Annotation of OpenXM_contrib2/asir2000/engine/nd.c, Revision 1.102
1.102 ! noro 1: /* $OpenXM: OpenXM_contrib2/asir2000/engine/nd.c,v 1.101 2004/09/14 10:00:26 noro Exp $ */
1.2 noro 2:
1.94 noro 3: #include "nd.h"
1.63 noro 4:
1.61 noro 5: int (*ndl_compare_function)(UINT *a1,UINT *a2);
1.94 noro 6: int nd_dcomp;
7: NM _nm_free_list;
8: ND _nd_free_list;
9: ND_pairs _ndp_free_list;
1.32 noro 10:
1.74 noro 11: static int ndv_alloc;
1.87 noro 12: #if 1
1.69 noro 13: static int nd_f4_nsp=0x7fffffff;
1.87 noro 14: #else
15: static int nd_f4_nsp=50;
16: #endif
1.42 noro 17: static double nd_scale=2;
1.61 noro 18: static UINT **nd_bound;
1.42 noro 19: static struct order_spec *nd_ord;
20: static EPOS nd_epos;
1.43 noro 21: static BlockMask nd_blockmask;
1.42 noro 22: static int nd_nvar;
23: static int nd_isrlex;
24: static int nd_epw,nd_bpe,nd_wpd,nd_exporigin;
1.61 noro 25: static UINT nd_mask[32];
26: static UINT nd_mask0,nd_mask1;
1.42 noro 27:
1.20 noro 28: static NDV *nd_ps;
1.53 noro 29: static NDV *nd_ps_trace;
1.42 noro 30: static RHist *nd_psh;
31: static int nd_psn,nd_pslen;
32: static RHist *nd_red;
1.96 noro 33: static int *nd_work_vector;
34: static int **nd_matrix;
35: static int nd_matrix_len;
1.97 noro 36: static struct weight_or_block *nd_worb;
37: static int nd_worb_len;
1.42 noro 38: static int nd_found,nd_create,nd_notfirst;
39: static int nmv_adv;
1.77 noro 40: static int nd_demand;
1.1 noro 41:
42: void nd_free_private_storage()
43: {
44: _nm_free_list = 0;
1.5 noro 45: _ndp_free_list = 0;
1.71 noro 46: #if 0
1.1 noro 47: GC_gcollect();
1.71 noro 48: #endif
1.1 noro 49: }
50:
51: void _NM_alloc()
52: {
53: NM p;
54: int i;
55:
1.11 noro 56: for ( i = 0; i < 1024; i++ ) {
1.61 noro 57: p = (NM)GC_malloc(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT));
1.1 noro 58: p->next = _nm_free_list; _nm_free_list = p;
59: }
60: }
61:
62: void _ND_alloc()
63: {
64: ND p;
65: int i;
66:
67: for ( i = 0; i < 1024; i++ ) {
68: p = (ND)GC_malloc(sizeof(struct oND));
69: p->body = (NM)_nd_free_list; _nd_free_list = p;
70: }
71: }
72:
73: void _NDP_alloc()
74: {
75: ND_pairs p;
76: int i;
77:
1.11 noro 78: for ( i = 0; i < 1024; i++ ) {
1.1 noro 79: p = (ND_pairs)GC_malloc(sizeof(struct oND_pairs)
1.61 noro 80: +(nd_wpd-1)*sizeof(UINT));
1.1 noro 81: p->next = _ndp_free_list; _ndp_free_list = p;
82: }
83: }
84:
1.30 noro 85: INLINE int nd_length(ND p)
1.1 noro 86: {
87: NM m;
88: int i;
89:
90: if ( !p )
91: return 0;
92: else {
93: for ( i = 0, m = BDY(p); m; m = NEXT(m), i++ );
94: return i;
95: }
96: }
97:
1.61 noro 98: INLINE int ndl_reducible(UINT *d1,UINT *d2)
1.1 noro 99: {
1.61 noro 100: UINT u1,u2;
1.1 noro 101: int i,j;
102:
1.34 noro 103: if ( TD(d1) < TD(d2) ) return 0;
1.65 noro 104: #if USE_UNROLL
1.1 noro 105: switch ( nd_bpe ) {
1.62 noro 106: case 3:
107: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
108: u1 = d1[i]; u2 = d2[i];
109: if ( (u1&0x38000000) < (u2&0x38000000) ) return 0;
110: if ( (u1& 0x7000000) < (u2& 0x7000000) ) return 0;
111: if ( (u1& 0xe00000) < (u2& 0xe00000) ) return 0;
112: if ( (u1& 0x1c0000) < (u2& 0x1c0000) ) return 0;
113: if ( (u1& 0x38000) < (u2& 0x38000) ) return 0;
114: if ( (u1& 0x7000) < (u2& 0x7000) ) return 0;
115: if ( (u1& 0xe00) < (u2& 0xe00) ) return 0;
116: if ( (u1& 0x1c0) < (u2& 0x1c0) ) return 0;
117: if ( (u1& 0x38) < (u2& 0x38) ) return 0;
118: if ( (u1& 0x7) < (u2& 0x7) ) return 0;
119: }
120: return 1;
121: break;
1.1 noro 122: case 4:
1.41 noro 123: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 124: u1 = d1[i]; u2 = d2[i];
125: if ( (u1&0xf0000000) < (u2&0xf0000000) ) return 0;
1.62 noro 126: if ( (u1& 0xf000000) < (u2& 0xf000000) ) return 0;
127: if ( (u1& 0xf00000) < (u2& 0xf00000) ) return 0;
128: if ( (u1& 0xf0000) < (u2& 0xf0000) ) return 0;
129: if ( (u1& 0xf000) < (u2& 0xf000) ) return 0;
130: if ( (u1& 0xf00) < (u2& 0xf00) ) return 0;
131: if ( (u1& 0xf0) < (u2& 0xf0) ) return 0;
132: if ( (u1& 0xf) < (u2& 0xf) ) return 0;
1.1 noro 133: }
134: return 1;
135: break;
136: case 6:
1.41 noro 137: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 138: u1 = d1[i]; u2 = d2[i];
139: if ( (u1&0x3f000000) < (u2&0x3f000000) ) return 0;
1.62 noro 140: if ( (u1& 0xfc0000) < (u2& 0xfc0000) ) return 0;
141: if ( (u1& 0x3f000) < (u2& 0x3f000) ) return 0;
142: if ( (u1& 0xfc0) < (u2& 0xfc0) ) return 0;
143: if ( (u1& 0x3f) < (u2& 0x3f) ) return 0;
1.1 noro 144: }
145: return 1;
146: break;
147: case 8:
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&0xff000000) < (u2&0xff000000) ) return 0;
1.62 noro 151: if ( (u1& 0xff0000) < (u2& 0xff0000) ) return 0;
152: if ( (u1& 0xff00) < (u2& 0xff00) ) return 0;
153: if ( (u1& 0xff) < (u2& 0xff) ) return 0;
1.1 noro 154: }
155: return 1;
156: break;
157: case 16:
1.41 noro 158: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 159: u1 = d1[i]; u2 = d2[i];
160: if ( (u1&0xffff0000) < (u2&0xffff0000) ) return 0;
1.62 noro 161: if ( (u1& 0xffff) < (u2& 0xffff) ) return 0;
1.1 noro 162: }
163: return 1;
164: break;
165: case 32:
1.41 noro 166: for ( i = nd_exporigin; i < nd_wpd; i++ )
1.1 noro 167: if ( d1[i] < d2[i] ) return 0;
168: return 1;
169: break;
170: default:
1.41 noro 171: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 172: u1 = d1[i]; u2 = d2[i];
173: for ( j = 0; j < nd_epw; j++ )
174: if ( (u1&nd_mask[j]) < (u2&nd_mask[j]) ) return 0;
175: }
176: return 1;
177: }
1.65 noro 178: #else
179: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
180: u1 = d1[i]; u2 = d2[i];
181: for ( j = 0; j < nd_epw; j++ )
182: if ( (u1&nd_mask[j]) < (u2&nd_mask[j]) ) return 0;
183: }
184: return 1;
185: #endif
1.1 noro 186: }
187:
1.61 noro 188: /*
189: * If the current order is a block order,
190: * then the last block is length 1 and contains
191: * the homo variable. Otherwise, the original
192: * order is either 0 or 2.
193: */
194:
195: void ndl_homogenize(UINT *d,UINT *r,int obpe,EPOS oepos,int weight)
1.23 noro 196: {
1.61 noro 197: int w,i,e,n,omask0;
198:
199: omask0 = (1<<obpe)-1;
200: n = nd_nvar-1;
201: ndl_zero(r);
202: for ( i = 0; i < n; i++ ) {
203: e = GET_EXP_OLD(d,i);
204: PUT_EXP(r,i,e);
205: }
206: w = TD(d);
207: PUT_EXP(r,nd_nvar-1,weight-w);
208: TD(r) = weight;
209: if ( nd_blockmask ) ndl_weight_mask(r);
210: }
211:
212: void ndl_dehomogenize(UINT *d)
213: {
214: UINT mask;
215: UINT h;
1.31 noro 216: int i,bits;
1.23 noro 217:
1.44 noro 218: if ( nd_blockmask ) {
219: h = GET_EXP(d,nd_nvar-1);
1.45 noro 220: XOR_EXP(d,nd_nvar-1,h);
1.44 noro 221: TD(d) -= h;
222: d[nd_exporigin-1] -= h;
223: } else {
224: if ( nd_isrlex ) {
225: if ( nd_bpe == 32 ) {
226: h = d[nd_exporigin];
227: for ( i = nd_exporigin+1; i < nd_wpd; i++ )
228: d[i-1] = d[i];
229: d[i-1] = 0;
230: TD(d) -= h;
231: } else {
232: bits = nd_epw*nd_bpe;
233: mask = bits==32?0xffffffff:((1<<(nd_epw*nd_bpe))-1);
234: h = (d[nd_exporigin]>>((nd_epw-1)*nd_bpe))&nd_mask0;
235: for ( i = nd_exporigin; i < nd_wpd; i++ )
236: d[i] = ((d[i]<<nd_bpe)&mask)
237: |(i+1<nd_wpd?((d[i+1]>>((nd_epw-1)*nd_bpe))&nd_mask0):0);
238: TD(d) -= h;
239: }
1.45 noro 240: } else {
241: h = GET_EXP(d,nd_nvar-1);
242: XOR_EXP(d,nd_nvar-1,h);
243: TD(d) -= h;
244: }
1.44 noro 245: }
1.23 noro 246: }
247:
1.61 noro 248: void ndl_lcm(UINT *d1,unsigned *d2,UINT *d)
1.1 noro 249: {
1.61 noro 250: UINT t1,t2,u,u1,u2;
1.43 noro 251: int i,j,l;
1.1 noro 252:
1.65 noro 253: #if USE_UNROLL
1.1 noro 254: switch ( nd_bpe ) {
1.62 noro 255: case 3:
256: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
257: u1 = d1[i]; u2 = d2[i];
258: t1 = (u1&0x38000000); t2 = (u2&0x38000000); u = t1>t2?t1:t2;
259: t1 = (u1& 0x7000000); t2 = (u2& 0x7000000); u |= t1>t2?t1:t2;
260: t1 = (u1& 0xe00000); t2 = (u2& 0xe00000); u |= t1>t2?t1:t2;
261: t1 = (u1& 0x1c0000); t2 = (u2& 0x1c0000); u |= t1>t2?t1:t2;
262: t1 = (u1& 0x38000); t2 = (u2& 0x38000); u |= t1>t2?t1:t2;
263: t1 = (u1& 0x7000); t2 = (u2& 0x7000); u |= t1>t2?t1:t2;
264: t1 = (u1& 0xe00); t2 = (u2& 0xe00); u |= t1>t2?t1:t2;
265: t1 = (u1& 0x1c0); t2 = (u2& 0x1c0); u |= t1>t2?t1:t2;
266: t1 = (u1& 0x38); t2 = (u2& 0x38); u |= t1>t2?t1:t2;
267: t1 = (u1& 0x7); t2 = (u2& 0x7); u |= t1>t2?t1:t2;
268: d[i] = u;
269: }
270: break;
1.1 noro 271: case 4:
1.41 noro 272: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 273: u1 = d1[i]; u2 = d2[i];
274: t1 = (u1&0xf0000000); t2 = (u2&0xf0000000); u = t1>t2?t1:t2;
1.62 noro 275: t1 = (u1& 0xf000000); t2 = (u2& 0xf000000); u |= t1>t2?t1:t2;
276: t1 = (u1& 0xf00000); t2 = (u2& 0xf00000); u |= t1>t2?t1:t2;
277: t1 = (u1& 0xf0000); t2 = (u2& 0xf0000); u |= t1>t2?t1:t2;
278: t1 = (u1& 0xf000); t2 = (u2& 0xf000); u |= t1>t2?t1:t2;
279: t1 = (u1& 0xf00); t2 = (u2& 0xf00); u |= t1>t2?t1:t2;
280: t1 = (u1& 0xf0); t2 = (u2& 0xf0); u |= t1>t2?t1:t2;
281: t1 = (u1& 0xf); t2 = (u2& 0xf); u |= t1>t2?t1:t2;
1.1 noro 282: d[i] = u;
283: }
284: break;
285: case 6:
1.41 noro 286: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 287: u1 = d1[i]; u2 = d2[i];
288: t1 = (u1&0x3f000000); t2 = (u2&0x3f000000); u = t1>t2?t1:t2;
1.62 noro 289: t1 = (u1& 0xfc0000); t2 = (u2& 0xfc0000); u |= t1>t2?t1:t2;
290: t1 = (u1& 0x3f000); t2 = (u2& 0x3f000); u |= t1>t2?t1:t2;
291: t1 = (u1& 0xfc0); t2 = (u2& 0xfc0); u |= t1>t2?t1:t2;
292: t1 = (u1& 0x3f); t2 = (u2& 0x3f); u |= t1>t2?t1:t2;
1.1 noro 293: d[i] = u;
294: }
295: break;
296: case 8:
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&0xff000000); t2 = (u2&0xff000000); u = t1>t2?t1:t2;
1.62 noro 300: t1 = (u1& 0xff0000); t2 = (u2& 0xff0000); u |= t1>t2?t1:t2;
301: t1 = (u1& 0xff00); t2 = (u2& 0xff00); u |= t1>t2?t1:t2;
302: t1 = (u1& 0xff); t2 = (u2& 0xff); u |= t1>t2?t1:t2;
1.1 noro 303: d[i] = u;
304: }
305: break;
306: case 16:
1.41 noro 307: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 308: u1 = d1[i]; u2 = d2[i];
309: t1 = (u1&0xffff0000); t2 = (u2&0xffff0000); u = t1>t2?t1:t2;
1.62 noro 310: t1 = (u1& 0xffff); t2 = (u2& 0xffff); u |= t1>t2?t1:t2;
1.1 noro 311: d[i] = u;
312: }
313: break;
314: case 32:
1.41 noro 315: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 316: u1 = d1[i]; u2 = d2[i];
317: d[i] = u1>u2?u1:u2;
318: }
319: break;
320: default:
1.41 noro 321: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 322: u1 = d1[i]; u2 = d2[i];
323: for ( j = 0, u = 0; j < nd_epw; j++ ) {
324: t1 = (u1&nd_mask[j]); t2 = (u2&nd_mask[j]); u |= t1>t2?t1:t2;
325: }
326: d[i] = u;
327: }
328: break;
329: }
1.65 noro 330: #else
331: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
332: u1 = d1[i]; u2 = d2[i];
333: for ( j = 0, u = 0; j < nd_epw; j++ ) {
334: t1 = (u1&nd_mask[j]); t2 = (u2&nd_mask[j]); u |= t1>t2?t1:t2;
335: }
336: d[i] = u;
337: }
338: #endif
1.39 noro 339: TD(d) = ndl_weight(d);
1.61 noro 340: if ( nd_blockmask ) ndl_weight_mask(d);
1.57 noro 341: }
342:
1.61 noro 343: int ndl_weight(UINT *d)
1.1 noro 344: {
1.61 noro 345: UINT t,u;
1.1 noro 346: int i,j;
347:
1.60 noro 348: if ( current_dl_weight_vector )
349: for ( i = 0, t = 0; i < nd_nvar; i++ ) {
350: u = GET_EXP(d,i);
351: t += MUL_WEIGHT(u,i);
352: }
353: else
354: for ( t = 0, i = nd_exporigin; i < nd_wpd; i++ ) {
355: u = d[i];
356: for ( j = 0; j < nd_epw; j++, u>>=nd_bpe )
357: t += (u&nd_mask0);
358: }
1.1 noro 359: return t;
360: }
361:
1.61 noro 362: void ndl_weight_mask(UINT *d)
1.43 noro 363: {
1.61 noro 364: UINT t,u;
365: UINT *mask;
366: int i,j,k,l;
1.43 noro 367:
1.61 noro 368: l = nd_blockmask->n;
369: for ( k = 0; k < l; k++ ) {
370: mask = nd_blockmask->mask[k];
371: if ( current_dl_weight_vector )
372: for ( i = 0, t = 0; i < nd_nvar; i++ ) {
373: u = GET_EXP_MASK(d,i,mask);
374: t += MUL_WEIGHT(u,i);
375: }
376: else
377: for ( t = 0, i = nd_exporigin; i < nd_wpd; i++ ) {
378: u = d[i]&mask[i];
379: for ( j = 0; j < nd_epw; j++, u>>=nd_bpe )
380: t += (u&nd_mask0);
381: }
382: d[k+1] = t;
383: }
1.43 noro 384: }
385:
1.61 noro 386: int ndl_lex_compare(UINT *d1,UINT *d2)
1.1 noro 387: {
388: int i;
389:
1.41 noro 390: d1 += nd_exporigin;
391: d2 += nd_exporigin;
392: for ( i = nd_exporigin; i < nd_wpd; i++, d1++, d2++ )
1.1 noro 393: if ( *d1 > *d2 )
1.32 noro 394: return nd_isrlex ? -1 : 1;
1.1 noro 395: else if ( *d1 < *d2 )
1.32 noro 396: return nd_isrlex ? 1 : -1;
1.1 noro 397: return 0;
398: }
399:
1.61 noro 400: int ndl_block_compare(UINT *d1,UINT *d2)
1.43 noro 401: {
402: int i,l,j,ord_o,ord_l;
403: struct order_pair *op;
1.61 noro 404: UINT t1,t2,m;
405: UINT *mask;
1.43 noro 406:
407: l = nd_blockmask->n;
408: op = nd_blockmask->order_pair;
409: for ( j = 0; j < l; j++ ) {
410: mask = nd_blockmask->mask[j];
411: ord_o = op[j].order;
412: if ( ord_o < 2 )
1.44 noro 413: if ( (t1=d1[j+1]) > (t2=d2[j+1]) ) return 1;
414: else if ( t1 < t2 ) return -1;
1.43 noro 415: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.44 noro 416: m = mask[i];
417: t1 = d1[i]&m;
418: t2 = d2[i]&m;
1.43 noro 419: if ( t1 > t2 )
420: return !ord_o ? -1 : 1;
421: else if ( t1 < t2 )
422: return !ord_o ? 1 : -1;
423: }
424: }
425: return 0;
426: }
427:
1.96 noro 428: int ndl_matrix_compare(UINT *d1,UINT *d2)
429: {
430: int i,j,s;
431: int *v;
432:
433: for ( j = 0; j < nd_nvar; j++ )
434: nd_work_vector[j] = GET_EXP(d1,j)-GET_EXP(d2,j);
435: for ( i = 0; i < nd_matrix_len; i++ ) {
436: v = nd_matrix[i];
437: for ( j = 0, s = 0; j < nd_nvar; j++ )
438: s += v[j]*nd_work_vector[j];
439: if ( s > 0 ) return 1;
440: else if ( s < 0 ) return -1;
441: }
442: return 0;
443: }
444:
1.97 noro 445: int ndl_composite_compare(UINT *d1,UINT *d2)
446: {
447: int i,j,s,start,end,len,o;
448: int *v;
449: struct sparse_weight *sw;
450:
451: for ( j = 0; j < nd_nvar; j++ )
452: nd_work_vector[j] = GET_EXP(d1,j)-GET_EXP(d2,j);
453: for ( i = 0; i < nd_worb_len; i++ ) {
454: len = nd_worb[i].length;
455: switch ( nd_worb[i].type ) {
456: case IS_DENSE_WEIGHT:
457: v = nd_worb[i].body.dense_weight;
458: for ( j = 0, s = 0; j < len; j++ )
459: s += v[j]*nd_work_vector[j];
460: if ( s > 0 ) return 1;
461: else if ( s < 0 ) return -1;
462: break;
463: case IS_SPARSE_WEIGHT:
464: sw = nd_worb[i].body.sparse_weight;
465: for ( j = 0, s = 0; j < len; j++ )
466: s += sw[j].value*nd_work_vector[sw[j].pos];
467: if ( s > 0 ) return 1;
468: else if ( s < 0 ) return -1;
469: break;
470: case IS_BLOCK:
471: o = nd_worb[i].body.block.order;
472: start = nd_worb[i].body.block.start;
473: switch ( o ) {
474: case 0:
475: end = start+len;
476: for ( j = start, s = 0; j < end; j++ )
477: s += MUL_WEIGHT(nd_work_vector[j],j);
478: if ( s > 0 ) return 1;
479: else if ( s < 0 ) return -1;
480: for ( j = end-1; j >= start; j-- )
481: if ( nd_work_vector[j] < 0 ) return 1;
482: else if ( nd_work_vector[j] > 0 ) return -1;
483: break;
484: case 1:
485: end = start+len;
486: for ( j = start, s = 0; j < end; j++ )
487: s += MUL_WEIGHT(nd_work_vector[j],j);
488: if ( s > 0 ) return 1;
489: else if ( s < 0 ) return -1;
490: for ( j = start; j < end; j++ )
491: if ( nd_work_vector[j] > 0 ) return 1;
492: else if ( nd_work_vector[j] < 0 ) return -1;
493: break;
494: case 2:
495: for ( j = start; j < end; j++ )
496: if ( nd_work_vector[j] > 0 ) return 1;
497: else if ( nd_work_vector[j] < 0 ) return -1;
498: break;
499: }
500: break;
501: }
502: }
503: return 0;
504: }
505:
1.58 noro 506: /* TDH -> WW -> TD-> RL */
507:
1.61 noro 508: int ndl_ww_lex_compare(UINT *d1,UINT *d2)
1.58 noro 509: {
510: int i,m,e1,e2;
511:
512: if ( TD(d1) > TD(d2) ) return 1;
513: else if ( TD(d1) < TD(d2) ) return -1;
514: m = nd_nvar>>1;
515: for ( i = 0, e1 = e2 = 0; i < m; i++ ) {
516: e1 += current_weyl_weight_vector[i]*(GET_EXP(d1,m+i)-GET_EXP(d1,i));
517: e2 += current_weyl_weight_vector[i]*(GET_EXP(d2,m+i)-GET_EXP(d2,i));
518: }
519: if ( e1 > e2 ) return 1;
520: else if ( e1 < e2 ) return -1;
521: return ndl_lex_compare(d1,d2);
522: }
523:
1.61 noro 524: INLINE int ndl_equal(UINT *d1,UINT *d2)
1.1 noro 525: {
526: int i;
527:
1.81 noro 528: switch ( nd_wpd ) {
529: case 2:
530: if ( TD(d2) != TD(d1) ) return 0;
531: if ( d2[1] != d1[1] ) return 0;
532: return 1;
533: break;
534: case 3:
535: if ( TD(d2) != TD(d1) ) return 0;
536: if ( d2[1] != d1[1] ) return 0;
537: if ( d2[2] != d1[2] ) return 0;
538: return 1;
539: break;
540: default:
541: for ( i = 0; i < nd_wpd; i++ )
542: if ( *d1++ != *d2++ ) return 0;
543: return 1;
544: break;
545: }
1.1 noro 546: }
547:
1.61 noro 548: INLINE void ndl_copy(UINT *d1,UINT *d2)
1.6 noro 549: {
550: int i;
551:
552: switch ( nd_wpd ) {
1.41 noro 553: case 2:
1.34 noro 554: TD(d2) = TD(d1);
555: d2[1] = d1[1];
1.6 noro 556: break;
1.41 noro 557: case 3:
1.34 noro 558: TD(d2) = TD(d1);
1.6 noro 559: d2[1] = d1[1];
1.34 noro 560: d2[2] = d1[2];
1.6 noro 561: break;
562: default:
1.41 noro 563: for ( i = 0; i < nd_wpd; i++ )
1.6 noro 564: d2[i] = d1[i];
565: break;
566: }
567: }
568:
1.61 noro 569: INLINE void ndl_zero(UINT *d)
570: {
571: int i;
572: for ( i = 0; i < nd_wpd; i++ ) d[i] = 0;
573: }
574:
575: INLINE void ndl_add(UINT *d1,UINT *d2,UINT *d)
1.1 noro 576: {
577: int i;
578:
1.43 noro 579: #if 1
1.6 noro 580: switch ( nd_wpd ) {
1.41 noro 581: case 2:
582: TD(d) = TD(d1)+TD(d2);
1.34 noro 583: d[1] = d1[1]+d2[1];
1.6 noro 584: break;
1.41 noro 585: case 3:
586: TD(d) = TD(d1)+TD(d2);
