Annotation of OpenXM_contrib2/asir2000/engine/nd.c, Revision 1.115
1.115 ! noro 1: /* $OpenXM: OpenXM_contrib2/asir2000/engine/nd.c,v 1.114 2004/10/25 04:19:50 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:
1.103 noro 11: #if 0
1.74 noro 12: static int ndv_alloc;
1.103 noro 13: #endif
1.87 noro 14: #if 1
1.69 noro 15: static int nd_f4_nsp=0x7fffffff;
1.87 noro 16: #else
17: static int nd_f4_nsp=50;
18: #endif
1.42 noro 19: static double nd_scale=2;
1.61 noro 20: static UINT **nd_bound;
1.42 noro 21: static struct order_spec *nd_ord;
22: static EPOS nd_epos;
1.43 noro 23: static BlockMask nd_blockmask;
1.42 noro 24: static int nd_nvar;
25: static int nd_isrlex;
26: static int nd_epw,nd_bpe,nd_wpd,nd_exporigin;
1.61 noro 27: static UINT nd_mask[32];
28: static UINT nd_mask0,nd_mask1;
1.42 noro 29:
1.20 noro 30: static NDV *nd_ps;
1.53 noro 31: static NDV *nd_ps_trace;
1.42 noro 32: static RHist *nd_psh;
33: static int nd_psn,nd_pslen;
34: static RHist *nd_red;
1.96 noro 35: static int *nd_work_vector;
36: static int **nd_matrix;
37: static int nd_matrix_len;
1.97 noro 38: static struct weight_or_block *nd_worb;
39: static int nd_worb_len;
1.42 noro 40: static int nd_found,nd_create,nd_notfirst;
41: static int nmv_adv;
1.77 noro 42: static int nd_demand;
1.1 noro 43:
1.114 noro 44: UINT *nd_det_compute_bound(NDV **dm,int n,int j);
45: void nd_det_reconstruct(NDV **dm,int n,int j,NDV d);
46:
1.1 noro 47: void nd_free_private_storage()
48: {
49: _nm_free_list = 0;
1.5 noro 50: _ndp_free_list = 0;
1.71 noro 51: #if 0
1.1 noro 52: GC_gcollect();
1.71 noro 53: #endif
1.1 noro 54: }
55:
56: void _NM_alloc()
57: {
58: NM p;
59: int i;
60:
1.11 noro 61: for ( i = 0; i < 1024; i++ ) {
1.61 noro 62: p = (NM)GC_malloc(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT));
1.1 noro 63: p->next = _nm_free_list; _nm_free_list = p;
64: }
65: }
66:
67: void _ND_alloc()
68: {
69: ND p;
70: int i;
71:
72: for ( i = 0; i < 1024; i++ ) {
73: p = (ND)GC_malloc(sizeof(struct oND));
74: p->body = (NM)_nd_free_list; _nd_free_list = p;
75: }
76: }
77:
78: void _NDP_alloc()
79: {
80: ND_pairs p;
81: int i;
82:
1.11 noro 83: for ( i = 0; i < 1024; i++ ) {
1.1 noro 84: p = (ND_pairs)GC_malloc(sizeof(struct oND_pairs)
1.61 noro 85: +(nd_wpd-1)*sizeof(UINT));
1.1 noro 86: p->next = _ndp_free_list; _ndp_free_list = p;
87: }
88: }
89:
1.30 noro 90: INLINE int nd_length(ND p)
1.1 noro 91: {
92: NM m;
93: int i;
94:
95: if ( !p )
96: return 0;
97: else {
98: for ( i = 0, m = BDY(p); m; m = NEXT(m), i++ );
99: return i;
100: }
101: }
102:
1.61 noro 103: INLINE int ndl_reducible(UINT *d1,UINT *d2)
1.1 noro 104: {
1.61 noro 105: UINT u1,u2;
1.1 noro 106: int i,j;
107:
1.34 noro 108: if ( TD(d1) < TD(d2) ) return 0;
1.65 noro 109: #if USE_UNROLL
1.1 noro 110: switch ( nd_bpe ) {
1.62 noro 111: case 3:
112: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
113: u1 = d1[i]; u2 = d2[i];
114: if ( (u1&0x38000000) < (u2&0x38000000) ) return 0;
115: if ( (u1& 0x7000000) < (u2& 0x7000000) ) return 0;
116: if ( (u1& 0xe00000) < (u2& 0xe00000) ) return 0;
117: if ( (u1& 0x1c0000) < (u2& 0x1c0000) ) return 0;
118: if ( (u1& 0x38000) < (u2& 0x38000) ) return 0;
119: if ( (u1& 0x7000) < (u2& 0x7000) ) return 0;
120: if ( (u1& 0xe00) < (u2& 0xe00) ) return 0;
121: if ( (u1& 0x1c0) < (u2& 0x1c0) ) return 0;
122: if ( (u1& 0x38) < (u2& 0x38) ) return 0;
123: if ( (u1& 0x7) < (u2& 0x7) ) return 0;
124: }
125: return 1;
126: break;
1.1 noro 127: case 4:
1.41 noro 128: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 129: u1 = d1[i]; u2 = d2[i];
130: if ( (u1&0xf0000000) < (u2&0xf0000000) ) return 0;
1.62 noro 131: if ( (u1& 0xf000000) < (u2& 0xf000000) ) return 0;
132: if ( (u1& 0xf00000) < (u2& 0xf00000) ) return 0;
133: if ( (u1& 0xf0000) < (u2& 0xf0000) ) return 0;
134: if ( (u1& 0xf000) < (u2& 0xf000) ) return 0;
135: if ( (u1& 0xf00) < (u2& 0xf00) ) return 0;
136: if ( (u1& 0xf0) < (u2& 0xf0) ) return 0;
137: if ( (u1& 0xf) < (u2& 0xf) ) return 0;
1.1 noro 138: }
139: return 1;
140: break;
141: case 6:
1.41 noro 142: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 143: u1 = d1[i]; u2 = d2[i];
144: if ( (u1&0x3f000000) < (u2&0x3f000000) ) return 0;
1.62 noro 145: if ( (u1& 0xfc0000) < (u2& 0xfc0000) ) return 0;
146: if ( (u1& 0x3f000) < (u2& 0x3f000) ) return 0;
147: if ( (u1& 0xfc0) < (u2& 0xfc0) ) return 0;
148: if ( (u1& 0x3f) < (u2& 0x3f) ) return 0;
1.1 noro 149: }
150: return 1;
151: break;
152: case 8:
1.41 noro 153: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 154: u1 = d1[i]; u2 = d2[i];
155: if ( (u1&0xff000000) < (u2&0xff000000) ) return 0;
1.62 noro 156: if ( (u1& 0xff0000) < (u2& 0xff0000) ) return 0;
157: if ( (u1& 0xff00) < (u2& 0xff00) ) return 0;
158: if ( (u1& 0xff) < (u2& 0xff) ) return 0;
1.1 noro 159: }
160: return 1;
161: break;
162: case 16:
1.41 noro 163: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 164: u1 = d1[i]; u2 = d2[i];
165: if ( (u1&0xffff0000) < (u2&0xffff0000) ) return 0;
1.62 noro 166: if ( (u1& 0xffff) < (u2& 0xffff) ) return 0;
1.1 noro 167: }
168: return 1;
169: break;
170: case 32:
1.41 noro 171: for ( i = nd_exporigin; i < nd_wpd; i++ )
1.1 noro 172: if ( d1[i] < d2[i] ) return 0;
173: return 1;
174: break;
175: default:
1.41 noro 176: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 177: u1 = d1[i]; u2 = d2[i];
178: for ( j = 0; j < nd_epw; j++ )
179: if ( (u1&nd_mask[j]) < (u2&nd_mask[j]) ) return 0;
180: }
181: return 1;
182: }
1.65 noro 183: #else
184: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
185: u1 = d1[i]; u2 = d2[i];
186: for ( j = 0; j < nd_epw; j++ )
187: if ( (u1&nd_mask[j]) < (u2&nd_mask[j]) ) return 0;
188: }
189: return 1;
190: #endif
1.1 noro 191: }
192:
1.61 noro 193: /*
194: * If the current order is a block order,
195: * then the last block is length 1 and contains
196: * the homo variable. Otherwise, the original
197: * order is either 0 or 2.
198: */
199:
200: void ndl_homogenize(UINT *d,UINT *r,int obpe,EPOS oepos,int weight)
1.23 noro 201: {
1.61 noro 202: int w,i,e,n,omask0;
203:
204: omask0 = (1<<obpe)-1;
205: n = nd_nvar-1;
206: ndl_zero(r);
207: for ( i = 0; i < n; i++ ) {
208: e = GET_EXP_OLD(d,i);
209: PUT_EXP(r,i,e);
210: }
211: w = TD(d);
212: PUT_EXP(r,nd_nvar-1,weight-w);
213: TD(r) = weight;
214: if ( nd_blockmask ) ndl_weight_mask(r);
215: }
216:
217: void ndl_dehomogenize(UINT *d)
218: {
219: UINT mask;
220: UINT h;
1.31 noro 221: int i,bits;
1.23 noro 222:
1.44 noro 223: if ( nd_blockmask ) {
224: h = GET_EXP(d,nd_nvar-1);
1.45 noro 225: XOR_EXP(d,nd_nvar-1,h);
1.44 noro 226: TD(d) -= h;
227: d[nd_exporigin-1] -= h;
228: } else {
229: if ( nd_isrlex ) {
230: if ( nd_bpe == 32 ) {
231: h = d[nd_exporigin];
232: for ( i = nd_exporigin+1; i < nd_wpd; i++ )
233: d[i-1] = d[i];
234: d[i-1] = 0;
235: TD(d) -= h;
236: } else {
237: bits = nd_epw*nd_bpe;
238: mask = bits==32?0xffffffff:((1<<(nd_epw*nd_bpe))-1);
239: h = (d[nd_exporigin]>>((nd_epw-1)*nd_bpe))&nd_mask0;
240: for ( i = nd_exporigin; i < nd_wpd; i++ )
241: d[i] = ((d[i]<<nd_bpe)&mask)
242: |(i+1<nd_wpd?((d[i+1]>>((nd_epw-1)*nd_bpe))&nd_mask0):0);
243: TD(d) -= h;
244: }
1.45 noro 245: } else {
246: h = GET_EXP(d,nd_nvar-1);
247: XOR_EXP(d,nd_nvar-1,h);
248: TD(d) -= h;
249: }
1.44 noro 250: }
1.23 noro 251: }
252:
1.61 noro 253: void ndl_lcm(UINT *d1,unsigned *d2,UINT *d)
1.1 noro 254: {
1.61 noro 255: UINT t1,t2,u,u1,u2;
1.43 noro 256: int i,j,l;
1.1 noro 257:
1.65 noro 258: #if USE_UNROLL
1.1 noro 259: switch ( nd_bpe ) {
1.62 noro 260: case 3:
261: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
262: u1 = d1[i]; u2 = d2[i];
263: t1 = (u1&0x38000000); t2 = (u2&0x38000000); u = t1>t2?t1:t2;
264: t1 = (u1& 0x7000000); t2 = (u2& 0x7000000); u |= t1>t2?t1:t2;
265: t1 = (u1& 0xe00000); t2 = (u2& 0xe00000); u |= t1>t2?t1:t2;
266: t1 = (u1& 0x1c0000); t2 = (u2& 0x1c0000); u |= t1>t2?t1:t2;
267: t1 = (u1& 0x38000); t2 = (u2& 0x38000); u |= t1>t2?t1:t2;
268: t1 = (u1& 0x7000); t2 = (u2& 0x7000); u |= t1>t2?t1:t2;
269: t1 = (u1& 0xe00); t2 = (u2& 0xe00); u |= t1>t2?t1:t2;
270: t1 = (u1& 0x1c0); t2 = (u2& 0x1c0); u |= t1>t2?t1:t2;
271: t1 = (u1& 0x38); t2 = (u2& 0x38); u |= t1>t2?t1:t2;
272: t1 = (u1& 0x7); t2 = (u2& 0x7); u |= t1>t2?t1:t2;
273: d[i] = u;
274: }
275: break;
1.1 noro 276: case 4:
1.41 noro 277: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 278: u1 = d1[i]; u2 = d2[i];
279: t1 = (u1&0xf0000000); t2 = (u2&0xf0000000); u = t1>t2?t1:t2;
1.62 noro 280: t1 = (u1& 0xf000000); t2 = (u2& 0xf000000); u |= t1>t2?t1:t2;
281: t1 = (u1& 0xf00000); t2 = (u2& 0xf00000); u |= t1>t2?t1:t2;
282: t1 = (u1& 0xf0000); t2 = (u2& 0xf0000); u |= t1>t2?t1:t2;
283: t1 = (u1& 0xf000); t2 = (u2& 0xf000); u |= t1>t2?t1:t2;
284: t1 = (u1& 0xf00); t2 = (u2& 0xf00); u |= t1>t2?t1:t2;
285: t1 = (u1& 0xf0); t2 = (u2& 0xf0); u |= t1>t2?t1:t2;
286: t1 = (u1& 0xf); t2 = (u2& 0xf); u |= t1>t2?t1:t2;
1.1 noro 287: d[i] = u;
288: }
289: break;
290: case 6:
1.41 noro 291: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 292: u1 = d1[i]; u2 = d2[i];
293: t1 = (u1&0x3f000000); t2 = (u2&0x3f000000); u = t1>t2?t1:t2;
1.62 noro 294: t1 = (u1& 0xfc0000); t2 = (u2& 0xfc0000); u |= t1>t2?t1:t2;
295: t1 = (u1& 0x3f000); t2 = (u2& 0x3f000); u |= t1>t2?t1:t2;
296: t1 = (u1& 0xfc0); t2 = (u2& 0xfc0); u |= t1>t2?t1:t2;
297: t1 = (u1& 0x3f); t2 = (u2& 0x3f); u |= t1>t2?t1:t2;
1.1 noro 298: d[i] = u;
299: }
300: break;
301: case 8:
1.41 noro 302: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 303: u1 = d1[i]; u2 = d2[i];
304: t1 = (u1&0xff000000); t2 = (u2&0xff000000); u = t1>t2?t1:t2;
1.62 noro 305: t1 = (u1& 0xff0000); t2 = (u2& 0xff0000); u |= t1>t2?t1:t2;
306: t1 = (u1& 0xff00); t2 = (u2& 0xff00); u |= t1>t2?t1:t2;
307: t1 = (u1& 0xff); t2 = (u2& 0xff); u |= t1>t2?t1:t2;
1.1 noro 308: d[i] = u;
309: }
310: break;
311: case 16:
1.41 noro 312: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 313: u1 = d1[i]; u2 = d2[i];
314: t1 = (u1&0xffff0000); t2 = (u2&0xffff0000); u = t1>t2?t1:t2;
1.62 noro 315: t1 = (u1& 0xffff); t2 = (u2& 0xffff); u |= t1>t2?t1:t2;
1.1 noro 316: d[i] = u;
317: }
318: break;
319: case 32:
1.41 noro 320: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 321: u1 = d1[i]; u2 = d2[i];
322: d[i] = u1>u2?u1:u2;
323: }
324: break;
325: default:
1.41 noro 326: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 327: u1 = d1[i]; u2 = d2[i];
328: for ( j = 0, u = 0; j < nd_epw; j++ ) {
329: t1 = (u1&nd_mask[j]); t2 = (u2&nd_mask[j]); u |= t1>t2?t1:t2;
330: }
331: d[i] = u;
332: }
333: break;
334: }
1.65 noro 335: #else
336: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
337: u1 = d1[i]; u2 = d2[i];
338: for ( j = 0, u = 0; j < nd_epw; j++ ) {
339: t1 = (u1&nd_mask[j]); t2 = (u2&nd_mask[j]); u |= t1>t2?t1:t2;
340: }
341: d[i] = u;
342: }
343: #endif
1.39 noro 344: TD(d) = ndl_weight(d);
1.61 noro 345: if ( nd_blockmask ) ndl_weight_mask(d);
1.57 noro 346: }
347:
1.61 noro 348: int ndl_weight(UINT *d)
1.1 noro 349: {
1.61 noro 350: UINT t,u;
1.1 noro 351: int i,j;
352:
1.60 noro 353: if ( current_dl_weight_vector )
354: for ( i = 0, t = 0; i < nd_nvar; i++ ) {
355: u = GET_EXP(d,i);
356: t += MUL_WEIGHT(u,i);
357: }
358: else
359: for ( t = 0, i = nd_exporigin; i < nd_wpd; i++ ) {
360: u = d[i];
361: for ( j = 0; j < nd_epw; j++, u>>=nd_bpe )
362: t += (u&nd_mask0);
363: }
1.1 noro 364: return t;
365: }
366:
1.61 noro 367: void ndl_weight_mask(UINT *d)
1.43 noro 368: {
1.61 noro 369: UINT t,u;
370: UINT *mask;
371: int i,j,k,l;
1.43 noro 372:
1.61 noro 373: l = nd_blockmask->n;
374: for ( k = 0; k < l; k++ ) {
375: mask = nd_blockmask->mask[k];
376: if ( current_dl_weight_vector )
377: for ( i = 0, t = 0; i < nd_nvar; i++ ) {
378: u = GET_EXP_MASK(d,i,mask);
379: t += MUL_WEIGHT(u,i);
380: }
381: else
382: for ( t = 0, i = nd_exporigin; i < nd_wpd; i++ ) {
383: u = d[i]&mask[i];
384: for ( j = 0; j < nd_epw; j++, u>>=nd_bpe )
385: t += (u&nd_mask0);
386: }
387: d[k+1] = t;
388: }
1.43 noro 389: }
390:
1.61 noro 391: int ndl_lex_compare(UINT *d1,UINT *d2)
1.1 noro 392: {
393: int i;
394:
1.41 noro 395: d1 += nd_exporigin;
396: d2 += nd_exporigin;
397: for ( i = nd_exporigin; i < nd_wpd; i++, d1++, d2++ )
1.1 noro 398: if ( *d1 > *d2 )
1.32 noro 399: return nd_isrlex ? -1 : 1;
1.1 noro 400: else if ( *d1 < *d2 )
1.32 noro 401: return nd_isrlex ? 1 : -1;
1.1 noro 402: return 0;
403: }
404:
1.61 noro 405: int ndl_block_compare(UINT *d1,UINT *d2)
1.43 noro 406: {
407: int i,l,j,ord_o,ord_l;
408: struct order_pair *op;
1.61 noro 409: UINT t1,t2,m;
410: UINT *mask;
1.43 noro 411:
412: l = nd_blockmask->n;
413: op = nd_blockmask->order_pair;
414: for ( j = 0; j < l; j++ ) {
415: mask = nd_blockmask->mask[j];
416: ord_o = op[j].order;
417: if ( ord_o < 2 )
1.44 noro 418: if ( (t1=d1[j+1]) > (t2=d2[j+1]) ) return 1;
419: else if ( t1 < t2 ) return -1;
1.43 noro 420: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.44 noro 421: m = mask[i];
422: t1 = d1[i]&m;
423: t2 = d2[i]&m;
1.43 noro 424: if ( t1 > t2 )
425: return !ord_o ? -1 : 1;
426: else if ( t1 < t2 )
427: return !ord_o ? 1 : -1;
428: }
429: }
430: return 0;
431: }
432:
1.96 noro 433: int ndl_matrix_compare(UINT *d1,UINT *d2)
434: {
435: int i,j,s;
436: int *v;
437:
438: for ( j = 0; j < nd_nvar; j++ )
439: nd_work_vector[j] = GET_EXP(d1,j)-GET_EXP(d2,j);
440: for ( i = 0; i < nd_matrix_len; i++ ) {
441: v = nd_matrix[i];
442: for ( j = 0, s = 0; j < nd_nvar; j++ )
443: s += v[j]*nd_work_vector[j];
444: if ( s > 0 ) return 1;
445: else if ( s < 0 ) return -1;
446: }
447: return 0;
448: }
449:
1.97 noro 450: int ndl_composite_compare(UINT *d1,UINT *d2)
451: {
452: int i,j,s,start,end,len,o;
453: int *v;
454: struct sparse_weight *sw;
455:
456: for ( j = 0; j < nd_nvar; j++ )
457: nd_work_vector[j] = GET_EXP(d1,j)-GET_EXP(d2,j);
458: for ( i = 0; i < nd_worb_len; i++ ) {
459: len = nd_worb[i].length;
460: switch ( nd_worb[i].type ) {
461: case IS_DENSE_WEIGHT:
462: v = nd_worb[i].body.dense_weight;
463: for ( j = 0, s = 0; j < len; j++ )
464: s += v[j]*nd_work_vector[j];
465: if ( s > 0 ) return 1;
466: else if ( s < 0 ) return -1;
467: break;
468: case IS_SPARSE_WEIGHT:
469: sw = nd_worb[i].body.sparse_weight;
470: for ( j = 0, s = 0; j < len; j++ )
471: s += sw[j].value*nd_work_vector[sw[j].pos];
472: if ( s > 0 ) return 1;
473: else if ( s < 0 ) return -1;
474: break;
475: case IS_BLOCK:
476: o = nd_worb[i].body.block.order;
477: start = nd_worb[i].body.block.start;
478: switch ( o ) {
479: case 0:
480: end = start+len;
481: for ( j = start, s = 0; j < end; j++ )
482: s += MUL_WEIGHT(nd_work_vector[j],j);
483: if ( s > 0 ) return 1;
484: else if ( s < 0 ) return -1;
485: for ( j = end-1; j >= start; j-- )
486: if ( nd_work_vector[j] < 0 ) return 1;
487: else if ( nd_work_vector[j] > 0 ) return -1;
488: break;
489: case 1:
490: end = start+len;
491: for ( j = start, s = 0; j < end; j++ )
492: s += MUL_WEIGHT(nd_work_vector[j],j);
493: if ( s > 0 ) return 1;
494: else if ( s < 0 ) return -1;
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: case 2:
500: for ( j = start; j < end; j++ )
501: if ( nd_work_vector[j] > 0 ) return 1;
502: else if ( nd_work_vector[j] < 0 ) return -1;
503: break;
504: }
505: break;
506: }
507: }
508: return 0;
509: }
510:
1.58 noro 511: /* TDH -> WW -> TD-> RL */
512:
1.61 noro 513: int ndl_ww_lex_compare(UINT *d1,UINT *d2)
1.58 noro 514: {
515: int i,m,e1,e2;
516:
517: if ( TD(d1) > TD(d2) ) return 1;
518: else if ( TD(d1) < TD(d2) ) return -1;
519: m = nd_nvar>>1;
520: for ( i = 0, e1 = e2 = 0; i < m; i++ ) {
521: e1 += current_weyl_weight_vector[i]*(GET_EXP(d1,m+i)-GET_EXP(d1,i));
522: e2 += current_weyl_weight_vector[i]*(GET_EXP(d2,m+i)-GET_EXP(d2,i));
523: }
524: if ( e1 > e2 ) return 1;
525: else if ( e1 < e2 ) return -1;
526: return ndl_lex_compare(d1,d2);
527: }
528:
1.61 noro 529: INLINE int ndl_equal(UINT *d1,UINT *d2)
1.1 noro 530: {
531: int i;
532:
1.81 noro 533: switch ( nd_wpd ) {
534: case 2:
535: if ( TD(d2) != TD(d1) ) return 0;
536: if ( d2[1] != d1[1] ) return 0;
537: return 1;
538: break;
539: case 3:
540: if ( TD(d2) != TD(d1) ) return 0;
541: if ( d2[1] != d1[1] ) return 0;
542: if ( d2[2] != d1[2] ) return 0;
543: return 1;
544: break;
545: default:
546: for ( i = 0; i < nd_wpd; i++ )
547: if ( *d1++ != *d2++ ) return 0;
548: return 1;
549: break;
550: }
1.1 noro 551: }
552:
1.61 noro 553: INLINE void ndl_copy(UINT *d1,UINT *d2)
1.6 noro 554: {
555: int i;
556:
557: switch ( nd_wpd ) {
1.41 noro 558: case 2:
1.34 noro 559: TD(d2) = TD(d1);
560: d2[1] = d1[1];
1.6 noro 561: break;
1.41 noro 562: case 3:
1.34 noro 563: TD(d2) = TD(d1);
1.6 noro 564: d2[1] = d1[1];
1.34 noro 565: d2[2] = d1[2];
1.6 noro 566: break;
567: default:
1.41 noro 568: for ( i = 0; i < nd_wpd; i++ )
1.6 noro 569: d2[i] = d1[i];
570: break;
571: }
572: }
573:
1.61 noro 574: INLINE void ndl_zero(UINT *d)
575: {
576: int i;
577: for ( i = 0; i < nd_wpd; i++ ) d[i] = 0;
578: }
579:
580: INLINE void ndl_add(UINT *d1,UINT *d2,UINT *d)
1.1 noro 581: {
582: int i;
583:
1.43 noro 584: #if 1
1.6 noro 585: switch ( nd_wpd ) {
1.41 noro 586: case 2:
587: TD(d) = TD(d1)+TD(d2);
1.34 noro 588: d[1] = d1[1]+d2[1];
1.6 noro 589: break;
1.41 noro 590: case 3:
591: TD(d) = TD(d1)+TD(d2);
1.6 noro 592: d[1] = d1[1]+d2[1];
1.34 noro 593: d[2] = d1[2]+d2[2];
1.6 noro 594: break;
595: default:
1.43 noro 596: for ( i = 0; i < nd_wpd; i++ ) d[i] = d1[i]+d2[i];
1.6 noro 597: break;
598: }
1.43 noro 599: #else
600: for ( i = 0; i < nd_wpd; i++ ) d[i] = d1[i]+d2[i];
601: #endif
1.6 noro 602: }
603:
1.55 noro 604: /* d1 += d2 */
1.61 noro 605: INLINE void ndl_addto(UINT *d1,UINT *d2)
1.55 noro 606: {
607: int i;
608:
609: #if 1
610: switch ( nd_wpd ) {
611: case 2:
612: TD(d1) += TD(d2);
613: d1[1] += d2[1];
614: break;
615: case 3:
616: TD(d1) += TD(d2);
617: d1[1] += d2[1];
618: d1[2] += d2[2];
619: break;
620: default:
621: for ( i = 0; i < nd_wpd; i++ ) d1[i] += d2[i];
622: break;
623: }
624: #else
625: for ( i = 0; i < nd_wpd; i++ ) d1[i] += d2[i];
626: #endif
627: }
628:
1.61 noro 629: INLINE void ndl_sub(UINT *d1,UINT *d2,UINT *d)
1.6 noro 630: {
631: int i;
632:
1.43 noro 633: for ( i = 0; i < nd_wpd; i++ ) d[i] = d1[i]-d2[i];
1.1 noro 634: }
635:
1.61 noro 636: int ndl_disjoint(UINT *d1,UINT *d2)
1.1 noro 637: {
1.61 noro 638: UINT t1,t2,u,u1,u2;
1.1 noro 639: int i,j;
640:
1.65 noro 641: #if USE_UNROLL
1.1 noro 642: switch ( nd_bpe ) {
1.62 noro 643: case 3:
644: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
645: u1 = d1[i]; u2 = d2[i];
646: t1 = u1&0x38000000; t2 = u2&0x38000000; if ( t1&&t2 ) return 0;
647: t1 = u1& 0x7000000; t2 = u2& 0x7000000; if ( t1&&t2 ) return 0;
