 1499: if ( j < 0 ) {
1500: j = i;
1501: gj = g->body[j];
1.113 noro 1502: sum = HCQ(gj);
1.26 noro 1503: } else {
1504: nv = NV(gi);
1.34 noro 1505: c = DL_COMPARE(HDL(gi),HDL(gj));
1.26 noro 1506: if ( c > 0 ) {
1.113 noro 1507: if ( sum ) HCQ(gj) = sum;
1.34 noro 1508: else g->body[j] = nd_remove_head(gj);
1.26 noro 1509: j = i;
1510: gj = g->body[j];
1.113 noro 1511: sum = HCQ(gj);
1.26 noro 1512: } else if ( c == 0 ) {
1.113 noro 1513: addq(sum,HCQ(gi),&t);
1514: sum = t;
1.26 noro 1515: g->body[i] = nd_remove_head(gi);
1516: }
1517: }
1518: }
1.34 noro 1519: if ( j < 0 ) return -1;
1.26 noro 1520: else if ( sum ) {
1.113 noro 1521: HCQ(gj) = sum;
1.1 noro 1522: return j;
1.31 noro 1523: } else
1.1 noro 1524: g->body[j] = nd_remove_head(gj);
1525: }
1526: }
1527:
1.25 noro 1528: ND normalize_pbucket(int mod,PGeoBucket g)
1.1 noro 1529: {
1.31 noro 1530: int i;
1.1 noro 1531: ND r,t;
1532:
1533: r = 0;
1.28 noro 1534: for ( i = 0; i <= g->m; i++ ) {
1.31 noro 1535: r = nd_add(mod,r,g->body[i]);
1.28 noro 1536: g->body[i] = 0;
1537: }
1538: g->m = -1;
1.1 noro 1539: return r;
1540: }
1541:
1.122 noro 1542: int do_diagonalize(int sugar,int m)
1.92 noro 1543: {
1544: int i,nh,stat;
1545: NODE r,g,t;
1546: ND h,nf,s,head;
1547: NDV nfv;
1.113 noro 1548: Q q,num,den;
1549: union oNDC dn;
1.92 noro 1550:
1.143 noro 1551: for ( i = nd_psn-1; i >= 0 && SG(nd_psh[i]) == sugar; i-- ) {
1.92 noro 1552: if ( nd_demand )
1553: nfv = ndv_load(i);
1554: else
1555: nfv = nd_ps[i];
1556: s = ndvtond(m,nfv);
1557: s = nd_separate_head(s,&head);
1.122 noro 1558: stat = nd_nf(m,s,nd_ps,1,&dn,&nf);
1559: if ( !stat ) return 0;
1.92 noro 1560: if ( !m ) {
1.113 noro 1561: NTOQ(NM(dn.z),SGN(dn.z),num);
1562: mulq(HCQ(head),num,&q); HCQ(head) = q;
1563: if ( DN(dn.z) ) {
1564: NTOQ(DN(dn.z),1,den);
1565: nd_mul_c_q(nf,den);
1566: }
1.92 noro 1567: }
1568: nf = nd_add(m,head,nf);
1569: ndv_free(nfv);
1570: nd_removecont(m,nf);
1571: nfv = ndtondv(m,nf);
1572: nd_free(nf);
1573: nd_bound[i] = ndv_compute_bound(nfv);
1574: if ( nd_demand ) {
1575: ndv_save(nfv,i);
1576: ndv_free(nfv);
1577: } else
1578: nd_ps[i] = nfv;
1579: }
1.122 noro 1580: return 1;
1.92 noro 1581: }
1582:
1.27 noro 1583: /* return value = 0 => input is not a GB */
1584:
1.77 noro 1585: NODE nd_gb(int m,int ishomo,int checkonly)
1.1 noro 1586: {
1587: int i,nh,sugar,stat;
1.23 noro 1588: NODE r,g,t;
1.1 noro 1589: ND_pairs d;
1590: ND_pairs l;
1.117 noro 1591: ND h,nf,s,head,nf1;
1.63 noro 1592: NDV nfv;
1.113 noro 1593: Q q,num,den;
1594: union oNDC dn;
1.140 noro 1595: int diag_count = 0;
1.1 noro 1596:
1.23 noro 1597: g = 0; d = 0;
1598: for ( i = 0; i < nd_psn; i++ ) {
1.1 noro 1599: d = update_pairs(d,g,i);
1600: g = update_base(g,i);
1601: }
1602: sugar = 0;
1603: while ( d ) {
1604: again:
1605: l = nd_minp(d,&d);
1.14 noro 1606: if ( SG(l) != sugar ) {
1.122 noro 1607: if ( ishomo ) {
1.140 noro 1608: diag_count = 0;
1.122 noro 1609: stat = do_diagonalize(sugar,m);
1610: if ( !stat ) {
1611: NEXT(l) = d; d = l;
1612: d = nd_reconstruct(0,d);
1613: goto again;
1614: }
1615: }
1.14 noro 1616: sugar = SG(l);
1.71 noro 1617: if ( DP_Print ) fprintf(asir_out,"%d",sugar);
1.1 noro 1618: }
1.53 noro 1619: stat = nd_sp(m,0,l,&h);
1.1 noro 1620: if ( !stat ) {
1621: NEXT(l) = d; d = l;
1.103 noro 1622: d = nd_reconstruct(0,d);
1.1 noro 1623: goto again;
1624: }
1.41 noro 1625: #if USE_GEOBUCKET
1.69 noro 1626: stat = m?nd_nf_pbucket(m,h,nd_ps,!Top,&nf):nd_nf(m,h,nd_ps,!Top,0,&nf);
1.41 noro 1627: #else
1.69 noro 1628: stat = nd_nf(m,h,nd_ps,!Top,0,&nf);
1.41 noro 1629: #endif
1.1 noro 1630: if ( !stat ) {
1631: NEXT(l) = d; d = l;
1.103 noro 1632: d = nd_reconstruct(0,d);
1.1 noro 1633: goto again;
1634: } else if ( nf ) {
1.27 noro 1635: if ( checkonly ) return 0;
1.71 noro 1636: if ( DP_Print ) { printf("+"); fflush(stdout); }
1.63 noro 1637: nd_removecont(m,nf);
1.118 noro 1638: if ( !m && nd_nalg ) {
1639: nd_monic(0,&nf);
1640: nd_removecont(m,nf);
1.117 noro 1641: }
1.63 noro 1642: nfv = ndtondv(m,nf); nd_free(nf);
1.77 noro 1643: nh = ndv_newps(m,nfv,0);
1.141 noro 1644: if ( !m && (ishomo && ++diag_count == diag_period) ) {
1.140 noro 1645: diag_count = 0;
1646: stat = do_diagonalize(sugar,m);
1647: if ( !stat ) {
1648: NEXT(l) = d; d = l;
1649: d = nd_reconstruct(1,d);
1650: goto again;
1651: }
1652: }
1.1 noro 1653: d = update_pairs(d,g,nh);
1654: g = update_base(g,nh);
1655: FREENDP(l);
1656: } else {
1.71 noro 1657: if ( DP_Print ) { printf("."); fflush(stdout); }
1.1 noro 1658: FREENDP(l);
1659: }
1660: }
1.77 noro 1661: if ( nd_demand )
1662: for ( t = g; t; t = NEXT(t) )
1663: BDY(t) = (pointer)ndv_load((int)BDY(t));
1664: else
1665: for ( t = g; t; t = NEXT(t) )
1666: BDY(t) = (pointer)nd_ps[(int)BDY(t)];
1.1 noro 1667: return g;
1668: }
1669:
1.122 noro 1670: int do_diagonalize_trace(int sugar,int m)
1.91 noro 1671: {
1672: int i,nh,stat;
1673: NODE r,g,t;
1674: ND h,nf,nfq,s,head;
1675: NDV nfv,nfqv;
1.113 noro 1676: Q q,den,num;
1677: union oNDC dn;
1.91 noro 1678:
1.143 noro 1679: for ( i = nd_psn-1; i >= 0 && SG(nd_psh[i]) == sugar; i-- ) {
1.91 noro 1680: /* for nd_ps */
1681: s = ndvtond(m,nd_ps[i]);
1682: s = nd_separate_head(s,&head);
1.122 noro 1683: stat = nd_nf_pbucket(m,s,nd_ps,1,&nf);
1684: if ( !stat ) return 0;
1.91 noro 1685: nf = nd_add(m,head,nf);
1686: ndv_free(nd_ps[i]);
1687: nd_ps[i] = ndtondv(m,nf);
1688: nd_free(nf);
1689:
1690: /* for nd_ps_trace */
1691: if ( nd_demand )
1692: nfv = ndv_load(i);
1693: else
1694: nfv = nd_ps_trace[i];
1695: s = ndvtond(0,nfv);
1696: s = nd_separate_head(s,&head);
1.122 noro 1697: stat = nd_nf(0,s,nd_ps_trace,1,&dn,&nf);
1698: if ( !stat ) return 0;
1.113 noro 1699: NTOQ(NM(dn.z),SGN(dn.z),num);
1700: mulq(HCQ(head),num,&q); HCQ(head) = q;
1701: if ( DN(dn.z) ) {
1702: NTOQ(DN(dn.z),1,den);
1703: nd_mul_c_q(nf,den);
1704: }
1.91 noro 1705: nf = nd_add(0,head,nf);
1706: ndv_free(nfv);
1707: nd_removecont(0,nf);
1708: nfv = ndtondv(0,nf);
1709: nd_free(nf);
1710: nd_bound[i] = ndv_compute_bound(nfv);
1711: if ( nd_demand ) {
1712: ndv_save(nfv,i);
1713: ndv_free(nfv);
1714: } else
1715: nd_ps_trace[i] = nfv;
1716: }
1.122 noro 1717: return 1;
1.91 noro 1718: }
1719:
1.118 noro 1720: static struct oEGT eg_invdalg;
1721: struct oEGT eg_le;
1722:
1.82 noro 1723: NODE nd_gb_trace(int m,int ishomo)
1.20 noro 1724: {
1725: int i,nh,sugar,stat;
1.23 noro 1726: NODE r,g,t;
1.20 noro 1727: ND_pairs d;
1728: ND_pairs l;
1.82 noro 1729: ND h,nf,nfq,s,head;
1.63 noro 1730: NDV nfv,nfqv;
1.113 noro 1731: Q q,den,num;
1732: union oNDC dn;
1.118 noro 1733: struct oEGT eg_monic,egm0,egm1;
1.130 noro 1734: int diag_count = 0;
1.20 noro 1735:
1.118 noro 1736: init_eg(&eg_monic);
1737: init_eg(&eg_invdalg);
1738: init_eg(&eg_le);
1.23 noro 1739: g = 0; d = 0;
1740: for ( i = 0; i < nd_psn; i++ ) {
1.20 noro 1741: d = update_pairs(d,g,i);
1742: g = update_base(g,i);
1743: }
1744: sugar = 0;
1745: while ( d ) {
1746: again:
1747: l = nd_minp(d,&d);
1748: if ( SG(l) != sugar ) {
1.130 noro 1749: #if 1
1.122 noro 1750: if ( ishomo ) {
1751: stat = do_diagonalize_trace(sugar,m);
1.130 noro 1752: diag_count = 0;
1.122 noro 1753: if ( !stat ) {
1754: NEXT(l) = d; d = l;
1755: d = nd_reconstruct(1,d);
1756: goto again;
1757: }
1758: }
1.130 noro 1759: #endif
1.20 noro 1760: sugar = SG(l);
1.71 noro 1761: if ( DP_Print ) fprintf(asir_out,"%d",sugar);
1.20 noro 1762: }
1.53 noro 1763: stat = nd_sp(m,0,l,&h);
1.20 noro 1764: if ( !stat ) {
1765: NEXT(l) = d; d = l;
1.103 noro 1766: d = nd_reconstruct(1,d);
1.20 noro 1767: goto again;
1768: }
1.41 noro 1769: #if USE_GEOBUCKET
1.53 noro 1770: stat = nd_nf_pbucket(m,h,nd_ps,!Top,&nf);
1.41 noro 1771: #else
1.69 noro 1772: stat = nd_nf(m,h,nd_ps,!Top,0,&nf);
1.41 noro 1773: #endif
1.20 noro 1774: if ( !stat ) {
1775: NEXT(l) = d; d = l;
1.103 noro 1776: d = nd_reconstruct(1,d);
1.20 noro 1777: goto again;
1778: } else if ( nf ) {
1.77 noro 1779: if ( nd_demand ) {
1780: nfqv = ndv_load(nd_psn);
1781: nfq = ndvtond(0,nfqv);
1782: } else
1783: nfq = 0;
1784: if ( !nfq ) {
1785: if ( !nd_sp(0,1,l,&h) || !nd_nf(0,h,nd_ps_trace,!Top,0,&nfq) ) {
1786: NEXT(l) = d; d = l;
1.103 noro 1787: d = nd_reconstruct(1,d);
1.77 noro 1788: goto again;
1789: }
1.70 noro 1790: }
1.20 noro 1791: if ( nfq ) {
1.70 noro 1792: /* m|HC(nfq) => failure */
1.113 noro 1793: if ( !rem(NM(HCQ(nfq)),m) ) return 0;
1.63 noro 1794:
1.71 noro 1795: if ( DP_Print ) { printf("+"); fflush(stdout); }
1.118 noro 1796: if ( nd_nalg ) {
1797: /* m|DN(HC(nf)^(-1)) => failure */
1798: get_eg(&egm0);
1799: if ( !nd_monic(m,&nfq) ) return 0;
1800: get_eg(&egm1); add_eg(&eg_monic,&egm0,&egm1);
1801: nd_removecont(0,nfq); nfqv = ndtondv(0,nfq); nd_free(nfq);
1802: nfv = ndv_dup(0,nfqv); ndv_mod(m,nfv); nd_free(nf);
1803: } else {
1804: nd_removecont(0,nfq); nfqv = ndtondv(0,nfq); nd_free(nfq);
1805: nd_removecont(m,nf); nfv = ndtondv(m,nf); nd_free(nf);
1806: }
1.77 noro 1807: nh = ndv_newps(0,nfv,nfqv);
1.130 noro 1808: if ( ishomo && ++diag_count == diag_period ) {
1809: diag_count = 0;
1810: stat = do_diagonalize_trace(sugar,m);
1811: if ( !stat ) {
1812: NEXT(l) = d; d = l;
1813: d = nd_reconstruct(1,d);
1814: goto again;
1815: }
1816: }
1.20 noro 1817: d = update_pairs(d,g,nh);
1818: g = update_base(g,nh);
1819: } else {
1.71 noro 1820: if ( DP_Print ) { printf("*"); fflush(stdout); }
1.20 noro 1821: }
1822: } else {
1.71 noro 1823: if ( DP_Print ) { printf("."); fflush(stdout); }
1.20 noro 1824: }
1825: FREENDP(l);
1826: }
1.77 noro 1827: if ( nd_demand )
1828: for ( t = g; t; t = NEXT(t) )
1829: BDY(t) = (pointer)ndv_load((int)BDY(t));
1830: else
1831: for ( t = g; t; t = NEXT(t) )
1832: BDY(t) = (pointer)nd_ps_trace[(int)BDY(t)];
1.118 noro 1833: if ( nd_nalg ) {
1834: print_eg("monic",&eg_monic);
1835: print_eg("invdalg",&eg_invdalg);
1836: print_eg("le",&eg_le);
1837: }
1.20 noro 1838: return g;
1839: }
1840:
1.23 noro 1841: int ndv_compare(NDV *p1,NDV *p2)
1842: {
1.34 noro 1843: return DL_COMPARE(HDL(*p1),HDL(*p2));
1.23 noro 1844: }
1845:
1846: int ndv_compare_rev(NDV *p1,NDV *p2)
1847: {
1.34 noro 1848: return -DL_COMPARE(HDL(*p1),HDL(*p2));
1.23 noro 1849: }
1850:
1.61 noro 1851: NODE ndv_reduceall(int m,NODE f)
1.23 noro 1852: {
1.69 noro 1853: int i,n,stat;
1854: ND nf,g,head;
1.23 noro 1855: NODE t,a0,a;
1.113 noro 1856: union oNDC dn;
1.69 noro 1857: NDV *w;
1.113 noro 1858: Q q,num,den;
1.23 noro 1859:
1.69 noro 1860: n = length(f);
1861: #if 0
1862: w = (NDV *)ALLOCA(n*sizeof(NDV));
1863: for ( i = 0, t = f; i < n; i++, t = NEXT(t) ) w[i] = (NDV)BDY(t);
1864: qsort(w,n,sizeof(NDV),
1865: (int (*)(const void *,const void *))ndv_compare);
1866: for ( t = f, i = 0; t; i++, t = NEXT(t) ) BDY(t) = (pointer)w[i];
1867: #endif
1.128 noro 1868: ndv_setup(m,0,f,0);
1.69 noro 1869: for ( i = 0; i < n; ) {
1870: g = ndvtond(m,nd_ps[i]);
1871: g = nd_separate_head(g,&head);
1872: stat = nd_nf(m,g,nd_ps,1,&dn,&nf);
1.50 noro 1873: if ( !stat )
1.103 noro 1874: nd_reconstruct(0,0);
1.69 noro 1875: else {
1.71 noro 1876: if ( DP_Print ) { printf("."); fflush(stdout); }
1.85 noro 1877: if ( !m ) {
1.113 noro 1878: NTOQ(NM(dn.z),SGN(dn.z),num);
1879: mulq(HCQ(head),num,&q); HCQ(head) = q;
1880: if ( DN(dn.z) ) {
1881: NTOQ(DN(dn.z),1,den);
1882: nd_mul_c_q(nf,den);
1883: }
1.85 noro 1884: }
1.69 noro 1885: nf = nd_add(m,head,nf);
1886: ndv_free(nd_ps[i]);
1.24 noro 1887: nd_removecont(m,nf);
1.69 noro 1888: nd_ps[i] = ndtondv(m,nf); nd_free(nf);
1889: nd_bound[i] = ndv_compute_bound(nd_ps[i]);
1.50 noro 1890: i++;
1.23 noro 1891: }
1892: }
1.71 noro 1893: if ( DP_Print ) { printf("\n"); }
1.23 noro 1894: for ( a0 = 0, i = 0; i < n; i++ ) {
1895: NEXTNODE(a0,a);
1.69 noro 1896: BDY(a) = (pointer)nd_ps[i];
1.23 noro 1897: }
1898: NEXT(a) = 0;
1899: return a0;
1900: }
1901:
1.1 noro 1902: ND_pairs update_pairs( ND_pairs d, NODE /* of index */ g, int t)
1903: {
1904: ND_pairs d1,nd,cur,head,prev,remove;
1905:
1906: if ( !g ) return d;
1907: d = crit_B(d,t);
1908: d1 = nd_newpairs(g,t);
1909: d1 = crit_M(d1);
1910: d1 = crit_F(d1);
1.55 noro 1911: if ( do_weyl )
1912: head = d1;
1913: else {
1914: prev = 0; cur = head = d1;
1915: while ( cur ) {
1916: if ( crit_2( cur->i1,cur->i2 ) ) {
1917: remove = cur;
1918: if ( !prev ) head = cur = NEXT(cur);
1919: else cur = NEXT(prev) = NEXT(cur);
1920: FREENDP(remove);
1921: } else {
1922: prev = cur; cur = NEXT(cur);
1923: }
1.1 noro 1924: }
1925: }
1926: if ( !d )
1927: return head;
1928: else {
1929: nd = d;
1.34 noro 1930: while ( NEXT(nd) ) nd = NEXT(nd);
1.1 noro 1931: NEXT(nd) = head;
1932: return d;
1933: }
1934: }
1935:
1936: ND_pairs nd_newpairs( NODE g, int t )
1937: {
1938: NODE h;
1.61 noro 1939: UINT *dl;
1.34 noro 1940: int ts,s;
1.1 noro 1941: ND_pairs r,r0;
1942:
1.20 noro 1943: dl = DL(nd_psh[t]);
1.34 noro 1944: ts = SG(nd_psh[t]) - TD(dl);
1.1 noro 1945: for ( r0 = 0, h = g; h; h = NEXT(h) ) {
1946: NEXTND_pairs(r0,r);
1947: r->i1 = (int)BDY(h);
1948: r->i2 = t;
1.20 noro 1949: ndl_lcm(DL(nd_psh[r->i1]),dl,r->lcm);
1.34 noro 1950: s = SG(nd_psh[r->i1])-TD(DL(nd_psh[r->i1]));
1951: SG(r) = MAX(s,ts) + TD(LCM(r));
1.1 noro 1952: }
1953: NEXT(r) = 0;
1954: return r0;
1955: }
1956:
1957: ND_pairs crit_B( ND_pairs d, int s )
1958: {
1959: ND_pairs cur,head,prev,remove;
1.61 noro 1960: UINT *t,*tl,*lcm;
1.1 noro 1961: int td,tdl;
1962:
1963: if ( !d ) return 0;
1.20 noro 1964: t = DL(nd_psh[s]);
1.1 noro 1965: prev = 0;
1966: head = cur = d;
1.61 noro 1967: lcm = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
1.1 noro 1968: while ( cur ) {
1969: tl = cur->lcm;
1970: if ( ndl_reducible(tl,t)
1.20 noro 1971: && (ndl_lcm(DL(nd_psh[cur->i1]),t,lcm),!ndl_equal(lcm,tl))
1972: && (ndl_lcm(DL(nd_psh[cur->i2]),t,lcm),!ndl_equal(lcm,tl)) ) {
1.1 noro 1973: remove = cur;
1974: if ( !prev ) {
1975: head = cur = NEXT(cur);
1976: } else {
1977: cur = NEXT(prev) = NEXT(cur);
1978: }
1979: FREENDP(remove);
1980: } else {
1.34 noro 1981: prev = cur; cur = NEXT(cur);
1.1 noro 1982: }
1983: }
1984: return head;
1985: }
1986:
1987: ND_pairs crit_M( ND_pairs d1 )
1988: {
1989: ND_pairs e,d2,d3,dd,p;
1.61 noro 1990: UINT *id,*jd;
1.1 noro 1991:
1992: for ( dd = 0, e = d1; e; e = d3 ) {
1993: if ( !(d2 = NEXT(e)) ) {
1994: NEXT(e) = dd;
1995: return e;
1996: }
1.34 noro 1997: id = LCM(e);
1.1 noro 1998: for ( d3 = 0; d2; d2 = p ) {
1.34 noro 1999: p = NEXT(d2);
2000: jd = LCM(d2);
2001: if ( ndl_equal(jd,id) )
2002: ;
2003: else if ( TD(jd) > TD(id) )
1.1 noro 2004: if ( ndl_reducible(jd,id) ) continue;
2005: else ;
1.34 noro 2006: else if ( ndl_reducible(id,jd) ) goto delit;
1.1 noro 2007: NEXT(d2) = d3;
2008: d3 = d2;
2009: }
2010: NEXT(e) = dd;
2011: dd = e;
2012: continue;
2013: /**/
2014: delit: NEXT(d2) = d3;
2015: d3 = d2;
2016: for ( ; p; p = d2 ) {
2017: d2 = NEXT(p);
2018: NEXT(p) = d3;
2019: d3 = p;
2020: }
2021: FREENDP(e);
2022: }
2023: return dd;
2024: }
2025:
2026: ND_pairs crit_F( ND_pairs d1 )
2027: {
2028: ND_pairs rest, head,remove;
2029: ND_pairs last, p, r, w;
2030: int s;
2031:
2032: for ( head = last = 0, p = d1; NEXT(p); ) {
2033: r = w = equivalent_pairs(p,&rest);
1.14 noro 2034: s = SG(r);
1.1 noro 2035: w = NEXT(w);
2036: while ( w ) {
2037: if ( crit_2(w->i1,w->i2) ) {
2038: r = w;
2039: w = NEXT(w);
2040: while ( w ) {
2041: remove = w;
2042: w = NEXT(w);
2043: FREENDP(remove);
2044: }
2045: break;
1.14 noro 2046: } else if ( SG(w) < s ) {
1.1 noro 2047: FREENDP(r);
2048: r = w;
1.14 noro 2049: s = SG(r);
1.1 noro 2050: w = NEXT(w);
2051: } else {
2052: remove = w;
2053: w = NEXT(w);
2054: FREENDP(remove);
2055: }
2056: }
2057: if ( last ) NEXT(last) = r;
2058: else head = r;
2059: NEXT(last = r) = 0;
2060: p = rest;
2061: if ( !p ) return head;
2062: }
2063: if ( !last ) return p;
2064: NEXT(last) = p;
2065: return head;
2066: }
2067:
2068: int crit_2( int dp1, int dp2 )
2069: {
1.20 noro 2070: return ndl_disjoint(DL(nd_psh[dp1]),DL(nd_psh[dp2]));
1.1 noro 2071: }
2072:
1.40 noro 2073: ND_pairs equivalent_pairs( ND_pairs d1, ND_pairs *prest )
1.1 noro 2074: {
2075: ND_pairs w,p,r,s;