1.6 noro 587: d[1] = d1[1]+d2[1];
1.34 noro 588: d[2] = d1[2]+d2[2];
1.6 noro 589: break;
590: default:
1.43 noro 591: for ( i = 0; i < nd_wpd; i++ ) d[i] = d1[i]+d2[i];
1.6 noro 592: break;
593: }
1.43 noro 594: #else
595: for ( i = 0; i < nd_wpd; i++ ) d[i] = d1[i]+d2[i];
596: #endif
1.6 noro 597: }
598:
1.55 noro 599: /* d1 += d2 */
1.61 noro 600: INLINE void ndl_addto(UINT *d1,UINT *d2)
1.55 noro 601: {
602: int i;
603:
604: #if 1
605: switch ( nd_wpd ) {
606: case 2:
607: TD(d1) += TD(d2);
608: d1[1] += d2[1];
609: break;
610: case 3:
611: TD(d1) += TD(d2);
612: d1[1] += d2[1];
613: d1[2] += d2[2];
614: break;
615: default:
616: for ( i = 0; i < nd_wpd; i++ ) d1[i] += d2[i];
617: break;
618: }
619: #else
620: for ( i = 0; i < nd_wpd; i++ ) d1[i] += d2[i];
621: #endif
622: }
623:
1.61 noro 624: INLINE void ndl_sub(UINT *d1,UINT *d2,UINT *d)
1.6 noro 625: {
626: int i;
627:
1.43 noro 628: for ( i = 0; i < nd_wpd; i++ ) d[i] = d1[i]-d2[i];
1.1 noro 629: }
630:
1.61 noro 631: int ndl_disjoint(UINT *d1,UINT *d2)
1.1 noro 632: {
1.61 noro 633: UINT t1,t2,u,u1,u2;
1.1 noro 634: int i,j;
635:
1.65 noro 636: #if USE_UNROLL
1.1 noro 637: switch ( nd_bpe ) {
1.62 noro 638: case 3:
639: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
640: u1 = d1[i]; u2 = d2[i];
641: t1 = u1&0x38000000; t2 = u2&0x38000000; if ( t1&&t2 ) return 0;
642: t1 = u1& 0x7000000; t2 = u2& 0x7000000; if ( t1&&t2 ) return 0;
643: t1 = u1& 0xe00000; t2 = u2& 0xe00000; if ( t1&&t2 ) return 0;
644: t1 = u1& 0x1c0000; t2 = u2& 0x1c0000; if ( t1&&t2 ) return 0;
645: t1 = u1& 0x38000; t2 = u2& 0x38000; if ( t1&&t2 ) return 0;
646: t1 = u1& 0x7000; t2 = u2& 0x7000; if ( t1&&t2 ) return 0;
647: t1 = u1& 0xe00; t2 = u2& 0xe00; if ( t1&&t2 ) return 0;
648: t1 = u1& 0x1c0; t2 = u2& 0x1c0; if ( t1&&t2 ) return 0;
649: t1 = u1& 0x38; t2 = u2& 0x38; if ( t1&&t2 ) return 0;
650: t1 = u1& 0x7; t2 = u2& 0x7; if ( t1&&t2 ) return 0;
651: }
652: return 1;
653: break;
1.1 noro 654: case 4:
1.41 noro 655: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 656: u1 = d1[i]; u2 = d2[i];
657: t1 = u1&0xf0000000; t2 = u2&0xf0000000; if ( t1&&t2 ) return 0;
1.62 noro 658: t1 = u1& 0xf000000; t2 = u2& 0xf000000; if ( t1&&t2 ) return 0;
659: t1 = u1& 0xf00000; t2 = u2& 0xf00000; if ( t1&&t2 ) return 0;
660: t1 = u1& 0xf0000; t2 = u2& 0xf0000; if ( t1&&t2 ) return 0;
661: t1 = u1& 0xf000; t2 = u2& 0xf000; if ( t1&&t2 ) return 0;
662: t1 = u1& 0xf00; t2 = u2& 0xf00; if ( t1&&t2 ) return 0;
663: t1 = u1& 0xf0; t2 = u2& 0xf0; if ( t1&&t2 ) return 0;
664: t1 = u1& 0xf; t2 = u2& 0xf; if ( t1&&t2 ) return 0;
1.1 noro 665: }
666: return 1;
667: break;
668: case 6:
1.41 noro 669: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 670: u1 = d1[i]; u2 = d2[i];
671: t1 = u1&0x3f000000; t2 = u2&0x3f000000; if ( t1&&t2 ) return 0;
1.62 noro 672: t1 = u1& 0xfc0000; t2 = u2& 0xfc0000; if ( t1&&t2 ) return 0;
673: t1 = u1& 0x3f000; t2 = u2& 0x3f000; if ( t1&&t2 ) return 0;
674: t1 = u1& 0xfc0; t2 = u2& 0xfc0; if ( t1&&t2 ) return 0;
675: t1 = u1& 0x3f; t2 = u2& 0x3f; if ( t1&&t2 ) return 0;
1.1 noro 676: }
677: return 1;
678: break;
679: case 8:
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&0xff000000; t2 = u2&0xff000000; if ( t1&&t2 ) return 0;
1.62 noro 683: t1 = u1& 0xff0000; t2 = u2& 0xff0000; if ( t1&&t2 ) return 0;
684: t1 = u1& 0xff00; t2 = u2& 0xff00; if ( t1&&t2 ) return 0;
685: t1 = u1& 0xff; t2 = u2& 0xff; if ( t1&&t2 ) return 0;
1.1 noro 686: }
687: return 1;
688: break;
689: case 16:
1.41 noro 690: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 691: u1 = d1[i]; u2 = d2[i];
692: t1 = u1&0xffff0000; t2 = u2&0xffff0000; if ( t1&&t2 ) return 0;
1.62 noro 693: t1 = u1& 0xffff; t2 = u2& 0xffff; if ( t1&&t2 ) return 0;
1.1 noro 694: }
695: return 1;
696: break;
697: case 32:
1.41 noro 698: for ( i = nd_exporigin; i < nd_wpd; i++ )
1.1 noro 699: if ( d1[i] && d2[i] ) return 0;
700: return 1;
701: break;
702: default:
1.41 noro 703: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 704: u1 = d1[i]; u2 = d2[i];
705: for ( j = 0; j < nd_epw; j++ ) {
706: if ( (u1&nd_mask0) && (u2&nd_mask0) ) return 0;
707: u1 >>= nd_bpe; u2 >>= nd_bpe;
708: }
709: }
710: return 1;
711: break;
712: }
1.65 noro 713: #else
714: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
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: #endif
1.1 noro 723: }
724:
1.61 noro 725: int ndl_check_bound2(int index,UINT *d2)
1.1 noro 726: {
1.61 noro 727: UINT u2;
728: UINT *d1;
1.5 noro 729: int i,j,ind,k;
1.1 noro 730:
1.5 noro 731: d1 = nd_bound[index];
732: ind = 0;
1.65 noro 733: #if USE_UNROLL
1.5 noro 734: switch ( nd_bpe ) {
1.62 noro 735: case 3:
736: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
737: u2 = d2[i];
738: if ( d1[ind++]+((u2>>27)&0x7) >= 0x8 ) return 1;
739: if ( d1[ind++]+((u2>>24)&0x7) >= 0x8 ) return 1;
740: if ( d1[ind++]+((u2>>21)&0x7) >= 0x8 ) return 1;
741: if ( d1[ind++]+((u2>>18)&0x7) >= 0x8 ) return 1;
742: if ( d1[ind++]+((u2>>15)&0x7) >= 0x8 ) return 1;
743: if ( d1[ind++]+((u2>>12)&0x7) >= 0x8 ) return 1;
744: if ( d1[ind++]+((u2>>9)&0x7) >= 0x8 ) return 1;
745: if ( d1[ind++]+((u2>>6)&0x7) >= 0x8 ) return 1;
746: if ( d1[ind++]+((u2>>3)&0x7) >= 0x8 ) return 1;
747: if ( d1[ind++]+(u2&0x7) >= 0x8 ) return 1;
748: }
749: return 0;
750: break;
1.5 noro 751: case 4:
1.41 noro 752: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 753: u2 = d2[i];
754: if ( d1[ind++]+((u2>>28)&0xf) >= 0x10 ) return 1;
755: if ( d1[ind++]+((u2>>24)&0xf) >= 0x10 ) return 1;
756: if ( d1[ind++]+((u2>>20)&0xf) >= 0x10 ) return 1;
757: if ( d1[ind++]+((u2>>16)&0xf) >= 0x10 ) return 1;
758: if ( d1[ind++]+((u2>>12)&0xf) >= 0x10 ) return 1;
759: if ( d1[ind++]+((u2>>8)&0xf) >= 0x10 ) return 1;
760: if ( d1[ind++]+((u2>>4)&0xf) >= 0x10 ) return 1;
761: if ( d1[ind++]+(u2&0xf) >= 0x10 ) return 1;
762: }
763: return 0;
764: break;
765: case 6:
1.41 noro 766: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 767: u2 = d2[i];
768: if ( d1[ind++]+((u2>>24)&0x3f) >= 0x40 ) return 1;
769: if ( d1[ind++]+((u2>>18)&0x3f) >= 0x40 ) return 1;
770: if ( d1[ind++]+((u2>>12)&0x3f) >= 0x40 ) return 1;
771: if ( d1[ind++]+((u2>>6)&0x3f) >= 0x40 ) return 1;
772: if ( d1[ind++]+(u2&0x3f) >= 0x40 ) return 1;
773: }
774: return 0;
775: break;
776: case 8:
1.41 noro 777: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 778: u2 = d2[i];
779: if ( d1[ind++]+((u2>>24)&0xff) >= 0x100 ) return 1;
780: if ( d1[ind++]+((u2>>16)&0xff) >= 0x100 ) return 1;
781: if ( d1[ind++]+((u2>>8)&0xff) >= 0x100 ) return 1;
782: if ( d1[ind++]+(u2&0xff) >= 0x100 ) return 1;
783: }
784: return 0;
785: break;
786: case 16:
1.41 noro 787: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 788: u2 = d2[i];
789: if ( d1[ind++]+((u2>>16)&0xffff) > 0x10000 ) return 1;
790: if ( d1[ind++]+(u2&0xffff) > 0x10000 ) return 1;
791: }
792: return 0;
793: break;
794: case 32:
1.41 noro 795: for ( i = nd_exporigin; i < nd_wpd; i++ )
1.5 noro 796: if ( d1[i]+d2[i]<d1[i] ) return 1;
797: return 0;
798: break;
799: default:
1.41 noro 800: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 801: u2 = d2[i];
802: k = (nd_epw-1)*nd_bpe;
803: for ( j = 0; j < nd_epw; j++, k -= nd_bpe )
804: if ( d1[ind++]+((u2>>k)&nd_mask0) > nd_mask0 ) return 1;
805: }
806: return 0;
807: break;
808: }
1.65 noro 809: #else
810: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
811: u2 = d2[i];
812: k = (nd_epw-1)*nd_bpe;
813: for ( j = 0; j < nd_epw; j++, k -= nd_bpe )
814: if ( d1[ind++]+((u2>>k)&nd_mask0) > nd_mask0 ) return 1;
815: }
816: return 0;
817: #endif
1.1 noro 818: }
819:
1.61 noro 820: INLINE int ndl_hash_value(UINT *d)
1.1 noro 821: {
822: int i;
823: int r;
824:
1.34 noro 825: r = 0;
1.41 noro 826: for ( i = 0; i < nd_wpd; i++ )
1.1 noro 827: r = ((r<<16)+d[i])%REDTAB_LEN;
828: return r;
829: }
830:
1.63 noro 831: INLINE int ndl_find_reducer(UINT *dg)
1.1 noro 832: {
1.13 noro 833: RHist r;
1.6 noro 834: int d,k,i;
1.1 noro 835:
1.63 noro 836: d = ndl_hash_value(dg);
1.13 noro 837: for ( r = nd_red[d], k = 0; r; r = NEXT(r), k++ ) {
1.34 noro 838: if ( ndl_equal(dg,DL(r)) ) {
1.1 noro 839: if ( k > 0 ) nd_notfirst++;
840: nd_found++;
1.13 noro 841: return r->index;
1.1 noro 842: }
843: }
1.13 noro 844: if ( Reverse )
845: for ( i = nd_psn-1; i >= 0; i-- ) {
846: r = nd_psh[i];
1.34 noro 847: if ( ndl_reducible(dg,DL(r)) ) {
1.13 noro 848: nd_create++;
1.34 noro 849: nd_append_red(dg,i);
1.13 noro 850: return i;
851: }
852: }
853: else
854: for ( i = 0; i < nd_psn; i++ ) {
855: r = nd_psh[i];
1.34 noro 856: if ( ndl_reducible(dg,DL(r)) ) {
1.13 noro 857: nd_create++;
1.34 noro 858: nd_append_red(dg,i);
1.13 noro 859: return i;
860: }
1.1 noro 861: }
1.6 noro 862: return -1;
1.1 noro 863: }
864:
1.63 noro 865: ND nd_merge(ND p1,ND p2)
866: {
867: int n,c;
868: int t,can,td1,td2;
869: ND r;
870: NM m1,m2,mr0,mr,s;
871:
872: if ( !p1 ) return p2;
873: else if ( !p2 ) return p1;
874: else {
875: can = 0;
876: for ( n = NV(p1), m1 = BDY(p1), m2 = BDY(p2), mr0 = 0; m1 && m2; ) {
877: c = DL_COMPARE(DL(m1),DL(m2));
878: switch ( c ) {
879: case 0:
880: s = m1; m1 = NEXT(m1);
881: can++; NEXTNM2(mr0,mr,s);
882: s = m2; m2 = NEXT(m2); FREENM(s);
883: break;
884: case 1:
885: s = m1; m1 = NEXT(m1); NEXTNM2(mr0,mr,s);
886: break;
887: case -1:
888: s = m2; m2 = NEXT(m2); NEXTNM2(mr0,mr,s);
889: break;
890: }
891: }
892: if ( !mr0 )
893: if ( m1 ) mr0 = m1;
894: else if ( m2 ) mr0 = m2;
895: else return 0;
896: else if ( m1 ) NEXT(mr) = m1;
897: else if ( m2 ) NEXT(mr) = m2;
898: else NEXT(mr) = 0;
899: BDY(p1) = mr0;
900: SG(p1) = MAX(SG(p1),SG(p2));
901: LEN(p1) = LEN(p1)+LEN(p2)-can;
902: FREEND(p2);
903: return p1;
904: }
905: }
906:
1.31 noro 907: ND nd_add(int mod,ND p1,ND p2)
1.1 noro 908: {
909: int n,c;
1.34 noro 910: int t,can,td1,td2;
1.1 noro 911: ND r;
912: NM m1,m2,mr0,mr,s;
913:
1.34 noro 914: if ( !p1 ) return p2;
915: else if ( !p2 ) return p1;
1.71 noro 916: else if ( mod == -1 ) return nd_add_sf(p1,p2);
1.34 noro 917: else if ( !mod ) return nd_add_q(p1,p2);
1.1 noro 918: else {
1.30 noro 919: can = 0;
1.1 noro 920: for ( n = NV(p1), m1 = BDY(p1), m2 = BDY(p2), mr0 = 0; m1 && m2; ) {
1.34 noro 921: c = DL_COMPARE(DL(m1),DL(m2));
1.1 noro 922: switch ( c ) {
923: case 0:
1.19 noro 924: t = ((CM(m1))+(CM(m2))) - mod;
1.34 noro 925: if ( t < 0 ) t += mod;
1.1 noro 926: s = m1; m1 = NEXT(m1);
927: if ( t ) {
1.34 noro 928: can++; NEXTNM2(mr0,mr,s); CM(mr) = (t);
1.17 noro 929: } else {
1.34 noro 930: can += 2; FREENM(s);
1.17 noro 931: }
932: s = m2; m2 = NEXT(m2); FREENM(s);
1.95 noro 933: break;
934: case 1:
935: s = m1; m1 = NEXT(m1); NEXTNM2(mr0,mr,s);
936: break;
937: case -1:
938: s = m2; m2 = NEXT(m2); NEXTNM2(mr0,mr,s);
939: break;
940: }
941: }
942: if ( !mr0 )
943: if ( m1 ) mr0 = m1;
944: else if ( m2 ) mr0 = m2;
945: else return 0;
946: else if ( m1 ) NEXT(mr) = m1;
947: else if ( m2 ) NEXT(mr) = m2;
948: else NEXT(mr) = 0;
949: BDY(p1) = mr0;
950: SG(p1) = MAX(SG(p1),SG(p2));
951: LEN(p1) = LEN(p1)+LEN(p2)-can;
952: FREEND(p2);
953: return p1;
954: }
955: }
956:
957: /* XXX on opteron, the inlined manipulation of destructive additon of
958: * two NM seems to make gcc optimizer get confused, so the part is
959: * done in a function.
960: */
961:
962: int nm_destructive_add_q(NM *m1,NM *m2,NM *mr0,NM *mr)
963: {
964: NM s;
965: Q t;
966: int can;
967:
968: addq(CQ(*m1),CQ(*m2),&t);
969: s = *m1; *m1 = NEXT(*m1);
970: if ( t ) {
971: can = 1; NEXTNM2(*mr0,*mr,s); CQ(*mr) = (t);
972: } else {
973: can = 2; FREENM(s);
974: }
975: s = *m2; *m2 = NEXT(*m2); FREENM(s);
976: return can;
977: }
978:
979: ND nd_add_q(ND p1,ND p2)
980: {
981: int n,c,can;
982: ND r;
983: NM m1,m2,mr0,mr,s;
984: Q t;
985:
986: if ( !p1 ) return p2;
987: else if ( !p2 ) return p1;
988: else {
989: can = 0;
990: for ( n = NV(p1), m1 = BDY(p1), m2 = BDY(p2), mr0 = 0; m1 && m2; ) {
991: c = DL_COMPARE(DL(m1),DL(m2));
992: switch ( c ) {
993: case 0:
994: #if defined(__x86_64__)
995: can += nm_destructive_add_q(&m1,&m2,&mr0,&mr);
996: #else
997: addq(CQ(m1),CQ(m2),&t);
998: s = m1; m1 = NEXT(m1);
999: if ( t ) {
1000: can++; NEXTNM2(mr0,mr,s); CQ(mr) = (t);
1001: } else {
1002: can += 2; FREENM(s);
1003: }
1004: s = m2; m2 = NEXT(m2); FREENM(s);
1005: #endif
1.17 noro 1006: break;
1007: case 1:
1008: s = m1; m1 = NEXT(m1); NEXTNM2(mr0,mr,s);
1009: break;
1010: case -1:
1011: s = m2; m2 = NEXT(m2); NEXTNM2(mr0,mr,s);
1012: break;
1013: }
1014: }
1015: if ( !mr0 )
1.34 noro 1016: if ( m1 ) mr0 = m1;
1017: else if ( m2 ) mr0 = m2;
1018: else return 0;
1019: else if ( m1 ) NEXT(mr) = m1;
1020: else if ( m2 ) NEXT(mr) = m2;
1021: else NEXT(mr) = 0;
1.17 noro 1022: BDY(p1) = mr0;
1023: SG(p1) = MAX(SG(p1),SG(p2));
1.31 noro 1024: LEN(p1) = LEN(p1)+LEN(p2)-can;
1.17 noro 1025: FREEND(p2);
1026: return p1;
1027: }
1028: }
1029:
1.71 noro 1030: ND nd_add_sf(ND p1,ND p2)
1031: {
1032: int n,c,can;
1033: ND r;
1034: NM m1,m2,mr0,mr,s;
1035: int t;
1036:
1037: if ( !p1 ) return p2;
1038: else if ( !p2 ) return p1;
1039: else {
1040: can = 0;
1041: for ( n = NV(p1), m1 = BDY(p1), m2 = BDY(p2), mr0 = 0; m1 && m2; ) {
1042: c = DL_COMPARE(DL(m1),DL(m2));
1043: switch ( c ) {
1044: case 0:
1045: t = _addsf(CM(m1),CM(m2));
1046: s = m1; m1 = NEXT(m1);
1047: if ( t ) {
1048: can++; NEXTNM2(mr0,mr,s); CM(mr) = (t);
1049: } else {
1050: can += 2; FREENM(s);
1051: }
1052: s = m2; m2 = NEXT(m2); FREENM(s);
1053: break;
1054: case 1:
1055: s = m1; m1 = NEXT(m1); NEXTNM2(mr0,mr,s);
1056: break;
1057: case -1:
1058: s = m2; m2 = NEXT(m2); NEXTNM2(mr0,mr,s);
1059: break;
1060: }
1061: }
1062: if ( !mr0 )
1063: if ( m1 ) mr0 = m1;
1064: else if ( m2 ) mr0 = m2;
1065: else return 0;
1066: else if ( m1 ) NEXT(mr) = m1;
1067: else if ( m2 ) NEXT(mr) = m2;
1068: else NEXT(mr) = 0;
1069: BDY(p1) = mr0;
1070: SG(p1) = MAX(SG(p1),SG(p2));
1071: LEN(p1) = LEN(p1)+LEN(p2)-can;
1072: FREEND(p2);
1073: return p1;
1074: }
1075: }
1076:
1.1 noro 1077: /* ret=1 : success, ret=0 : overflow */
1.69 noro 1078: int nd_nf(int mod,ND g,NDV *ps,int full,NDC dn,ND *rp)
1.1 noro 1079: {
1.11 noro 1080: ND d;
1.1 noro 1081: NM m,mrd,tail;
1.7 noro 1082: NM mul;
1.10 noro 1083: int n,sugar,psugar,sugar0,stat,index;
1.30 noro 1084: int c,c1,c2,dummy;
1.17 noro 1085: RHist h;
1.11 noro 1086: NDV p,red;
1.69 noro 1087: Q cg,cred,gcd,tq,qq;
1.21 noro 1088: double hmag;
1.1 noro 1089:
1.69 noro 1090: if ( dn ) {
1091: if ( mod )
1092: dn->m = 1;
1093: else
1094: dn->z = ONE;
1095: }
1.1 noro 1096: if ( !g ) {
1097: *rp = 0;
1098: return 1;
1099: }
1.34 noro 1100: if ( !mod ) hmag = ((double)p_mag((P)HCQ(g)))*nd_scale;
1.21 noro 1101:
1.14 noro 1102: sugar0 = sugar = SG(g);
1.1 noro 1103: n = NV(g);
1.61 noro 1104: mul = (NM)ALLOCA(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT));
1.1 noro 1105: for ( d = 0; g; ) {
1.63 noro 1106: index = ndl_find_reducer(HDL(g));
1.6 noro 1107: if ( index >= 0 ) {
1.17 noro 1108: h = nd_psh[index];
1109: ndl_sub(HDL(g),DL(h),DL(mul));
1.14 noro 1110: if ( ndl_check_bound2(index,DL(mul)) ) {
1.6 noro 1111: nd_free(g); nd_free(d);
1112: return 0;
1113: }
1.77 noro 1114: if ( nd_demand )
1115: p = ndv_load(index);
1116: else
1117: p = ps[index];
1.71 noro 1118: if ( mod == -1 )
1119: CM(mul) = _mulsf(_invsf(HCM(p)),_chsgnsf(HCM(g)));
1120: else if ( mod ) {
1.19 noro 1121: c1 = invm(HCM(p),mod); c2 = mod-HCM(g);
1122: DMAR(c1,c2,0,mod,c); CM(mul) = c;
1.16 noro 1123: } else {
1.17 noro 1124: igcd_cofactor(HCQ(g),HCQ(p),&gcd,&cg,&cred);
1.16 noro 1125: chsgnq(cg,&CQ(mul));
1.20 noro 1126: nd_mul_c_q(d,cred); nd_mul_c_q(g,cred);
1.69 noro 1127: if ( dn ) {
1128: mulq(dn->z,cred,&tq); dn->z = tq;
1129: }
1.16 noro 1130: }
1.55 noro 1131: g = nd_add(mod,g,ndv_mul_nm(mod,mul,p));
1.34 noro 1132: sugar = MAX(sugar,SG(p)+TD(DL(mul)));
1.22 noro 1133: if ( !mod && hmag && g && ((double)(p_mag((P)HCQ(g))) > hmag) ) {
1.69 noro 1134: tq = HCQ(g);
1.21 noro 1135: nd_removecont2(d,g);
1.69 noro 1136: if ( dn ) {
1137: divq(tq,HCQ(g),&qq); divq(dn->z,qq,&tq); dn->z = tq;
1138: }
1.21 noro 1139: hmag = ((double)p_mag((P)HCQ(g)))*nd_scale;
1140: }
1.1 noro 1141: } else if ( !full ) {
1142: *rp = g;
1143: return 1;
1144: } else {
1145: m = BDY(g);
1146: if ( NEXT(m) ) {
1.34 noro 1147: BDY(g) = NEXT(m); NEXT(m) = 0; LEN(g)--;
1.1 noro 1148: } else {
1149: FREEND(g); g = 0;
1150: }
1151: if ( d ) {
1.34 noro 1152: NEXT(tail)=m; tail=m; LEN(d)++;
1.1 noro 1153: } else {
1.34 noro 1154: MKND(n,m,1,d); tail = BDY(d);
1.1 noro 1155: }
1156: }
1157: }
1.34 noro 1158: if ( d ) SG(d) = sugar;
1.1 noro 1159: *rp = d;
1160: return 1;
1161: }
1.28 noro 1162:
1.53 noro 1163: int nd_nf_pbucket(int mod,ND g,NDV *ps,int full,ND *rp)
1.25 noro 1164: {
1165: int hindex,index;
1166: NDV p;
1167: ND u,d,red;
1168: NODE l;
1.31 noro 1169: NM mul,m,mrd,tail;
1.25 noro 1170: int sugar,psugar,n,h_reducible;
1171: PGeoBucket bucket;
1172: int c,c1,c2;
1.26 noro 1173: Q cg,cred,gcd,zzz;
1.25 noro 1174: RHist h;
1.28 noro 1175: double hmag,gmag;
1.77 noro 1176: int count = 0;
1177: int hcount = 0;
1.25 noro 1178:
1179: if ( !g ) {
1180: *rp = 0;
1181: return 1;
1182: }
1183: sugar = SG(g);
1184: n = NV(g);
1.34 noro 1185: if ( !mod ) hmag = ((double)p_mag((P)HCQ(g)))*nd_scale;