648: t1 = u1& 0xe00000; t2 = u2& 0xe00000; if ( t1&&t2 ) return 0;
649: t1 = u1& 0x1c0000; t2 = u2& 0x1c0000; if ( t1&&t2 ) return 0;
650: t1 = u1& 0x38000; t2 = u2& 0x38000; if ( t1&&t2 ) return 0;
651: t1 = u1& 0x7000; t2 = u2& 0x7000; if ( t1&&t2 ) return 0;
652: t1 = u1& 0xe00; t2 = u2& 0xe00; if ( t1&&t2 ) return 0;
653: t1 = u1& 0x1c0; t2 = u2& 0x1c0; if ( t1&&t2 ) return 0;
654: t1 = u1& 0x38; t2 = u2& 0x38; if ( t1&&t2 ) return 0;
655: t1 = u1& 0x7; t2 = u2& 0x7; if ( t1&&t2 ) return 0;
656: }
657: return 1;
658: break;
1.1 noro 659: case 4:
1.41 noro 660: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 661: u1 = d1[i]; u2 = d2[i];
662: t1 = u1&0xf0000000; t2 = u2&0xf0000000; if ( t1&&t2 ) return 0;
1.62 noro 663: t1 = u1& 0xf000000; t2 = u2& 0xf000000; if ( t1&&t2 ) return 0;
664: t1 = u1& 0xf00000; t2 = u2& 0xf00000; if ( t1&&t2 ) return 0;
665: t1 = u1& 0xf0000; t2 = u2& 0xf0000; if ( t1&&t2 ) return 0;
666: t1 = u1& 0xf000; t2 = u2& 0xf000; if ( t1&&t2 ) return 0;
667: t1 = u1& 0xf00; t2 = u2& 0xf00; if ( t1&&t2 ) return 0;
668: t1 = u1& 0xf0; t2 = u2& 0xf0; if ( t1&&t2 ) return 0;
669: t1 = u1& 0xf; t2 = u2& 0xf; if ( t1&&t2 ) return 0;
1.1 noro 670: }
671: return 1;
672: break;
673: case 6:
1.41 noro 674: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 675: u1 = d1[i]; u2 = d2[i];
676: t1 = u1&0x3f000000; t2 = u2&0x3f000000; if ( t1&&t2 ) return 0;
1.62 noro 677: t1 = u1& 0xfc0000; t2 = u2& 0xfc0000; if ( t1&&t2 ) return 0;
678: t1 = u1& 0x3f000; t2 = u2& 0x3f000; if ( t1&&t2 ) return 0;
679: t1 = u1& 0xfc0; t2 = u2& 0xfc0; if ( t1&&t2 ) return 0;
680: t1 = u1& 0x3f; t2 = u2& 0x3f; if ( t1&&t2 ) return 0;
1.1 noro 681: }
682: return 1;
683: break;
684: case 8:
1.41 noro 685: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 686: u1 = d1[i]; u2 = d2[i];
687: t1 = u1&0xff000000; t2 = u2&0xff000000; if ( t1&&t2 ) return 0;
1.62 noro 688: t1 = u1& 0xff0000; t2 = u2& 0xff0000; if ( t1&&t2 ) return 0;
689: t1 = u1& 0xff00; t2 = u2& 0xff00; if ( t1&&t2 ) return 0;
690: t1 = u1& 0xff; t2 = u2& 0xff; if ( t1&&t2 ) return 0;
1.1 noro 691: }
692: return 1;
693: break;
694: case 16:
1.41 noro 695: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 696: u1 = d1[i]; u2 = d2[i];
697: t1 = u1&0xffff0000; t2 = u2&0xffff0000; if ( t1&&t2 ) return 0;
1.62 noro 698: t1 = u1& 0xffff; t2 = u2& 0xffff; if ( t1&&t2 ) return 0;
1.1 noro 699: }
700: return 1;
701: break;
702: case 32:
1.41 noro 703: for ( i = nd_exporigin; i < nd_wpd; i++ )
1.1 noro 704: if ( d1[i] && d2[i] ) return 0;
705: return 1;
706: break;
707: default:
1.41 noro 708: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 709: u1 = d1[i]; u2 = d2[i];
710: for ( j = 0; j < nd_epw; j++ ) {
711: if ( (u1&nd_mask0) && (u2&nd_mask0) ) return 0;
712: u1 >>= nd_bpe; u2 >>= nd_bpe;
713: }
714: }
715: return 1;
716: break;
717: }
1.65 noro 718: #else
719: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
720: u1 = d1[i]; u2 = d2[i];
721: for ( j = 0; j < nd_epw; j++ ) {
722: if ( (u1&nd_mask0) && (u2&nd_mask0) ) return 0;
723: u1 >>= nd_bpe; u2 >>= nd_bpe;
724: }
725: }
726: return 1;
727: #endif
1.1 noro 728: }
729:
1.114 noro 730: int ndl_check_bound(UINT *d1,UINT *d2)
1.1 noro 731: {
1.61 noro 732: UINT u2;
1.5 noro 733: int i,j,ind,k;
1.1 noro 734:
1.5 noro 735: ind = 0;
1.65 noro 736: #if USE_UNROLL
1.5 noro 737: switch ( nd_bpe ) {
1.62 noro 738: case 3:
739: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
740: u2 = d2[i];
741: if ( d1[ind++]+((u2>>27)&0x7) >= 0x8 ) return 1;
742: if ( d1[ind++]+((u2>>24)&0x7) >= 0x8 ) return 1;
743: if ( d1[ind++]+((u2>>21)&0x7) >= 0x8 ) return 1;
744: if ( d1[ind++]+((u2>>18)&0x7) >= 0x8 ) return 1;
745: if ( d1[ind++]+((u2>>15)&0x7) >= 0x8 ) return 1;
746: if ( d1[ind++]+((u2>>12)&0x7) >= 0x8 ) return 1;
747: if ( d1[ind++]+((u2>>9)&0x7) >= 0x8 ) return 1;
748: if ( d1[ind++]+((u2>>6)&0x7) >= 0x8 ) return 1;
749: if ( d1[ind++]+((u2>>3)&0x7) >= 0x8 ) return 1;
750: if ( d1[ind++]+(u2&0x7) >= 0x8 ) return 1;
751: }
752: return 0;
753: break;
1.5 noro 754: case 4:
1.41 noro 755: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 756: u2 = d2[i];
757: if ( d1[ind++]+((u2>>28)&0xf) >= 0x10 ) return 1;
758: if ( d1[ind++]+((u2>>24)&0xf) >= 0x10 ) return 1;
759: if ( d1[ind++]+((u2>>20)&0xf) >= 0x10 ) return 1;
760: if ( d1[ind++]+((u2>>16)&0xf) >= 0x10 ) return 1;
761: if ( d1[ind++]+((u2>>12)&0xf) >= 0x10 ) return 1;
762: if ( d1[ind++]+((u2>>8)&0xf) >= 0x10 ) return 1;
763: if ( d1[ind++]+((u2>>4)&0xf) >= 0x10 ) return 1;
764: if ( d1[ind++]+(u2&0xf) >= 0x10 ) return 1;
765: }
766: return 0;
767: break;
768: case 6:
1.41 noro 769: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 770: u2 = d2[i];
771: if ( d1[ind++]+((u2>>24)&0x3f) >= 0x40 ) return 1;
772: if ( d1[ind++]+((u2>>18)&0x3f) >= 0x40 ) return 1;
773: if ( d1[ind++]+((u2>>12)&0x3f) >= 0x40 ) return 1;
774: if ( d1[ind++]+((u2>>6)&0x3f) >= 0x40 ) return 1;
775: if ( d1[ind++]+(u2&0x3f) >= 0x40 ) return 1;
776: }
777: return 0;
778: break;
779: case 8:
1.41 noro 780: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 781: u2 = d2[i];
782: if ( d1[ind++]+((u2>>24)&0xff) >= 0x100 ) return 1;
783: if ( d1[ind++]+((u2>>16)&0xff) >= 0x100 ) return 1;
784: if ( d1[ind++]+((u2>>8)&0xff) >= 0x100 ) return 1;
785: if ( d1[ind++]+(u2&0xff) >= 0x100 ) return 1;
786: }
787: return 0;
788: break;
789: case 16:
1.41 noro 790: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 791: u2 = d2[i];
792: if ( d1[ind++]+((u2>>16)&0xffff) > 0x10000 ) return 1;
793: if ( d1[ind++]+(u2&0xffff) > 0x10000 ) return 1;
794: }
795: return 0;
796: break;
797: case 32:
1.41 noro 798: for ( i = nd_exporigin; i < nd_wpd; i++ )
1.5 noro 799: if ( d1[i]+d2[i]<d1[i] ) return 1;
800: return 0;
801: break;
802: default:
1.41 noro 803: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 804: u2 = d2[i];
805: k = (nd_epw-1)*nd_bpe;
806: for ( j = 0; j < nd_epw; j++, k -= nd_bpe )
807: if ( d1[ind++]+((u2>>k)&nd_mask0) > nd_mask0 ) return 1;
808: }
809: return 0;
810: break;
811: }
1.65 noro 812: #else
813: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
814: u2 = d2[i];
815: k = (nd_epw-1)*nd_bpe;
816: for ( j = 0; j < nd_epw; j++, k -= nd_bpe )
817: if ( d1[ind++]+((u2>>k)&nd_mask0) > nd_mask0 ) return 1;
818: }
819: return 0;
820: #endif
1.1 noro 821: }
822:
1.114 noro 823: int ndl_check_bound2(int index,UINT *d2)
824: {
825: return ndl_check_bound(nd_bound[index],d2);
826: }
827:
1.61 noro 828: INLINE int ndl_hash_value(UINT *d)
1.1 noro 829: {
830: int i;
831: int r;
832:
1.34 noro 833: r = 0;
1.41 noro 834: for ( i = 0; i < nd_wpd; i++ )
1.1 noro 835: r = ((r<<16)+d[i])%REDTAB_LEN;
836: return r;
837: }
838:
1.63 noro 839: INLINE int ndl_find_reducer(UINT *dg)
1.1 noro 840: {
1.13 noro 841: RHist r;
1.6 noro 842: int d,k,i;
1.1 noro 843:
1.63 noro 844: d = ndl_hash_value(dg);
1.13 noro 845: for ( r = nd_red[d], k = 0; r; r = NEXT(r), k++ ) {
1.34 noro 846: if ( ndl_equal(dg,DL(r)) ) {
1.1 noro 847: if ( k > 0 ) nd_notfirst++;
848: nd_found++;
1.13 noro 849: return r->index;
1.1 noro 850: }
851: }
1.13 noro 852: if ( Reverse )
853: for ( i = nd_psn-1; i >= 0; i-- ) {
854: r = nd_psh[i];
1.34 noro 855: if ( ndl_reducible(dg,DL(r)) ) {
1.13 noro 856: nd_create++;
1.34 noro 857: nd_append_red(dg,i);
1.13 noro 858: return i;
859: }
860: }
861: else
862: for ( i = 0; i < nd_psn; i++ ) {
863: r = nd_psh[i];
1.34 noro 864: if ( ndl_reducible(dg,DL(r)) ) {
1.13 noro 865: nd_create++;
1.34 noro 866: nd_append_red(dg,i);
1.13 noro 867: return i;
868: }
1.1 noro 869: }
1.6 noro 870: return -1;
1.1 noro 871: }
872:
1.63 noro 873: ND nd_merge(ND p1,ND p2)
874: {
875: int n,c;
876: int t,can,td1,td2;
877: ND r;
878: NM m1,m2,mr0,mr,s;
879:
880: if ( !p1 ) return p2;
881: else if ( !p2 ) return p1;
882: else {
883: can = 0;
884: for ( n = NV(p1), m1 = BDY(p1), m2 = BDY(p2), mr0 = 0; m1 && m2; ) {
885: c = DL_COMPARE(DL(m1),DL(m2));
886: switch ( c ) {
887: case 0:
888: s = m1; m1 = NEXT(m1);
889: can++; NEXTNM2(mr0,mr,s);
890: s = m2; m2 = NEXT(m2); FREENM(s);
891: break;
892: case 1:
893: s = m1; m1 = NEXT(m1); NEXTNM2(mr0,mr,s);
894: break;
895: case -1:
896: s = m2; m2 = NEXT(m2); NEXTNM2(mr0,mr,s);
897: break;
898: }
899: }
900: if ( !mr0 )
901: if ( m1 ) mr0 = m1;
902: else if ( m2 ) mr0 = m2;
903: else return 0;
904: else if ( m1 ) NEXT(mr) = m1;
905: else if ( m2 ) NEXT(mr) = m2;
906: else NEXT(mr) = 0;
907: BDY(p1) = mr0;
908: SG(p1) = MAX(SG(p1),SG(p2));
909: LEN(p1) = LEN(p1)+LEN(p2)-can;
910: FREEND(p2);
911: return p1;
912: }
913: }
914:
1.31 noro 915: ND nd_add(int mod,ND p1,ND p2)
1.1 noro 916: {
917: int n,c;
1.34 noro 918: int t,can,td1,td2;
1.1 noro 919: ND r;
920: NM m1,m2,mr0,mr,s;
921:
1.34 noro 922: if ( !p1 ) return p2;
923: else if ( !p2 ) return p1;
1.71 noro 924: else if ( mod == -1 ) return nd_add_sf(p1,p2);
1.113 noro 925: else if ( !mod ) return nd_add_q(p1,p2);
1.1 noro 926: else {
1.30 noro 927: can = 0;
1.1 noro 928: for ( n = NV(p1), m1 = BDY(p1), m2 = BDY(p2), mr0 = 0; m1 && m2; ) {
1.34 noro 929: c = DL_COMPARE(DL(m1),DL(m2));
1.1 noro 930: switch ( c ) {
931: case 0:
1.19 noro 932: t = ((CM(m1))+(CM(m2))) - mod;
1.34 noro 933: if ( t < 0 ) t += mod;
1.1 noro 934: s = m1; m1 = NEXT(m1);
935: if ( t ) {
1.34 noro 936: can++; NEXTNM2(mr0,mr,s); CM(mr) = (t);
1.17 noro 937: } else {
1.34 noro 938: can += 2; FREENM(s);
1.17 noro 939: }
940: s = m2; m2 = NEXT(m2); FREENM(s);
1.95 noro 941: break;
942: case 1:
943: s = m1; m1 = NEXT(m1); NEXTNM2(mr0,mr,s);
944: break;
945: case -1:
946: s = m2; m2 = NEXT(m2); NEXTNM2(mr0,mr,s);
947: break;
948: }
949: }
950: if ( !mr0 )
951: if ( m1 ) mr0 = m1;
952: else if ( m2 ) mr0 = m2;
953: else return 0;
954: else if ( m1 ) NEXT(mr) = m1;
955: else if ( m2 ) NEXT(mr) = m2;
956: else NEXT(mr) = 0;
957: BDY(p1) = mr0;
958: SG(p1) = MAX(SG(p1),SG(p2));
959: LEN(p1) = LEN(p1)+LEN(p2)-can;
960: FREEND(p2);
961: return p1;
962: }
963: }
964:
965: /* XXX on opteron, the inlined manipulation of destructive additon of
966: * two NM seems to make gcc optimizer get confused, so the part is
967: * done in a function.
968: */
969:
1.113 noro 970: int nm_destructive_add_q(NM *m1,NM *m2,NM *mr0,NM *mr)
1.95 noro 971: {
972: NM s;
1.113 noro 973: Q t;
1.95 noro 974: int can;
975:
1.113 noro 976: addq(CQ(*m1),CQ(*m2),&t);
1.95 noro 977: s = *m1; *m1 = NEXT(*m1);
978: if ( t ) {
1.113 noro 979: can = 1; NEXTNM2(*mr0,*mr,s); CQ(*mr) = (t);
1.95 noro 980: } else {
981: can = 2; FREENM(s);
982: }
983: s = *m2; *m2 = NEXT(*m2); FREENM(s);
984: return can;
985: }
986:
1.113 noro 987: ND nd_add_q(ND p1,ND p2)
1.95 noro 988: {
989: int n,c,can;
990: ND r;
991: NM m1,m2,mr0,mr,s;
1.113 noro 992: Q t;
1.95 noro 993:
994: if ( !p1 ) return p2;
995: else if ( !p2 ) return p1;
996: else {
997: can = 0;
998: for ( n = NV(p1), m1 = BDY(p1), m2 = BDY(p2), mr0 = 0; m1 && m2; ) {
999: c = DL_COMPARE(DL(m1),DL(m2));
1000: switch ( c ) {
1001: case 0:
1002: #if defined(__x86_64__)
1.113 noro 1003: can += nm_destructive_add_q(&m1,&m2,&mr0,&mr);
1.95 noro 1004: #else
1.113 noro 1005: addq(CQ(m1),CQ(m2),&t);
1.95 noro 1006: s = m1; m1 = NEXT(m1);
1007: if ( t ) {
1.113 noro 1008: can++; NEXTNM2(mr0,mr,s); CQ(mr) = (t);
1.95 noro 1009: } else {
1010: can += 2; FREENM(s);
1011: }
1012: s = m2; m2 = NEXT(m2); FREENM(s);
1013: #endif
1.17 noro 1014: break;
1015: case 1:
1016: s = m1; m1 = NEXT(m1); NEXTNM2(mr0,mr,s);
1017: break;
1018: case -1:
1019: s = m2; m2 = NEXT(m2); NEXTNM2(mr0,mr,s);
1020: break;
1021: }
1022: }
1023: if ( !mr0 )
1.34 noro 1024: if ( m1 ) mr0 = m1;
1025: else if ( m2 ) mr0 = m2;
1026: else return 0;
1027: else if ( m1 ) NEXT(mr) = m1;
1028: else if ( m2 ) NEXT(mr) = m2;
1029: else NEXT(mr) = 0;
1.17 noro 1030: BDY(p1) = mr0;
1031: SG(p1) = MAX(SG(p1),SG(p2));
1.31 noro 1032: LEN(p1) = LEN(p1)+LEN(p2)-can;
1.17 noro 1033: FREEND(p2);
1034: return p1;
1035: }
1036: }
1037:
1.71 noro 1038: ND nd_add_sf(ND p1,ND p2)
1039: {
1040: int n,c,can;
1041: ND r;
1042: NM m1,m2,mr0,mr,s;
1043: int t;
1044:
1045: if ( !p1 ) return p2;
1046: else if ( !p2 ) return p1;
1047: else {
1048: can = 0;
1049: for ( n = NV(p1), m1 = BDY(p1), m2 = BDY(p2), mr0 = 0; m1 && m2; ) {
1050: c = DL_COMPARE(DL(m1),DL(m2));
1051: switch ( c ) {
1052: case 0:
1053: t = _addsf(CM(m1),CM(m2));
1054: s = m1; m1 = NEXT(m1);
1055: if ( t ) {
1056: can++; NEXTNM2(mr0,mr,s); CM(mr) = (t);
1057: } else {
1058: can += 2; FREENM(s);
1059: }
1060: s = m2; m2 = NEXT(m2); FREENM(s);
1061: break;
1062: case 1:
1063: s = m1; m1 = NEXT(m1); NEXTNM2(mr0,mr,s);
1064: break;
1065: case -1:
1066: s = m2; m2 = NEXT(m2); NEXTNM2(mr0,mr,s);
1067: break;
1068: }
1069: }
1070: if ( !mr0 )
1071: if ( m1 ) mr0 = m1;
1072: else if ( m2 ) mr0 = m2;
1073: else return 0;
1074: else if ( m1 ) NEXT(mr) = m1;
1075: else if ( m2 ) NEXT(mr) = m2;
1076: else NEXT(mr) = 0;
1077: BDY(p1) = mr0;
1078: SG(p1) = MAX(SG(p1),SG(p2));
1079: LEN(p1) = LEN(p1)+LEN(p2)-can;
1080: FREEND(p2);
1081: return p1;
1082: }
1083: }
1084:
1.1 noro 1085: /* ret=1 : success, ret=0 : overflow */
1.113 noro 1086: int nd_nf(int mod,ND g,NDV *ps,int full,NDC dn,ND *rp)
1.1 noro 1087: {
1.11 noro 1088: ND d;
1.1 noro 1089: NM m,mrd,tail;
1.7 noro 1090: NM mul;
1.10 noro 1091: int n,sugar,psugar,sugar0,stat,index;
1.30 noro 1092: int c,c1,c2,dummy;
1.17 noro 1093: RHist h;
1.11 noro 1094: NDV p,red;
1.113 noro 1095: Q cg,cred,gcd,tq,qq;
1.21 noro 1096: double hmag;
1.1 noro 1097:
1.113 noro 1098: if ( dn ) {
1099: if ( mod )
1100: dn->m = 1;
1101: else
1102: dn->z = ONE;
1103: }
1.1 noro 1104: if ( !g ) {
1105: *rp = 0;
1106: return 1;
1107: }
1.113 noro 1108: if ( !mod ) hmag = ((double)p_mag((P)HCQ(g)))*nd_scale;
1.21 noro 1109:
1.14 noro 1110: sugar0 = sugar = SG(g);
1.1 noro 1111: n = NV(g);
1.61 noro 1112: mul = (NM)ALLOCA(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT));
1.1 noro 1113: for ( d = 0; g; ) {
1.63 noro 1114: index = ndl_find_reducer(HDL(g));
1.6 noro 1115: if ( index >= 0 ) {
1.17 noro 1116: h = nd_psh[index];
1117: ndl_sub(HDL(g),DL(h),DL(mul));
1.14 noro 1118: if ( ndl_check_bound2(index,DL(mul)) ) {
1.6 noro 1119: nd_free(g); nd_free(d);
1120: return 0;
1121: }
1.77 noro 1122: if ( nd_demand )
1123: p = ndv_load(index);
1124: else
1125: p = ps[index];
1.71 noro 1126: if ( mod == -1 )
1127: CM(mul) = _mulsf(_invsf(HCM(p)),_chsgnsf(HCM(g)));
1128: else if ( mod ) {
1.19 noro 1129: c1 = invm(HCM(p),mod); c2 = mod-HCM(g);
1130: DMAR(c1,c2,0,mod,c); CM(mul) = c;
1.16 noro 1131: } else {
1.113 noro 1132: igcd_cofactor(HCQ(g),HCQ(p),&gcd,&cg,&cred);
1133: chsgnq(cg,&CQ(mul));
1134: nd_mul_c_q(d,cred); nd_mul_c_q(g,cred);
1.69 noro 1135: if ( dn ) {
1.113 noro 1136: mulq(dn->z,cred,&tq); dn->z = tq;
1.69 noro 1137: }
1.16 noro 1138: }
1.55 noro 1139: g = nd_add(mod,g,ndv_mul_nm(mod,mul,p));
1.34 noro 1140: sugar = MAX(sugar,SG(p)+TD(DL(mul)));
1.113 noro 1141: if ( !mod && hmag && g && ((double)(p_mag((P)HCQ(g))) > hmag) ) {
1142: tq = HCQ(g);
1.21 noro 1143: nd_removecont2(d,g);
1.69 noro 1144: if ( dn ) {
1.113 noro 1145: divq(tq,HCQ(g),&qq); divq(dn->z,qq,&tq); dn->z = tq;
1.69 noro 1146: }
1.113 noro 1147: hmag = ((double)p_mag((P)HCQ(g)))*nd_scale;
1.21 noro 1148: }
1.1 noro 1149: } else if ( !full ) {
1150: *rp = g;
1151: return 1;
1152: } else {
1153: m = BDY(g);
1154: if ( NEXT(m) ) {
1.34 noro 1155: BDY(g) = NEXT(m); NEXT(m) = 0; LEN(g)--;
1.1 noro 1156: } else {
1157: FREEND(g); g = 0;
1158: }
1159: if ( d ) {
1.34 noro 1160: NEXT(tail)=m; tail=m; LEN(d)++;
1.1 noro 1161: } else {
1.34 noro 1162: MKND(n,m,1,d); tail = BDY(d);
1.1 noro 1163: }
1164: }
1165: }
1.34 noro 1166: if ( d ) SG(d) = sugar;
1.1 noro 1167: *rp = d;
1168: return 1;
1169: }
1.28 noro 1170:
1.53 noro 1171: int nd_nf_pbucket(int mod,ND g,NDV *ps,int full,ND *rp)
1.25 noro 1172: {
1173: int hindex,index;
1174: NDV p;
1175: ND u,d,red;
1176: NODE l;
1.31 noro 1177: NM mul,m,mrd,tail;
1.25 noro 1178: int sugar,psugar,n,h_reducible;
1179: PGeoBucket bucket;
1180: int c,c1,c2;
1.113 noro 1181: Q cg,cred,gcd,zzz;
1.25 noro 1182: RHist h;
1.28 noro 1183: double hmag,gmag;
1.77 noro 1184: int count = 0;
1185: int hcount = 0;
1.25 noro 1186:
1187: if ( !g ) {
1188: *rp = 0;
1189: return 1;
1190: }
1191: sugar = SG(g);
1192: n = NV(g);
1.113 noro 1193: if ( !mod ) hmag = ((double)p_mag((P)HCQ(g)))*nd_scale;
1.25 noro 1194: bucket = create_pbucket();
1.31 noro 1195: add_pbucket(mod,bucket,g);
1.25 noro 1196: d = 0;
1.61 noro 1197: mul = (NM)ALLOCA(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT));
1.25 noro 1198: while ( 1 ) {
1.113 noro 1199: hindex = mod?head_pbucket(mod,bucket):head_pbucket_q(bucket);
1.25 noro 1200: if ( hindex < 0 ) {
1.77 noro 1201: if ( DP_Print > 3 ) printf("(%d %d)",count,hcount);
1.34 noro 1202: if ( d ) SG(d) = sugar;
1.25 noro 1203: *rp = d;
1204: return 1;
1205: }
1206: g = bucket->body[hindex];
1.63 noro 1207: index = ndl_find_reducer(HDL(g));
1.25 noro 1208: if ( index >= 0 ) {
1.77 noro 1209: count++;
1210: if ( !d ) hcount++;
1.25 noro 1211: h = nd_psh[index];
1212: ndl_sub(HDL(g),DL(h),DL(mul));
1213: if ( ndl_check_bound2(index,DL(mul)) ) {
1.26 noro 1214: nd_free(d);
1.25 noro 1215: free_pbucket(bucket);
1216: *rp = 0;
1217: return 0;
1218: }
1.53 noro 1219: p = ps[index];
1.71 noro 1220: if ( mod == -1 )
1221: CM(mul) = _mulsf(_invsf(HCM(p)),_chsgnsf(HCM(g)));
1222: else if ( mod ) {
1.25 noro 1223: c1 = invm(HCM(p),mod); c2 = mod-HCM(g);
1224: DMAR(c1,c2,0,mod,c); CM(mul) = c;
1225: } else {
1.113 noro 1226: igcd_cofactor(HCQ(g),HCQ(p),&gcd,&cg,&cred);
1227: chsgnq(cg,&CQ(mul));
1228: nd_mul_c_q(d,cred);
1229: mulq_pbucket(bucket,cred);
1.26 noro 1230: g = bucket->body[hindex];
1.113 noro 1231: gmag = (double)p_mag((P)HCQ(g));
1.25 noro 1232: }
1.55 noro 1233: red = ndv_mul_nm(mod,mul,p);
1.25 noro 1234: bucket->body[hindex] = nd_remove_head(g);
1235: red = nd_remove_head(red);
1.31 noro 1236: add_pbucket(mod,bucket,red);
1.34 noro 1237: psugar = SG(p)+TD(DL(mul));
1238: sugar = MAX(sugar,psugar);
1.28 noro 1239: if ( !mod && hmag && (gmag > hmag) ) {
1240: g = normalize_pbucket(mod,bucket);
1241: if ( !g ) {
1.34 noro 1242: if ( d ) SG(d) = sugar;
1.28 noro 1243: *rp = d;
1244: return 1;
1245: }
1246: nd_removecont2(d,g);
1.113 noro 1247: hmag = ((double)p_mag((P)HCQ(g)))*nd_scale;
1.31 noro 1248: add_pbucket(mod,bucket,g);
1.28 noro 1249: }
1.25 noro 1250: } else if ( !full ) {
1251: g = normalize_pbucket(mod,bucket);
1.34 noro 1252: if ( g ) SG(g) = sugar;
1.25 noro 1253: *rp = g;
1254: return 1;
1255: } else {
1256: m = BDY(g);
1257: if ( NEXT(m) ) {
1.34 noro 1258: BDY(g) = NEXT(m); NEXT(m) = 0; LEN(g)--;