1.61 noro 2076: UINT *d;
1.1 noro 2077:
2078: w = d1;
1.34 noro 2079: d = LCM(w);
1.1 noro 2080: s = NEXT(w);
2081: NEXT(w) = 0;
2082: for ( r = 0; s; s = p ) {
2083: p = NEXT(s);
1.34 noro 2084: if ( ndl_equal(d,LCM(s)) ) {
1.39 noro 2085: NEXT(s) = w; w = s;
1.1 noro 2086: } else {
1.39 noro 2087: NEXT(s) = r; r = s;
1.1 noro 2088: }
2089: }
2090: *prest = r;
2091: return w;
2092: }
2093:
2094: NODE update_base(NODE nd,int ndp)
2095: {
1.61 noro 2096: UINT *dl, *dln;
1.137 noro 2097: NODE last, p, head;
1.1 noro 2098:
1.20 noro 2099: dl = DL(nd_psh[ndp]);
1.1 noro 2100: for ( head = last = 0, p = nd; p; ) {
1.20 noro 2101: dln = DL(nd_psh[(int)BDY(p)]);
1.34 noro 2102: if ( ndl_reducible( dln, dl ) ) {
1.1 noro 2103: p = NEXT(p);
2104: if ( last ) NEXT(last) = p;
2105: } else {
2106: if ( !last ) head = p;
2107: p = NEXT(last = p);
2108: }
2109: }
2110: head = append_one(head,ndp);
2111: return head;
2112: }
2113:
2114: ND_pairs nd_minp( ND_pairs d, ND_pairs *prest )
2115: {
2116: ND_pairs m,ml,p,l;
1.61 noro 2117: UINT *lcm;
1.33 noro 2118: int s,td,len,tlen,c,c1;
1.1 noro 2119:
2120: if ( !(p = NEXT(m = d)) ) {
2121: *prest = p;
2122: NEXT(m) = 0;
2123: return m;
2124: }
1.14 noro 2125: s = SG(m);
1.77 noro 2126: if ( !NoSugar ) {
2127: for ( ml = 0, l = m; p; p = NEXT(l = p) )
2128: if ( (SG(p) < s)
2129: || ((SG(p) == s) && (DL_COMPARE(LCM(p),LCM(m)) < 0)) ) {
2130: ml = l; m = p; s = SG(m);
2131: }
2132: } else {
2133: for ( ml = 0, l = m; p; p = NEXT(l = p) )
2134: if ( DL_COMPARE(LCM(p),LCM(m)) < 0 ) {
2135: ml = l; m = p; s = SG(m);
2136: }
2137: }
1.1 noro 2138: if ( !ml ) *prest = NEXT(m);
2139: else {
2140: NEXT(ml) = NEXT(m);
2141: *prest = d;
2142: }
2143: NEXT(m) = 0;
2144: return m;
2145: }
2146:
1.63 noro 2147: ND_pairs nd_minsugarp( ND_pairs d, ND_pairs *prest )
2148: {
1.69 noro 2149: int msugar,i;
1.63 noro 2150: ND_pairs t,dm0,dm,dr0,dr;
2151:
2152: for ( msugar = SG(d), t = NEXT(d); t; t = NEXT(t) )
2153: if ( SG(t) < msugar ) msugar = SG(t);
2154: dm0 = 0; dr0 = 0;
1.69 noro 2155: for ( i = 0, t = d; t; t = NEXT(t) )
2156: if ( i < nd_f4_nsp && SG(t) == msugar ) {
1.63 noro 2157: if ( dm0 ) NEXT(dm) = t;
2158: else dm0 = t;
2159: dm = t;
1.69 noro 2160: i++;
1.63 noro 2161: } else {
2162: if ( dr0 ) NEXT(dr) = t;
2163: else dr0 = t;
2164: dr = t;
2165: }
2166: NEXT(dm) = 0;
2167: if ( dr0 ) NEXT(dr) = 0;
2168: *prest = dr0;
2169: return dm0;
2170: }
2171:
1.77 noro 2172: int ndv_newps(int m,NDV a,NDV aq)
1.1 noro 2173: {
1.3 noro 2174: int len;
1.13 noro 2175: RHist r;
1.20 noro 2176: NDV b;
1.3 noro 2177:
1.1 noro 2178: if ( nd_psn == nd_pslen ) {
2179: nd_pslen *= 2;
1.11 noro 2180: nd_ps = (NDV *)REALLOC((char *)nd_ps,nd_pslen*sizeof(NDV));
1.53 noro 2181: nd_ps_trace = (NDV *)REALLOC((char *)nd_ps_trace,nd_pslen*sizeof(NDV));
1.13 noro 2182: nd_psh = (RHist *)REALLOC((char *)nd_psh,nd_pslen*sizeof(RHist));
1.61 noro 2183: nd_bound = (UINT **)
2184: REALLOC((char *)nd_bound,nd_pslen*sizeof(UINT *));
1.1 noro 2185: }
1.39 noro 2186: NEWRHist(r); nd_psh[nd_psn] = r;
1.63 noro 2187: nd_ps[nd_psn] = a;
1.39 noro 2188: if ( aq ) {
1.63 noro 2189: nd_ps_trace[nd_psn] = aq;
2190: nd_bound[nd_psn] = ndv_compute_bound(aq);
1.39 noro 2191: SG(r) = SG(aq); ndl_copy(HDL(aq),DL(r));
1.53 noro 2192: } else {
1.63 noro 2193: nd_bound[nd_psn] = ndv_compute_bound(a);
1.53 noro 2194: SG(r) = SG(a); ndl_copy(HDL(a),DL(r));
1.39 noro 2195: }
1.77 noro 2196: if ( nd_demand ) {
2197: if ( aq ) {
2198: ndv_save(nd_ps_trace[nd_psn],nd_psn);
2199: nd_ps_trace[nd_psn] = 0;
2200: } else {
2201: ndv_save(nd_ps[nd_psn],nd_psn);
2202: nd_ps[nd_psn] = 0;
2203: }
2204: }
1.1 noro 2205: return nd_psn++;
2206: }
2207:
1.128 noro 2208: void ndv_setup(int mod,int trace,NODE f,int dont_sort)
1.1 noro 2209: {
1.5 noro 2210: int i,j,td,len,max;
1.1 noro 2211: NODE s,s0,f0;
1.61 noro 2212: UINT *d;
1.13 noro 2213: RHist r;
1.69 noro 2214: NDV *w;
1.61 noro 2215: NDV a,am;
1.11 noro 2216:
2217: nd_found = 0; nd_notfirst = 0; nd_create = 0;
1.1 noro 2218:
1.69 noro 2219: for ( nd_psn = 0, s = f; s; s = NEXT(s) ) if ( BDY(s) ) nd_psn++;
2220: w = (NDV *)ALLOCA(nd_psn*sizeof(NDV));
2221: for ( i = 0, s = f; s; s = NEXT(s) ) if ( BDY(s) ) w[i++] = BDY(s);
1.139 noro 2222: if ( !dont_sort ) {
2223: /* XXX heuristic */
2224: if ( !nd_ord->id && (nd_ord->ord.simple<2) )
2225: qsort(w,nd_psn,sizeof(NDV),
2226: (int (*)(const void *,const void *))ndv_compare_rev);
2227: else
2228: qsort(w,nd_psn,sizeof(NDV),
2229: (int (*)(const void *,const void *))ndv_compare);
2230: }
1.69 noro 2231: nd_pslen = 2*nd_psn;
1.11 noro 2232: nd_ps = (NDV *)MALLOC(nd_pslen*sizeof(NDV));
1.53 noro 2233: nd_ps_trace = (NDV *)MALLOC(nd_pslen*sizeof(NDV));
1.13 noro 2234: nd_psh = (RHist *)MALLOC(nd_pslen*sizeof(RHist));
1.61 noro 2235: nd_bound = (UINT **)MALLOC(nd_pslen*sizeof(UINT *));
1.57 noro 2236:
1.11 noro 2237: if ( !nd_red )
1.13 noro 2238: nd_red = (RHist *)MALLOC(REDTAB_LEN*sizeof(RHist));
1.71 noro 2239: for ( i = 0; i < REDTAB_LEN; i++ ) nd_red[i] = 0;
1.69 noro 2240: for ( i = 0; i < nd_psn; i++ ) {
1.39 noro 2241: if ( trace ) {
1.69 noro 2242: a = nd_ps_trace[i] = ndv_dup(0,w[i]);
1.61 noro 2243: ndv_removecont(0,a);
2244: am = nd_ps[i] = ndv_dup(mod,a);
2245: ndv_mod(mod,am);
2246: ndv_removecont(mod,am);
2247: } else {
1.69 noro 2248: a = nd_ps[i] = ndv_dup(mod,w[i]);
1.61 noro 2249: ndv_removecont(mod,a);
1.39 noro 2250: }
1.61 noro 2251: NEWRHist(r); SG(r) = HTD(a); ndl_copy(HDL(a),DL(r));
1.57 noro 2252: nd_bound[i] = ndv_compute_bound(a);
1.20 noro 2253: nd_psh[i] = r;
1.77 noro 2254: if ( nd_demand ) {
2255: if ( trace ) {
2256: ndv_save(nd_ps_trace[i],i);
2257: nd_ps_trace[i] = 0;
2258: } else {
2259: ndv_save(nd_ps[i],i);
2260: nd_ps[i] = 0;
2261: }
2262: }
1.20 noro 2263: }
2264: }
2265:
1.119 noro 2266: struct order_spec *append_block(struct order_spec *spec,
2267: int nv,int nalg,int ord);
2268:
1.121 noro 2269: extern VECT current_dl_weight_vector_obj;
2270: static VECT prev_weight_vector_obj;
2271:
1.120 noro 2272: void preprocess_algcoef(VL vv,VL av,struct order_spec *ord,LIST f,
2273: struct order_spec **ord1p,LIST *f1p,NODE *alistp)
2274: {
1.121 noro 2275: NODE alist,t,s,r0,r,arg;
1.120 noro 2276: VL tv;
2277: P poly;
2278: DP d;
2279: Alg alpha,dp;
2280: DAlg inv,da,hc;
2281: MP m;
1.121 noro 2282: int i,nvar,nalg,n;
1.120 noro 2283: NumberField nf;
2284: LIST f1,f2;
2285: struct order_spec *current_spec;
1.121 noro 2286: VECT obj,obj0;
2287: Obj tmp;
1.120 noro 2288:
2289: for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++);
2290: for ( nalg = 0, tv = av; tv; tv = NEXT(tv), nalg++);
2291:
2292: for ( alist = 0, tv = av; tv; tv = NEXT(tv) ) {
2293: NEXTNODE(alist,t); MKV(tv->v,poly);
2294: MKAlg(poly,alpha); BDY(t) = (pointer)alpha;
2295: tv->v = tv->v->priv;
2296: }
2297: NEXT(t) = 0;
2298:
2299: /* simplification, makeing polynomials monic */
2300: setfield_dalg(alist);
2301: obj_algtodalg(f,&f1);
2302: for ( t = BDY(f); t; t = NEXT(t) ) {
2303: initd(ord); ptod(vv,vv,(P)BDY(t),&d);
2304: hc = (DAlg)BDY(d)->c;
2305: if ( NID(hc) == N_DA ) {
2306: invdalg(hc,&inv);
2307: for ( m = BDY(d); m; m = NEXT(m) ) {
2308: muldalg(inv,(DAlg)m->c,&da); m->c = (P)da;
2309: }
2310: }
2311: initd(ord); dtop(vv,vv,d,&poly); BDY(f) = (pointer)poly;
2312: }
2313: obj_dalgtoalg(f1,&f);
2314:
2315: /* append alg vars to the var list */
2316: for ( tv = vv; NEXT(tv); tv = NEXT(tv) );
2317: NEXT(tv) = av;
2318:
2319: /* append a block to ord */
2320: *ord1p = append_block(ord,nvar,nalg,2);
2321:
2322: /* create generator list */
2323: nf = get_numberfield();
2324: for ( i = nalg-1, t = BDY(f); i >= 0; i-- ) {
2325: MKAlg(nf->defpoly[i],dp);
2326: MKNODE(s,dp,t); t = s;
2327: }
2328: MKLIST(f1,t);
2329: *alistp = alist;
2330: algobjtorat(f1,f1p);
1.121 noro 2331:
2332: /* creating a new weight vector */
2333: prev_weight_vector_obj = obj0 = current_dl_weight_vector_obj;
2334: n = nvar+nalg+1;
2335: MKVECT(obj,n);
2336: if ( obj0 && obj0->len == nvar )
2337: for ( i = 0; i < nvar; i++ ) BDY(obj)[i] = BDY(obj0)[i];
2338: else
2339: for ( i = 0; i < nvar; i++ ) BDY(obj)[i] = (pointer)ONE;
2340: for ( i = 0; i < nalg; i++ ) BDY(obj)[i+nvar] = 0;
2341: BDY(obj)[n-1] = (pointer)ONE;
2342: arg = mknode(1,obj);
2343: Pdp_set_weight(arg,&tmp);
2344: }
2345:
2346: NODE postprocess_algcoef(VL av,NODE alist,NODE r)
2347: {
2348: NODE s,t,u0,u;
2349: P p;
2350: VL tv;
2351: Obj obj,tmp;
2352: NODE arg;
2353:
2354: u0 = 0;
2355: for ( t = r; t; t = NEXT(t) ) {
2356: p = (P)BDY(t);
2357: for ( tv = av, s = alist; tv; tv = NEXT(tv), s = NEXT(s) ) {
2358: substr(CO,0,(Obj)p,tv->v,(Obj)BDY(s),&obj); p = (P)obj;
2359: }
2360: if ( OID(p) == O_P || (OID(p) == O_N && NID((Num)p) != N_A) ) {
2361: NEXTNODE(u0,u);
2362: BDY(u) = (pointer)p;
2363: }
2364: }
2365: arg = mknode(1,prev_weight_vector_obj);
2366: Pdp_set_weight(arg,&tmp);
2367:
2368: return u0;
1.120 noro 2369: }
2370:
1.63 noro 2371: void nd_gr(LIST f,LIST v,int m,int f4,struct order_spec *ord,LIST *rp)
1.1 noro 2372: {
1.119 noro 2373: VL tv,fv,vv,vc,av;
2374: NODE fd,fd0,r,r0,t,x,s,xx,alist;
2375: int e,max,nvar,i;
1.69 noro 2376: NDV b;
1.119 noro 2377: int ishomo,nalg;
2378: Alg alpha,dp;
1.144 ! noro 2379: P p,zp;
! 2380: Q dmy;
1.119 noro 2381: LIST f1,f2;
2382: Obj obj;
2383: NumberField nf;
1.120 noro 2384: struct order_spec *ord1;
1.1 noro 2385:
1.78 noro 2386: if ( !m && Demand ) nd_demand = 1;
2387: else nd_demand = 0;
2388:
1.103 noro 2389: #if 0
1.74 noro 2390: ndv_alloc = 0;
1.103 noro 2391: #endif
1.1 noro 2392: get_vars((Obj)f,&fv); pltovl(v,&vv);
1.74 noro 2393: for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++ );
1.88 noro 2394: switch ( ord->id ) {
2395: case 1:
2396: if ( ord->nv != nvar )
2397: error("nd_{gr,f4} : invalid order specification");
2398: break;
2399: default:
2400: break;
2401: }
1.119 noro 2402: nd_nalg = 0;
1.123 noro 2403: av = 0;
1.119 noro 2404: if ( !m ) {
2405: get_algtree((Obj)f,&av);
2406: for ( nalg = 0, tv = av; tv; tv = NEXT(tv), nalg++ );
1.121 noro 2407: nd_ntrans = nvar;
1.119 noro 2408: nd_nalg = nalg;
2409: /* #i -> t#i */
2410: if ( nalg ) {
1.120 noro 2411: preprocess_algcoef(vv,av,ord,f,&ord1,&f1,&alist);
2412: ord = ord1;
2413: f = f1;
1.119 noro 2414: }
1.120 noro 2415: nvar += nalg;
1.119 noro 2416: }
1.32 noro 2417: nd_init_ord(ord);
1.61 noro 2418: for ( t = BDY(f), max = 0; t; t = NEXT(t) )
2419: for ( tv = vv; tv; tv = NEXT(tv) ) {
2420: e = getdeg(tv->v,(P)BDY(t));
2421: max = MAX(e,max);
2422: }
2423: nd_setup_parameters(nvar,max);
1.77 noro 2424: ishomo = 1;
1.1 noro 2425: for ( fd0 = 0, t = BDY(f); t; t = NEXT(t) ) {
1.144 ! noro 2426: ptozp((P)BDY(t),1,&dmy,&zp);
! 2427: b = (pointer)ptondv(CO,vv,zp);
1.77 noro 2428: if ( ishomo )
2429: ishomo = ishomo && ndv_ishomo(b);
1.69 noro 2430: if ( m ) ndv_mod(m,b);
1.61 noro 2431: if ( b ) { NEXTNODE(fd0,fd); BDY(fd) = (pointer)b; }
1.1 noro 2432: }
2433: if ( fd0 ) NEXT(fd) = 0;
1.128 noro 2434: ndv_setup(m,0,fd0,0);
1.77 noro 2435: x = f4?nd_f4(m):nd_gb(m,ishomo,0);
1.82 noro 2436: nd_demand = 0;
1.61 noro 2437: x = ndv_reducebase(x);
2438: x = ndv_reduceall(m,x);
1.23 noro 2439: for ( r0 = 0, t = x; t; t = NEXT(t) ) {
1.1 noro 2440: NEXTNODE(r0,r);
1.61 noro 2441: BDY(r) = ndvtop(m,CO,vv,BDY(t));
1.20 noro 2442: }
2443: if ( r0 ) NEXT(r) = 0;
1.121 noro 2444: if ( nalg )
2445: r0 = postprocess_algcoef(av,alist,r0);
1.20 noro 2446: MKLIST(*rp,r0);
1.103 noro 2447: #if 0
1.74 noro 2448: fprintf(asir_out,"ndv_alloc=%d\n",ndv_alloc);
1.103 noro 2449: #endif
1.127 noro 2450: }
2451:
2452: void nd_gr_postproc(LIST f,LIST v,int m,struct order_spec *ord,int do_check,LIST *rp)
2453: {
2454: VL tv,fv,vv,vc,av;
2455: NODE fd,fd0,r,r0,t,x,s,xx,alist;
2456: int e,max,nvar,i;
2457: NDV b;
2458: int ishomo,nalg;
2459: Alg alpha,dp;
1.144 ! noro 2460: P p,zp;
! 2461: Q dmy;
1.127 noro 2462: LIST f1,f2;
2463: Obj obj;
2464: NumberField nf;
2465: struct order_spec *ord1;
2466:
2467: get_vars((Obj)f,&fv); pltovl(v,&vv);
2468: for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++ );
2469: switch ( ord->id ) {
2470: case 1:
2471: if ( ord->nv != nvar )
2472: error("nd_check : invalid order specification");
2473: break;
2474: default:
2475: break;
2476: }
2477: nd_nalg = 0;
2478: av = 0;
2479: if ( !m ) {
2480: get_algtree((Obj)f,&av);
2481: for ( nalg = 0, tv = av; tv; tv = NEXT(tv), nalg++ );
2482: nd_ntrans = nvar;
2483: nd_nalg = nalg;
2484: /* #i -> t#i */
2485: if ( nalg ) {
2486: preprocess_algcoef(vv,av,ord,f,&ord1,&f1,&alist);
2487: ord = ord1;
2488: f = f1;
2489: }
2490: nvar += nalg;
2491: }
2492: nd_init_ord(ord);
2493: for ( t = BDY(f), max = 0; t; t = NEXT(t) )
2494: for ( tv = vv; tv; tv = NEXT(tv) ) {
2495: e = getdeg(tv->v,(P)BDY(t));
2496: max = MAX(e,max);
2497: }
2498: nd_setup_parameters(nvar,max);
2499: ishomo = 1;
2500: for ( fd0 = 0, t = BDY(f); t; t = NEXT(t) ) {
1.144 ! noro 2501: ptozp((P)BDY(t),1,&dmy,&zp);
! 2502: b = (pointer)ptondv(CO,vv,zp);
1.127 noro 2503: if ( ishomo )
2504: ishomo = ishomo && ndv_ishomo(b);
2505: if ( m ) ndv_mod(m,b);
2506: if ( b ) { NEXTNODE(fd0,fd); BDY(fd) = (pointer)b; }
2507: }
2508: if ( fd0 ) NEXT(fd) = 0;
1.128 noro 2509: ndv_setup(m,0,fd0,0);
1.127 noro 2510: for ( x = 0, i = 0; i < nd_psn; i++ )
2511: x = update_base(x,i);
2512: if ( do_check ) {
2513: x = nd_gb(m,ishomo,1);
2514: if ( !x ) {
2515: *rp = 0;
2516: return;
2517: }
2518: } else {
2519: for ( t = x; t; t = NEXT(t) )
2520: BDY(t) = (pointer)nd_ps[(int)BDY(t)];
2521: }
2522: x = ndv_reducebase(x);
2523: x = ndv_reduceall(m,x);
2524: for ( r0 = 0, t = x; t; t = NEXT(t) ) {
2525: NEXTNODE(r0,r);
2526: BDY(r) = ndvtop(m,CO,vv,BDY(t));
2527: }
2528: if ( r0 ) NEXT(r) = 0;
2529: if ( nalg )
2530: r0 = postprocess_algcoef(av,alist,r0);
2531: MKLIST(*rp,r0);
1.20 noro 2532: }
2533:
1.133 noro 2534: void nd_gr_trace(LIST f,LIST v,int trace,int homo,int f4,struct order_spec *ord,LIST *rp)
1.20 noro 2535: {
1.119 noro 2536: VL tv,fv,vv,vc,av;
2537: NODE fd,fd0,in0,in,r,r0,t,s,cand,alist;
1.61 noro 2538: int m,nocheck,nvar,mindex,e,max;
2539: NDV c;
2540: NMV a;
1.144 ! noro 2541: P p,zp;
! 2542: Q dmy;
1.61 noro 2543: EPOS oepos;
1.119 noro 2544: int obpe,oadv,wmax,i,len,cbpe,ishomo,nalg;
2545: Alg alpha,dp;
2546: P poly;
2547: LIST f1,f2;
2548: Obj obj;
2549: NumberField nf;
1.120 noro 2550: struct order_spec *ord1;
1.20 noro 2551:
2552: get_vars((Obj)f,&fv); pltovl(v,&vv);
1.74 noro 2553: for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++ );
1.88 noro 2554: switch ( ord->id ) {
2555: case 1:
2556: if ( ord->nv != nvar )
2557: error("nd_gr_trace : invalid order specification");
2558: break;
2559: default:
2560: break;
2561: }
1.119 noro 2562:
2563: get_algtree((Obj)f,&av);
2564: for ( nalg = 0, tv = av; tv; tv = NEXT(tv), nalg++ );
1.121 noro 2565: nd_ntrans = nvar;
1.119 noro 2566: nd_nalg = nalg;
2567: /* #i -> t#i */
2568: if ( nalg ) {
1.120 noro 2569: preprocess_algcoef(vv,av,ord,f,&ord1,&f1,&alist);
2570: ord = ord1;
2571: f = f1;
1.119 noro 2572: }
1.120 noro 2573: nvar += nalg;
1.119 noro 2574:
1.52 noro 2575: nocheck = 0;
2576: mindex = 0;
2577:
1.78 noro 2578: if ( Demand ) nd_demand = 1;
2579: else nd_demand = 0;
2580:
1.52 noro 2581: /* setup modulus */
2582: if ( trace < 0 ) {
2583: trace = -trace;
2584: nocheck = 1;
2585: }
2586: m = trace > 1 ? trace : get_lprime(mindex);
1.61 noro 2587: for ( t = BDY(f), max = 0; t; t = NEXT(t) )
2588: for ( tv = vv; tv; tv = NEXT(tv) ) {
2589: e = getdeg(tv->v,(P)BDY(t));
2590: max = MAX(e,max);
1.23 noro 2591: }
1.61 noro 2592: nd_init_ord(ord);
2593: nd_setup_parameters(nvar,max);
2594: obpe = nd_bpe; oadv = nmv_adv; oepos = nd_epos;
1.82 noro 2595: ishomo = 1;
1.61 noro 2596: for ( in0 = 0, fd0 = 0, t = BDY(f); t; t = NEXT(t) ) {
1.144 ! noro 2597: ptozp((P)BDY(t),1,&dmy,&zp);
! 2598: c = (pointer)ptondv(CO,vv,zp);
1.82 noro 2599: if ( ishomo )
2600: ishomo = ishomo && ndv_ishomo(c);
1.61 noro 2601: if ( c ) {
2602: NEXTNODE(in0,in); BDY(in) = (pointer)c;
2603: NEXTNODE(fd0,fd); BDY(fd) = (pointer)ndv_dup(0,c);
1.23 noro 2604: }
1.61 noro 2605: }