1.25 noro 1186: bucket = create_pbucket();
1.31 noro 1187: add_pbucket(mod,bucket,g);
1.25 noro 1188: d = 0;
1.61 noro 1189: mul = (NM)ALLOCA(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT));
1.25 noro 1190: while ( 1 ) {
1.26 noro 1191: hindex = mod?head_pbucket(mod,bucket):head_pbucket_q(bucket);
1.25 noro 1192: if ( hindex < 0 ) {
1.77 noro 1193: if ( DP_Print > 3 ) printf("(%d %d)",count,hcount);
1.34 noro 1194: if ( d ) SG(d) = sugar;
1.25 noro 1195: *rp = d;
1196: return 1;
1197: }
1198: g = bucket->body[hindex];
1.63 noro 1199: index = ndl_find_reducer(HDL(g));
1.25 noro 1200: if ( index >= 0 ) {
1.77 noro 1201: count++;
1202: if ( !d ) hcount++;
1.25 noro 1203: h = nd_psh[index];
1204: ndl_sub(HDL(g),DL(h),DL(mul));
1205: if ( ndl_check_bound2(index,DL(mul)) ) {
1.26 noro 1206: nd_free(d);
1.25 noro 1207: free_pbucket(bucket);
1208: *rp = 0;
1209: return 0;
1210: }
1.53 noro 1211: p = ps[index];
1.71 noro 1212: if ( mod == -1 )
1213: CM(mul) = _mulsf(_invsf(HCM(p)),_chsgnsf(HCM(g)));
1214: else if ( mod ) {
1.25 noro 1215: c1 = invm(HCM(p),mod); c2 = mod-HCM(g);
1216: DMAR(c1,c2,0,mod,c); CM(mul) = c;
1217: } else {
1218: igcd_cofactor(HCQ(g),HCQ(p),&gcd,&cg,&cred);
1219: chsgnq(cg,&CQ(mul));
1.26 noro 1220: nd_mul_c_q(d,cred);
1221: mulq_pbucket(bucket,cred);
1222: g = bucket->body[hindex];
1.28 noro 1223: gmag = (double)p_mag((P)HCQ(g));
1.25 noro 1224: }
1.55 noro 1225: red = ndv_mul_nm(mod,mul,p);
1.25 noro 1226: bucket->body[hindex] = nd_remove_head(g);
1227: red = nd_remove_head(red);
1.31 noro 1228: add_pbucket(mod,bucket,red);
1.34 noro 1229: psugar = SG(p)+TD(DL(mul));
1230: sugar = MAX(sugar,psugar);
1.28 noro 1231: if ( !mod && hmag && (gmag > hmag) ) {
1232: g = normalize_pbucket(mod,bucket);
1233: if ( !g ) {
1.34 noro 1234: if ( d ) SG(d) = sugar;
1.28 noro 1235: *rp = d;
1236: return 1;
1237: }
1238: nd_removecont2(d,g);
1239: hmag = ((double)p_mag((P)HCQ(g)))*nd_scale;
1.31 noro 1240: add_pbucket(mod,bucket,g);
1.28 noro 1241: }
1.25 noro 1242: } else if ( !full ) {
1243: g = normalize_pbucket(mod,bucket);
1.34 noro 1244: if ( g ) SG(g) = sugar;
1.25 noro 1245: *rp = g;
1246: return 1;
1247: } else {
1248: m = BDY(g);
1249: if ( NEXT(m) ) {
1.34 noro 1250: BDY(g) = NEXT(m); NEXT(m) = 0; LEN(g)--;
1.25 noro 1251: } else {
1252: FREEND(g); g = 0;
1253: }
1254: bucket->body[hindex] = g;
1255: NEXT(m) = 0;
1256: if ( d ) {
1.34 noro 1257: NEXT(tail)=m; tail=m; LEN(d)++;
1.25 noro 1258: } else {
1.34 noro 1259: MKND(n,m,1,d); tail = BDY(d);
1.25 noro 1260: }
1261: }
1262: }
1263: }
1.27 noro 1264:
1.61 noro 1265: /* input : list of NDV, cand : list of NDV */
1.28 noro 1266:
1.61 noro 1267: int ndv_check_candidate(NODE input,int obpe,int oadv,EPOS oepos,NODE cand)
1.28 noro 1268: {
1269: int n,i,stat;
1270: ND nf,d;
1.61 noro 1271: NDV r;
1.45 noro 1272: NODE t,s;
1.69 noro 1273: union oNDC dn;
1.45 noro 1274:
1.61 noro 1275: ndv_setup(0,0,cand);
1.31 noro 1276: n = length(cand);
1.28 noro 1277:
1278: /* membercheck : list is a subset of Id(cand) ? */
1279: for ( t = input; t; t = NEXT(t) ) {
1.45 noro 1280: again:
1.61 noro 1281: if ( nd_bpe > obpe )
1282: r = ndv_dup_realloc((NDV)BDY(t),obpe,oadv,oepos);
1283: else
1284: r = (NDV)BDY(t);
1285: d = ndvtond(0,r);
1.69 noro 1286: stat = nd_nf(0,d,nd_ps,0,0,&nf);
1.45 noro 1287: if ( !stat ) {
1288: nd_reconstruct(0,0,0);
1289: goto again;
1290: } else if ( nf ) return 0;
1.71 noro 1291: if ( DP_Print ) { printf("."); fflush(stdout); }
1.28 noro 1292: }
1.71 noro 1293: if ( DP_Print ) { printf("\n"); }
1.28 noro 1294: /* gbcheck : cand is a GB of Id(cand) ? */
1.77 noro 1295: if ( !nd_gb(0,0,1) ) return 0;
1.28 noro 1296: /* XXX */
1.23 noro 1297: return 1;
1298: }
1.1 noro 1299:
1300: ND nd_remove_head(ND p)
1301: {
1302: NM m;
1303:
1304: m = BDY(p);
1305: if ( !NEXT(m) ) {
1.34 noro 1306: FREEND(p); p = 0;
1.31 noro 1307: } else {
1.34 noro 1308: BDY(p) = NEXT(m); LEN(p)--;
1.31 noro 1309: }
1.1 noro 1310: FREENM(m);
1311: return p;
1312: }
1313:
1.69 noro 1314: ND nd_separate_head(ND p,ND *head)
1315: {
1316: NM m,m0;
1317: ND r;
1318:
1319: m = BDY(p);
1320: if ( !NEXT(m) ) {
1321: *head = p; p = 0;
1322: } else {
1323: m0 = m;
1324: BDY(p) = NEXT(m); LEN(p)--;
1325: NEXT(m0) = 0;
1326: MKND(NV(p),m0,1,r);
1327: *head = r;
1328: }
1329: return p;
1330: }
1331:
1.1 noro 1332: PGeoBucket create_pbucket()
1333: {
1334: PGeoBucket g;
1335:
1336: g = CALLOC(1,sizeof(struct oPGeoBucket));
1337: g->m = -1;
1338: return g;
1339: }
1340:
1.25 noro 1341: void free_pbucket(PGeoBucket b) {
1342: int i;
1343:
1.26 noro 1344: for ( i = 0; i <= b->m; i++ )
1.25 noro 1345: if ( b->body[i] ) {
1346: nd_free(b->body[i]);
1347: b->body[i] = 0;
1348: }
1349: GC_free(b);
1350: }
1351:
1.63 noro 1352: void add_pbucket_symbolic(PGeoBucket g,ND d)
1353: {
1354: int l,i,k,m;
1355:
1356: if ( !d )
1357: return;
1358: l = LEN(d);
1359: for ( k = 0, m = 1; l > m; k++, m <<= 1 );
1360: /* 2^(k-1) < l <= 2^k (=m) */
1361: d = nd_merge(g->body[k],d);
1362: for ( ; d && LEN(d) > m; k++, m <<= 1 ) {
1363: g->body[k] = 0;
1364: d = nd_merge(g->body[k+1],d);
1365: }
1366: g->body[k] = d;
1367: g->m = MAX(g->m,k);
1368: }
1369:
1.31 noro 1370: void add_pbucket(int mod,PGeoBucket g,ND d)
1.1 noro 1371: {
1.31 noro 1372: int l,i,k,m;
1.1 noro 1373:
1.31 noro 1374: if ( !d )
1375: return;
1376: l = LEN(d);
1.29 noro 1377: for ( k = 0, m = 1; l > m; k++, m <<= 1 );
1378: /* 2^(k-1) < l <= 2^k (=m) */
1.31 noro 1379: d = nd_add(mod,g->body[k],d);
1380: for ( ; d && LEN(d) > m; k++, m <<= 1 ) {
1.1 noro 1381: g->body[k] = 0;
1.31 noro 1382: d = nd_add(mod,g->body[k+1],d);
1.1 noro 1383: }
1384: g->body[k] = d;
1385: g->m = MAX(g->m,k);
1386: }
1387:
1.26 noro 1388: void mulq_pbucket(PGeoBucket g,Q c)
1389: {
1390: int k;
1391:
1392: for ( k = 0; k <= g->m; k++ )
1393: nd_mul_c_q(g->body[k],c);
1394: }
1395:
1.63 noro 1396: NM remove_head_pbucket_symbolic(PGeoBucket g)
1397: {
1398: int j,i,k,c;
1399: NM head;
1400:
1401: k = g->m;
1402: j = -1;
1403: for ( i = 0; i <= k; i++ ) {
1404: if ( !g->body[i] ) continue;
1405: if ( j < 0 ) j = i;
1406: else {
1407: c = DL_COMPARE(HDL(g->body[i]),HDL(g->body[j]));
1408: if ( c > 0 )
1409: j = i;
1410: else if ( c == 0 )
1411: g->body[i] = nd_remove_head(g->body[i]);
1412: }
1413: }
1414: if ( j < 0 ) return 0;
1415: else {
1416: head = BDY(g->body[j]);
1417: if ( !NEXT(head) ) {
1418: FREEND(g->body[j]);
1419: g->body[j] = 0;
1420: } else {
1421: BDY(g->body[j]) = NEXT(head);
1422: LEN(g->body[j])--;
1423: }
1424: return head;
1425: }
1426: }
1427:
1.19 noro 1428: int head_pbucket(int mod,PGeoBucket g)
1.1 noro 1429: {
1430: int j,i,c,k,nv,sum;
1.61 noro 1431: UINT *di,*dj;
1.1 noro 1432: ND gi,gj;
1433:
1434: k = g->m;
1435: while ( 1 ) {
1436: j = -1;
1437: for ( i = 0; i <= k; i++ ) {
1438: if ( !(gi = g->body[i]) )
1439: continue;
1440: if ( j < 0 ) {
1441: j = i;
1442: gj = g->body[j];
1443: dj = HDL(gj);
1.14 noro 1444: sum = HCM(gj);
1.1 noro 1445: } else {
1.34 noro 1446: c = DL_COMPARE(HDL(gi),dj);
1.1 noro 1447: if ( c > 0 ) {
1.34 noro 1448: if ( sum ) HCM(gj) = sum;
1449: else g->body[j] = nd_remove_head(gj);
1.1 noro 1450: j = i;
1451: gj = g->body[j];
1452: dj = HDL(gj);
1.14 noro 1453: sum = HCM(gj);
1.1 noro 1454: } else if ( c == 0 ) {
1.71 noro 1455: if ( mod == -1 )
1456: sum = _addsf(sum,HCM(gi));
1457: else {
1458: sum = sum+HCM(gi)-mod;
1459: if ( sum < 0 ) sum += mod;
1460: }
1.1 noro 1461: g->body[i] = nd_remove_head(gi);
1462: }
1463: }
1464: }
1.34 noro 1465: if ( j < 0 ) return -1;
1.1 noro 1466: else if ( sum ) {
1.14 noro 1467: HCM(gj) = sum;
1.26 noro 1468: return j;
1.31 noro 1469: } else
1.26 noro 1470: g->body[j] = nd_remove_head(gj);
1471: }
1472: }
1473:
1474: int head_pbucket_q(PGeoBucket g)
1475: {
1476: int j,i,c,k,nv;
1477: Q sum,t;
1478: ND gi,gj;
1479:
1480: k = g->m;
1481: while ( 1 ) {
1482: j = -1;
1483: for ( i = 0; i <= k; i++ ) {
1.34 noro 1484: if ( !(gi = g->body[i]) ) continue;
1.26 noro 1485: if ( j < 0 ) {
1486: j = i;
1487: gj = g->body[j];
1488: sum = HCQ(gj);
1489: } else {
1490: nv = NV(gi);
1.34 noro 1491: c = DL_COMPARE(HDL(gi),HDL(gj));
1.26 noro 1492: if ( c > 0 ) {
1.34 noro 1493: if ( sum ) HCQ(gj) = sum;
1494: else g->body[j] = nd_remove_head(gj);
1.26 noro 1495: j = i;
1496: gj = g->body[j];
1497: sum = HCQ(gj);
1498: } else if ( c == 0 ) {
1499: addq(sum,HCQ(gi),&t);
1500: sum = t;
1501: g->body[i] = nd_remove_head(gi);
1502: }
1503: }
1504: }
1.34 noro 1505: if ( j < 0 ) return -1;
1.26 noro 1506: else if ( sum ) {
1507: HCQ(gj) = sum;
1.1 noro 1508: return j;
1.31 noro 1509: } else
1.1 noro 1510: g->body[j] = nd_remove_head(gj);
1511: }
1512: }
1513:
1.25 noro 1514: ND normalize_pbucket(int mod,PGeoBucket g)
1.1 noro 1515: {
1.31 noro 1516: int i;
1.1 noro 1517: ND r,t;
1518:
1519: r = 0;
1.28 noro 1520: for ( i = 0; i <= g->m; i++ ) {
1.31 noro 1521: r = nd_add(mod,r,g->body[i]);
1.28 noro 1522: g->body[i] = 0;
1523: }
1524: g->m = -1;
1.1 noro 1525: return r;
1526: }
1527:
1.92 noro 1528: void do_diagonalize(int sugar,int m)
1529: {
1530: int i,nh,stat;
1531: NODE r,g,t;
1532: ND h,nf,s,head;
1533: NDV nfv;
1534: Q q,num,den;
1535: union oNDC dn;
1536:
1537: for ( i = nd_psn-1; i >= 0 && SG(nd_psh[i]) == sugar; i-- ) {
1538: if ( nd_demand )
1539: nfv = ndv_load(i);
1540: else
1541: nfv = nd_ps[i];
1542: s = ndvtond(m,nfv);
1543: s = nd_separate_head(s,&head);
1544: nd_nf(m,s,nd_ps,1,&dn,&nf);
1545: if ( !m ) {
1546: NTOQ(NM(dn.z),SGN(dn.z),num);
1547: mulq(HCQ(head),num,&q); HCQ(head) = q;
1548: if ( DN(dn.z) ) {
1549: NTOQ(DN(dn.z),1,den);
1550: nd_mul_c_q(nf,den);
1551: }
1552: }
1553: nf = nd_add(m,head,nf);
1554: ndv_free(nfv);
1555: nd_removecont(m,nf);
1556: nfv = ndtondv(m,nf);
1557: nd_free(nf);
1558: nd_bound[i] = ndv_compute_bound(nfv);
1559: if ( nd_demand ) {
1560: ndv_save(nfv,i);
1561: ndv_free(nfv);
1562: } else
1563: nd_ps[i] = nfv;
1564: }
1565: }
1566:
1.27 noro 1567: /* return value = 0 => input is not a GB */
1568:
1.77 noro 1569: NODE nd_gb(int m,int ishomo,int checkonly)
1.1 noro 1570: {
1571: int i,nh,sugar,stat;
1.23 noro 1572: NODE r,g,t;
1.1 noro 1573: ND_pairs d;
1574: ND_pairs l;
1.77 noro 1575: ND h,nf,s,head;
1.63 noro 1576: NDV nfv;
1.85 noro 1577: Q q,num,den;
1.69 noro 1578: union oNDC dn;
1.1 noro 1579:
1.23 noro 1580: g = 0; d = 0;
1581: for ( i = 0; i < nd_psn; i++ ) {
1.1 noro 1582: d = update_pairs(d,g,i);
1583: g = update_base(g,i);
1584: }
1585: sugar = 0;
1586: while ( d ) {
1587: again:
1588: l = nd_minp(d,&d);
1.14 noro 1589: if ( SG(l) != sugar ) {
1.92 noro 1590: if ( ishomo ) do_diagonalize(sugar,m);
1591:
1.14 noro 1592: sugar = SG(l);
1.71 noro 1593: if ( DP_Print ) fprintf(asir_out,"%d",sugar);
1.1 noro 1594: }
1.53 noro 1595: stat = nd_sp(m,0,l,&h);
1.1 noro 1596: if ( !stat ) {
1597: NEXT(l) = d; d = l;
1.20 noro 1598: d = nd_reconstruct(m,0,d);
1.1 noro 1599: goto again;
1600: }
1.41 noro 1601: #if USE_GEOBUCKET
1.69 noro 1602: stat = m?nd_nf_pbucket(m,h,nd_ps,!Top,&nf):nd_nf(m,h,nd_ps,!Top,0,&nf);
1.41 noro 1603: #else
1.69 noro 1604: stat = nd_nf(m,h,nd_ps,!Top,0,&nf);
1.41 noro 1605: #endif
1.1 noro 1606: if ( !stat ) {
1607: NEXT(l) = d; d = l;
1.20 noro 1608: d = nd_reconstruct(m,0,d);
1.1 noro 1609: goto again;
1610: } else if ( nf ) {
1.27 noro 1611: if ( checkonly ) return 0;
1.71 noro 1612: if ( DP_Print ) { printf("+"); fflush(stdout); }
1.63 noro 1613: nd_removecont(m,nf);
1614: nfv = ndtondv(m,nf); nd_free(nf);
1.77 noro 1615: nh = ndv_newps(m,nfv,0);
1.1 noro 1616: d = update_pairs(d,g,nh);
1617: g = update_base(g,nh);
1618: FREENDP(l);
1619: } else {
1.71 noro 1620: if ( DP_Print ) { printf("."); fflush(stdout); }
1.1 noro 1621: FREENDP(l);
1622: }
1623: }
1.77 noro 1624: if ( nd_demand )
1625: for ( t = g; t; t = NEXT(t) )
1626: BDY(t) = (pointer)ndv_load((int)BDY(t));
1627: else
1628: for ( t = g; t; t = NEXT(t) )
1629: BDY(t) = (pointer)nd_ps[(int)BDY(t)];
1.1 noro 1630: return g;
1631: }
1632:
1.92 noro 1633: void do_diagonalize_trace(int sugar,int m)
1.91 noro 1634: {
1635: int i,nh,stat;
1636: NODE r,g,t;
1637: ND h,nf,nfq,s,head;
1638: NDV nfv,nfqv;
1639: Q q,den,num;
1640: union oNDC dn;
1641:
1642: for ( i = nd_psn-1; i >= 0 && SG(nd_psh[i]) == sugar; i-- ) {
1643: /* for nd_ps */
1644: s = ndvtond(m,nd_ps[i]);
1645: s = nd_separate_head(s,&head);
1646: nd_nf_pbucket(m,s,nd_ps,1,&nf);
1647: nf = nd_add(m,head,nf);
1648: ndv_free(nd_ps[i]);
1649: nd_ps[i] = ndtondv(m,nf);
1650: nd_free(nf);
1651:
1652: /* for nd_ps_trace */
1653: if ( nd_demand )
1654: nfv = ndv_load(i);
1655: else
1656: nfv = nd_ps_trace[i];
1657: s = ndvtond(0,nfv);
1658: s = nd_separate_head(s,&head);
1659: nd_nf(0,s,nd_ps_trace,1,&dn,&nf);
1660: NTOQ(NM(dn.z),SGN(dn.z),num);
1661: mulq(HCQ(head),num,&q); HCQ(head) = q;
1662: if ( DN(dn.z) ) {
1663: NTOQ(DN(dn.z),1,den);
1664: nd_mul_c_q(nf,den);
1665: }
1666: nf = nd_add(0,head,nf);
1667: ndv_free(nfv);
1668: nd_removecont(0,nf);
1669: nfv = ndtondv(0,nf);
1670: nd_free(nf);
1671: nd_bound[i] = ndv_compute_bound(nfv);
1672: if ( nd_demand ) {
1673: ndv_save(nfv,i);
1674: ndv_free(nfv);
1675: } else
1676: nd_ps_trace[i] = nfv;
1677: }
1678: }
1679:
1.82 noro 1680: NODE nd_gb_trace(int m,int ishomo)
1.20 noro 1681: {
1682: int i,nh,sugar,stat;
1.23 noro 1683: NODE r,g,t;
1.20 noro 1684: ND_pairs d;
1685: ND_pairs l;
1.82 noro 1686: ND h,nf,nfq,s,head;
1.63 noro 1687: NDV nfv,nfqv;
1.85 noro 1688: Q q,den,num;
1.69 noro 1689: union oNDC dn;
1.20 noro 1690:
1.23 noro 1691: g = 0; d = 0;
1692: for ( i = 0; i < nd_psn; i++ ) {
1.20 noro 1693: d = update_pairs(d,g,i);
1694: g = update_base(g,i);
1695: }
1696: sugar = 0;
1697: while ( d ) {
1698: again:
1699: l = nd_minp(d,&d);
1700: if ( SG(l) != sugar ) {
1.93 noro 1701: if ( ishomo ) do_diagonalize_trace(sugar,m);
1.20 noro 1702: sugar = SG(l);
1.71 noro 1703: if ( DP_Print ) fprintf(asir_out,"%d",sugar);
1.20 noro 1704: }
1.53 noro 1705: stat = nd_sp(m,0,l,&h);
1.20 noro 1706: if ( !stat ) {
1707: NEXT(l) = d; d = l;
1708: d = nd_reconstruct(m,1,d);
1709: goto again;
1710: }
1.41 noro 1711: #if USE_GEOBUCKET
1.53 noro 1712: stat = nd_nf_pbucket(m,h,nd_ps,!Top,&nf);
1.41 noro 1713: #else
1.69 noro 1714: stat = nd_nf(m,h,nd_ps,!Top,0,&nf);
1.41 noro 1715: #endif
1.20 noro 1716: if ( !stat ) {
1717: NEXT(l) = d; d = l;
1718: d = nd_reconstruct(m,1,d);
1719: goto again;
1720: } else if ( nf ) {
1.77 noro 1721: if ( nd_demand ) {
1722: nfqv = ndv_load(nd_psn);
1723: nfq = ndvtond(0,nfqv);
1724: } else
1725: nfq = 0;
1726: if ( !nfq ) {
1727: if ( !nd_sp(0,1,l,&h) || !nd_nf(0,h,nd_ps_trace,!Top,0,&nfq) ) {
1728: NEXT(l) = d; d = l;
1729: d = nd_reconstruct(m,1,d);
1730: goto again;
1731: }
1.70 noro 1732: }
1.20 noro 1733: if ( nfq ) {
1.70 noro 1734: /* m|HC(nfq) => failure */
1.63 noro 1735: if ( !rem(NM(HCQ(nfq)),m) ) return 0;
1736:
1.71 noro 1737: if ( DP_Print ) { printf("+"); fflush(stdout); }
1.63 noro 1738: nd_removecont(m,nf); nfv = ndtondv(m,nf); nd_free(nf);
1739: nd_removecont(0,nfq); nfqv = ndtondv(0,nfq); nd_free(nfq);
1.77 noro 1740: nh = ndv_newps(0,nfv,nfqv);
1.20 noro 1741: d = update_pairs(d,g,nh);
1742: g = update_base(g,nh);
1743: } else {
1.71 noro 1744: if ( DP_Print ) { printf("*"); fflush(stdout); }
1.20 noro 1745: }
1746: } else {
1.71 noro 1747: if ( DP_Print ) { printf("."); fflush(stdout); }
1.20 noro 1748: }
1749: FREENDP(l);
1750: }
1.77 noro 1751: if ( nd_demand )
1752: for ( t = g; t; t = NEXT(t) )
1753: BDY(t) = (pointer)ndv_load((int)BDY(t));
1754: else
1755: for ( t = g; t; t = NEXT(t) )
1756: BDY(t) = (pointer)nd_ps_trace[(int)BDY(t)];
1.20 noro 1757: return g;
1758: }
1759:
1.23 noro 1760: int ndv_compare(NDV *p1,NDV *p2)
1761: {
1.34 noro 1762: return DL_COMPARE(HDL(*p1),HDL(*p2));
1.23 noro 1763: }
1764:
1765: int ndv_compare_rev(NDV *p1,NDV *p2)
1766: {
1.34 noro 1767: return -DL_COMPARE(HDL(*p1),HDL(*p2));
1.23 noro 1768: }
1769:
1.61 noro 1770: NODE ndv_reduceall(int m,NODE f)
1.23 noro 1771: {
1.69 noro 1772: int i,n,stat;
1773: ND nf,g,head;
1.23 noro 1774: NODE t,a0,a;
1.69 noro 1775: union oNDC dn;
1776: NDV *w;
1.85 noro 1777: Q q,num,den;
1.23 noro 1778:
1.69 noro 1779: n = length(f);
1780: #if 0
1781: w = (NDV *)ALLOCA(n*sizeof(NDV));
1782: for ( i = 0, t = f; i < n; i++, t = NEXT(t) ) w[i] = (NDV)BDY(t);
1783: qsort(w,n,sizeof(NDV),
1784: (int (*)(const void *,const void *))ndv_compare);
1785: for ( t = f, i = 0; t; i++, t = NEXT(t) ) BDY(t) = (pointer)w[i];
1786: #endif
1787: ndv_setup(m,0,f);
1788: for ( i = 0; i < n; ) {
1789: g = ndvtond(m,nd_ps[i]);
1790: g = nd_separate_head(g,&head);
1791: stat = nd_nf(m,g,nd_ps,1,&dn,&nf);
1.50 noro 1792: if ( !stat )
1.69 noro 1793: nd_reconstruct(m,0,0);
1794: else {
1.71 noro 1795: if ( DP_Print ) { printf("."); fflush(stdout); }
1.85 noro 1796: if ( !m ) {
1797: NTOQ(NM(dn.z),SGN(dn.z),num);
1798: mulq(HCQ(head),num,&q); HCQ(head) = q;
1799: if ( DN(dn.z) ) {
1800: NTOQ(DN(dn.z),1,den);
1801: nd_mul_c_q(nf,den);
1802: }
1803: }
1.69 noro 1804: nf = nd_add(m,head,nf);
1805: ndv_free(nd_ps[i]);
1.24 noro 1806: nd_removecont(m,nf);
1.69 noro 1807: nd_ps[i] = ndtondv(m,nf); nd_free(nf);
1808: nd_bound[i] = ndv_compute_bound(nd_ps[i]);
1.50 noro 1809: i++;