1.25 noro 1259: } else {
1260: FREEND(g); g = 0;
1261: }
1262: bucket->body[hindex] = g;
1263: NEXT(m) = 0;
1264: if ( d ) {
1.34 noro 1265: NEXT(tail)=m; tail=m; LEN(d)++;
1.25 noro 1266: } else {
1.34 noro 1267: MKND(n,m,1,d); tail = BDY(d);
1.25 noro 1268: }
1269: }
1270: }
1271: }
1.27 noro 1272:
1.61 noro 1273: /* input : list of NDV, cand : list of NDV */
1.28 noro 1274:
1.61 noro 1275: int ndv_check_candidate(NODE input,int obpe,int oadv,EPOS oepos,NODE cand)
1.28 noro 1276: {
1277: int n,i,stat;
1278: ND nf,d;
1.61 noro 1279: NDV r;
1.45 noro 1280: NODE t,s;
1.113 noro 1281: union oNDC dn;
1.45 noro 1282:
1.61 noro 1283: ndv_setup(0,0,cand);
1.31 noro 1284: n = length(cand);
1.28 noro 1285:
1286: /* membercheck : list is a subset of Id(cand) ? */
1287: for ( t = input; t; t = NEXT(t) ) {
1.45 noro 1288: again:
1.61 noro 1289: if ( nd_bpe > obpe )
1290: r = ndv_dup_realloc((NDV)BDY(t),obpe,oadv,oepos);
1291: else
1292: r = (NDV)BDY(t);
1293: d = ndvtond(0,r);
1.69 noro 1294: stat = nd_nf(0,d,nd_ps,0,0,&nf);
1.45 noro 1295: if ( !stat ) {
1.103 noro 1296: nd_reconstruct(0,0);
1.45 noro 1297: goto again;
1298: } else if ( nf ) return 0;
1.71 noro 1299: if ( DP_Print ) { printf("."); fflush(stdout); }
1.28 noro 1300: }
1.71 noro 1301: if ( DP_Print ) { printf("\n"); }
1.28 noro 1302: /* gbcheck : cand is a GB of Id(cand) ? */
1.77 noro 1303: if ( !nd_gb(0,0,1) ) return 0;
1.28 noro 1304: /* XXX */
1.23 noro 1305: return 1;
1306: }
1.1 noro 1307:
1308: ND nd_remove_head(ND p)
1309: {
1310: NM m;
1311:
1312: m = BDY(p);
1313: if ( !NEXT(m) ) {
1.34 noro 1314: FREEND(p); p = 0;
1.31 noro 1315: } else {
1.34 noro 1316: BDY(p) = NEXT(m); LEN(p)--;
1.31 noro 1317: }
1.1 noro 1318: FREENM(m);
1319: return p;
1320: }
1321:
1.69 noro 1322: ND nd_separate_head(ND p,ND *head)
1323: {
1324: NM m,m0;
1325: ND r;
1326:
1327: m = BDY(p);
1328: if ( !NEXT(m) ) {
1329: *head = p; p = 0;
1330: } else {
1331: m0 = m;
1332: BDY(p) = NEXT(m); LEN(p)--;
1333: NEXT(m0) = 0;
1334: MKND(NV(p),m0,1,r);
1335: *head = r;
1336: }
1337: return p;
1338: }
1339:
1.1 noro 1340: PGeoBucket create_pbucket()
1341: {
1342: PGeoBucket g;
1343:
1344: g = CALLOC(1,sizeof(struct oPGeoBucket));
1345: g->m = -1;
1346: return g;
1347: }
1348:
1.25 noro 1349: void free_pbucket(PGeoBucket b) {
1350: int i;
1351:
1.26 noro 1352: for ( i = 0; i <= b->m; i++ )
1.25 noro 1353: if ( b->body[i] ) {
1354: nd_free(b->body[i]);
1355: b->body[i] = 0;
1356: }
1357: GC_free(b);
1358: }
1359:
1.63 noro 1360: void add_pbucket_symbolic(PGeoBucket g,ND d)
1361: {
1362: int l,i,k,m;
1363:
1364: if ( !d )
1365: return;
1366: l = LEN(d);
1367: for ( k = 0, m = 1; l > m; k++, m <<= 1 );
1368: /* 2^(k-1) < l <= 2^k (=m) */
1369: d = nd_merge(g->body[k],d);
1370: for ( ; d && LEN(d) > m; k++, m <<= 1 ) {
1371: g->body[k] = 0;
1372: d = nd_merge(g->body[k+1],d);
1373: }
1374: g->body[k] = d;
1375: g->m = MAX(g->m,k);
1376: }
1377:
1.31 noro 1378: void add_pbucket(int mod,PGeoBucket g,ND d)
1.1 noro 1379: {
1.31 noro 1380: int l,i,k,m;
1.1 noro 1381:
1.31 noro 1382: if ( !d )
1383: return;
1384: l = LEN(d);
1.29 noro 1385: for ( k = 0, m = 1; l > m; k++, m <<= 1 );
1386: /* 2^(k-1) < l <= 2^k (=m) */
1.31 noro 1387: d = nd_add(mod,g->body[k],d);
1388: for ( ; d && LEN(d) > m; k++, m <<= 1 ) {
1.1 noro 1389: g->body[k] = 0;
1.31 noro 1390: d = nd_add(mod,g->body[k+1],d);
1.1 noro 1391: }
1392: g->body[k] = d;
1393: g->m = MAX(g->m,k);
1394: }
1395:
1.113 noro 1396: void mulq_pbucket(PGeoBucket g,Q c)
1.26 noro 1397: {
1398: int k;
1399:
1400: for ( k = 0; k <= g->m; k++ )
1.113 noro 1401: nd_mul_c_q(g->body[k],c);
1.26 noro 1402: }
1403:
1.63 noro 1404: NM remove_head_pbucket_symbolic(PGeoBucket g)
1405: {
1406: int j,i,k,c;
1407: NM head;
1408:
1409: k = g->m;
1410: j = -1;
1411: for ( i = 0; i <= k; i++ ) {
1412: if ( !g->body[i] ) continue;
1413: if ( j < 0 ) j = i;
1414: else {
1415: c = DL_COMPARE(HDL(g->body[i]),HDL(g->body[j]));
1416: if ( c > 0 )
1417: j = i;
1418: else if ( c == 0 )
1419: g->body[i] = nd_remove_head(g->body[i]);
1420: }
1421: }
1422: if ( j < 0 ) return 0;
1423: else {
1424: head = BDY(g->body[j]);
1425: if ( !NEXT(head) ) {
1426: FREEND(g->body[j]);
1427: g->body[j] = 0;
1428: } else {
1429: BDY(g->body[j]) = NEXT(head);
1430: LEN(g->body[j])--;
1431: }
1432: return head;
1433: }
1434: }
1435:
1.19 noro 1436: int head_pbucket(int mod,PGeoBucket g)
1.1 noro 1437: {
1438: int j,i,c,k,nv,sum;
1.61 noro 1439: UINT *di,*dj;
1.1 noro 1440: ND gi,gj;
1441:
1442: k = g->m;
1443: while ( 1 ) {
1444: j = -1;
1445: for ( i = 0; i <= k; i++ ) {
1446: if ( !(gi = g->body[i]) )
1447: continue;
1448: if ( j < 0 ) {
1449: j = i;
1450: gj = g->body[j];
1451: dj = HDL(gj);
1.14 noro 1452: sum = HCM(gj);
1.1 noro 1453: } else {
1.34 noro 1454: c = DL_COMPARE(HDL(gi),dj);
1.1 noro 1455: if ( c > 0 ) {
1.34 noro 1456: if ( sum ) HCM(gj) = sum;
1457: else g->body[j] = nd_remove_head(gj);
1.1 noro 1458: j = i;
1459: gj = g->body[j];
1460: dj = HDL(gj);
1.14 noro 1461: sum = HCM(gj);
1.1 noro 1462: } else if ( c == 0 ) {
1.71 noro 1463: if ( mod == -1 )
1464: sum = _addsf(sum,HCM(gi));
1465: else {
1466: sum = sum+HCM(gi)-mod;
1467: if ( sum < 0 ) sum += mod;
1468: }
1.1 noro 1469: g->body[i] = nd_remove_head(gi);
1470: }
1471: }
1472: }
1.34 noro 1473: if ( j < 0 ) return -1;
1.1 noro 1474: else if ( sum ) {
1.14 noro 1475: HCM(gj) = sum;
1.26 noro 1476: return j;
1.31 noro 1477: } else
1.26 noro 1478: g->body[j] = nd_remove_head(gj);
1479: }
1480: }
1481:
1.113 noro 1482: int head_pbucket_q(PGeoBucket g)
1.26 noro 1483: {
1484: int j,i,c,k,nv;
1.113 noro 1485: Q sum,t;
1.26 noro 1486: ND gi,gj;
1487:
1488: k = g->m;
1489: while ( 1 ) {
1490: j = -1;
1491: for ( i = 0; i <= k; i++ ) {
1.34 noro 1492: if ( !(gi = g->body[i]) ) continue;
1.26 noro 1493: if ( j < 0 ) {
1494: j = i;
1495: gj = g->body[j];
1.113 noro 1496: sum = HCQ(gj);
1.26 noro 1497: } else {
1498: nv = NV(gi);
1.34 noro 1499: c = DL_COMPARE(HDL(gi),HDL(gj));
1.26 noro 1500: if ( c > 0 ) {
1.113 noro 1501: if ( sum ) HCQ(gj) = sum;
1.34 noro 1502: else g->body[j] = nd_remove_head(gj);
1.26 noro 1503: j = i;
1504: gj = g->body[j];
1.113 noro 1505: sum = HCQ(gj);
1.26 noro 1506: } else if ( c == 0 ) {
1.113 noro 1507: addq(sum,HCQ(gi),&t);
1508: sum = t;
1.26 noro 1509: g->body[i] = nd_remove_head(gi);
1510: }
1511: }
1512: }
1.34 noro 1513: if ( j < 0 ) return -1;
1.26 noro 1514: else if ( sum ) {
1.113 noro 1515: HCQ(gj) = sum;
1.1 noro 1516: return j;
1.31 noro 1517: } else
1.1 noro 1518: g->body[j] = nd_remove_head(gj);
1519: }
1520: }
1521:
1.25 noro 1522: ND normalize_pbucket(int mod,PGeoBucket g)
1.1 noro 1523: {
1.31 noro 1524: int i;
1.1 noro 1525: ND r,t;
1526:
1527: r = 0;
1.28 noro 1528: for ( i = 0; i <= g->m; i++ ) {
1.31 noro 1529: r = nd_add(mod,r,g->body[i]);
1.28 noro 1530: g->body[i] = 0;
1531: }
1532: g->m = -1;
1.1 noro 1533: return r;
1534: }
1535:
1.92 noro 1536: void do_diagonalize(int sugar,int m)
1537: {
1538: int i,nh,stat;
1539: NODE r,g,t;
1540: ND h,nf,s,head;
1541: NDV nfv;
1.113 noro 1542: Q q,num,den;
1543: union oNDC dn;
1.92 noro 1544:
1545: for ( i = nd_psn-1; i >= 0 && SG(nd_psh[i]) == sugar; i-- ) {
1546: if ( nd_demand )
1547: nfv = ndv_load(i);
1548: else
1549: nfv = nd_ps[i];
1550: s = ndvtond(m,nfv);
1551: s = nd_separate_head(s,&head);
1552: nd_nf(m,s,nd_ps,1,&dn,&nf);
1553: if ( !m ) {
1.113 noro 1554: NTOQ(NM(dn.z),SGN(dn.z),num);
1555: mulq(HCQ(head),num,&q); HCQ(head) = q;
1556: if ( DN(dn.z) ) {
1557: NTOQ(DN(dn.z),1,den);
1558: nd_mul_c_q(nf,den);
1559: }
1.92 noro 1560: }
1561: nf = nd_add(m,head,nf);
1562: ndv_free(nfv);
1563: nd_removecont(m,nf);
1564: nfv = ndtondv(m,nf);
1565: nd_free(nf);
1566: nd_bound[i] = ndv_compute_bound(nfv);
1567: if ( nd_demand ) {
1568: ndv_save(nfv,i);
1569: ndv_free(nfv);
1570: } else
1571: nd_ps[i] = nfv;
1572: }
1573: }
1574:
1.27 noro 1575: /* return value = 0 => input is not a GB */
1576:
1.77 noro 1577: NODE nd_gb(int m,int ishomo,int checkonly)
1.1 noro 1578: {
1579: int i,nh,sugar,stat;
1.23 noro 1580: NODE r,g,t;
1.1 noro 1581: ND_pairs d;
1582: ND_pairs l;
1.77 noro 1583: ND h,nf,s,head;
1.63 noro 1584: NDV nfv;
1.113 noro 1585: Q q,num,den;
1586: union oNDC dn;
1.1 noro 1587:
1.23 noro 1588: g = 0; d = 0;
1589: for ( i = 0; i < nd_psn; i++ ) {
1.1 noro 1590: d = update_pairs(d,g,i);
1591: g = update_base(g,i);
1592: }
1593: sugar = 0;
1594: while ( d ) {
1595: again:
1596: l = nd_minp(d,&d);
1.14 noro 1597: if ( SG(l) != sugar ) {
1.92 noro 1598: if ( ishomo ) do_diagonalize(sugar,m);
1599:
1.14 noro 1600: sugar = SG(l);
1.71 noro 1601: if ( DP_Print ) fprintf(asir_out,"%d",sugar);
1.1 noro 1602: }
1.53 noro 1603: stat = nd_sp(m,0,l,&h);
1.1 noro 1604: if ( !stat ) {
1605: NEXT(l) = d; d = l;
1.103 noro 1606: d = nd_reconstruct(0,d);
1.1 noro 1607: goto again;
1608: }
1.41 noro 1609: #if USE_GEOBUCKET
1.69 noro 1610: stat = m?nd_nf_pbucket(m,h,nd_ps,!Top,&nf):nd_nf(m,h,nd_ps,!Top,0,&nf);
1.41 noro 1611: #else
1.69 noro 1612: stat = nd_nf(m,h,nd_ps,!Top,0,&nf);
1.41 noro 1613: #endif
1.1 noro 1614: if ( !stat ) {
1615: NEXT(l) = d; d = l;
1.103 noro 1616: d = nd_reconstruct(0,d);
1.1 noro 1617: goto again;
1618: } else if ( nf ) {
1.27 noro 1619: if ( checkonly ) return 0;
1.71 noro 1620: if ( DP_Print ) { printf("+"); fflush(stdout); }
1.63 noro 1621: nd_removecont(m,nf);
1622: nfv = ndtondv(m,nf); nd_free(nf);
1.77 noro 1623: nh = ndv_newps(m,nfv,0);
1.1 noro 1624: d = update_pairs(d,g,nh);
1625: g = update_base(g,nh);
1626: FREENDP(l);
1627: } else {
1.71 noro 1628: if ( DP_Print ) { printf("."); fflush(stdout); }
1.1 noro 1629: FREENDP(l);
1630: }
1631: }
1.77 noro 1632: if ( nd_demand )
1633: for ( t = g; t; t = NEXT(t) )
1634: BDY(t) = (pointer)ndv_load((int)BDY(t));
1635: else
1636: for ( t = g; t; t = NEXT(t) )
1637: BDY(t) = (pointer)nd_ps[(int)BDY(t)];
1.1 noro 1638: return g;
1639: }
1640:
1.92 noro 1641: void do_diagonalize_trace(int sugar,int m)
1.91 noro 1642: {
1643: int i,nh,stat;
1644: NODE r,g,t;
1645: ND h,nf,nfq,s,head;
1646: NDV nfv,nfqv;
1.113 noro 1647: Q q,den,num;
1648: union oNDC dn;
1.91 noro 1649:
1650: for ( i = nd_psn-1; i >= 0 && SG(nd_psh[i]) == sugar; i-- ) {
1651: /* for nd_ps */
1652: s = ndvtond(m,nd_ps[i]);
1653: s = nd_separate_head(s,&head);
1654: nd_nf_pbucket(m,s,nd_ps,1,&nf);
1655: nf = nd_add(m,head,nf);
1656: ndv_free(nd_ps[i]);
1657: nd_ps[i] = ndtondv(m,nf);
1658: nd_free(nf);
1659:
1660: /* for nd_ps_trace */
1661: if ( nd_demand )
1662: nfv = ndv_load(i);
1663: else
1664: nfv = nd_ps_trace[i];
1665: s = ndvtond(0,nfv);
1666: s = nd_separate_head(s,&head);
1667: nd_nf(0,s,nd_ps_trace,1,&dn,&nf);
1.113 noro 1668: NTOQ(NM(dn.z),SGN(dn.z),num);
1669: mulq(HCQ(head),num,&q); HCQ(head) = q;
1670: if ( DN(dn.z) ) {
1671: NTOQ(DN(dn.z),1,den);
1672: nd_mul_c_q(nf,den);
1673: }
1.91 noro 1674: nf = nd_add(0,head,nf);
1675: ndv_free(nfv);
1676: nd_removecont(0,nf);
1677: nfv = ndtondv(0,nf);
1678: nd_free(nf);
1679: nd_bound[i] = ndv_compute_bound(nfv);
1680: if ( nd_demand ) {
1681: ndv_save(nfv,i);
1682: ndv_free(nfv);
1683: } else
1684: nd_ps_trace[i] = nfv;
1685: }
1686: }
1687:
1.82 noro 1688: NODE nd_gb_trace(int m,int ishomo)
1.20 noro 1689: {
1690: int i,nh,sugar,stat;
1.23 noro 1691: NODE r,g,t;
1.20 noro 1692: ND_pairs d;
1693: ND_pairs l;
1.82 noro 1694: ND h,nf,nfq,s,head;
1.63 noro 1695: NDV nfv,nfqv;
1.113 noro 1696: Q q,den,num;
1697: union oNDC dn;
1.20 noro 1698:
1.23 noro 1699: g = 0; d = 0;
1700: for ( i = 0; i < nd_psn; i++ ) {
1.20 noro 1701: d = update_pairs(d,g,i);
1702: g = update_base(g,i);
1703: }
1704: sugar = 0;
1705: while ( d ) {
1706: again:
1707: l = nd_minp(d,&d);
1708: if ( SG(l) != sugar ) {
1.93 noro 1709: if ( ishomo ) do_diagonalize_trace(sugar,m);
1.20 noro 1710: sugar = SG(l);
1.71 noro 1711: if ( DP_Print ) fprintf(asir_out,"%d",sugar);
1.20 noro 1712: }
1.53 noro 1713: stat = nd_sp(m,0,l,&h);
1.20 noro 1714: if ( !stat ) {
1715: NEXT(l) = d; d = l;
1.103 noro 1716: d = nd_reconstruct(1,d);
1.20 noro 1717: goto again;
1718: }
1.41 noro 1719: #if USE_GEOBUCKET
1.53 noro 1720: stat = nd_nf_pbucket(m,h,nd_ps,!Top,&nf);
1.41 noro 1721: #else
1.69 noro 1722: stat = nd_nf(m,h,nd_ps,!Top,0,&nf);
1.41 noro 1723: #endif
1.20 noro 1724: if ( !stat ) {
1725: NEXT(l) = d; d = l;
1.103 noro 1726: d = nd_reconstruct(1,d);
1.20 noro 1727: goto again;
1728: } else if ( nf ) {
1.77 noro 1729: if ( nd_demand ) {
1730: nfqv = ndv_load(nd_psn);
1731: nfq = ndvtond(0,nfqv);
1732: } else
1733: nfq = 0;
1734: if ( !nfq ) {
1735: if ( !nd_sp(0,1,l,&h) || !nd_nf(0,h,nd_ps_trace,!Top,0,&nfq) ) {
1736: NEXT(l) = d; d = l;
1.103 noro 1737: d = nd_reconstruct(1,d);
1.77 noro 1738: goto again;
1739: }
1.70 noro 1740: }
1.20 noro 1741: if ( nfq ) {
1.70 noro 1742: /* m|HC(nfq) => failure */
1.113 noro 1743: if ( !rem(NM(HCQ(nfq)),m) ) return 0;
1.63 noro 1744:
1.71 noro 1745: if ( DP_Print ) { printf("+"); fflush(stdout); }
1.63 noro 1746: nd_removecont(m,nf); nfv = ndtondv(m,nf); nd_free(nf);
1747: nd_removecont(0,nfq); nfqv = ndtondv(0,nfq); nd_free(nfq);
1.77 noro 1748: nh = ndv_newps(0,nfv,nfqv);
1.20 noro 1749: d = update_pairs(d,g,nh);
1750: g = update_base(g,nh);
1751: } else {
1.71 noro 1752: if ( DP_Print ) { printf("*"); fflush(stdout); }
1.20 noro 1753: }
1754: } else {
1.71 noro 1755: if ( DP_Print ) { printf("."); fflush(stdout); }
1.20 noro 1756: }
1757: FREENDP(l);
1758: }
1.77 noro 1759: if ( nd_demand )
1760: for ( t = g; t; t = NEXT(t) )
1761: BDY(t) = (pointer)ndv_load((int)BDY(t));
1762: else
1763: for ( t = g; t; t = NEXT(t) )
1764: BDY(t) = (pointer)nd_ps_trace[(int)BDY(t)];
1.20 noro 1765: return g;
1766: }
1767:
1.23 noro 1768: int ndv_compare(NDV *p1,NDV *p2)
1769: {
1.34 noro 1770: return DL_COMPARE(HDL(*p1),HDL(*p2));
1.23 noro 1771: }
1772:
1773: int ndv_compare_rev(NDV *p1,NDV *p2)
1774: {
1.34 noro 1775: return -DL_COMPARE(HDL(*p1),HDL(*p2));
1.23 noro 1776: }
1777:
1.61 noro 1778: NODE ndv_reduceall(int m,NODE f)
1.23 noro 1779: {
1.69 noro 1780: int i,n,stat;
1781: ND nf,g,head;
1.23 noro 1782: NODE t,a0,a;
1.113 noro 1783: union oNDC dn;
1.69 noro 1784: NDV *w;
1.113 noro 1785: Q q,num,den;
1.23 noro 1786:
1.69 noro 1787: n = length(f);
1788: #if 0
1789: w = (NDV *)ALLOCA(n*sizeof(NDV));
1790: for ( i = 0, t = f; i < n; i++, t = NEXT(t) ) w[i] = (NDV)BDY(t);
1791: qsort(w,n,sizeof(NDV),
1792: (int (*)(const void *,const void *))ndv_compare);
1793: for ( t = f, i = 0; t; i++, t = NEXT(t) ) BDY(t) = (pointer)w[i];
1794: #endif
1795: ndv_setup(m,0,f);
1796: for ( i = 0; i < n; ) {
1797: g = ndvtond(m,nd_ps[i]);
1798: g = nd_separate_head(g,&head);
1799: stat = nd_nf(m,g,nd_ps,1,&dn,&nf);
1.50 noro 1800: if ( !stat )
1.103 noro 1801: nd_reconstruct(0,0);
1.69 noro 1802: else {
1.71 noro 1803: if ( DP_Print ) { printf("."); fflush(stdout); }
1.85 noro 1804: if ( !m ) {
1.113 noro 1805: NTOQ(NM(dn.z),SGN(dn.z),num);
1806: mulq(HCQ(head),num,&q); HCQ(head) = q;
1807: if ( DN(dn.z) ) {
1808: NTOQ(DN(dn.z),1,den);
1809: nd_mul_c_q(nf,den);
1810: }
1.85 noro 1811: }
1.69 noro 1812: nf = nd_add(m,head,nf);
1813: ndv_free(nd_ps[i]);
1.24 noro 1814: nd_removecont(m,nf);
1.69 noro 1815: nd_ps[i] = ndtondv(m,nf); nd_free(nf);
1816: nd_bound[i] = ndv_compute_bound(nd_ps[i]);
1.50 noro 1817: i++;
1.23 noro 1818: }
1819: }
1.71 noro 1820: if ( DP_Print ) { printf("\n"); }
1.23 noro 1821: for ( a0 = 0, i = 0; i < n; i++ ) {
1822: NEXTNODE(a0,a);
1.69 noro 1823: BDY(a) = (pointer)nd_ps[i];
1.23 noro 1824: }
1825: NEXT(a) = 0;
1826: return a0;
1827: }
1828:
1.1 noro 1829: ND_pairs update_pairs( ND_pairs d, NODE /* of index */ g, int t)
1830: {
1831: ND_pairs d1,nd,cur,head,prev,remove;
1832:
1833: if ( !g ) return d;
1834: d = crit_B(d,t);
1835: d1 = nd_newpairs(g,t);
1836: d1 = crit_M(d1);
1837: d1 = crit_F(d1);
1.55 noro 1838: if ( do_weyl )
1839: head = d1;
1840: else {
1841: prev = 0; cur = head = d1;
1842: while ( cur ) {
1843: if ( crit_2( cur->i1,cur->i2 ) ) {
1844: remove = cur;
1845: if ( !prev ) head = cur = NEXT(cur);
1846: else cur = NEXT(prev) = NEXT(cur);
1847: FREENDP(remove);
1848: } else {
1849: prev = cur; cur = NEXT(cur);
1850: }
1.1 noro 1851: }
1852: }
1853: if ( !d )
1854: return head;
1855: else {
1856: nd = d;
1.34 noro 1857: while ( NEXT(nd) ) nd = NEXT(nd);
1.1 noro 1858: NEXT(nd) = head;
1859: return d;
1860: }
1861: }
1862:
1863: ND_pairs nd_newpairs( NODE g, int t )
1864: {
1865: NODE h;
1.61 noro 1866: UINT *dl;
1.34 noro 1867: int ts,s;
1.1 noro 1868: ND_pairs r,r0;
1869:
1.20 noro 1870: dl = DL(nd_psh[t]);
1.34 noro 1871: ts = SG(nd_psh[t]) - TD(dl);
1.1 noro 1872: for ( r0 = 0, h = g; h; h = NEXT(h) ) {
1873: NEXTND_pairs(r0,r);
1874: r->i1 = (int)BDY(h);
1875: r->i2 = t;
1.20 noro 1876: ndl_lcm(DL(nd_psh[r->i1]),dl,r->lcm);
1.34 noro 1877: s = SG(nd_psh[r->i1])-TD(DL(nd_psh[r->i1]));
1878: SG(r) = MAX(s,ts) + TD(LCM(r));
1.1 noro 1879: }
1880: NEXT(r) = 0;
1881: return r0;
1882: }
1883:
1884: ND_pairs crit_B( ND_pairs d, int s )
1885: {
1886: ND_pairs cur,head,prev,remove;
1.61 noro 1887: UINT *t,*tl,*lcm;
1.1 noro 1888: int td,tdl;
1889:
1890: if ( !d ) return 0;
1.20 noro 1891: t = DL(nd_psh[s]);
1.1 noro 1892: prev = 0;
1893: head = cur = d;
1.61 noro 1894: lcm = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
1.1 noro 1895: while ( cur ) {
1896: tl = cur->lcm;
1897: if ( ndl_reducible(tl,t)
1.20 noro 1898: && (ndl_lcm(DL(nd_psh[cur->i1]),t,lcm),!ndl_equal(lcm,tl))
1899: && (ndl_lcm(DL(nd_psh[cur->i2]),t,lcm),!ndl_equal(lcm,tl)) ) {
1.1 noro 1900: remove = cur;
1901: if ( !prev ) {
1902: head = cur = NEXT(cur);
1903: } else {
1904: cur = NEXT(prev) = NEXT(cur);
1905: }
1906: FREENDP(remove);
1907: } else {
1.34 noro 1908: prev = cur; cur = NEXT(cur);
1.1 noro 1909: }
1910: }
1911: return head;
1912: }
1913:
1914: ND_pairs crit_M( ND_pairs d1 )
1915: {
1916: ND_pairs e,d2,d3,dd,p;
1.61 noro 1917: UINT *id,*jd;
1.1 noro 1918:
1919: for ( dd = 0, e = d1; e; e = d3 ) {
1920: if ( !(d2 = NEXT(e)) ) {
1921: NEXT(e) = dd;
1922: return e;
1923: }
1.34 noro 1924: id = LCM(e);
1.1 noro 1925: for ( d3 = 0; d2; d2 = p ) {
1.34 noro 1926: p = NEXT(d2);
1927: jd = LCM(d2);
1928: if ( ndl_equal(jd,id) )
1929: ;
1930: else if ( TD(jd) > TD(id) )
1.1 noro 1931: if ( ndl_reducible(jd,id) ) continue;
1932: else ;
1.34 noro 1933: else if ( ndl_reducible(id,jd) ) goto delit;
1.1 noro 1934: NEXT(d2) = d3;
1935: d3 = d2;
1936: }
1937: NEXT(e) = dd;
1938: dd = e;
1939: continue;
1940: /**/
1941: delit: NEXT(d2) = d3;
1942: d3 = d2;
1943: for ( ; p; p = d2 ) {
1944: d2 = NEXT(p);
1945: NEXT(p) = d3;
1946: d3 = p;
1947: }
1948: FREENDP(e);
1949: }
1950: return dd;
1951: }
1952:
1953: ND_pairs crit_F( ND_pairs d1 )
1954: {
1955: ND_pairs rest, head,remove;
1956: ND_pairs last, p, r, w;
1957: int s;
1958:
1959: for ( head = last = 0, p = d1; NEXT(p); ) {
1960: r = w = equivalent_pairs(p,&rest);
1.14 noro 1961: s = SG(r);
1.1 noro 1962: w = NEXT(w);
1963: while ( w ) {
1964: if ( crit_2(w->i1,w->i2) ) {
1965: r = w;
1966: w = NEXT(w);
1967: while ( w ) {
1968: remove = w;
1969: w = NEXT(w);
1970: FREENDP(remove);
1971: }
1972: break;
1.14 noro 1973: } else if ( SG(w) < s ) {
1.1 noro 1974: FREENDP(r);
1975: r = w;
1.14 noro 1976: s = SG(r);
1.1 noro 1977: w = NEXT(w);
1978: } else {
1979: remove = w;
1980: w = NEXT(w);
1981: FREENDP(remove);
1982: }
1983: }
1984: if ( last ) NEXT(last) = r;
1985: else head = r;
1986: NEXT(last = r) = 0;
1987: p = rest;
1988: if ( !p ) return head;
1989: }
1990: if ( !last ) return p;
1991: NEXT(last) = p;
1992: return head;
1993: }
1994:
1995: int crit_2( int dp1, int dp2 )
1996: {
1.20 noro 1997: return ndl_disjoint(DL(nd_psh[dp1]),DL(nd_psh[dp2]));
1.1 noro 1998: }
1999:
1.40 noro 2000: ND_pairs equivalent_pairs( ND_pairs d1, ND_pairs *prest )
1.1 noro 2001: {
2002: ND_pairs w,p,r,s;
1.61 noro 2003: UINT *d;
1.1 noro 2004:
2005: w = d1;
1.34 noro 2006: d = LCM(w);
1.1 noro 2007: s = NEXT(w);
2008: NEXT(w) = 0;
2009: for ( r = 0; s; s = p ) {
2010: p = NEXT(s);
1.34 noro 2011: if ( ndl_equal(d,LCM(s)) ) {
1.39 noro 2012: NEXT(s) = w; w = s;
1.1 noro 2013: } else {
1.39 noro 2014: NEXT(s) = r; r = s;
1.1 noro 2015: }
2016: }
2017: *prest = r;
2018: return w;
2019: }
2020:
2021: NODE update_base(NODE nd,int ndp)
2022: {
1.61 noro 2023: UINT *dl, *dln;
1.1 noro 2024: NODE last, p, head;
2025:
1.20 noro 2026: dl = DL(nd_psh[ndp]);
1.1 noro 2027: for ( head = last = 0, p = nd; p; ) {
1.20 noro 2028: dln = DL(nd_psh[(int)BDY(p)]);
1.34 noro 2029: if ( ndl_reducible( dln, dl ) ) {
1.1 noro 2030: p = NEXT(p);
2031: if ( last ) NEXT(last) = p;
2032: } else {
2033: if ( !last ) head = p;
2034: p = NEXT(last = p);
2035: }
2036: }
2037: head = append_one(head,ndp);
2038: return head;
2039: }
2040:
2041: ND_pairs nd_minp( ND_pairs d, ND_pairs *prest )
2042: {
2043: ND_pairs m,ml,p,l;
1.61 noro 2044: UINT *lcm;
1.33 noro 2045: int s,td,len,tlen,c,c1;
1.1 noro 2046:
2047: if ( !(p = NEXT(m = d)) ) {
2048: *prest = p;
2049: NEXT(m) = 0;
2050: return m;
2051: }
1.14 noro 2052: s = SG(m);
1.77 noro 2053: if ( !NoSugar ) {
2054: for ( ml = 0, l = m; p; p = NEXT(l = p) )
2055: if ( (SG(p) < s)
2056: || ((SG(p) == s) && (DL_COMPARE(LCM(p),LCM(m)) < 0)) ) {
2057: ml = l; m = p; s = SG(m);
2058: }
2059: } else {
2060: for ( ml = 0, l = m; p; p = NEXT(l = p) )
2061: if ( DL_COMPARE(LCM(p),LCM(m)) < 0 ) {
2062: ml = l; m = p; s = SG(m);
2063: }
2064: }
1.1 noro 2065: if ( !ml ) *prest = NEXT(m);
2066: else {
2067: NEXT(ml) = NEXT(m);
2068: *prest = d;
2069: }
2070: NEXT(m) = 0;
2071: return m;
2072: }
2073:
1.63 noro 2074: ND_pairs nd_minsugarp( ND_pairs d, ND_pairs *prest )
2075: {
1.69 noro 2076: int msugar,i;
1.63 noro 2077: ND_pairs t,dm0,dm,dr0,dr;
2078:
2079: for ( msugar = SG(d), t = NEXT(d); t; t = NEXT(t) )
2080: if ( SG(t) < msugar ) msugar = SG(t);
2081: dm0 = 0; dr0 = 0;
1.69 noro 2082: for ( i = 0, t = d; t; t = NEXT(t) )
2083: if ( i < nd_f4_nsp && SG(t) == msugar ) {
1.63 noro 2084: if ( dm0 ) NEXT(dm) = t;
2085: else dm0 = t;
2086: dm = t;
1.69 noro 2087: i++;
1.63 noro 2088: } else {
2089: if ( dr0 ) NEXT(dr) = t;
2090: else dr0 = t;
2091: dr = t;
2092: }
2093: NEXT(dm) = 0;
2094: if ( dr0 ) NEXT(dr) = 0;
2095: *prest = dr0;
2096: return dm0;
2097: }
2098:
1.77 noro 2099: int ndv_newps(int m,NDV a,NDV aq)
1.1 noro 2100: {
1.3 noro 2101: int len;
1.13 noro 2102: RHist r;
1.20 noro 2103: NDV b;
1.3 noro 2104:
1.1 noro 2105: if ( nd_psn == nd_pslen ) {
2106: nd_pslen *= 2;
1.11 noro 2107: nd_ps = (NDV *)REALLOC((char *)nd_ps,nd_pslen*sizeof(NDV));
1.53 noro 2108: nd_ps_trace = (NDV *)REALLOC((char *)nd_ps_trace,nd_pslen*sizeof(NDV));
1.13 noro 2109: nd_psh = (RHist *)REALLOC((char *)nd_psh,nd_pslen*sizeof(RHist));
1.61 noro 2110: nd_bound = (UINT **)
2111: REALLOC((char *)nd_bound,nd_pslen*sizeof(UINT *));
1.1 noro 2112: }
1.39 noro 2113: NEWRHist(r); nd_psh[nd_psn] = r;
1.63 noro 2114: nd_ps[nd_psn] = a;
1.39 noro 2115: if ( aq ) {
1.63 noro 2116: nd_ps_trace[nd_psn] = aq;
2117: nd_bound[nd_psn] = ndv_compute_bound(aq);
1.39 noro 2118: SG(r) = SG(aq); ndl_copy(HDL(aq),DL(r));
1.53 noro 2119: } else {
1.63 noro 2120: nd_bound[nd_psn] = ndv_compute_bound(a);
1.53 noro 2121: SG(r) = SG(a); ndl_copy(HDL(a),DL(r));
1.39 noro 2122: }
1.77 noro 2123: if ( nd_demand ) {
2124: if ( aq ) {
2125: ndv_save(nd_ps_trace[nd_psn],nd_psn);
2126: nd_ps_trace[nd_psn] = 0;
2127: } else {
2128: ndv_save(nd_ps[nd_psn],nd_psn);
2129: nd_ps[nd_psn] = 0;
2130: }
2131: }
1.1 noro 2132: return nd_psn++;
2133: }
2134:
1.61 noro 2135: void ndv_setup(int mod,int trace,NODE f)
1.1 noro 2136: {
1.5 noro 2137: int i,j,td,len,max;
1.1 noro 2138: NODE s,s0,f0;
1.61 noro 2139: UINT *d;
1.13 noro 2140: RHist r;
1.69 noro 2141: NDV *w;
1.61 noro 2142: NDV a,am;
1.11 noro 2143:
2144: nd_found = 0; nd_notfirst = 0; nd_create = 0;
1.1 noro 2145:
1.69 noro 2146: for ( nd_psn = 0, s = f; s; s = NEXT(s) ) if ( BDY(s) ) nd_psn++;
2147: w = (NDV *)ALLOCA(nd_psn*sizeof(NDV));
2148: for ( i = 0, s = f; s; s = NEXT(s) ) if ( BDY(s) ) w[i++] = BDY(s);
2149: qsort(w,nd_psn,sizeof(NDV),
2150: (int (*)(const void *,const void *))ndv_compare);
2151: nd_pslen = 2*nd_psn;
1.11 noro 2152: nd_ps = (NDV *)MALLOC(nd_pslen*sizeof(NDV));
1.53 noro 2153: nd_ps_trace = (NDV *)MALLOC(nd_pslen*sizeof(NDV));
1.13 noro 2154: nd_psh = (RHist *)MALLOC(nd_pslen*sizeof(RHist));
1.61 noro 2155: nd_bound = (UINT **)MALLOC(nd_pslen*sizeof(UINT *));
1.57 noro 2156:
1.11 noro 2157: if ( !nd_red )
1.13 noro 2158: nd_red = (RHist *)MALLOC(REDTAB_LEN*sizeof(RHist));
1.71 noro 2159: for ( i = 0; i < REDTAB_LEN; i++ ) nd_red[i] = 0;
1.69 noro 2160: for ( i = 0; i < nd_psn; i++ ) {
1.39 noro 2161: if ( trace ) {
1.69 noro 2162: a = nd_ps_trace[i] = ndv_dup(0,w[i]);
1.61 noro 2163: ndv_removecont(0,a);
2164: am = nd_ps[i] = ndv_dup(mod,a);
2165: ndv_mod(mod,am);
2166: ndv_removecont(mod,am);
2167: } else {
1.69 noro 2168: a = nd_ps[i] = ndv_dup(mod,w[i]);
1.61 noro 2169: ndv_removecont(mod,a);
1.39 noro 2170: }
1.61 noro 2171: NEWRHist(r); SG(r) = HTD(a); ndl_copy(HDL(a),DL(r));
1.57 noro 2172: nd_bound[i] = ndv_compute_bound(a);
1.20 noro 2173: nd_psh[i] = r;
1.77 noro 2174: if ( nd_demand ) {
2175: if ( trace ) {
2176: ndv_save(nd_ps_trace[i],i);
2177: nd_ps_trace[i] = 0;
2178: } else {
2179: ndv_save(nd_ps[i],i);
2180: nd_ps[i] = 0;
2181: }
2182: }
1.20 noro 2183: }
2184: }
2185:
1.63 noro 2186: void nd_gr(LIST f,LIST v,int m,int f4,struct order_spec *ord,LIST *rp)
1.1 noro 2187: {
1.61 noro 2188: VL tv,fv,vv,vc;
1.1 noro 2189: NODE fd,fd0,r,r0,t,x,s,xx;
1.61 noro 2190: int e,max,nvar;
1.69 noro 2191: NDV b;
1.77 noro 2192: int ishomo;
1.1 noro 2193:
1.78 noro 2194: if ( !m && Demand ) nd_demand = 1;
2195: else nd_demand = 0;
2196:
1.103 noro 2197: #if 0
1.74 noro 2198: ndv_alloc = 0;
1.103 noro 2199: #endif
1.1 noro 2200: get_vars((Obj)f,&fv); pltovl(v,&vv);
1.74 noro 2201: for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++ );
1.88 noro 2202: switch ( ord->id ) {
2203: case 1:
2204: if ( ord->nv != nvar )
2205: error("nd_{gr,f4} : invalid order specification");
2206: break;
2207: default:
2208: break;
2209: }
1.32 noro 2210: nd_init_ord(ord);
1.61 noro 2211: for ( t = BDY(f), max = 0; t; t = NEXT(t) )
2212: for ( tv = vv; tv; tv = NEXT(tv) ) {
2213: e = getdeg(tv->v,(P)BDY(t));
2214: max = MAX(e,max);
2215: }
2216: nd_setup_parameters(nvar,max);
1.77 noro 2217: ishomo = 1;
1.1 noro 2218: for ( fd0 = 0, t = BDY(f); t; t = NEXT(t) ) {
1.61 noro 2219: b = (pointer)ptondv(CO,vv,(P)BDY(t));
1.77 noro 2220: if ( ishomo )
2221: ishomo = ishomo && ndv_ishomo(b);
1.69 noro 2222: if ( m ) ndv_mod(m,b);
1.61 noro 2223: if ( b ) { NEXTNODE(fd0,fd); BDY(fd) = (pointer)b; }
1.1 noro 2224: }
2225: if ( fd0 ) NEXT(fd) = 0;
1.61 noro 2226: ndv_setup(m,0,fd0);
1.77 noro 2227: x = f4?nd_f4(m):nd_gb(m,ishomo,0);
1.82 noro 2228: nd_demand = 0;
1.61 noro 2229: x = ndv_reducebase(x);
2230: x = ndv_reduceall(m,x);
1.23 noro 2231: for ( r0 = 0, t = x; t; t = NEXT(t) ) {
1.1 noro 2232: NEXTNODE(r0,r);
1.61 noro 2233: BDY(r) = ndvtop(m,CO,vv,BDY(t));
1.20 noro 2234: }
2235: if ( r0 ) NEXT(r) = 0;
2236: MKLIST(*rp,r0);
1.103 noro 2237: #if 0
1.74 noro 2238: fprintf(asir_out,"ndv_alloc=%d\n",ndv_alloc);
1.103 noro 2239: #endif
1.20 noro 2240: }
2241:
1.52 noro 2242: void nd_gr_trace(LIST f,LIST v,int trace,int homo,struct order_spec *ord,LIST *rp)
1.20 noro 2243: {
1.89 noro 2244: struct order_spec *ord1;
1.61 noro 2245: VL tv,fv,vv,vc;
1.27 noro 2246: NODE fd,fd0,in0,in,r,r0,t,s,cand;
1.61 noro 2247: int m,nocheck,nvar,mindex,e,max;
2248: NDV c;
2249: NMV a;
1.27 noro 2250: P p;
1.61 noro 2251: EPOS oepos;
1.82 noro 2252: int obpe,oadv,wmax,i,len,cbpe,ishomo;
1.20 noro 2253:
2254: get_vars((Obj)f,&fv); pltovl(v,&vv);
1.74 noro 2255: for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++ );
1.88 noro 2256: switch ( ord->id ) {
2257: case 1:
2258: if ( ord->nv != nvar )
2259: error("nd_gr_trace : invalid order specification");
2260: break;
2261: default:
2262: break;
2263: }
1.52 noro 2264: nocheck = 0;
2265: mindex = 0;
2266:
1.78 noro 2267: if ( Demand ) nd_demand = 1;
2268: else nd_demand = 0;
2269:
1.52 noro 2270: /* setup modulus */
2271: if ( trace < 0 ) {
2272: trace = -trace;
2273: nocheck = 1;
2274: }
2275: m = trace > 1 ? trace : get_lprime(mindex);
1.61 noro 2276: for ( t = BDY(f), max = 0; t; t = NEXT(t) )
2277: for ( tv = vv; tv; tv = NEXT(tv) ) {
2278: e = getdeg(tv->v,(P)BDY(t));
2279: max = MAX(e,max);
1.23 noro 2280: }
1.61 noro 2281: nd_init_ord(ord);
2282: nd_setup_parameters(nvar,max);
2283: obpe = nd_bpe; oadv = nmv_adv; oepos = nd_epos;
1.82 noro 2284: ishomo = 1;
1.61 noro 2285: for ( in0 = 0, fd0 = 0, t = BDY(f); t; t = NEXT(t) ) {
2286: c = ptondv(CO,vv,(P)BDY(t));
1.82 noro 2287: if ( ishomo )
2288: ishomo = ishomo && ndv_ishomo(c);
1.61 noro 2289: if ( c ) {
2290: NEXTNODE(in0,in); BDY(in) = (pointer)c;
2291: NEXTNODE(fd0,fd); BDY(fd) = (pointer)ndv_dup(0,c);
1.23 noro 2292: }
1.61 noro 2293: }
2294: if ( in0 ) NEXT(in) = 0;
2295: if ( fd0 ) NEXT(fd) = 0;
1.82 noro 2296: if ( !ishomo && homo ) {
1.61 noro 2297: for ( t = in0, wmax = 0; t; t = NEXT(t) ) {
2298: c = (NDV)BDY(t); len = LEN(c);
2299: for ( a = BDY(c), i = 0; i < len; i++, NMV_ADV(a) )
2300: wmax = MAX(TD(DL(a)),wmax);
2301: }
2302: homogenize_order(ord,nvar,&ord1);
1.89 noro 2303: nd_init_ord(ord1);
1.61 noro 2304: nd_setup_parameters(nvar+1,wmax);
2305: for ( t = fd0; t; t = NEXT(t) )
2306: ndv_homogenize((NDV)BDY(t),obpe,oadv,oepos);
1.27 noro 2307: }
1.52 noro 2308: while ( 1 ) {
1.77 noro 2309: if ( Demand )
2310: nd_demand = 1;
1.61 noro 2311: ndv_setup(m,1,fd0);
1.82 noro 2312: cand = nd_gb_trace(m,ishomo || homo);
1.52 noro 2313: if ( !cand ) {
2314: /* failure */
1.61 noro 2315: if ( trace > 1 ) { *rp = 0; return; }
2316: else m = get_lprime(++mindex);
1.52 noro 2317: continue;
2318: }
1.82 noro 2319: if ( !ishomo && homo ) {
1.27 noro 2320: /* dehomogenization */
1.61 noro 2321: for ( t = cand; t; t = NEXT(t) ) ndv_dehomogenize((NDV)BDY(t),ord);
1.45 noro 2322: nd_init_ord(ord);
1.61 noro 2323: nd_setup_parameters(nvar,0);
1.27 noro 2324: }
1.77 noro 2325: nd_demand = 0;
1.61 noro 2326: cand = ndv_reducebase(cand);
2327: cand = ndv_reduceall(0,cand);
1.78 noro 2328: cbpe = nd_bpe;
1.61 noro 2329: if ( nocheck )
2330: break;
2331: if ( ndv_check_candidate(in0,obpe,oadv,oepos,cand) )
1.52 noro 2332: /* success */
2333: break;
2334: else if ( trace > 1 ) {
2335: /* failure */
2336: *rp = 0; return;
1.61 noro 2337: } else {
1.52 noro 2338: /* try the next modulus */
2339: m = get_lprime(++mindex);
1.61 noro 2340: /* reset the parameters */
1.82 noro 2341: if ( !ishomo && homo ) {
1.89 noro 2342: nd_init_ord(ord1);
1.61 noro 2343: nd_setup_parameters(nvar+1,wmax);
2344: } else {
2345: nd_init_ord(ord);
2346: nd_setup_parameters(nvar,max);
2347: }
2348: }
1.52 noro 2349: }
1.27 noro 2350: /* dp->p */
1.61 noro 2351: nd_bpe = cbpe;
1.73 noro 2352: nd_setup_parameters(nd_nvar,0);
1.61 noro 2353: for ( r = cand; r; r = NEXT(r) ) BDY(r) = (pointer)ndvtop(0,CO,vv,BDY(r));
1.27 noro 2354: MKLIST(*rp,cand);
1.1 noro 2355: }
2356:
1.61 noro 2357: void dltondl(int n,DL dl,UINT *r)
1.1 noro 2358: {
1.61 noro 2359: UINT *d;
1.57 noro 2360: int i,j,l,s,ord_l;
1.43 noro 2361: struct order_pair *op;
1.1 noro 2362:
2363: d = dl->d;
1.41 noro 2364: for ( i = 0; i < nd_wpd; i++ ) r[i] = 0;
1.43 noro 2365: if ( nd_blockmask ) {
2366: l = nd_blockmask->n;
2367: op = nd_blockmask->order_pair;
2368: for ( j = 0, s = 0; j < l; j++ ) {
2369: ord_l = op[j].length;
1.57 noro 2370: for ( i = 0; i < ord_l; i++, s++ ) PUT_EXP(r,s,d[s]);
1.43 noro 2371: }
2372: TD(r) = ndl_weight(r);
1.61 noro 2373: ndl_weight_mask(r);
1.43 noro 2374: } else {
1.56 noro 2375: for ( i = 0; i < n; i++ ) PUT_EXP(r,i,d[i]);
1.43 noro 2376: TD(r) = ndl_weight(r);
2377: }
1.1 noro 2378: }
2379:
1.61 noro 2380: DL ndltodl(int n,UINT *ndl)
1.1 noro 2381: {
2382: DL dl;
2383: int *d;
1.57 noro 2384: int i,j,l,s,ord_l;
1.43 noro 2385: struct order_pair *op;
1.1 noro 2386:
2387: NEWDL(dl,n);
1.34 noro 2388: dl->td = TD(ndl);
1.1 noro 2389: d = dl->d;
1.43 noro 2390: if ( nd_blockmask ) {
2391: l = nd_blockmask->n;
2392: op = nd_blockmask->order_pair;
2393: for ( j = 0, s = 0; j < l; j++ ) {
2394: ord_l = op[j].length;
1.57 noro 2395: for ( i = 0; i < ord_l; i++, s++ ) d[s] = GET_EXP(ndl,s);
1.43 noro 2396: }
2397: } else {
1.56 noro 2398: for ( i = 0; i < n; i++ ) d[i] = GET_EXP(ndl,i);
1.43 noro 2399: }
1.1 noro 2400: return dl;
2401: }
2402:
1.61 noro 2403: void ndl_print(UINT *dl)
1.1 noro 2404: {
2405: int n;
1.57 noro 2406: int i,j,l,ord_l,s,s0;
1.43 noro 2407: struct order_pair *op;
1.1 noro 2408:
2409: n = nd_nvar;
2410: printf("<<");
1.43 noro 2411: if ( nd_blockmask ) {
2412: l = nd_blockmask->n;
2413: op = nd_blockmask->order_pair;
2414: for ( j = 0, s = s0 = 0; j < l; j++ ) {
2415: ord_l = op[j].length;
1.57 noro 2416: for ( i = 0; i < ord_l; i++, s++ )
2417: printf(s==n-1?"%d":"%d,",GET_EXP(dl,s));
1.43 noro 2418: }
2419: } else {
1.56 noro 2420: for ( i = 0; i < n; i++ ) printf(i==n-1?"%d":"%d,",GET_EXP(dl,i));
1.43 noro 2421: }
1.1 noro 2422: printf(">>");
2423: }
2424:
2425: void nd_print(ND p)
2426: {
2427: NM m;
2428:
2429: if ( !p )
2430: printf("0\n");
2431: else {
2432: for ( m = BDY(p); m; m = NEXT(m) ) {
1.71 noro 2433: if ( CM(m) & 0x80000000 ) printf("+@_%d*",IFTOF(CM(m)));
2434: else printf("+%d*",CM(m));
1.14 noro 2435: ndl_print(DL(m));
1.1 noro 2436: }
2437: printf("\n");
2438: }
2439: }
2440:
1.113 noro 2441: void nd_print_q(ND p)
1.16 noro 2442: {
2443: NM m;
2444:
2445: if ( !p )
2446: printf("0\n");
2447: else {
2448: for ( m = BDY(p); m; m = NEXT(m) ) {
2449: printf("+");
1.113 noro 2450: printexpr(CO,(Obj)CQ(m));
1.16 noro 2451: printf("*");
2452: ndl_print(DL(m));
2453: }
2454: printf("\n");
2455: }
2456: }
2457:
1.1 noro 2458: void ndp_print(ND_pairs d)
2459: {
2460: ND_pairs t;
2461:
1.34 noro 2462: for ( t = d; t; t = NEXT(t) ) printf("%d,%d ",t->i1,t->i2);
1.1 noro 2463: printf("\n");
2464: }
2465:
1.20 noro 2466: void nd_removecont(int mod,ND p)
1.16 noro 2467: {
2468: int i,n;
1.113 noro 2469: Q *w;
2470: Q dvr,t;
1.16 noro 2471: NM m;
1.21 noro 2472: struct oVECT v;
1.113 noro 2473: N q,r;
1.16 noro 2474:
1.71 noro 2475: if ( mod == -1 ) nd_mul_c(mod,p,_invsf(HCM(p)));
2476: else if ( mod ) nd_mul_c(mod,p,invm(HCM(p),mod));
1.20 noro 2477: else {
2478: for ( m = BDY(p), n = 0; m; m = NEXT(m), n++ );
1.113 noro 2479: w = (Q *)ALLOCA(n*sizeof(Q));
1.21 noro 2480: v.len = n;
2481: v.body = (pointer *)w;
1.113 noro 2482: for ( m = BDY(p), i = 0; i < n; m = NEXT(m), i++ ) w[i] = CQ(m);
1.21 noro 2483: removecont_array(w,n);
1.113 noro 2484: for ( m = BDY(p), i = 0; i < n; m = NEXT(m), i++ ) CQ(m) = w[i];
1.16 noro 2485: }
2486: }
2487:
1.21 noro 2488: void nd_removecont2(ND p1,ND p2)
2489: {
2490: int i,n1,n2,n;
1.113 noro 2491: Q *w;
2492: Q dvr,t;
1.21 noro 2493: NM m;
2494: struct oVECT v;
1.113 noro 2495: N q,r;
1.21 noro 2496:
2497: if ( !p1 ) {
2498: nd_removecont(0,p2); return;
2499: } else if ( !p2 ) {
2500: nd_removecont(0,p1); return;
2501: }
2502: n1 = nd_length(p1);
2503: n2 = nd_length(p2);
2504: n = n1+n2;
1.113 noro 2505: w = (Q *)ALLOCA(n*sizeof(Q));
1.21 noro 2506: v.len = n;
2507: v.body = (pointer *)w;
1.113 noro 2508: for ( m = BDY(p1), i = 0; i < n1; m = NEXT(m), i++ ) w[i] = CQ(m);
2509: for ( m = BDY(p2); i < n; m = NEXT(m), i++ ) w[i] = CQ(m);
1.21 noro 2510: removecont_array(w,n);
1.113 noro 2511: for ( m = BDY(p1), i = 0; i < n1; m = NEXT(m), i++ ) CQ(m) = w[i];
2512: for ( m = BDY(p2); i < n; m = NEXT(m), i++ ) CQ(m) = w[i];
1.21 noro 2513: }
2514:
1.20 noro 2515: void ndv_removecont(int mod,NDV p)
1.16 noro 2516: {
2517: int i,len;
1.113 noro 2518: Q *w;
2519: Q dvr,t;
1.16 noro 2520: NMV m;
2521:
1.71 noro 2522: if ( mod == -1 )
2523: ndv_mul_c(mod,p,_invsf(HCM(p)));
2524: else if ( mod )
1.20 noro 2525: ndv_mul_c(mod,p,invm(HCM(p),mod));
2526: else {
2527: len = p->len;
1.113 noro 2528: w = (Q *)ALLOCA(len*sizeof(Q));
2529: for ( m = BDY(p), i = 0; i < len; NMV_ADV(m), i++ ) w[i] = CQ(m);
2530: sortbynm(w,len);
2531: qltozl(w,len,&dvr);
2532: for ( m = BDY(p), i = 0; i < len; NMV_ADV(m), i++ ) {
2533: divq(CQ(m),dvr,&t); CQ(m) = t;
2534: }
1.16 noro 2535: }
1.21 noro 2536: }
2537:
1.61 noro 2538: void ndv_homogenize(NDV p,int obpe,int oadv,EPOS oepos)
2539: {
2540: int len,i,max;
2541: NMV m,mr0,mr,t;
2542:
2543: len = p->len;
2544: for ( m = BDY(p), i = 0, max = 0; i < len; NMV_OADV(m), i++ )
2545: max = MAX(max,TD(DL(m)));
2546: mr0 = nmv_adv>oadv?(NMV)REALLOC(BDY(p),len*nmv_adv):BDY(p);
2547: m = (NMV)((char *)mr0+(len-1)*oadv);
2548: mr = (NMV)((char *)mr0+(len-1)*nmv_adv);
2549: t = (NMV)ALLOCA(nmv_adv);
2550: for ( i = 0; i < len; i++, NMV_OPREV(m), NMV_PREV(mr) ) {
2551: ndl_homogenize(DL(m),DL(t),obpe,oepos,max);
1.113 noro 2552: CQ(mr) = CQ(m);
1.61 noro 2553: ndl_copy(DL(t),DL(mr));
2554: }
2555: NV(p)++;
2556: BDY(p) = mr0;
2557: }
2558:
1.45 noro 2559: void ndv_dehomogenize(NDV p,struct order_spec *ord)
1.23 noro 2560: {
1.45 noro 2561: int i,j,adj,len,newnvar,newwpd,newadv,newexporigin;
1.113 noro 2562: Q *w;
2563: Q dvr,t;
1.23 noro 2564: NMV m,r;
2565:
2566: len = p->len;
2567: newnvar = nd_nvar-1;
1.48 noro 2568: newexporigin = nd_get_exporigin(ord);
1.45 noro 2569: newwpd = newnvar/nd_epw+(newnvar%nd_epw?1:0)+newexporigin;
