Annotation of OpenXM_contrib2/asir2000/engine/nd.c, Revision 1.112
1.112 ! noro 1: /* $OpenXM: OpenXM_contrib2/asir2000/engine/nd.c,v 1.111 2004/09/21 07:19:01 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:
44: void nd_free_private_storage()
45: {
46: _nm_free_list = 0;
1.5 noro 47: _ndp_free_list = 0;
1.71 noro 48: #if 0
1.1 noro 49: GC_gcollect();
1.71 noro 50: #endif
1.1 noro 51: }
52:
53: void _NM_alloc()
54: {
55: NM p;
56: int i;
57:
1.11 noro 58: for ( i = 0; i < 1024; i++ ) {
1.61 noro 59: p = (NM)GC_malloc(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT));
1.1 noro 60: p->next = _nm_free_list; _nm_free_list = p;
61: }
62: }
63:
64: void _ND_alloc()
65: {
66: ND p;
67: int i;
68:
69: for ( i = 0; i < 1024; i++ ) {
70: p = (ND)GC_malloc(sizeof(struct oND));
71: p->body = (NM)_nd_free_list; _nd_free_list = p;
72: }
73: }
74:
75: void _NDP_alloc()
76: {
77: ND_pairs p;
78: int i;
79:
1.11 noro 80: for ( i = 0; i < 1024; i++ ) {
1.1 noro 81: p = (ND_pairs)GC_malloc(sizeof(struct oND_pairs)
1.61 noro 82: +(nd_wpd-1)*sizeof(UINT));
1.1 noro 83: p->next = _ndp_free_list; _ndp_free_list = p;
84: }
85: }
86:
1.30 noro 87: INLINE int nd_length(ND p)
1.1 noro 88: {
89: NM m;
90: int i;
91:
92: if ( !p )
93: return 0;
94: else {
95: for ( i = 0, m = BDY(p); m; m = NEXT(m), i++ );
96: return i;
97: }
98: }
99:
1.61 noro 100: INLINE int ndl_reducible(UINT *d1,UINT *d2)
1.1 noro 101: {
1.61 noro 102: UINT u1,u2;
1.1 noro 103: int i,j;
104:
1.34 noro 105: if ( TD(d1) < TD(d2) ) return 0;
1.65 noro 106: #if USE_UNROLL
1.1 noro 107: switch ( nd_bpe ) {
1.62 noro 108: case 3:
109: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
110: u1 = d1[i]; u2 = d2[i];
111: if ( (u1&0x38000000) < (u2&0x38000000) ) return 0;
112: if ( (u1& 0x7000000) < (u2& 0x7000000) ) return 0;
113: if ( (u1& 0xe00000) < (u2& 0xe00000) ) return 0;
114: if ( (u1& 0x1c0000) < (u2& 0x1c0000) ) return 0;
115: if ( (u1& 0x38000) < (u2& 0x38000) ) return 0;
116: if ( (u1& 0x7000) < (u2& 0x7000) ) return 0;
117: if ( (u1& 0xe00) < (u2& 0xe00) ) return 0;
118: if ( (u1& 0x1c0) < (u2& 0x1c0) ) return 0;
119: if ( (u1& 0x38) < (u2& 0x38) ) return 0;
120: if ( (u1& 0x7) < (u2& 0x7) ) return 0;
121: }
122: return 1;
123: break;
1.1 noro 124: case 4:
1.41 noro 125: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 126: u1 = d1[i]; u2 = d2[i];
127: if ( (u1&0xf0000000) < (u2&0xf0000000) ) return 0;
1.62 noro 128: if ( (u1& 0xf000000) < (u2& 0xf000000) ) return 0;
129: if ( (u1& 0xf00000) < (u2& 0xf00000) ) return 0;
130: if ( (u1& 0xf0000) < (u2& 0xf0000) ) return 0;
131: if ( (u1& 0xf000) < (u2& 0xf000) ) return 0;
132: if ( (u1& 0xf00) < (u2& 0xf00) ) return 0;
133: if ( (u1& 0xf0) < (u2& 0xf0) ) return 0;
134: if ( (u1& 0xf) < (u2& 0xf) ) return 0;
1.1 noro 135: }
136: return 1;
137: break;
138: case 6:
1.41 noro 139: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 140: u1 = d1[i]; u2 = d2[i];
141: if ( (u1&0x3f000000) < (u2&0x3f000000) ) return 0;
1.62 noro 142: if ( (u1& 0xfc0000) < (u2& 0xfc0000) ) return 0;
143: if ( (u1& 0x3f000) < (u2& 0x3f000) ) return 0;
144: if ( (u1& 0xfc0) < (u2& 0xfc0) ) return 0;
145: if ( (u1& 0x3f) < (u2& 0x3f) ) return 0;
1.1 noro 146: }
147: return 1;
148: break;
149: case 8:
1.41 noro 150: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 151: u1 = d1[i]; u2 = d2[i];
152: if ( (u1&0xff000000) < (u2&0xff000000) ) return 0;
1.62 noro 153: if ( (u1& 0xff0000) < (u2& 0xff0000) ) return 0;
154: if ( (u1& 0xff00) < (u2& 0xff00) ) return 0;
155: if ( (u1& 0xff) < (u2& 0xff) ) return 0;
1.1 noro 156: }
157: return 1;
158: break;
159: case 16:
1.41 noro 160: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 161: u1 = d1[i]; u2 = d2[i];
162: if ( (u1&0xffff0000) < (u2&0xffff0000) ) return 0;
1.62 noro 163: if ( (u1& 0xffff) < (u2& 0xffff) ) return 0;
1.1 noro 164: }
165: return 1;
166: break;
167: case 32:
1.41 noro 168: for ( i = nd_exporigin; i < nd_wpd; i++ )
1.1 noro 169: if ( d1[i] < d2[i] ) return 0;
170: return 1;
171: break;
172: default:
1.41 noro 173: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 174: u1 = d1[i]; u2 = d2[i];
175: for ( j = 0; j < nd_epw; j++ )
176: if ( (u1&nd_mask[j]) < (u2&nd_mask[j]) ) return 0;
177: }
178: return 1;
179: }
1.65 noro 180: #else
181: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
182: u1 = d1[i]; u2 = d2[i];
183: for ( j = 0; j < nd_epw; j++ )
184: if ( (u1&nd_mask[j]) < (u2&nd_mask[j]) ) return 0;
185: }
186: return 1;
187: #endif
1.1 noro 188: }
189:
1.61 noro 190: /*
191: * If the current order is a block order,
192: * then the last block is length 1 and contains
193: * the homo variable. Otherwise, the original
194: * order is either 0 or 2.