2606: if ( in0 ) NEXT(in) = 0;
2607: if ( fd0 ) NEXT(fd) = 0;
1.82 noro 2608: if ( !ishomo && homo ) {
1.122 noro 2609: for ( t = in0, wmax = max; t; t = NEXT(t) ) {
1.61 noro 2610: c = (NDV)BDY(t); len = LEN(c);
2611: for ( a = BDY(c), i = 0; i < len; i++, NMV_ADV(a) )
2612: wmax = MAX(TD(DL(a)),wmax);
2613: }
2614: homogenize_order(ord,nvar,&ord1);
1.89 noro 2615: nd_init_ord(ord1);
1.61 noro 2616: nd_setup_parameters(nvar+1,wmax);
2617: for ( t = fd0; t; t = NEXT(t) )
2618: ndv_homogenize((NDV)BDY(t),obpe,oadv,oepos);
1.27 noro 2619: }
1.52 noro 2620: while ( 1 ) {
1.77 noro 2621: if ( Demand )
2622: nd_demand = 1;
1.128 noro 2623: ndv_setup(m,1,fd0,0);
1.133 noro 2624: cand = f4?nd_f4_trace(m):nd_gb_trace(m,ishomo || homo);
1.52 noro 2625: if ( !cand ) {
2626: /* failure */
1.61 noro 2627: if ( trace > 1 ) { *rp = 0; return; }
2628: else m = get_lprime(++mindex);
1.52 noro 2629: continue;
2630: }
1.82 noro 2631: if ( !ishomo && homo ) {
1.27 noro 2632: /* dehomogenization */
1.61 noro 2633: for ( t = cand; t; t = NEXT(t) ) ndv_dehomogenize((NDV)BDY(t),ord);
1.45 noro 2634: nd_init_ord(ord);
1.61 noro 2635: nd_setup_parameters(nvar,0);
1.27 noro 2636: }
1.77 noro 2637: nd_demand = 0;
1.61 noro 2638: cand = ndv_reducebase(cand);
2639: cand = ndv_reduceall(0,cand);
1.78 noro 2640: cbpe = nd_bpe;
1.61 noro 2641: if ( nocheck )
2642: break;
2643: if ( ndv_check_candidate(in0,obpe,oadv,oepos,cand) )
1.52 noro 2644: /* success */
2645: break;
2646: else if ( trace > 1 ) {
2647: /* failure */
2648: *rp = 0; return;
1.61 noro 2649: } else {
1.52 noro 2650: /* try the next modulus */
2651: m = get_lprime(++mindex);
1.61 noro 2652: /* reset the parameters */
1.82 noro 2653: if ( !ishomo && homo ) {
1.89 noro 2654: nd_init_ord(ord1);
1.61 noro 2655: nd_setup_parameters(nvar+1,wmax);
2656: } else {
2657: nd_init_ord(ord);
2658: nd_setup_parameters(nvar,max);
2659: }
2660: }
1.52 noro 2661: }
1.27 noro 2662: /* dp->p */
1.61 noro 2663: nd_bpe = cbpe;
1.73 noro 2664: nd_setup_parameters(nd_nvar,0);
1.121 noro 2665: for ( r = cand; r; r = NEXT(r) )
1.119 noro 2666: BDY(r) = (pointer)ndvtop(0,CO,vv,BDY(r));
1.121 noro 2667: if ( nalg )
2668: cand = postprocess_algcoef(av,alist,cand);
1.27 noro 2669: MKLIST(*rp,cand);
1.1 noro 2670: }
2671:
1.61 noro 2672: void dltondl(int n,DL dl,UINT *r)
1.1 noro 2673: {
1.61 noro 2674: UINT *d;
1.57 noro 2675: int i,j,l,s,ord_l;
1.43 noro 2676: struct order_pair *op;
1.1 noro 2677:
2678: d = dl->d;
1.41 noro 2679: for ( i = 0; i < nd_wpd; i++ ) r[i] = 0;
1.43 noro 2680: if ( nd_blockmask ) {
2681: l = nd_blockmask->n;
2682: op = nd_blockmask->order_pair;
2683: for ( j = 0, s = 0; j < l; j++ ) {
2684: ord_l = op[j].length;
1.57 noro 2685: for ( i = 0; i < ord_l; i++, s++ ) PUT_EXP(r,s,d[s]);
1.43 noro 2686: }
2687: TD(r) = ndl_weight(r);
1.61 noro 2688: ndl_weight_mask(r);
1.43 noro 2689: } else {
1.56 noro 2690: for ( i = 0; i < n; i++ ) PUT_EXP(r,i,d[i]);
1.43 noro 2691: TD(r) = ndl_weight(r);
2692: }
1.1 noro 2693: }
2694:
1.61 noro 2695: DL ndltodl(int n,UINT *ndl)
1.1 noro 2696: {
2697: DL dl;
2698: int *d;
1.57 noro 2699: int i,j,l,s,ord_l;
1.43 noro 2700: struct order_pair *op;
1.1 noro 2701:
2702: NEWDL(dl,n);
1.34 noro 2703: dl->td = TD(ndl);
1.1 noro 2704: d = dl->d;
1.43 noro 2705: if ( nd_blockmask ) {
2706: l = nd_blockmask->n;
2707: op = nd_blockmask->order_pair;
2708: for ( j = 0, s = 0; j < l; j++ ) {
2709: ord_l = op[j].length;
1.57 noro 2710: for ( i = 0; i < ord_l; i++, s++ ) d[s] = GET_EXP(ndl,s);
1.43 noro 2711: }
2712: } else {
1.56 noro 2713: for ( i = 0; i < n; i++ ) d[i] = GET_EXP(ndl,i);
1.43 noro 2714: }
1.1 noro 2715: return dl;
2716: }
2717:
1.61 noro 2718: void ndl_print(UINT *dl)
1.1 noro 2719: {
2720: int n;
1.57 noro 2721: int i,j,l,ord_l,s,s0;
1.43 noro 2722: struct order_pair *op;
1.1 noro 2723:
2724: n = nd_nvar;
2725: printf("<<");
1.43 noro 2726: if ( nd_blockmask ) {
2727: l = nd_blockmask->n;
2728: op = nd_blockmask->order_pair;
2729: for ( j = 0, s = s0 = 0; j < l; j++ ) {
2730: ord_l = op[j].length;
1.57 noro 2731: for ( i = 0; i < ord_l; i++, s++ )
2732: printf(s==n-1?"%d":"%d,",GET_EXP(dl,s));
1.43 noro 2733: }
2734: } else {
1.56 noro 2735: for ( i = 0; i < n; i++ ) printf(i==n-1?"%d":"%d,",GET_EXP(dl,i));
1.43 noro 2736: }
1.1 noro 2737: printf(">>");
2738: }
2739:
2740: void nd_print(ND p)
2741: {
2742: NM m;
2743:
2744: if ( !p )
2745: printf("0\n");
2746: else {
2747: for ( m = BDY(p); m; m = NEXT(m) ) {
1.71 noro 2748: if ( CM(m) & 0x80000000 ) printf("+@_%d*",IFTOF(CM(m)));
2749: else printf("+%d*",CM(m));
1.14 noro 2750: ndl_print(DL(m));
1.1 noro 2751: }
2752: printf("\n");
2753: }
2754: }
2755:
1.113 noro 2756: void nd_print_q(ND p)
1.16 noro 2757: {
2758: NM m;
2759:
2760: if ( !p )
2761: printf("0\n");
2762: else {
2763: for ( m = BDY(p); m; m = NEXT(m) ) {
2764: printf("+");
1.113 noro 2765: printexpr(CO,(Obj)CQ(m));
1.16 noro 2766: printf("*");
2767: ndl_print(DL(m));
2768: }
2769: printf("\n");
2770: }
2771: }
2772:
1.1 noro 2773: void ndp_print(ND_pairs d)
2774: {
2775: ND_pairs t;
2776:
1.34 noro 2777: for ( t = d; t; t = NEXT(t) ) printf("%d,%d ",t->i1,t->i2);
1.1 noro 2778: printf("\n");
2779: }
2780:
1.20 noro 2781: void nd_removecont(int mod,ND p)
1.16 noro 2782: {
2783: int i,n;
1.113 noro 2784: Q *w;
2785: Q dvr,t;
1.16 noro 2786: NM m;
1.21 noro 2787: struct oVECT v;
1.113 noro 2788: N q,r;
1.16 noro 2789:
1.71 noro 2790: if ( mod == -1 ) nd_mul_c(mod,p,_invsf(HCM(p)));
2791: else if ( mod ) nd_mul_c(mod,p,invm(HCM(p),mod));
1.20 noro 2792: else {
2793: for ( m = BDY(p), n = 0; m; m = NEXT(m), n++ );
1.113 noro 2794: w = (Q *)ALLOCA(n*sizeof(Q));
1.21 noro 2795: v.len = n;
2796: v.body = (pointer *)w;
1.113 noro 2797: for ( m = BDY(p), i = 0; i < n; m = NEXT(m), i++ ) w[i] = CQ(m);
1.21 noro 2798: removecont_array(w,n);
1.113 noro 2799: for ( m = BDY(p), i = 0; i < n; m = NEXT(m), i++ ) CQ(m) = w[i];
1.16 noro 2800: }
2801: }
2802:
1.21 noro 2803: void nd_removecont2(ND p1,ND p2)
2804: {
2805: int i,n1,n2,n;
1.113 noro 2806: Q *w;
2807: Q dvr,t;
1.21 noro 2808: NM m;
2809: struct oVECT v;
1.113 noro 2810: N q,r;
1.21 noro 2811:
2812: if ( !p1 ) {
2813: nd_removecont(0,p2); return;
2814: } else if ( !p2 ) {
2815: nd_removecont(0,p1); return;
2816: }
2817: n1 = nd_length(p1);
2818: n2 = nd_length(p2);
2819: n = n1+n2;
1.113 noro 2820: w = (Q *)ALLOCA(n*sizeof(Q));
1.21 noro 2821: v.len = n;
2822: v.body = (pointer *)w;
1.113 noro 2823: for ( m = BDY(p1), i = 0; i < n1; m = NEXT(m), i++ ) w[i] = CQ(m);
2824: for ( m = BDY(p2); i < n; m = NEXT(m), i++ ) w[i] = CQ(m);
1.21 noro 2825: removecont_array(w,n);
1.113 noro 2826: for ( m = BDY(p1), i = 0; i < n1; m = NEXT(m), i++ ) CQ(m) = w[i];
2827: for ( m = BDY(p2); i < n; m = NEXT(m), i++ ) CQ(m) = w[i];
1.21 noro 2828: }
2829:
1.20 noro 2830: void ndv_removecont(int mod,NDV p)
1.16 noro 2831: {
2832: int i,len;
1.113 noro 2833: Q *w;
2834: Q dvr,t;
1.16 noro 2835: NMV m;
2836:
1.71 noro 2837: if ( mod == -1 )
2838: ndv_mul_c(mod,p,_invsf(HCM(p)));
2839: else if ( mod )
1.20 noro 2840: ndv_mul_c(mod,p,invm(HCM(p),mod));
2841: else {
2842: len = p->len;
1.113 noro 2843: w = (Q *)ALLOCA(len*sizeof(Q));
2844: for ( m = BDY(p), i = 0; i < len; NMV_ADV(m), i++ ) w[i] = CQ(m);
2845: sortbynm(w,len);
2846: qltozl(w,len,&dvr);
2847: for ( m = BDY(p), i = 0; i < len; NMV_ADV(m), i++ ) {
2848: divq(CQ(m),dvr,&t); CQ(m) = t;
2849: }
1.16 noro 2850: }
1.21 noro 2851: }
2852:
1.61 noro 2853: void ndv_homogenize(NDV p,int obpe,int oadv,EPOS oepos)
2854: {
2855: int len,i,max;
2856: NMV m,mr0,mr,t;
2857:
2858: len = p->len;
2859: for ( m = BDY(p), i = 0, max = 0; i < len; NMV_OADV(m), i++ )
2860: max = MAX(max,TD(DL(m)));
2861: mr0 = nmv_adv>oadv?(NMV)REALLOC(BDY(p),len*nmv_adv):BDY(p);
2862: m = (NMV)((char *)mr0+(len-1)*oadv);
2863: mr = (NMV)((char *)mr0+(len-1)*nmv_adv);
2864: t = (NMV)ALLOCA(nmv_adv);
2865: for ( i = 0; i < len; i++, NMV_OPREV(m), NMV_PREV(mr) ) {
2866: ndl_homogenize(DL(m),DL(t),obpe,oepos,max);
1.113 noro 2867: CQ(mr) = CQ(m);
1.61 noro 2868: ndl_copy(DL(t),DL(mr));
2869: }
2870: NV(p)++;
2871: BDY(p) = mr0;
2872: }
2873:
1.45 noro 2874: void ndv_dehomogenize(NDV p,struct order_spec *ord)
1.23 noro 2875: {
1.45 noro 2876: int i,j,adj,len,newnvar,newwpd,newadv,newexporigin;
1.113 noro 2877: Q *w;
2878: Q dvr,t;
1.23 noro 2879: NMV m,r;
2880:
2881: len = p->len;
2882: newnvar = nd_nvar-1;
1.48 noro 2883: newexporigin = nd_get_exporigin(ord);
1.45 noro 2884: newwpd = newnvar/nd_epw+(newnvar%nd_epw?1:0)+newexporigin;
1.23 noro 2885: for ( m = BDY(p), i = 0; i < len; NMV_ADV(m), i++ )
1.34 noro 2886: ndl_dehomogenize(DL(m));
1.23 noro 2887: if ( newwpd != nd_wpd ) {
1.90 noro 2888: newadv = ROUND_FOR_ALIGN(sizeof(struct oNMV)+(newwpd-1)*sizeof(UINT));
1.61 noro 2889: for ( m = r = BDY(p), i = 0; i < len; NMV_ADV(m), NDV_NADV(r), i++ ) {
1.113 noro 2890: CQ(r) = CQ(m);
1.45 noro 2891: for ( j = 0; j < newexporigin; j++ ) DL(r)[j] = DL(m)[j];
2892: adj = nd_exporigin-newexporigin;
2893: for ( ; j < newwpd; j++ ) DL(r)[j] = DL(m)[j+adj];
1.23 noro 2894: }
2895: }
2896: NV(p)--;
2897: }
2898:
1.113 noro 2899: void removecont_array(Q *c,int n)
1.21 noro 2900: {
2901: struct oVECT v;
1.113 noro 2902: Q d0,d1,a,u,u1,gcd;
1.110 noro 2903: int i,j;
1.113 noro 2904: N qn,rn,gn;
2905: Q *q,*r;
1.21 noro 2906:
1.113 noro 2907: q = (Q *)ALLOCA(n*sizeof(Q));
2908: r = (Q *)ALLOCA(n*sizeof(Q));
1.21 noro 2909: v.id = O_VECT; v.len = n; v.body = (pointer *)c;
1.113 noro 2910: igcdv_estimate(&v,&d0);
2911: for ( i = 0; i < n; i++ ) {
2912: divn(NM(c[i]),NM(d0),&qn,&rn);
2913: NTOQ(qn,SGN(c[i])*SGN(d0),q[i]);
2914: NTOQ(rn,SGN(c[i]),r[i]);
2915: }
1.34 noro 2916: for ( i = 0; i < n; i++ ) if ( r[i] ) break;
1.21 noro 2917: if ( i < n ) {
2918: v.id = O_VECT; v.len = n; v.body = (pointer *)r;
1.113 noro 2919: igcdv(&v,&d1);
2920: gcdn(NM(d0),NM(d1),&gn); NTOQ(gn,1,gcd);
2921: divsn(NM(d0),gn,&qn); NTOQ(qn,1,a);
1.21 noro 2922: for ( i = 0; i < n; i++ ) {
1.113 noro 2923: mulq(a,q[i],&u);
2924: if ( r[i] ) {
2925: divsn(NM(r[i]),gn,&qn); NTOQ(qn,SGN(r[i]),u1);
2926: addq(u,u1,&q[i]);
2927: } else
2928: q[i] = u;
1.21 noro 2929: }
2930: }
1.34 noro 2931: for ( i = 0; i < n; i++ ) c[i] = q[i];
1.16 noro 2932: }
2933:
1.19 noro 2934: void nd_mul_c(int mod,ND p,int mul)
1.1 noro 2935: {
2936: NM m;
2937: int c,c1;
2938:
1.34 noro 2939: if ( !p ) return;
1.115 noro 2940: if ( mul == 1 ) return;
1.71 noro 2941: if ( mod == -1 )
2942: for ( m = BDY(p); m; m = NEXT(m) )
2943: CM(m) = _mulsf(CM(m),mul);
2944: else
2945: for ( m = BDY(p); m; m = NEXT(m) ) {
2946: c1 = CM(m); DMAR(c1,mul,0,mod,c); CM(m) = c;
2947: }
1.1 noro 2948: }
2949:
1.113 noro 2950: void nd_mul_c_q(ND p,Q mul)
1.16 noro 2951: {
2952: NM m;
1.113 noro 2953: Q c;
1.16 noro 2954:
1.113 noro 2955: if ( !p ) return;
1.115 noro 2956: if ( UNIQ(mul) ) return;
1.16 noro 2957: for ( m = BDY(p); m; m = NEXT(m) ) {
1.113 noro 2958: mulq(CQ(m),mul,&c); CQ(m) = c;
1.16 noro 2959: }
2960: }
2961:
1.61 noro 2962: void nd_mul_c_p(VL vl,ND p,P mul)
2963: {
2964: NM m;
2965: P c;
2966:
2967: if ( !p ) return;
2968: for ( m = BDY(p); m; m = NEXT(m) ) {
2969: mulp(vl,CP(m),mul,&c); CP(m) = c;
2970: }
2971: }
2972:
1.1 noro 2973: void nd_free(ND p)
2974: {
2975: NM t,s;
2976:
1.34 noro 2977: if ( !p ) return;
1.1 noro 2978: t = BDY(p);
2979: while ( t ) {
2980: s = NEXT(t);
2981: FREENM(t);
2982: t = s;
2983: }
2984: FREEND(p);
2985: }
2986:
1.23 noro 2987: void ndv_free(NDV p)
2988: {
2989: GC_free(BDY(p));
2990: }
2991:
1.61 noro 2992: void nd_append_red(UINT *d,int i)
1.1 noro 2993: {
1.13 noro 2994: RHist m,m0;
1.1 noro 2995: int h;
2996:
1.13 noro 2997: NEWRHist(m);
1.34 noro 2998: h = ndl_hash_value(d);
1.13 noro 2999: m->index = i;
1.14 noro 3000: ndl_copy(d,DL(m));
1.1 noro 3001: NEXT(m) = nd_red[h];
3002: nd_red[h] = m;
3003: }
3004:
1.61 noro 3005: UINT *ndv_compute_bound(NDV p)
1.1 noro 3006: {
1.61 noro 3007: UINT *d1,*d2,*t;
3008: UINT u;
1.57 noro 3009: int i,j,k,l,len,ind;
1.45 noro 3010: NMV m;
1.1 noro 3011:
3012: if ( !p )
3013: return 0;
1.61 noro 3014: d1 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
3015: d2 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
1.45 noro 3016: len = LEN(p);
3017: m = BDY(p); ndl_copy(DL(m),d1); NMV_ADV(m);
3018: for ( i = 1; i < len; i++, NMV_ADV(m) ) {
1.14 noro 3019: ndl_lcm(DL(m),d1,d2);
1.1 noro 3020: t = d1; d1 = d2; d2 = t;
3021: }
1.12 noro 3022: l = nd_nvar+31;
1.61 noro 3023: t = (UINT *)MALLOC_ATOMIC(l*sizeof(UINT));
1.57 noro 3024: for ( i = nd_exporigin, ind = 0; i < nd_wpd; i++ ) {
3025: u = d1[i];
3026: k = (nd_epw-1)*nd_bpe;
3027: for ( j = 0; j < nd_epw; j++, k -= nd_bpe, ind++ )
3028: t[ind] = (u>>k)&nd_mask0;
3029: }
3030: for ( ; ind < l; ind++ ) t[ind] = 0;
1.1 noro 3031: return t;
3032: }
3033:
1.99 noro 3034: UINT *nd_compute_bound(ND p)
3035: {
3036: UINT *d1,*d2,*t;
3037: UINT u;
3038: int i,j,k,l,len,ind;
3039: NM m;
3040:
3041: if ( !p )
3042: return 0;
3043: d1 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
3044: d2 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
3045: len = LEN(p);
3046: m = BDY(p); ndl_copy(DL(m),d1); m = NEXT(m);
3047: for ( m = NEXT(m); m; m = NEXT(m) ) {
3048: ndl_lcm(DL(m),d1,d2);
3049: t = d1; d1 = d2; d2 = t;
3050: }
3051: l = nd_nvar+31;
3052: t = (UINT *)MALLOC_ATOMIC(l*sizeof(UINT));
3053: for ( i = nd_exporigin, ind = 0; i < nd_wpd; i++ ) {
3054: u = d1[i];
3055: k = (nd_epw-1)*nd_bpe;
3056: for ( j = 0; j < nd_epw; j++, k -= nd_bpe, ind++ )
3057: t[ind] = (u>>k)&nd_mask0;
3058: }
3059: for ( ; ind < l; ind++ ) t[ind] = 0;
3060: return t;
3061: }
3062:
1.48 noro 3063: int nd_get_exporigin(struct order_spec *ord)
3064: {
1.51 noro 3065: switch ( ord->id ) {
1.96 noro 3066: case 0: case 2:
1.48 noro 3067: return 1;
1.41 noro 3068: case 1:
3069: /* block order */
1.43 noro 3070: /* d[0]:weight d[1]:w0,...,d[nd_exporigin-1]:w(n-1) */
1.48 noro 3071: return ord->ord.block.length+1;
1.96 noro 3072: case 3:
3073: error("nd_get_exporigin : composite order is not supported yet.");
1.41 noro 3074: }
1.48 noro 3075: }
3076:
1.61 noro 3077: void nd_setup_parameters(int nvar,int max) {
1.79 noro 3078: int i,j,n,elen,ord_o,ord_l,l,s,wpd;
1.57 noro 3079: struct order_pair *op;
1.48 noro 3080:
1.73 noro 3081: nd_nvar = nvar;
1.79 noro 3082: if ( max ) {
1.83 noro 3083: /* XXX */
3084: if ( do_weyl ) nd_bpe = 32;
3085: else if ( max < 2 ) nd_bpe = 1;
1.79 noro 3086: else if ( max < 4 ) nd_bpe = 2;
3087: else if ( max < 8 ) nd_bpe = 3;
3088: else if ( max < 16 ) nd_bpe = 4;
3089: else if ( max < 32 ) nd_bpe = 5;
3090: else if ( max < 64 ) nd_bpe = 6;
3091: else if ( max < 256 ) nd_bpe = 8;
3092: else if ( max < 1024 ) nd_bpe = 10;
3093: else if ( max < 65536 ) nd_bpe = 16;
3094: else nd_bpe = 32;
3095: }
1.61 noro 3096: nd_epw = (sizeof(UINT)*8)/nd_bpe;
1.48 noro 3097: elen = nd_nvar/nd_epw+(nd_nvar%nd_epw?1:0);
3098: nd_exporigin = nd_get_exporigin(nd_ord);
1.79 noro 3099: wpd = nd_exporigin+elen;
3100: if ( wpd != nd_wpd ) {
3101: nd_free_private_storage();
3102: nd_wpd = wpd;
3103: }
1.1 noro 3104: if ( nd_bpe < 32 ) {
3105: nd_mask0 = (1<<nd_bpe)-1;
3106: } else {
3107: nd_mask0 = 0xffffffff;
3108: }
3109: bzero(nd_mask,sizeof(nd_mask));
3110: nd_mask1 = 0;
3111: for ( i = 0; i < nd_epw; i++ ) {
3112: nd_mask[nd_epw-i-1] = (nd_mask0<<(i*nd_bpe));
3113: nd_mask1 |= (1<<(nd_bpe-1))<<(i*nd_bpe);
3114: }
1.90 noro 3115: nmv_adv = ROUND_FOR_ALIGN(sizeof(struct oNMV)+(nd_wpd-1)*sizeof(UINT));
1.57 noro 3116: nd_epos = nd_create_epos(nd_ord);
1.43 noro 3117: nd_blockmask = nd_create_blockmask(nd_ord);
1.96 noro 3118: nd_work_vector = (int *)REALLOC(nd_work_vector,nd_nvar*sizeof(int));
1.1 noro 3119: }
3120:
1.103 noro 3121: ND_pairs nd_reconstruct(int trace,ND_pairs d)
1.1 noro 3122: {
1.37 noro 3123: int i,obpe,oadv,h;
1.71 noro 3124: static NM prev_nm_free_list;
3125: static ND_pairs prev_ndp_free_list;
1.13 noro 3126: RHist mr0,mr;
3127: RHist r;
1.37 noro 3128: RHist *old_red;
1.71 noro 3129: ND_pairs s0,s,t;
1.43 noro 3130: EPOS oepos;
1.15 noro 3131:
1.1 noro 3132: obpe = nd_bpe;