1.23 noro 2570: for ( m = BDY(p), i = 0; i < len; NMV_ADV(m), i++ )
1.34 noro 2571: ndl_dehomogenize(DL(m));
1.23 noro 2572: if ( newwpd != nd_wpd ) {
1.90 noro 2573: newadv = ROUND_FOR_ALIGN(sizeof(struct oNMV)+(newwpd-1)*sizeof(UINT));
1.61 noro 2574: for ( m = r = BDY(p), i = 0; i < len; NMV_ADV(m), NDV_NADV(r), i++ ) {
1.113 noro 2575: CQ(r) = CQ(m);
1.45 noro 2576: for ( j = 0; j < newexporigin; j++ ) DL(r)[j] = DL(m)[j];
2577: adj = nd_exporigin-newexporigin;
2578: for ( ; j < newwpd; j++ ) DL(r)[j] = DL(m)[j+adj];
1.23 noro 2579: }
2580: }
2581: NV(p)--;
2582: }
2583:
1.113 noro 2584: void removecont_array(Q *c,int n)
1.21 noro 2585: {
2586: struct oVECT v;
1.113 noro 2587: Q d0,d1,a,u,u1,gcd;
1.110 noro 2588: int i,j;
1.113 noro 2589: N qn,rn,gn;
2590: Q *q,*r;
1.21 noro 2591:
1.113 noro 2592: q = (Q *)ALLOCA(n*sizeof(Q));
2593: r = (Q *)ALLOCA(n*sizeof(Q));
1.21 noro 2594: v.id = O_VECT; v.len = n; v.body = (pointer *)c;
1.113 noro 2595: igcdv_estimate(&v,&d0);
2596: for ( i = 0; i < n; i++ ) {
2597: divn(NM(c[i]),NM(d0),&qn,&rn);
2598: NTOQ(qn,SGN(c[i])*SGN(d0),q[i]);
2599: NTOQ(rn,SGN(c[i]),r[i]);
2600: }
1.34 noro 2601: for ( i = 0; i < n; i++ ) if ( r[i] ) break;
1.21 noro 2602: if ( i < n ) {
2603: v.id = O_VECT; v.len = n; v.body = (pointer *)r;
1.113 noro 2604: igcdv(&v,&d1);
2605: gcdn(NM(d0),NM(d1),&gn); NTOQ(gn,1,gcd);
2606: divsn(NM(d0),gn,&qn); NTOQ(qn,1,a);
1.21 noro 2607: for ( i = 0; i < n; i++ ) {
1.113 noro 2608: mulq(a,q[i],&u);
2609: if ( r[i] ) {
2610: divsn(NM(r[i]),gn,&qn); NTOQ(qn,SGN(r[i]),u1);
2611: addq(u,u1,&q[i]);
2612: } else
2613: q[i] = u;
1.21 noro 2614: }
2615: }
1.34 noro 2616: for ( i = 0; i < n; i++ ) c[i] = q[i];
1.16 noro 2617: }
2618:
1.19 noro 2619: void nd_mul_c(int mod,ND p,int mul)
1.1 noro 2620: {
2621: NM m;
2622: int c,c1;
2623:
1.34 noro 2624: if ( !p ) return;
1.115 ! noro 2625: if ( mul == 1 ) return;
1.71 noro 2626: if ( mod == -1 )
2627: for ( m = BDY(p); m; m = NEXT(m) )
2628: CM(m) = _mulsf(CM(m),mul);
2629: else
2630: for ( m = BDY(p); m; m = NEXT(m) ) {
2631: c1 = CM(m); DMAR(c1,mul,0,mod,c); CM(m) = c;
2632: }
1.1 noro 2633: }
2634:
1.113 noro 2635: void nd_mul_c_q(ND p,Q mul)
1.16 noro 2636: {
2637: NM m;
1.113 noro 2638: Q c;
1.16 noro 2639:
1.113 noro 2640: if ( !p ) return;
1.115 ! noro 2641: if ( UNIQ(mul) ) return;
1.16 noro 2642: for ( m = BDY(p); m; m = NEXT(m) ) {
1.113 noro 2643: mulq(CQ(m),mul,&c); CQ(m) = c;
1.16 noro 2644: }
2645: }
2646:
1.61 noro 2647: void nd_mul_c_p(VL vl,ND p,P mul)
2648: {
2649: NM m;
2650: P c;
2651:
2652: if ( !p ) return;
2653: for ( m = BDY(p); m; m = NEXT(m) ) {
2654: mulp(vl,CP(m),mul,&c); CP(m) = c;
2655: }
2656: }
2657:
1.1 noro 2658: void nd_free(ND p)
2659: {
2660: NM t,s;
2661:
1.34 noro 2662: if ( !p ) return;
1.1 noro 2663: t = BDY(p);
2664: while ( t ) {
2665: s = NEXT(t);
2666: FREENM(t);
2667: t = s;
2668: }
2669: FREEND(p);
2670: }
2671:
1.23 noro 2672: void ndv_free(NDV p)
2673: {
2674: GC_free(BDY(p));
2675: }
2676:
1.61 noro 2677: void nd_append_red(UINT *d,int i)
1.1 noro 2678: {
1.13 noro 2679: RHist m,m0;
1.1 noro 2680: int h;
2681:
1.13 noro 2682: NEWRHist(m);
1.34 noro 2683: h = ndl_hash_value(d);
1.13 noro 2684: m->index = i;
1.14 noro 2685: ndl_copy(d,DL(m));
1.1 noro 2686: NEXT(m) = nd_red[h];
2687: nd_red[h] = m;
2688: }
2689:
1.61 noro 2690: UINT *ndv_compute_bound(NDV p)
1.1 noro 2691: {
1.61 noro 2692: UINT *d1,*d2,*t;
2693: UINT u;
1.57 noro 2694: int i,j,k,l,len,ind;
1.45 noro 2695: NMV m;
1.1 noro 2696:
2697: if ( !p )
2698: return 0;
1.61 noro 2699: d1 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
2700: d2 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
1.45 noro 2701: len = LEN(p);
2702: m = BDY(p); ndl_copy(DL(m),d1); NMV_ADV(m);
2703: for ( i = 1; i < len; i++, NMV_ADV(m) ) {
1.14 noro 2704: ndl_lcm(DL(m),d1,d2);
1.1 noro 2705: t = d1; d1 = d2; d2 = t;
2706: }
1.12 noro 2707: l = nd_nvar+31;
1.61 noro 2708: t = (UINT *)MALLOC_ATOMIC(l*sizeof(UINT));
1.57 noro 2709: for ( i = nd_exporigin, ind = 0; i < nd_wpd; i++ ) {
2710: u = d1[i];
2711: k = (nd_epw-1)*nd_bpe;
2712: for ( j = 0; j < nd_epw; j++, k -= nd_bpe, ind++ )
2713: t[ind] = (u>>k)&nd_mask0;
2714: }
2715: for ( ; ind < l; ind++ ) t[ind] = 0;
1.1 noro 2716: return t;
2717: }
2718:
1.99 noro 2719: UINT *nd_compute_bound(ND p)
2720: {
2721: UINT *d1,*d2,*t;
2722: UINT u;
2723: int i,j,k,l,len,ind;
2724: NM m;
2725:
2726: if ( !p )
2727: return 0;
2728: d1 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
2729: d2 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
2730: len = LEN(p);
2731: m = BDY(p); ndl_copy(DL(m),d1); m = NEXT(m);
2732: for ( m = NEXT(m); m; m = NEXT(m) ) {
2733: ndl_lcm(DL(m),d1,d2);
2734: t = d1; d1 = d2; d2 = t;
2735: }
2736: l = nd_nvar+31;
2737: t = (UINT *)MALLOC_ATOMIC(l*sizeof(UINT));
2738: for ( i = nd_exporigin, ind = 0; i < nd_wpd; i++ ) {
2739: u = d1[i];
2740: k = (nd_epw-1)*nd_bpe;
2741: for ( j = 0; j < nd_epw; j++, k -= nd_bpe, ind++ )
2742: t[ind] = (u>>k)&nd_mask0;
2743: }
2744: for ( ; ind < l; ind++ ) t[ind] = 0;
2745: return t;
2746: }
2747:
1.48 noro 2748: int nd_get_exporigin(struct order_spec *ord)
2749: {
1.51 noro 2750: switch ( ord->id ) {
1.96 noro 2751: case 0: case 2:
1.48 noro 2752: return 1;
1.41 noro 2753: case 1:
2754: /* block order */
1.43 noro 2755: /* d[0]:weight d[1]:w0,...,d[nd_exporigin-1]:w(n-1) */
1.48 noro 2756: return ord->ord.block.length+1;
1.96 noro 2757: case 3:
2758: error("nd_get_exporigin : composite order is not supported yet.");
1.41 noro 2759: }
1.48 noro 2760: }
2761:
1.61 noro 2762: void nd_setup_parameters(int nvar,int max) {
1.79 noro 2763: int i,j,n,elen,ord_o,ord_l,l,s,wpd;
1.57 noro 2764: struct order_pair *op;
1.48 noro 2765:
1.73 noro 2766: nd_nvar = nvar;
1.79 noro 2767: if ( max ) {
1.83 noro 2768: /* XXX */
2769: if ( do_weyl ) nd_bpe = 32;
2770: else if ( max < 2 ) nd_bpe = 1;
1.79 noro 2771: else if ( max < 4 ) nd_bpe = 2;
2772: else if ( max < 8 ) nd_bpe = 3;
2773: else if ( max < 16 ) nd_bpe = 4;
2774: else if ( max < 32 ) nd_bpe = 5;
2775: else if ( max < 64 ) nd_bpe = 6;
2776: else if ( max < 256 ) nd_bpe = 8;
2777: else if ( max < 1024 ) nd_bpe = 10;
2778: else if ( max < 65536 ) nd_bpe = 16;
2779: else nd_bpe = 32;
2780: }
1.61 noro 2781: nd_epw = (sizeof(UINT)*8)/nd_bpe;
1.48 noro 2782: elen = nd_nvar/nd_epw+(nd_nvar%nd_epw?1:0);
2783: nd_exporigin = nd_get_exporigin(nd_ord);
1.79 noro 2784: wpd = nd_exporigin+elen;
2785: if ( wpd != nd_wpd ) {
2786: nd_free_private_storage();
2787: nd_wpd = wpd;
2788: }
1.1 noro 2789: if ( nd_bpe < 32 ) {
2790: nd_mask0 = (1<<nd_bpe)-1;
2791: } else {
2792: nd_mask0 = 0xffffffff;
2793: }
2794: bzero(nd_mask,sizeof(nd_mask));
2795: nd_mask1 = 0;
2796: for ( i = 0; i < nd_epw; i++ ) {
2797: nd_mask[nd_epw-i-1] = (nd_mask0<<(i*nd_bpe));
2798: nd_mask1 |= (1<<(nd_bpe-1))<<(i*nd_bpe);
2799: }
1.90 noro 2800: nmv_adv = ROUND_FOR_ALIGN(sizeof(struct oNMV)+(nd_wpd-1)*sizeof(UINT));
1.57 noro 2801: nd_epos = nd_create_epos(nd_ord);
1.43 noro 2802: nd_blockmask = nd_create_blockmask(nd_ord);
1.96 noro 2803: nd_work_vector = (int *)REALLOC(nd_work_vector,nd_nvar*sizeof(int));
1.1 noro 2804: }
2805:
1.103 noro 2806: ND_pairs nd_reconstruct(int trace,ND_pairs d)
1.1 noro 2807: {
1.37 noro 2808: int i,obpe,oadv,h;
1.71 noro 2809: static NM prev_nm_free_list;
2810: static ND_pairs prev_ndp_free_list;
1.13 noro 2811: RHist mr0,mr;
2812: RHist r;
1.37 noro 2813: RHist *old_red;
1.71 noro 2814: ND_pairs s0,s,t;
1.43 noro 2815: EPOS oepos;
1.15 noro 2816:
1.1 noro 2817: obpe = nd_bpe;
1.11 noro 2818: oadv = nmv_adv;
1.43 noro 2819: oepos = nd_epos;
1.63 noro 2820: if ( obpe < 2 ) nd_bpe = 2;
2821: else if ( obpe < 3 ) nd_bpe = 3;
1.62 noro 2822: else if ( obpe < 4 ) nd_bpe = 4;
1.63 noro 2823: else if ( obpe < 5 ) nd_bpe = 5;
1.34 noro 2824: else if ( obpe < 6 ) nd_bpe = 6;
2825: else if ( obpe < 8 ) nd_bpe = 8;
1.63 noro 2826: else if ( obpe < 10 ) nd_bpe = 10;
1.34 noro 2827: else if ( obpe < 16 ) nd_bpe = 16;
2828: else if ( obpe < 32 ) nd_bpe = 32;
2829: else error("nd_reconstruct : exponent too large");
1.5 noro 2830:
1.73 noro 2831: nd_setup_parameters(nd_nvar,0);
1.1 noro 2832: prev_nm_free_list = _nm_free_list;
2833: prev_ndp_free_list = _ndp_free_list;
2834: _nm_free_list = 0;
2835: _ndp_free_list = 0;
1.53 noro 2836: for ( i = nd_psn-1; i >= 0; i-- ) ndv_realloc(nd_ps[i],obpe,oadv,oepos);
2837: if ( trace )
2838: for ( i = nd_psn-1; i >= 0; i-- )
2839: ndv_realloc(nd_ps_trace[i],obpe,oadv,oepos);
1.1 noro 2840: s0 = 0;
2841: for ( t = d; t; t = NEXT(t) ) {
2842: NEXTND_pairs(s0,s);
2843: s->i1 = t->i1;
2844: s->i2 = t->i2;
1.14 noro 2845: SG(s) = SG(t);
1.61 noro 2846: ndl_reconstruct(LCM(t),LCM(s),obpe,oepos);
1.1 noro 2847: }
1.37 noro 2848:
2849: old_red = (RHist *)ALLOCA(REDTAB_LEN*sizeof(RHist));
1.6 noro 2850: for ( i = 0; i < REDTAB_LEN; i++ ) {
1.37 noro 2851: old_red[i] = nd_red[i];
2852: nd_red[i] = 0;
2853: }
2854: for ( i = 0; i < REDTAB_LEN; i++ )
2855: for ( r = old_red[i]; r; r = NEXT(r) ) {
2856: NEWRHist(mr);
1.13 noro 2857: mr->index = r->index;
1.20 noro 2858: SG(mr) = SG(r);
1.61 noro 2859: ndl_reconstruct(DL(r),DL(mr),obpe,oepos);
1.37 noro 2860: h = ndl_hash_value(DL(mr));
2861: NEXT(mr) = nd_red[h];
2862: nd_red[h] = mr;
1.6 noro 2863: }
1.37 noro 2864: for ( i = 0; i < REDTAB_LEN; i++ ) old_red[i] = 0;
2865: old_red = 0;
1.11 noro 2866: for ( i = 0; i < nd_psn; i++ ) {
1.20 noro 2867: NEWRHist(r); SG(r) = SG(nd_psh[i]);
1.61 noro 2868: ndl_reconstruct(DL(nd_psh[i]),DL(r),obpe,oepos);
1.13 noro 2869: nd_psh[i] = r;
1.11 noro 2870: }
1.1 noro 2871: if ( s0 ) NEXT(s) = 0;
2872: prev_nm_free_list = 0;
2873: prev_ndp_free_list = 0;
1.71 noro 2874: #if 0
1.1 noro 2875: GC_gcollect();
1.71 noro 2876: #endif
1.1 noro 2877: return s0;
2878: }
2879:
1.61 noro 2880: void ndl_reconstruct(UINT *d,UINT *r,int obpe,EPOS oepos)
1.1 noro 2881: {
1.57 noro 2882: int n,i,ei,oepw,omask0,j,s,ord_l,l;
1.43 noro 2883: struct order_pair *op;
1.1 noro 2884:
2885: n = nd_nvar;
1.61 noro 2886: oepw = (sizeof(UINT)*8)/obpe;
1.43 noro 2887: omask0 = (1<<obpe)-1;
1.34 noro 2888: TD(r) = TD(d);
1.41 noro 2889: for ( i = nd_exporigin; i < nd_wpd; i++ ) r[i] = 0;
1.43 noro 2890: if ( nd_blockmask ) {
2891: l = nd_blockmask->n;
2892: op = nd_blockmask->order_pair;
2893: for ( i = 1; i < nd_exporigin; i++ )
2894: r[i] = d[i];
2895: for ( j = 0, s = 0; j < l; j++ ) {
2896: ord_l = op[j].length;
1.57 noro 2897: for ( i = 0; i < ord_l; i++, s++ ) {
2898: ei = GET_EXP_OLD(d,s);
2899: PUT_EXP(r,s,ei);
2900: }
1.1 noro 2901: }
1.43 noro 2902: } else {
1.56 noro 2903: for ( i = 0; i < n; i++ ) {
2904: ei = GET_EXP_OLD(d,i);
2905: PUT_EXP(r,i,ei);
2906: }
1.1 noro 2907: }
2908: }
1.3 noro 2909:
1.6 noro 2910: ND nd_copy(ND p)
2911: {
2912: NM m,mr,mr0;
1.41 noro 2913: int c,n;
1.6 noro 2914: ND r;
2915:
2916: if ( !p )
2917: return 0;
2918: else {
2919: for ( mr0 = 0, m = BDY(p); m; m = NEXT(m) ) {
2920: NEXTNM(mr0,mr);
1.14 noro 2921: CM(mr) = CM(m);
2922: ndl_copy(DL(m),DL(mr));
1.6 noro 2923: }
2924: NEXT(mr) = 0;
1.31 noro 2925: MKND(NV(p),mr0,LEN(p),r);
1.14 noro 2926: SG(r) = SG(p);
1.6 noro 2927: return r;
2928: }
2929: }
2930:
1.53 noro 2931: int nd_sp(int mod,int trace,ND_pairs p,ND *rp)
1.11 noro 2932: {
2933: NM m;
2934: NDV p1,p2;
2935: ND t1,t2;
1.61 noro 2936: UINT *lcm;
1.31 noro 2937: int td;
1.11 noro 2938:
1.77 noro 2939: if ( !mod && nd_demand ) {
2940: p1 = ndv_load(p->i1); p2 = ndv_load(p->i2);
1.53 noro 2941: } else {
1.77 noro 2942: if ( trace ) {
2943: p1 = nd_ps_trace[p->i1]; p2 = nd_ps_trace[p->i2];
2944: } else {
2945: p1 = nd_ps[p->i1]; p2 = nd_ps[p->i2];
2946: }
1.20 noro 2947: }
1.34 noro 2948: lcm = LCM(p);
1.11 noro 2949: NEWNM(m);
1.113 noro 2950: CQ(m) = HCQ(p2);
1.34 noro 2951: ndl_sub(lcm,HDL(p1),DL(m));
1.56 noro 2952: if ( ndl_check_bound2(p->i1,DL(m)) )
2953: return 0;
1.55 noro 2954: t1 = ndv_mul_nm(mod,m,p1);
1.71 noro 2955: if ( mod == -1 ) CM(m) = _chsgnsf(HCM(p1));
2956: else if ( mod ) CM(m) = mod-HCM(p1);
1.113 noro 2957: else chsgnq(HCQ(p1),&CQ(m));
1.34 noro 2958: ndl_sub(lcm,HDL(p2),DL(m));
1.14 noro 2959: if ( ndl_check_bound2(p->i2,DL(m)) ) {
1.11 noro 2960: nd_free(t1);
2961: return 0;
2962: }
1.55 noro 2963: t2 = ndv_mul_nm(mod,m,p2);
1.31 noro 2964: *rp = nd_add(mod,t1,t2);
1.11 noro 2965: FREENM(m);
2966: return 1;
2967: }
2968:
1.19 noro 2969: void ndv_mul_c(int mod,NDV p,int mul)
1.11 noro 2970: {
2971: NMV m;
2972: int c,c1,len,i;
2973:
1.34 noro 2974: if ( !p ) return;
1.14 noro 2975: len = LEN(p);
1.71 noro 2976: if ( mod == -1 )
2977: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) )
2978: CM(m) = _mulsf(CM(m),mul);
2979: else
2980: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
2981: c1 = CM(m); DMAR(c1,mul,0,mod,c); CM(m) = c;
2982: }
1.11 noro 2983: }
2984:
1.113 noro 2985: void ndv_mul_c_q(NDV p,Q mul)
1.16 noro 2986: {
2987: NMV m;
1.113 noro 2988: Q c;
1.16 noro 2989: int len,i;
2990:
1.34 noro 2991: if ( !p ) return;
1.16 noro 2992: len = LEN(p);
2993: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
1.113 noro 2994: mulq(CQ(m),mul,&c); CQ(m) = c;
1.16 noro 2995: }
2996: }
2997:
1.55 noro 2998: ND weyl_ndv_mul_nm(int mod,NM m0,NDV p) {
2999: int n2,i,j,l,n,tlen;
1.61 noro 3000: UINT *d0;
1.55 noro 3001: NM *tab,*psum;
3002: ND s,r;
3003: NM t;
3004: NMV m1;
3005:
3006: if ( !p ) return 0;
3007: n = NV(p); n2 = n>>1;
3008: d0 = DL(m0);
3009: l = LEN(p);
3010: for ( i = 0, tlen = 1; i < n2; i++ ) tlen *= (GET_EXP(d0,n2+i)+1);
3011: tab = (NM *)ALLOCA(tlen*sizeof(NM));
3012: psum = (NM *)ALLOCA(tlen*sizeof(NM));
3013: for ( i = 0; i < tlen; i++ ) psum[i] = 0;
1.56 noro 3014: m1 = (NMV)(((char *)BDY(p))+nmv_adv*(l-1));
3015: for ( i = l-1; i >= 0; i--, NMV_PREV(m1) ) {
1.55 noro 3016: /* m0(NM) * m1(NMV) => tab(NM) */
1.56 noro 3017: weyl_mul_nm_nmv(n,mod,m0,m1,tab,tlen);
1.55 noro 3018: for ( j = 0; j < tlen; j++ ) {
3019: if ( tab[j] ) {
3020: NEXT(tab[j]) = psum[j]; psum[j] = tab[j];
3021: }
3022: }
3023: }
3024: for ( i = tlen-1, r = 0; i >= 0; i-- )
3025: if ( psum[i] ) {
3026: for ( j = 0, t = psum[i]; t; t = NEXT(t), j++ );
3027: MKND(n,psum[i],j,s);
3028: r = nd_add(mod,r,s);
3029: }
1.56 noro 3030: if ( r ) SG(r) = SG(p)+TD(d0);
3031: return r;
1.55 noro 3032: }
3033:
1.56 noro 3034: /* product of monomials */
3035: /* XXX block order is not handled correctly */
3036:
1.55 noro 3037: void weyl_mul_nm_nmv(int n,int mod,NM m0,NMV m1,NM *tab,int tlen)
3038: {
1.56 noro 3039: int i,n2,j,s,curlen,homo,h,a,b,k,l,u,min;
1.61 noro 3040: UINT *d0,*d1,*d,*dt,*ctab;
1.113 noro 3041: Q *ctab_q;
3042: Q q,q1;
1.61 noro 3043: UINT c0,c1,c;
1.55 noro 3044: NM *p;
3045: NM m,t;
3046:
3047: for ( i = 0; i < tlen; i++ ) tab[i] = 0;
3048: if ( !m0 || !m1 ) return;
3049: d0 = DL(m0); d1 = DL(m1); n2 = n>>1;
3050: NEWNM(m); d = DL(m);
1.56 noro 3051: if ( mod ) {
3052: c0 = CM(m0); c1 = CM(m1); DMAR(c0,c1,0,mod,c); CM(m) = c;
3053: } else
1.113 noro 3054: mulq(CQ(m0),CQ(m1),&CQ(m));
1.55 noro 3055: for ( i = 0; i < nd_wpd; i++ ) d[i] = 0;
3056: homo = n&1 ? 1 : 0;
3057: if ( homo ) {
3058: /* offset of h-degree */
3059: h = GET_EXP(d0,n-1)+GET_EXP(d1,n-1);
3060: PUT_EXP(DL(m),n-1,h);
3061: TD(DL(m)) = h;
1.61 noro 3062: if ( nd_blockmask ) ndl_weight_mask(DL(m));
1.55 noro 3063: }
3064: tab[0] = m;
3065: NEWNM(m); d = DL(m);
1.57 noro 3066: for ( i = 0, curlen = 1; i < n2; i++ ) {
1.55 noro 3067: a = GET_EXP(d0,i); b = GET_EXP(d1,n2+i);
3068: k = GET_EXP(d0,n2+i); l = GET_EXP(d1,i);
3069: /* xi^a*(Di^k*xi^l)*Di^b */
3070: a += l; b += k;
1.56 noro 3071: s = MUL_WEIGHT(a,i)+MUL_WEIGHT(b,n2+i);
1.55 noro 3072: if ( !k || !l ) {
3073: for ( j = 0; j < curlen; j++ )
1.56 noro 3074: if ( t = tab[j] ) {
3075: dt = DL(t);
3076: PUT_EXP(dt,i,a); PUT_EXP(dt,n2+i,b); TD(dt) += s;
1.61 noro 3077: if ( nd_blockmask ) ndl_weight_mask(dt);
1.55 noro 3078: }
3079: curlen *= k+1;
3080: continue;
3081: }
3082: min = MIN(k,l);
1.56 noro 3083: if ( mod ) {
1.61 noro 3084: ctab = (UINT *)ALLOCA((min+1)*sizeof(UINT));
1.56 noro 3085: mkwcm(k,l,mod,ctab);
3086: } else {
1.113 noro 3087: ctab_q = (Q *)ALLOCA((min+1)*sizeof(Q));
3088: mkwc(k,l,ctab_q);
1.56 noro 3089: }
1.57 noro 3090: for ( j = min; j >= 0; j-- ) {
1.56 noro 3091: for ( u = 0; u < nd_wpd; u++ ) d[u] = 0;
1.55 noro 3092: PUT_EXP(d,i,a-j); PUT_EXP(d,n2+i,b-j);
1.56 noro 3093: h = MUL_WEIGHT(a-j,i)+MUL_WEIGHT(b-j,n2+i);
1.55 noro 3094: if ( homo ) {
3095: TD(d) = s;
1.56 noro 3096: PUT_EXP(d,n-1,s-h);
1.55 noro 3097: } else TD(d) = h;
1.61 noro 3098: if ( nd_blockmask ) ndl_weight_mask(d);
1.56 noro 3099: if ( mod ) c = ctab[j];
1.113 noro 3100: else q = ctab_q[j];
1.57 noro 3101: p = tab+curlen*j;
3102: if ( j == 0 ) {
3103: for ( u = 0; u < curlen; u++, p++ ) {
3104: if ( tab[u] ) {
3105: ndl_addto(DL(tab[u]),d);
3106: if ( mod ) {
3107: c0 = CM(tab[u]); DMAR(c0,c,0,mod,c1); CM(tab[u]) = c1;
3108: } else {
1.113 noro 3109: mulq(CQ(tab[u]),q,&q1); CQ(tab[u]) = q1;
1.57 noro 3110: }
3111: }
1.56 noro 3112: }
1.57 noro 3113: } else {
3114: for ( u = 0; u < curlen; u++, p++ ) {
3115: if ( tab[u] ) {
3116: NEWNM(t);
3117: ndl_add(DL(tab[u]),d,DL(t));
3118: if ( mod ) {
3119: c0 = CM(tab[u]); DMAR(c0,c,0,mod,c1); CM(t) = c1;
3120: } else
1.113 noro 3121: mulq(CQ(tab[u]),q,&CQ(t));
1.57 noro 3122: *p = t;
3123: }
1.55 noro 3124: }
3125: }
3126: }
3127: curlen *= k+1;
3128: }
3129: FREENM(m);
3130: }
3131:
1.63 noro 3132: ND ndv_mul_nm_symbolic(NM m0,NDV p)
3133: {
3134: NM mr,mr0;
3135: NMV m;
3136: UINT *d,*dt,*dm;
3137: int c,n,td,i,c1,c2,len;
1.113 noro 3138: Q q;
1.63 noro 3139: ND r;
3140:
3141: if ( !p ) return 0;
3142: else {
3143: n = NV(p); m = BDY(p);
3144: d = DL(m0);
3145: len = LEN(p);
3146: mr0 = 0;
3147: td = TD(d);
3148: c = CM(m0);
3149: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
3150: NEXTNM(mr0,mr);
3151: CM(mr) = 1;
3152: ndl_add(DL(m),d,DL(mr));
3153: }
3154: NEXT(mr) = 0;
3155: MKND(NV(p),mr0,len,r);
3156: SG(r) = SG(p) + TD(d);
3157: return r;
3158: }
3159: }
3160:
1.55 noro 3161: ND ndv_mul_nm(int mod,NM m0,NDV p)
1.9 noro 3162: {
3163: NM mr,mr0;
3164: NMV m;
1.61 noro 3165: UINT *d,*dt,*dm;
1.9 noro 3166: int c,n,td,i,c1,c2,len;
1.113 noro 3167: Q q;
1.9 noro 3168: ND r;
3169:
1.34 noro 3170: if ( !p ) return 0;
1.55 noro 3171: else if ( do_weyl )
1.71 noro 3172: if ( mod == -1 )
3173: error("ndv_mul_nm : not implemented (weyl)");
3174: else
3175: return weyl_ndv_mul_nm(mod,m0,p);
1.9 noro 3176: else {
3177: n = NV(p); m = BDY(p);
1.34 noro 3178: d = DL(m0);
1.14 noro 3179: len = LEN(p);
1.9 noro 3180: mr0 = 0;
1.34 noro 3181: td = TD(d);
1.71 noro 3182: if ( mod == -1 ) {