195: */
196:
197: void ndl_homogenize(UINT *d,UINT *r,int obpe,EPOS oepos,int weight)
1.23 noro 198: {
1.61 noro 199: int w,i,e,n,omask0;
200:
201: omask0 = (1<<obpe)-1;
202: n = nd_nvar-1;
203: ndl_zero(r);
204: for ( i = 0; i < n; i++ ) {
205: e = GET_EXP_OLD(d,i);
206: PUT_EXP(r,i,e);
207: }
208: w = TD(d);
209: PUT_EXP(r,nd_nvar-1,weight-w);
210: TD(r) = weight;
211: if ( nd_blockmask ) ndl_weight_mask(r);
212: }
213:
214: void ndl_dehomogenize(UINT *d)
215: {
216: UINT mask;
217: UINT h;
1.31 noro 218: int i,bits;
1.23 noro 219:
1.44 noro 220: if ( nd_blockmask ) {
221: h = GET_EXP(d,nd_nvar-1);
1.45 noro 222: XOR_EXP(d,nd_nvar-1,h);
1.44 noro 223: TD(d) -= h;
224: d[nd_exporigin-1] -= h;
225: } else {
226: if ( nd_isrlex ) {
227: if ( nd_bpe == 32 ) {
228: h = d[nd_exporigin];
229: for ( i = nd_exporigin+1; i < nd_wpd; i++ )
230: d[i-1] = d[i];
231: d[i-1] = 0;
232: TD(d) -= h;
233: } else {
234: bits = nd_epw*nd_bpe;
235: mask = bits==32?0xffffffff:((1<<(nd_epw*nd_bpe))-1);
236: h = (d[nd_exporigin]>>((nd_epw-1)*nd_bpe))&nd_mask0;
237: for ( i = nd_exporigin; i < nd_wpd; i++ )
238: d[i] = ((d[i]<<nd_bpe)&mask)
239: |(i+1<nd_wpd?((d[i+1]>>((nd_epw-1)*nd_bpe))&nd_mask0):0);
240: TD(d) -= h;
241: }
1.45 noro 242: } else {
243: h = GET_EXP(d,nd_nvar-1);
244: XOR_EXP(d,nd_nvar-1,h);
245: TD(d) -= h;
246: }
1.44 noro 247: }
1.23 noro 248: }
249:
1.61 noro 250: void ndl_lcm(UINT *d1,unsigned *d2,UINT *d)
1.1 noro 251: {
1.61 noro 252: UINT t1,t2,u,u1,u2;
1.43 noro 253: int i,j,l;
1.1 noro 254:
1.65 noro 255: #if USE_UNROLL
1.1 noro 256: switch ( nd_bpe ) {
1.62 noro 257: case 3:
258: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
259: u1 = d1[i]; u2 = d2[i];
260: t1 = (u1&0x38000000); t2 = (u2&0x38000000); u = t1>t2?t1:t2;
261: t1 = (u1& 0x7000000); t2 = (u2& 0x7000000); u |= t1>t2?t1:t2;
262: t1 = (u1& 0xe00000); t2 = (u2& 0xe00000); u |= t1>t2?t1:t2;
263: t1 = (u1& 0x1c0000); t2 = (u2& 0x1c0000); u |= t1>t2?t1:t2;
264: t1 = (u1& 0x38000); t2 = (u2& 0x38000); u |= t1>t2?t1:t2;
265: t1 = (u1& 0x7000); t2 = (u2& 0x7000); u |= t1>t2?t1:t2;
266: t1 = (u1& 0xe00); t2 = (u2& 0xe00); u |= t1>t2?t1:t2;
267: t1 = (u1& 0x1c0); t2 = (u2& 0x1c0); u |= t1>t2?t1:t2;
268: t1 = (u1& 0x38); t2 = (u2& 0x38); u |= t1>t2?t1:t2;
269: t1 = (u1& 0x7); t2 = (u2& 0x7); u |= t1>t2?t1:t2;
270: d[i] = u;
271: }
272: break;
1.1 noro 273: case 4:
1.41 noro 274: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 275: u1 = d1[i]; u2 = d2[i];
276: t1 = (u1&0xf0000000); t2 = (u2&0xf0000000); u = t1>t2?t1:t2;
1.62 noro 277: t1 = (u1& 0xf000000); t2 = (u2& 0xf000000); u |= t1>t2?t1:t2;
278: t1 = (u1& 0xf00000); t2 = (u2& 0xf00000); u |= t1>t2?t1:t2;
279: t1 = (u1& 0xf0000); t2 = (u2& 0xf0000); u |= t1>t2?t1:t2;
280: t1 = (u1& 0xf000); t2 = (u2& 0xf000); u |= t1>t2?t1:t2;
281: t1 = (u1& 0xf00); t2 = (u2& 0xf00); u |= t1>t2?t1:t2;
282: t1 = (u1& 0xf0); t2 = (u2& 0xf0); u |= t1>t2?t1:t2;
283: t1 = (u1& 0xf); t2 = (u2& 0xf); u |= t1>t2?t1:t2;
1.1 noro 284: d[i] = u;
285: }
286: break;
287: case 6:
1.41 noro 288: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 289: u1 = d1[i]; u2 = d2[i];
290: t1 = (u1&0x3f000000); t2 = (u2&0x3f000000); u = t1>t2?t1:t2;
1.62 noro 291: t1 = (u1& 0xfc0000); t2 = (u2& 0xfc0000); u |= t1>t2?t1:t2;
292: t1 = (u1& 0x3f000); t2 = (u2& 0x3f000); u |= t1>t2?t1:t2;
293: t1 = (u1& 0xfc0); t2 = (u2& 0xfc0); u |= t1>t2?t1:t2;
294: t1 = (u1& 0x3f); t2 = (u2& 0x3f); u |= t1>t2?t1:t2;
1.1 noro 295: d[i] = u;
296: }
297: break;
298: case 8:
1.41 noro 299: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 300: u1 = d1[i]; u2 = d2[i];
301: t1 = (u1&0xff000000); t2 = (u2&0xff000000); u = t1>t2?t1:t2;
1.62 noro 302: t1 = (u1& 0xff0000); t2 = (u2& 0xff0000); u |= t1>t2?t1:t2;
303: t1 = (u1& 0xff00); t2 = (u2& 0xff00); u |= t1>t2?t1:t2;