1.11 noro 3133: oadv = nmv_adv;
1.43 noro 3134: oepos = nd_epos;
1.63 noro 3135: if ( obpe < 2 ) nd_bpe = 2;
3136: else if ( obpe < 3 ) nd_bpe = 3;
1.62 noro 3137: else if ( obpe < 4 ) nd_bpe = 4;
1.63 noro 3138: else if ( obpe < 5 ) nd_bpe = 5;
1.34 noro 3139: else if ( obpe < 6 ) nd_bpe = 6;
3140: else if ( obpe < 8 ) nd_bpe = 8;
1.63 noro 3141: else if ( obpe < 10 ) nd_bpe = 10;
1.34 noro 3142: else if ( obpe < 16 ) nd_bpe = 16;
3143: else if ( obpe < 32 ) nd_bpe = 32;
3144: else error("nd_reconstruct : exponent too large");
1.5 noro 3145:
1.73 noro 3146: nd_setup_parameters(nd_nvar,0);
1.1 noro 3147: prev_nm_free_list = _nm_free_list;
3148: prev_ndp_free_list = _ndp_free_list;
3149: _nm_free_list = 0;
3150: _ndp_free_list = 0;
1.53 noro 3151: for ( i = nd_psn-1; i >= 0; i-- ) ndv_realloc(nd_ps[i],obpe,oadv,oepos);
3152: if ( trace )
3153: for ( i = nd_psn-1; i >= 0; i-- )
3154: ndv_realloc(nd_ps_trace[i],obpe,oadv,oepos);
1.1 noro 3155: s0 = 0;
3156: for ( t = d; t; t = NEXT(t) ) {
3157: NEXTND_pairs(s0,s);
3158: s->i1 = t->i1;
3159: s->i2 = t->i2;
1.14 noro 3160: SG(s) = SG(t);
1.61 noro 3161: ndl_reconstruct(LCM(t),LCM(s),obpe,oepos);
1.1 noro 3162: }
1.37 noro 3163:
3164: old_red = (RHist *)ALLOCA(REDTAB_LEN*sizeof(RHist));
1.6 noro 3165: for ( i = 0; i < REDTAB_LEN; i++ ) {
1.37 noro 3166: old_red[i] = nd_red[i];
3167: nd_red[i] = 0;
3168: }
3169: for ( i = 0; i < REDTAB_LEN; i++ )
3170: for ( r = old_red[i]; r; r = NEXT(r) ) {
3171: NEWRHist(mr);
1.13 noro 3172: mr->index = r->index;
1.20 noro 3173: SG(mr) = SG(r);
1.61 noro 3174: ndl_reconstruct(DL(r),DL(mr),obpe,oepos);
1.37 noro 3175: h = ndl_hash_value(DL(mr));
3176: NEXT(mr) = nd_red[h];
3177: nd_red[h] = mr;
1.6 noro 3178: }
1.37 noro 3179: for ( i = 0; i < REDTAB_LEN; i++ ) old_red[i] = 0;
3180: old_red = 0;
1.11 noro 3181: for ( i = 0; i < nd_psn; i++ ) {
1.20 noro 3182: NEWRHist(r); SG(r) = SG(nd_psh[i]);
1.61 noro 3183: ndl_reconstruct(DL(nd_psh[i]),DL(r),obpe,oepos);
1.13 noro 3184: nd_psh[i] = r;
1.11 noro 3185: }
1.1 noro 3186: if ( s0 ) NEXT(s) = 0;
3187: prev_nm_free_list = 0;
3188: prev_ndp_free_list = 0;
1.71 noro 3189: #if 0
1.1 noro 3190: GC_gcollect();
1.71 noro 3191: #endif
1.1 noro 3192: return s0;
3193: }
3194:
1.61 noro 3195: void ndl_reconstruct(UINT *d,UINT *r,int obpe,EPOS oepos)
1.1 noro 3196: {
1.57 noro 3197: int n,i,ei,oepw,omask0,j,s,ord_l,l;
1.43 noro 3198: struct order_pair *op;
1.1 noro 3199:
3200: n = nd_nvar;
1.61 noro 3201: oepw = (sizeof(UINT)*8)/obpe;
1.43 noro 3202: omask0 = (1<<obpe)-1;
1.34 noro 3203: TD(r) = TD(d);
1.41 noro 3204: for ( i = nd_exporigin; i < nd_wpd; i++ ) r[i] = 0;
1.43 noro 3205: if ( nd_blockmask ) {
3206: l = nd_blockmask->n;
3207: op = nd_blockmask->order_pair;
3208: for ( i = 1; i < nd_exporigin; i++ )
3209: r[i] = d[i];
3210: for ( j = 0, s = 0; j < l; j++ ) {
3211: ord_l = op[j].length;
1.57 noro 3212: for ( i = 0; i < ord_l; i++, s++ ) {
3213: ei = GET_EXP_OLD(d,s);
3214: PUT_EXP(r,s,ei);
3215: }
1.1 noro 3216: }
1.43 noro 3217: } else {
1.56 noro 3218: for ( i = 0; i < n; i++ ) {
3219: ei = GET_EXP_OLD(d,i);
3220: PUT_EXP(r,i,ei);
3221: }
1.1 noro 3222: }
3223: }
1.3 noro 3224:
1.6 noro 3225: ND nd_copy(ND p)
3226: {
3227: NM m,mr,mr0;
1.41 noro 3228: int c,n;
1.6 noro 3229: ND r;
3230:
3231: if ( !p )
3232: return 0;
3233: else {
3234: for ( mr0 = 0, m = BDY(p); m; m = NEXT(m) ) {
3235: NEXTNM(mr0,mr);
1.14 noro 3236: CM(mr) = CM(m);
3237: ndl_copy(DL(m),DL(mr));
1.6 noro 3238: }
3239: NEXT(mr) = 0;
1.31 noro 3240: MKND(NV(p),mr0,LEN(p),r);
1.14 noro 3241: SG(r) = SG(p);
1.6 noro 3242: return r;
3243: }
3244: }
3245:
1.53 noro 3246: int nd_sp(int mod,int trace,ND_pairs p,ND *rp)
1.11 noro 3247: {
3248: NM m;
3249: NDV p1,p2;
3250: ND t1,t2;
1.61 noro 3251: UINT *lcm;
1.31 noro 3252: int td;
1.11 noro 3253:
1.77 noro 3254: if ( !mod && nd_demand ) {
3255: p1 = ndv_load(p->i1); p2 = ndv_load(p->i2);
1.53 noro 3256: } else {
1.77 noro 3257: if ( trace ) {
3258: p1 = nd_ps_trace[p->i1]; p2 = nd_ps_trace[p->i2];
3259: } else {
3260: p1 = nd_ps[p->i1]; p2 = nd_ps[p->i2];
3261: }
1.20 noro 3262: }
1.34 noro 3263: lcm = LCM(p);
1.11 noro 3264: NEWNM(m);
1.113 noro 3265: CQ(m) = HCQ(p2);
1.34 noro 3266: ndl_sub(lcm,HDL(p1),DL(m));
1.56 noro 3267: if ( ndl_check_bound2(p->i1,DL(m)) )
3268: return 0;
1.55 noro 3269: t1 = ndv_mul_nm(mod,m,p1);
1.71 noro 3270: if ( mod == -1 ) CM(m) = _chsgnsf(HCM(p1));
3271: else if ( mod ) CM(m) = mod-HCM(p1);
1.113 noro 3272: else chsgnq(HCQ(p1),&CQ(m));
1.34 noro 3273: ndl_sub(lcm,HDL(p2),DL(m));
1.14 noro 3274: if ( ndl_check_bound2(p->i2,DL(m)) ) {
1.11 noro 3275: nd_free(t1);
3276: return 0;
3277: }
1.55 noro 3278: t2 = ndv_mul_nm(mod,m,p2);
1.31 noro 3279: *rp = nd_add(mod,t1,t2);
1.11 noro 3280: FREENM(m);
3281: return 1;
3282: }
3283:
1.19 noro 3284: void ndv_mul_c(int mod,NDV p,int mul)
1.11 noro 3285: {
3286: NMV m;
3287: int c,c1,len,i;
3288:
1.34 noro 3289: if ( !p ) return;
1.14 noro 3290: len = LEN(p);
1.71 noro 3291: if ( mod == -1 )
3292: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) )
3293: CM(m) = _mulsf(CM(m),mul);
3294: else
3295: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
3296: c1 = CM(m); DMAR(c1,mul,0,mod,c); CM(m) = c;
3297: }
1.11 noro 3298: }
3299:
1.113 noro 3300: void ndv_mul_c_q(NDV p,Q mul)
1.16 noro 3301: {
3302: NMV m;
1.113 noro 3303: Q c;
1.16 noro 3304: int len,i;
3305:
1.34 noro 3306: if ( !p ) return;
1.16 noro 3307: len = LEN(p);
3308: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
1.113 noro 3309: mulq(CQ(m),mul,&c); CQ(m) = c;
1.16 noro 3310: }
3311: }
3312:
1.55 noro 3313: ND weyl_ndv_mul_nm(int mod,NM m0,NDV p) {
3314: int n2,i,j,l,n,tlen;
1.61 noro 3315: UINT *d0;
1.55 noro 3316: NM *tab,*psum;
3317: ND s,r;
3318: NM t;
3319: NMV m1;
3320:
3321: if ( !p ) return 0;
3322: n = NV(p); n2 = n>>1;
3323: d0 = DL(m0);
3324: l = LEN(p);
3325: for ( i = 0, tlen = 1; i < n2; i++ ) tlen *= (GET_EXP(d0,n2+i)+1);
3326: tab = (NM *)ALLOCA(tlen*sizeof(NM));
3327: psum = (NM *)ALLOCA(tlen*sizeof(NM));
3328: for ( i = 0; i < tlen; i++ ) psum[i] = 0;
1.56 noro 3329: m1 = (NMV)(((char *)BDY(p))+nmv_adv*(l-1));
3330: for ( i = l-1; i >= 0; i--, NMV_PREV(m1) ) {
1.55 noro 3331: /* m0(NM) * m1(NMV) => tab(NM) */
1.56 noro 3332: weyl_mul_nm_nmv(n,mod,m0,m1,tab,tlen);
1.55 noro 3333: for ( j = 0; j < tlen; j++ ) {
3334: if ( tab[j] ) {
3335: NEXT(tab[j]) = psum[j]; psum[j] = tab[j];
3336: }
3337: }
3338: }
3339: for ( i = tlen-1, r = 0; i >= 0; i-- )
3340: if ( psum[i] ) {
3341: for ( j = 0, t = psum[i]; t; t = NEXT(t), j++ );
3342: MKND(n,psum[i],j,s);
3343: r = nd_add(mod,r,s);
3344: }
1.56 noro 3345: if ( r ) SG(r) = SG(p)+TD(d0);
3346: return r;
1.55 noro 3347: }
3348:
1.56 noro 3349: /* product of monomials */
3350: /* XXX block order is not handled correctly */
3351:
1.55 noro 3352: void weyl_mul_nm_nmv(int n,int mod,NM m0,NMV m1,NM *tab,int tlen)
3353: {
1.56 noro 3354: int i,n2,j,s,curlen,homo,h,a,b,k,l,u,min;
1.61 noro 3355: UINT *d0,*d1,*d,*dt,*ctab;
1.113 noro 3356: Q *ctab_q;
3357: Q q,q1;
1.61 noro 3358: UINT c0,c1,c;
1.55 noro 3359: NM *p;
3360: NM m,t;
3361:
3362: for ( i = 0; i < tlen; i++ ) tab[i] = 0;
3363: if ( !m0 || !m1 ) return;
3364: d0 = DL(m0); d1 = DL(m1); n2 = n>>1;
3365: NEWNM(m); d = DL(m);
1.56 noro 3366: if ( mod ) {
3367: c0 = CM(m0); c1 = CM(m1); DMAR(c0,c1,0,mod,c); CM(m) = c;
3368: } else
1.113 noro 3369: mulq(CQ(m0),CQ(m1),&CQ(m));
1.55 noro 3370: for ( i = 0; i < nd_wpd; i++ ) d[i] = 0;
3371: homo = n&1 ? 1 : 0;
3372: if ( homo ) {
3373: /* offset of h-degree */
3374: h = GET_EXP(d0,n-1)+GET_EXP(d1,n-1);
3375: PUT_EXP(DL(m),n-1,h);
3376: TD(DL(m)) = h;
1.61 noro 3377: if ( nd_blockmask ) ndl_weight_mask(DL(m));
1.55 noro 3378: }
3379: tab[0] = m;
3380: NEWNM(m); d = DL(m);
1.57 noro 3381: for ( i = 0, curlen = 1; i < n2; i++ ) {
1.55 noro 3382: a = GET_EXP(d0,i); b = GET_EXP(d1,n2+i);
3383: k = GET_EXP(d0,n2+i); l = GET_EXP(d1,i);
3384: /* xi^a*(Di^k*xi^l)*Di^b */
3385: a += l; b += k;
1.56 noro 3386: s = MUL_WEIGHT(a,i)+MUL_WEIGHT(b,n2+i);
1.55 noro 3387: if ( !k || !l ) {
3388: for ( j = 0; j < curlen; j++ )
1.56 noro 3389: if ( t = tab[j] ) {
3390: dt = DL(t);
3391: PUT_EXP(dt,i,a); PUT_EXP(dt,n2+i,b); TD(dt) += s;
1.61 noro 3392: if ( nd_blockmask ) ndl_weight_mask(dt);
1.55 noro 3393: }
3394: curlen *= k+1;
3395: continue;
3396: }
3397: min = MIN(k,l);
1.56 noro 3398: if ( mod ) {
1.61 noro 3399: ctab = (UINT *)ALLOCA((min+1)*sizeof(UINT));
1.56 noro 3400: mkwcm(k,l,mod,ctab);
3401: } else {
1.113 noro 3402: ctab_q = (Q *)ALLOCA((min+1)*sizeof(Q));
3403: mkwc(k,l,ctab_q);
1.56 noro 3404: }
1.57 noro 3405: for ( j = min; j >= 0; j-- ) {
1.56 noro 3406: for ( u = 0; u < nd_wpd; u++ ) d[u] = 0;
1.55 noro 3407: PUT_EXP(d,i,a-j); PUT_EXP(d,n2+i,b-j);
1.56 noro 3408: h = MUL_WEIGHT(a-j,i)+MUL_WEIGHT(b-j,n2+i);
1.55 noro 3409: if ( homo ) {
3410: TD(d) = s;
1.56 noro 3411: PUT_EXP(d,n-1,s-h);
1.55 noro 3412: } else TD(d) = h;
1.61 noro 3413: if ( nd_blockmask ) ndl_weight_mask(d);
1.56 noro 3414: if ( mod ) c = ctab[j];
1.113 noro 3415: else q = ctab_q[j];
1.57 noro 3416: p = tab+curlen*j;
3417: if ( j == 0 ) {
3418: for ( u = 0; u < curlen; u++, p++ ) {
3419: if ( tab[u] ) {
3420: ndl_addto(DL(tab[u]),d);
3421: if ( mod ) {
3422: c0 = CM(tab[u]); DMAR(c0,c,0,mod,c1); CM(tab[u]) = c1;
3423: } else {
1.113 noro 3424: mulq(CQ(tab[u]),q,&q1); CQ(tab[u]) = q1;
1.57 noro 3425: }
3426: }
1.56 noro 3427: }
1.57 noro 3428: } else {
3429: for ( u = 0; u < curlen; u++, p++ ) {
3430: if ( tab[u] ) {
3431: NEWNM(t);
3432: ndl_add(DL(tab[u]),d,DL(t));
3433: if ( mod ) {
3434: c0 = CM(tab[u]); DMAR(c0,c,0,mod,c1); CM(t) = c1;
3435: } else
1.113 noro 3436: mulq(CQ(tab[u]),q,&CQ(t));
1.57 noro 3437: *p = t;
3438: }
1.55 noro 3439: }
3440: }
3441: }
3442: curlen *= k+1;
3443: }
3444: FREENM(m);
3445: }
3446:
1.63 noro 3447: ND ndv_mul_nm_symbolic(NM m0,NDV p)
3448: {
3449: NM mr,mr0;
3450: NMV m;
3451: UINT *d,*dt,*dm;
3452: int c,n,td,i,c1,c2,len;
1.113 noro 3453: Q q;
1.63 noro 3454: ND r;
3455:
3456: if ( !p ) return 0;
3457: else {
3458: n = NV(p); m = BDY(p);
3459: d = DL(m0);
3460: len = LEN(p);
3461: mr0 = 0;
3462: td = TD(d);
3463: c = CM(m0);
3464: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
3465: NEXTNM(mr0,mr);
3466: CM(mr) = 1;
3467: ndl_add(DL(m),d,DL(mr));
3468: }
3469: NEXT(mr) = 0;
3470: MKND(NV(p),mr0,len,r);
3471: SG(r) = SG(p) + TD(d);
3472: return r;
3473: }
3474: }
3475:
1.55 noro 3476: ND ndv_mul_nm(int mod,NM m0,NDV p)
1.9 noro 3477: {
3478: NM mr,mr0;
3479: NMV m;
1.61 noro 3480: UINT *d,*dt,*dm;
1.9 noro 3481: int c,n,td,i,c1,c2,len;
1.113 noro 3482: Q q;
1.9 noro 3483: ND r;
3484:
1.34 noro 3485: if ( !p ) return 0;
1.55 noro 3486: else if ( do_weyl )
1.71 noro 3487: if ( mod == -1 )
3488: error("ndv_mul_nm : not implemented (weyl)");
3489: else
3490: return weyl_ndv_mul_nm(mod,m0,p);
1.9 noro 3491: else {
3492: n = NV(p); m = BDY(p);
1.34 noro 3493: d = DL(m0);
1.14 noro 3494: len = LEN(p);
1.9 noro 3495: mr0 = 0;
1.34 noro 3496: td = TD(d);
1.71 noro 3497: if ( mod == -1 ) {
3498: c = CM(m0);
3499: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
3500: NEXTNM(mr0,mr);
3501: CM(mr) = _mulsf(CM(m),c);
3502: ndl_add(DL(m),d,DL(mr));
3503: }
3504: } else if ( mod ) {
1.16 noro 3505: c = CM(m0);
3506: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
3507: NEXTNM(mr0,mr);
3508: c1 = CM(m);
1.19 noro 3509: DMAR(c1,c,0,mod,c2);
1.16 noro 3510: CM(mr) = c2;
3511: ndl_add(DL(m),d,DL(mr));
3512: }
3513: } else {
1.113 noro 3514: q = CQ(m0);
1.16 noro 3515: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
3516: NEXTNM(mr0,mr);
1.113 noro 3517: mulq(CQ(m),q,&CQ(mr));
1.16 noro 3518: ndl_add(DL(m),d,DL(mr));
3519: }
1.4 noro 3520: }
1.9 noro 3521: NEXT(mr) = 0;
1.31 noro 3522: MKND(NV(p),mr0,len,r);
1.34 noro 3523: SG(r) = SG(p) + TD(d);
1.9 noro 3524: return r;
1.4 noro 3525: }
3526: }
3527:
1.104 noro 3528: ND nd_quo(int mod,PGeoBucket bucket,NDV d)
1.99 noro 3529: {
3530: NM mq0,mq;
1.102 noro 3531: NMV tm;
1.113 noro 3532: Q q;
1.104 noro 3533: int i,nv,sg,c,c1,c2,hindex;
3534: ND p,t,r;
3535: N tnm;
3536:
1.124 noro 3537: if ( bucket->m < 0 ) return 0;
1.99 noro 3538: else {
1.104 noro 3539: nv = NV(d);
1.99 noro 3540: mq0 = 0;
1.102 noro 3541: tm = (NMV)ALLOCA(nmv_adv);
1.104 noro 3542: while ( 1 ) {
1.113 noro 3543: hindex = mod?head_pbucket(mod,bucket):head_pbucket_q(bucket);
1.104 noro 3544: if ( hindex < 0 ) break;
3545: p = bucket->body[hindex];
1.99 noro 3546: NEXTNM(mq0,mq);
3547: ndl_sub(HDL(p),HDL(d),DL(mq));
1.102 noro 3548: ndl_copy(DL(mq),DL(tm));
3549: if ( mod ) {
3550: c1 = invm(HCM(d),mod); c2 = HCM(p);
3551: DMAR(c1,c2,0,mod,c); CM(mq) = c;
3552: CM(tm) = mod-c;
3553: } else {
1.113 noro 3554: divsn(NM(HCQ(p)),NM(HCQ(d)),&tnm);
3555: NTOQ(tnm,SGN(HCQ(p))*SGN(HCQ(d)),CQ(mq));
3556: chsgnq(CQ(mq),&CQ(tm));
1.102 noro 3557: }
3558: t = ndv_mul_nmv_trunc(mod,tm,d,HDL(d));
1.104 noro 3559: bucket->body[hindex] = nd_remove_head(p);
3560: t = nd_remove_head(t);
3561: add_pbucket(mod,bucket,t);
3562: }
3563: if ( !mq0 )
3564: r = 0;
3565: else {
3566: NEXT(mq) = 0;
3567: for ( i = 0, mq = mq0; mq; mq = NEXT(mq), i++ );
3568: MKND(nv,mq0,i,r);
3569: /* XXX */
3570: SG(r) = HTD(r);
1.99 noro 3571: }
3572: return r;
3573: }
3574: }
3575:
1.43 noro 3576: void ndv_realloc(NDV p,int obpe,int oadv,EPOS oepos)
1.11 noro 3577: {
1.13 noro 3578: NMV m,mr,mr0,t;
3579: int len,i,k;
1.11 noro 3580:
1.61 noro 3581: if ( !p ) return;
3582: m = BDY(p); len = LEN(p);
3583: mr0 = nmv_adv>oadv?(NMV)REALLOC(BDY(p),len*nmv_adv):BDY(p);
3584: m = (NMV)((char *)mr0+(len-1)*oadv);
3585: mr = (NMV)((char *)mr0+(len-1)*nmv_adv);
3586: t = (NMV)ALLOCA(nmv_adv);
3587: for ( i = 0; i < len; i++, NMV_OPREV(m), NMV_PREV(mr) ) {
1.113 noro 3588: CQ(t) = CQ(m);
1.61 noro 3589: for ( k = 0; k < nd_wpd; k++ ) DL(t)[k] = 0;
3590: ndl_reconstruct(DL(m),DL(t),obpe,oepos);
1.113 noro 3591: CQ(mr) = CQ(t);
1.61 noro 3592: ndl_copy(DL(t),DL(mr));
3593: }
3594: BDY(p) = mr0;
3595: }
3596:
3597: NDV ndv_dup_realloc(NDV p,int obpe,int oadv,EPOS oepos)
3598: {
3599: NMV m,mr,mr0;
3600: int len,i;
3601: NDV r;
1.11 noro 3602:
1.61 noro 3603: if ( !p ) return 0;
3604: m = BDY(p); len = LEN(p);
3605: mr0 = mr = (NMV)MALLOC(len*nmv_adv);
3606: for ( i = 0; i < len; i++, NMV_OADV(m), NMV_ADV(mr) ) {
3607: ndl_zero(DL(mr));
3608: ndl_reconstruct(DL(m),DL(mr),obpe,oepos);
1.113 noro 3609: CQ(mr) = CQ(m);
1.11 noro 3610: }
1.61 noro 3611: MKNDV(NV(p),mr0,len,r);
3612: SG(r) = SG(p);
3613: return r;
1.11 noro 3614: }
3615:
1.61 noro 3616: /* duplicate p */
3617:
3618: NDV ndv_dup(int mod,NDV p)
1.3 noro 3619: {
3620: NDV d;
1.61 noro 3621: NMV t,m,m0;
1.3 noro 3622: int i,len;
3623:
1.34 noro 3624: if ( !p ) return 0;
1.31 noro 3625: len = LEN(p);
1.34 noro 3626: m0 = m = (NMV)(mod?MALLOC_ATOMIC(len*nmv_adv):MALLOC(len*nmv_adv));
1.61 noro 3627: for ( t = BDY(p), i = 0; i < len; i++, NMV_ADV(t), NMV_ADV(m) ) {
1.14 noro 3628: ndl_copy(DL(t),DL(m));
1.113 noro 3629: CQ(m) = CQ(t);
1.3 noro 3630: }
3631: MKNDV(NV(p),m0,len,d);
1.23 noro 3632: SG(d) = SG(p);
3633: return d;
3634: }
3635:
1.63 noro 3636: ND nd_dup(ND p)
3637: {
3638: ND d;
3639: NM t,m,m0;
3640:
3641: if ( !p ) return 0;
3642: for ( m0 = 0, t = BDY(p); t; t = NEXT(t) ) {
3643: NEXTNM(m0,m);
3644: ndl_copy(DL(t),DL(m));
1.113 noro 3645: CQ(m) = CQ(t);
1.63 noro 3646: }
3647: if ( m0 ) NEXT(m) = 0;
3648: MKND(NV(p),m0,LEN(p),d);
3649: SG(d) = SG(p);
3650: return d;
3651: }
3652:
1.61 noro 3653: /* XXX if p->len == 0 then it represents 0 */
3654:
3655: void ndv_mod(int mod,NDV p)
3656: {
3657: NMV t,d;
1.125 noro 3658: int r,s,u;
1.61 noro 3659: int i,len,dlen;
1.71 noro 3660: Obj gfs;
1.61 noro 3661:
3662: if ( !p ) return;
3663: len = LEN(p);
3664: dlen = 0;
1.71 noro 3665: if ( mod == -1 )
3666: for ( t = d = BDY(p), i = 0; i < len; i++, NMV_ADV(t) ) {
3667: simp_ff((Obj)CP(t),&gfs);
3668: r = FTOIF(CONT((GFS)gfs));
1.61 noro 3669: CM(d) = r;
3670: ndl_copy(DL(t),DL(d));
3671: NMV_ADV(d);
3672: dlen++;
3673: }
1.71 noro 3674: else