3183: c = CM(m0);
3184: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
3185: NEXTNM(mr0,mr);
3186: CM(mr) = _mulsf(CM(m),c);
3187: ndl_add(DL(m),d,DL(mr));
3188: }
3189: } else if ( mod ) {
1.16 noro 3190: c = CM(m0);
3191: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
3192: NEXTNM(mr0,mr);
3193: c1 = CM(m);
1.19 noro 3194: DMAR(c1,c,0,mod,c2);
1.16 noro 3195: CM(mr) = c2;
3196: ndl_add(DL(m),d,DL(mr));
3197: }
3198: } else {
1.113 noro 3199: q = CQ(m0);
1.16 noro 3200: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
3201: NEXTNM(mr0,mr);
1.113 noro 3202: mulq(CQ(m),q,&CQ(mr));
1.16 noro 3203: ndl_add(DL(m),d,DL(mr));
3204: }
1.4 noro 3205: }
1.9 noro 3206: NEXT(mr) = 0;
1.31 noro 3207: MKND(NV(p),mr0,len,r);
1.34 noro 3208: SG(r) = SG(p) + TD(d);
1.9 noro 3209: return r;
1.4 noro 3210: }
3211: }
3212:
1.104 noro 3213: ND nd_quo(int mod,PGeoBucket bucket,NDV d)
1.99 noro 3214: {
3215: NM mq0,mq;
1.102 noro 3216: NMV tm;
1.113 noro 3217: Q q;
1.104 noro 3218: int i,nv,sg,c,c1,c2,hindex;
3219: ND p,t,r;
3220: N tnm;
3221:
1.99 noro 3222: if ( !p ) return 0;
3223: else {
1.104 noro 3224: nv = NV(d);
1.99 noro 3225: mq0 = 0;
1.102 noro 3226: tm = (NMV)ALLOCA(nmv_adv);
1.104 noro 3227: while ( 1 ) {
1.113 noro 3228: hindex = mod?head_pbucket(mod,bucket):head_pbucket_q(bucket);
1.104 noro 3229: if ( hindex < 0 ) break;
3230: p = bucket->body[hindex];
1.99 noro 3231: NEXTNM(mq0,mq);
3232: ndl_sub(HDL(p),HDL(d),DL(mq));
1.102 noro 3233: ndl_copy(DL(mq),DL(tm));
3234: if ( mod ) {
3235: c1 = invm(HCM(d),mod); c2 = HCM(p);
3236: DMAR(c1,c2,0,mod,c); CM(mq) = c;
3237: CM(tm) = mod-c;
3238: } else {
1.113 noro 3239: divsn(NM(HCQ(p)),NM(HCQ(d)),&tnm);
3240: NTOQ(tnm,SGN(HCQ(p))*SGN(HCQ(d)),CQ(mq));
3241: chsgnq(CQ(mq),&CQ(tm));
1.102 noro 3242: }
3243: t = ndv_mul_nmv_trunc(mod,tm,d,HDL(d));
1.104 noro 3244: bucket->body[hindex] = nd_remove_head(p);
3245: t = nd_remove_head(t);
3246: add_pbucket(mod,bucket,t);
3247: }
3248: if ( !mq0 )
3249: r = 0;
3250: else {
3251: NEXT(mq) = 0;
3252: for ( i = 0, mq = mq0; mq; mq = NEXT(mq), i++ );
3253: MKND(nv,mq0,i,r);
3254: /* XXX */
3255: SG(r) = HTD(r);
1.99 noro 3256: }
3257: return r;
3258: }
3259: }
3260:
1.43 noro 3261: void ndv_realloc(NDV p,int obpe,int oadv,EPOS oepos)
1.11 noro 3262: {
1.13 noro 3263: NMV m,mr,mr0,t;
3264: int len,i,k;
1.11 noro 3265:
1.61 noro 3266: if ( !p ) return;
3267: m = BDY(p); len = LEN(p);
3268: mr0 = nmv_adv>oadv?(NMV)REALLOC(BDY(p),len*nmv_adv):BDY(p);
3269: m = (NMV)((char *)mr0+(len-1)*oadv);
3270: mr = (NMV)((char *)mr0+(len-1)*nmv_adv);
3271: t = (NMV)ALLOCA(nmv_adv);
3272: for ( i = 0; i < len; i++, NMV_OPREV(m), NMV_PREV(mr) ) {
1.113 noro 3273: CQ(t) = CQ(m);
1.61 noro 3274: for ( k = 0; k < nd_wpd; k++ ) DL(t)[k] = 0;
3275: ndl_reconstruct(DL(m),DL(t),obpe,oepos);
1.113 noro 3276: CQ(mr) = CQ(t);
1.61 noro 3277: ndl_copy(DL(t),DL(mr));
3278: }
3279: BDY(p) = mr0;
3280: }
3281:
3282: NDV ndv_dup_realloc(NDV p,int obpe,int oadv,EPOS oepos)
3283: {
3284: NMV m,mr,mr0;
3285: int len,i;
3286: NDV r;
1.11 noro 3287:
1.61 noro 3288: if ( !p ) return 0;
3289: m = BDY(p); len = LEN(p);
3290: mr0 = mr = (NMV)MALLOC(len*nmv_adv);
3291: for ( i = 0; i < len; i++, NMV_OADV(m), NMV_ADV(mr) ) {
3292: ndl_zero(DL(mr));
3293: ndl_reconstruct(DL(m),DL(mr),obpe,oepos);
1.113 noro 3294: CQ(mr) = CQ(m);
1.11 noro 3295: }
1.61 noro 3296: MKNDV(NV(p),mr0,len,r);
3297: SG(r) = SG(p);
3298: return r;
1.11 noro 3299: }
3300:
1.61 noro 3301: /* duplicate p */
3302:
3303: NDV ndv_dup(int mod,NDV p)
1.3 noro 3304: {
3305: NDV d;
1.61 noro 3306: NMV t,m,m0;
1.3 noro 3307: int i,len;
3308:
1.34 noro 3309: if ( !p ) return 0;
1.31 noro 3310: len = LEN(p);
1.34 noro 3311: m0 = m = (NMV)(mod?MALLOC_ATOMIC(len*nmv_adv):MALLOC(len*nmv_adv));
1.61 noro 3312: for ( t = BDY(p), i = 0; i < len; i++, NMV_ADV(t), NMV_ADV(m) ) {
1.14 noro 3313: ndl_copy(DL(t),DL(m));
1.113 noro 3314: CQ(m) = CQ(t);
1.3 noro 3315: }
3316: MKNDV(NV(p),m0,len,d);
1.23 noro 3317: SG(d) = SG(p);
3318: return d;
3319: }
3320:
1.63 noro 3321: ND nd_dup(ND p)
3322: {
3323: ND d;
3324: NM t,m,m0;
3325:
3326: if ( !p ) return 0;
3327: for ( m0 = 0, t = BDY(p); t; t = NEXT(t) ) {
3328: NEXTNM(m0,m);
3329: ndl_copy(DL(t),DL(m));
1.113 noro 3330: CQ(m) = CQ(t);
1.63 noro 3331: }
3332: if ( m0 ) NEXT(m) = 0;
3333: MKND(NV(p),m0,LEN(p),d);
3334: SG(d) = SG(p);
3335: return d;
3336: }
3337:
1.61 noro 3338: /* XXX if p->len == 0 then it represents 0 */
3339:
3340: void ndv_mod(int mod,NDV p)
3341: {
3342: NMV t,d;
3343: int r;
3344: int i,len,dlen;
1.71 noro 3345: Obj gfs;
1.61 noro 3346:
3347: if ( !p ) return;
3348: len = LEN(p);
3349: dlen = 0;
1.71 noro 3350: if ( mod == -1 )
3351: for ( t = d = BDY(p), i = 0; i < len; i++, NMV_ADV(t) ) {
3352: simp_ff((Obj)CP(t),&gfs);
3353: r = FTOIF(CONT((GFS)gfs));
1.61 noro 3354: CM(d) = r;
3355: ndl_copy(DL(t),DL(d));
3356: NMV_ADV(d);
3357: dlen++;
3358: }
1.71 noro 3359: else
3360: for ( t = d = BDY(p), i = 0; i < len; i++, NMV_ADV(t) ) {
1.113 noro 3361: r = rem(NM(CQ(t)),mod);
1.71 noro 3362: if ( r ) {
1.113 noro 3363: if ( SGN(CQ(t)) < 0 )
3364: r = mod-r;
1.71 noro 3365: CM(d) = r;
3366: ndl_copy(DL(t),DL(d));
3367: NMV_ADV(d);
3368: dlen++;
3369: }
3370: }
1.61 noro 3371: LEN(p) = dlen;
3372: }
3373:
3374: NDV ptondv(VL vl,VL dvl,P p)
3375: {
3376: ND nd;
3377:
3378: nd = ptond(vl,dvl,p);
3379: return ndtondv(0,nd);
3380: }
3381:
3382: ND ptond(VL vl,VL dvl,P p)
1.23 noro 3383: {
1.61 noro 3384: int n,i,j,k,e;
3385: VL tvl;
3386: V v;
3387: DCP dc;
3388: DCP *w;
3389: ND r,s,t,u;
3390: P x;
3391: int c;
3392: UINT *d;
1.23 noro 3393: NM m,m0;
1.61 noro 3394:
3395: if ( !p )
3396: return 0;
3397: else if ( NUM(p) ) {
3398: NEWNM(m);
3399: ndl_zero(DL(m));
1.113 noro 3400: CQ(m) = (Q)p;
1.61 noro 3401: NEXT(m) = 0;
3402: MKND(nd_nvar,m,1,r);
3403: SG(r) = 0;
3404: return r;
3405: } else {
3406: for ( dc = DC(p), k = 0; dc; dc = NEXT(dc), k++ );
3407: w = (DCP *)ALLOCA(k*sizeof(DCP));
3408: for ( dc = DC(p), j = 0; j < k; dc = NEXT(dc), j++ ) w[j] = dc;
3409: for ( i = 0, tvl = dvl, v = VR(p);
1.98 noro 3410: tvl && tvl->v != v; tvl = NEXT(tvl), i++ );
1.61 noro 3411: if ( !tvl ) {
3412: for ( j = k-1, s = 0, MKV(v,x); j >= 0; j-- ) {
3413: t = ptond(vl,dvl,COEF(w[j]));
3414: pwrp(vl,x,DEG(w[j]),&p);
3415: nd_mul_c_p(CO,t,p); s = nd_add(0,s,t);
3416: }
3417: return s;
3418: } else {
3419: NEWNM(m0); d = DL(m0);
3420: for ( j = k-1, s = 0; j >= 0; j-- ) {
3421: ndl_zero(d); e = QTOS(DEG(w[j])); PUT_EXP(d,i,e);
3422: TD(d) = MUL_WEIGHT(e,i);
3423: if ( nd_blockmask) ndl_weight_mask(d);
3424: t = ptond(vl,dvl,COEF(w[j]));
3425: for ( m = BDY(t); m; m = NEXT(m) )
3426: ndl_addto(DL(m),d);
3427: SG(t) += TD(d);
3428: s = nd_add(0,s,t);
3429: }
3430: FREENM(m0);
3431: return s;
3432: }
3433: }
3434: }
3435:
3436: P ndvtop(int mod,VL vl,VL dvl,NDV p)
3437: {
3438: VL tvl;
3439: int len,n,j,i,e;
3440: NMV m;
3441: Q q;
3442: P c;
3443: UINT *d;
3444: P s,r,u,t,w;
1.71 noro 3445: GFS gfs;
1.23 noro 3446:
1.34 noro 3447: if ( !p ) return 0;
1.61 noro 3448: else {
3449: len = LEN(p);
3450: n = NV(p);
3451: m = (NMV)(((char *)BDY(p))+nmv_adv*(len-1));
3452: for ( j = len-1, s = 0; j >= 0; j--, NMV_PREV(m) ) {
1.71 noro 3453: if ( mod == -1 ) {
3454: e = IFTOF(CM(m)); MKGFS(e,gfs); c = (P)gfs;
3455: } else if ( mod ) {
1.61 noro 3456: STOQ(CM(m),q); c = (P)q;
3457: } else
1.113 noro 3458: c = CP(m);
1.61 noro 3459: d = DL(m);
3460: for ( i = 0, t = c, tvl = dvl; i < n; tvl = NEXT(tvl), i++ ) {
3461: MKV(tvl->v,r); e = GET_EXP(d,i); STOQ(e,q);
3462: pwrp(vl,r,q,&u); mulp(vl,t,u,&w); t = w;
3463: }
3464: addp(vl,s,t,&u); s = u;
3465: }
3466: return s;
1.23 noro 3467: }
1.3 noro 3468: }
3469:
1.61 noro 3470: NDV ndtondv(int mod,ND p)
1.11 noro 3471: {
3472: NDV d;
1.61 noro 3473: NMV m,m0;
3474: NM t;
3475: int i,len;
1.11 noro 3476:
1.34 noro 3477: if ( !p ) return 0;
1.61 noro 3478: len = LEN(p);
1.74 noro 3479: if ( mod )
3480: m0 = m = (NMV)GC_malloc_atomic_ignore_off_page(len*nmv_adv);
3481: else
3482: m0 = m = MALLOC(len*nmv_adv);
1.103 noro 3483: #if 0
1.74 noro 3484: ndv_alloc += nmv_adv*len;
1.103 noro 3485: #endif
1.61 noro 3486: for ( t = BDY(p), i = 0; t; t = NEXT(t), i++, NMV_ADV(m) ) {
3487: ndl_copy(DL(t),DL(m));
1.113 noro 3488: CQ(m) = CQ(t);
1.11 noro 3489: }
1.61 noro 3490: MKNDV(NV(p),m0,len,d);
1.14 noro 3491: SG(d) = SG(p);
1.11 noro 3492: return d;
3493: }
3494:
1.61 noro 3495: ND ndvtond(int mod,NDV p)
1.11 noro 3496: {
1.61 noro 3497: ND d;
3498: NM m,m0;
1.11 noro 3499: NMV t;
1.61 noro 3500: int i,len;
1.11 noro 3501:
1.34 noro 3502: if ( !p ) return 0;
1.11 noro 3503: m0 = 0;
1.61 noro 3504: len = p->len;
3505: for ( t = BDY(p), i = 0; i < len; NMV_ADV(t), i++ ) {
3506: NEXTNM(m0,m);
3507: ndl_copy(DL(t),DL(m));
1.113 noro 3508: CQ(m) = CQ(t);
1.11 noro 3509: }
3510: NEXT(m) = 0;
1.61 noro 3511: MKND(NV(p),m0,len,d);
1.14 noro 3512: SG(d) = SG(p);
1.11 noro 3513: return d;
3514: }
3515:
1.3 noro 3516: void ndv_print(NDV p)
3517: {
3518: NMV m;
3519: int i,len;
3520:
1.34 noro 3521: if ( !p ) printf("0\n");
1.3 noro 3522: else {
1.14 noro 3523: len = LEN(p);
1.3 noro 3524: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
1.71 noro 3525: if ( CM(m) & 0x80000000 ) printf("+@_%d*",IFTOF(CM(m)));
3526: else printf("+%d*",CM(m));
1.16 noro 3527: ndl_print(DL(m));
3528: }
3529: printf("\n");
3530: }
3531: }
3532:
1.113 noro 3533: void ndv_print_q(NDV p)
1.16 noro 3534: {
3535: NMV m;
3536: int i,len;
3537:
1.34 noro 3538: if ( !p ) printf("0\n");
1.16 noro 3539: else {
3540: len = LEN(p);
3541: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
3542: printf("+");
1.113 noro 3543: printexpr(CO,(Obj)CQ(m));
1.16 noro 3544: printf("*");
1.14 noro 3545: ndl_print(DL(m));
1.3 noro 3546: }
3547: printf("\n");
3548: }
1.25 noro 3549: }
3550:
1.61 noro 3551: NODE ndv_reducebase(NODE x)
1.27 noro 3552: {
3553: int len,i,j;
3554: NDV *w;
3555: NODE t,t0;
3556:
3557: len = length(x);
3558: w = (NDV *)ALLOCA(len*sizeof(NDV));
3559: for ( i = 0, t = x; i < len; i++, t = NEXT(t) ) w[i] = BDY(t);
3560: for ( i = 0; i < len; i++ ) {
3561: for ( j = 0; j < i; j++ ) {
3562: if ( w[i] && w[j] )
3563: if ( ndl_reducible(HDL(w[i]),HDL(w[j])) ) w[i] = 0;
3564: else if ( ndl_reducible(HDL(w[j]),HDL(w[i])) ) w[j] = 0;
3565: }
3566: }
3567: for ( i = len-1, t0 = 0; i >= 0; i-- ) {
3568: if ( w[i] ) { NEXTNODE(t0,t); BDY(t) = (pointer)w[i]; }
3569: }
3570: NEXT(t) = 0; x = t0;
3571: return x;
1.11 noro 3572: }
1.32 noro 3573:
1.43 noro 3574: /* XXX incomplete */
3575:
1.32 noro 3576: void nd_init_ord(struct order_spec *ord)
3577: {
1.43 noro 3578: switch ( ord->id ) {
1.32 noro 3579: case 0:
1.43 noro 3580: switch ( ord->ord.simple ) {
3581: case 0:
3582: nd_dcomp = 1;
3583: nd_isrlex = 1;
3584: break;
3585: case 1:
3586: nd_dcomp = 1;
3587: nd_isrlex = 0;
3588: break;
3589: case 2:
3590: nd_dcomp = 0;
3591: nd_isrlex = 0;
1.45 noro 3592: ndl_compare_function = ndl_lex_compare;
1.58 noro 3593: break;
3594: case 11:
3595: /* XXX */
3596: nd_dcomp = 0;
3597: nd_isrlex = 1;
3598: ndl_compare_function = ndl_ww_lex_compare;
1.43 noro 3599: break;
3600: default:
3601: error("nd_gr : unsupported order");
3602: }
1.32 noro 3603: break;
3604: case 1:
1.96 noro 3605: /* block order */
1.43 noro 3606: /* XXX */
3607: nd_dcomp = -1;
1.32 noro 3608: nd_isrlex = 0;
1.45 noro 3609: ndl_compare_function = ndl_block_compare;
1.34 noro 3610: break;
1.43 noro 3611: case 2:
1.96 noro 3612: /* matrix order */
3613: /* XXX */
3614: nd_dcomp = -1;
3615: nd_isrlex = 0;
3616: nd_matrix_len = ord->ord.matrix.row;
3617: nd_matrix = ord->ord.matrix.matrix;
3618: ndl_compare_function = ndl_matrix_compare;
3619: break;
3620: case 3:
1.97 noro 3621: /* composite order */
3622: nd_dcomp = -1;
3623: nd_isrlex = 0;
3624: nd_worb_len = ord->ord.composite.length;
3625: nd_worb = ord->ord.composite.w_or_b;
3626: ndl_compare_function = ndl_composite_compare;
1.32 noro 3627: break;
3628: }
1.41 noro 3629: nd_ord = ord;
1.32 noro 3630: }
3631:
1.43 noro 3632: BlockMask nd_create_blockmask(struct order_spec *ord)
3633: {
3634: int n,i,j,s,l;
1.61 noro 3635: UINT *t;
1.43 noro 3636: BlockMask bm;
3637:
1.96 noro 3638: /* we only create mask table for block order */
3639: if ( ord->id != 1 )
1.43 noro 3640: return 0;
3641: n = ord->ord.block.length;
3642: bm = (BlockMask)MALLOC(sizeof(struct oBlockMask));
3643: bm->n = n;
3644: bm->order_pair = ord->ord.block.order_pair;
1.61 noro 3645: bm->mask = (UINT **)MALLOC(n*sizeof(UINT *));
1.43 noro 3646: for ( i = 0, s = 0; i < n; i++ ) {
1.61 noro 3647: bm->mask[i] = t = (UINT *)MALLOC_ATOMIC(nd_wpd*sizeof(UINT));
1.43 noro 3648: for ( j = 0; j < nd_wpd; j++ ) t[j] = 0;
3649: l = bm->order_pair[i].length;
3650: for ( j = 0; j < l; j++, s++ ) PUT_EXP(t,s,nd_mask0);
3651: }
3652: return bm;
1.57 noro 3653: }
3654:
3655: EPOS nd_create_epos(struct order_spec *ord)
3656: {
3657: int i,j,l,s,ord_l,ord_o;
3658: EPOS epos;
3659: struct order_pair *op;
3660:
3661: epos = (EPOS)MALLOC_ATOMIC(nd_nvar*sizeof(struct oEPOS));
3662: switch ( ord->id ) {
3663: case 0:
3664: if ( nd_isrlex ) {
3665: for ( i = 0; i < nd_nvar; i++ ) {
3666: epos[i].i = nd_exporigin + (nd_nvar-1-i)/nd_epw;
3667: epos[i].s = (nd_epw-((nd_nvar-1-i)%nd_epw)-1)*nd_bpe;
3668: }
3669: } else {
3670: for ( i = 0; i < nd_nvar; i++ ) {
3671: epos[i].i = nd_exporigin + i/nd_epw;
3672: epos[i].s = (nd_epw-(i%nd_epw)-1)*nd_bpe;
3673: }
3674: }
3675: break;
3676: case 1:
3677: /* block order */
3678: l = ord->ord.block.length;
3679: op = ord->ord.block.order_pair;
3680: for ( j = 0, s = 0; j < l; j++ ) {
3681: ord_o = op[j].order;
3682: ord_l = op[j].length;
3683: if ( !ord_o )
3684: for ( i = 0; i < ord_l; i++ ) {
3685: epos[s+i].i = nd_exporigin + (s+ord_l-i-1)/nd_epw;
3686: epos[s+i].s = (nd_epw-((s+ord_l-i-1)%nd_epw)-1)*nd_bpe;
3687: }
3688: else
3689: for ( i = 0; i < ord_l; i++ ) {
3690: epos[s+i].i = nd_exporigin + (s+i)/nd_epw;
3691: epos[s+i].s = (nd_epw-((s+i)%nd_epw)-1)*nd_bpe;
3692: }
3693: s += ord_l;
3694: }
3695: break;
3696: case 2:
1.96 noro 3697: /* matrix order */
3698: case 3:
3699: /* composite order */
3700: for ( i = 0; i < nd_nvar; i++ ) {
3701: epos[i].i = nd_exporigin + i/nd_epw;
3702: epos[i].s = (nd_epw-(i%nd_epw)-1)*nd_bpe;
3703: }
3704: break;
1.57 noro 3705: }
3706: return epos;
1.43 noro 3707: }
1.59 noro 3708:
3709: /* external interface */
3710:
3711: void nd_nf_p(P f,LIST g,LIST v,int m,struct order_spec *ord,P *rp)
3712: {
1.61 noro 3713: NODE t,in0,in;
1.59 noro 3714: ND nd,nf;
1.61 noro 3715: NDV ndv;
3716: VL vv,tv;
3717: int stat,nvar,max,e;
1.113 noro 3718: union oNDC dn;
1.59 noro 3719:
3720: pltovl(v,&vv);
1.74 noro 3721: for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++ );
1.61 noro 3722:
3723: /* get the degree bound */
3724: for ( t = BDY(g), max = 0; t; t = NEXT(t) )
3725: for ( tv = vv; tv; tv = NEXT(tv) ) {
3726: e = getdeg(tv->v,(P)BDY(t));
3727: max = MAX(e,max);
3728: }
3729: for ( tv = vv; tv; tv = NEXT(tv) ) {
3730: e = getdeg(tv->v,f);
3731: max = MAX(e,max);
3732: }
3733:
1.59 noro 3734: nd_init_ord(ord);
1.61 noro 3735: nd_setup_parameters(nvar,max);
3736:
3737: /* conversion to ndv */
3738: for ( in0 = 0, t = BDY(g); t; t = NEXT(t) ) {
3739: NEXTNODE(in0,in);
3740: BDY(in) = (pointer)ptondv(CO,vv,(P)BDY(t));
1.69 noro 3741: if ( m ) ndv_mod(m,(NDV)BDY(in));
1.61 noro 3742: }
3743: NEXTNODE(in0,in);
3744: BDY(in) = (pointer)ptondv(CO,vv,f);
1.69 noro 3745: if ( m ) ndv_mod(m,(NDV)BDY(in));
1.61 noro 3746: NEXT(in) = 0;
3747:
3748: ndv_setup(m,0,in0);
1.59 noro 3749: nd_psn--;
3750: nd_scale=2;
3751: while ( 1 ) {
3752: nd = (pointer)ndvtond(m,nd_ps[nd_psn]);
1.69 noro 3753: stat = nd_nf(m,nd,nd_ps,1,0,&nf);
1.59 noro 3754: if ( !stat ) {
3755: nd_psn++;
1.103 noro 3756: nd_reconstruct(0,0);
1.59 noro 3757: nd_psn--;
3758: } else
3759: break;
3760: }
1.61 noro 3761: *rp = ndvtop(m,CO,vv,ndtondv(m,nf));
1.63 noro 3762: }
3763:
3764: int nd_to_vect(int mod,UINT *s0,int n,ND d,UINT *r)
3765: {
3766: NM m;
3767: UINT *t,*s;
3768: int i;
3769:
3770: for ( i = 0; i < n; i++ ) r[i] = 0;
3771: for ( i = 0, s = s0, m = BDY(d); m; m = NEXT(m) ) {
3772: t = DL(m);
3773: for ( ; !ndl_equal(t,s); s += nd_wpd, i++ );
3774: r[i] = CM(m);
3775: }
3776: for ( i = 0; !r[i]; i++ );
3777: return i;
3778: }
3779:
1.113 noro 3780: int nd_to_vect_q(UINT *s0,int n,ND d,Q *r)
1.74 noro 3781: {
1.107 noro 3782: NM m;
1.74 noro 3783: UINT *t,*s;
1.107 noro 3784: int i;
1.74 noro 3785:
3786: for ( i = 0; i < n; i++ ) r[i] = 0;
1.107 noro 3787: for ( i = 0, s = s0, m = BDY(d); m; m = NEXT(m) ) {
1.74 noro 3788: t = DL(m);
3789: for ( ; !ndl_equal(t,s); s += nd_wpd, i++ );
1.113 noro 3790: r[i] = CQ(m);
1.74 noro 3791: }
3792: for ( i = 0; !r[i]; i++ );
3793: return i;
3794: }
3795:
1.67 noro 3796: IndArray nm_ind_pair_to_vect_compress(int mod,UINT *s0,int n,NM_ind_pair pair)
1.64 noro 3797: {
3798: NM m;
3799: NMV mr;
3800: UINT *d,*t,*s;
3801: NDV p;
1.67 noro 3802: unsigned char *ivc;
3803: unsigned short *ivs;
1.81 noro 3804: UINT *v,*ivi,*s0v;
1.67 noro 3805: int i,j,len,prev,diff,cdiff;
3806: IndArray r;
1.64 noro 3807:
3808: m = pair->mul;
3809: d = DL(m);
3810: p = nd_ps[pair->index];
3811: len = LEN(p);
3812: t = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
1.67 noro 3813: v = (unsigned int *)ALLOCA(len*sizeof(unsigned int));
1.64 noro 3814: for ( i = j = 0, s = s0, mr = BDY(p); j < len; j++, NMV_ADV(mr) ) {
3815: ndl_add(d,DL(mr),t);
3816: for ( ; !ndl_equal(t,s); s += nd_wpd, i++ );
1.67 noro 3817: v[j] = i;
3818: }
1.81 noro 3819: r = (IndArray)MALLOC(sizeof(struct oIndArray));
1.67 noro 3820: r->head = v[0];
3821: diff = 0;
3822: for ( i = 1; i < len; i++ ) {
3823: cdiff = v[i]-v[i-1]; diff = MAX(cdiff,diff);
3824: }
3825: if ( diff < 256 ) {
3826: r->width = 1;
3827: ivc = (unsigned char *)MALLOC_ATOMIC(len*sizeof(unsigned char));
3828: r->index.c = ivc;
3829: for ( i = 1, ivc[0] = 0; i < len; i++ ) ivc[i] = v[i]-v[i-1];
3830: } else if ( diff < 65536 ) {
3831: r->width = 2;
3832: ivs = (unsigned short *)MALLOC_ATOMIC(len*sizeof(unsigned short));
3833: r->index.s = ivs;
3834: for ( i = 1, ivs[0] = 0; i < len; i++ ) ivs[i] = v[i]-v[i-1];
3835: } else {
3836: r->width = 4;
3837: ivi = (unsigned int *)MALLOC_ATOMIC(len*sizeof(unsigned int));
3838: r->index.i = ivi;
3839: for ( i = 1, ivi[0] = 0; i < len; i++ ) ivi[i] = v[i]-v[i-1];
1.64 noro 3840: }
1.67 noro 3841: return r;
1.64 noro 3842: }
3843:
1.65 noro 3844:
1.113 noro 3845: int ndv_reduce_vect_q(Q *svect,int col,IndArray *imat,NM_ind_pair *rp0,int nred)
1.107 noro 3846: {
3847: int i,j,k,len,pos,prev;
1.113 noro 3848: Q cs,mcs,c1,c2,cr,gcd,t;