304: t1 = (u1& 0xff); t2 = (u2& 0xff); u |= t1>t2?t1:t2;
1.1 noro 305: d[i] = u;
306: }
307: break;
308: case 16:
1.41 noro 309: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 310: u1 = d1[i]; u2 = d2[i];
311: t1 = (u1&0xffff0000); t2 = (u2&0xffff0000); u = t1>t2?t1:t2;
1.62 noro 312: t1 = (u1& 0xffff); t2 = (u2& 0xffff); u |= t1>t2?t1:t2;
1.1 noro 313: d[i] = u;
314: }
315: break;
316: case 32:
1.41 noro 317: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 318: u1 = d1[i]; u2 = d2[i];
319: d[i] = u1>u2?u1:u2;
320: }
321: break;
322: default:
1.41 noro 323: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 324: u1 = d1[i]; u2 = d2[i];
325: for ( j = 0, u = 0; j < nd_epw; j++ ) {
326: t1 = (u1&nd_mask[j]); t2 = (u2&nd_mask[j]); u |= t1>t2?t1:t2;
327: }
328: d[i] = u;
329: }
330: break;
331: }
1.65 noro 332: #else
333: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
334: u1 = d1[i]; u2 = d2[i];
335: for ( j = 0, u = 0; j < nd_epw; j++ ) {
336: t1 = (u1&nd_mask[j]); t2 = (u2&nd_mask[j]); u |= t1>t2?t1:t2;
337: }
338: d[i] = u;
339: }
340: #endif
1.39 noro 341: TD(d) = ndl_weight(d);
1.61 noro 342: if ( nd_blockmask ) ndl_weight_mask(d);
1.57 noro 343: }
344:
1.61 noro 345: int ndl_weight(UINT *d)
1.1 noro 346: {
1.61 noro 347: UINT t,u;
1.1 noro 348: int i,j;
349:
1.60 noro 350: if ( current_dl_weight_vector )
351: for ( i = 0, t = 0; i < nd_nvar; i++ ) {
352: u = GET_EXP(d,i);
353: t += MUL_WEIGHT(u,i);
354: }
355: else
356: for ( t = 0, i = nd_exporigin; i < nd_wpd; i++ ) {
357: u = d[i];
358: for ( j = 0; j < nd_epw; j++, u>>=nd_bpe )
359: t += (u&nd_mask0);
360: }
1.1 noro 361: return t;
362: }
363:
1.61 noro 364: void ndl_weight_mask(UINT *d)
1.43 noro 365: {
1.61 noro 366: UINT t,u;
367: UINT *mask;
368: int i,j,k,l;
1.43 noro 369:
1.61 noro 370: l = nd_blockmask->n;
371: for ( k = 0; k < l; k++ ) {
372: mask = nd_blockmask->mask[k];
373: if ( current_dl_weight_vector )
374: for ( i = 0, t = 0; i < nd_nvar; i++ ) {
375: u = GET_EXP_MASK(d,i,mask);
376: t += MUL_WEIGHT(u,i);
377: }
378: else
379: for ( t = 0, i = nd_exporigin; i < nd_wpd; i++ ) {
380: u = d[i]&mask[i];
381: for ( j = 0; j < nd_epw; j++, u>>=nd_bpe )
382: t += (u&nd_mask0);
383: }
384: d[k+1] = t;
385: }
1.43 noro 386: }
387:
1.61 noro 388: int ndl_lex_compare(UINT *d1,UINT *d2)
1.1 noro 389: {
390: int i;
391:
1.41 noro 392: d1 += nd_exporigin;
393: d2 += nd_exporigin;
394: for ( i = nd_exporigin; i < nd_wpd; i++, d1++, d2++ )
1.1 noro 395: if ( *d1 > *d2 )
1.32 noro 396: return nd_isrlex ? -1 : 1;
1.1 noro 397: else if ( *d1 < *d2 )
1.32 noro 398: return nd_isrlex ? 1 : -1;
1.1 noro 399: return 0;
400: }
401:
1.61 noro 402: int ndl_block_compare(UINT *d1,UINT *d2)
1.43 noro 403: {
404: int i,l,j,ord_o,ord_l;
405: struct order_pair *op;
1.61 noro 406: UINT t1,t2,m;
407: UINT *mask;
1.43 noro 408:
409: l = nd_blockmask->n;
410: op = nd_blockmask->order_pair;
411: for ( j = 0; j < l; j++ ) {
412: mask = nd_blockmask->mask[j];
413: ord_o = op[j].order;
414: if ( ord_o < 2 )
1.44 noro 415: if ( (t1=d1[j+1]) > (t2=d2[j+1]) ) return 1;
416: else if ( t1 < t2 ) return -1;
1.43 noro 417: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.44 noro 418: m = mask[i];
419: t1 = d1[i]&m;
420: t2 = d2[i]&m;
1.43 noro 421: if ( t1 > t2 )
422: return !ord_o ? -1 : 1;
423: else if ( t1 < t2 )
424: return !ord_o ? 1 : -1;
425: }
426: }
427: return 0;
428: }
429:
1.96 noro 430: int ndl_matrix_compare(UINT *d1,UINT *d2)
431: {
432: int i,j,s;
433: int *v;
434:
435: for ( j = 0; j < nd_nvar; j++ )
436: nd_work_vector[j] = GET_EXP(d1,j)-GET_EXP(d2,j);
437: for ( i = 0; i < nd_matrix_len; i++ ) {
438: v = nd_matrix[i];
439: for ( j = 0, s = 0; j < nd_nvar; j++ )
440: s += v[j]*nd_work_vector[j];
441: if ( s > 0 ) return 1;
442: else if ( s < 0 ) return -1;
443: }
444: return 0;