1.23 noro 1810: }
1811: }
1.71 noro 1812: if ( DP_Print ) { printf("\n"); }
1.23 noro 1813: for ( a0 = 0, i = 0; i < n; i++ ) {
1814: NEXTNODE(a0,a);
1.69 noro 1815: BDY(a) = (pointer)nd_ps[i];
1.23 noro 1816: }
1817: NEXT(a) = 0;
1818: return a0;
1819: }
1820:
1.1 noro 1821: ND_pairs update_pairs( ND_pairs d, NODE /* of index */ g, int t)
1822: {
1823: ND_pairs d1,nd,cur,head,prev,remove;
1824:
1825: if ( !g ) return d;
1826: d = crit_B(d,t);
1827: d1 = nd_newpairs(g,t);
1828: d1 = crit_M(d1);
1829: d1 = crit_F(d1);
1.55 noro 1830: if ( do_weyl )
1831: head = d1;
1832: else {
1833: prev = 0; cur = head = d1;
1834: while ( cur ) {
1835: if ( crit_2( cur->i1,cur->i2 ) ) {
1836: remove = cur;
1837: if ( !prev ) head = cur = NEXT(cur);
1838: else cur = NEXT(prev) = NEXT(cur);
1839: FREENDP(remove);
1840: } else {
1841: prev = cur; cur = NEXT(cur);
1842: }
1.1 noro 1843: }
1844: }
1845: if ( !d )
1846: return head;
1847: else {
1848: nd = d;
1.34 noro 1849: while ( NEXT(nd) ) nd = NEXT(nd);
1.1 noro 1850: NEXT(nd) = head;
1851: return d;
1852: }
1853: }
1854:
1855: ND_pairs nd_newpairs( NODE g, int t )
1856: {
1857: NODE h;
1.61 noro 1858: UINT *dl;
1.34 noro 1859: int ts,s;
1.1 noro 1860: ND_pairs r,r0;
1861:
1.20 noro 1862: dl = DL(nd_psh[t]);
1.34 noro 1863: ts = SG(nd_psh[t]) - TD(dl);
1.1 noro 1864: for ( r0 = 0, h = g; h; h = NEXT(h) ) {
1865: NEXTND_pairs(r0,r);
1866: r->i1 = (int)BDY(h);
1867: r->i2 = t;
1.20 noro 1868: ndl_lcm(DL(nd_psh[r->i1]),dl,r->lcm);
1.34 noro 1869: s = SG(nd_psh[r->i1])-TD(DL(nd_psh[r->i1]));
1870: SG(r) = MAX(s,ts) + TD(LCM(r));
1.1 noro 1871: }
1872: NEXT(r) = 0;
1873: return r0;
1874: }
1875:
1876: ND_pairs crit_B( ND_pairs d, int s )
1877: {
1878: ND_pairs cur,head,prev,remove;
1.61 noro 1879: UINT *t,*tl,*lcm;
1.1 noro 1880: int td,tdl;
1881:
1882: if ( !d ) return 0;
1.20 noro 1883: t = DL(nd_psh[s]);
1.1 noro 1884: prev = 0;
1885: head = cur = d;
1.61 noro 1886: lcm = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
1.1 noro 1887: while ( cur ) {
1888: tl = cur->lcm;
1889: if ( ndl_reducible(tl,t)
1.20 noro 1890: && (ndl_lcm(DL(nd_psh[cur->i1]),t,lcm),!ndl_equal(lcm,tl))
1891: && (ndl_lcm(DL(nd_psh[cur->i2]),t,lcm),!ndl_equal(lcm,tl)) ) {
1.1 noro 1892: remove = cur;
1893: if ( !prev ) {
1894: head = cur = NEXT(cur);
1895: } else {
1896: cur = NEXT(prev) = NEXT(cur);
1897: }
1898: FREENDP(remove);
1899: } else {
1.34 noro 1900: prev = cur; cur = NEXT(cur);
1.1 noro 1901: }
1902: }
1903: return head;
1904: }
1905:
1906: ND_pairs crit_M( ND_pairs d1 )
1907: {
1908: ND_pairs e,d2,d3,dd,p;
1.61 noro 1909: UINT *id,*jd;
1.1 noro 1910:
1911: for ( dd = 0, e = d1; e; e = d3 ) {
1912: if ( !(d2 = NEXT(e)) ) {
1913: NEXT(e) = dd;
1914: return e;
1915: }
1.34 noro 1916: id = LCM(e);
1.1 noro 1917: for ( d3 = 0; d2; d2 = p ) {
1.34 noro 1918: p = NEXT(d2);
1919: jd = LCM(d2);
1920: if ( ndl_equal(jd,id) )
1921: ;
1922: else if ( TD(jd) > TD(id) )
1.1 noro 1923: if ( ndl_reducible(jd,id) ) continue;
1924: else ;
1.34 noro 1925: else if ( ndl_reducible(id,jd) ) goto delit;
1.1 noro 1926: NEXT(d2) = d3;
1927: d3 = d2;
1928: }
1929: NEXT(e) = dd;
1930: dd = e;
1931: continue;
1932: /**/
1933: delit: NEXT(d2) = d3;
1934: d3 = d2;
1935: for ( ; p; p = d2 ) {
1936: d2 = NEXT(p);
1937: NEXT(p) = d3;
1938: d3 = p;
1939: }
1940: FREENDP(e);
1941: }
1942: return dd;
1943: }
1944:
1945: ND_pairs crit_F( ND_pairs d1 )
1946: {
1947: ND_pairs rest, head,remove;
1948: ND_pairs last, p, r, w;
1949: int s;
1950:
1951: for ( head = last = 0, p = d1; NEXT(p); ) {
1952: r = w = equivalent_pairs(p,&rest);
1.14 noro 1953: s = SG(r);
1.1 noro 1954: w = NEXT(w);
1955: while ( w ) {
1956: if ( crit_2(w->i1,w->i2) ) {
1957: r = w;
1958: w = NEXT(w);
1959: while ( w ) {
1960: remove = w;
1961: w = NEXT(w);
1962: FREENDP(remove);
1963: }
1964: break;
1.14 noro 1965: } else if ( SG(w) < s ) {
1.1 noro 1966: FREENDP(r);
1967: r = w;
1.14 noro 1968: s = SG(r);
1.1 noro 1969: w = NEXT(w);
1970: } else {
1971: remove = w;
1972: w = NEXT(w);
1973: FREENDP(remove);
1974: }
1975: }
1976: if ( last ) NEXT(last) = r;
1977: else head = r;
1978: NEXT(last = r) = 0;
1979: p = rest;
1980: if ( !p ) return head;
1981: }
1982: if ( !last ) return p;
1983: NEXT(last) = p;
1984: return head;
1985: }
1986:
1987: int crit_2( int dp1, int dp2 )
1988: {
1.20 noro 1989: return ndl_disjoint(DL(nd_psh[dp1]),DL(nd_psh[dp2]));
1.1 noro 1990: }
1991:
1.40 noro 1992: ND_pairs equivalent_pairs( ND_pairs d1, ND_pairs *prest )
1.1 noro 1993: {
1994: ND_pairs w,p,r,s;
1.61 noro 1995: UINT *d;
1.1 noro 1996:
1997: w = d1;
1.34 noro 1998: d = LCM(w);
1.1 noro 1999: s = NEXT(w);
2000: NEXT(w) = 0;
2001: for ( r = 0; s; s = p ) {
2002: p = NEXT(s);
1.34 noro 2003: if ( ndl_equal(d,LCM(s)) ) {
1.39 noro 2004: NEXT(s) = w; w = s;
1.1 noro 2005: } else {
1.39 noro 2006: NEXT(s) = r; r = s;
1.1 noro 2007: }
2008: }
2009: *prest = r;
2010: return w;
2011: }
2012:
2013: NODE update_base(NODE nd,int ndp)
2014: {
1.61 noro 2015: UINT *dl, *dln;
1.1 noro 2016: NODE last, p, head;
2017:
1.20 noro 2018: dl = DL(nd_psh[ndp]);
1.1 noro 2019: for ( head = last = 0, p = nd; p; ) {
1.20 noro 2020: dln = DL(nd_psh[(int)BDY(p)]);
1.34 noro 2021: if ( ndl_reducible( dln, dl ) ) {
1.1 noro 2022: p = NEXT(p);
2023: if ( last ) NEXT(last) = p;
2024: } else {
2025: if ( !last ) head = p;
2026: p = NEXT(last = p);
2027: }
2028: }
2029: head = append_one(head,ndp);
2030: return head;
2031: }
2032:
2033: ND_pairs nd_minp( ND_pairs d, ND_pairs *prest )
2034: {
2035: ND_pairs m,ml,p,l;
1.61 noro 2036: UINT *lcm;
1.33 noro 2037: int s,td,len,tlen,c,c1;
1.1 noro 2038:
2039: if ( !(p = NEXT(m = d)) ) {
2040: *prest = p;
2041: NEXT(m) = 0;
2042: return m;
2043: }
1.14 noro 2044: s = SG(m);
1.77 noro 2045: if ( !NoSugar ) {
2046: for ( ml = 0, l = m; p; p = NEXT(l = p) )
2047: if ( (SG(p) < s)
2048: || ((SG(p) == s) && (DL_COMPARE(LCM(p),LCM(m)) < 0)) ) {
2049: ml = l; m = p; s = SG(m);
2050: }
2051: } else {
2052: for ( ml = 0, l = m; p; p = NEXT(l = p) )
2053: if ( DL_COMPARE(LCM(p),LCM(m)) < 0 ) {
2054: ml = l; m = p; s = SG(m);
2055: }
2056: }
1.1 noro 2057: if ( !ml ) *prest = NEXT(m);
2058: else {
2059: NEXT(ml) = NEXT(m);
2060: *prest = d;
2061: }
2062: NEXT(m) = 0;
2063: return m;
2064: }
2065:
1.63 noro 2066: ND_pairs nd_minsugarp( ND_pairs d, ND_pairs *prest )
2067: {
1.69 noro 2068: int msugar,i;
1.63 noro 2069: ND_pairs t,dm0,dm,dr0,dr;
2070:
2071: for ( msugar = SG(d), t = NEXT(d); t; t = NEXT(t) )
2072: if ( SG(t) < msugar ) msugar = SG(t);
2073: dm0 = 0; dr0 = 0;
1.69 noro 2074: for ( i = 0, t = d; t; t = NEXT(t) )
2075: if ( i < nd_f4_nsp && SG(t) == msugar ) {
1.63 noro 2076: if ( dm0 ) NEXT(dm) = t;
2077: else dm0 = t;
2078: dm = t;
1.69 noro 2079: i++;
1.63 noro 2080: } else {
2081: if ( dr0 ) NEXT(dr) = t;
2082: else dr0 = t;
2083: dr = t;
2084: }
2085: NEXT(dm) = 0;
2086: if ( dr0 ) NEXT(dr) = 0;
2087: *prest = dr0;
2088: return dm0;
2089: }
2090:
1.77 noro 2091: int ndv_newps(int m,NDV a,NDV aq)
1.1 noro 2092: {
1.3 noro 2093: int len;
1.13 noro 2094: RHist r;
1.20 noro 2095: NDV b;
1.3 noro 2096:
1.1 noro 2097: if ( nd_psn == nd_pslen ) {
2098: nd_pslen *= 2;
1.11 noro 2099: nd_ps = (NDV *)REALLOC((char *)nd_ps,nd_pslen*sizeof(NDV));
1.53 noro 2100: nd_ps_trace = (NDV *)REALLOC((char *)nd_ps_trace,nd_pslen*sizeof(NDV));
1.13 noro 2101: nd_psh = (RHist *)REALLOC((char *)nd_psh,nd_pslen*sizeof(RHist));
1.61 noro 2102: nd_bound = (UINT **)
2103: REALLOC((char *)nd_bound,nd_pslen*sizeof(UINT *));
1.1 noro 2104: }
1.39 noro 2105: NEWRHist(r); nd_psh[nd_psn] = r;
1.63 noro 2106: nd_ps[nd_psn] = a;
1.39 noro 2107: if ( aq ) {
1.63 noro 2108: nd_ps_trace[nd_psn] = aq;
2109: nd_bound[nd_psn] = ndv_compute_bound(aq);
1.39 noro 2110: SG(r) = SG(aq); ndl_copy(HDL(aq),DL(r));
1.53 noro 2111: } else {
1.63 noro 2112: nd_bound[nd_psn] = ndv_compute_bound(a);
1.53 noro 2113: SG(r) = SG(a); ndl_copy(HDL(a),DL(r));
1.39 noro 2114: }
1.77 noro 2115: if ( nd_demand ) {
2116: if ( aq ) {
2117: ndv_save(nd_ps_trace[nd_psn],nd_psn);
2118: nd_ps_trace[nd_psn] = 0;
2119: } else {
2120: ndv_save(nd_ps[nd_psn],nd_psn);
2121: nd_ps[nd_psn] = 0;
2122: }
2123: }
1.1 noro 2124: return nd_psn++;
2125: }
2126:
1.61 noro 2127: void ndv_setup(int mod,int trace,NODE f)
1.1 noro 2128: {
1.5 noro 2129: int i,j,td,len,max;
1.1 noro 2130: NODE s,s0,f0;
1.61 noro 2131: UINT *d;
1.13 noro 2132: RHist r;
1.69 noro 2133: NDV *w;
1.61 noro 2134: NDV a,am;
1.11 noro 2135:
2136: nd_found = 0; nd_notfirst = 0; nd_create = 0;
1.1 noro 2137:
1.69 noro 2138: for ( nd_psn = 0, s = f; s; s = NEXT(s) ) if ( BDY(s) ) nd_psn++;
2139: w = (NDV *)ALLOCA(nd_psn*sizeof(NDV));
2140: for ( i = 0, s = f; s; s = NEXT(s) ) if ( BDY(s) ) w[i++] = BDY(s);
2141: qsort(w,nd_psn,sizeof(NDV),
2142: (int (*)(const void *,const void *))ndv_compare);
2143: nd_pslen = 2*nd_psn;
1.11 noro 2144: nd_ps = (NDV *)MALLOC(nd_pslen*sizeof(NDV));
1.53 noro 2145: nd_ps_trace = (NDV *)MALLOC(nd_pslen*sizeof(NDV));
1.13 noro 2146: nd_psh = (RHist *)MALLOC(nd_pslen*sizeof(RHist));
1.61 noro 2147: nd_bound = (UINT **)MALLOC(nd_pslen*sizeof(UINT *));
1.57 noro 2148:
1.11 noro 2149: if ( !nd_red )
1.13 noro 2150: nd_red = (RHist *)MALLOC(REDTAB_LEN*sizeof(RHist));
1.71 noro 2151: for ( i = 0; i < REDTAB_LEN; i++ ) nd_red[i] = 0;
1.69 noro 2152: for ( i = 0; i < nd_psn; i++ ) {
1.39 noro 2153: if ( trace ) {
1.69 noro 2154: a = nd_ps_trace[i] = ndv_dup(0,w[i]);
1.61 noro 2155: ndv_removecont(0,a);
2156: am = nd_ps[i] = ndv_dup(mod,a);
2157: ndv_mod(mod,am);
2158: ndv_removecont(mod,am);
2159: } else {
1.69 noro 2160: a = nd_ps[i] = ndv_dup(mod,w[i]);
1.61 noro 2161: ndv_removecont(mod,a);
1.39 noro 2162: }
1.61 noro 2163: NEWRHist(r); SG(r) = HTD(a); ndl_copy(HDL(a),DL(r));
1.57 noro 2164: nd_bound[i] = ndv_compute_bound(a);
1.20 noro 2165: nd_psh[i] = r;
1.77 noro 2166: if ( nd_demand ) {
2167: if ( trace ) {
2168: ndv_save(nd_ps_trace[i],i);
2169: nd_ps_trace[i] = 0;
2170: } else {
2171: ndv_save(nd_ps[i],i);
2172: nd_ps[i] = 0;
2173: }
2174: }
1.20 noro 2175: }
2176: }
2177:
1.63 noro 2178: void nd_gr(LIST f,LIST v,int m,int f4,struct order_spec *ord,LIST *rp)
1.1 noro 2179: {
1.61 noro 2180: VL tv,fv,vv,vc;
1.1 noro 2181: NODE fd,fd0,r,r0,t,x,s,xx;
1.61 noro 2182: int e,max,nvar;
1.69 noro 2183: NDV b;
1.77 noro 2184: int ishomo;
1.1 noro 2185:
1.78 noro 2186: if ( !m && Demand ) nd_demand = 1;
2187: else nd_demand = 0;
2188:
1.74 noro 2189: ndv_alloc = 0;
1.1 noro 2190: get_vars((Obj)f,&fv); pltovl(v,&vv);
1.74 noro 2191: for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++ );
1.88 noro 2192: switch ( ord->id ) {
2193: case 1:
2194: if ( ord->nv != nvar )
2195: error("nd_{gr,f4} : invalid order specification");
2196: break;
2197: default:
2198: break;
2199: }
1.32 noro 2200: nd_init_ord(ord);
1.61 noro 2201: for ( t = BDY(f), max = 0; t; t = NEXT(t) )
2202: for ( tv = vv; tv; tv = NEXT(tv) ) {
2203: e = getdeg(tv->v,(P)BDY(t));
2204: max = MAX(e,max);
2205: }
2206: nd_setup_parameters(nvar,max);
1.77 noro 2207: ishomo = 1;
1.1 noro 2208: for ( fd0 = 0, t = BDY(f); t; t = NEXT(t) ) {
1.61 noro 2209: b = (pointer)ptondv(CO,vv,(P)BDY(t));
1.77 noro 2210: if ( ishomo )
2211: ishomo = ishomo && ndv_ishomo(b);
1.69 noro 2212: if ( m ) ndv_mod(m,b);
1.61 noro 2213: if ( b ) { NEXTNODE(fd0,fd); BDY(fd) = (pointer)b; }
1.1 noro 2214: }
2215: if ( fd0 ) NEXT(fd) = 0;
1.61 noro 2216: ndv_setup(m,0,fd0);
1.77 noro 2217: x = f4?nd_f4(m):nd_gb(m,ishomo,0);
1.82 noro 2218: nd_demand = 0;
1.61 noro 2219: x = ndv_reducebase(x);
2220: x = ndv_reduceall(m,x);
1.23 noro 2221: for ( r0 = 0, t = x; t; t = NEXT(t) ) {
1.1 noro 2222: NEXTNODE(r0,r);
1.61 noro 2223: BDY(r) = ndvtop(m,CO,vv,BDY(t));
1.20 noro 2224: }
2225: if ( r0 ) NEXT(r) = 0;
2226: MKLIST(*rp,r0);
1.74 noro 2227: fprintf(asir_out,"ndv_alloc=%d\n",ndv_alloc);
1.20 noro 2228: }
2229:
1.52 noro 2230: void nd_gr_trace(LIST f,LIST v,int trace,int homo,struct order_spec *ord,LIST *rp)
1.20 noro 2231: {
1.89 noro 2232: struct order_spec *ord1;
1.61 noro 2233: VL tv,fv,vv,vc;
1.27 noro 2234: NODE fd,fd0,in0,in,r,r0,t,s,cand;
1.61 noro 2235: int m,nocheck,nvar,mindex,e,max;
2236: NDV c;
2237: NMV a;
1.27 noro 2238: P p;
1.61 noro 2239: EPOS oepos;
1.82 noro 2240: int obpe,oadv,wmax,i,len,cbpe,ishomo;
1.20 noro 2241:
2242: get_vars((Obj)f,&fv); pltovl(v,&vv);
1.74 noro 2243: for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++ );
1.88 noro 2244: switch ( ord->id ) {
2245: case 1:
2246: if ( ord->nv != nvar )
2247: error("nd_gr_trace : invalid order specification");
2248: break;
2249: default:
2250: break;
2251: }
1.52 noro 2252: nocheck = 0;
2253: mindex = 0;
2254:
1.78 noro 2255: if ( Demand ) nd_demand = 1;
2256: else nd_demand = 0;
2257:
1.52 noro 2258: /* setup modulus */
2259: if ( trace < 0 ) {
2260: trace = -trace;
2261: nocheck = 1;
2262: }
2263: m = trace > 1 ? trace : get_lprime(mindex);
1.61 noro 2264: for ( t = BDY(f), max = 0; t; t = NEXT(t) )
2265: for ( tv = vv; tv; tv = NEXT(tv) ) {
2266: e = getdeg(tv->v,(P)BDY(t));
2267: max = MAX(e,max);
1.23 noro 2268: }
1.61 noro 2269: nd_init_ord(ord);
2270: nd_setup_parameters(nvar,max);
2271: obpe = nd_bpe; oadv = nmv_adv; oepos = nd_epos;
1.82 noro 2272: ishomo = 1;
1.61 noro 2273: for ( in0 = 0, fd0 = 0, t = BDY(f); t; t = NEXT(t) ) {
2274: c = ptondv(CO,vv,(P)BDY(t));
1.82 noro 2275: if ( ishomo )
2276: ishomo = ishomo && ndv_ishomo(c);
1.61 noro 2277: if ( c ) {
2278: NEXTNODE(in0,in); BDY(in) = (pointer)c;
2279: NEXTNODE(fd0,fd); BDY(fd) = (pointer)ndv_dup(0,c);
1.23 noro 2280: }
1.61 noro 2281: }
2282: if ( in0 ) NEXT(in) = 0;
2283: if ( fd0 ) NEXT(fd) = 0;
1.82 noro 2284: if ( !ishomo && homo ) {
1.61 noro 2285: for ( t = in0, wmax = 0; t; t = NEXT(t) ) {
2286: c = (NDV)BDY(t); len = LEN(c);
2287: for ( a = BDY(c), i = 0; i < len; i++, NMV_ADV(a) )
2288: wmax = MAX(TD(DL(a)),wmax);
2289: }
2290: homogenize_order(ord,nvar,&ord1);
1.89 noro 2291: nd_init_ord(ord1);
1.61 noro 2292: nd_setup_parameters(nvar+1,wmax);
2293: for ( t = fd0; t; t = NEXT(t) )
2294: ndv_homogenize((NDV)BDY(t),obpe,oadv,oepos);
1.27 noro 2295: }
1.52 noro 2296: while ( 1 ) {
1.77 noro 2297: if ( Demand )
2298: nd_demand = 1;
1.61 noro 2299: ndv_setup(m,1,fd0);
1.82 noro 2300: cand = nd_gb_trace(m,ishomo || homo);
1.52 noro 2301: if ( !cand ) {
2302: /* failure */
1.61 noro 2303: if ( trace > 1 ) { *rp = 0; return; }
2304: else m = get_lprime(++mindex);
1.52 noro 2305: continue;
2306: }
1.82 noro 2307: if ( !ishomo && homo ) {
1.27 noro 2308: /* dehomogenization */
1.61 noro 2309: for ( t = cand; t; t = NEXT(t) ) ndv_dehomogenize((NDV)BDY(t),ord);
1.45 noro 2310: nd_init_ord(ord);
1.61 noro 2311: nd_setup_parameters(nvar,0);
1.27 noro 2312: }
1.77 noro 2313: nd_demand = 0;
1.61 noro 2314: cand = ndv_reducebase(cand);
2315: cand = ndv_reduceall(0,cand);
1.78 noro 2316: cbpe = nd_bpe;
1.61 noro 2317: if ( nocheck )
2318: break;
2319: if ( ndv_check_candidate(in0,obpe,oadv,oepos,cand) )
1.52 noro 2320: /* success */
2321: break;
2322: else if ( trace > 1 ) {
2323: /* failure */
2324: *rp = 0; return;
1.61 noro 2325: } else {
1.52 noro 2326: /* try the next modulus */
2327: m = get_lprime(++mindex);
1.61 noro 2328: /* reset the parameters */
1.82 noro 2329: if ( !ishomo && homo ) {
1.89 noro 2330: nd_init_ord(ord1);
1.61 noro 2331: nd_setup_parameters(nvar+1,wmax);
2332: } else {
2333: nd_init_ord(ord);
2334: nd_setup_parameters(nvar,max);
2335: }
2336: }
1.52 noro 2337: }
1.27 noro 2338: /* dp->p */
1.61 noro 2339: nd_bpe = cbpe;
1.73 noro 2340: nd_setup_parameters(nd_nvar,0);
1.61 noro 2341: for ( r = cand; r; r = NEXT(r) ) BDY(r) = (pointer)ndvtop(0,CO,vv,BDY(r));
1.27 noro 2342: MKLIST(*rp,cand);
1.1 noro 2343: }
2344:
1.61 noro 2345: void dltondl(int n,DL dl,UINT *r)
1.1 noro 2346: {
1.61 noro 2347: UINT *d;
1.57 noro 2348: int i,j,l,s,ord_l;
1.43 noro 2349: struct order_pair *op;
1.1 noro 2350:
2351: d = dl->d;
1.41 noro 2352: for ( i = 0; i < nd_wpd; i++ ) r[i] = 0;
1.43 noro 2353: if ( nd_blockmask ) {
2354: l = nd_blockmask->n;
2355: op = nd_blockmask->order_pair;
2356: for ( j = 0, s = 0; j < l; j++ ) {
2357: ord_l = op[j].length;
1.57 noro 2358: for ( i = 0; i < ord_l; i++, s++ ) PUT_EXP(r,s,d[s]);
1.43 noro 2359: }
2360: TD(r) = ndl_weight(r);
1.61 noro 2361: ndl_weight_mask(r);
1.43 noro 2362: } else {
1.56 noro 2363: for ( i = 0; i < n; i++ ) PUT_EXP(r,i,d[i]);
1.43 noro 2364: TD(r) = ndl_weight(r);
2365: }
1.1 noro 2366: }
2367:
1.61 noro 2368: DL ndltodl(int n,UINT *ndl)
1.1 noro 2369: {
2370: DL dl;
2371: int *d;
1.57 noro 2372: int i,j,l,s,ord_l;
1.43 noro 2373: struct order_pair *op;
1.1 noro 2374:
2375: NEWDL(dl,n);
1.34 noro 2376: dl->td = TD(ndl);
1.1 noro 2377: d = dl->d;
1.43 noro 2378: if ( nd_blockmask ) {
2379: l = nd_blockmask->n;
2380: op = nd_blockmask->order_pair;
2381: for ( j = 0, s = 0; j < l; j++ ) {
2382: ord_l = op[j].length;
1.57 noro 2383: for ( i = 0; i < ord_l; i++, s++ ) d[s] = GET_EXP(ndl,s);
1.43 noro 2384: }
2385: } else {
1.56 noro 2386: for ( i = 0; i < n; i++ ) d[i] = GET_EXP(ndl,i);
1.43 noro 2387: }
1.1 noro 2388: return dl;
2389: }
2390:
1.61 noro 2391: void ndl_print(UINT *dl)
1.1 noro 2392: {
2393: int n;
1.57 noro 2394: int i,j,l,ord_l,s,s0;