1.107 noro 3849: IndArray ivect;
3850: unsigned char *ivc;
3851: unsigned short *ivs;
3852: unsigned int *ivi;
3853: NDV redv;
3854: NMV mr;
3855: NODE rp;
3856: int maxrs;
3857:
3858: maxrs = 0;
3859: for ( i = 0; i < nred; i++ ) {
3860: ivect = imat[i];
3861: k = ivect->head;
3862: if ( svect[k] ) {
3863: maxrs = MAX(maxrs,rp0[i]->sugar);
3864: redv = nd_ps[rp0[i]->index];
3865: len = LEN(redv); mr = BDY(redv);
1.113 noro 3866: igcd_cofactor(svect[k],CQ(mr),&gcd,&cs,&cr);
3867: chsgnq(cs,&mcs);
3868: if ( !UNIQ(cr) ) {
3869: for ( j = 0; j < col; j++ ) {
3870: mulq(svect[j],cr,&c1); svect[j] = c1;
3871: }
3872: }
1.107 noro 3873: svect[k] = 0; prev = k;
3874: switch ( ivect->width ) {
3875: case 1:
3876: ivc = ivect->index.c;
3877: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3878: pos = prev+ivc[j]; prev = pos;
1.113 noro 3879: mulq(CQ(mr),mcs,&c2); addq(svect[pos],c2,&t); svect[pos] = t;
1.107 noro 3880: }
3881: break;
3882: case 2:
3883: ivs = ivect->index.s;
3884: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3885: pos = prev+ivs[j]; prev = pos;
1.113 noro 3886: mulq(CQ(mr),mcs,&c2); addq(svect[pos],c2,&t); svect[pos] = t;
1.107 noro 3887: }
3888: break;
3889: case 4:
3890: ivi = ivect->index.i;
3891: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3892: pos = prev+ivi[j]; prev = pos;
1.113 noro 3893: mulq(CQ(mr),mcs,&c2); addq(svect[pos],c2,&t); svect[pos] = t;
1.107 noro 3894: }
3895: break;
3896: }
3897: }
3898: }
3899: return maxrs;
3900: }
3901:
1.76 noro 3902: int ndv_reduce_vect(int m,UINT *svect,int col,IndArray *imat,NM_ind_pair *rp0,int nred)
1.65 noro 3903: {
1.67 noro 3904: int i,j,k,len,pos,prev;
1.66 noro 3905: UINT c,c1,c2,c3,up,lo,dmy;
1.67 noro 3906: IndArray ivect;
3907: unsigned char *ivc;
3908: unsigned short *ivs;
3909: unsigned int *ivi;
1.65 noro 3910: NDV redv;
1.67 noro 3911: NMV mr;
1.65 noro 3912: NODE rp;
1.76 noro 3913: int maxrs;
1.65 noro 3914:
1.76 noro 3915: maxrs = 0;
1.74 noro 3916: for ( i = 0; i < nred; i++ ) {
1.65 noro 3917: ivect = imat[i];
1.67 noro 3918: k = ivect->head; svect[k] %= m;
1.65 noro 3919: if ( c = svect[k] ) {
1.76 noro 3920: maxrs = MAX(maxrs,rp0[i]->sugar);
1.74 noro 3921: c = m-c; redv = nd_ps[rp0[i]->index];
1.67 noro 3922: len = LEN(redv); mr = BDY(redv);
3923: svect[k] = 0; prev = k;
3924: switch ( ivect->width ) {
3925: case 1:
3926: ivc = ivect->index.c;
3927: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3928: pos = prev+ivc[j]; c1 = CM(mr); c2 = svect[pos];
3929: prev = pos;
3930: DMA(c1,c,c2,up,lo);
3931: if ( up ) { DSAB(m,up,lo,dmy,c3); svect[pos] = c3;
3932: } else svect[pos] = lo;
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]; c1 = CM(mr); c2 = svect[pos];
3939: prev = pos;
3940: DMA(c1,c,c2,up,lo);
3941: if ( up ) { DSAB(m,up,lo,dmy,c3); svect[pos] = c3;
3942: } else svect[pos] = lo;
3943: }
3944: break;
3945: case 4:
3946: ivi = ivect->index.i;
3947: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3948: pos = prev+ivi[j]; c1 = CM(mr); c2 = svect[pos];
3949: prev = pos;
3950: DMA(c1,c,c2,up,lo);
3951: if ( up ) { DSAB(m,up,lo,dmy,c3); svect[pos] = c3;
3952: } else svect[pos] = lo;
3953: }
3954: break;
1.65 noro 3955: }
3956: }
3957: }
1.66 noro 3958: for ( i = 0; i < col; i++ )
3959: if ( svect[i] >= (UINT)m ) svect[i] %= m;
1.76 noro 3960: return maxrs;
1.65 noro 3961: }
3962:
1.76 noro 3963: int ndv_reduce_vect_sf(int m,UINT *svect,int col,IndArray *imat,NM_ind_pair *rp0,int nred)
1.72 noro 3964: {
3965: int i,j,k,len,pos,prev;
3966: UINT c,c1,c2,c3,up,lo,dmy;
3967: IndArray ivect;
3968: unsigned char *ivc;
3969: unsigned short *ivs;
3970: unsigned int *ivi;
3971: NDV redv;
3972: NMV mr;
3973: NODE rp;
1.76 noro 3974: int maxrs;
1.72 noro 3975:
1.76 noro 3976: maxrs = 0;
1.74 noro 3977: for ( i = 0; i < nred; i++ ) {
1.72 noro 3978: ivect = imat[i];
3979: k = ivect->head; svect[k] %= m;
3980: if ( c = svect[k] ) {
1.76 noro 3981: maxrs = MAX(maxrs,rp0[i]->sugar);
1.74 noro 3982: c = _chsgnsf(c); redv = nd_ps[rp0[i]->index];
1.72 noro 3983: len = LEN(redv); mr = BDY(redv);
3984: svect[k] = 0; prev = k;
3985: switch ( ivect->width ) {
3986: case 1:
3987: ivc = ivect->index.c;
3988: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3989: pos = prev+ivc[j]; prev = pos;
3990: svect[pos] = _addsf(_mulsf(CM(mr),c),svect[pos]);
3991: }
3992: break;
3993: case 2:
3994: ivs = ivect->index.s;
3995: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3996: pos = prev+ivs[j]; prev = pos;
3997: svect[pos] = _addsf(_mulsf(CM(mr),c),svect[pos]);
3998: }
3999: break;
4000: case 4:
4001: ivi = ivect->index.i;
4002: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
4003: pos = prev+ivi[j]; prev = pos;
4004: svect[pos] = _addsf(_mulsf(CM(mr),c),svect[pos]);
4005: }
4006: break;
4007: }
4008: }
4009: }
1.76 noro 4010: return maxrs;
1.72 noro 4011: }
4012:
1.65 noro 4013: NDV vect_to_ndv(UINT *vect,int spcol,int col,int *rhead,UINT *s0vect)
4014: {
4015: int j,k,len;
4016: UINT *p;
4017: UINT c;
4018: NDV r;
4019: NMV mr0,mr;
4020:
4021: for ( j = 0, len = 0; j < spcol; j++ ) if ( vect[j] ) len++;
4022: if ( !len ) return 0;
4023: else {
1.74 noro 4024: mr0 = (NMV)GC_malloc_atomic_ignore_off_page(nmv_adv*len);
1.103 noro 4025: #if 0
1.74 noro 4026: ndv_alloc += nmv_adv*len;
1.103 noro 4027: #endif
1.65 noro 4028: mr = mr0;
4029: p = s0vect;
4030: for ( j = k = 0; j < col; j++, p += nd_wpd )
4031: if ( !rhead[j] ) {
4032: if ( c = vect[k++] ) {
4033: ndl_copy(p,DL(mr)); CM(mr) = c; NMV_ADV(mr);
4034: }
4035: }
4036: MKNDV(nd_nvar,mr0,len,r);
4037: return r;
4038: }
4039: }
4040:
1.113 noro 4041: NDV vect_to_ndv_q(Q *vect,int spcol,int col,int *rhead,UINT *s0vect)
1.107 noro 4042: {
4043: int j,k,len;
4044: UINT *p;
1.113 noro 4045: Q c;
1.107 noro 4046: NDV r;
4047: NMV mr0,mr;
4048:
4049: for ( j = 0, len = 0; j < spcol; j++ ) if ( vect[j] ) len++;
4050: if ( !len ) return 0;
4051: else {
1.109 noro 4052: mr0 = (NMV)GC_malloc(nmv_adv*len);
1.107 noro 4053: #if 0
4054: ndv_alloc += nmv_adv*len;
4055: #endif
4056: mr = mr0;
4057: p = s0vect;
4058: for ( j = k = 0; j < col; j++, p += nd_wpd )
4059: if ( !rhead[j] ) {
4060: if ( c = vect[k++] ) {
1.113 noro 4061: if ( DN(c) )
4062: error("afo");
4063: ndl_copy(p,DL(mr)); CQ(mr) = c; NMV_ADV(mr);
1.107 noro 4064: }
4065: }
4066: MKNDV(nd_nvar,mr0,len,r);
4067: return r;
4068: }
4069: }
4070:
1.74 noro 4071: int nd_sp_f4(int m,ND_pairs l,PGeoBucket bucket)
1.65 noro 4072: {
4073: ND_pairs t;
4074: NODE sp0,sp;
4075: int stat;
4076: ND spol;
4077:
4078: for ( t = l; t; t = NEXT(t) ) {
4079: stat = nd_sp(m,0,t,&spol);
4080: if ( !stat ) return 0;
4081: if ( spol ) {
4082: add_pbucket_symbolic(bucket,spol);
4083: }
4084: }
1.68 noro 4085: return 1;
1.65 noro 4086: }
4087:
4088: int nd_symbolic_preproc(PGeoBucket bucket,UINT **s0vect,NODE *r)
4089: {
4090: NODE rp0,rp;
4091: NM mul,head,s0,s;
1.76 noro 4092: int index,col,i,sugar;
1.65 noro 4093: RHist h;
4094: UINT *s0v,*p;
4095: NM_ind_pair pair;
4096: ND red;
4097:
4098: s0 = 0; rp0 = 0; col = 0;
4099: while ( 1 ) {
4100: head = remove_head_pbucket_symbolic(bucket);
4101: if ( !head ) break;
4102: if ( !s0 ) s0 = head;
4103: else NEXT(s) = head;
4104: s = head;
4105: index = ndl_find_reducer(DL(head));
4106: if ( index >= 0 ) {
4107: h = nd_psh[index];
4108: NEWNM(mul);
4109: ndl_sub(DL(head),DL(h),DL(mul));
4110: if ( ndl_check_bound2(index,DL(mul)) ) return 0;
1.76 noro 4111: sugar = TD(DL(mul))+SG(nd_ps[index]);
4112: MKNM_ind_pair(pair,mul,index,sugar);
1.65 noro 4113: red = ndv_mul_nm_symbolic(mul,nd_ps[index]);
4114: add_pbucket_symbolic(bucket,nd_remove_head(red));
4115: NEXTNODE(rp0,rp); BDY(rp) = (pointer)pair;
4116: }
4117: col++;
4118: }
1.72 noro 4119: if ( rp0 ) NEXT(rp) = 0;
4120: NEXT(s) = 0;
1.65 noro 4121: s0v = (UINT *)MALLOC_ATOMIC(col*nd_wpd*sizeof(UINT));
4122: for ( i = 0, p = s0v, s = s0; i < col;
4123: i++, p += nd_wpd, s = NEXT(s) ) ndl_copy(DL(s),p);
4124: *s0vect = s0v;
4125: *r = rp0;
4126: return col;
4127: }
4128:
1.69 noro 4129: NODE nd_f4(int m)
4130: {
4131: int i,nh,stat,index;
4132: NODE r,g;
4133: ND_pairs d,l,t;
4134: ND spol,red;
4135: NDV nf,redv;
4136: NM s0,s;
1.84 noro 4137: NODE rp0,srp0,nflist;
1.69 noro 4138: int nsp,nred,col,rank,len,k,j,a;
4139: UINT c;
1.74 noro 4140: UINT **spmat;
1.69 noro 4141: UINT *s0vect,*svect,*p,*v;
4142: int *colstat;
4143: IndArray *imat;
4144: int *rhead;
4145: int spcol,sprow;
4146: int sugar;
4147: PGeoBucket bucket;
4148: struct oEGT eg0,eg1,eg_f4;
4149:
1.103 noro 4150: #if 0
1.74 noro 4151: ndv_alloc = 0;
1.103 noro 4152: #endif
1.69 noro 4153: g = 0; d = 0;
4154: for ( i = 0; i < nd_psn; i++ ) {
4155: d = update_pairs(d,g,i);
4156: g = update_base(g,i);
4157: }
4158: while ( d ) {
4159: get_eg(&eg0);
4160: l = nd_minsugarp(d,&d);
4161: sugar = SG(l);
4162: bucket = create_pbucket();
1.74 noro 4163: stat = nd_sp_f4(m,l,bucket);
1.69 noro 4164: if ( !stat ) {
4165: for ( t = l; NEXT(t); t = NEXT(t) );
4166: NEXT(t) = d; d = l;
1.103 noro 4167: d = nd_reconstruct(0,d);
1.69 noro 4168: continue;
4169: }
1.86 noro 4170: if ( bucket->m < 0 ) continue;
1.69 noro 4171: col = nd_symbolic_preproc(bucket,&s0vect,&rp0);
4172: if ( !col ) {
4173: for ( t = l; NEXT(t); t = NEXT(t) );
4174: NEXT(t) = d; d = l;
1.103 noro 4175: d = nd_reconstruct(0,d);
1.69 noro 4176: continue;
4177: }
4178: get_eg(&eg1); init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg1);
1.74 noro 4179: if ( DP_Print )
4180: fprintf(asir_out,"sugar=%d,symb=%fsec,",
4181: sugar,eg_f4.exectime+eg_f4.gctime);
4182: if ( 1 )
4183: nflist = nd_f4_red(m,l,s0vect,col,rp0);
4184: else
4185: nflist = nd_f4_red_dist(m,l,s0vect,col,rp0);
1.69 noro 4186: /* adding new bases */
1.74 noro 4187: for ( r = nflist; r; r = NEXT(r) ) {
4188: nf = (NDV)BDY(r);
1.69 noro 4189: ndv_removecont(m,nf);
1.77 noro 4190: nh = ndv_newps(m,nf,0);
1.69 noro 4191: d = update_pairs(d,g,nh);
4192: g = update_base(g,nh);
4193: }
4194: }
4195: for ( r = g; r; r = NEXT(r) ) BDY(r) = (pointer)nd_ps[(int)BDY(r)];
1.103 noro 4196: #if 0
1.74 noro 4197: fprintf(asir_out,"ndv_alloc=%d\n",ndv_alloc);
1.103 noro 4198: #endif
1.69 noro 4199: return g;
4200: }
1.74 noro 4201:
4202: NODE nd_f4_red(int m,ND_pairs sp0,UINT *s0vect,int col,NODE rp0)
1.63 noro 4203: {
1.67 noro 4204: IndArray *imat;
1.106 noro 4205: int nsp,nred,i;
1.65 noro 4206: int *rhead;
1.106 noro 4207: NODE r0,rp;
1.74 noro 4208: ND_pairs sp;
4209: NM_ind_pair *rvect;
1.63 noro 4210:
1.74 noro 4211: for ( sp = sp0, nsp = 0; sp; sp = NEXT(sp), nsp++ );
1.106 noro 4212: nred = length(rp0);
1.74 noro 4213: imat = (IndArray *)ALLOCA(nred*sizeof(IndArray));
4214: rhead = (int *)ALLOCA(col*sizeof(int));
4215: for ( i = 0; i < col; i++ ) rhead[i] = 0;
1.63 noro 4216:
1.74 noro 4217: /* construction of index arrays */
4218: rvect = (NM_ind_pair *)ALLOCA(nred*sizeof(NM_ind_pair));
4219: for ( rp = rp0, i = 0; rp; i++, rp = NEXT(rp) ) {
4220: rvect[i] = (NM_ind_pair)BDY(rp);
4221: imat[i] = nm_ind_pair_to_vect_compress(m,s0vect,col,rvect[i]);
4222: rhead[imat[i]->head] = 1;
4223: }
1.107 noro 4224: if ( m )
4225: r0 = nd_f4_red_main(m,sp0,nsp,s0vect,col,rvect,rhead,imat,nred);
4226: else
1.113 noro 4227: r0 = nd_f4_red_q_main(sp0,nsp,s0vect,col,rvect,rhead,imat,nred);
1.106 noro 4228: return r0;
4229: }
1.74 noro 4230:
1.106 noro 4231: NODE nd_f4_red_main(int m,ND_pairs sp0,int nsp,UINT *s0vect,int col,
4232: NM_ind_pair *rvect,int *rhead,IndArray *imat,int nred)
4233: {
4234: int spcol,sprow,a;
4235: int i,j,k,l,rank;
4236: NODE r0,r;
4237: ND_pairs sp;
4238: ND spol;
4239: int **spmat;
4240: UINT *svect,*v;
4241: int *colstat;
4242: struct oEGT eg0,eg1,eg2,eg_f4,eg_f4_1,eg_f4_2;
4243: int maxrs;
4244: int *spsugar;
4245:
4246: spcol = col-nred;
4247: get_eg(&eg0);
1.74 noro 4248: /* elimination (1st step) */
4249: spmat = (int **)ALLOCA(nsp*sizeof(UINT *));
4250: svect = (UINT *)ALLOCA(col*sizeof(UINT));
1.76 noro 4251: spsugar = (int *)ALLOCA(nsp*sizeof(UINT));
1.74 noro 4252: for ( a = sprow = 0, sp = sp0; a < nsp; a++, sp = NEXT(sp) ) {
4253: nd_sp(m,0,sp,&spol);
1.75 noro 4254: if ( !spol ) continue;
1.74 noro 4255: nd_to_vect(m,s0vect,col,spol,svect);
1.76 noro 4256: if ( m == -1 )
4257: maxrs = ndv_reduce_vect_sf(m,svect,col,imat,rvect,nred);
4258: else
4259: maxrs = ndv_reduce_vect(m,svect,col,imat,rvect,nred);
1.74 noro 4260: for ( i = 0; i < col; i++ ) if ( svect[i] ) break;
4261: if ( i < col ) {
4262: spmat[sprow] = v = (UINT *)MALLOC_ATOMIC(spcol*sizeof(UINT));
4263: for ( j = k = 0; j < col; j++ )
4264: if ( !rhead[j] ) v[k++] = svect[j];
1.76 noro 4265: spsugar[sprow] = MAX(maxrs,SG(spol));
1.74 noro 4266: sprow++;
4267: }
1.76 noro 4268: nd_free(spol);
1.74 noro 4269: }
1.80 noro 4270: get_eg(&eg1); init_eg(&eg_f4_1); add_eg(&eg_f4_1,&eg0,&eg1);
4271: if ( DP_Print ) {
4272: fprintf(asir_out,"elim1=%fsec,",eg_f4_1.exectime+eg_f4_1.gctime);
4273: fflush(asir_out);
4274: }
1.74 noro 4275: /* free index arrays */
4276: for ( i = 0; i < nred; i++ ) GC_free(imat[i]->index.c);
4277:
4278: /* elimination (2nd step) */
4279: colstat = (int *)ALLOCA(spcol*sizeof(int));
4280: if ( m == -1 )
1.76 noro 4281: rank = nd_gauss_elim_sf(spmat,spsugar,sprow,spcol,m,colstat);
1.74 noro 4282: else
1.76 noro 4283: rank = nd_gauss_elim_mod(spmat,spsugar,sprow,spcol,m,colstat);
1.74 noro 4284: r0 = 0;
4285: for ( i = 0; i < rank; i++ ) {
4286: NEXTNODE(r0,r); BDY(r) =
4287: (pointer)vect_to_ndv(spmat[i],spcol,col,rhead,s0vect);
1.76 noro 4288: SG((NDV)BDY(r)) = spsugar[i];
1.74 noro 4289: GC_free(spmat[i]);
4290: }
1.109 noro 4291: if ( r0 ) NEXT(r) = 0;
1.74 noro 4292: for ( ; i < sprow; i++ ) GC_free(spmat[i]);
1.80 noro 4293: get_eg(&eg2); init_eg(&eg_f4_2); add_eg(&eg_f4_2,&eg1,&eg2);
4294: init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg2);
1.74 noro 4295: if ( DP_Print ) {
1.80 noro 4296: fprintf(asir_out,"elim2=%fsec\n",eg_f4_2.exectime+eg_f4_2.gctime);
1.74 noro 4297: fprintf(asir_out,"nsp=%d,nred=%d,spmat=(%d,%d),rank=%d ",
4298: nsp,nred,sprow,spcol,rank);
4299: fprintf(asir_out,"%fsec\n",eg_f4.exectime+eg_f4.gctime);
1.63 noro 4300: }
1.74 noro 4301: return r0;
4302: }
4303:
1.113 noro 4304: NODE nd_f4_red_q_main(ND_pairs sp0,int nsp,UINT *s0vect,int col,
1.107 noro 4305: NM_ind_pair *rvect,int *rhead,IndArray *imat,int nred)
4306: {
4307: int spcol,sprow,a;
4308: int i,j,k,l,rank;
4309: NODE r0,r;
4310: ND_pairs sp;
4311: ND spol;
1.113 noro 4312: Q **spmat;
4313: Q *svect,*v;
1.107 noro 4314: int *colstat;
4315: struct oEGT eg0,eg1,eg2,eg_f4,eg_f4_1,eg_f4_2;
4316: int maxrs;
4317: int *spsugar;
4318:
4319: spcol = col-nred;
4320: get_eg(&eg0);
4321: /* elimination (1st step) */
1.113 noro 4322: spmat = (Q **)ALLOCA(nsp*sizeof(Q *));
4323: svect = (Q *)ALLOCA(col*sizeof(Q));
4324: spsugar = (int *)ALLOCA(nsp*sizeof(Q));
1.107 noro 4325: for ( a = sprow = 0, sp = sp0; a < nsp; a++, sp = NEXT(sp) ) {
4326: nd_sp(0,0,sp,&spol);
4327: if ( !spol ) continue;
1.113 noro 4328: nd_to_vect_q(s0vect,col,spol,svect);
4329: maxrs = ndv_reduce_vect_q(svect,col,imat,rvect,nred);
1.107 noro 4330: for ( i = 0; i < col; i++ ) if ( svect[i] ) break;
4331: if ( i < col ) {
1.113 noro 4332: spmat[sprow] = v = (Q *)MALLOC(spcol*sizeof(Q));
1.107 noro 4333: for ( j = k = 0; j < col; j++ )
4334: if ( !rhead[j] ) v[k++] = svect[j];
4335: spsugar[sprow] = MAX(maxrs,SG(spol));
4336: sprow++;
4337: }
1.109 noro 4338: /* nd_free(spol); */
1.107 noro 4339: }
4340: get_eg(&eg1); init_eg(&eg_f4_1); add_eg(&eg_f4_1,&eg0,&eg1);
4341: if ( DP_Print ) {
4342: fprintf(asir_out,"elim1=%fsec,",eg_f4_1.exectime+eg_f4_1.gctime);
4343: fflush(asir_out);
4344: }
4345: /* free index arrays */
1.109 noro 4346: /* for ( i = 0; i < nred; i++ ) GC_free(imat[i]->index.c); */
1.107 noro 4347:
4348: /* elimination (2nd step) */
4349: colstat = (int *)ALLOCA(spcol*sizeof(int));
1.113 noro 4350: rank = nd_gauss_elim_q(spmat,spsugar,sprow,spcol,colstat);
1.107 noro 4351: r0 = 0;
4352: for ( i = 0; i < rank; i++ ) {
4353: NEXTNODE(r0,r); BDY(r) =
1.113 noro 4354: (pointer)vect_to_ndv_q(spmat[i],spcol,col,rhead,s0vect);
1.107 noro 4355: SG((NDV)BDY(r)) = spsugar[i];
1.109 noro 4356: /* GC_free(spmat[i]); */
1.107 noro 4357: }
1.109 noro 4358: if ( r0 ) NEXT(r) = 0;
4359:
4360: /* for ( ; i < sprow; i++ ) GC_free(spmat[i]); */
1.107 noro 4361: get_eg(&eg2); init_eg(&eg_f4_2); add_eg(&eg_f4_2,&eg1,&eg2);
4362: init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg2);
4363: if ( DP_Print ) {
4364: fprintf(asir_out,"elim2=%fsec\n",eg_f4_2.exectime+eg_f4_2.gctime);
4365: fprintf(asir_out,"nsp=%d,nred=%d,spmat=(%d,%d),rank=%d ",
4366: nsp,nred,sprow,spcol,rank);
4367: fprintf(asir_out,"%fsec\n",eg_f4.exectime+eg_f4.gctime);
4368: }
4369: return r0;
4370: }
4371:
1.74 noro 4372: FILE *nd_write,*nd_read;
4373:
4374: void nd_send_int(int a) {
4375: write_int(nd_write,&a);
4376: }
4377:
4378: void nd_send_intarray(int *p,int len) {
4379: write_intarray(nd_write,p,len);
4380: }
4381:
4382: int nd_recv_int() {
4383: int a;
4384:
4385: read_int(nd_read,&a);
4386: return a;
4387: }
4388:
4389: void nd_recv_intarray(int *p,int len) {
4390: read_intarray(nd_read,p,len);
4391: }
4392:
4393: void nd_send_ndv(NDV p) {
4394: int len,i;
4395: NMV m;
4396:
4397: if ( !p ) nd_send_int(0);
4398: else {
4399: len = LEN(p);
4400: nd_send_int(len);
4401: m = BDY(p);
4402: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
4403: nd_send_int(CM(m));
4404: nd_send_intarray(DL(m),nd_wpd);
1.68 noro 4405: }
1.74 noro 4406: }
4407: }
4408:
4409: void nd_send_nd(ND p) {
4410: int len,i;
4411: NM m;
4412:
4413: if ( !p ) nd_send_int(0);
4414: else {
4415: len = LEN(p);
4416: nd_send_int(len);
4417: m = BDY(p);
4418: for ( i = 0; i < len; i++, m = NEXT(m) ) {
4419: nd_send_int(CM(m));
4420: nd_send_intarray(DL(m),nd_wpd);
1.65 noro 4421: }
1.74 noro 4422: }
4423: }
1.65 noro 4424:
1.74 noro 4425: NDV nd_recv_ndv()
4426: {
4427: int len,i;
4428: NMV m,m0;
4429: NDV r;
1.65 noro 4430:
1.74 noro 4431: len = nd_recv_int();
4432: if ( !len ) return 0;
4433: else {
4434: m0 = m = (NMV)GC_malloc_atomic_ignore_off_page(nmv_adv*len);
1.103 noro 4435: #if 0
1.74 noro 4436: ndv_alloc += len*nmv_adv;
1.103 noro 4437: #endif
1.74 noro 4438: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
4439: CM(m) = nd_recv_int();
4440: nd_recv_intarray(DL(m),nd_wpd);
1.71 noro 4441: }
1.74 noro 4442: MKNDV(nd_nvar,m0,len,r);
4443: return r;
4444: }
4445: }
1.65 noro 4446:
1.74 noro 4447: int ox_exec_f4_red(Q proc)
4448: {
4449: Obj obj;
4450: STRING fname;
4451: NODE arg;
4452: int s;
4453: extern int ox_need_conv,ox_file_io;
4454:
4455: MKSTR(fname,"nd_exec_f4_red");