445: }
446:
1.97 noro 447: int ndl_composite_compare(UINT *d1,UINT *d2)
448: {
449: int i,j,s,start,end,len,o;
450: int *v;
451: struct sparse_weight *sw;
452:
453: for ( j = 0; j < nd_nvar; j++ )
454: nd_work_vector[j] = GET_EXP(d1,j)-GET_EXP(d2,j);
455: for ( i = 0; i < nd_worb_len; i++ ) {
456: len = nd_worb[i].length;
457: switch ( nd_worb[i].type ) {
458: case IS_DENSE_WEIGHT:
459: v = nd_worb[i].body.dense_weight;
460: for ( j = 0, s = 0; j < len; j++ )
461: s += v[j]*nd_work_vector[j];
462: if ( s > 0 ) return 1;
463: else if ( s < 0 ) return -1;
464: break;
465: case IS_SPARSE_WEIGHT:
466: sw = nd_worb[i].body.sparse_weight;
467: for ( j = 0, s = 0; j < len; j++ )
468: s += sw[j].value*nd_work_vector[sw[j].pos];
469: if ( s > 0 ) return 1;
470: else if ( s < 0 ) return -1;
471: break;
472: case IS_BLOCK:
473: o = nd_worb[i].body.block.order;
474: start = nd_worb[i].body.block.start;
475: switch ( o ) {
476: case 0:
477: end = start+len;
478: for ( j = start, s = 0; j < end; j++ )
479: s += MUL_WEIGHT(nd_work_vector[j],j);
480: if ( s > 0 ) return 1;
481: else if ( s < 0 ) return -1;
482: for ( j = end-1; j >= start; j-- )
483: if ( nd_work_vector[j] < 0 ) return 1;
484: else if ( nd_work_vector[j] > 0 ) return -1;
485: break;
486: case 1:
487: end = start+len;
488: for ( j = start, s = 0; j < end; j++ )
489: s += MUL_WEIGHT(nd_work_vector[j],j);
490: if ( s > 0 ) return 1;
491: else if ( s < 0 ) return -1;
492: for ( j = start; j < end; j++ )
493: if ( nd_work_vector[j] > 0 ) return 1;
494: else if ( nd_work_vector[j] < 0 ) return -1;
495: break;
496: case 2:
497: for ( j = start; j < end; j++ )
498: if ( nd_work_vector[j] > 0 ) return 1;
499: else if ( nd_work_vector[j] < 0 ) return -1;
500: break;
501: }
502: break;
503: }
504: }
505: return 0;
506: }
507:
1.58 noro 508: /* TDH -> WW -> TD-> RL */
509:
1.61 noro 510: int ndl_ww_lex_compare(UINT *d1,UINT *d2)
1.58 noro 511: {
512: int i,m,e1,e2;
513:
514: if ( TD(d1) > TD(d2) ) return 1;
515: else if ( TD(d1) < TD(d2) ) return -1;
516: m = nd_nvar>>1;
517: for ( i = 0, e1 = e2 = 0; i < m; i++ ) {
518: e1 += current_weyl_weight_vector[i]*(GET_EXP(d1,m+i)-GET_EXP(d1,i));
519: e2 += current_weyl_weight_vector[i]*(GET_EXP(d2,m+i)-GET_EXP(d2,i));
520: }
521: if ( e1 > e2 ) return 1;
522: else if ( e1 < e2 ) return -1;
523: return ndl_lex_compare(d1,d2);
524: }
525:
1.61 noro 526: INLINE int ndl_equal(UINT *d1,UINT *d2)
1.1 noro 527: {
528: int i;
529:
1.81 noro 530: switch ( nd_wpd ) {
531: case 2:
532: if ( TD(d2) != TD(d1) ) return 0;
533: if ( d2[1] != d1[1] ) return 0;
534: return 1;
535: break;
536: case 3:
537: if ( TD(d2) != TD(d1) ) return 0;
538: if ( d2[1] != d1[1] ) return 0;
539: if ( d2[2] != d1[2] ) return 0;
540: return 1;
541: break;
542: default:
543: for ( i = 0; i < nd_wpd; i++ )
544: if ( *d1++ != *d2++ ) return 0;
545: return 1;
546: break;
547: }
1.1 noro 548: }
549:
1.61 noro 550: INLINE void ndl_copy(UINT *d1,UINT *d2)
1.6 noro 551: {
552: int i;
553:
554: switch ( nd_wpd ) {
1.41 noro 555: case 2:
1.34 noro 556: TD(d2) = TD(d1);
557: d2[1] = d1[1];
1.6 noro 558: break;
1.41 noro 559: case 3:
1.34 noro 560: TD(d2) = TD(d1);
1.6 noro 561: d2[1] = d1[1];
1.34 noro 562: d2[2] = d1[2];
1.6 noro 563: break;
564: default:
1.41 noro 565: for ( i = 0; i < nd_wpd; i++ )
1.6 noro 566: d2[i] = d1[i];
567: break;
568: }
569: }
570:
1.61 noro 571: INLINE void ndl_zero(UINT *d)
572: {
573: int i;
574: for ( i = 0; i < nd_wpd; i++ ) d[i] = 0;
575: }
576:
577: INLINE void ndl_add(UINT *d1,UINT *d2,UINT *d)
1.1 noro 578: {
579: int i;
580:
1.43 noro 581: #if 1
1.6 noro 582: switch ( nd_wpd ) {
1.41 noro 583: case 2:
584: TD(d) = TD(d1)+TD(d2);
1.34 noro 585: d[1] = d1[1]+d2[1];
1.6 noro 586: break;
1.41 noro 587: case 3:
588: TD(d) = TD(d1)+TD(d2);
1.6 noro 589: d[1] = d1[1]+d2[1];