1.43 noro 2395: struct order_pair *op;
1.1 noro 2396:
2397: n = nd_nvar;
2398: printf("<<");
1.43 noro 2399: if ( nd_blockmask ) {
2400: l = nd_blockmask->n;
2401: op = nd_blockmask->order_pair;
2402: for ( j = 0, s = s0 = 0; j < l; j++ ) {
2403: ord_l = op[j].length;
1.57 noro 2404: for ( i = 0; i < ord_l; i++, s++ )
2405: printf(s==n-1?"%d":"%d,",GET_EXP(dl,s));
1.43 noro 2406: }
2407: } else {
1.56 noro 2408: for ( i = 0; i < n; i++ ) printf(i==n-1?"%d":"%d,",GET_EXP(dl,i));
1.43 noro 2409: }
1.1 noro 2410: printf(">>");
2411: }
2412:
2413: void nd_print(ND p)
2414: {
2415: NM m;
2416:
2417: if ( !p )
2418: printf("0\n");
2419: else {
2420: for ( m = BDY(p); m; m = NEXT(m) ) {
1.71 noro 2421: if ( CM(m) & 0x80000000 ) printf("+@_%d*",IFTOF(CM(m)));
2422: else printf("+%d*",CM(m));
1.14 noro 2423: ndl_print(DL(m));
1.1 noro 2424: }
2425: printf("\n");
2426: }
2427: }
2428:
1.16 noro 2429: void nd_print_q(ND p)
2430: {
2431: NM m;
2432:
2433: if ( !p )
2434: printf("0\n");
2435: else {
2436: for ( m = BDY(p); m; m = NEXT(m) ) {
2437: printf("+");
1.74 noro 2438: printexpr(CO,(Obj)CQ(m));
1.16 noro 2439: printf("*");
2440: ndl_print(DL(m));
2441: }
2442: printf("\n");
2443: }
2444: }
2445:
1.1 noro 2446: void ndp_print(ND_pairs d)
2447: {
2448: ND_pairs t;
2449:
1.34 noro 2450: for ( t = d; t; t = NEXT(t) ) printf("%d,%d ",t->i1,t->i2);
1.1 noro 2451: printf("\n");
2452: }
2453:
1.20 noro 2454: void nd_removecont(int mod,ND p)
1.16 noro 2455: {
2456: int i,n;
2457: Q *w;
2458: Q dvr,t;
2459: NM m;
1.21 noro 2460: struct oVECT v;
2461: N q,r;
1.16 noro 2462:
1.71 noro 2463: if ( mod == -1 ) nd_mul_c(mod,p,_invsf(HCM(p)));
2464: else if ( mod ) nd_mul_c(mod,p,invm(HCM(p),mod));
1.20 noro 2465: else {
2466: for ( m = BDY(p), n = 0; m; m = NEXT(m), n++ );
2467: w = (Q *)ALLOCA(n*sizeof(Q));
1.21 noro 2468: v.len = n;
2469: v.body = (pointer *)w;
1.34 noro 2470: for ( m = BDY(p), i = 0; i < n; m = NEXT(m), i++ ) w[i] = CQ(m);
1.21 noro 2471: removecont_array(w,n);
2472: for ( m = BDY(p), i = 0; i < n; m = NEXT(m), i++ ) CQ(m) = w[i];
1.16 noro 2473: }
2474: }
2475:
1.21 noro 2476: void nd_removecont2(ND p1,ND p2)
2477: {
2478: int i,n1,n2,n;
2479: Q *w;
2480: Q dvr,t;
2481: NM m;
2482: struct oVECT v;
2483: N q,r;
2484:
2485: if ( !p1 ) {
2486: nd_removecont(0,p2); return;
2487: } else if ( !p2 ) {
2488: nd_removecont(0,p1); return;
2489: }
2490: n1 = nd_length(p1);
2491: n2 = nd_length(p2);
2492: n = n1+n2;
2493: w = (Q *)ALLOCA(n*sizeof(Q));
2494: v.len = n;
2495: v.body = (pointer *)w;
1.34 noro 2496: for ( m = BDY(p1), i = 0; i < n1; m = NEXT(m), i++ ) w[i] = CQ(m);
2497: for ( m = BDY(p2); i < n; m = NEXT(m), i++ ) w[i] = CQ(m);
1.21 noro 2498: removecont_array(w,n);
2499: for ( m = BDY(p1), i = 0; i < n1; m = NEXT(m), i++ ) CQ(m) = w[i];
2500: for ( m = BDY(p2); i < n; m = NEXT(m), i++ ) CQ(m) = w[i];
2501: }
2502:
1.20 noro 2503: void ndv_removecont(int mod,NDV p)
1.16 noro 2504: {
2505: int i,len;
2506: Q *w;
2507: Q dvr,t;
2508: NMV m;
2509:
1.71 noro 2510: if ( mod == -1 )
2511: ndv_mul_c(mod,p,_invsf(HCM(p)));
2512: else if ( mod )
1.20 noro 2513: ndv_mul_c(mod,p,invm(HCM(p),mod));
2514: else {
2515: len = p->len;
2516: w = (Q *)ALLOCA(len*sizeof(Q));
1.34 noro 2517: for ( m = BDY(p), i = 0; i < len; NMV_ADV(m), i++ ) w[i] = CQ(m);
1.20 noro 2518: sortbynm(w,len);
2519: qltozl(w,len,&dvr);
2520: for ( m = BDY(p), i = 0; i < len; NMV_ADV(m), i++ ) {
2521: divq(CQ(m),dvr,&t); CQ(m) = t;
2522: }
1.16 noro 2523: }
1.21 noro 2524: }
2525:
1.61 noro 2526: void ndv_homogenize(NDV p,int obpe,int oadv,EPOS oepos)
2527: {
2528: int len,i,max;
2529: NMV m,mr0,mr,t;
2530:
2531: len = p->len;
2532: for ( m = BDY(p), i = 0, max = 0; i < len; NMV_OADV(m), i++ )
2533: max = MAX(max,TD(DL(m)));
2534: mr0 = nmv_adv>oadv?(NMV)REALLOC(BDY(p),len*nmv_adv):BDY(p);
2535: m = (NMV)((char *)mr0+(len-1)*oadv);
2536: mr = (NMV)((char *)mr0+(len-1)*nmv_adv);
2537: t = (NMV)ALLOCA(nmv_adv);
2538: for ( i = 0; i < len; i++, NMV_OPREV(m), NMV_PREV(mr) ) {
2539: ndl_homogenize(DL(m),DL(t),obpe,oepos,max);
2540: CQ(mr) = CQ(m);
2541: ndl_copy(DL(t),DL(mr));
2542: }
2543: NV(p)++;
2544: BDY(p) = mr0;
2545: }
2546:
1.45 noro 2547: void ndv_dehomogenize(NDV p,struct order_spec *ord)
1.23 noro 2548: {
1.45 noro 2549: int i,j,adj,len,newnvar,newwpd,newadv,newexporigin;
1.23 noro 2550: Q *w;
2551: Q dvr,t;
2552: NMV m,r;
2553:
2554: len = p->len;
2555: newnvar = nd_nvar-1;
1.48 noro 2556: newexporigin = nd_get_exporigin(ord);
1.45 noro 2557: newwpd = newnvar/nd_epw+(newnvar%nd_epw?1:0)+newexporigin;
1.23 noro 2558: for ( m = BDY(p), i = 0; i < len; NMV_ADV(m), i++ )
1.34 noro 2559: ndl_dehomogenize(DL(m));
1.23 noro 2560: if ( newwpd != nd_wpd ) {
1.90 noro 2561: newadv = ROUND_FOR_ALIGN(sizeof(struct oNMV)+(newwpd-1)*sizeof(UINT));
1.61 noro 2562: for ( m = r = BDY(p), i = 0; i < len; NMV_ADV(m), NDV_NADV(r), i++ ) {
1.45 noro 2563: CQ(r) = CQ(m);
2564: for ( j = 0; j < newexporigin; j++ ) DL(r)[j] = DL(m)[j];
2565: adj = nd_exporigin-newexporigin;
2566: for ( ; j < newwpd; j++ ) DL(r)[j] = DL(m)[j+adj];
1.23 noro 2567: }
2568: }
2569: NV(p)--;
2570: }
2571:
1.21 noro 2572: void removecont_array(Q *c,int n)
2573: {
2574: struct oVECT v;
2575: Q d0,d1,a,u,u1,gcd;
2576: int i;
2577: N qn,rn,gn;
2578: Q *q,*r;
2579:
2580: q = (Q *)ALLOCA(n*sizeof(Q));
2581: r = (Q *)ALLOCA(n*sizeof(Q));
2582: v.id = O_VECT; v.len = n; v.body = (pointer *)c;
2583: igcdv_estimate(&v,&d0);
2584: for ( i = 0; i < n; i++ ) {
2585: divn(NM(c[i]),NM(d0),&qn,&rn);
2586: NTOQ(qn,SGN(c[i])*SGN(d0),q[i]);
2587: NTOQ(rn,SGN(c[i]),r[i]);
2588: }
1.34 noro 2589: for ( i = 0; i < n; i++ ) if ( r[i] ) break;
1.21 noro 2590: if ( i < n ) {
2591: v.id = O_VECT; v.len = n; v.body = (pointer *)r;
2592: igcdv(&v,&d1);
2593: gcdn(NM(d0),NM(d1),&gn); NTOQ(gn,1,gcd);
2594: divsn(NM(d0),gn,&qn); NTOQ(qn,1,a);
2595: for ( i = 0; i < n; i++ ) {
2596: mulq(a,q[i],&u);
2597: if ( r[i] ) {
2598: divsn(NM(r[i]),gn,&qn); NTOQ(qn,SGN(r[i]),u1);
2599: addq(u,u1,&q[i]);
2600: } else
2601: q[i] = u;
2602: }
2603: }
1.34 noro 2604: for ( i = 0; i < n; i++ ) c[i] = q[i];
1.16 noro 2605: }
2606:
1.19 noro 2607: void nd_mul_c(int mod,ND p,int mul)
1.1 noro 2608: {
2609: NM m;
2610: int c,c1;
2611:
1.34 noro 2612: if ( !p ) return;
1.71 noro 2613: if ( mod == -1 )
2614: for ( m = BDY(p); m; m = NEXT(m) )
2615: CM(m) = _mulsf(CM(m),mul);
2616: else
2617: for ( m = BDY(p); m; m = NEXT(m) ) {
2618: c1 = CM(m); DMAR(c1,mul,0,mod,c); CM(m) = c;
2619: }
1.1 noro 2620: }
2621:
1.16 noro 2622: void nd_mul_c_q(ND p,Q mul)
2623: {
2624: NM m;
2625: Q c;
2626:
1.34 noro 2627: if ( !p ) return;
1.16 noro 2628: for ( m = BDY(p); m; m = NEXT(m) ) {
2629: mulq(CQ(m),mul,&c); CQ(m) = c;
2630: }
2631: }
2632:
1.61 noro 2633: void nd_mul_c_p(VL vl,ND p,P mul)
2634: {
2635: NM m;
2636: P c;
2637:
2638: if ( !p ) return;
2639: for ( m = BDY(p); m; m = NEXT(m) ) {
2640: mulp(vl,CP(m),mul,&c); CP(m) = c;
2641: }
2642: }
2643:
1.1 noro 2644: void nd_free(ND p)
2645: {
2646: NM t,s;
2647:
1.34 noro 2648: if ( !p ) return;
1.1 noro 2649: t = BDY(p);
2650: while ( t ) {
2651: s = NEXT(t);
2652: FREENM(t);
2653: t = s;
2654: }
2655: FREEND(p);
2656: }
2657:
1.23 noro 2658: void ndv_free(NDV p)
2659: {
2660: GC_free(BDY(p));
2661: }
2662:
1.61 noro 2663: void nd_append_red(UINT *d,int i)
1.1 noro 2664: {
1.13 noro 2665: RHist m,m0;
1.1 noro 2666: int h;
2667:
1.13 noro 2668: NEWRHist(m);
1.34 noro 2669: h = ndl_hash_value(d);
1.13 noro 2670: m->index = i;
1.14 noro 2671: ndl_copy(d,DL(m));
1.1 noro 2672: NEXT(m) = nd_red[h];
2673: nd_red[h] = m;
2674: }
2675:
1.61 noro 2676: UINT *ndv_compute_bound(NDV p)
1.1 noro 2677: {
1.61 noro 2678: UINT *d1,*d2,*t;
2679: UINT u;
1.57 noro 2680: int i,j,k,l,len,ind;
1.45 noro 2681: NMV m;
1.1 noro 2682:
2683: if ( !p )
2684: return 0;
1.61 noro 2685: d1 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
2686: d2 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
1.45 noro 2687: len = LEN(p);
2688: m = BDY(p); ndl_copy(DL(m),d1); NMV_ADV(m);
2689: for ( i = 1; i < len; i++, NMV_ADV(m) ) {
1.14 noro 2690: ndl_lcm(DL(m),d1,d2);
1.1 noro 2691: t = d1; d1 = d2; d2 = t;
2692: }
1.12 noro 2693: l = nd_nvar+31;
1.61 noro 2694: t = (UINT *)MALLOC_ATOMIC(l*sizeof(UINT));
1.57 noro 2695: for ( i = nd_exporigin, ind = 0; i < nd_wpd; i++ ) {
2696: u = d1[i];
2697: k = (nd_epw-1)*nd_bpe;
2698: for ( j = 0; j < nd_epw; j++, k -= nd_bpe, ind++ )
2699: t[ind] = (u>>k)&nd_mask0;
2700: }
2701: for ( ; ind < l; ind++ ) t[ind] = 0;
1.1 noro 2702: return t;
2703: }
2704:
1.99 noro 2705: UINT *nd_compute_bound(ND p)
2706: {
2707: UINT *d1,*d2,*t;
2708: UINT u;
2709: int i,j,k,l,len,ind;
2710: NM m;
2711:
2712: if ( !p )
2713: return 0;
2714: d1 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
2715: d2 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
2716: len = LEN(p);
2717: m = BDY(p); ndl_copy(DL(m),d1); m = NEXT(m);
2718: for ( m = NEXT(m); m; m = NEXT(m) ) {
2719: ndl_lcm(DL(m),d1,d2);
2720: t = d1; d1 = d2; d2 = t;
2721: }
2722: l = nd_nvar+31;
2723: t = (UINT *)MALLOC_ATOMIC(l*sizeof(UINT));
2724: for ( i = nd_exporigin, ind = 0; i < nd_wpd; i++ ) {
2725: u = d1[i];
2726: k = (nd_epw-1)*nd_bpe;
2727: for ( j = 0; j < nd_epw; j++, k -= nd_bpe, ind++ )
2728: t[ind] = (u>>k)&nd_mask0;
2729: }
2730: for ( ; ind < l; ind++ ) t[ind] = 0;
2731: return t;
2732: }
2733:
1.48 noro 2734: int nd_get_exporigin(struct order_spec *ord)
2735: {
1.51 noro 2736: switch ( ord->id ) {
1.96 noro 2737: case 0: case 2:
1.48 noro 2738: return 1;
1.41 noro 2739: case 1:
2740: /* block order */
1.43 noro 2741: /* d[0]:weight d[1]:w0,...,d[nd_exporigin-1]:w(n-1) */
1.48 noro 2742: return ord->ord.block.length+1;
1.96 noro 2743: case 3:
2744: error("nd_get_exporigin : composite order is not supported yet.");
1.41 noro 2745: }
1.48 noro 2746: }
2747:
1.61 noro 2748: void nd_setup_parameters(int nvar,int max) {
1.79 noro 2749: int i,j,n,elen,ord_o,ord_l,l,s,wpd;
1.57 noro 2750: struct order_pair *op;
1.48 noro 2751:
1.73 noro 2752: nd_nvar = nvar;
1.79 noro 2753: if ( max ) {
1.83 noro 2754: /* XXX */
2755: if ( do_weyl ) nd_bpe = 32;
2756: else if ( max < 2 ) nd_bpe = 1;
1.79 noro 2757: else if ( max < 4 ) nd_bpe = 2;
2758: else if ( max < 8 ) nd_bpe = 3;
2759: else if ( max < 16 ) nd_bpe = 4;
2760: else if ( max < 32 ) nd_bpe = 5;
2761: else if ( max < 64 ) nd_bpe = 6;
2762: else if ( max < 256 ) nd_bpe = 8;
2763: else if ( max < 1024 ) nd_bpe = 10;
2764: else if ( max < 65536 ) nd_bpe = 16;
2765: else nd_bpe = 32;
2766: }
1.61 noro 2767: nd_epw = (sizeof(UINT)*8)/nd_bpe;
1.48 noro 2768: elen = nd_nvar/nd_epw+(nd_nvar%nd_epw?1:0);
2769: nd_exporigin = nd_get_exporigin(nd_ord);
1.79 noro 2770: wpd = nd_exporigin+elen;
2771: if ( wpd != nd_wpd ) {
2772: nd_free_private_storage();
2773: nd_wpd = wpd;
2774: }
1.1 noro 2775: if ( nd_bpe < 32 ) {
2776: nd_mask0 = (1<<nd_bpe)-1;
2777: } else {
2778: nd_mask0 = 0xffffffff;
2779: }
2780: bzero(nd_mask,sizeof(nd_mask));
2781: nd_mask1 = 0;
2782: for ( i = 0; i < nd_epw; i++ ) {
2783: nd_mask[nd_epw-i-1] = (nd_mask0<<(i*nd_bpe));
2784: nd_mask1 |= (1<<(nd_bpe-1))<<(i*nd_bpe);
2785: }
1.90 noro 2786: nmv_adv = ROUND_FOR_ALIGN(sizeof(struct oNMV)+(nd_wpd-1)*sizeof(UINT));
1.57 noro 2787: nd_epos = nd_create_epos(nd_ord);
1.43 noro 2788: nd_blockmask = nd_create_blockmask(nd_ord);
1.96 noro 2789: nd_work_vector = (int *)REALLOC(nd_work_vector,nd_nvar*sizeof(int));
1.1 noro 2790: }
2791:
1.20 noro 2792: ND_pairs nd_reconstruct(int mod,int trace,ND_pairs d)
1.1 noro 2793: {
1.37 noro 2794: int i,obpe,oadv,h;
1.71 noro 2795: static NM prev_nm_free_list;
2796: static ND_pairs prev_ndp_free_list;
1.13 noro 2797: RHist mr0,mr;
2798: RHist r;
1.37 noro 2799: RHist *old_red;
1.71 noro 2800: ND_pairs s0,s,t;
1.43 noro 2801: EPOS oepos;
1.15 noro 2802:
1.1 noro 2803: obpe = nd_bpe;
1.11 noro 2804: oadv = nmv_adv;
1.43 noro 2805: oepos = nd_epos;
1.63 noro 2806: if ( obpe < 2 ) nd_bpe = 2;
2807: else if ( obpe < 3 ) nd_bpe = 3;
1.62 noro 2808: else if ( obpe < 4 ) nd_bpe = 4;
1.63 noro 2809: else if ( obpe < 5 ) nd_bpe = 5;
1.34 noro 2810: else if ( obpe < 6 ) nd_bpe = 6;
2811: else if ( obpe < 8 ) nd_bpe = 8;
1.63 noro 2812: else if ( obpe < 10 ) nd_bpe = 10;
1.34 noro 2813: else if ( obpe < 16 ) nd_bpe = 16;
2814: else if ( obpe < 32 ) nd_bpe = 32;
2815: else error("nd_reconstruct : exponent too large");
1.5 noro 2816:
1.73 noro 2817: nd_setup_parameters(nd_nvar,0);
1.1 noro 2818: prev_nm_free_list = _nm_free_list;
2819: prev_ndp_free_list = _ndp_free_list;
2820: _nm_free_list = 0;
2821: _ndp_free_list = 0;
1.53 noro 2822: for ( i = nd_psn-1; i >= 0; i-- ) ndv_realloc(nd_ps[i],obpe,oadv,oepos);
2823: if ( trace )
2824: for ( i = nd_psn-1; i >= 0; i-- )
2825: ndv_realloc(nd_ps_trace[i],obpe,oadv,oepos);
1.1 noro 2826: s0 = 0;
2827: for ( t = d; t; t = NEXT(t) ) {
2828: NEXTND_pairs(s0,s);
2829: s->i1 = t->i1;
2830: s->i2 = t->i2;
1.14 noro 2831: SG(s) = SG(t);
1.61 noro 2832: ndl_reconstruct(LCM(t),LCM(s),obpe,oepos);
1.1 noro 2833: }
1.37 noro 2834:
2835: old_red = (RHist *)ALLOCA(REDTAB_LEN*sizeof(RHist));
1.6 noro 2836: for ( i = 0; i < REDTAB_LEN; i++ ) {
1.37 noro 2837: old_red[i] = nd_red[i];
2838: nd_red[i] = 0;
2839: }
2840: for ( i = 0; i < REDTAB_LEN; i++ )
2841: for ( r = old_red[i]; r; r = NEXT(r) ) {
2842: NEWRHist(mr);
1.13 noro 2843: mr->index = r->index;
1.20 noro 2844: SG(mr) = SG(r);
1.61 noro 2845: ndl_reconstruct(DL(r),DL(mr),obpe,oepos);
1.37 noro 2846: h = ndl_hash_value(DL(mr));
2847: NEXT(mr) = nd_red[h];
2848: nd_red[h] = mr;
1.6 noro 2849: }
1.37 noro 2850: for ( i = 0; i < REDTAB_LEN; i++ ) old_red[i] = 0;
2851: old_red = 0;
1.11 noro 2852: for ( i = 0; i < nd_psn; i++ ) {
1.20 noro 2853: NEWRHist(r); SG(r) = SG(nd_psh[i]);
1.61 noro 2854: ndl_reconstruct(DL(nd_psh[i]),DL(r),obpe,oepos);
1.13 noro 2855: nd_psh[i] = r;
1.11 noro 2856: }
1.1 noro 2857: if ( s0 ) NEXT(s) = 0;
2858: prev_nm_free_list = 0;
2859: prev_ndp_free_list = 0;
1.71 noro 2860: #if 0
1.1 noro 2861: GC_gcollect();
1.71 noro 2862: #endif
1.1 noro 2863: return s0;
2864: }
2865:
1.61 noro 2866: void ndl_reconstruct(UINT *d,UINT *r,int obpe,EPOS oepos)
1.1 noro 2867: {
1.57 noro 2868: int n,i,ei,oepw,omask0,j,s,ord_l,l;
1.43 noro 2869: struct order_pair *op;
1.1 noro 2870:
2871: n = nd_nvar;
1.61 noro 2872: oepw = (sizeof(UINT)*8)/obpe;
1.43 noro 2873: omask0 = (1<<obpe)-1;
1.34 noro 2874: TD(r) = TD(d);
1.41 noro 2875: for ( i = nd_exporigin; i < nd_wpd; i++ ) r[i] = 0;
1.43 noro 2876: if ( nd_blockmask ) {
2877: l = nd_blockmask->n;
2878: op = nd_blockmask->order_pair;
2879: for ( i = 1; i < nd_exporigin; i++ )
2880: r[i] = d[i];
2881: for ( j = 0, s = 0; j < l; j++ ) {
2882: ord_l = op[j].length;
1.57 noro 2883: for ( i = 0; i < ord_l; i++, s++ ) {
2884: ei = GET_EXP_OLD(d,s);
2885: PUT_EXP(r,s,ei);
2886: }
1.1 noro 2887: }
1.43 noro 2888: } else {
1.56 noro 2889: for ( i = 0; i < n; i++ ) {
2890: ei = GET_EXP_OLD(d,i);
2891: PUT_EXP(r,i,ei);
2892: }
1.1 noro 2893: }
2894: }
1.3 noro 2895:
1.6 noro 2896: ND nd_copy(ND p)
2897: {
2898: NM m,mr,mr0;
1.41 noro 2899: int c,n;
1.6 noro 2900: ND r;
2901:
2902: if ( !p )
2903: return 0;
2904: else {
2905: for ( mr0 = 0, m = BDY(p); m; m = NEXT(m) ) {
2906: NEXTNM(mr0,mr);
1.14 noro 2907: CM(mr) = CM(m);
2908: ndl_copy(DL(m),DL(mr));
1.6 noro 2909: }
2910: NEXT(mr) = 0;
1.31 noro 2911: MKND(NV(p),mr0,LEN(p),r);
1.14 noro 2912: SG(r) = SG(p);
1.6 noro 2913: return r;
2914: }
2915: }
2916:
1.53 noro 2917: int nd_sp(int mod,int trace,ND_pairs p,ND *rp)
1.11 noro 2918: {
2919: NM m;
2920: NDV p1,p2;
2921: ND t1,t2;
1.61 noro 2922: UINT *lcm;
1.31 noro 2923: int td;
1.11 noro 2924:
1.77 noro 2925: if ( !mod && nd_demand ) {
2926: p1 = ndv_load(p->i1); p2 = ndv_load(p->i2);
1.53 noro 2927: } else {
1.77 noro 2928: if ( trace ) {
2929: p1 = nd_ps_trace[p->i1]; p2 = nd_ps_trace[p->i2];
2930: } else {
2931: p1 = nd_ps[p->i1]; p2 = nd_ps[p->i2];
2932: }
1.20 noro 2933: }
1.34 noro 2934: lcm = LCM(p);
1.11 noro 2935: NEWNM(m);
1.20 noro 2936: CQ(m) = HCQ(p2);
1.34 noro 2937: ndl_sub(lcm,HDL(p1),DL(m));
1.56 noro 2938: if ( ndl_check_bound2(p->i1,DL(m)) )
2939: return 0;
1.55 noro 2940: t1 = ndv_mul_nm(mod,m,p1);
1.71 noro 2941: if ( mod == -1 ) CM(m) = _chsgnsf(HCM(p1));
2942: else if ( mod ) CM(m) = mod-HCM(p1);
1.34 noro 2943: else chsgnq(HCQ(p1),&CQ(m));
2944: ndl_sub(lcm,HDL(p2),DL(m));
1.14 noro 2945: if ( ndl_check_bound2(p->i2,DL(m)) ) {
1.11 noro 2946: nd_free(t1);
2947: return 0;
2948: }
1.55 noro 2949: t2 = ndv_mul_nm(mod,m,p2);
1.31 noro 2950: *rp = nd_add(mod,t1,t2);
1.11 noro 2951: FREENM(m);
2952: return 1;
2953: }
2954:
1.19 noro 2955: void ndv_mul_c(int mod,NDV p,int mul)
1.11 noro 2956: {
2957: NMV m;
2958: int c,c1,len,i;
2959:
1.34 noro 2960: if ( !p ) return;
1.14 noro 2961: len = LEN(p);
1.71 noro 2962: if ( mod == -1 )
2963: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) )
2964: CM(m) = _mulsf(CM(m),mul);
2965: else
2966: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
2967: c1 = CM(m); DMAR(c1,mul,0,mod,c); CM(m) = c;
2968: }
1.11 noro 2969: }
2970:
1.16 noro 2971: void ndv_mul_c_q(NDV p,Q mul)
2972: {
2973: NMV m;
2974: Q c;
2975: int len,i;
2976:
1.34 noro 2977: if ( !p ) return;