3675: for ( t = d = BDY(p), i = 0; i < len; i++, NMV_ADV(t) ) {
1.113 noro 3676: r = rem(NM(CQ(t)),mod);
1.71 noro 3677: if ( r ) {
1.113 noro 3678: if ( SGN(CQ(t)) < 0 )
3679: r = mod-r;
1.125 noro 3680: if ( DN(CQ(t)) ) {
3681: s = rem(DN(CQ(t)),mod);
3682: if ( !s )
3683: error("ndv_mod : division by 0");
3684: s = invm(s,mod);
3685: DMAR(r,s,0,mod,u); r = u;
3686: }
1.71 noro 3687: CM(d) = r;
3688: ndl_copy(DL(t),DL(d));
3689: NMV_ADV(d);
3690: dlen++;
3691: }
3692: }
1.61 noro 3693: LEN(p) = dlen;
3694: }
3695:
3696: NDV ptondv(VL vl,VL dvl,P p)
3697: {
3698: ND nd;
3699:
3700: nd = ptond(vl,dvl,p);
3701: return ndtondv(0,nd);
3702: }
3703:
3704: ND ptond(VL vl,VL dvl,P p)
1.23 noro 3705: {
1.61 noro 3706: int n,i,j,k,e;
3707: VL tvl;
3708: V v;
3709: DCP dc;
3710: DCP *w;
3711: ND r,s,t,u;
3712: P x;
3713: int c;
3714: UINT *d;
1.23 noro 3715: NM m,m0;
1.61 noro 3716:
3717: if ( !p )
3718: return 0;
3719: else if ( NUM(p) ) {
3720: NEWNM(m);
3721: ndl_zero(DL(m));
1.113 noro 3722: CQ(m) = (Q)p;
1.61 noro 3723: NEXT(m) = 0;
3724: MKND(nd_nvar,m,1,r);
3725: SG(r) = 0;
3726: return r;
3727: } else {
3728: for ( dc = DC(p), k = 0; dc; dc = NEXT(dc), k++ );
3729: w = (DCP *)ALLOCA(k*sizeof(DCP));
3730: for ( dc = DC(p), j = 0; j < k; dc = NEXT(dc), j++ ) w[j] = dc;
3731: for ( i = 0, tvl = dvl, v = VR(p);
1.98 noro 3732: tvl && tvl->v != v; tvl = NEXT(tvl), i++ );
1.61 noro 3733: if ( !tvl ) {
3734: for ( j = k-1, s = 0, MKV(v,x); j >= 0; j-- ) {
3735: t = ptond(vl,dvl,COEF(w[j]));
3736: pwrp(vl,x,DEG(w[j]),&p);
3737: nd_mul_c_p(CO,t,p); s = nd_add(0,s,t);
3738: }
3739: return s;
3740: } else {
3741: NEWNM(m0); d = DL(m0);
3742: for ( j = k-1, s = 0; j >= 0; j-- ) {
3743: ndl_zero(d); e = QTOS(DEG(w[j])); PUT_EXP(d,i,e);
3744: TD(d) = MUL_WEIGHT(e,i);
3745: if ( nd_blockmask) ndl_weight_mask(d);
3746: t = ptond(vl,dvl,COEF(w[j]));
3747: for ( m = BDY(t); m; m = NEXT(m) )
3748: ndl_addto(DL(m),d);
3749: SG(t) += TD(d);
3750: s = nd_add(0,s,t);
3751: }
3752: FREENM(m0);
3753: return s;
3754: }
3755: }
3756: }
3757:
3758: P ndvtop(int mod,VL vl,VL dvl,NDV p)
3759: {
3760: VL tvl;
3761: int len,n,j,i,e;
3762: NMV m;
3763: Q q;
3764: P c;
3765: UINT *d;
3766: P s,r,u,t,w;
1.71 noro 3767: GFS gfs;
1.23 noro 3768:
1.34 noro 3769: if ( !p ) return 0;
1.61 noro 3770: else {
3771: len = LEN(p);
3772: n = NV(p);
3773: m = (NMV)(((char *)BDY(p))+nmv_adv*(len-1));
3774: for ( j = len-1, s = 0; j >= 0; j--, NMV_PREV(m) ) {
1.71 noro 3775: if ( mod == -1 ) {
3776: e = IFTOF(CM(m)); MKGFS(e,gfs); c = (P)gfs;
3777: } else if ( mod ) {
1.61 noro 3778: STOQ(CM(m),q); c = (P)q;
3779: } else
1.113 noro 3780: c = CP(m);
1.61 noro 3781: d = DL(m);
3782: for ( i = 0, t = c, tvl = dvl; i < n; tvl = NEXT(tvl), i++ ) {
3783: MKV(tvl->v,r); e = GET_EXP(d,i); STOQ(e,q);
3784: pwrp(vl,r,q,&u); mulp(vl,t,u,&w); t = w;
3785: }
3786: addp(vl,s,t,&u); s = u;
3787: }
3788: return s;
1.23 noro 3789: }
1.3 noro 3790: }
3791:
1.61 noro 3792: NDV ndtondv(int mod,ND p)
1.11 noro 3793: {
3794: NDV d;
1.61 noro 3795: NMV m,m0;
3796: NM t;
3797: int i,len;
1.11 noro 3798:
1.34 noro 3799: if ( !p ) return 0;
1.61 noro 3800: len = LEN(p);
1.74 noro 3801: if ( mod )
3802: m0 = m = (NMV)GC_malloc_atomic_ignore_off_page(len*nmv_adv);
3803: else
3804: m0 = m = MALLOC(len*nmv_adv);
1.103 noro 3805: #if 0
1.74 noro 3806: ndv_alloc += nmv_adv*len;
1.103 noro 3807: #endif
1.61 noro 3808: for ( t = BDY(p), i = 0; t; t = NEXT(t), i++, NMV_ADV(m) ) {
3809: ndl_copy(DL(t),DL(m));
1.113 noro 3810: CQ(m) = CQ(t);
1.11 noro 3811: }
1.61 noro 3812: MKNDV(NV(p),m0,len,d);
1.14 noro 3813: SG(d) = SG(p);
1.11 noro 3814: return d;
3815: }
3816:
1.61 noro 3817: ND ndvtond(int mod,NDV p)
1.11 noro 3818: {
1.61 noro 3819: ND d;
3820: NM m,m0;
1.11 noro 3821: NMV t;
1.61 noro 3822: int i,len;
1.11 noro 3823:
1.34 noro 3824: if ( !p ) return 0;
1.11 noro 3825: m0 = 0;
1.61 noro 3826: len = p->len;
3827: for ( t = BDY(p), i = 0; i < len; NMV_ADV(t), i++ ) {
3828: NEXTNM(m0,m);
3829: ndl_copy(DL(t),DL(m));
1.113 noro 3830: CQ(m) = CQ(t);
1.11 noro 3831: }
3832: NEXT(m) = 0;
1.61 noro 3833: MKND(NV(p),m0,len,d);
1.14 noro 3834: SG(d) = SG(p);
1.11 noro 3835: return d;
3836: }
3837:
1.3 noro 3838: void ndv_print(NDV p)
3839: {
3840: NMV m;
3841: int i,len;
3842:
1.34 noro 3843: if ( !p ) printf("0\n");
1.3 noro 3844: else {
1.14 noro 3845: len = LEN(p);
1.3 noro 3846: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
1.71 noro 3847: if ( CM(m) & 0x80000000 ) printf("+@_%d*",IFTOF(CM(m)));
3848: else printf("+%d*",CM(m));
1.16 noro 3849: ndl_print(DL(m));
3850: }
3851: printf("\n");
3852: }
3853: }
3854:
1.113 noro 3855: void ndv_print_q(NDV p)
1.16 noro 3856: {
3857: NMV m;
3858: int i,len;
3859:
1.34 noro 3860: if ( !p ) printf("0\n");
1.16 noro 3861: else {
3862: len = LEN(p);
3863: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
3864: printf("+");
1.113 noro 3865: printexpr(CO,(Obj)CQ(m));
1.16 noro 3866: printf("*");
1.14 noro 3867: ndl_print(DL(m));
1.3 noro 3868: }
3869: printf("\n");
3870: }
1.25 noro 3871: }
3872:
1.61 noro 3873: NODE ndv_reducebase(NODE x)
1.27 noro 3874: {
3875: int len,i,j;
3876: NDV *w;
3877: NODE t,t0;
3878:
3879: len = length(x);
3880: w = (NDV *)ALLOCA(len*sizeof(NDV));
3881: for ( i = 0, t = x; i < len; i++, t = NEXT(t) ) w[i] = BDY(t);
3882: for ( i = 0; i < len; i++ ) {
3883: for ( j = 0; j < i; j++ ) {
3884: if ( w[i] && w[j] )
3885: if ( ndl_reducible(HDL(w[i]),HDL(w[j])) ) w[i] = 0;
3886: else if ( ndl_reducible(HDL(w[j]),HDL(w[i])) ) w[j] = 0;
3887: }
3888: }
3889: for ( i = len-1, t0 = 0; i >= 0; i-- ) {
3890: if ( w[i] ) { NEXTNODE(t0,t); BDY(t) = (pointer)w[i]; }
3891: }
3892: NEXT(t) = 0; x = t0;
3893: return x;
1.11 noro 3894: }
1.32 noro 3895:
1.43 noro 3896: /* XXX incomplete */
3897:
1.32 noro 3898: void nd_init_ord(struct order_spec *ord)
3899: {
1.43 noro 3900: switch ( ord->id ) {
1.32 noro 3901: case 0:
1.43 noro 3902: switch ( ord->ord.simple ) {
3903: case 0:
3904: nd_dcomp = 1;
3905: nd_isrlex = 1;
3906: break;
3907: case 1:
3908: nd_dcomp = 1;
3909: nd_isrlex = 0;
3910: break;
3911: case 2:
3912: nd_dcomp = 0;
3913: nd_isrlex = 0;
1.45 noro 3914: ndl_compare_function = ndl_lex_compare;
1.58 noro 3915: break;
3916: case 11:
3917: /* XXX */
3918: nd_dcomp = 0;
3919: nd_isrlex = 1;
3920: ndl_compare_function = ndl_ww_lex_compare;
1.43 noro 3921: break;
3922: default:
3923: error("nd_gr : unsupported order");
3924: }
1.32 noro 3925: break;
3926: case 1:
1.96 noro 3927: /* block order */
1.43 noro 3928: /* XXX */
3929: nd_dcomp = -1;
1.32 noro 3930: nd_isrlex = 0;
1.45 noro 3931: ndl_compare_function = ndl_block_compare;
1.34 noro 3932: break;
1.43 noro 3933: case 2:
1.96 noro 3934: /* matrix order */
3935: /* XXX */
3936: nd_dcomp = -1;
3937: nd_isrlex = 0;
3938: nd_matrix_len = ord->ord.matrix.row;
3939: nd_matrix = ord->ord.matrix.matrix;
3940: ndl_compare_function = ndl_matrix_compare;
3941: break;
3942: case 3:
1.97 noro 3943: /* composite order */
3944: nd_dcomp = -1;
3945: nd_isrlex = 0;
3946: nd_worb_len = ord->ord.composite.length;
3947: nd_worb = ord->ord.composite.w_or_b;
3948: ndl_compare_function = ndl_composite_compare;
1.32 noro 3949: break;
3950: }
1.41 noro 3951: nd_ord = ord;
1.32 noro 3952: }
3953:
1.43 noro 3954: BlockMask nd_create_blockmask(struct order_spec *ord)
3955: {
3956: int n,i,j,s,l;
1.61 noro 3957: UINT *t;
1.43 noro 3958: BlockMask bm;
3959:
1.96 noro 3960: /* we only create mask table for block order */
3961: if ( ord->id != 1 )
1.43 noro 3962: return 0;
3963: n = ord->ord.block.length;
3964: bm = (BlockMask)MALLOC(sizeof(struct oBlockMask));
3965: bm->n = n;
3966: bm->order_pair = ord->ord.block.order_pair;
1.61 noro 3967: bm->mask = (UINT **)MALLOC(n*sizeof(UINT *));
1.43 noro 3968: for ( i = 0, s = 0; i < n; i++ ) {
1.61 noro 3969: bm->mask[i] = t = (UINT *)MALLOC_ATOMIC(nd_wpd*sizeof(UINT));
1.43 noro 3970: for ( j = 0; j < nd_wpd; j++ ) t[j] = 0;
3971: l = bm->order_pair[i].length;
3972: for ( j = 0; j < l; j++, s++ ) PUT_EXP(t,s,nd_mask0);
3973: }
3974: return bm;
1.57 noro 3975: }
3976:
3977: EPOS nd_create_epos(struct order_spec *ord)
3978: {
3979: int i,j,l,s,ord_l,ord_o;
3980: EPOS epos;
3981: struct order_pair *op;
3982:
3983: epos = (EPOS)MALLOC_ATOMIC(nd_nvar*sizeof(struct oEPOS));
3984: switch ( ord->id ) {
3985: case 0:
3986: if ( nd_isrlex ) {
3987: for ( i = 0; i < nd_nvar; i++ ) {
3988: epos[i].i = nd_exporigin + (nd_nvar-1-i)/nd_epw;
3989: epos[i].s = (nd_epw-((nd_nvar-1-i)%nd_epw)-1)*nd_bpe;
3990: }
3991: } else {
3992: for ( i = 0; i < nd_nvar; i++ ) {
3993: epos[i].i = nd_exporigin + i/nd_epw;
3994: epos[i].s = (nd_epw-(i%nd_epw)-1)*nd_bpe;
3995: }
3996: }
3997: break;
3998: case 1:
3999: /* block order */
4000: l = ord->ord.block.length;
4001: op = ord->ord.block.order_pair;
4002: for ( j = 0, s = 0; j < l; j++ ) {
4003: ord_o = op[j].order;
4004: ord_l = op[j].length;
4005: if ( !ord_o )
4006: for ( i = 0; i < ord_l; i++ ) {
4007: epos[s+i].i = nd_exporigin + (s+ord_l-i-1)/nd_epw;
4008: epos[s+i].s = (nd_epw-((s+ord_l-i-1)%nd_epw)-1)*nd_bpe;
4009: }
4010: else
4011: for ( i = 0; i < ord_l; i++ ) {
4012: epos[s+i].i = nd_exporigin + (s+i)/nd_epw;
4013: epos[s+i].s = (nd_epw-((s+i)%nd_epw)-1)*nd_bpe;
4014: }
4015: s += ord_l;
4016: }
4017: break;
4018: case 2:
1.96 noro 4019: /* matrix order */
4020: case 3:
4021: /* composite order */
4022: for ( i = 0; i < nd_nvar; i++ ) {
4023: epos[i].i = nd_exporigin + i/nd_epw;
4024: epos[i].s = (nd_epw-(i%nd_epw)-1)*nd_bpe;
4025: }
4026: break;
1.57 noro 4027: }
4028: return epos;
1.43 noro 4029: }
1.59 noro 4030:
4031: /* external interface */
4032:
4033: void nd_nf_p(P f,LIST g,LIST v,int m,struct order_spec *ord,P *rp)
4034: {
1.61 noro 4035: NODE t,in0,in;
1.59 noro 4036: ND nd,nf;
1.61 noro 4037: NDV ndv;
4038: VL vv,tv;
4039: int stat,nvar,max,e;
1.113 noro 4040: union oNDC dn;
1.59 noro 4041:
1.116 noro 4042: if ( !f ) {
4043: *rp = 0;
4044: return;
4045: }
1.59 noro 4046: pltovl(v,&vv);
1.74 noro 4047: for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++ );
1.61 noro 4048:
4049: /* get the degree bound */
4050: for ( t = BDY(g), max = 0; t; t = NEXT(t) )
4051: for ( tv = vv; tv; tv = NEXT(tv) ) {
4052: e = getdeg(tv->v,(P)BDY(t));
4053: max = MAX(e,max);
4054: }
4055: for ( tv = vv; tv; tv = NEXT(tv) ) {
4056: e = getdeg(tv->v,f);
4057: max = MAX(e,max);
4058: }
4059:
1.59 noro 4060: nd_init_ord(ord);
1.61 noro 4061: nd_setup_parameters(nvar,max);
4062:
4063: /* conversion to ndv */
4064: for ( in0 = 0, t = BDY(g); t; t = NEXT(t) ) {
4065: NEXTNODE(in0,in);
4066: BDY(in) = (pointer)ptondv(CO,vv,(P)BDY(t));
1.69 noro 4067: if ( m ) ndv_mod(m,(NDV)BDY(in));
1.61 noro 4068: }
4069: NEXTNODE(in0,in);
4070: BDY(in) = (pointer)ptondv(CO,vv,f);
1.69 noro 4071: if ( m ) ndv_mod(m,(NDV)BDY(in));
1.61 noro 4072: NEXT(in) = 0;
4073:
1.128 noro 4074: /* dont sort */
4075: ndv_setup(m,0,in0,1);
1.59 noro 4076: nd_psn--;
4077: nd_scale=2;
4078: while ( 1 ) {
4079: nd = (pointer)ndvtond(m,nd_ps[nd_psn]);
1.69 noro 4080: stat = nd_nf(m,nd,nd_ps,1,0,&nf);
1.59 noro 4081: if ( !stat ) {
4082: nd_psn++;
1.103 noro 4083: nd_reconstruct(0,0);
1.59 noro 4084: nd_psn--;
4085: } else
4086: break;
4087: }
1.61 noro 4088: *rp = ndvtop(m,CO,vv,ndtondv(m,nf));
1.63 noro 4089: }
4090:
4091: int nd_to_vect(int mod,UINT *s0,int n,ND d,UINT *r)
4092: {
4093: NM m;
4094: UINT *t,*s;
4095: int i;
4096:
4097: for ( i = 0; i < n; i++ ) r[i] = 0;
4098: for ( i = 0, s = s0, m = BDY(d); m; m = NEXT(m) ) {
4099: t = DL(m);
4100: for ( ; !ndl_equal(t,s); s += nd_wpd, i++ );
4101: r[i] = CM(m);
4102: }
4103: for ( i = 0; !r[i]; i++ );
4104: return i;
4105: }
4106:
1.113 noro 4107: int nd_to_vect_q(UINT *s0,int n,ND d,Q *r)
1.74 noro 4108: {
1.107 noro 4109: NM m;
1.74 noro 4110: UINT *t,*s;
1.107 noro 4111: int i;
1.74 noro 4112:
4113: for ( i = 0; i < n; i++ ) r[i] = 0;
1.107 noro 4114: for ( i = 0, s = s0, m = BDY(d); m; m = NEXT(m) ) {
1.74 noro 4115: t = DL(m);
4116: for ( ; !ndl_equal(t,s); s += nd_wpd, i++ );
1.113 noro 4117: r[i] = CQ(m);
1.74 noro 4118: }
4119: for ( i = 0; !r[i]; i++ );
4120: return i;
4121: }
4122:
1.129 noro 4123: Q *nm_ind_pair_to_vect(int mod,UINT *s0,int n,NM_ind_pair pair)
4124: {
4125: NM m;
4126: NMV mr;
4127: UINT *d,*t,*s;
4128: NDV p;
4129: int i,j,len;
4130: Q *r;
4131:
4132: m = pair->mul;
4133: d = DL(m);
4134: p = nd_ps[pair->index];
4135: len = LEN(p);
4136: r = (Q *)CALLOC(n,sizeof(Q));
4137: t = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
4138: for ( i = j = 0, s = s0, mr = BDY(p); j < len; j++, NMV_ADV(mr) ) {
4139: ndl_add(d,DL(mr),t);
4140: for ( ; !ndl_equal(t,s); s += nd_wpd, i++ );
4141: r[i] = CQ(mr);
4142: }
4143: return r;
4144: }
4145:
1.67 noro 4146: IndArray nm_ind_pair_to_vect_compress(int mod,UINT *s0,int n,NM_ind_pair pair)
1.64 noro 4147: {
4148: NM m;
4149: NMV mr;
4150: UINT *d,*t,*s;
4151: NDV p;
1.67 noro 4152: unsigned char *ivc;
4153: unsigned short *ivs;
1.81 noro 4154: UINT *v,*ivi,*s0v;
1.67 noro 4155: int i,j,len,prev,diff,cdiff;
4156: IndArray r;
1.64 noro 4157:
4158: m = pair->mul;
4159: d = DL(m);
4160: p = nd_ps[pair->index];
4161: len = LEN(p);
4162: t = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
1.67 noro 4163: v = (unsigned int *)ALLOCA(len*sizeof(unsigned int));
1.64 noro 4164: for ( i = j = 0, s = s0, mr = BDY(p); j < len; j++, NMV_ADV(mr) ) {
4165: ndl_add(d,DL(mr),t);
4166: for ( ; !ndl_equal(t,s); s += nd_wpd, i++ );
1.67 noro 4167: v[j] = i;
4168: }
1.81 noro 4169: r = (IndArray)MALLOC(sizeof(struct oIndArray));
1.67 noro 4170: r->head = v[0];
4171: diff = 0;
4172: for ( i = 1; i < len; i++ ) {
4173: cdiff = v[i]-v[i-1]; diff = MAX(cdiff,diff);
4174: }
4175: if ( diff < 256 ) {
4176: r->width = 1;
4177: ivc = (unsigned char *)MALLOC_ATOMIC(len*sizeof(unsigned char));
4178: r->index.c = ivc;
4179: for ( i = 1, ivc[0] = 0; i < len; i++ ) ivc[i] = v[i]-v[i-1];
4180: } else if ( diff < 65536 ) {
4181: r->width = 2;
4182: ivs = (unsigned short *)MALLOC_ATOMIC(len*sizeof(unsigned short));
4183: r->index.s = ivs;
4184: for ( i = 1, ivs[0] = 0; i < len; i++ ) ivs[i] = v[i]-v[i-1];
4185: } else {
4186: r->width = 4;
4187: ivi = (unsigned int *)MALLOC_ATOMIC(len*sizeof(unsigned int));
4188: r->index.i = ivi;
4189: for ( i = 1, ivi[0] = 0; i < len; i++ ) ivi[i] = v[i]-v[i-1];
1.64 noro 4190: }
1.67 noro 4191: return r;
1.64 noro 4192: }
4193:
1.135 noro 4194: int compress_array(Q *svect,Q *cvect,int n)
4195: {
4196: int i,j;
4197:
4198: for ( i = j = 0; i < n; i++ )
4199: if ( svect[i] ) cvect[j++] = svect[i];
4200: return j;
4201: }
4202:
4203: void expand_array(Q *svect,Q *cvect,int n)
4204: {
4205: int i,j;
4206:
4207: for ( i = j = 0; j < n; i++ )
4208: if ( svect[i] ) svect[i] = cvect[j++];
4209: }
4210:
1.133 noro 4211: int ndv_reduce_vect_q(Q *svect,int trace,int col,IndArray *imat,NM_ind_pair *rp0,int nred)
1.107 noro 4212: {
1.135 noro 4213: int i,j,k,len,pos,prev,nz;
1.113 noro 4214: Q cs,mcs,c1,c2,cr,gcd,t;
1.107 noro 4215: IndArray ivect;
4216: unsigned char *ivc;
4217: unsigned short *ivs;
4218: unsigned int *ivi;
4219: NDV redv;
4220: NMV mr;
4221: NODE rp;
4222: int maxrs;
1.133 noro 4223: double hmag;
1.135 noro 4224: Q *cvect;
1.107 noro 4225:
4226: maxrs = 0;
1.134 noro 4227: for ( i = 0; i < col && !svect[i]; i++ );
4228: if ( i == col ) return maxrs;
4229: hmag = p_mag((P)svect[i])*nd_scale;
1.135 noro 4230: cvect = (Q *)ALLOCA(col*sizeof(Q));
1.107 noro 4231: for ( i = 0; i < nred; i++ ) {
4232: ivect = imat[i];
4233: k = ivect->head;
4234: if ( svect[k] ) {
4235: maxrs = MAX(maxrs,rp0[i]->sugar);
1.133 noro 4236: redv = trace?nd_ps_trace[rp0[i]->index]:nd_ps[rp0[i]->index];
1.107 noro 4237: len = LEN(redv); mr = BDY(redv);
1.113 noro 4238: igcd_cofactor(svect[k],CQ(mr),&gcd,&cs,&cr);
4239: chsgnq(cs,&mcs);
4240: if ( !UNIQ(cr) ) {
4241: for ( j = 0; j < col; j++ ) {
4242: mulq(svect[j],cr,&c1); svect[j] = c1;
4243: }
4244: }
1.107 noro 4245: svect[k] = 0; prev = k;
4246: switch ( ivect->width ) {
4247: case 1:
4248: ivc = ivect->index.c;
4249: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