1.34 noro 590: d[2] = d1[2]+d2[2];
1.6 noro 591: break;
592: default:
1.43 noro 593: for ( i = 0; i < nd_wpd; i++ ) d[i] = d1[i]+d2[i];
1.6 noro 594: break;
595: }
1.43 noro 596: #else
597: for ( i = 0; i < nd_wpd; i++ ) d[i] = d1[i]+d2[i];
598: #endif
1.6 noro 599: }
600:
1.55 noro 601: /* d1 += d2 */
1.61 noro 602: INLINE void ndl_addto(UINT *d1,UINT *d2)
1.55 noro 603: {
604: int i;
605:
606: #if 1
607: switch ( nd_wpd ) {
608: case 2:
609: TD(d1) += TD(d2);
610: d1[1] += d2[1];
611: break;
612: case 3:
613: TD(d1) += TD(d2);
614: d1[1] += d2[1];
615: d1[2] += d2[2];
616: break;
617: default:
618: for ( i = 0; i < nd_wpd; i++ ) d1[i] += d2[i];
619: break;
620: }
621: #else
622: for ( i = 0; i < nd_wpd; i++ ) d1[i] += d2[i];
623: #endif
624: }
625:
1.61 noro 626: INLINE void ndl_sub(UINT *d1,UINT *d2,UINT *d)
1.6 noro 627: {
628: int i;
629:
1.43 noro 630: for ( i = 0; i < nd_wpd; i++ ) d[i] = d1[i]-d2[i];
1.1 noro 631: }
632:
1.61 noro 633: int ndl_disjoint(UINT *d1,UINT *d2)
1.1 noro 634: {
1.61 noro 635: UINT t1,t2,u,u1,u2;
1.1 noro 636: int i,j;
637:
1.65 noro 638: #if USE_UNROLL
1.1 noro 639: switch ( nd_bpe ) {
1.62 noro 640: case 3:
641: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
642: u1 = d1[i]; u2 = d2[i];
643: t1 = u1&0x38000000; t2 = u2&0x38000000; if ( t1&&t2 ) return 0;
644: t1 = u1& 0x7000000; t2 = u2& 0x7000000; if ( t1&&t2 ) return 0;
645: t1 = u1& 0xe00000; t2 = u2& 0xe00000; if ( t1&&t2 ) return 0;
646: t1 = u1& 0x1c0000; t2 = u2& 0x1c0000; if ( t1&&t2 ) return 0;
647: t1 = u1& 0x38000; t2 = u2& 0x38000; if ( t1&&t2 ) return 0;
648: t1 = u1& 0x7000; t2 = u2& 0x7000; if ( t1&&t2 ) return 0;
649: t1 = u1& 0xe00; t2 = u2& 0xe00; if ( t1&&t2 ) return 0;
650: t1 = u1& 0x1c0; t2 = u2& 0x1c0; if ( t1&&t2 ) return 0;
651: t1 = u1& 0x38; t2 = u2& 0x38; if ( t1&&t2 ) return 0;
652: t1 = u1& 0x7; t2 = u2& 0x7; if ( t1&&t2 ) return 0;
653: }
654: return 1;
655: break;
1.1 noro 656: case 4:
1.41 noro 657: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 658: u1 = d1[i]; u2 = d2[i];
659: t1 = u1&0xf0000000; t2 = u2&0xf0000000; if ( t1&&t2 ) return 0;
1.62 noro 660: t1 = u1& 0xf000000; t2 = u2& 0xf000000; if ( t1&&t2 ) return 0;
661: t1 = u1& 0xf00000; t2 = u2& 0xf00000; if ( t1&&t2 ) return 0;
662: t1 = u1& 0xf0000; t2 = u2& 0xf0000; if ( t1&&t2 ) return 0;
663: t1 = u1& 0xf000; t2 = u2& 0xf000; if ( t1&&t2 ) return 0;
664: t1 = u1& 0xf00; t2 = u2& 0xf00; if ( t1&&t2 ) return 0;
665: t1 = u1& 0xf0; t2 = u2& 0xf0; if ( t1&&t2 ) return 0;
666: t1 = u1& 0xf; t2 = u2& 0xf; if ( t1&&t2 ) return 0;
1.1 noro 667: }
668: return 1;
669: break;
670: case 6:
1.41 noro 671: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 672: u1 = d1[i]; u2 = d2[i];
673: t1 = u1&0x3f000000; t2 = u2&0x3f000000; if ( t1&&t2 ) return 0;
1.62 noro 674: t1 = u1& 0xfc0000; t2 = u2& 0xfc0000; if ( t1&&t2 ) return 0;
675: t1 = u1& 0x3f000; t2 = u2& 0x3f000; if ( t1&&t2 ) return 0;
676: t1 = u1& 0xfc0; t2 = u2& 0xfc0; if ( t1&&t2 ) return 0;
677: t1 = u1& 0x3f; t2 = u2& 0x3f; if ( t1&&t2 ) return 0;
1.1 noro 678: }
679: return 1;
680: break;
681: case 8:
1.41 noro 682: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 683: u1 = d1[i]; u2 = d2[i];
684: t1 = u1&0xff000000; t2 = u2&0xff000000; if ( t1&&t2 ) return 0;
1.62 noro 685: t1 = u1& 0xff0000; t2 = u2& 0xff0000; if ( t1&&t2 ) return 0;
686: t1 = u1& 0xff00; t2 = u2& 0xff00; if ( t1&&t2 ) return 0;
687: t1 = u1& 0xff; t2 = u2& 0xff; if ( t1&&t2 ) return 0;
1.1 noro 688: }
689: return 1;
690: break;
691: case 16:
1.41 noro 692: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 693: u1 = d1[i]; u2 = d2[i];
694: t1 = u1&0xffff0000; t2 = u2&0xffff0000; if ( t1&&t2 ) return 0;
1.62 noro 695: t1 = u1& 0xffff; t2 = u2& 0xffff; if ( t1&&t2 ) return 0;
1.1 noro 696: }
697: return 1;
698: break;