1.16 noro 2978: len = LEN(p);
2979: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
2980: mulq(CQ(m),mul,&c); CQ(m) = c;
2981: }
2982: }
2983:
1.55 noro 2984: ND weyl_ndv_mul_nm(int mod,NM m0,NDV p) {
2985: int n2,i,j,l,n,tlen;
1.61 noro 2986: UINT *d0;
1.55 noro 2987: NM *tab,*psum;
2988: ND s,r;
2989: NM t;
2990: NMV m1;
2991:
2992: if ( !p ) return 0;
2993: n = NV(p); n2 = n>>1;
2994: d0 = DL(m0);
2995: l = LEN(p);
2996: for ( i = 0, tlen = 1; i < n2; i++ ) tlen *= (GET_EXP(d0,n2+i)+1);
2997: tab = (NM *)ALLOCA(tlen*sizeof(NM));
2998: psum = (NM *)ALLOCA(tlen*sizeof(NM));
2999: for ( i = 0; i < tlen; i++ ) psum[i] = 0;
1.56 noro 3000: m1 = (NMV)(((char *)BDY(p))+nmv_adv*(l-1));
3001: for ( i = l-1; i >= 0; i--, NMV_PREV(m1) ) {
1.55 noro 3002: /* m0(NM) * m1(NMV) => tab(NM) */
1.56 noro 3003: weyl_mul_nm_nmv(n,mod,m0,m1,tab,tlen);
1.55 noro 3004: for ( j = 0; j < tlen; j++ ) {
3005: if ( tab[j] ) {
3006: NEXT(tab[j]) = psum[j]; psum[j] = tab[j];
3007: }
3008: }
3009: }
3010: for ( i = tlen-1, r = 0; i >= 0; i-- )
3011: if ( psum[i] ) {
3012: for ( j = 0, t = psum[i]; t; t = NEXT(t), j++ );
3013: MKND(n,psum[i],j,s);
3014: r = nd_add(mod,r,s);
3015: }
1.56 noro 3016: if ( r ) SG(r) = SG(p)+TD(d0);
3017: return r;
1.55 noro 3018: }
3019:
1.56 noro 3020: /* product of monomials */
3021: /* XXX block order is not handled correctly */
3022:
1.55 noro 3023: void weyl_mul_nm_nmv(int n,int mod,NM m0,NMV m1,NM *tab,int tlen)
3024: {
1.56 noro 3025: int i,n2,j,s,curlen,homo,h,a,b,k,l,u,min;
1.61 noro 3026: UINT *d0,*d1,*d,*dt,*ctab;
1.56 noro 3027: Q *ctab_q;
3028: Q q,q1;
1.61 noro 3029: UINT c0,c1,c;
1.55 noro 3030: NM *p;
3031: NM m,t;
3032:
3033: for ( i = 0; i < tlen; i++ ) tab[i] = 0;
3034: if ( !m0 || !m1 ) return;
3035: d0 = DL(m0); d1 = DL(m1); n2 = n>>1;
3036: NEWNM(m); d = DL(m);
1.56 noro 3037: if ( mod ) {
3038: c0 = CM(m0); c1 = CM(m1); DMAR(c0,c1,0,mod,c); CM(m) = c;
3039: } else
3040: mulq(CQ(m0),CQ(m1),&CQ(m));
1.55 noro 3041: for ( i = 0; i < nd_wpd; i++ ) d[i] = 0;
3042: homo = n&1 ? 1 : 0;
3043: if ( homo ) {
3044: /* offset of h-degree */
3045: h = GET_EXP(d0,n-1)+GET_EXP(d1,n-1);
3046: PUT_EXP(DL(m),n-1,h);
3047: TD(DL(m)) = h;
1.61 noro 3048: if ( nd_blockmask ) ndl_weight_mask(DL(m));
1.55 noro 3049: }
3050: tab[0] = m;
3051: NEWNM(m); d = DL(m);
1.57 noro 3052: for ( i = 0, curlen = 1; i < n2; i++ ) {
1.55 noro 3053: a = GET_EXP(d0,i); b = GET_EXP(d1,n2+i);
3054: k = GET_EXP(d0,n2+i); l = GET_EXP(d1,i);
3055: /* xi^a*(Di^k*xi^l)*Di^b */
3056: a += l; b += k;
1.56 noro 3057: s = MUL_WEIGHT(a,i)+MUL_WEIGHT(b,n2+i);
1.55 noro 3058: if ( !k || !l ) {
3059: for ( j = 0; j < curlen; j++ )
1.56 noro 3060: if ( t = tab[j] ) {
3061: dt = DL(t);
3062: PUT_EXP(dt,i,a); PUT_EXP(dt,n2+i,b); TD(dt) += s;
1.61 noro 3063: if ( nd_blockmask ) ndl_weight_mask(dt);
1.55 noro 3064: }
3065: curlen *= k+1;
3066: continue;
3067: }
3068: min = MIN(k,l);
1.56 noro 3069: if ( mod ) {
1.61 noro 3070: ctab = (UINT *)ALLOCA((min+1)*sizeof(UINT));
1.56 noro 3071: mkwcm(k,l,mod,ctab);
3072: } else {
3073: ctab_q = (Q *)ALLOCA((min+1)*sizeof(Q));
3074: mkwc(k,l,ctab_q);
3075: }
1.57 noro 3076: for ( j = min; j >= 0; j-- ) {
1.56 noro 3077: for ( u = 0; u < nd_wpd; u++ ) d[u] = 0;
1.55 noro 3078: PUT_EXP(d,i,a-j); PUT_EXP(d,n2+i,b-j);
1.56 noro 3079: h = MUL_WEIGHT(a-j,i)+MUL_WEIGHT(b-j,n2+i);
1.55 noro 3080: if ( homo ) {
3081: TD(d) = s;
1.56 noro 3082: PUT_EXP(d,n-1,s-h);
1.55 noro 3083: } else TD(d) = h;
1.61 noro 3084: if ( nd_blockmask ) ndl_weight_mask(d);
1.56 noro 3085: if ( mod ) c = ctab[j];
3086: else q = ctab_q[j];
1.57 noro 3087: p = tab+curlen*j;
3088: if ( j == 0 ) {
3089: for ( u = 0; u < curlen; u++, p++ ) {
3090: if ( tab[u] ) {
3091: ndl_addto(DL(tab[u]),d);
3092: if ( mod ) {
3093: c0 = CM(tab[u]); DMAR(c0,c,0,mod,c1); CM(tab[u]) = c1;
3094: } else {
3095: mulq(CQ(tab[u]),q,&q1); CQ(tab[u]) = q1;
3096: }
3097: }
1.56 noro 3098: }
1.57 noro 3099: } else {
3100: for ( u = 0; u < curlen; u++, p++ ) {
3101: if ( tab[u] ) {
3102: NEWNM(t);
3103: ndl_add(DL(tab[u]),d,DL(t));
3104: if ( mod ) {
3105: c0 = CM(tab[u]); DMAR(c0,c,0,mod,c1); CM(t) = c1;
3106: } else
3107: mulq(CQ(tab[u]),q,&CQ(t));
3108: *p = t;
3109: }
1.55 noro 3110: }
3111: }
3112: }
3113: curlen *= k+1;
3114: }
3115: FREENM(m);
3116: }
3117:
1.63 noro 3118: ND ndv_mul_nm_symbolic(NM m0,NDV p)
3119: {
3120: NM mr,mr0;
3121: NMV m;
3122: UINT *d,*dt,*dm;
3123: int c,n,td,i,c1,c2,len;
3124: Q q;
3125: ND r;
3126:
3127: if ( !p ) return 0;
3128: else {
3129: n = NV(p); m = BDY(p);
3130: d = DL(m0);
3131: len = LEN(p);
3132: mr0 = 0;
3133: td = TD(d);
3134: c = CM(m0);
3135: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
3136: NEXTNM(mr0,mr);
3137: CM(mr) = 1;
3138: ndl_add(DL(m),d,DL(mr));
3139: }
3140: NEXT(mr) = 0;
3141: MKND(NV(p),mr0,len,r);
3142: SG(r) = SG(p) + TD(d);
3143: return r;
3144: }
3145: }
3146:
1.55 noro 3147: ND ndv_mul_nm(int mod,NM m0,NDV p)
1.9 noro 3148: {
3149: NM mr,mr0;
3150: NMV m;
1.61 noro 3151: UINT *d,*dt,*dm;
1.9 noro 3152: int c,n,td,i,c1,c2,len;
1.16 noro 3153: Q q;
1.9 noro 3154: ND r;
3155:
1.34 noro 3156: if ( !p ) return 0;
1.55 noro 3157: else if ( do_weyl )
1.71 noro 3158: if ( mod == -1 )
3159: error("ndv_mul_nm : not implemented (weyl)");
3160: else
3161: return weyl_ndv_mul_nm(mod,m0,p);
1.9 noro 3162: else {
3163: n = NV(p); m = BDY(p);
1.34 noro 3164: d = DL(m0);
1.14 noro 3165: len = LEN(p);
1.9 noro 3166: mr0 = 0;
1.34 noro 3167: td = TD(d);
1.71 noro 3168: if ( mod == -1 ) {
3169: c = CM(m0);
3170: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
3171: NEXTNM(mr0,mr);
3172: CM(mr) = _mulsf(CM(m),c);
3173: ndl_add(DL(m),d,DL(mr));
3174: }
3175: } else if ( mod ) {
1.16 noro 3176: c = CM(m0);
3177: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
3178: NEXTNM(mr0,mr);
3179: c1 = CM(m);
1.19 noro 3180: DMAR(c1,c,0,mod,c2);
1.16 noro 3181: CM(mr) = c2;
3182: ndl_add(DL(m),d,DL(mr));
3183: }
3184: } else {
3185: q = CQ(m0);
3186: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
3187: NEXTNM(mr0,mr);
3188: mulq(CQ(m),q,&CQ(mr));
3189: ndl_add(DL(m),d,DL(mr));
3190: }
1.4 noro 3191: }
1.9 noro 3192: NEXT(mr) = 0;
1.31 noro 3193: MKND(NV(p),mr0,len,r);
1.34 noro 3194: SG(r) = SG(p) + TD(d);
1.9 noro 3195: return r;
1.4 noro 3196: }
3197: }
3198:
1.102 ! noro 3199: ND nd_quo(int mod,ND p,NDV d)
1.99 noro 3200: {
3201: NM mq0,mq;
1.102 ! noro 3202: NMV tm;
1.99 noro 3203: Q q;
1.102 ! noro 3204: int i,nv,sg,c,c1,c2;
1.99 noro 3205: ND t,r;
3206:
3207: if ( !p ) return 0;
3208: else {
3209: nv = NV(p);
3210: sg = SG(p);
3211: mq0 = 0;
1.102 ! noro 3212: tm = (NMV)ALLOCA(nmv_adv);
1.99 noro 3213: while ( p ) {
3214: NEXTNM(mq0,mq);
3215: ndl_sub(HDL(p),HDL(d),DL(mq));
1.102 ! noro 3216: ndl_copy(DL(mq),DL(tm));
! 3217: if ( mod ) {
! 3218: c1 = invm(HCM(d),mod); c2 = HCM(p);
! 3219: DMAR(c1,c2,0,mod,c); CM(mq) = c;
! 3220: CM(tm) = mod-c;
! 3221: } else {
! 3222: divq(HCQ(p),HCQ(d),&CQ(mq));
! 3223: chsgnq(CQ(mq),&CQ(tm));
! 3224: }
! 3225: t = ndv_mul_nmv_trunc(mod,tm,d,HDL(d));
! 3226: p = nd_add(mod,p,t);
1.99 noro 3227: }
3228: NEXT(mq) = 0;
3229: for ( i = 0, mq = mq0; mq; mq = NEXT(mq), i++ );
3230: MKND(nv,mq0,i,r);
3231: /* XXX */
3232: SG(r) = sg-SG(d);
3233: return r;
3234: }
3235: }
3236:
1.43 noro 3237: void ndv_realloc(NDV p,int obpe,int oadv,EPOS oepos)
1.11 noro 3238: {
1.13 noro 3239: NMV m,mr,mr0,t;
3240: int len,i,k;
1.11 noro 3241:
1.61 noro 3242: if ( !p ) return;
3243: m = BDY(p); len = LEN(p);
3244: mr0 = nmv_adv>oadv?(NMV)REALLOC(BDY(p),len*nmv_adv):BDY(p);
3245: m = (NMV)((char *)mr0+(len-1)*oadv);
3246: mr = (NMV)((char *)mr0+(len-1)*nmv_adv);
3247: t = (NMV)ALLOCA(nmv_adv);
3248: for ( i = 0; i < len; i++, NMV_OPREV(m), NMV_PREV(mr) ) {
3249: CQ(t) = CQ(m);
3250: for ( k = 0; k < nd_wpd; k++ ) DL(t)[k] = 0;
3251: ndl_reconstruct(DL(m),DL(t),obpe,oepos);
3252: CQ(mr) = CQ(t);
3253: ndl_copy(DL(t),DL(mr));
3254: }
3255: BDY(p) = mr0;
3256: }
3257:
3258: NDV ndv_dup_realloc(NDV p,int obpe,int oadv,EPOS oepos)
3259: {
3260: NMV m,mr,mr0;
3261: int len,i;
3262: NDV r;
1.11 noro 3263:
1.61 noro 3264: if ( !p ) return 0;
3265: m = BDY(p); len = LEN(p);
3266: mr0 = mr = (NMV)MALLOC(len*nmv_adv);
3267: for ( i = 0; i < len; i++, NMV_OADV(m), NMV_ADV(mr) ) {
3268: ndl_zero(DL(mr));
3269: ndl_reconstruct(DL(m),DL(mr),obpe,oepos);
3270: CQ(mr) = CQ(m);
1.11 noro 3271: }
1.61 noro 3272: MKNDV(NV(p),mr0,len,r);
3273: SG(r) = SG(p);
3274: return r;
1.11 noro 3275: }
3276:
1.61 noro 3277: /* duplicate p */
3278:
3279: NDV ndv_dup(int mod,NDV p)
1.3 noro 3280: {
3281: NDV d;
1.61 noro 3282: NMV t,m,m0;
1.3 noro 3283: int i,len;
3284:
1.34 noro 3285: if ( !p ) return 0;
1.31 noro 3286: len = LEN(p);
1.34 noro 3287: m0 = m = (NMV)(mod?MALLOC_ATOMIC(len*nmv_adv):MALLOC(len*nmv_adv));
1.61 noro 3288: for ( t = BDY(p), i = 0; i < len; i++, NMV_ADV(t), NMV_ADV(m) ) {
1.14 noro 3289: ndl_copy(DL(t),DL(m));
1.16 noro 3290: CQ(m) = CQ(t);
1.3 noro 3291: }
3292: MKNDV(NV(p),m0,len,d);
1.23 noro 3293: SG(d) = SG(p);
3294: return d;
3295: }
3296:
1.63 noro 3297: ND nd_dup(ND p)
3298: {
3299: ND d;
3300: NM t,m,m0;
3301:
3302: if ( !p ) return 0;
3303: for ( m0 = 0, t = BDY(p); t; t = NEXT(t) ) {
3304: NEXTNM(m0,m);
3305: ndl_copy(DL(t),DL(m));
3306: CQ(m) = CQ(t);
3307: }
3308: if ( m0 ) NEXT(m) = 0;
3309: MKND(NV(p),m0,LEN(p),d);
3310: SG(d) = SG(p);
3311: return d;
3312: }
3313:
1.61 noro 3314: /* XXX if p->len == 0 then it represents 0 */
3315:
3316: void ndv_mod(int mod,NDV p)
3317: {
3318: NMV t,d;
3319: int r;
3320: int i,len,dlen;
1.71 noro 3321: Obj gfs;
1.61 noro 3322:
3323: if ( !p ) return;
3324: len = LEN(p);
3325: dlen = 0;
1.71 noro 3326: if ( mod == -1 )
3327: for ( t = d = BDY(p), i = 0; i < len; i++, NMV_ADV(t) ) {
3328: simp_ff((Obj)CP(t),&gfs);
3329: r = FTOIF(CONT((GFS)gfs));
1.61 noro 3330: CM(d) = r;
3331: ndl_copy(DL(t),DL(d));
3332: NMV_ADV(d);
3333: dlen++;
3334: }
1.71 noro 3335: else
3336: for ( t = d = BDY(p), i = 0; i < len; i++, NMV_ADV(t) ) {
3337: r = rem(NM(CQ(t)),mod);
3338: if ( r ) {
3339: if ( SGN(CQ(t)) < 0 )
3340: r = mod-r;
3341: CM(d) = r;
3342: ndl_copy(DL(t),DL(d));
3343: NMV_ADV(d);
3344: dlen++;
3345: }
3346: }
1.61 noro 3347: LEN(p) = dlen;
3348: }
3349:
3350: NDV ptondv(VL vl,VL dvl,P p)
3351: {
3352: ND nd;
3353:
3354: nd = ptond(vl,dvl,p);
3355: return ndtondv(0,nd);
3356: }
3357:
3358: ND ptond(VL vl,VL dvl,P p)
1.23 noro 3359: {
1.61 noro 3360: int n,i,j,k,e;
3361: VL tvl;
3362: V v;
3363: DCP dc;
3364: DCP *w;
3365: ND r,s,t,u;
3366: P x;
3367: int c;
3368: UINT *d;
1.23 noro 3369: NM m,m0;
1.61 noro 3370:
3371: if ( !p )
3372: return 0;
3373: else if ( NUM(p) ) {
3374: NEWNM(m);
3375: ndl_zero(DL(m));
3376: CQ(m) = (Q)p;
3377: NEXT(m) = 0;
3378: MKND(nd_nvar,m,1,r);
3379: SG(r) = 0;
3380: return r;
3381: } else {
3382: for ( dc = DC(p), k = 0; dc; dc = NEXT(dc), k++ );
3383: w = (DCP *)ALLOCA(k*sizeof(DCP));
3384: for ( dc = DC(p), j = 0; j < k; dc = NEXT(dc), j++ ) w[j] = dc;
3385: for ( i = 0, tvl = dvl, v = VR(p);
1.98 noro 3386: tvl && tvl->v != v; tvl = NEXT(tvl), i++ );
1.61 noro 3387: if ( !tvl ) {
3388: for ( j = k-1, s = 0, MKV(v,x); j >= 0; j-- ) {
3389: t = ptond(vl,dvl,COEF(w[j]));
3390: pwrp(vl,x,DEG(w[j]),&p);
3391: nd_mul_c_p(CO,t,p); s = nd_add(0,s,t);
3392: }
3393: return s;
3394: } else {
3395: NEWNM(m0); d = DL(m0);
3396: for ( j = k-1, s = 0; j >= 0; j-- ) {
3397: ndl_zero(d); e = QTOS(DEG(w[j])); PUT_EXP(d,i,e);
3398: TD(d) = MUL_WEIGHT(e,i);
3399: if ( nd_blockmask) ndl_weight_mask(d);
3400: t = ptond(vl,dvl,COEF(w[j]));
3401: for ( m = BDY(t); m; m = NEXT(m) )
3402: ndl_addto(DL(m),d);
3403: SG(t) += TD(d);
3404: s = nd_add(0,s,t);
3405: }
3406: FREENM(m0);
3407: return s;
3408: }
3409: }
3410: }
3411:
3412: P ndvtop(int mod,VL vl,VL dvl,NDV p)
3413: {
3414: VL tvl;
3415: int len,n,j,i,e;
3416: NMV m;
3417: Q q;
3418: P c;
3419: UINT *d;
3420: P s,r,u,t,w;
1.71 noro 3421: GFS gfs;
1.23 noro 3422:
1.34 noro 3423: if ( !p ) return 0;
1.61 noro 3424: else {
3425: len = LEN(p);
3426: n = NV(p);
3427: m = (NMV)(((char *)BDY(p))+nmv_adv*(len-1));
3428: for ( j = len-1, s = 0; j >= 0; j--, NMV_PREV(m) ) {
1.71 noro 3429: if ( mod == -1 ) {
3430: e = IFTOF(CM(m)); MKGFS(e,gfs); c = (P)gfs;
3431: } else if ( mod ) {
1.61 noro 3432: STOQ(CM(m),q); c = (P)q;
3433: } else
3434: c = CP(m);
3435: d = DL(m);
3436: for ( i = 0, t = c, tvl = dvl; i < n; tvl = NEXT(tvl), i++ ) {
3437: MKV(tvl->v,r); e = GET_EXP(d,i); STOQ(e,q);
3438: pwrp(vl,r,q,&u); mulp(vl,t,u,&w); t = w;
3439: }
3440: addp(vl,s,t,&u); s = u;
3441: }
3442: return s;
1.23 noro 3443: }
1.3 noro 3444: }
3445:
1.61 noro 3446: NDV ndtondv(int mod,ND p)
1.11 noro 3447: {
3448: NDV d;
1.61 noro 3449: NMV m,m0;
3450: NM t;
3451: int i,len;
1.11 noro 3452:
1.34 noro 3453: if ( !p ) return 0;
1.61 noro 3454: len = LEN(p);
1.74 noro 3455: if ( mod )
3456: m0 = m = (NMV)GC_malloc_atomic_ignore_off_page(len*nmv_adv);
3457: else
3458: m0 = m = MALLOC(len*nmv_adv);
3459: ndv_alloc += nmv_adv*len;
1.61 noro 3460: for ( t = BDY(p), i = 0; t; t = NEXT(t), i++, NMV_ADV(m) ) {
3461: ndl_copy(DL(t),DL(m));
3462: CQ(m) = CQ(t);
1.11 noro 3463: }
1.61 noro 3464: MKNDV(NV(p),m0,len,d);
1.14 noro 3465: SG(d) = SG(p);
1.11 noro 3466: return d;
3467: }
3468:
1.61 noro 3469: ND ndvtond(int mod,NDV p)
1.11 noro 3470: {
1.61 noro 3471: ND d;
3472: NM m,m0;
1.11 noro 3473: NMV t;
1.61 noro 3474: int i,len;
1.11 noro 3475:
1.34 noro 3476: if ( !p ) return 0;
1.11 noro 3477: m0 = 0;
1.61 noro 3478: len = p->len;
3479: for ( t = BDY(p), i = 0; i < len; NMV_ADV(t), i++ ) {
3480: NEXTNM(m0,m);
3481: ndl_copy(DL(t),DL(m));
3482: CQ(m) = CQ(t);
1.11 noro 3483: }
3484: NEXT(m) = 0;
1.61 noro 3485: MKND(NV(p),m0,len,d);
1.14 noro 3486: SG(d) = SG(p);
1.11 noro 3487: return d;
3488: }
3489:
1.3 noro 3490: void ndv_print(NDV p)
3491: {
3492: NMV m;
3493: int i,len;
3494:
1.34 noro 3495: if ( !p ) printf("0\n");
1.3 noro 3496: else {
1.14 noro 3497: len = LEN(p);
1.3 noro 3498: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
1.71 noro 3499: if ( CM(m) & 0x80000000 ) printf("+@_%d*",IFTOF(CM(m)));
3500: else printf("+%d*",CM(m));
1.16 noro 3501: ndl_print(DL(m));
3502: }
3503: printf("\n");
3504: }
3505: }
3506:
3507: void ndv_print_q(NDV p)
3508: {
3509: NMV m;
3510: int i,len;
3511:
1.34 noro 3512: if ( !p ) printf("0\n");
1.16 noro 3513: else {
3514: len = LEN(p);
3515: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
3516: printf("+");
1.74 noro 3517: printexpr(CO,(Obj)CQ(m));
1.16 noro 3518: printf("*");
1.14 noro 3519: ndl_print(DL(m));
1.3 noro 3520: }
3521: printf("\n");
3522: }
1.25 noro 3523: }
3524:
1.61 noro 3525: NODE ndv_reducebase(NODE x)
1.27 noro 3526: {
3527: int len,i,j;
3528: NDV *w;
3529: NODE t,t0;
3530:
3531: len = length(x);
3532: w = (NDV *)ALLOCA(len*sizeof(NDV));
3533: for ( i = 0, t = x; i < len; i++, t = NEXT(t) ) w[i] = BDY(t);
3534: for ( i = 0; i < len; i++ ) {
3535: for ( j = 0; j < i; j++ ) {
3536: if ( w[i] && w[j] )
3537: if ( ndl_reducible(HDL(w[i]),HDL(w[j])) ) w[i] = 0;
3538: else if ( ndl_reducible(HDL(w[j]),HDL(w[i])) ) w[j] = 0;
3539: }
3540: }
3541: for ( i = len-1, t0 = 0; i >= 0; i-- ) {
3542: if ( w[i] ) { NEXTNODE(t0,t); BDY(t) = (pointer)w[i]; }
3543: }
3544: NEXT(t) = 0; x = t0;
3545: return x;
1.11 noro 3546: }
1.32 noro 3547:
1.43 noro 3548: /* XXX incomplete */
3549:
1.32 noro 3550: void nd_init_ord(struct order_spec *ord)
3551: {
1.43 noro 3552: switch ( ord->id ) {
1.32 noro 3553: case 0:
1.43 noro 3554: switch ( ord->ord.simple ) {
3555: case 0:
3556: nd_dcomp = 1;
3557: nd_isrlex = 1;
3558: break;
3559: case 1:
3560: nd_dcomp = 1;
3561: nd_isrlex = 0;
3562: break;
3563: case 2:
3564: nd_dcomp = 0;
3565: nd_isrlex = 0;
1.45 noro 3566: ndl_compare_function = ndl_lex_compare;
1.58 noro 3567: break;
3568: case 11:
3569: /* XXX */
3570: nd_dcomp = 0;
3571: nd_isrlex = 1;
3572: ndl_compare_function = ndl_ww_lex_compare;
1.43 noro 3573: break;
3574: default:
3575: error("nd_gr : unsupported order");
3576: }
1.32 noro 3577: break;
3578: case 1:
1.96 noro 3579: /* block order */
1.43 noro 3580: /* XXX */
3581: nd_dcomp = -1;
1.32 noro 3582: nd_isrlex = 0;
1.45 noro 3583: ndl_compare_function = ndl_block_compare;
1.34 noro 3584: break;
1.43 noro 3585: case 2:
1.96 noro 3586: /* matrix order */
3587: /* XXX */
3588: nd_dcomp = -1;
3589: nd_isrlex = 0;
3590: nd_matrix_len = ord->ord.matrix.row;
3591: nd_matrix = ord->ord.matrix.matrix;
3592: ndl_compare_function = ndl_matrix_compare;
3593: break;
3594: case 3:
1.97 noro 3595: /* composite order */
3596: nd_dcomp = -1;
3597: nd_isrlex = 0;
3598: nd_worb_len = ord->ord.composite.length;
3599: nd_worb = ord->ord.composite.w_or_b;
3600: ndl_compare_function = ndl_composite_compare;
1.32 noro 3601: break;
3602: }
1.41 noro 3603: nd_ord = ord;
1.32 noro 3604: }
3605:
1.43 noro 3606: BlockMask nd_create_blockmask(struct order_spec *ord)
3607: {
3608: int n,i,j,s,l;
1.61 noro 3609: UINT *t;
1.43 noro 3610: BlockMask bm;
3611:
1.96 noro 3612: /* we only create mask table for block order */
3613: if ( ord->id != 1 )
1.43 noro 3614: return 0;
3615: n = ord->ord.block.length;
3616: bm = (BlockMask)MALLOC(sizeof(struct oBlockMask));
3617: bm->n = n;