4456: arg = mknode(2,proc,fname);
4457: Pox_cmo_rpc(arg,&obj);
4458: s = get_ox_server_id(QTOS(proc));
4459: nd_write = iofp[s].out;
4460: nd_read = iofp[s].in;
4461: ox_need_conv = ox_file_io = 0;
4462: return s;
4463: }
4464:
4465: NODE nd_f4_red_dist(int m,ND_pairs sp0,UINT *s0vect,int col,NODE rp0)
4466: {
4467: int nsp,nred;
4468: int i,rank,s;
4469: NODE rp,r0,r;
4470: ND_pairs sp;
4471: NM_ind_pair pair;
4472: NMV nmv;
4473: NM nm;
4474: NDV nf;
4475: Obj proc,dmy;
4476:
4477: ox_launch_main(0,0,&proc);
4478: s = ox_exec_f4_red((Q)proc);
4479:
4480: nd_send_int(m);
4481: nd_send_int(nd_nvar);
4482: nd_send_int(nd_bpe);
4483: nd_send_int(nd_wpd);
4484: nd_send_int(nmv_adv);
4485:
1.89 noro 4486: saveobj(nd_write,dp_current_spec->obj); fflush(nd_write);
1.74 noro 4487:
4488: nd_send_int(nd_psn);
4489: for ( i = 0; i < nd_psn; i++ ) nd_send_ndv(nd_ps[i]);
4490:
4491: for ( sp = sp0, nsp = 0; sp; sp = NEXT(sp), nsp++ );
4492: nd_send_int(nsp);
4493: for ( i = 0, sp = sp0; i < nsp; i++, sp = NEXT(sp) ) {
4494: nd_send_int(sp->i1); nd_send_int(sp->i2);
4495: }
4496:
4497: nd_send_int(col); nd_send_intarray(s0vect,col*nd_wpd);
4498:
4499: nred = length(rp0); nd_send_int(nred);
4500: for ( i = 0, rp = rp0; i < nred; i++, rp = NEXT(rp) ) {
4501: pair = (NM_ind_pair)BDY(rp);
4502: nd_send_int(pair->index);
4503: nd_send_intarray(pair->mul->dl,nd_wpd);
4504: }
4505: fflush(nd_write);
4506: rank = nd_recv_int();
4507: fprintf(asir_out,"rank=%d\n",rank);
4508: r0 = 0;
4509: for ( i = 0; i < rank; i++ ) {
4510: nf = nd_recv_ndv();
4511: NEXTNODE(r0,r); BDY(r) = (pointer)nf;
4512: }
4513: Pox_shutdown(mknode(1,proc),&dmy);
4514: return r0;
4515: }
4516:
4517: /* server side */
4518:
4519: void nd_exec_f4_red_dist()
4520: {
4521: int m,i,nsp,col,s0size,nred,spcol,j,k;
4522: NM_ind_pair *rp0;
4523: NDV nf;
4524: UINT *s0vect;
4525: IndArray *imat;
4526: int *rhead;
4527: int **spmat;
4528: UINT *svect,*v;
4529: ND_pairs *sp0;
4530: int *colstat;
4531: int a,sprow,rank;
1.89 noro 4532: struct order_spec *ord;
1.74 noro 4533: Obj ordspec;
4534: ND spol;
1.76 noro 4535: int maxrs;
4536: int *spsugar;
1.74 noro 4537:
4538: nd_read = iofp[0].in;
4539: nd_write = iofp[0].out;
4540: m = nd_recv_int();
4541: nd_nvar = nd_recv_int();
4542: nd_bpe = nd_recv_int();
4543: nd_wpd = nd_recv_int();
4544: nmv_adv = nd_recv_int();
4545:
4546: loadobj(nd_read,&ordspec);
1.89 noro 4547: create_order_spec(0,ordspec,&ord);
4548: nd_init_ord(ord);
1.74 noro 4549: nd_setup_parameters(nd_nvar,0);
4550:
4551: nd_psn = nd_recv_int();
4552: nd_ps = (NDV *)MALLOC(nd_psn*sizeof(NDV));
4553: nd_bound = (UINT **)MALLOC(nd_psn*sizeof(UINT *));
4554: for ( i = 0; i < nd_psn; i++ ) {
4555: nd_ps[i] = nd_recv_ndv();
4556: nd_bound[i] = ndv_compute_bound(nd_ps[i]);
4557: }
4558:
4559: nsp = nd_recv_int();
4560: sp0 = (ND_pairs *)MALLOC(nsp*sizeof(ND_pairs));
4561: for ( i = 0; i < nsp; i++ ) {
4562: NEWND_pairs(sp0[i]);
4563: sp0[i]->i1 = nd_recv_int(); sp0[i]->i2 = nd_recv_int();
4564: ndl_lcm(HDL(nd_ps[sp0[i]->i1]),HDL(nd_ps[sp0[i]->i2]),LCM(sp0[i]));
4565: }
4566:
4567: col = nd_recv_int();
4568: s0size = col*nd_wpd;
4569: s0vect = (UINT *)MALLOC(s0size*sizeof(UINT));
4570: nd_recv_intarray(s0vect,s0size);
4571:
4572: nred = nd_recv_int();
4573: rp0 = (NM_ind_pair *)MALLOC(nred*sizeof(NM_ind_pair));
4574: for ( i = 0; i < nred; i++ ) {
4575: rp0[i] = (NM_ind_pair)MALLOC(sizeof(struct oNM_ind_pair));
4576: rp0[i]->index = nd_recv_int();
4577: rp0[i]->mul = (NM)MALLOC(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT));
4578: nd_recv_intarray(rp0[i]->mul->dl,nd_wpd);
4579: }
4580:
4581: spcol = col-nred;
4582: imat = (IndArray *)MALLOC(nred*sizeof(IndArray));
4583: rhead = (int *)MALLOC(col*sizeof(int));
4584: for ( i = 0; i < col; i++ ) rhead[i] = 0;
4585:
4586: /* construction of index arrays */
4587: for ( i = 0; i < nred; i++ ) {
4588: imat[i] = nm_ind_pair_to_vect_compress(m,s0vect,col,rp0[i]);
4589: rhead[imat[i]->head] = 1;
4590: }
4591:
4592: /* elimination (1st step) */
4593: spmat = (int **)MALLOC(nsp*sizeof(UINT *));
4594: svect = (UINT *)MALLOC(col*sizeof(UINT));
1.76 noro 4595: spsugar = (int *)ALLOCA(nsp*sizeof(UINT));
1.74 noro 4596: for ( a = sprow = 0; a < nsp; a++ ) {
4597: nd_sp(m,0,sp0[a],&spol);
1.75 noro 4598: if ( !spol ) continue;
1.74 noro 4599: nd_to_vect(m,s0vect,col,spol,svect);
1.76 noro 4600: if ( m == -1 )
4601: maxrs = ndv_reduce_vect_sf(m,svect,col,imat,rp0,nred);
4602: else
4603: maxrs = ndv_reduce_vect(m,svect,col,imat,rp0,nred);
1.74 noro 4604: for ( i = 0; i < col; i++ ) if ( svect[i] ) break;
4605: if ( i < col ) {
4606: spmat[sprow] = v = (UINT *)MALLOC(spcol*sizeof(UINT));
4607: for ( j = k = 0; j < col; j++ )
4608: if ( !rhead[j] ) v[k++] = svect[j];
1.76 noro 4609: spsugar[sprow] = MAX(maxrs,SG(spol));
1.74 noro 4610: sprow++;
4611: }
1.76 noro 4612: nd_free(spol);
1.74 noro 4613: }
4614: /* elimination (2nd step) */
4615: colstat = (int *)ALLOCA(spcol*sizeof(int));
4616: if ( m == -1 )
1.76 noro 4617: rank = nd_gauss_elim_sf(spmat,spsugar,sprow,spcol,m,colstat);
1.74 noro 4618: else
1.76 noro 4619: rank = nd_gauss_elim_mod(spmat,spsugar,sprow,spcol,m,colstat);
1.74 noro 4620: nd_send_int(rank);
4621: for ( i = 0; i < rank; i++ ) {
4622: nf = vect_to_ndv(spmat[i],spcol,col,rhead,s0vect);
4623: nd_send_ndv(nf);
1.63 noro 4624: }
1.74 noro 4625: fflush(nd_write);
1.107 noro 4626: }
4627:
1.113 noro 4628: int nd_gauss_elim_q(Q **mat0,int *sugar,int row,int col,int *colstat)
1.107 noro 4629: {
1.109 noro 4630: int mod,i,j,t,c,rank,rank0,inv;
4631: int *ci,*ri;
1.113 noro 4632: Q dn;
1.109 noro 4633: MAT m,nm;
1.108 noro 4634: int **wmat;
4635:
4636: /* XXX */
4637: mod = 99999989;
4638: wmat = (int **)ALLOCA(row*sizeof(int *));
4639: for ( i = 0; i < row; i++ ) {
4640: wmat[i] = (int *)ALLOCA(col*sizeof(int));
4641: for ( j = 0; j < col; j++ ) {
1.113 noro 4642: if ( mat0[i][j] ) {
4643: t = rem(NM(mat0[i][j]),mod);
4644: if ( SGN(mat0[i][j]) < 0 ) t = mod-t;
4645: wmat[i][j] = t;
4646: } else
1.108 noro 4647: wmat[i][j] = 0;
4648: }
4649: }
1.109 noro 4650: rank0 = nd_gauss_elim_mod(wmat,sugar,row,col,mod,colstat);
4651: NEWMAT(m); m->row = row; m->col = col; m->body = (pointer **)mat0;
1.113 noro 4652: rank = generic_gauss_elim(m,&nm,&dn,&ri,&ci);
1.109 noro 4653: if ( rank != rank0 )
4654: error("afo");
4655: for ( i = 0; i < row; i++ )
4656: for ( j = 0; j < col; j++ )
4657: mat0[i][j] = 0;
4658: c = col-rank;
4659: for ( i = 0; i < rank; i++ ) {
1.113 noro 4660: mat0[i][ri[i]] = dn;
1.109 noro 4661: for ( j = 0; j < c; j++ )
1.113 noro 4662: mat0[i][ci[j]] = (Q)BDY(nm)[i][j];
1.109 noro 4663: }
1.113 noro 4664: inv = invm(rem(NM(dn),mod),mod);
4665: if ( SGN(dn) < 0 ) inv = mod-inv;
1.109 noro 4666: for ( i = 0; i < row; i++ )
4667: for ( j = 0; j < col; j++ ) {
1.113 noro 4668: if ( mat0[i][j] ) {
4669: t = rem(NM(mat0[i][j]),mod);
4670: if ( SGN(mat0[i][j]) < 0 ) t = mod-t;
4671: } else
1.109 noro 4672: t = 0;
4673: c = dmar(t,inv,0,mod);
4674: if ( wmat[i][j] != c )
4675: error("afo");
4676: }
4677: return rank;
1.76 noro 4678: }
4679:
4680: int nd_gauss_elim_mod(int **mat0,int *sugar,int row,int col,int md,int *colstat)
4681: {
4682: int i,j,k,l,inv,a,rank,s;
4683: unsigned int *t,*pivot,*pk;
4684: unsigned int **mat;
4685:
4686: mat = (unsigned int **)mat0;
4687: for ( rank = 0, j = 0; j < col; j++ ) {
4688: for ( i = rank; i < row; i++ )
4689: mat[i][j] %= md;
4690: for ( i = rank; i < row; i++ )
4691: if ( mat[i][j] )
4692: break;
4693: if ( i == row ) {
4694: colstat[j] = 0;
4695: continue;
4696: } else
4697: colstat[j] = 1;
4698: if ( i != rank ) {
4699: t = mat[i]; mat[i] = mat[rank]; mat[rank] = t;
4700: s = sugar[i]; sugar[i] = sugar[rank]; sugar[rank] = s;
4701: }
4702: pivot = mat[rank];
4703: s = sugar[rank];
4704: inv = invm(pivot[j],md);
4705: for ( k = j, pk = pivot+k; k < col; k++, pk++ )
4706: if ( *pk ) {
4707: if ( *pk >= (unsigned int)md )
4708: *pk %= md;
4709: DMAR(*pk,inv,0,md,*pk)
4710: }
4711: for ( i = rank+1; i < row; i++ ) {
4712: t = mat[i];
4713: if ( a = t[j] ) {
4714: sugar[i] = MAX(sugar[i],s);
4715: red_by_vect(md,t+j,pivot+j,md-a,col-j);
4716: }
4717: }
4718: rank++;
4719: }
4720: for ( j = col-1, l = rank-1; j >= 0; j-- )
4721: if ( colstat[j] ) {
4722: pivot = mat[l];
4723: s = sugar[l];
4724: for ( i = 0; i < l; i++ ) {
4725: t = mat[i];
4726: t[j] %= md;
4727: if ( a = t[j] ) {
4728: sugar[i] = MAX(sugar[i],s);
4729: red_by_vect(md,t+j,pivot+j,md-a,col-j);
4730: }
4731: }
4732: l--;
4733: }
4734: for ( j = 0, l = 0; l < rank; j++ )
4735: if ( colstat[j] ) {
4736: t = mat[l];
4737: for ( k = j; k < col; k++ )
4738: if ( t[k] >= (unsigned int)md )
4739: t[k] %= md;
4740: l++;
4741: }
4742: return rank;
4743: }
4744:
4745: int nd_gauss_elim_sf(int **mat0,int *sugar,int row,int col,int md,int *colstat)
4746: {
4747: int i,j,k,l,inv,a,rank,s;
4748: unsigned int *t,*pivot,*pk;
4749: unsigned int **mat;
4750:
4751: mat = (unsigned int **)mat0;
4752: for ( rank = 0, j = 0; j < col; j++ ) {
4753: for ( i = rank; i < row; i++ )
4754: if ( mat[i][j] )
4755: break;
4756: if ( i == row ) {
4757: colstat[j] = 0;
4758: continue;
4759: } else
4760: colstat[j] = 1;
4761: if ( i != rank ) {
4762: t = mat[i]; mat[i] = mat[rank]; mat[rank] = t;
4763: s = sugar[i]; sugar[i] = sugar[rank]; sugar[rank] = s;
4764: }
4765: pivot = mat[rank];
4766: s = sugar[rank];
4767: inv = _invsf(pivot[j]);
4768: for ( k = j, pk = pivot+k; k < col; k++, pk++ )
4769: if ( *pk )
4770: *pk = _mulsf(*pk,inv);
4771: for ( i = rank+1; i < row; i++ ) {
4772: t = mat[i];
4773: if ( a = t[j] ) {
4774: sugar[i] = MAX(sugar[i],s);
4775: red_by_vect_sf(md,t+j,pivot+j,_chsgnsf(a),col-j);
4776: }
4777: }
4778: rank++;
4779: }
4780: for ( j = col-1, l = rank-1; j >= 0; j-- )
4781: if ( colstat[j] ) {
4782: pivot = mat[l];
4783: s = sugar[l];
4784: for ( i = 0; i < l; i++ ) {
4785: t = mat[i];
4786: if ( a = t[j] ) {
4787: sugar[i] = MAX(sugar[i],s);
4788: red_by_vect_sf(md,t+j,pivot+j,_chsgnsf(a),col-j);
4789: }
4790: }
4791: l--;
4792: }
4793: return rank;
1.77 noro 4794: }
4795:
4796: int ndv_ishomo(NDV p)
4797: {
4798: NMV m;
4799: int len,h;
4800:
4801: if ( !p ) return 1;
4802: len = LEN(p);
4803: m = BDY(p);
4804: h = TD(DL(m));
4805: NMV_ADV(m);
4806: for ( len--; len; len--, NMV_ADV(m) )
4807: if ( TD(DL(m)) != h ) return 0;
4808: return 1;
4809: }
4810:
4811: void ndv_save(NDV p,int index)
4812: {
4813: FILE *s;
4814: char name[BUFSIZ];
4815: short id;
4816: int nv,sugar,len,n,i,td,e,j;
4817: NMV m;
4818: unsigned int *dl;
4819:
4820: sprintf(name,"%s/%d",Demand,index);
4821: s = fopen(name,"w");
4822: savevl(s,0);
4823: if ( !p ) {
4824: saveobj(s,0);
4825: return;
4826: }
4827: id = O_DP;
4828: nv = NV(p);
4829: sugar = SG(p);
4830: len = LEN(p);
4831: write_short(s,&id); write_int(s,&nv); write_int(s,&sugar);
4832: write_int(s,&len);
4833:
4834: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
1.113 noro 4835: saveobj(s,(Obj)CQ(m));
1.77 noro 4836: dl = DL(m);
4837: td = TD(dl);
4838: write_int(s,&td);
4839: for ( j = 0; j < nv; j++ ) {
4840: e = GET_EXP(dl,j);
4841: write_int(s,&e);
4842: }
4843: }
4844: fclose(s);
4845: }
4846:
4847: NDV ndv_load(int index)
4848: {
4849: FILE *s;
4850: char name[BUFSIZ];
4851: short id;
4852: int nv,sugar,len,n,i,td,e,j;
4853: NDV d;
4854: NMV m0,m;
4855: unsigned int *dl;
4856: Obj obj;
4857:
4858: sprintf(name,"%s/%d",Demand,index);
4859: s = fopen(name,"r");
4860: if ( !s ) return 0;
4861:
4862: skipvl(s);
4863: read_short(s,&id);
4864: if ( !id ) return 0;
4865: read_int(s,&nv);
4866: read_int(s,&sugar);
4867: read_int(s,&len);
4868:
4869: m0 = m = MALLOC(len*nmv_adv);
4870: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
1.113 noro 4871: loadobj(s,&obj); CQ(m) = (Q)obj;
1.77 noro 4872: dl = DL(m);
4873: ndl_zero(dl);
4874: read_int(s,&td); TD(dl) = td;
4875: for ( j = 0; j < nv; j++ ) {
4876: read_int(s,&e);
4877: PUT_EXP(dl,j,e);
4878: }
4879: if ( nd_blockmask ) ndl_weight_mask(dl);
4880: }
4881: fclose(s);
4882: MKNDV(nv,m0,len,d);
4883: SG(d) = sugar;
4884: return d;
1.99 noro 4885: }
4886:
1.102 noro 4887: void nd_det(int mod,MAT f,P *rp)
1.99 noro 4888: {
4889: VL fv,tv;
1.102 noro 4890: int n,i,j,max,e,nvar,sgn,k0,l0,len0,len,k,l,a;
1.99 noro 4891: pointer **m;
1.113 noro 4892: Q mone;
1.102 noro 4893: NDV **dm;
4894: NDV *t,*mi,*mj;
4895: NDV d,s,mij,mjj;
4896: ND u;
4897: NMV nmv;
1.114 noro 4898: UINT *bound;
1.99 noro 4899: PGeoBucket bucket;
4900: struct order_spec *ord;
4901:
4902: create_order_spec(0,0,&ord);
4903: nd_init_ord(ord);
4904: get_vars((Obj)f,&fv);
4905: if ( f->row != f->col )
4906: error("nd_det : non-square matrix");
4907: n = f->row;
4908: for ( nvar = 0, tv = fv; tv; tv = NEXT(tv), nvar++ );
4909: m = f->body;
4910: for ( i = 0, max = 0; i < n; i++ )
4911: for ( j = 0; j < n; j++ )
4912: for ( tv = fv; tv; tv = NEXT(tv) ) {
4913: e = getdeg(tv->v,(P)m[i][j]);
4914: max = MAX(e,max);
4915: }
1.114 noro 4916: nd_setup_parameters(nvar,max);
1.102 noro 4917: dm = (NDV **)almat_pointer(n,n);
1.99 noro 4918: for ( i = 0, max = 0; i < n; i++ )
1.102 noro 4919: for ( j = 0; j < n; j++ ) {
4920: dm[i][j] = ptondv(CO,fv,m[i][j]);
4921: if ( mod ) ndv_mod(mod,dm[i][j]);
4922: if ( dm[i][j] && !LEN(dm[i][j]) ) dm[i][j] = 0;
4923: }
4924: d = ptondv(CO,fv,(P)ONE);
4925: if ( mod ) ndv_mod(mod,d);
1.113 noro 4926: chsgnq(ONE,&mone);
1.99 noro 4927: for ( j = 0, sgn = 1; j < n; j++ ) {
1.104 noro 4928: if ( DP_Print ) fprintf(stderr,"j=%d\n",j);
1.99 noro 4929: for ( i = j; i < n && !dm[i][j]; i++ );
4930: if ( i == n ) {
4931: *rp = 0;
4932: return;
4933: }
1.102 noro 4934: k0 = i; l0 = j; len0 = LEN(dm[k0][l0]);
1.99 noro 4935: for ( k = j; k < n; k++ )
4936: for ( l = j; l < n; l++ )
1.102 noro 4937: if ( dm[k][l] && LEN(dm[k][l]) < len0 ) {
4938: k0 = k; l0 = l; len0 = LEN(dm[k][l]);
1.99 noro 4939: }
4940: if ( k0 != j ) {
4941: t = dm[j]; dm[j] = dm[k0]; dm[k0] = t;
4942: sgn = -sgn;
4943: }
4944: if ( l0 != j ) {
4945: for ( k = j; k < n; k++ ) {
4946: s = dm[k][j]; dm[k][j] = dm[k][l0]; dm[k][l0] = s;
4947: }
4948: sgn = -sgn;
4949: }
1.114 noro 4950: bound = nd_det_compute_bound(dm,n,j);
4951: if ( ndl_check_bound(bound,bound) )
4952: nd_det_reconstruct(dm,n,j,d);
4953:
1.99 noro 4954: for ( i = j+1, mj = dm[j], mjj = mj[j]; i < n; i++ ) {
1.114 noro 4955: /* if ( DP_Print ) fprintf(stderr," i=%d\n ",i); */
1.99 noro 4956: mi = dm[i]; mij = mi[j];
1.102 noro 4957: if ( mod )
4958: ndv_mul_c(mod,mij,mod-1);
4959: else
1.113 noro 4960: ndv_mul_c_q(mij,mone);
1.99 noro 4961: for ( k = j+1; k < n; k++ ) {
1.114 noro 4962: /* if ( DP_Print ) fprintf(stderr,"k=%d ",k); */
1.99 noro 4963: bucket = create_pbucket();
1.104 noro 4964: if ( mi[k] ) {
1.102 noro 4965: nmv = BDY(mjj); len = LEN(mjj);
4966: for ( a = 0; a < len; a++, NMV_ADV(nmv) ) {
4967: u = ndv_mul_nmv_trunc(mod,nmv,mi[k],DL(BDY(d)));
4968: add_pbucket(mod,bucket,u);
1.99 noro 4969: }
1.104 noro 4970: }
1.99 noro 4971: if ( mj[k] && mij ) {
1.102 noro 4972: nmv = BDY(mij); len = LEN(mij);
4973: for ( a = 0; a < len; a++, NMV_ADV(nmv) ) {
4974: u = ndv_mul_nmv_trunc(mod,nmv,mj[k],DL(BDY(d)));
4975: add_pbucket(mod,bucket,u);
1.99 noro 4976: }
4977: }
1.104 noro 4978: u = nd_quo(mod,bucket,d);
1.102 noro 4979: mi[k] = ndtondv(mod,u);
1.99 noro 4980: }
1.114 noro 4981: /* if ( DP_Print ) fprintf(stderr,"\n",k); */
1.99 noro 4982: }
4983: d = mjj;
4984: }
1.100 noro 4985: if ( sgn < 0 )
1.102 noro 4986: if ( mod )
4987: ndv_mul_c(mod,d,mod-1);
4988: else
1.113 noro 4989: ndv_mul_c_q(d,mone);
1.102 noro 4990: *rp = ndvtop(mod,CO,fv,d);
1.99 noro 4991: }
4992:
1.102 noro 4993: ND ndv_mul_nmv_trunc(int mod,NMV m0,NDV p,UINT *d)
1.99 noro 4994: {
4995: NM mr,mr0;
1.102 noro 4996: NM tnm;
4997: NMV m;
1.99 noro 4998: UINT *d0,*dt,*dm;
4999: int c,n,td,i,c1,c2,len;
1.113 noro 5000: Q q;
1.99 noro 5001: ND r;
5002:
5003: if ( !p ) return 0;
5004: else {
1.102 noro 5005: n = NV(p); m = BDY(p); len = LEN(p);
1.99 noro 5006: d0 = DL(m0);
1.102 noro 5007: td = TD(d);
1.99 noro 5008: mr0 = 0;
5009: NEWNM(tnm);
1.102 noro 5010: if ( mod ) {
5011: c = CM(m0);
5012: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
5013: ndl_add(DL(m),d0,DL(tnm));
5014: if ( ndl_reducible(DL(tnm),d) ) {
5015: NEXTNM(mr0,mr);
5016: c1 = CM(m); DMAR(c1,c,0,mod,c2); CM(mr) = c2;
1.104 noro 5017: ndl_copy(DL(tnm),DL(mr));
1.102 noro 5018: }
5019: }
5020: } else {
1.113 noro 5021: q = CQ(m0);
1.102 noro 5022: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
5023: ndl_add(DL(m),d0,DL(tnm));
5024: if ( ndl_reducible(DL(tnm),d) ) {
5025: NEXTNM(mr0,mr);
1.113 noro 5026: mulq(CQ(m),q,&CQ(mr));
1.104 noro 5027: ndl_copy(DL(tnm),DL(mr));
1.102 noro 5028: }
1.99 noro 5029: }
5030: }
5031: if ( !mr0 )
5032: return 0;
5033: else {
5034: NEXT(mr) = 0;
1.105 noro 5035: for ( len = 0, mr = mr0; mr; mr = NEXT(mr), len++ );
1.99 noro 5036: MKND(NV(p),mr0,len,r);
5037: SG(r) = SG(p) + TD(d0);
5038: return r;
5039: }
5040: }
1.114 noro 5041: }
5042:
5043: void nd_det_reconstruct(NDV **dm,int n,int j,NDV d)
5044: {
5045: int i,obpe,oadv,h,k,l;
5046: static NM prev_nm_free_list;
5047: EPOS oepos;
5048:
5049: obpe = nd_bpe;
5050: oadv = nmv_adv;
5051: oepos = nd_epos;
5052: if ( obpe < 2 ) nd_bpe = 2;
5053: else if ( obpe < 3 ) nd_bpe = 3;
5054: else if ( obpe < 4 ) nd_bpe = 4;
5055: else if ( obpe < 5 ) nd_bpe = 5;
5056: else if ( obpe < 6 ) nd_bpe = 6;
5057: else if ( obpe < 8 ) nd_bpe = 8;
5058: else if ( obpe < 10 ) nd_bpe = 10;
5059: else if ( obpe < 16 ) nd_bpe = 16;
5060: else if ( obpe < 32 ) nd_bpe = 32;
5061: else error("nd_det_reconstruct : exponent too large");
5062:
5063: nd_setup_parameters(nd_nvar,0);
5064: prev_nm_free_list = _nm_free_list;
5065: _nm_free_list = 0;
5066: for ( k = j; k < n; k++ )
5067: for (l = j; l < n; l++ )
5068: ndv_realloc(dm[k][l],obpe,oadv,oepos);
5069: ndv_realloc(d,obpe,oadv,oepos);
5070: prev_nm_free_list = 0;
5071: #if 0
5072: GC_gcollect();
5073: #endif
5074: }
5075:
5076: UINT *nd_det_compute_bound(NDV **dm,int n,int j)
5077: {
5078: UINT *d0,*d1,*d,*t,*r;
5079: int k,l;
5080:
5081: d0 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
5082: d1 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
5083: for ( k = 0; k < nd_wpd; k++ ) d0[k] = 0;
5084: for ( k = j; k < n; k++ )
5085: for ( l = j; l < n; l++ )
5086: if ( dm[k][l] ) {
5087: d = ndv_compute_bound(dm[k][l]);
5088: ndl_lcm(d,d0,d1);
5089: t = d1; d1 = d0; d0 = t;
5090: }
5091: r = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
5092: for ( k = 0; k < nd_wpd; k++ ) r[k] = d0[k];
5093: return r;
1.59 noro 5094: }
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>