699: case 32:
1.41 noro 700: for ( i = nd_exporigin; i < nd_wpd; i++ )
1.1 noro 701: if ( d1[i] && d2[i] ) return 0;
702: return 1;
703: break;
704: default:
1.41 noro 705: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.1 noro 706: u1 = d1[i]; u2 = d2[i];
707: for ( j = 0; j < nd_epw; j++ ) {
708: if ( (u1&nd_mask0) && (u2&nd_mask0) ) return 0;
709: u1 >>= nd_bpe; u2 >>= nd_bpe;
710: }
711: }
712: return 1;
713: break;
714: }
1.65 noro 715: #else
716: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
717: u1 = d1[i]; u2 = d2[i];
718: for ( j = 0; j < nd_epw; j++ ) {
719: if ( (u1&nd_mask0) && (u2&nd_mask0) ) return 0;
720: u1 >>= nd_bpe; u2 >>= nd_bpe;
721: }
722: }
723: return 1;
724: #endif
1.1 noro 725: }
726:
1.61 noro 727: int ndl_check_bound2(int index,UINT *d2)
1.1 noro 728: {
1.61 noro 729: UINT u2;
730: UINT *d1;
1.5 noro 731: int i,j,ind,k;
1.1 noro 732:
1.5 noro 733: d1 = nd_bound[index];
734: ind = 0;
1.65 noro 735: #if USE_UNROLL
1.5 noro 736: switch ( nd_bpe ) {
1.62 noro 737: case 3:
738: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
739: u2 = d2[i];
740: if ( d1[ind++]+((u2>>27)&0x7) >= 0x8 ) return 1;
741: if ( d1[ind++]+((u2>>24)&0x7) >= 0x8 ) return 1;
742: if ( d1[ind++]+((u2>>21)&0x7) >= 0x8 ) return 1;
743: if ( d1[ind++]+((u2>>18)&0x7) >= 0x8 ) return 1;
744: if ( d1[ind++]+((u2>>15)&0x7) >= 0x8 ) return 1;
745: if ( d1[ind++]+((u2>>12)&0x7) >= 0x8 ) return 1;
746: if ( d1[ind++]+((u2>>9)&0x7) >= 0x8 ) return 1;
747: if ( d1[ind++]+((u2>>6)&0x7) >= 0x8 ) return 1;
748: if ( d1[ind++]+((u2>>3)&0x7) >= 0x8 ) return 1;
749: if ( d1[ind++]+(u2&0x7) >= 0x8 ) return 1;
750: }
751: return 0;
752: break;
1.5 noro 753: case 4:
1.41 noro 754: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 755: u2 = d2[i];
756: if ( d1[ind++]+((u2>>28)&0xf) >= 0x10 ) return 1;
757: if ( d1[ind++]+((u2>>24)&0xf) >= 0x10 ) return 1;
758: if ( d1[ind++]+((u2>>20)&0xf) >= 0x10 ) return 1;
759: if ( d1[ind++]+((u2>>16)&0xf) >= 0x10 ) return 1;
760: if ( d1[ind++]+((u2>>12)&0xf) >= 0x10 ) return 1;
761: if ( d1[ind++]+((u2>>8)&0xf) >= 0x10 ) return 1;
762: if ( d1[ind++]+((u2>>4)&0xf) >= 0x10 ) return 1;
763: if ( d1[ind++]+(u2&0xf) >= 0x10 ) return 1;
764: }
765: return 0;
766: break;
767: case 6:
1.41 noro 768: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 769: u2 = d2[i];
770: if ( d1[ind++]+((u2>>24)&0x3f) >= 0x40 ) return 1;
771: if ( d1[ind++]+((u2>>18)&0x3f) >= 0x40 ) return 1;
772: if ( d1[ind++]+((u2>>12)&0x3f) >= 0x40 ) return 1;
773: if ( d1[ind++]+((u2>>6)&0x3f) >= 0x40 ) return 1;
774: if ( d1[ind++]+(u2&0x3f) >= 0x40 ) return 1;
775: }
776: return 0;
777: break;
778: case 8:
1.41 noro 779: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 780: u2 = d2[i];
781: if ( d1[ind++]+((u2>>24)&0xff) >= 0x100 ) return 1;
782: if ( d1[ind++]+((u2>>16)&0xff) >= 0x100 ) return 1;
783: if ( d1[ind++]+((u2>>8)&0xff) >= 0x100 ) return 1;
784: if ( d1[ind++]+(u2&0xff) >= 0x100 ) return 1;
785: }
786: return 0;
787: break;
788: case 16:
1.41 noro 789: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 790: u2 = d2[i];
791: if ( d1[ind++]+((u2>>16)&0xffff) > 0x10000 ) return 1;
792: if ( d1[ind++]+(u2&0xffff) > 0x10000 ) return 1;
793: }
794: return 0;
795: break;
796: case 32:
1.41 noro 797: for ( i = nd_exporigin; i < nd_wpd; i++ )
1.5 noro 798: if ( d1[i]+d2[i]<d1[i] ) return 1;
799: return 0;
800: break;
801: default:
1.41 noro 802: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
1.5 noro 803: u2 = d2[i];
804: k = (nd_epw-1)*nd_bpe;
805: for ( j = 0; j < nd_epw; j++, k -= nd_bpe )
806: if ( d1[ind++]+((u2>>k)&nd_mask0) > nd_mask0 ) return 1;
807: }
808: return 0;
809: break;
810: }
1.65 noro 811: #else
812: for ( i = nd_exporigin; i < nd_wpd; i++ ) {
813: u2 = d2[i];
814: k = (nd_epw-1)*nd_bpe;