4250: pos = prev+ivc[j]; prev = pos;
1.113 noro 4251: mulq(CQ(mr),mcs,&c2); addq(svect[pos],c2,&t); svect[pos] = t;
1.107 noro 4252: }
4253: break;
4254: case 2:
4255: ivs = ivect->index.s;
4256: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
4257: pos = prev+ivs[j]; prev = pos;
1.113 noro 4258: mulq(CQ(mr),mcs,&c2); addq(svect[pos],c2,&t); svect[pos] = t;
1.107 noro 4259: }
4260: break;
4261: case 4:
4262: ivi = ivect->index.i;
4263: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
4264: pos = prev+ivi[j]; prev = pos;
1.113 noro 4265: mulq(CQ(mr),mcs,&c2); addq(svect[pos],c2,&t); svect[pos] = t;
1.107 noro 4266: }
4267: break;
4268: }
1.134 noro 4269: for ( j = k+1; j < col && !svect[j]; j++ );
4270: if ( j == col ) break;
4271: if ( hmag && ((double)p_mag((P)svect[j]) > hmag) ) {
1.135 noro 4272: nz = compress_array(svect,cvect,col);
4273: removecont_array(cvect,nz);
4274: expand_array(svect,cvect,nz);
1.134 noro 4275: hmag = ((double)p_mag((P)svect[j]))*nd_scale;
4276: }
1.107 noro 4277: }
1.133 noro 4278: }
1.135 noro 4279: nz = compress_array(svect,cvect,col);
4280: removecont_array(cvect,nz);
4281: expand_array(svect,cvect,nz);
1.133 noro 4282: if ( DP_Print ) {
4283: fprintf(asir_out,"-"); fflush(asir_out);
1.107 noro 4284: }
4285: return maxrs;
4286: }
4287:
1.76 noro 4288: int ndv_reduce_vect(int m,UINT *svect,int col,IndArray *imat,NM_ind_pair *rp0,int nred)
1.65 noro 4289: {
1.67 noro 4290: int i,j,k,len,pos,prev;
1.66 noro 4291: UINT c,c1,c2,c3,up,lo,dmy;
1.67 noro 4292: IndArray ivect;
4293: unsigned char *ivc;
4294: unsigned short *ivs;
4295: unsigned int *ivi;
1.65 noro 4296: NDV redv;
1.67 noro 4297: NMV mr;
1.65 noro 4298: NODE rp;
1.76 noro 4299: int maxrs;
1.65 noro 4300:
1.76 noro 4301: maxrs = 0;
1.74 noro 4302: for ( i = 0; i < nred; i++ ) {
1.65 noro 4303: ivect = imat[i];
1.67 noro 4304: k = ivect->head; svect[k] %= m;
1.65 noro 4305: if ( c = svect[k] ) {
1.76 noro 4306: maxrs = MAX(maxrs,rp0[i]->sugar);
1.74 noro 4307: c = m-c; redv = nd_ps[rp0[i]->index];
1.67 noro 4308: len = LEN(redv); mr = BDY(redv);
4309: svect[k] = 0; prev = k;
4310: switch ( ivect->width ) {
4311: case 1:
4312: ivc = ivect->index.c;
4313: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
4314: pos = prev+ivc[j]; c1 = CM(mr); c2 = svect[pos];
4315: prev = pos;
4316: DMA(c1,c,c2,up,lo);
4317: if ( up ) { DSAB(m,up,lo,dmy,c3); svect[pos] = c3;
4318: } else svect[pos] = lo;
4319: }
4320: break;
4321: case 2:
4322: ivs = ivect->index.s;
4323: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
4324: pos = prev+ivs[j]; c1 = CM(mr); c2 = svect[pos];
4325: prev = pos;
4326: DMA(c1,c,c2,up,lo);
4327: if ( up ) { DSAB(m,up,lo,dmy,c3); svect[pos] = c3;
4328: } else svect[pos] = lo;
4329: }
4330: break;
4331: case 4:
4332: ivi = ivect->index.i;
4333: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
4334: pos = prev+ivi[j]; c1 = CM(mr); c2 = svect[pos];
4335: prev = pos;
4336: DMA(c1,c,c2,up,lo);
4337: if ( up ) { DSAB(m,up,lo,dmy,c3); svect[pos] = c3;
4338: } else svect[pos] = lo;
4339: }
4340: break;
1.65 noro 4341: }
4342: }
4343: }
1.66 noro 4344: for ( i = 0; i < col; i++ )
4345: if ( svect[i] >= (UINT)m ) svect[i] %= m;
1.76 noro 4346: return maxrs;
1.65 noro 4347: }
4348:
1.76 noro 4349: int ndv_reduce_vect_sf(int m,UINT *svect,int col,IndArray *imat,NM_ind_pair *rp0,int nred)
1.72 noro 4350: {
4351: int i,j,k,len,pos,prev;
4352: UINT c,c1,c2,c3,up,lo,dmy;
4353: IndArray ivect;
4354: unsigned char *ivc;
4355: unsigned short *ivs;
4356: unsigned int *ivi;
4357: NDV redv;
4358: NMV mr;
4359: NODE rp;
1.76 noro 4360: int maxrs;
1.72 noro 4361:
1.76 noro 4362: maxrs = 0;
1.74 noro 4363: for ( i = 0; i < nred; i++ ) {
1.72 noro 4364: ivect = imat[i];
4365: k = ivect->head; svect[k] %= m;
4366: if ( c = svect[k] ) {
1.76 noro 4367: maxrs = MAX(maxrs,rp0[i]->sugar);
1.74 noro 4368: c = _chsgnsf(c); redv = nd_ps[rp0[i]->index];
1.72 noro 4369: len = LEN(redv); mr = BDY(redv);
4370: svect[k] = 0; prev = k;
4371: switch ( ivect->width ) {
4372: case 1:
4373: ivc = ivect->index.c;
4374: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
4375: pos = prev+ivc[j]; prev = pos;
4376: svect[pos] = _addsf(_mulsf(CM(mr),c),svect[pos]);
4377: }
4378: break;
4379: case 2:
4380: ivs = ivect->index.s;
4381: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
4382: pos = prev+ivs[j]; prev = pos;
4383: svect[pos] = _addsf(_mulsf(CM(mr),c),svect[pos]);
4384: }
4385: break;
4386: case 4:
4387: ivi = ivect->index.i;
4388: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
4389: pos = prev+ivi[j]; prev = pos;
4390: svect[pos] = _addsf(_mulsf(CM(mr),c),svect[pos]);
4391: }
4392: break;
4393: }
4394: }
4395: }
1.76 noro 4396: return maxrs;
1.72 noro 4397: }
4398:
1.65 noro 4399: NDV vect_to_ndv(UINT *vect,int spcol,int col,int *rhead,UINT *s0vect)
4400: {
4401: int j,k,len;
4402: UINT *p;
4403: UINT c;
4404: NDV r;
4405: NMV mr0,mr;
4406:
4407: for ( j = 0, len = 0; j < spcol; j++ ) if ( vect[j] ) len++;
4408: if ( !len ) return 0;
4409: else {
1.74 noro 4410: mr0 = (NMV)GC_malloc_atomic_ignore_off_page(nmv_adv*len);
1.103 noro 4411: #if 0
1.74 noro 4412: ndv_alloc += nmv_adv*len;
1.103 noro 4413: #endif
1.65 noro 4414: mr = mr0;
4415: p = s0vect;
4416: for ( j = k = 0; j < col; j++, p += nd_wpd )
4417: if ( !rhead[j] ) {
4418: if ( c = vect[k++] ) {
4419: ndl_copy(p,DL(mr)); CM(mr) = c; NMV_ADV(mr);
4420: }
4421: }
4422: MKNDV(nd_nvar,mr0,len,r);
4423: return r;
4424: }
4425: }
4426:
1.129 noro 4427: /* for preprocessed vector */
4428:
1.113 noro 4429: NDV vect_to_ndv_q(Q *vect,int spcol,int col,int *rhead,UINT *s0vect)
1.107 noro 4430: {
4431: int j,k,len;
4432: UINT *p;
1.113 noro 4433: Q c;
1.107 noro 4434: NDV r;
4435: NMV mr0,mr;
4436:
4437: for ( j = 0, len = 0; j < spcol; j++ ) if ( vect[j] ) len++;
4438: if ( !len ) return 0;
4439: else {
1.109 noro 4440: mr0 = (NMV)GC_malloc(nmv_adv*len);
1.107 noro 4441: #if 0
4442: ndv_alloc += nmv_adv*len;
4443: #endif
4444: mr = mr0;
4445: p = s0vect;
4446: for ( j = k = 0; j < col; j++, p += nd_wpd )
4447: if ( !rhead[j] ) {
4448: if ( c = vect[k++] ) {
1.113 noro 4449: if ( DN(c) )
4450: error("afo");
4451: ndl_copy(p,DL(mr)); CQ(mr) = c; NMV_ADV(mr);
1.107 noro 4452: }
4453: }
4454: MKNDV(nd_nvar,mr0,len,r);
4455: return r;
4456: }
4457: }
4458:
1.129 noro 4459: /* for plain vector */
4460:
4461: NDV plain_vect_to_ndv_q(Q *vect,int col,UINT *s0vect)
4462: {
4463: int j,k,len;
4464: UINT *p;
4465: Q c;
4466: NDV r;
4467: NMV mr0,mr;
4468:
4469: for ( j = 0, len = 0; j < col; j++ ) if ( vect[j] ) len++;
4470: if ( !len ) return 0;
4471: else {
4472: mr0 = (NMV)GC_malloc(nmv_adv*len);
4473: #if 0
4474: ndv_alloc += nmv_adv*len;
4475: #endif
4476: mr = mr0;
4477: p = s0vect;
4478: for ( j = k = 0; j < col; j++, p += nd_wpd, k++ )
4479: if ( c = vect[k] ) {
4480: if ( DN(c) )
4481: error("afo");
4482: ndl_copy(p,DL(mr)); CQ(mr) = c; NMV_ADV(mr);
4483: }
4484: MKNDV(nd_nvar,mr0,len,r);
4485: return r;
4486: }
4487: }
4488:
1.133 noro 4489: int nd_sp_f4(int m,int trace,ND_pairs l,PGeoBucket bucket)
1.65 noro 4490: {
4491: ND_pairs t;
4492: NODE sp0,sp;
4493: int stat;
4494: ND spol;
4495:
4496: for ( t = l; t; t = NEXT(t) ) {
1.133 noro 4497: stat = nd_sp(m,trace,t,&spol);
1.65 noro 4498: if ( !stat ) return 0;
4499: if ( spol ) {
4500: add_pbucket_symbolic(bucket,spol);
4501: }
4502: }
1.68 noro 4503: return 1;
1.65 noro 4504: }
4505:
1.133 noro 4506: int nd_symbolic_preproc(PGeoBucket bucket,int trace,UINT **s0vect,NODE *r)
1.65 noro 4507: {
4508: NODE rp0,rp;
4509: NM mul,head,s0,s;
1.76 noro 4510: int index,col,i,sugar;
1.65 noro 4511: RHist h;
4512: UINT *s0v,*p;
4513: NM_ind_pair pair;
4514: ND red;
1.133 noro 4515: NDV *ps;
1.65 noro 4516:
4517: s0 = 0; rp0 = 0; col = 0;
1.133 noro 4518: ps = trace?nd_ps_trace:nd_ps;
1.65 noro 4519: while ( 1 ) {
4520: head = remove_head_pbucket_symbolic(bucket);
4521: if ( !head ) break;
4522: if ( !s0 ) s0 = head;
4523: else NEXT(s) = head;
4524: s = head;
4525: index = ndl_find_reducer(DL(head));
4526: if ( index >= 0 ) {
4527: h = nd_psh[index];
4528: NEWNM(mul);
4529: ndl_sub(DL(head),DL(h),DL(mul));
4530: if ( ndl_check_bound2(index,DL(mul)) ) return 0;
1.133 noro 4531: sugar = TD(DL(mul))+SG(ps[index]);
1.76 noro 4532: MKNM_ind_pair(pair,mul,index,sugar);
1.133 noro 4533: red = ndv_mul_nm_symbolic(mul,ps[index]);
1.65 noro 4534: add_pbucket_symbolic(bucket,nd_remove_head(red));
4535: NEXTNODE(rp0,rp); BDY(rp) = (pointer)pair;
4536: }
4537: col++;
4538: }
1.72 noro 4539: if ( rp0 ) NEXT(rp) = 0;
4540: NEXT(s) = 0;
1.65 noro 4541: s0v = (UINT *)MALLOC_ATOMIC(col*nd_wpd*sizeof(UINT));
4542: for ( i = 0, p = s0v, s = s0; i < col;
4543: i++, p += nd_wpd, s = NEXT(s) ) ndl_copy(DL(s),p);
4544: *s0vect = s0v;
4545: *r = rp0;
4546: return col;
4547: }
4548:
1.69 noro 4549: NODE nd_f4(int m)
4550: {
4551: int i,nh,stat,index;
4552: NODE r,g;
4553: ND_pairs d,l,t;
4554: ND spol,red;
4555: NDV nf,redv;
4556: NM s0,s;
1.84 noro 4557: NODE rp0,srp0,nflist;
1.69 noro 4558: int nsp,nred,col,rank,len,k,j,a;
4559: UINT c;
1.74 noro 4560: UINT **spmat;
1.69 noro 4561: UINT *s0vect,*svect,*p,*v;
4562: int *colstat;
4563: IndArray *imat;
4564: int *rhead;
4565: int spcol,sprow;
4566: int sugar;
4567: PGeoBucket bucket;
4568: struct oEGT eg0,eg1,eg_f4;
4569:
1.103 noro 4570: #if 0
1.74 noro 4571: ndv_alloc = 0;
1.103 noro 4572: #endif
1.69 noro 4573: g = 0; d = 0;
4574: for ( i = 0; i < nd_psn; i++ ) {
4575: d = update_pairs(d,g,i);
4576: g = update_base(g,i);
4577: }
4578: while ( d ) {
4579: get_eg(&eg0);
4580: l = nd_minsugarp(d,&d);
4581: sugar = SG(l);
4582: bucket = create_pbucket();
1.133 noro 4583: stat = nd_sp_f4(m,0,l,bucket);
1.69 noro 4584: if ( !stat ) {
4585: for ( t = l; NEXT(t); t = NEXT(t) );
4586: NEXT(t) = d; d = l;
1.103 noro 4587: d = nd_reconstruct(0,d);
1.69 noro 4588: continue;
4589: }
1.86 noro 4590: if ( bucket->m < 0 ) continue;
1.133 noro 4591: col = nd_symbolic_preproc(bucket,0,&s0vect,&rp0);
1.69 noro 4592: if ( !col ) {
4593: for ( t = l; NEXT(t); t = NEXT(t) );
4594: NEXT(t) = d; d = l;
1.103 noro 4595: d = nd_reconstruct(0,d);
1.69 noro 4596: continue;
4597: }
4598: get_eg(&eg1); init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg1);
1.74 noro 4599: if ( DP_Print )
4600: fprintf(asir_out,"sugar=%d,symb=%fsec,",
4601: sugar,eg_f4.exectime+eg_f4.gctime);
4602: if ( 1 )
1.133 noro 4603: nflist = nd_f4_red(m,l,0,s0vect,col,rp0,0);
1.74 noro 4604: else
1.133 noro 4605: nflist = nd_f4_red_dist(m,l,s0vect,col,rp0,0);
1.69 noro 4606: /* adding new bases */
1.74 noro 4607: for ( r = nflist; r; r = NEXT(r) ) {
4608: nf = (NDV)BDY(r);
1.69 noro 4609: ndv_removecont(m,nf);
1.129 noro 4610: if ( !m && nd_nalg ) {
4611: ND nf1;
4612:
4613: nf1 = ndvtond(m,nf);
4614: nd_monic(0,&nf1);
4615: nd_removecont(m,nf1);
4616: nf = ndtondv(m,nf1);
4617: }
1.77 noro 4618: nh = ndv_newps(m,nf,0);
1.69 noro 4619: d = update_pairs(d,g,nh);
4620: g = update_base(g,nh);
4621: }
4622: }
4623: for ( r = g; r; r = NEXT(r) ) BDY(r) = (pointer)nd_ps[(int)BDY(r)];
1.103 noro 4624: #if 0
1.74 noro 4625: fprintf(asir_out,"ndv_alloc=%d\n",ndv_alloc);
1.103 noro 4626: #endif
1.69 noro 4627: return g;
4628: }
1.74 noro 4629:
1.133 noro 4630: NODE nd_f4_trace(int m)
4631: {
4632: int i,nh,stat,index;
4633: NODE r,g;
4634: ND_pairs d,l,l0,t;
4635: ND spol,red;
4636: NDV nf,redv,nfqv,nfv;
4637: NM s0,s;
4638: NODE rp0,srp0,nflist;
4639: int nsp,nred,col,rank,len,k,j,a;
4640: UINT c;
4641: UINT **spmat;
4642: UINT *s0vect,*svect,*p,*v;
4643: int *colstat;
4644: IndArray *imat;
4645: int *rhead;
4646: int spcol,sprow;
4647: int sugar;
4648: PGeoBucket bucket;
4649: struct oEGT eg0,eg1,eg_f4;
4650:
4651: g = 0; d = 0;
4652: for ( i = 0; i < nd_psn; i++ ) {
4653: d = update_pairs(d,g,i);
4654: g = update_base(g,i);
4655: }
4656: while ( d ) {
4657: get_eg(&eg0);
4658: l = nd_minsugarp(d,&d);
4659: sugar = SG(l);
4660: bucket = create_pbucket();
4661: stat = nd_sp_f4(m,0,l,bucket);
4662: if ( !stat ) {
4663: for ( t = l; NEXT(t); t = NEXT(t) );
4664: NEXT(t) = d; d = l;
4665: d = nd_reconstruct(1,d);
4666: continue;
4667: }
4668: if ( bucket->m < 0 ) continue;
4669: col = nd_symbolic_preproc(bucket,0,&s0vect,&rp0);
4670: if ( !col ) {
4671: for ( t = l; NEXT(t); t = NEXT(t) );
4672: NEXT(t) = d; d = l;
4673: d = nd_reconstruct(1,d);
4674: continue;
4675: }
4676: get_eg(&eg1); init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg1);
4677: if ( DP_Print )
4678: fprintf(asir_out,"sugar=%d,symb=%fsec,",
4679: sugar,eg_f4.exectime+eg_f4.gctime);
4680: nflist = nd_f4_red(m,l,0,s0vect,col,rp0,&l0);
4681: if ( !l0 ) continue;
4682: l = l0;
4683:
4684: /* over Q */
4685: bucket = create_pbucket();
4686: stat = nd_sp_f4(0,1,l,bucket);
4687: if ( !stat ) {
4688: for ( t = l; NEXT(t); t = NEXT(t) );
4689: NEXT(t) = d; d = l;
4690: d = nd_reconstruct(1,d);
4691: continue;
4692: }
4693: if ( bucket->m < 0 ) continue;
4694: col = nd_symbolic_preproc(bucket,1,&s0vect,&rp0);
4695: if ( !col ) {
4696: for ( t = l; NEXT(t); t = NEXT(t) );
4697: NEXT(t) = d; d = l;
4698: d = nd_reconstruct(1,d);
4699: continue;
4700: }
4701: nflist = nd_f4_red(0,l,1,s0vect,col,rp0,0);
4702: /* adding new bases */
4703: for ( r = nflist; r; r = NEXT(r) ) {
4704: nfqv = (NDV)BDY(r);
4705: ndv_removecont(0,nfqv);
4706: if ( !rem(NM(HCQ(nfqv)),m) ) return 0;
4707: if ( nd_nalg ) {
4708: ND nf1;
4709:
4710: nf1 = ndvtond(m,nfqv);
4711: nd_monic(0,&nf1);
4712: nd_removecont(0,nf1);
4713: nfqv = ndtondv(0,nf1); nd_free(nf1);
4714: }
4715: nfv = ndv_dup(0,nfqv);
4716: ndv_mod(m,nfv);
4717: ndv_removecont(m,nfv);
4718: nh = ndv_newps(0,nfv,nfqv);
4719: d = update_pairs(d,g,nh);
4720: g = update_base(g,nh);
4721: }
4722: }
4723: for ( r = g; r; r = NEXT(r) ) BDY(r) = (pointer)nd_ps_trace[(int)BDY(r)];
4724: #if 0
4725: fprintf(asir_out,"ndv_alloc=%d\n",ndv_alloc);
4726: #endif
4727: return g;
4728: }
4729:
4730: NODE nd_f4_red(int m,ND_pairs sp0,int trace,UINT *s0vect,int col,NODE rp0,ND_pairs *nz)
1.63 noro 4731: {
1.67 noro 4732: IndArray *imat;
1.106 noro 4733: int nsp,nred,i;
1.65 noro 4734: int *rhead;
1.106 noro 4735: NODE r0,rp;
1.74 noro 4736: ND_pairs sp;
4737: NM_ind_pair *rvect;
1.63 noro 4738:
1.74 noro 4739: for ( sp = sp0, nsp = 0; sp; sp = NEXT(sp), nsp++ );
1.106 noro 4740: nred = length(rp0);
1.74 noro 4741: imat = (IndArray *)ALLOCA(nred*sizeof(IndArray));
4742: rhead = (int *)ALLOCA(col*sizeof(int));
4743: for ( i = 0; i < col; i++ ) rhead[i] = 0;
1.63 noro 4744:
1.74 noro 4745: /* construction of index arrays */
4746: rvect = (NM_ind_pair *)ALLOCA(nred*sizeof(NM_ind_pair));
4747: for ( rp = rp0, i = 0; rp; i++, rp = NEXT(rp) ) {
4748: rvect[i] = (NM_ind_pair)BDY(rp);
4749: imat[i] = nm_ind_pair_to_vect_compress(m,s0vect,col,rvect[i]);
4750: rhead[imat[i]->head] = 1;
4751: }
1.107 noro 4752: if ( m )
1.133 noro 4753: r0 = nd_f4_red_main(m,sp0,nsp,s0vect,col,rvect,rhead,imat,nred,nz);
1.107 noro 4754: else
1.133 noro 4755: r0 = nd_f4_red_q_main(sp0,nsp,trace,s0vect,col,rvect,rhead,imat,nred);
1.106 noro 4756: return r0;
4757: }
1.74 noro 4758:
1.106 noro 4759: NODE nd_f4_red_main(int m,ND_pairs sp0,int nsp,UINT *s0vect,int col,
1.133 noro 4760: NM_ind_pair *rvect,int *rhead,IndArray *imat,int nred,ND_pairs *nz)
1.106 noro 4761: {
4762: int spcol,sprow,a;
4763: int i,j,k,l,rank;
4764: NODE r0,r;
4765: ND_pairs sp;
4766: ND spol;
4767: int **spmat;
4768: UINT *svect,*v;
4769: int *colstat;
4770: struct oEGT eg0,eg1,eg2,eg_f4,eg_f4_1,eg_f4_2;
4771: int maxrs;
4772: int *spsugar;
1.133 noro 4773: ND_pairs *spactive;
1.106 noro 4774:
4775: spcol = col-nred;
4776: get_eg(&eg0);
1.74 noro 4777: /* elimination (1st step) */
4778: spmat = (int **)ALLOCA(nsp*sizeof(UINT *));
4779: svect = (UINT *)ALLOCA(col*sizeof(UINT));
1.76 noro 4780: spsugar = (int *)ALLOCA(nsp*sizeof(UINT));
1.133 noro 4781: spactive = !nz?0:(ND_pairs *)ALLOCA(nsp*sizeof(ND_pairs));
1.74 noro 4782: for ( a = sprow = 0, sp = sp0; a < nsp; a++, sp = NEXT(sp) ) {
4783: nd_sp(m,0,sp,&spol);
1.75 noro 4784: if ( !spol ) continue;
1.74 noro 4785: nd_to_vect(m,s0vect,col,spol,svect);