3618: bm->order_pair = ord->ord.block.order_pair;
1.61 noro 3619: bm->mask = (UINT **)MALLOC(n*sizeof(UINT *));
1.43 noro 3620: for ( i = 0, s = 0; i < n; i++ ) {
1.61 noro 3621: bm->mask[i] = t = (UINT *)MALLOC_ATOMIC(nd_wpd*sizeof(UINT));
1.43 noro 3622: for ( j = 0; j < nd_wpd; j++ ) t[j] = 0;
3623: l = bm->order_pair[i].length;
3624: for ( j = 0; j < l; j++, s++ ) PUT_EXP(t,s,nd_mask0);
3625: }
3626: return bm;
1.57 noro 3627: }
3628:
3629: EPOS nd_create_epos(struct order_spec *ord)
3630: {
3631: int i,j,l,s,ord_l,ord_o;
3632: EPOS epos;
3633: struct order_pair *op;
3634:
3635: epos = (EPOS)MALLOC_ATOMIC(nd_nvar*sizeof(struct oEPOS));
3636: switch ( ord->id ) {
3637: case 0:
3638: if ( nd_isrlex ) {
3639: for ( i = 0; i < nd_nvar; i++ ) {
3640: epos[i].i = nd_exporigin + (nd_nvar-1-i)/nd_epw;
3641: epos[i].s = (nd_epw-((nd_nvar-1-i)%nd_epw)-1)*nd_bpe;
3642: }
3643: } else {
3644: for ( i = 0; i < nd_nvar; i++ ) {
3645: epos[i].i = nd_exporigin + i/nd_epw;
3646: epos[i].s = (nd_epw-(i%nd_epw)-1)*nd_bpe;
3647: }
3648: }
3649: break;
3650: case 1:
3651: /* block order */
3652: l = ord->ord.block.length;
3653: op = ord->ord.block.order_pair;
3654: for ( j = 0, s = 0; j < l; j++ ) {
3655: ord_o = op[j].order;
3656: ord_l = op[j].length;
3657: if ( !ord_o )
3658: for ( i = 0; i < ord_l; i++ ) {
3659: epos[s+i].i = nd_exporigin + (s+ord_l-i-1)/nd_epw;
3660: epos[s+i].s = (nd_epw-((s+ord_l-i-1)%nd_epw)-1)*nd_bpe;
3661: }
3662: else
3663: for ( i = 0; i < ord_l; i++ ) {
3664: epos[s+i].i = nd_exporigin + (s+i)/nd_epw;
3665: epos[s+i].s = (nd_epw-((s+i)%nd_epw)-1)*nd_bpe;
3666: }
3667: s += ord_l;
3668: }
3669: break;
3670: case 2:
1.96 noro 3671: /* matrix order */
3672: case 3:
3673: /* composite order */
3674: for ( i = 0; i < nd_nvar; i++ ) {
3675: epos[i].i = nd_exporigin + i/nd_epw;
3676: epos[i].s = (nd_epw-(i%nd_epw)-1)*nd_bpe;
3677: }
3678: break;
1.57 noro 3679: }
3680: return epos;
1.43 noro 3681: }
1.59 noro 3682:
3683: /* external interface */
3684:
3685: void nd_nf_p(P f,LIST g,LIST v,int m,struct order_spec *ord,P *rp)
3686: {
1.61 noro 3687: NODE t,in0,in;
1.59 noro 3688: ND nd,nf;
1.61 noro 3689: NDV ndv;
3690: VL vv,tv;
3691: int stat,nvar,max,e;
1.69 noro 3692: union oNDC dn;
1.59 noro 3693:
3694: pltovl(v,&vv);
1.74 noro 3695: for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++ );
1.61 noro 3696:
3697: /* get the degree bound */
3698: for ( t = BDY(g), max = 0; t; t = NEXT(t) )
3699: for ( tv = vv; tv; tv = NEXT(tv) ) {
3700: e = getdeg(tv->v,(P)BDY(t));
3701: max = MAX(e,max);
3702: }
3703: for ( tv = vv; tv; tv = NEXT(tv) ) {
3704: e = getdeg(tv->v,f);
3705: max = MAX(e,max);
3706: }
3707:
1.59 noro 3708: nd_init_ord(ord);
1.61 noro 3709: nd_setup_parameters(nvar,max);
3710:
3711: /* conversion to ndv */
3712: for ( in0 = 0, t = BDY(g); t; t = NEXT(t) ) {
3713: NEXTNODE(in0,in);
3714: BDY(in) = (pointer)ptondv(CO,vv,(P)BDY(t));
1.69 noro 3715: if ( m ) ndv_mod(m,(NDV)BDY(in));
1.61 noro 3716: }
3717: NEXTNODE(in0,in);
3718: BDY(in) = (pointer)ptondv(CO,vv,f);
1.69 noro 3719: if ( m ) ndv_mod(m,(NDV)BDY(in));
1.61 noro 3720: NEXT(in) = 0;
3721:
3722: ndv_setup(m,0,in0);
1.59 noro 3723: nd_psn--;
3724: nd_scale=2;
3725: while ( 1 ) {
3726: nd = (pointer)ndvtond(m,nd_ps[nd_psn]);
1.69 noro 3727: stat = nd_nf(m,nd,nd_ps,1,0,&nf);
1.59 noro 3728: if ( !stat ) {
3729: nd_psn++;
3730: nd_reconstruct(m,0,0);
3731: nd_psn--;
3732: } else
3733: break;
3734: }
1.61 noro 3735: *rp = ndvtop(m,CO,vv,ndtondv(m,nf));
1.63 noro 3736: }
3737:
3738: int nd_to_vect(int mod,UINT *s0,int n,ND d,UINT *r)
3739: {
3740: NM m;
3741: UINT *t,*s;
3742: int i;
3743:
3744: for ( i = 0; i < n; i++ ) r[i] = 0;
3745: for ( i = 0, s = s0, m = BDY(d); m; m = NEXT(m) ) {
3746: t = DL(m);
3747: for ( ; !ndl_equal(t,s); s += nd_wpd, i++ );
3748: r[i] = CM(m);
3749: }
3750: for ( i = 0; !r[i]; i++ );
3751: return i;
3752: }
3753:
1.74 noro 3754: int ndv_to_vect(int mod,UINT *s0,int n,NDV d,UINT *r)
3755: {
3756: NMV m;
3757: UINT *t,*s;
3758: int i,j,len;
3759:
3760: for ( i = 0; i < n; i++ ) r[i] = 0;
3761: m = BDY(d);
3762: len = LEN(d);
3763: for ( i = j = 0, s = s0; j < len; j++, NMV_ADV(m)) {
3764: t = DL(m);
3765: for ( ; !ndl_equal(t,s); s += nd_wpd, i++ );
3766: r[i] = CM(m);
3767: }
3768: for ( i = 0; !r[i]; i++ );
3769: return i;
3770: }
3771:
1.63 noro 3772: int nm_ind_pair_to_vect(int mod,UINT *s0,int n,NM_ind_pair pair,UINT *r)
3773: {
3774: NM m;
3775: NMV mr;
3776: UINT *d,*t,*s;
3777: NDV p;
3778: int i,j,len;
3779:
3780: m = pair->mul;
3781: d = DL(m);
3782: p = nd_ps[pair->index];
3783: t = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
3784: for ( i = 0; i < n; i++ ) r[i] = 0;
3785: len = LEN(p);
3786: for ( i = j = 0, s = s0, mr = BDY(p); j < len; j++, NMV_ADV(mr) ) {
3787: ndl_add(d,DL(mr),t);
3788: for ( ; !ndl_equal(t,s); s += nd_wpd, i++ );
3789: r[i] = CM(mr);
3790: }
3791: for ( i = 0; !r[i]; i++ );
3792: return i;
3793: }
3794:
1.67 noro 3795: IndArray nm_ind_pair_to_vect_compress(int mod,UINT *s0,int n,NM_ind_pair pair)
1.64 noro 3796: {
3797: NM m;
3798: NMV mr;
3799: UINT *d,*t,*s;
3800: NDV p;
1.67 noro 3801: unsigned char *ivc;
3802: unsigned short *ivs;
1.81 noro 3803: UINT *v,*ivi,*s0v;
1.67 noro 3804: int i,j,len,prev,diff,cdiff;
3805: IndArray r;
1.64 noro 3806:
3807: m = pair->mul;
3808: d = DL(m);
3809: p = nd_ps[pair->index];
3810: len = LEN(p);
3811: t = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
1.67 noro 3812: v = (unsigned int *)ALLOCA(len*sizeof(unsigned int));
1.64 noro 3813: for ( i = j = 0, s = s0, mr = BDY(p); j < len; j++, NMV_ADV(mr) ) {
3814: ndl_add(d,DL(mr),t);
3815: for ( ; !ndl_equal(t,s); s += nd_wpd, i++ );
1.67 noro 3816: v[j] = i;
3817: }
1.81 noro 3818: r = (IndArray)MALLOC(sizeof(struct oIndArray));
1.67 noro 3819: r->head = v[0];
3820: diff = 0;
3821: for ( i = 1; i < len; i++ ) {
3822: cdiff = v[i]-v[i-1]; diff = MAX(cdiff,diff);
3823: }
3824: if ( diff < 256 ) {
3825: r->width = 1;
3826: ivc = (unsigned char *)MALLOC_ATOMIC(len*sizeof(unsigned char));
3827: r->index.c = ivc;
3828: for ( i = 1, ivc[0] = 0; i < len; i++ ) ivc[i] = v[i]-v[i-1];
3829: } else if ( diff < 65536 ) {
3830: r->width = 2;
3831: ivs = (unsigned short *)MALLOC_ATOMIC(len*sizeof(unsigned short));
3832: r->index.s = ivs;
3833: for ( i = 1, ivs[0] = 0; i < len; i++ ) ivs[i] = v[i]-v[i-1];
3834: } else {
3835: r->width = 4;
3836: ivi = (unsigned int *)MALLOC_ATOMIC(len*sizeof(unsigned int));
3837: r->index.i = ivi;
3838: for ( i = 1, ivi[0] = 0; i < len; i++ ) ivi[i] = v[i]-v[i-1];
1.64 noro 3839: }
1.67 noro 3840: return r;
1.64 noro 3841: }
3842:
1.65 noro 3843:
1.76 noro 3844: int ndv_reduce_vect(int m,UINT *svect,int col,IndArray *imat,NM_ind_pair *rp0,int nred)
1.65 noro 3845: {
1.67 noro 3846: int i,j,k,len,pos,prev;
1.66 noro 3847: UINT c,c1,c2,c3,up,lo,dmy;
1.67 noro 3848: IndArray ivect;
3849: unsigned char *ivc;
3850: unsigned short *ivs;
3851: unsigned int *ivi;
1.65 noro 3852: NDV redv;
1.67 noro 3853: NMV mr;
1.65 noro 3854: NODE rp;
1.76 noro 3855: int maxrs;
1.65 noro 3856:
1.76 noro 3857: maxrs = 0;
1.74 noro 3858: for ( i = 0; i < nred; i++ ) {
1.65 noro 3859: ivect = imat[i];
1.67 noro 3860: k = ivect->head; svect[k] %= m;
1.65 noro 3861: if ( c = svect[k] ) {
1.76 noro 3862: maxrs = MAX(maxrs,rp0[i]->sugar);
1.74 noro 3863: c = m-c; redv = nd_ps[rp0[i]->index];
1.67 noro 3864: len = LEN(redv); mr = BDY(redv);
3865: svect[k] = 0; prev = k;
3866: switch ( ivect->width ) {
3867: case 1:
3868: ivc = ivect->index.c;
3869: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3870: pos = prev+ivc[j]; c1 = CM(mr); c2 = svect[pos];
3871: prev = pos;
3872: DMA(c1,c,c2,up,lo);
3873: if ( up ) { DSAB(m,up,lo,dmy,c3); svect[pos] = c3;
3874: } else svect[pos] = lo;
3875: }
3876: break;
3877: case 2:
3878: ivs = ivect->index.s;
3879: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3880: pos = prev+ivs[j]; c1 = CM(mr); c2 = svect[pos];
3881: prev = pos;
3882: DMA(c1,c,c2,up,lo);
3883: if ( up ) { DSAB(m,up,lo,dmy,c3); svect[pos] = c3;
3884: } else svect[pos] = lo;
3885: }
3886: break;
3887: case 4:
3888: ivi = ivect->index.i;
3889: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3890: pos = prev+ivi[j]; c1 = CM(mr); c2 = svect[pos];
3891: prev = pos;
3892: DMA(c1,c,c2,up,lo);
3893: if ( up ) { DSAB(m,up,lo,dmy,c3); svect[pos] = c3;
3894: } else svect[pos] = lo;
3895: }
3896: break;
1.65 noro 3897: }
3898: }
3899: }
1.66 noro 3900: for ( i = 0; i < col; i++ )
3901: if ( svect[i] >= (UINT)m ) svect[i] %= m;
1.76 noro 3902: return maxrs;
1.65 noro 3903: }
3904:
1.76 noro 3905: int ndv_reduce_vect_sf(int m,UINT *svect,int col,IndArray *imat,NM_ind_pair *rp0,int nred)
1.72 noro 3906: {
3907: int i,j,k,len,pos,prev;
3908: UINT c,c1,c2,c3,up,lo,dmy;
3909: IndArray ivect;
3910: unsigned char *ivc;
3911: unsigned short *ivs;
3912: unsigned int *ivi;
3913: NDV redv;
3914: NMV mr;
3915: NODE rp;
1.76 noro 3916: int maxrs;
1.72 noro 3917:
1.76 noro 3918: maxrs = 0;
1.74 noro 3919: for ( i = 0; i < nred; i++ ) {
1.72 noro 3920: ivect = imat[i];
3921: k = ivect->head; svect[k] %= m;
3922: if ( c = svect[k] ) {
1.76 noro 3923: maxrs = MAX(maxrs,rp0[i]->sugar);
1.74 noro 3924: c = _chsgnsf(c); redv = nd_ps[rp0[i]->index];
1.72 noro 3925: len = LEN(redv); mr = BDY(redv);
3926: svect[k] = 0; prev = k;
3927: switch ( ivect->width ) {
3928: case 1:
3929: ivc = ivect->index.c;
3930: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3931: pos = prev+ivc[j]; prev = pos;
3932: svect[pos] = _addsf(_mulsf(CM(mr),c),svect[pos]);
3933: }
3934: break;
3935: case 2:
3936: ivs = ivect->index.s;
3937: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3938: pos = prev+ivs[j]; prev = pos;
3939: svect[pos] = _addsf(_mulsf(CM(mr),c),svect[pos]);
3940: }
3941: break;
3942: case 4:
3943: ivi = ivect->index.i;
3944: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3945: pos = prev+ivi[j]; prev = pos;
3946: svect[pos] = _addsf(_mulsf(CM(mr),c),svect[pos]);
3947: }
3948: break;
3949: }
3950: }
3951: }
1.76 noro 3952: return maxrs;
1.72 noro 3953: }
3954:
1.65 noro 3955: NDV vect_to_ndv(UINT *vect,int spcol,int col,int *rhead,UINT *s0vect)
3956: {
3957: int j,k,len;
3958: UINT *p;
3959: UINT c;
3960: NDV r;
3961: NMV mr0,mr;
3962:
3963: for ( j = 0, len = 0; j < spcol; j++ ) if ( vect[j] ) len++;
3964: if ( !len ) return 0;
3965: else {
1.74 noro 3966: mr0 = (NMV)GC_malloc_atomic_ignore_off_page(nmv_adv*len);
3967: ndv_alloc += nmv_adv*len;
1.65 noro 3968: mr = mr0;
3969: p = s0vect;
3970: for ( j = k = 0; j < col; j++, p += nd_wpd )
3971: if ( !rhead[j] ) {
3972: if ( c = vect[k++] ) {
3973: ndl_copy(p,DL(mr)); CM(mr) = c; NMV_ADV(mr);
3974: }
3975: }
3976: MKNDV(nd_nvar,mr0,len,r);
3977: return r;
3978: }
3979: }
3980:
1.74 noro 3981: int nd_sp_f4(int m,ND_pairs l,PGeoBucket bucket)
1.65 noro 3982: {
3983: ND_pairs t;
3984: NODE sp0,sp;
3985: int stat;
3986: ND spol;
3987:
3988: for ( t = l; t; t = NEXT(t) ) {
3989: stat = nd_sp(m,0,t,&spol);
3990: if ( !stat ) return 0;
3991: if ( spol ) {
3992: add_pbucket_symbolic(bucket,spol);
3993: }
3994: }
1.68 noro 3995: return 1;
1.65 noro 3996: }
3997:
3998: int nd_symbolic_preproc(PGeoBucket bucket,UINT **s0vect,NODE *r)
3999: {
4000: NODE rp0,rp;
4001: NM mul,head,s0,s;
1.76 noro 4002: int index,col,i,sugar;
1.65 noro 4003: RHist h;
4004: UINT *s0v,*p;
4005: NM_ind_pair pair;
4006: ND red;
4007:
4008: s0 = 0; rp0 = 0; col = 0;
4009: while ( 1 ) {
4010: head = remove_head_pbucket_symbolic(bucket);
4011: if ( !head ) break;
4012: if ( !s0 ) s0 = head;
4013: else NEXT(s) = head;
4014: s = head;
4015: index = ndl_find_reducer(DL(head));
4016: if ( index >= 0 ) {
4017: h = nd_psh[index];
4018: NEWNM(mul);
4019: ndl_sub(DL(head),DL(h),DL(mul));
4020: if ( ndl_check_bound2(index,DL(mul)) ) return 0;
1.76 noro 4021: sugar = TD(DL(mul))+SG(nd_ps[index]);
4022: MKNM_ind_pair(pair,mul,index,sugar);
1.65 noro 4023: red = ndv_mul_nm_symbolic(mul,nd_ps[index]);
4024: add_pbucket_symbolic(bucket,nd_remove_head(red));
4025: NEXTNODE(rp0,rp); BDY(rp) = (pointer)pair;
4026: }
4027: col++;
4028: }
1.72 noro 4029: if ( rp0 ) NEXT(rp) = 0;
4030: NEXT(s) = 0;
1.65 noro 4031: s0v = (UINT *)MALLOC_ATOMIC(col*nd_wpd*sizeof(UINT));
4032: for ( i = 0, p = s0v, s = s0; i < col;
4033: i++, p += nd_wpd, s = NEXT(s) ) ndl_copy(DL(s),p);
4034: *s0vect = s0v;
4035: *r = rp0;
4036: return col;
4037: }
4038:
1.69 noro 4039: NODE nd_f4(int m)
4040: {
4041: int i,nh,stat,index;
4042: NODE r,g;
4043: ND_pairs d,l,t;
4044: ND spol,red;
4045: NDV nf,redv;
4046: NM s0,s;
1.84 noro 4047: NODE rp0,srp0,nflist;
1.69 noro 4048: int nsp,nred,col,rank,len,k,j,a;
4049: UINT c;
1.74 noro 4050: UINT **spmat;
1.69 noro 4051: UINT *s0vect,*svect,*p,*v;
4052: int *colstat;
4053: IndArray *imat;
4054: int *rhead;
4055: int spcol,sprow;
4056: int sugar;
4057: PGeoBucket bucket;
4058: struct oEGT eg0,eg1,eg_f4;
4059:
4060: if ( !m )
4061: error("nd_f4 : not implemented");
1.74 noro 4062: ndv_alloc = 0;
1.69 noro 4063: g = 0; d = 0;
4064: for ( i = 0; i < nd_psn; i++ ) {
4065: d = update_pairs(d,g,i);
4066: g = update_base(g,i);
4067: }
4068: while ( d ) {
4069: get_eg(&eg0);
4070: l = nd_minsugarp(d,&d);
4071: sugar = SG(l);
4072: bucket = create_pbucket();
1.74 noro 4073: stat = nd_sp_f4(m,l,bucket);
1.69 noro 4074: if ( !stat ) {
4075: for ( t = l; NEXT(t); t = NEXT(t) );
4076: NEXT(t) = d; d = l;
4077: d = nd_reconstruct(m,0,d);
4078: continue;
4079: }
1.86 noro 4080: if ( bucket->m < 0 ) continue;
1.69 noro 4081: col = nd_symbolic_preproc(bucket,&s0vect,&rp0);
4082: if ( !col ) {
4083: for ( t = l; NEXT(t); t = NEXT(t) );
4084: NEXT(t) = d; d = l;
4085: d = nd_reconstruct(m,0,d);
4086: continue;
4087: }
4088: get_eg(&eg1); init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg1);
1.74 noro 4089: if ( DP_Print )
4090: fprintf(asir_out,"sugar=%d,symb=%fsec,",
4091: sugar,eg_f4.exectime+eg_f4.gctime);
4092: if ( 1 )
4093: nflist = nd_f4_red(m,l,s0vect,col,rp0);
4094: else
4095: nflist = nd_f4_red_dist(m,l,s0vect,col,rp0);
1.69 noro 4096: /* adding new bases */
1.74 noro 4097: for ( r = nflist; r; r = NEXT(r) ) {
4098: nf = (NDV)BDY(r);
1.69 noro 4099: ndv_removecont(m,nf);
1.77 noro 4100: nh = ndv_newps(m,nf,0);
1.69 noro 4101: d = update_pairs(d,g,nh);
4102: g = update_base(g,nh);
4103: }
4104: }
4105: for ( r = g; r; r = NEXT(r) ) BDY(r) = (pointer)nd_ps[(int)BDY(r)];
1.74 noro 4106: fprintf(asir_out,"ndv_alloc=%d\n",ndv_alloc);
1.69 noro 4107: return g;
4108: }
1.74 noro 4109:
4110: NODE nd_f4_red(int m,ND_pairs sp0,UINT *s0vect,int col,NODE rp0)
1.63 noro 4111: {
1.67 noro 4112: IndArray *imat;
1.74 noro 4113: int nsp,nred,spcol,sprow,a;
1.65 noro 4114: int *rhead;
1.74 noro 4115: int i,j,k,l,rank;
4116: NODE rp,r0,r;
4117: ND_pairs sp;
4118: ND spol;
4119: int **spmat;
4120: UINT *svect,*v;
4121: int *colstat;
1.80 noro 4122: struct oEGT eg0,eg1,eg2,eg_f4,eg_f4_1,eg_f4_2;
1.74 noro 4123: NM_ind_pair *rvect;
1.76 noro 4124: int maxrs;
4125: int *spsugar;
1.63 noro 4126:
1.74 noro 4127: get_eg(&eg0);
4128: for ( sp = sp0, nsp = 0; sp; sp = NEXT(sp), nsp++ );
4129: nred = length(rp0); spcol = col-nred;
4130: imat = (IndArray *)ALLOCA(nred*sizeof(IndArray));
4131: rhead = (int *)ALLOCA(col*sizeof(int));
4132: for ( i = 0; i < col; i++ ) rhead[i] = 0;
1.63 noro 4133:
1.74 noro 4134: /* construction of index arrays */
4135: rvect = (NM_ind_pair *)ALLOCA(nred*sizeof(NM_ind_pair));
4136: for ( rp = rp0, i = 0; rp; i++, rp = NEXT(rp) ) {
4137: rvect[i] = (NM_ind_pair)BDY(rp);
4138: imat[i] = nm_ind_pair_to_vect_compress(m,s0vect,col,rvect[i]);
4139: rhead[imat[i]->head] = 1;
4140: }
4141:
4142: /* elimination (1st step) */
4143: spmat = (int **)ALLOCA(nsp*sizeof(UINT *));
4144: svect = (UINT *)ALLOCA(col*sizeof(UINT));
1.76 noro 4145: spsugar = (int *)ALLOCA(nsp*sizeof(UINT));
1.74 noro 4146: for ( a = sprow = 0, sp = sp0; a < nsp; a++, sp = NEXT(sp) ) {
4147: nd_sp(m,0,sp,&spol);
1.75 noro 4148: if ( !spol ) continue;
1.74 noro 4149: nd_to_vect(m,s0vect,col,spol,svect);
1.76 noro 4150: if ( m == -1 )
4151: maxrs = ndv_reduce_vect_sf(m,svect,col,imat,rvect,nred);
4152: else
4153: maxrs = ndv_reduce_vect(m,svect,col,imat,rvect,nred);
1.74 noro 4154: for ( i = 0; i < col; i++ ) if ( svect[i] ) break;
4155: if ( i < col ) {
4156: spmat[sprow] = v = (UINT *)MALLOC_ATOMIC(spcol*sizeof(UINT));
4157: for ( j = k = 0; j < col; j++ )
4158: if ( !rhead[j] ) v[k++] = svect[j];
1.76 noro 4159: spsugar[sprow] = MAX(maxrs,SG(spol));
1.74 noro 4160: sprow++;
4161: }
1.76 noro 4162: nd_free(spol);
1.74 noro 4163: }
1.80 noro 4164: get_eg(&eg1); init_eg(&eg_f4_1); add_eg(&eg_f4_1,&eg0,&eg1);
4165: if ( DP_Print ) {
4166: fprintf(asir_out,"elim1=%fsec,",eg_f4_1.exectime+eg_f4_1.gctime);
4167: fflush(asir_out);
4168: }
1.74 noro 4169: /* free index arrays */
4170: for ( i = 0; i < nred; i++ ) GC_free(imat[i]->index.c);
4171:
4172: /* elimination (2nd step) */
4173: colstat = (int *)ALLOCA(spcol*sizeof(int));
4174: if ( m == -1 )
1.76 noro 4175: rank = nd_gauss_elim_sf(spmat,spsugar,sprow,spcol,m,colstat);
1.74 noro 4176: else
1.76 noro 4177: rank = nd_gauss_elim_mod(spmat,spsugar,sprow,spcol,m,colstat);
1.74 noro 4178: r0 = 0;
4179: for ( i = 0; i < rank; i++ ) {
4180: NEXTNODE(r0,r); BDY(r) =