815: for ( j = 0; j < nd_epw; j++, k -= nd_bpe )
816: if ( d1[ind++]+((u2>>k)&nd_mask0) > nd_mask0 ) return 1;
817: }
818: return 0;
819: #endif
1.1 noro 820: }
821:
1.61 noro 822: INLINE int ndl_hash_value(UINT *d)
1.1 noro 823: {
824: int i;
825: int r;
826:
1.34 noro 827: r = 0;
1.41 noro 828: for ( i = 0; i < nd_wpd; i++ )
1.1 noro 829: r = ((r<<16)+d[i])%REDTAB_LEN;
830: return r;
831: }
832:
1.63 noro 833: INLINE int ndl_find_reducer(UINT *dg)
1.1 noro 834: {
1.13 noro 835: RHist r;
1.6 noro 836: int d,k,i;
1.1 noro 837:
1.63 noro 838: d = ndl_hash_value(dg);
1.13 noro 839: for ( r = nd_red[d], k = 0; r; r = NEXT(r), k++ ) {
1.34 noro 840: if ( ndl_equal(dg,DL(r)) ) {
1.1 noro 841: if ( k > 0 ) nd_notfirst++;
842: nd_found++;
1.13 noro 843: return r->index;
1.1 noro 844: }
845: }
1.13 noro 846: if ( Reverse )
847: for ( i = nd_psn-1; i >= 0; i-- ) {
848: r = nd_psh[i];
1.34 noro 849: if ( ndl_reducible(dg,DL(r)) ) {
1.13 noro 850: nd_create++;
1.34 noro 851: nd_append_red(dg,i);
1.13 noro 852: return i;
853: }
854: }
855: else
856: for ( i = 0; i < nd_psn; i++ ) {
857: r = nd_psh[i];
1.34 noro 858: if ( ndl_reducible(dg,DL(r)) ) {
1.13 noro 859: nd_create++;
1.34 noro 860: nd_append_red(dg,i);
1.13 noro 861: return i;
862: }
1.1 noro 863: }
1.6 noro 864: return -1;
1.1 noro 865: }
866:
1.63 noro 867: ND nd_merge(ND p1,ND p2)
868: {
869: int n,c;
870: int t,can,td1,td2;
871: ND r;
872: NM m1,m2,mr0,mr,s;
873:
874: if ( !p1 ) return p2;
875: else if ( !p2 ) return p1;
876: else {
877: can = 0;
878: for ( n = NV(p1), m1 = BDY(p1), m2 = BDY(p2), mr0 = 0; m1 && m2; ) {
879: c = DL_COMPARE(DL(m1),DL(m2));
880: switch ( c ) {
881: case 0:
882: s = m1; m1 = NEXT(m1);
883: can++; NEXTNM2(mr0,mr,s);
884: s = m2; m2 = NEXT(m2); FREENM(s);
885: break;
886: case 1:
887: s = m1; m1 = NEXT(m1); NEXTNM2(mr0,mr,s);
888: break;
889: case -1:
890: s = m2; m2 = NEXT(m2); NEXTNM2(mr0,mr,s);
891: break;
892: }
893: }
894: if ( !mr0 )
895: if ( m1 ) mr0 = m1;
896: else if ( m2 ) mr0 = m2;
897: else return 0;
898: else if ( m1 ) NEXT(mr) = m1;
899: else if ( m2 ) NEXT(mr) = m2;
900: else NEXT(mr) = 0;
901: BDY(p1) = mr0;
902: SG(p1) = MAX(SG(p1),SG(p2));
903: LEN(p1) = LEN(p1)+LEN(p2)-can;
904: FREEND(p2);
905: return p1;
906: }
907: }
908:
1.31 noro 909: ND nd_add(int mod,ND p1,ND p2)
1.1 noro 910: {
911: int n,c;
1.34 noro 912: int t,can,td1,td2;
1.1 noro 913: ND r;
914: NM m1,m2,mr0,mr,s;
915:
1.34 noro 916: if ( !p1 ) return p2;
917: else if ( !p2 ) return p1;
1.71 noro 918: else if ( mod == -1 ) return nd_add_sf(p1,p2);
1.112 ! noro 919: else if ( !mod ) return nd_add_z(p1,p2);
1.1 noro 920: else {
1.30 noro 921: can = 0;
1.1 noro 922: for ( n = NV(p1), m1 = BDY(p1), m2 = BDY(p2), mr0 = 0; m1 && m2; ) {
1.34 noro 923: c = DL_COMPARE(DL(m1),DL(m2));
1.1 noro 924: switch ( c ) {
925: case 0:
1.19 noro 926: t = ((CM(m1))+(CM(m2))) - mod;
1.34 noro 927: if ( t < 0 ) t += mod;
1.1 noro 928: s = m1; m1 = NEXT(m1);
929: if ( t ) {
1.34 noro 930: can++; NEXTNM2(mr0,mr,s); CM(mr) = (t);
1.17 noro 931: } else {
1.34 noro 932: can += 2; FREENM(s);
1.17 noro 933: }
934: s = m2; m2 = NEXT(m2); FREENM(s);
1.95 noro 935: break;
936: case 1:
937: s = m1; m1 = NEXT(m1); NEXTNM2(mr0,mr,s);
938: break;
939: case -1:
940: s = m2; m2 = NEXT(m2); NEXTNM2(mr0,mr,s);
941: break;
942: }
943: }
944: if ( !mr0 )
945: if ( m1 ) mr0 = m1;
946: else if ( m2 ) mr0 = m2;
947: else return 0;
948: else if ( m1 ) NEXT(mr) = m1;
949: else if ( m2 ) NEXT(mr) = m2;
950: else NEXT(mr) = 0;
951: BDY(p1) = mr0;
952: SG(p1) = MAX(SG(p1),SG(p2));
953: LEN(p1) = LEN(p1)+LEN(p2)-can;
954: FREEND(p2);
955: return p1;
956: }
957: }
958:
959: /* XXX on opteron, the inlined manipulation of destructive additon of
960: * two NM seems to make gcc optimizer get confused, so the part is
961: * done in a function.