1.76 noro 4786: if ( m == -1 )
4787: maxrs = ndv_reduce_vect_sf(m,svect,col,imat,rvect,nred);
4788: else
4789: maxrs = ndv_reduce_vect(m,svect,col,imat,rvect,nred);
1.74 noro 4790: for ( i = 0; i < col; i++ ) if ( svect[i] ) break;
4791: if ( i < col ) {
4792: spmat[sprow] = v = (UINT *)MALLOC_ATOMIC(spcol*sizeof(UINT));
4793: for ( j = k = 0; j < col; j++ )
4794: if ( !rhead[j] ) v[k++] = svect[j];
1.76 noro 4795: spsugar[sprow] = MAX(maxrs,SG(spol));
1.133 noro 4796: if ( nz )
4797: spactive[sprow] = sp;
1.74 noro 4798: sprow++;
4799: }
1.76 noro 4800: nd_free(spol);
1.74 noro 4801: }
1.80 noro 4802: get_eg(&eg1); init_eg(&eg_f4_1); add_eg(&eg_f4_1,&eg0,&eg1);
4803: if ( DP_Print ) {
4804: fprintf(asir_out,"elim1=%fsec,",eg_f4_1.exectime+eg_f4_1.gctime);
4805: fflush(asir_out);
4806: }
1.74 noro 4807: /* free index arrays */
4808: for ( i = 0; i < nred; i++ ) GC_free(imat[i]->index.c);
4809:
4810: /* elimination (2nd step) */
4811: colstat = (int *)ALLOCA(spcol*sizeof(int));
4812: if ( m == -1 )
1.76 noro 4813: rank = nd_gauss_elim_sf(spmat,spsugar,sprow,spcol,m,colstat);
1.74 noro 4814: else
1.133 noro 4815: rank = nd_gauss_elim_mod(spmat,spsugar,spactive,sprow,spcol,m,colstat);
1.74 noro 4816: r0 = 0;
4817: for ( i = 0; i < rank; i++ ) {
4818: NEXTNODE(r0,r); BDY(r) =
4819: (pointer)vect_to_ndv(spmat[i],spcol,col,rhead,s0vect);
1.76 noro 4820: SG((NDV)BDY(r)) = spsugar[i];
1.74 noro 4821: GC_free(spmat[i]);
4822: }
1.109 noro 4823: if ( r0 ) NEXT(r) = 0;
1.133 noro 4824:
1.74 noro 4825: for ( ; i < sprow; i++ ) GC_free(spmat[i]);
1.80 noro 4826: get_eg(&eg2); init_eg(&eg_f4_2); add_eg(&eg_f4_2,&eg1,&eg2);
4827: init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg2);
1.74 noro 4828: if ( DP_Print ) {
1.80 noro 4829: fprintf(asir_out,"elim2=%fsec\n",eg_f4_2.exectime+eg_f4_2.gctime);
1.74 noro 4830: fprintf(asir_out,"nsp=%d,nred=%d,spmat=(%d,%d),rank=%d ",
4831: nsp,nred,sprow,spcol,rank);
4832: fprintf(asir_out,"%fsec\n",eg_f4.exectime+eg_f4.gctime);
1.63 noro 4833: }
1.133 noro 4834: if ( nz ) {
4835: for ( i = 0; i < rank-1; i++ ) NEXT(spactive[i]) = spactive[i+1];
4836: if ( rank > 0 ) {
4837: NEXT(spactive[rank-1]) = 0;
4838: *nz = spactive[0];
4839: } else
4840: *nz = 0;
4841: }
1.74 noro 4842: return r0;
4843: }
4844:
1.133 noro 4845: #if 1
4846: NODE nd_f4_red_q_main(ND_pairs sp0,int nsp,int trace,UINT *s0vect,int col,
1.107 noro 4847: NM_ind_pair *rvect,int *rhead,IndArray *imat,int nred)
4848: {
4849: int spcol,sprow,a;
4850: int i,j,k,l,rank;
4851: NODE r0,r;
4852: ND_pairs sp;
4853: ND spol;
1.113 noro 4854: Q **spmat;
4855: Q *svect,*v;
1.107 noro 4856: int *colstat;
4857: struct oEGT eg0,eg1,eg2,eg_f4,eg_f4_1,eg_f4_2;
4858: int maxrs;
4859: int *spsugar;
1.137 noro 4860: pointer *w;
1.107 noro 4861:
4862: spcol = col-nred;
4863: get_eg(&eg0);
4864: /* elimination (1st step) */
1.113 noro 4865: spmat = (Q **)ALLOCA(nsp*sizeof(Q *));
4866: svect = (Q *)ALLOCA(col*sizeof(Q));
4867: spsugar = (int *)ALLOCA(nsp*sizeof(Q));
1.107 noro 4868: for ( a = sprow = 0, sp = sp0; a < nsp; a++, sp = NEXT(sp) ) {
1.133 noro 4869: nd_sp(0,trace,sp,&spol);
1.107 noro 4870: if ( !spol ) continue;
1.113 noro 4871: nd_to_vect_q(s0vect,col,spol,svect);
1.133 noro 4872: maxrs = ndv_reduce_vect_q(svect,trace,col,imat,rvect,nred);
1.107 noro 4873: for ( i = 0; i < col; i++ ) if ( svect[i] ) break;
4874: if ( i < col ) {
1.113 noro 4875: spmat[sprow] = v = (Q *)MALLOC(spcol*sizeof(Q));
1.107 noro 4876: for ( j = k = 0; j < col; j++ )
4877: if ( !rhead[j] ) v[k++] = svect[j];
4878: spsugar[sprow] = MAX(maxrs,SG(spol));
4879: sprow++;
4880: }
1.109 noro 4881: /* nd_free(spol); */
1.107 noro 4882: }
4883: get_eg(&eg1); init_eg(&eg_f4_1); add_eg(&eg_f4_1,&eg0,&eg1);
4884: if ( DP_Print ) {
4885: fprintf(asir_out,"elim1=%fsec,",eg_f4_1.exectime+eg_f4_1.gctime);
4886: fflush(asir_out);
4887: }
4888: /* free index arrays */
1.109 noro 4889: /* for ( i = 0; i < nred; i++ ) GC_free(imat[i]->index.c); */
1.107 noro 4890:
4891: /* elimination (2nd step) */
4892: colstat = (int *)ALLOCA(spcol*sizeof(int));
1.113 noro 4893: rank = nd_gauss_elim_q(spmat,spsugar,sprow,spcol,colstat);
1.137 noro 4894: w = (pointer *)ALLOCA(rank*sizeof(pointer));
4895: for ( i = 0; i < rank; i++ ) {
4896: w[rank-i-1] = (pointer)vect_to_ndv_q(spmat[i],spcol,col,rhead,s0vect);
4897: SG((NDV)w[rank-i-1]) = spsugar[i];
4898: /* GC_free(spmat[i]); */
4899: }
1.138 noro 4900: #if 0
1.137 noro 4901: qsort(w,rank,sizeof(NDV),
4902: (int (*)(const void *,const void *))ndv_compare);
4903: #endif
1.107 noro 4904: r0 = 0;
4905: for ( i = 0; i < rank; i++ ) {
1.137 noro 4906: NEXTNODE(r0,r); BDY(r) = w[i];
1.107 noro 4907: }
1.137 noro 4908: if ( r0 ) NEXT(r) = 0;
1.135 noro 4909:
1.109 noro 4910: /* for ( ; i < sprow; i++ ) GC_free(spmat[i]); */
1.107 noro 4911: get_eg(&eg2); init_eg(&eg_f4_2); add_eg(&eg_f4_2,&eg1,&eg2);
4912: init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg2);
4913: if ( DP_Print ) {
4914: fprintf(asir_out,"elim2=%fsec\n",eg_f4_2.exectime+eg_f4_2.gctime);
4915: fprintf(asir_out,"nsp=%d,nred=%d,spmat=(%d,%d),rank=%d ",
4916: nsp,nred,sprow,spcol,rank);
4917: fprintf(asir_out,"%fsec\n",eg_f4.exectime+eg_f4.gctime);
4918: }
4919: return r0;
4920: }
1.129 noro 4921: #else
4922: void printm(Q **mat,int row,int col)
4923: {
4924: int i,j;
4925: printf("[");
4926: for ( i = 0; i < row; i++ ) {
4927: for ( j = 0; j < col; j++ ) {
4928: printexpr(CO,mat[i][j]); printf(" ");
4929: }
4930: printf("]\n");
4931: }
4932: }
4933:
4934: NODE nd_f4_red_q_main(ND_pairs sp0,int nsp,UINT *s0vect,int col,
4935: NM_ind_pair *rvect,int *rhead,IndArray *imat,int nred)
4936: {
4937: int row,a;
4938: int i,j,rank;
4939: NODE r0,r;
4940: ND_pairs sp;
4941: ND spol;
4942: Q **mat;
4943: int *colstat;
4944: int *sugar;
4945:
4946: row = nsp+nred;
4947: /* make the matrix */
4948: mat = (Q **)ALLOCA(row*sizeof(Q *));
4949: sugar = (int *)ALLOCA(row*sizeof(int));
4950: for ( row = a = 0, sp = sp0; a < nsp; a++, sp = NEXT(sp) ) {
4951: nd_sp(0,0,sp,&spol);
4952: if ( !spol ) continue;
4953: mat[row] = (Q *)MALLOC(col*sizeof(Q));
4954: nd_to_vect_q(s0vect,col,spol,mat[row]);
4955: sugar[row] = SG(spol);
4956: row++;
4957: }
4958: for ( i = 0; i < nred; i++, row++ ) {
4959: mat[row] = nm_ind_pair_to_vect(0,s0vect,col,rvect[i]);
4960: sugar[row] = rvect[i]->sugar;
4961: }
4962: /* elimination */
4963: colstat = (int *)ALLOCA(col*sizeof(int));
4964: rank = nd_gauss_elim_q(mat,sugar,row,col,colstat);
4965: r0 = 0;
4966: for ( i = 0; i < rank; i++ ) {
4967: for ( j = 0; j < col; j++ ) if ( mat[i][j] ) break;
4968: if ( j == col ) error("nd_f4_red_q_main : cannot happen");
4969: if ( rhead[j] ) continue;
4970: NEXTNODE(r0,r); BDY(r) =
4971: (pointer)plain_vect_to_ndv_q(mat[i],col,s0vect);
4972: SG((NDV)BDY(r)) = sugar[i];
4973: }
4974: if ( r0 ) NEXT(r) = 0;
4975: printf("\n");
4976: return r0;
4977: }
4978: #endif
1.107 noro 4979:
1.74 noro 4980: FILE *nd_write,*nd_read;
4981:
4982: void nd_send_int(int a) {
4983: write_int(nd_write,&a);
4984: }
4985:
4986: void nd_send_intarray(int *p,int len) {
4987: write_intarray(nd_write,p,len);
4988: }
4989:
4990: int nd_recv_int() {
4991: int a;
4992:
4993: read_int(nd_read,&a);
4994: return a;
4995: }
4996:
4997: void nd_recv_intarray(int *p,int len) {
4998: read_intarray(nd_read,p,len);
4999: }
5000:
5001: void nd_send_ndv(NDV p) {
5002: int len,i;
5003: NMV m;
5004:
5005: if ( !p ) nd_send_int(0);
5006: else {
5007: len = LEN(p);
5008: nd_send_int(len);
5009: m = BDY(p);
5010: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
5011: nd_send_int(CM(m));
5012: nd_send_intarray(DL(m),nd_wpd);
1.68 noro 5013: }
1.74 noro 5014: }
5015: }
5016:
5017: void nd_send_nd(ND p) {
5018: int len,i;
5019: NM m;
5020:
5021: if ( !p ) nd_send_int(0);
5022: else {
5023: len = LEN(p);
5024: nd_send_int(len);
5025: m = BDY(p);
5026: for ( i = 0; i < len; i++, m = NEXT(m) ) {
5027: nd_send_int(CM(m));
5028: nd_send_intarray(DL(m),nd_wpd);
1.65 noro 5029: }
1.74 noro 5030: }
5031: }
1.65 noro 5032:
1.74 noro 5033: NDV nd_recv_ndv()
5034: {
5035: int len,i;
5036: NMV m,m0;
5037: NDV r;
1.65 noro 5038:
1.74 noro 5039: len = nd_recv_int();
5040: if ( !len ) return 0;
5041: else {
5042: m0 = m = (NMV)GC_malloc_atomic_ignore_off_page(nmv_adv*len);
1.103 noro 5043: #if 0
1.74 noro 5044: ndv_alloc += len*nmv_adv;
1.103 noro 5045: #endif
1.74 noro 5046: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
5047: CM(m) = nd_recv_int();
5048: nd_recv_intarray(DL(m),nd_wpd);
1.71 noro 5049: }
1.74 noro 5050: MKNDV(nd_nvar,m0,len,r);
5051: return r;
5052: }
5053: }
1.65 noro 5054:
1.74 noro 5055: int ox_exec_f4_red(Q proc)
5056: {
5057: Obj obj;
5058: STRING fname;
5059: NODE arg;
5060: int s;
5061: extern int ox_need_conv,ox_file_io;
5062:
5063: MKSTR(fname,"nd_exec_f4_red");
5064: arg = mknode(2,proc,fname);
5065: Pox_cmo_rpc(arg,&obj);
5066: s = get_ox_server_id(QTOS(proc));
5067: nd_write = iofp[s].out;
5068: nd_read = iofp[s].in;
5069: ox_need_conv = ox_file_io = 0;
5070: return s;
5071: }
5072:
1.133 noro 5073: NODE nd_f4_red_dist(int m,ND_pairs sp0,UINT *s0vect,int col,NODE rp0,ND_pairs *nz)
1.74 noro 5074: {
5075: int nsp,nred;
5076: int i,rank,s;
5077: NODE rp,r0,r;
5078: ND_pairs sp;
5079: NM_ind_pair pair;
5080: NMV nmv;
5081: NM nm;
5082: NDV nf;
5083: Obj proc,dmy;
5084:
5085: ox_launch_main(0,0,&proc);
5086: s = ox_exec_f4_red((Q)proc);
5087:
5088: nd_send_int(m);
5089: nd_send_int(nd_nvar);
5090: nd_send_int(nd_bpe);
5091: nd_send_int(nd_wpd);
5092: nd_send_int(nmv_adv);
5093:
1.89 noro 5094: saveobj(nd_write,dp_current_spec->obj); fflush(nd_write);
1.74 noro 5095:
5096: nd_send_int(nd_psn);
5097: for ( i = 0; i < nd_psn; i++ ) nd_send_ndv(nd_ps[i]);
5098:
5099: for ( sp = sp0, nsp = 0; sp; sp = NEXT(sp), nsp++ );
5100: nd_send_int(nsp);
5101: for ( i = 0, sp = sp0; i < nsp; i++, sp = NEXT(sp) ) {
5102: nd_send_int(sp->i1); nd_send_int(sp->i2);
5103: }
5104:
5105: nd_send_int(col); nd_send_intarray(s0vect,col*nd_wpd);
5106:
5107: nred = length(rp0); nd_send_int(nred);
5108: for ( i = 0, rp = rp0; i < nred; i++, rp = NEXT(rp) ) {
5109: pair = (NM_ind_pair)BDY(rp);
5110: nd_send_int(pair->index);
5111: nd_send_intarray(pair->mul->dl,nd_wpd);
5112: }
5113: fflush(nd_write);
5114: rank = nd_recv_int();
5115: fprintf(asir_out,"rank=%d\n",rank);
5116: r0 = 0;
5117: for ( i = 0; i < rank; i++ ) {
5118: nf = nd_recv_ndv();
5119: NEXTNODE(r0,r); BDY(r) = (pointer)nf;
5120: }
5121: Pox_shutdown(mknode(1,proc),&dmy);
5122: return r0;
5123: }
5124:
5125: /* server side */
5126:
5127: void nd_exec_f4_red_dist()
5128: {
5129: int m,i,nsp,col,s0size,nred,spcol,j,k;
5130: NM_ind_pair *rp0;
5131: NDV nf;
5132: UINT *s0vect;
5133: IndArray *imat;
5134: int *rhead;
5135: int **spmat;
5136: UINT *svect,*v;
5137: ND_pairs *sp0;
5138: int *colstat;
5139: int a,sprow,rank;
1.89 noro 5140: struct order_spec *ord;
1.74 noro 5141: Obj ordspec;
5142: ND spol;
1.76 noro 5143: int maxrs;
5144: int *spsugar;
1.74 noro 5145:
5146: nd_read = iofp[0].in;
5147: nd_write = iofp[0].out;
5148: m = nd_recv_int();
5149: nd_nvar = nd_recv_int();
5150: nd_bpe = nd_recv_int();
5151: nd_wpd = nd_recv_int();
5152: nmv_adv = nd_recv_int();
5153:
5154: loadobj(nd_read,&ordspec);
1.89 noro 5155: create_order_spec(0,ordspec,&ord);
5156: nd_init_ord(ord);
1.74 noro 5157: nd_setup_parameters(nd_nvar,0);
5158:
5159: nd_psn = nd_recv_int();
5160: nd_ps = (NDV *)MALLOC(nd_psn*sizeof(NDV));
5161: nd_bound = (UINT **)MALLOC(nd_psn*sizeof(UINT *));
5162: for ( i = 0; i < nd_psn; i++ ) {
5163: nd_ps[i] = nd_recv_ndv();
5164: nd_bound[i] = ndv_compute_bound(nd_ps[i]);
5165: }
5166:
5167: nsp = nd_recv_int();
5168: sp0 = (ND_pairs *)MALLOC(nsp*sizeof(ND_pairs));
5169: for ( i = 0; i < nsp; i++ ) {
5170: NEWND_pairs(sp0[i]);
5171: sp0[i]->i1 = nd_recv_int(); sp0[i]->i2 = nd_recv_int();
5172: ndl_lcm(HDL(nd_ps[sp0[i]->i1]),HDL(nd_ps[sp0[i]->i2]),LCM(sp0[i]));
5173: }
5174:
5175: col = nd_recv_int();
5176: s0size = col*nd_wpd;
5177: s0vect = (UINT *)MALLOC(s0size*sizeof(UINT));
5178: nd_recv_intarray(s0vect,s0size);
5179:
5180: nred = nd_recv_int();
5181: rp0 = (NM_ind_pair *)MALLOC(nred*sizeof(NM_ind_pair));
5182: for ( i = 0; i < nred; i++ ) {
5183: rp0[i] = (NM_ind_pair)MALLOC(sizeof(struct oNM_ind_pair));
5184: rp0[i]->index = nd_recv_int();
5185: rp0[i]->mul = (NM)MALLOC(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT));
5186: nd_recv_intarray(rp0[i]->mul->dl,nd_wpd);
5187: }
5188:
5189: spcol = col-nred;
5190: imat = (IndArray *)MALLOC(nred*sizeof(IndArray));
5191: rhead = (int *)MALLOC(col*sizeof(int));
5192: for ( i = 0; i < col; i++ ) rhead[i] = 0;
5193:
5194: /* construction of index arrays */
5195: for ( i = 0; i < nred; i++ ) {
5196: imat[i] = nm_ind_pair_to_vect_compress(m,s0vect,col,rp0[i]);
5197: rhead[imat[i]->head] = 1;
5198: }
5199:
5200: /* elimination (1st step) */
5201: spmat = (int **)MALLOC(nsp*sizeof(UINT *));
5202: svect = (UINT *)MALLOC(col*sizeof(UINT));
1.76 noro 5203: spsugar = (int *)ALLOCA(nsp*sizeof(UINT));
1.74 noro 5204: for ( a = sprow = 0; a < nsp; a++ ) {
5205: nd_sp(m,0,sp0[a],&spol);
1.75 noro 5206: if ( !spol ) continue;
1.74 noro 5207: nd_to_vect(m,s0vect,col,spol,svect);
1.76 noro 5208: if ( m == -1 )
5209: maxrs = ndv_reduce_vect_sf(m,svect,col,imat,rp0,nred);
5210: else
5211: maxrs = ndv_reduce_vect(m,svect,col,imat,rp0,nred);
1.74 noro 5212: for ( i = 0; i < col; i++ ) if ( svect[i] ) break;
5213: if ( i < col ) {
5214: spmat[sprow] = v = (UINT *)MALLOC(spcol*sizeof(UINT));
5215: for ( j = k = 0; j < col; j++ )
5216: if ( !rhead[j] ) v[k++] = svect[j];
1.76 noro 5217: spsugar[sprow] = MAX(maxrs,SG(spol));
1.74 noro 5218: sprow++;
5219: }
1.76 noro 5220: nd_free(spol);
1.74 noro 5221: }
5222: /* elimination (2nd step) */
5223: colstat = (int *)ALLOCA(spcol*sizeof(int));
5224: if ( m == -1 )
1.76 noro 5225: rank = nd_gauss_elim_sf(spmat,spsugar,sprow,spcol,m,colstat);
1.74 noro 5226: else
1.133 noro 5227: rank = nd_gauss_elim_mod(spmat,spsugar,0,sprow,spcol,m,colstat);
1.74 noro 5228: nd_send_int(rank);
5229: for ( i = 0; i < rank; i++ ) {
5230: nf = vect_to_ndv(spmat[i],spcol,col,rhead,s0vect);
5231: nd_send_ndv(nf);
1.63 noro 5232: }
1.74 noro 5233: fflush(nd_write);
1.107 noro 5234: }
5235:
1.113 noro 5236: int nd_gauss_elim_q(Q **mat0,int *sugar,int row,int col,int *colstat)
1.107 noro 5237: {
1.109 noro 5238: int mod,i,j,t,c,rank,rank0,inv;
5239: int *ci,*ri;
1.113 noro 5240: Q dn;
1.109 noro 5241: MAT m,nm;
1.108 noro 5242: int **wmat;
5243:
5244: /* XXX */
5245: mod = 99999989;
5246: wmat = (int **)ALLOCA(row*sizeof(int *));
5247: for ( i = 0; i < row; i++ ) {
5248: wmat[i] = (int *)ALLOCA(col*sizeof(int));
5249: for ( j = 0; j < col; j++ ) {
1.113 noro 5250: if ( mat0[i][j] ) {
5251: t = rem(NM(mat0[i][j]),mod);
5252: if ( SGN(mat0[i][j]) < 0 ) t = mod-t;
5253: wmat[i][j] = t;
5254: } else
1.108 noro 5255: wmat[i][j] = 0;
5256: }
5257: }
1.133 noro 5258: rank0 = nd_gauss_elim_mod(wmat,sugar,0,row,col,mod,colstat);
1.109 noro 5259: NEWMAT(m); m->row = row; m->col = col; m->body = (pointer **)mat0;
1.113 noro 5260: rank = generic_gauss_elim(m,&nm,&dn,&ri,&ci);
1.109 noro 5261: if ( rank != rank0 )
5262: error("afo");
5263: for ( i = 0; i < row; i++ )
5264: for ( j = 0; j < col; j++ )
5265: mat0[i][j] = 0;
5266: c = col-rank;
5267: for ( i = 0; i < rank; i++ ) {
1.113 noro 5268: mat0[i][ri[i]] = dn;
1.109 noro 5269: for ( j = 0; j < c; j++ )
1.113 noro 5270: mat0[i][ci[j]] = (Q)BDY(nm)[i][j];
1.109 noro 5271: }
1.113 noro 5272: inv = invm(rem(NM(dn),mod),mod);
5273: if ( SGN(dn) < 0 ) inv = mod-inv;
1.109 noro 5274: for ( i = 0; i < row; i++ )
5275: for ( j = 0; j < col; j++ ) {
1.113 noro 5276: if ( mat0[i][j] ) {
5277: t = rem(NM(mat0[i][j]),mod);
5278: if ( SGN(mat0[i][j]) < 0 ) t = mod-t;
5279: } else
1.109 noro 5280: t = 0;
5281: c = dmar(t,inv,0,mod);
5282: if ( wmat[i][j] != c )
5283: error("afo");
5284: }
5285: return rank;
1.76 noro 5286: }
5287:
1.133 noro 5288: int nd_gauss_elim_mod(int **mat0,int *sugar,ND_pairs *spactive,int row,int col,int md,int *colstat)