4181: (pointer)vect_to_ndv(spmat[i],spcol,col,rhead,s0vect);
1.76 noro 4182: SG((NDV)BDY(r)) = spsugar[i];
1.74 noro 4183: GC_free(spmat[i]);
4184: }
4185: for ( ; i < sprow; i++ ) GC_free(spmat[i]);
1.80 noro 4186: get_eg(&eg2); init_eg(&eg_f4_2); add_eg(&eg_f4_2,&eg1,&eg2);
4187: init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg2);
1.74 noro 4188: if ( DP_Print ) {
1.80 noro 4189: fprintf(asir_out,"elim2=%fsec\n",eg_f4_2.exectime+eg_f4_2.gctime);
1.74 noro 4190: fprintf(asir_out,"nsp=%d,nred=%d,spmat=(%d,%d),rank=%d ",
4191: nsp,nred,sprow,spcol,rank);
4192: fprintf(asir_out,"%fsec\n",eg_f4.exectime+eg_f4.gctime);
1.63 noro 4193: }
1.74 noro 4194: return r0;
4195: }
4196:
4197: FILE *nd_write,*nd_read;
4198:
4199: void nd_send_int(int a) {
4200: write_int(nd_write,&a);
4201: }
4202:
4203: void nd_send_intarray(int *p,int len) {
4204: write_intarray(nd_write,p,len);
4205: }
4206:
4207: int nd_recv_int() {
4208: int a;
4209:
4210: read_int(nd_read,&a);
4211: return a;
4212: }
4213:
4214: void nd_recv_intarray(int *p,int len) {
4215: read_intarray(nd_read,p,len);
4216: }
4217:
4218: void nd_send_ndv(NDV p) {
4219: int len,i;
4220: NMV m;
4221:
4222: if ( !p ) nd_send_int(0);
4223: else {
4224: len = LEN(p);
4225: nd_send_int(len);
4226: m = BDY(p);
4227: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
4228: nd_send_int(CM(m));
4229: nd_send_intarray(DL(m),nd_wpd);
1.68 noro 4230: }
1.74 noro 4231: }
4232: }
4233:
4234: void nd_send_nd(ND p) {
4235: int len,i;
4236: NM m;
4237:
4238: if ( !p ) nd_send_int(0);
4239: else {
4240: len = LEN(p);
4241: nd_send_int(len);
4242: m = BDY(p);
4243: for ( i = 0; i < len; i++, m = NEXT(m) ) {
4244: nd_send_int(CM(m));
4245: nd_send_intarray(DL(m),nd_wpd);
1.65 noro 4246: }
1.74 noro 4247: }
4248: }
1.65 noro 4249:
1.74 noro 4250: NDV nd_recv_ndv()
4251: {
4252: int len,i;
4253: NMV m,m0;
4254: NDV r;
1.65 noro 4255:
1.74 noro 4256: len = nd_recv_int();
4257: if ( !len ) return 0;
4258: else {
4259: m0 = m = (NMV)GC_malloc_atomic_ignore_off_page(nmv_adv*len);
4260: ndv_alloc += len*nmv_adv;
4261: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
4262: CM(m) = nd_recv_int();
4263: nd_recv_intarray(DL(m),nd_wpd);
1.71 noro 4264: }
1.74 noro 4265: MKNDV(nd_nvar,m0,len,r);
4266: return r;
4267: }
4268: }
1.65 noro 4269:
1.74 noro 4270: int ox_exec_f4_red(Q proc)
4271: {
4272: Obj obj;
4273: STRING fname;
4274: NODE arg;
4275: int s;
4276: extern int ox_need_conv,ox_file_io;
4277:
4278: MKSTR(fname,"nd_exec_f4_red");
4279: arg = mknode(2,proc,fname);
4280: Pox_cmo_rpc(arg,&obj);
4281: s = get_ox_server_id(QTOS(proc));
4282: nd_write = iofp[s].out;
4283: nd_read = iofp[s].in;
4284: ox_need_conv = ox_file_io = 0;
4285: return s;
4286: }
4287:
4288: NODE nd_f4_red_dist(int m,ND_pairs sp0,UINT *s0vect,int col,NODE rp0)
4289: {
4290: int nsp,nred;
4291: int i,rank,s;
4292: NODE rp,r0,r;
4293: ND_pairs sp;
4294: NM_ind_pair pair;
4295: NMV nmv;
4296: NM nm;
4297: NDV nf;
4298: Obj proc,dmy;
4299:
4300: ox_launch_main(0,0,&proc);
4301: s = ox_exec_f4_red((Q)proc);
4302:
4303: nd_send_int(m);
4304: nd_send_int(nd_nvar);
4305: nd_send_int(nd_bpe);
4306: nd_send_int(nd_wpd);
4307: nd_send_int(nmv_adv);
4308:
1.89 noro 4309: saveobj(nd_write,dp_current_spec->obj); fflush(nd_write);
1.74 noro 4310:
4311: nd_send_int(nd_psn);
4312: for ( i = 0; i < nd_psn; i++ ) nd_send_ndv(nd_ps[i]);
4313:
4314: for ( sp = sp0, nsp = 0; sp; sp = NEXT(sp), nsp++ );
4315: nd_send_int(nsp);
4316: for ( i = 0, sp = sp0; i < nsp; i++, sp = NEXT(sp) ) {
4317: nd_send_int(sp->i1); nd_send_int(sp->i2);
4318: }
4319:
4320: nd_send_int(col); nd_send_intarray(s0vect,col*nd_wpd);
4321:
4322: nred = length(rp0); nd_send_int(nred);
4323: for ( i = 0, rp = rp0; i < nred; i++, rp = NEXT(rp) ) {
4324: pair = (NM_ind_pair)BDY(rp);
4325: nd_send_int(pair->index);
4326: nd_send_intarray(pair->mul->dl,nd_wpd);
4327: }
4328: fflush(nd_write);
4329: rank = nd_recv_int();
4330: fprintf(asir_out,"rank=%d\n",rank);
4331: r0 = 0;
4332: for ( i = 0; i < rank; i++ ) {
4333: nf = nd_recv_ndv();
4334: NEXTNODE(r0,r); BDY(r) = (pointer)nf;
4335: }
4336: Pox_shutdown(mknode(1,proc),&dmy);
4337: return r0;
4338: }
4339:
4340: /* server side */
4341:
4342: void nd_exec_f4_red_dist()
4343: {
4344: int m,i,nsp,col,s0size,nred,spcol,j,k;
4345: NM_ind_pair *rp0;
4346: NDV nf;
4347: UINT *s0vect;
4348: IndArray *imat;
4349: int *rhead;
4350: int **spmat;
4351: UINT *svect,*v;
4352: ND_pairs *sp0;
4353: int *colstat;
4354: int a,sprow,rank;
1.89 noro 4355: struct order_spec *ord;
1.74 noro 4356: Obj ordspec;
4357: ND spol;
1.76 noro 4358: int maxrs;
4359: int *spsugar;
1.74 noro 4360:
4361: nd_read = iofp[0].in;
4362: nd_write = iofp[0].out;
4363: m = nd_recv_int();
4364: nd_nvar = nd_recv_int();
4365: nd_bpe = nd_recv_int();
4366: nd_wpd = nd_recv_int();
4367: nmv_adv = nd_recv_int();
4368:
4369: loadobj(nd_read,&ordspec);
1.89 noro 4370: create_order_spec(0,ordspec,&ord);
4371: nd_init_ord(ord);
1.74 noro 4372: nd_setup_parameters(nd_nvar,0);
4373:
4374: nd_psn = nd_recv_int();
4375: nd_ps = (NDV *)MALLOC(nd_psn*sizeof(NDV));
4376: nd_bound = (UINT **)MALLOC(nd_psn*sizeof(UINT *));
4377: for ( i = 0; i < nd_psn; i++ ) {
4378: nd_ps[i] = nd_recv_ndv();
4379: nd_bound[i] = ndv_compute_bound(nd_ps[i]);
4380: }
4381:
4382: nsp = nd_recv_int();
4383: sp0 = (ND_pairs *)MALLOC(nsp*sizeof(ND_pairs));
4384: for ( i = 0; i < nsp; i++ ) {
4385: NEWND_pairs(sp0[i]);
4386: sp0[i]->i1 = nd_recv_int(); sp0[i]->i2 = nd_recv_int();
4387: ndl_lcm(HDL(nd_ps[sp0[i]->i1]),HDL(nd_ps[sp0[i]->i2]),LCM(sp0[i]));
4388: }
4389:
4390: col = nd_recv_int();
4391: s0size = col*nd_wpd;
4392: s0vect = (UINT *)MALLOC(s0size*sizeof(UINT));
4393: nd_recv_intarray(s0vect,s0size);
4394:
4395: nred = nd_recv_int();
4396: rp0 = (NM_ind_pair *)MALLOC(nred*sizeof(NM_ind_pair));
4397: for ( i = 0; i < nred; i++ ) {
4398: rp0[i] = (NM_ind_pair)MALLOC(sizeof(struct oNM_ind_pair));
4399: rp0[i]->index = nd_recv_int();
4400: rp0[i]->mul = (NM)MALLOC(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT));
4401: nd_recv_intarray(rp0[i]->mul->dl,nd_wpd);
4402: }
4403:
4404: spcol = col-nred;
4405: imat = (IndArray *)MALLOC(nred*sizeof(IndArray));
4406: rhead = (int *)MALLOC(col*sizeof(int));
4407: for ( i = 0; i < col; i++ ) rhead[i] = 0;
4408:
4409: /* construction of index arrays */
4410: for ( i = 0; i < nred; i++ ) {
4411: imat[i] = nm_ind_pair_to_vect_compress(m,s0vect,col,rp0[i]);
4412: rhead[imat[i]->head] = 1;
4413: }
4414:
4415: /* elimination (1st step) */
4416: spmat = (int **)MALLOC(nsp*sizeof(UINT *));
4417: svect = (UINT *)MALLOC(col*sizeof(UINT));
1.76 noro 4418: spsugar = (int *)ALLOCA(nsp*sizeof(UINT));
1.74 noro 4419: for ( a = sprow = 0; a < nsp; a++ ) {
4420: nd_sp(m,0,sp0[a],&spol);
1.75 noro 4421: if ( !spol ) continue;
1.74 noro 4422: nd_to_vect(m,s0vect,col,spol,svect);
1.76 noro 4423: if ( m == -1 )
4424: maxrs = ndv_reduce_vect_sf(m,svect,col,imat,rp0,nred);
4425: else
4426: maxrs = ndv_reduce_vect(m,svect,col,imat,rp0,nred);
1.74 noro 4427: for ( i = 0; i < col; i++ ) if ( svect[i] ) break;
4428: if ( i < col ) {
4429: spmat[sprow] = v = (UINT *)MALLOC(spcol*sizeof(UINT));
4430: for ( j = k = 0; j < col; j++ )
4431: if ( !rhead[j] ) v[k++] = svect[j];
1.76 noro 4432: spsugar[sprow] = MAX(maxrs,SG(spol));
1.74 noro 4433: sprow++;
4434: }
1.76 noro 4435: nd_free(spol);
1.74 noro 4436: }
4437: /* elimination (2nd step) */
4438: colstat = (int *)ALLOCA(spcol*sizeof(int));
4439: if ( m == -1 )
1.76 noro 4440: rank = nd_gauss_elim_sf(spmat,spsugar,sprow,spcol,m,colstat);
1.74 noro 4441: else
1.76 noro 4442: rank = nd_gauss_elim_mod(spmat,spsugar,sprow,spcol,m,colstat);
1.74 noro 4443: nd_send_int(rank);
4444: for ( i = 0; i < rank; i++ ) {
4445: nf = vect_to_ndv(spmat[i],spcol,col,rhead,s0vect);
4446: nd_send_ndv(nf);
1.63 noro 4447: }
1.74 noro 4448: fflush(nd_write);
1.76 noro 4449: }
4450:
4451: int nd_gauss_elim_mod(int **mat0,int *sugar,int row,int col,int md,int *colstat)
4452: {
4453: int i,j,k,l,inv,a,rank,s;
4454: unsigned int *t,*pivot,*pk;
4455: unsigned int **mat;
4456:
4457: mat = (unsigned int **)mat0;
4458: for ( rank = 0, j = 0; j < col; j++ ) {
4459: for ( i = rank; i < row; i++ )
4460: mat[i][j] %= md;
4461: for ( i = rank; i < row; i++ )
4462: if ( mat[i][j] )
4463: break;
4464: if ( i == row ) {
4465: colstat[j] = 0;
4466: continue;
4467: } else
4468: colstat[j] = 1;
4469: if ( i != rank ) {
4470: t = mat[i]; mat[i] = mat[rank]; mat[rank] = t;
4471: s = sugar[i]; sugar[i] = sugar[rank]; sugar[rank] = s;
4472: }
4473: pivot = mat[rank];
4474: s = sugar[rank];
4475: inv = invm(pivot[j],md);
4476: for ( k = j, pk = pivot+k; k < col; k++, pk++ )
4477: if ( *pk ) {
4478: if ( *pk >= (unsigned int)md )
4479: *pk %= md;
4480: DMAR(*pk,inv,0,md,*pk)
4481: }
4482: for ( i = rank+1; i < row; i++ ) {
4483: t = mat[i];
4484: if ( a = t[j] ) {
4485: sugar[i] = MAX(sugar[i],s);
4486: red_by_vect(md,t+j,pivot+j,md-a,col-j);
4487: }
4488: }
4489: rank++;
4490: }
4491: for ( j = col-1, l = rank-1; j >= 0; j-- )
4492: if ( colstat[j] ) {
4493: pivot = mat[l];
4494: s = sugar[l];
4495: for ( i = 0; i < l; i++ ) {
4496: t = mat[i];
4497: t[j] %= md;
4498: if ( a = t[j] ) {
4499: sugar[i] = MAX(sugar[i],s);
4500: red_by_vect(md,t+j,pivot+j,md-a,col-j);
4501: }
4502: }
4503: l--;
4504: }
4505: for ( j = 0, l = 0; l < rank; j++ )
4506: if ( colstat[j] ) {
4507: t = mat[l];
4508: for ( k = j; k < col; k++ )
4509: if ( t[k] >= (unsigned int)md )
4510: t[k] %= md;
4511: l++;
4512: }
4513: return rank;
4514: }
4515:
4516: int nd_gauss_elim_sf(int **mat0,int *sugar,int row,int col,int md,int *colstat)
4517: {
4518: int i,j,k,l,inv,a,rank,s;
4519: unsigned int *t,*pivot,*pk;
4520: unsigned int **mat;
4521:
4522: mat = (unsigned int **)mat0;
4523: for ( rank = 0, j = 0; j < col; j++ ) {
4524: for ( i = rank; i < row; i++ )
4525: if ( mat[i][j] )
4526: break;
4527: if ( i == row ) {
4528: colstat[j] = 0;
4529: continue;
4530: } else
4531: colstat[j] = 1;
4532: if ( i != rank ) {
4533: t = mat[i]; mat[i] = mat[rank]; mat[rank] = t;
4534: s = sugar[i]; sugar[i] = sugar[rank]; sugar[rank] = s;
4535: }
4536: pivot = mat[rank];
4537: s = sugar[rank];
4538: inv = _invsf(pivot[j]);
4539: for ( k = j, pk = pivot+k; k < col; k++, pk++ )
4540: if ( *pk )
4541: *pk = _mulsf(*pk,inv);
4542: for ( i = rank+1; i < row; i++ ) {
4543: t = mat[i];
4544: if ( a = t[j] ) {
4545: sugar[i] = MAX(sugar[i],s);
4546: red_by_vect_sf(md,t+j,pivot+j,_chsgnsf(a),col-j);
4547: }
4548: }
4549: rank++;
4550: }
4551: for ( j = col-1, l = rank-1; j >= 0; j-- )
4552: if ( colstat[j] ) {
4553: pivot = mat[l];
4554: s = sugar[l];
4555: for ( i = 0; i < l; i++ ) {
4556: t = mat[i];
4557: if ( a = t[j] ) {
4558: sugar[i] = MAX(sugar[i],s);
4559: red_by_vect_sf(md,t+j,pivot+j,_chsgnsf(a),col-j);
4560: }
4561: }
4562: l--;
4563: }
4564: return rank;
1.77 noro 4565: }
4566:
4567: int ndv_ishomo(NDV p)
4568: {
4569: NMV m;
4570: int len,h;
4571:
4572: if ( !p ) return 1;
4573: len = LEN(p);
4574: m = BDY(p);
4575: h = TD(DL(m));
4576: NMV_ADV(m);
4577: for ( len--; len; len--, NMV_ADV(m) )
4578: if ( TD(DL(m)) != h ) return 0;
4579: return 1;
4580: }
4581:
4582: void ndv_save(NDV p,int index)
4583: {
4584: FILE *s;
4585: char name[BUFSIZ];
4586: short id;
4587: int nv,sugar,len,n,i,td,e,j;
4588: NMV m;
4589: unsigned int *dl;
4590:
4591: sprintf(name,"%s/%d",Demand,index);
4592: s = fopen(name,"w");
4593: savevl(s,0);
4594: if ( !p ) {
4595: saveobj(s,0);
4596: return;
4597: }
4598: id = O_DP;
4599: nv = NV(p);
4600: sugar = SG(p);
4601: len = LEN(p);
4602: write_short(s,&id); write_int(s,&nv); write_int(s,&sugar);
4603: write_int(s,&len);
4604:
4605: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
4606: saveobj(s,(Obj)CQ(m));
4607: dl = DL(m);
4608: td = TD(dl);
4609: write_int(s,&td);
4610: for ( j = 0; j < nv; j++ ) {
4611: e = GET_EXP(dl,j);
4612: write_int(s,&e);
4613: }
4614: }
4615: fclose(s);
4616: }
4617:
4618: NDV ndv_load(int index)
4619: {
4620: FILE *s;
4621: char name[BUFSIZ];
4622: short id;
4623: int nv,sugar,len,n,i,td,e,j;
4624: NDV d;
4625: NMV m0,m;
4626: unsigned int *dl;
4627: Obj obj;
4628:
4629: sprintf(name,"%s/%d",Demand,index);
4630: s = fopen(name,"r");
4631: if ( !s ) return 0;
4632:
4633: skipvl(s);
4634: read_short(s,&id);
4635: if ( !id ) return 0;
4636: read_int(s,&nv);
4637: read_int(s,&sugar);
4638: read_int(s,&len);
4639:
4640: m0 = m = MALLOC(len*nmv_adv);
4641: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
4642: loadobj(s,&obj); CQ(m) = (Q)obj;
4643: dl = DL(m);
4644: ndl_zero(dl);
4645: read_int(s,&td); TD(dl) = td;
4646: for ( j = 0; j < nv; j++ ) {
4647: read_int(s,&e);
4648: PUT_EXP(dl,j,e);
4649: }
4650: if ( nd_blockmask ) ndl_weight_mask(dl);
4651: }
4652: fclose(s);
4653: MKNDV(nv,m0,len,d);
4654: SG(d) = sugar;
4655: return d;
1.99 noro 4656: }
4657:
1.102 ! noro 4658: void nd_det(int mod,MAT f,P *rp)
1.99 noro 4659: {
4660: VL fv,tv;
1.102 ! noro 4661: int n,i,j,max,e,nvar,sgn,k0,l0,len0,len,k,l,a;
1.99 noro 4662: pointer **m;
4663: Q mone;
1.102 ! noro 4664: NDV **dm;
! 4665: NDV *t,*mi,*mj;
! 4666: NDV d,s,mij,mjj;
! 4667: ND u;
! 4668: NMV nmv;
1.99 noro 4669: PGeoBucket bucket;
4670: struct order_spec *ord;
4671:
4672: create_order_spec(0,0,&ord);
4673: nd_init_ord(ord);
4674: get_vars((Obj)f,&fv);
4675: if ( f->row != f->col )
4676: error("nd_det : non-square matrix");
4677: n = f->row;
4678: for ( nvar = 0, tv = fv; tv; tv = NEXT(tv), nvar++ );
4679: m = f->body;
4680: for ( i = 0, max = 0; i < n; i++ )
4681: for ( j = 0; j < n; j++ )
4682: for ( tv = fv; tv; tv = NEXT(tv) ) {
4683: e = getdeg(tv->v,(P)m[i][j]);
4684: max = MAX(e,max);
4685: }
4686: nd_setup_parameters(nvar,1024);
1.102 ! noro 4687: dm = (NDV **)almat_pointer(n,n);
1.99 noro 4688: for ( i = 0, max = 0; i < n; i++ )
1.102 ! noro 4689: for ( j = 0; j < n; j++ ) {
! 4690: dm[i][j] = ptondv(CO,fv,m[i][j]);
! 4691: if ( mod ) ndv_mod(mod,dm[i][j]);
! 4692: if ( dm[i][j] && !LEN(dm[i][j]) ) dm[i][j] = 0;
! 4693: }
! 4694: d = ptondv(CO,fv,(P)ONE);
! 4695: if ( mod ) ndv_mod(mod,d);
1.99 noro 4696: chsgnq(ONE,&mone);
4697: for ( j = 0, sgn = 1; j < n; j++ ) {
4698: for ( i = j; i < n && !dm[i][j]; i++ );
4699: if ( i == n ) {
4700: *rp = 0;
4701: return;
4702: }
1.102 ! noro 4703: k0 = i; l0 = j; len0 = LEN(dm[k0][l0]);
1.99 noro 4704: for ( k = j; k < n; k++ )
4705: for ( l = j; l < n; l++ )
1.102 ! noro 4706: if ( dm[k][l] && LEN(dm[k][l]) < len0 ) {
! 4707: k0 = k; l0 = l; len0 = LEN(dm[k][l]);
1.99 noro 4708: }
4709: if ( k0 != j ) {
4710: t = dm[j]; dm[j] = dm[k0]; dm[k0] = t;
4711: sgn = -sgn;
4712: }
4713: if ( l0 != j ) {
4714: for ( k = j; k < n; k++ ) {
4715: s = dm[k][j]; dm[k][j] = dm[k][l0]; dm[k][l0] = s;
4716: }
4717: sgn = -sgn;
4718: }
4719: for ( i = j+1, mj = dm[j], mjj = mj[j]; i < n; i++ ) {
4720: mi = dm[i]; mij = mi[j];
1.102 ! noro 4721: if ( mod )
! 4722: ndv_mul_c(mod,mij,mod-1);
! 4723: else
! 4724: ndv_mul_c_q(mij,mone);
1.99 noro 4725: for ( k = j+1; k < n; k++ ) {
4726: bucket = create_pbucket();
4727: if ( mi[k] )
1.102 ! noro 4728: nmv = BDY(mjj); len = LEN(mjj);
! 4729: for ( a = 0; a < len; a++, NMV_ADV(nmv) ) {
! 4730: u = ndv_mul_nmv_trunc(mod,nmv,mi[k],DL(BDY(d)));
! 4731: add_pbucket(mod,bucket,u);
1.99 noro 4732: }
4733: if ( mj[k] && mij ) {
1.102 ! noro 4734: nmv = BDY(mij); len = LEN(mij);
! 4735: for ( a = 0; a < len; a++, NMV_ADV(nmv) ) {
! 4736: u = ndv_mul_nmv_trunc(mod,nmv,mj[k],DL(BDY(d)));
! 4737: add_pbucket(mod,bucket,u);
1.99 noro 4738: }
4739: }
1.102 ! noro 4740: u = normalize_pbucket(mod,bucket);
! 4741: u = nd_quo(mod,u,d);
! 4742: mi[k] = ndtondv(mod,u);
1.99 noro 4743: }
4744: }
4745: d = mjj;
4746: }
1.100 noro 4747: if ( sgn < 0 )
1.102 ! noro 4748: if ( mod )
! 4749: ndv_mul_c(mod,d,mod-1);
! 4750: else
! 4751: ndv_mul_c_q(d,mone);
! 4752: *rp = ndvtop(mod,CO,fv,d);
1.99 noro 4753: }
4754:
1.102 ! noro 4755: ND ndv_mul_nmv_trunc(int mod,NMV m0,NDV p,UINT *d)
1.99 noro 4756: {
4757: NM mr,mr0;
1.102 ! noro 4758: NM tnm;
! 4759: NMV m;
1.99 noro 4760: UINT *d0,*dt,*dm;
4761: int c,n,td,i,c1,c2,len;
4762: Q q;
4763: ND r;
4764:
4765: if ( !p ) return 0;
4766: else {
1.102 ! noro 4767: n = NV(p); m = BDY(p); len = LEN(p);
1.99 noro 4768: d0 = DL(m0);
1.102 ! noro 4769: td = TD(d);
1.99 noro 4770: mr0 = 0;
4771: NEWNM(tnm);
1.102 ! noro 4772: if ( mod ) {
! 4773: c = CM(m0);
! 4774: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
! 4775: ndl_add(DL(m),d0,DL(tnm));
! 4776: if ( ndl_reducible(DL(tnm),d) ) {
! 4777: NEXTNM(mr0,mr);
! 4778: c1 = CM(m); DMAR(c1,c,0,mod,c2); CM(mr) = c2;
! 4779: ndl_add(DL(m),d0,DL(mr));
! 4780: }
! 4781: }
! 4782: } else {
! 4783: q = CQ(m0);
! 4784: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
! 4785: ndl_add(DL(m),d0,DL(tnm));
! 4786: if ( ndl_reducible(DL(tnm),d) ) {
! 4787: NEXTNM(mr0,mr);
! 4788: mulq(CQ(m),q,&CQ(mr));
! 4789: ndl_add(DL(m),d0,DL(mr));
! 4790: }
1.99 noro 4791: }
4792: }
4793: if ( !mr0 )
4794: return 0;
4795: else {
4796: NEXT(mr) = 0;
4797: MKND(NV(p),mr0,len,r);
4798: SG(r) = SG(p) + TD(d0);
4799: return r;
4800: }
4801: }
1.59 noro 4802: }
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>