1.76 noro 5289: {
5290: int i,j,k,l,inv,a,rank,s;
5291: unsigned int *t,*pivot,*pk;
5292: unsigned int **mat;
1.133 noro 5293: ND_pairs pair;
1.76 noro 5294:
5295: mat = (unsigned int **)mat0;
5296: for ( rank = 0, j = 0; j < col; j++ ) {
5297: for ( i = rank; i < row; i++ )
5298: mat[i][j] %= md;
5299: for ( i = rank; i < row; i++ )
5300: if ( mat[i][j] )
5301: break;
5302: if ( i == row ) {
5303: colstat[j] = 0;
5304: continue;
5305: } else
5306: colstat[j] = 1;
5307: if ( i != rank ) {
5308: t = mat[i]; mat[i] = mat[rank]; mat[rank] = t;
5309: s = sugar[i]; sugar[i] = sugar[rank]; sugar[rank] = s;
1.133 noro 5310: if ( spactive ) {
5311: pair = spactive[i]; spactive[i] = spactive[rank];
5312: spactive[rank] = pair;
5313: }
1.76 noro 5314: }
5315: pivot = mat[rank];
5316: s = sugar[rank];
5317: inv = invm(pivot[j],md);
5318: for ( k = j, pk = pivot+k; k < col; k++, pk++ )
5319: if ( *pk ) {
5320: if ( *pk >= (unsigned int)md )
5321: *pk %= md;
5322: DMAR(*pk,inv,0,md,*pk)
5323: }
5324: for ( i = rank+1; i < row; i++ ) {
5325: t = mat[i];
5326: if ( a = t[j] ) {
5327: sugar[i] = MAX(sugar[i],s);
5328: red_by_vect(md,t+j,pivot+j,md-a,col-j);
5329: }
5330: }
5331: rank++;
5332: }
5333: for ( j = col-1, l = rank-1; j >= 0; j-- )
5334: if ( colstat[j] ) {
5335: pivot = mat[l];
5336: s = sugar[l];
5337: for ( i = 0; i < l; i++ ) {
5338: t = mat[i];
5339: t[j] %= md;
5340: if ( a = t[j] ) {
5341: sugar[i] = MAX(sugar[i],s);
5342: red_by_vect(md,t+j,pivot+j,md-a,col-j);
5343: }
5344: }
5345: l--;
5346: }
5347: for ( j = 0, l = 0; l < rank; j++ )
5348: if ( colstat[j] ) {
5349: t = mat[l];
5350: for ( k = j; k < col; k++ )
5351: if ( t[k] >= (unsigned int)md )
5352: t[k] %= md;
5353: l++;
5354: }
5355: return rank;
5356: }
5357:
5358: int nd_gauss_elim_sf(int **mat0,int *sugar,int row,int col,int md,int *colstat)
5359: {
5360: int i,j,k,l,inv,a,rank,s;
5361: unsigned int *t,*pivot,*pk;
5362: unsigned int **mat;
5363:
5364: mat = (unsigned int **)mat0;
5365: for ( rank = 0, j = 0; j < col; j++ ) {
5366: for ( i = rank; i < row; i++ )
5367: if ( mat[i][j] )
5368: break;
5369: if ( i == row ) {
5370: colstat[j] = 0;
5371: continue;
5372: } else
5373: colstat[j] = 1;
5374: if ( i != rank ) {
5375: t = mat[i]; mat[i] = mat[rank]; mat[rank] = t;
5376: s = sugar[i]; sugar[i] = sugar[rank]; sugar[rank] = s;
5377: }
5378: pivot = mat[rank];
5379: s = sugar[rank];
5380: inv = _invsf(pivot[j]);
5381: for ( k = j, pk = pivot+k; k < col; k++, pk++ )
5382: if ( *pk )
5383: *pk = _mulsf(*pk,inv);
5384: for ( i = rank+1; i < row; i++ ) {
5385: t = mat[i];
5386: if ( a = t[j] ) {
5387: sugar[i] = MAX(sugar[i],s);
5388: red_by_vect_sf(md,t+j,pivot+j,_chsgnsf(a),col-j);
5389: }
5390: }
5391: rank++;
5392: }
5393: for ( j = col-1, l = rank-1; j >= 0; j-- )
5394: if ( colstat[j] ) {
5395: pivot = mat[l];
5396: s = sugar[l];
5397: for ( i = 0; i < l; i++ ) {
5398: t = mat[i];
5399: if ( a = t[j] ) {
5400: sugar[i] = MAX(sugar[i],s);
5401: red_by_vect_sf(md,t+j,pivot+j,_chsgnsf(a),col-j);
5402: }
5403: }
5404: l--;
5405: }
5406: return rank;
1.77 noro 5407: }
5408:
5409: int ndv_ishomo(NDV p)
5410: {
5411: NMV m;
5412: int len,h;
5413:
5414: if ( !p ) return 1;
5415: len = LEN(p);
5416: m = BDY(p);
5417: h = TD(DL(m));
5418: NMV_ADV(m);
5419: for ( len--; len; len--, NMV_ADV(m) )
5420: if ( TD(DL(m)) != h ) return 0;
5421: return 1;
5422: }
5423:
5424: void ndv_save(NDV p,int index)
5425: {
5426: FILE *s;
5427: char name[BUFSIZ];
5428: short id;
5429: int nv,sugar,len,n,i,td,e,j;
5430: NMV m;
5431: unsigned int *dl;
5432:
5433: sprintf(name,"%s/%d",Demand,index);
5434: s = fopen(name,"w");
5435: savevl(s,0);
5436: if ( !p ) {
5437: saveobj(s,0);
5438: return;
5439: }
5440: id = O_DP;
5441: nv = NV(p);
5442: sugar = SG(p);
5443: len = LEN(p);
5444: write_short(s,&id); write_int(s,&nv); write_int(s,&sugar);
5445: write_int(s,&len);
5446:
5447: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
1.113 noro 5448: saveobj(s,(Obj)CQ(m));
1.77 noro 5449: dl = DL(m);
5450: td = TD(dl);
5451: write_int(s,&td);
5452: for ( j = 0; j < nv; j++ ) {
5453: e = GET_EXP(dl,j);
5454: write_int(s,&e);
5455: }
5456: }
5457: fclose(s);
5458: }
5459:
5460: NDV ndv_load(int index)
5461: {
5462: FILE *s;
5463: char name[BUFSIZ];
5464: short id;
5465: int nv,sugar,len,n,i,td,e,j;
5466: NDV d;
5467: NMV m0,m;
5468: unsigned int *dl;
5469: Obj obj;
5470:
5471: sprintf(name,"%s/%d",Demand,index);
5472: s = fopen(name,"r");
5473: if ( !s ) return 0;
5474:
5475: skipvl(s);
5476: read_short(s,&id);
5477: if ( !id ) return 0;
5478: read_int(s,&nv);
5479: read_int(s,&sugar);
5480: read_int(s,&len);
5481:
5482: m0 = m = MALLOC(len*nmv_adv);
5483: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
1.113 noro 5484: loadobj(s,&obj); CQ(m) = (Q)obj;
1.77 noro 5485: dl = DL(m);
5486: ndl_zero(dl);
5487: read_int(s,&td); TD(dl) = td;
5488: for ( j = 0; j < nv; j++ ) {
5489: read_int(s,&e);
5490: PUT_EXP(dl,j,e);
5491: }
5492: if ( nd_blockmask ) ndl_weight_mask(dl);
5493: }
5494: fclose(s);
5495: MKNDV(nv,m0,len,d);
5496: SG(d) = sugar;
5497: return d;
1.99 noro 5498: }
5499:
1.102 noro 5500: void nd_det(int mod,MAT f,P *rp)
1.99 noro 5501: {
5502: VL fv,tv;
1.102 noro 5503: int n,i,j,max,e,nvar,sgn,k0,l0,len0,len,k,l,a;
1.99 noro 5504: pointer **m;
1.113 noro 5505: Q mone;
1.125 noro 5506: P **w;
1.126 noro 5507: P mp,r;
1.102 noro 5508: NDV **dm;
5509: NDV *t,*mi,*mj;
5510: NDV d,s,mij,mjj;
5511: ND u;
5512: NMV nmv;
1.114 noro 5513: UINT *bound;
1.99 noro 5514: PGeoBucket bucket;
5515: struct order_spec *ord;
1.126 noro 5516: Q dq,dt,ds;
5517: N gn,qn,dn0,nm,dn;
1.99 noro 5518:
5519: create_order_spec(0,0,&ord);
5520: nd_init_ord(ord);
5521: get_vars((Obj)f,&fv);
5522: if ( f->row != f->col )
5523: error("nd_det : non-square matrix");
5524: n = f->row;
1.125 noro 5525: m = f->body;
1.99 noro 5526: for ( nvar = 0, tv = fv; tv; tv = NEXT(tv), nvar++ );
1.125 noro 5527:
5528: if ( !nvar ) {
5529: if ( !mod )
5530: detp(CO,(P **)m,n,rp);
5531: else {
5532: w = (P **)almat_pointer(n,n);
5533: for ( i = 0; i < n; i++ )
5534: for ( j = 0; j < n; j++ )
5535: ptomp(mod,(P)m[i][j],&w[i][j]);
5536: detmp(CO,mod,w,n,&mp);
5537: mptop(mp,rp);
5538: }
5539: return;
5540: }
1.126 noro 5541:
5542: if ( !mod ) {
5543: w = (P **)almat_pointer(n,n);
5544: dq = ONE;
5545: for ( i = 0; i < n; i++ ) {
5546: dn0 = ONEN;
5547: for ( j = 0; j < n; j++ ) {
5548: if ( !m[i][j] ) continue;
5549: lgp(m[i][j],&nm,&dn);
5550: gcdn(dn0,dn,&gn); divsn(dn0,gn,&qn); muln(qn,dn,&dn0);
5551: }
5552: if ( !UNIN(dn0) ) {
5553: NTOQ(dn0,1,ds);
5554: for ( j = 0; j < n; j++ )
5555: mulp(CO,(P)m[i][j],(P)ds,&w[i][j]);
5556: mulq(dq,ds,&dt); dq = dt;
5557: } else
5558: for ( j = 0; j < n; j++ )
5559: w[i][j] = (P)m[i][j];
5560: }
5561: m = (pointer **)w;
5562: }
1.125 noro 5563:
1.99 noro 5564: for ( i = 0, max = 0; i < n; i++ )
5565: for ( j = 0; j < n; j++ )
5566: for ( tv = fv; tv; tv = NEXT(tv) ) {
5567: e = getdeg(tv->v,(P)m[i][j]);
5568: max = MAX(e,max);
5569: }
1.114 noro 5570: nd_setup_parameters(nvar,max);
1.102 noro 5571: dm = (NDV **)almat_pointer(n,n);
1.99 noro 5572: for ( i = 0, max = 0; i < n; i++ )
1.102 noro 5573: for ( j = 0; j < n; j++ ) {
5574: dm[i][j] = ptondv(CO,fv,m[i][j]);
5575: if ( mod ) ndv_mod(mod,dm[i][j]);
5576: if ( dm[i][j] && !LEN(dm[i][j]) ) dm[i][j] = 0;
5577: }
5578: d = ptondv(CO,fv,(P)ONE);
5579: if ( mod ) ndv_mod(mod,d);
1.113 noro 5580: chsgnq(ONE,&mone);
1.99 noro 5581: for ( j = 0, sgn = 1; j < n; j++ ) {
1.125 noro 5582: if ( DP_Print ) fprintf(stderr,".",j);
1.99 noro 5583: for ( i = j; i < n && !dm[i][j]; i++ );
5584: if ( i == n ) {
5585: *rp = 0;
5586: return;
5587: }
1.102 noro 5588: k0 = i; l0 = j; len0 = LEN(dm[k0][l0]);
1.99 noro 5589: for ( k = j; k < n; k++ )
5590: for ( l = j; l < n; l++ )
1.102 noro 5591: if ( dm[k][l] && LEN(dm[k][l]) < len0 ) {
5592: k0 = k; l0 = l; len0 = LEN(dm[k][l]);
1.99 noro 5593: }
5594: if ( k0 != j ) {
5595: t = dm[j]; dm[j] = dm[k0]; dm[k0] = t;
5596: sgn = -sgn;
5597: }
5598: if ( l0 != j ) {
5599: for ( k = j; k < n; k++ ) {
5600: s = dm[k][j]; dm[k][j] = dm[k][l0]; dm[k][l0] = s;
5601: }
5602: sgn = -sgn;
5603: }
1.114 noro 5604: bound = nd_det_compute_bound(dm,n,j);
5605: if ( ndl_check_bound(bound,bound) )
5606: nd_det_reconstruct(dm,n,j,d);
5607:
1.99 noro 5608: for ( i = j+1, mj = dm[j], mjj = mj[j]; i < n; i++ ) {
1.114 noro 5609: /* if ( DP_Print ) fprintf(stderr," i=%d\n ",i); */
1.99 noro 5610: mi = dm[i]; mij = mi[j];
1.102 noro 5611: if ( mod )
5612: ndv_mul_c(mod,mij,mod-1);
5613: else
1.113 noro 5614: ndv_mul_c_q(mij,mone);
1.99 noro 5615: for ( k = j+1; k < n; k++ ) {
1.114 noro 5616: /* if ( DP_Print ) fprintf(stderr,"k=%d ",k); */
1.99 noro 5617: bucket = create_pbucket();
1.104 noro 5618: if ( mi[k] ) {
1.102 noro 5619: nmv = BDY(mjj); len = LEN(mjj);
5620: for ( a = 0; a < len; a++, NMV_ADV(nmv) ) {
5621: u = ndv_mul_nmv_trunc(mod,nmv,mi[k],DL(BDY(d)));
5622: add_pbucket(mod,bucket,u);
1.99 noro 5623: }
1.104 noro 5624: }
1.99 noro 5625: if ( mj[k] && mij ) {
1.102 noro 5626: nmv = BDY(mij); len = LEN(mij);
5627: for ( a = 0; a < len; a++, NMV_ADV(nmv) ) {
5628: u = ndv_mul_nmv_trunc(mod,nmv,mj[k],DL(BDY(d)));
5629: add_pbucket(mod,bucket,u);
1.99 noro 5630: }
5631: }
1.104 noro 5632: u = nd_quo(mod,bucket,d);
1.102 noro 5633: mi[k] = ndtondv(mod,u);
1.99 noro 5634: }
1.114 noro 5635: /* if ( DP_Print ) fprintf(stderr,"\n",k); */
1.99 noro 5636: }
5637: d = mjj;
5638: }
1.125 noro 5639: if ( DP_Print ) fprintf(stderr,"\n",k);
1.100 noro 5640: if ( sgn < 0 )
1.102 noro 5641: if ( mod )
5642: ndv_mul_c(mod,d,mod-1);
5643: else
1.113 noro 5644: ndv_mul_c_q(d,mone);
1.126 noro 5645: r = ndvtop(mod,CO,fv,d);
5646: if ( !mod && !UNIQ(dq) )
5647: divsp(CO,r,(P)dq,rp);
5648: else
5649: *rp = r;
1.99 noro 5650: }
5651:
1.102 noro 5652: ND ndv_mul_nmv_trunc(int mod,NMV m0,NDV p,UINT *d)
1.99 noro 5653: {
5654: NM mr,mr0;
1.102 noro 5655: NM tnm;
5656: NMV m;
1.99 noro 5657: UINT *d0,*dt,*dm;
5658: int c,n,td,i,c1,c2,len;
1.113 noro 5659: Q q;
1.99 noro 5660: ND r;
5661:
5662: if ( !p ) return 0;
5663: else {
1.102 noro 5664: n = NV(p); m = BDY(p); len = LEN(p);
1.99 noro 5665: d0 = DL(m0);
1.102 noro 5666: td = TD(d);
1.99 noro 5667: mr0 = 0;
5668: NEWNM(tnm);
1.102 noro 5669: if ( mod ) {
5670: c = CM(m0);
5671: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
5672: ndl_add(DL(m),d0,DL(tnm));
5673: if ( ndl_reducible(DL(tnm),d) ) {
5674: NEXTNM(mr0,mr);
5675: c1 = CM(m); DMAR(c1,c,0,mod,c2); CM(mr) = c2;
1.104 noro 5676: ndl_copy(DL(tnm),DL(mr));
1.102 noro 5677: }
5678: }
5679: } else {
1.113 noro 5680: q = CQ(m0);
1.102 noro 5681: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
5682: ndl_add(DL(m),d0,DL(tnm));
5683: if ( ndl_reducible(DL(tnm),d) ) {
5684: NEXTNM(mr0,mr);
1.113 noro 5685: mulq(CQ(m),q,&CQ(mr));
1.104 noro 5686: ndl_copy(DL(tnm),DL(mr));
1.102 noro 5687: }
1.99 noro 5688: }
5689: }
5690: if ( !mr0 )
5691: return 0;
5692: else {
5693: NEXT(mr) = 0;
1.105 noro 5694: for ( len = 0, mr = mr0; mr; mr = NEXT(mr), len++ );
1.99 noro 5695: MKND(NV(p),mr0,len,r);
5696: SG(r) = SG(p) + TD(d0);
5697: return r;
5698: }
5699: }
1.114 noro 5700: }
5701:
5702: void nd_det_reconstruct(NDV **dm,int n,int j,NDV d)
5703: {
5704: int i,obpe,oadv,h,k,l;
5705: static NM prev_nm_free_list;
5706: EPOS oepos;
5707:
5708: obpe = nd_bpe;
5709: oadv = nmv_adv;
5710: oepos = nd_epos;
5711: if ( obpe < 2 ) nd_bpe = 2;
5712: else if ( obpe < 3 ) nd_bpe = 3;
5713: else if ( obpe < 4 ) nd_bpe = 4;
5714: else if ( obpe < 5 ) nd_bpe = 5;
5715: else if ( obpe < 6 ) nd_bpe = 6;
5716: else if ( obpe < 8 ) nd_bpe = 8;
5717: else if ( obpe < 10 ) nd_bpe = 10;
5718: else if ( obpe < 16 ) nd_bpe = 16;
5719: else if ( obpe < 32 ) nd_bpe = 32;
5720: else error("nd_det_reconstruct : exponent too large");
5721:
5722: nd_setup_parameters(nd_nvar,0);
5723: prev_nm_free_list = _nm_free_list;
5724: _nm_free_list = 0;
5725: for ( k = j; k < n; k++ )
5726: for (l = j; l < n; l++ )
5727: ndv_realloc(dm[k][l],obpe,oadv,oepos);
5728: ndv_realloc(d,obpe,oadv,oepos);
5729: prev_nm_free_list = 0;
5730: #if 0
5731: GC_gcollect();
5732: #endif
5733: }
5734:
5735: UINT *nd_det_compute_bound(NDV **dm,int n,int j)
5736: {
5737: UINT *d0,*d1,*d,*t,*r;
5738: int k,l;
5739:
5740: d0 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
5741: d1 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
5742: for ( k = 0; k < nd_wpd; k++ ) d0[k] = 0;
5743: for ( k = j; k < n; k++ )
5744: for ( l = j; l < n; l++ )
5745: if ( dm[k][l] ) {
5746: d = ndv_compute_bound(dm[k][l]);
5747: ndl_lcm(d,d0,d1);
5748: t = d1; d1 = d0; d0 = t;
5749: }
5750: r = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
5751: for ( k = 0; k < nd_wpd; k++ ) r[k] = d0[k];
5752: return r;
1.117 noro 5753: }
5754:
5755: DL nd_separate_d(UINT *d,UINT *trans)
5756: {
1.121 noro 5757: int n,td,i,e,j;
1.117 noro 5758: DL a;
5759:
5760: ndl_zero(trans);
5761: td = 0;
1.121 noro 5762: for ( i = 0; i < nd_ntrans; i++ ) {
5763: e = GET_EXP(d,i);
5764: PUT_EXP(trans,i,e);
5765: td += MUL_WEIGHT(e,i);
5766: }
5767: if ( nd_ntrans+nd_nalg < nd_nvar ) {
5768: /* homogenized */
5769: i = nd_nvar-1;
1.117 noro 5770: e = GET_EXP(d,i);
5771: PUT_EXP(trans,i,e);
5772: td += MUL_WEIGHT(e,i);
5773: }
5774: TD(trans) = td;
5775: if ( nd_blockmask) ndl_weight_mask(trans);
5776: NEWDL(a,nd_nalg);
5777: td = 0;
1.121 noro 5778: for ( i = 0; i < nd_nalg; i++ ) {
5779: j = nd_ntrans+i;
5780: e = GET_EXP(d,j);
5781: a->d[i] = e;
1.117 noro 5782: td += e;
5783: }
5784: a->td = td;
5785: return a;
5786: }
5787:
1.118 noro 5788: int nd_monic(int mod,ND *p)
1.117 noro 5789: {
5790: UINT *trans,*t;
5791: DL alg;
5792: MP mp0,mp;
1.118 noro 5793: NM m,m0,m1,ma0,ma,mb,mr0,mr;
5794: ND r;
1.117 noro 5795: DL dl;
5796: DP nm;
5797: NDV ndv;
1.118 noro 5798: DAlg inv,cd;
1.117 noro 5799: ND s,c;
1.118 noro 5800: Q l,mul;
5801: N ln;
5802: int n,ntrans,i,e,td,is_lc,len;
5803: NumberField nf;
5804: struct oEGT eg0,eg1;
5805:
5806: if ( !(nf = get_numberfield()) )
5807: error("nd_monic : current_numberfield is not set");
1.117 noro 5808:
1.118 noro 5809: /* Q coef -> DAlg coef */
5810: NEWNM(ma0); ma = ma0;
5811: m = BDY(*p);
5812: is_lc = 1;
5813: while ( 1 ) {
5814: NEWMP(mp0); mp = mp0;
5815: mp->c = (P)CQ(m);
5816: mp->dl = nd_separate_d(DL(m),DL(ma));
5817: NEWNM(mb);
5818: for ( m = NEXT(m); m; m = NEXT(m) ) {
5819: alg = nd_separate_d(DL(m),DL(mb));
5820: if ( !ndl_equal(DL(ma),DL(mb)) )
5821: break;
5822: NEXTMP(mp0,mp); mp->c = (P)CQ(m); mp->dl = alg;
5823: }
5824: NEXT(mp) = 0;
5825: MKDP(nd_nalg,mp0,nm);
5826: MKDAlg(nm,ONE,cd);
5827: if ( is_lc == 1 ) {
5828: /* if the lc is a rational number, we have nothing to do */
5829: if ( !mp0->dl->td )
5830: return 1;
5831:
5832: get_eg(&eg0);
5833: invdalg(cd,&inv);
5834: get_eg(&eg1); add_eg(&eg_invdalg,&eg0,&eg1);
5835: /* check the validity of inv */
5836: if ( mod && !rem(NM(inv->dn),mod) )
5837: return 0;
5838: CA(ma) = nf->one;
5839: is_lc = 0;
5840: ln = ONEN;
5841: } else {
5842: muldalg(cd,inv,&CA(ma));
5843: lcmn(ln,NM(CA(ma)->dn),&ln);
5844: }
5845: if ( m ) {
5846: NEXT(ma) = mb; ma = mb;
5847: } else {
5848: NEXT(ma) = 0;
5849: break;
5850: }
5851: }
5852: /* l = lcm(denoms) */
5853: NTOQ(ln,1,l);
5854: for ( mr0 = 0, m = ma0; m; m = NEXT(m) ) {
5855: divq(l,CA(m)->dn,&mul);
5856: for ( mp = BDY(CA(m)->nm); mp; mp = NEXT(mp) ) {
5857: NEXTNM(mr0,mr);
5858: mulq((Q)mp->c,mul,&CQ(mr));
5859: dl = mp->dl;
5860: td = TD(DL(m));
5861: ndl_copy(DL(m),DL(mr));
1.121 noro 5862: for ( i = 0; i < nd_nalg; i++ ) {
5863: e = dl->d[i];
5864: PUT_EXP(DL(mr),i+nd_ntrans,e);
5865: td += MUL_WEIGHT(e,i+nd_ntrans);
1.118 noro 5866: }
5867: TD(DL(mr)) = td;
5868: if ( nd_blockmask) ndl_weight_mask(DL(mr));
5869: }
1.117 noro 5870: }
1.118 noro 5871: NEXT(mr) = 0;
5872: for ( len = 0, mr = mr0; mr; mr = NEXT(mr), len++ );
5873: MKND(NV(*p),mr0,len,r);
5874: /* XXX */
5875: SG(r) = SG(*p);
5876: nd_free(*p);
5877: *p = r;
5878: return 1;
1.59 noro 5879: }
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>