962: */
963:
1.112 ! noro 964: int nm_destructive_add_z(NM *m1,NM *m2,NM *mr0,NM *mr)
1.95 noro 965: {
966: NM s;
1.112 ! noro 967: Z t;
1.95 noro 968: int can;
969:
1.112 ! noro 970: t = addz(CZ(*m1),CZ(*m2));
1.95 noro 971: s = *m1; *m1 = NEXT(*m1);
972: if ( t ) {
1.112 ! noro 973: can = 1; NEXTNM2(*mr0,*mr,s); CZ(*mr) = (t);
1.95 noro 974: } else {
975: can = 2; FREENM(s);
976: }
977: s = *m2; *m2 = NEXT(*m2); FREENM(s);
978: return can;
979: }
980:
1.112 ! noro 981: ND nd_add_z(ND p1,ND p2)
1.95 noro 982: {
983: int n,c,can;
984: ND r;
985: NM m1,m2,mr0,mr,s;
1.112 ! noro 986: Z t;
1.95 noro 987:
988: if ( !p1 ) return p2;
989: else if ( !p2 ) return p1;
990: else {
991: can = 0;
992: for ( n = NV(p1), m1 = BDY(p1), m2 = BDY(p2), mr0 = 0; m1 && m2; ) {
993: c = DL_COMPARE(DL(m1),DL(m2));
994: switch ( c ) {
995: case 0:
996: #if defined(__x86_64__)
1.112 ! noro 997: can += nm_destructive_add_z(&m1,&m2,&mr0,&mr);
1.95 noro 998: #else
1.112 ! noro 999: t = addz(CZ(m1),CZ(m2));
1.95 noro 1000: s = m1; m1 = NEXT(m1);
1001: if ( t ) {
1.112 ! noro 1002: can++; NEXTNM2(mr0,mr,s); CZ(mr) = (t);
1.95 noro 1003: } else {
1004: can += 2; FREENM(s);
1005: }
1006: s = m2; m2 = NEXT(m2); FREENM(s);
1007: #endif
1.17 noro 1008: break;
1009: case 1:
1010: s = m1; m1 = NEXT(m1); NEXTNM2(mr0,mr,s);
1011: break;
1012: case -1:
1013: s = m2; m2 = NEXT(m2); NEXTNM2(mr0,mr,s);
1014: break;
1015: }
1016: }
1017: if ( !mr0 )
1.34 noro 1018: if ( m1 ) mr0 = m1;
1019: else if ( m2 ) mr0 = m2;
1020: else return 0;
1021: else if ( m1 ) NEXT(mr) = m1;
1022: else if ( m2 ) NEXT(mr) = m2;
1023: else NEXT(mr) = 0;
1.17 noro 1024: BDY(p1) = mr0;
1025: SG(p1) = MAX(SG(p1),SG(p2));
1.31 noro 1026: LEN(p1) = LEN(p1)+LEN(p2)-can;
1.17 noro 1027: FREEND(p2);
1028: return p1;
1029: }
1030: }
1031:
1.71 noro 1032: ND nd_add_sf(ND p1,ND p2)
1033: {
1034: int n,c,can;
1035: ND r;
1036: NM m1,m2,mr0,mr,s;
1037: int t;
1038:
1039: if ( !p1 ) return p2;
1040: else if ( !p2 ) return p1;
1041: else {
1042: can = 0;
1043: for ( n = NV(p1), m1 = BDY(p1), m2 = BDY(p2), mr0 = 0; m1 && m2; ) {
1044: c = DL_COMPARE(DL(m1),DL(m2));
1045: switch ( c ) {
1046: case 0:
1047: t = _addsf(CM(m1),CM(m2));
1048: s = m1; m1 = NEXT(m1);
1049: if ( t ) {
1050: can++; NEXTNM2(mr0,mr,s); CM(mr) = (t);
1051: } else {
1052: can += 2; FREENM(s);
1053: }
1054: s = m2; m2 = NEXT(m2); FREENM(s);
1055: break;
1056: case 1:
1057: s = m1; m1 = NEXT(m1); NEXTNM2(mr0,mr,s);
1058: break;
1059: case -1:
1060: s = m2; m2 = NEXT(m2); NEXTNM2(mr0,mr,s);
1061: break;
1062: }
1063: }
1064: if ( !mr0 )
1065: if ( m1 ) mr0 = m1;
1066: else if ( m2 ) mr0 = m2;
1067: else return 0;
1068: else if ( m1 ) NEXT(mr) = m1;
1069: else if ( m2 ) NEXT(mr) = m2;
1070: else NEXT(mr) = 0;
1071: BDY(p1) = mr0;
1072: SG(p1) = MAX(SG(p1),SG(p2));
1073: LEN(p1) = LEN(p1)+LEN(p2)-can;
1074: FREEND(p2);
1075: return p1;
1076: }
1077: }
1078:
1.1 noro 1079: /* ret=1 : success, ret=0 : overflow */
1.112 ! noro 1080: int nd_nf(int mod,ND g,NDV *ps,int full,Q *dn,ND *rp)
1.1 noro 1081: {
1.11 noro 1082: ND d;
1.1 noro 1083: NM m,mrd,tail;
1.7 noro 1084: NM mul;
1.10 noro 1085: int n,sugar,psugar,sugar0,stat,index;
1.30 noro 1086: int c,c1,c2,dummy;
1.17 noro 1087: RHist h;
1.11 noro 1088: NDV p,red;
1.112 ! noro 1089: Z cg,cred,gcd;
! 1090: Q tq,qq;
1.21 noro 1091: double hmag;
1.1 noro 1092:
1.112 ! noro 1093: if ( dn && !mod )
! 1094: *dn = ONE;
1.1 noro 1095: if ( !g ) {
1096: *rp = 0;
1097: return 1;
1098: }
1.112 ! noro 1099: if ( !mod ) hmag = ((double)z_mag(HCZ(g)))*nd_scale;
1.21 noro 1100:
1.14 noro 1101: sugar0 = sugar = SG(g);
1.1 noro 1102: n = NV(g);
1.61 noro 1103: mul = (NM)ALLOCA(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT));
1.1 noro 1104: for ( d = 0; g; ) {
1.63 noro 1105: index = ndl_find_reducer(HDL(g));
1.6 noro 1106: if ( index >= 0 ) {
1.17 noro 1107: h = nd_psh[index];
1108: ndl_sub(HDL(g),DL(h),DL(mul));
1.14 noro 1109: if ( ndl_check_bound2(index,DL(mul)) ) {
1.6 noro 1110: nd_free(g); nd_free(d);
1111: return 0;
1112: }
1.77 noro 1113: if ( nd_demand )
1114: p = ndv_load(index);
1115: else
1116: p = ps[index];
1.71 noro 1117: if ( mod == -1 )
1118: CM(mul) = _mulsf(_invsf(HCM(p)),_chsgnsf(HCM(g)));
1119: else if ( mod ) {
1.19 noro 1120: c1 = invm(HCM(p),mod); c2 = mod-HCM(g);
1121: DMAR(c1,c2,0,mod,c); CM(mul) = c;
1.16 noro 1122: } else {
1.112 ! noro 1123: gcd = gcdz_cofactor(HCZ(g),HCZ(p),&cg,&cred);
! 1124: CZ(mul) = chsgnz(cg);
! 1125: nd_mul_c_z(d,cred);
! 1126: nd_mul_c_z(g,cred);
1.69 noro 1127: if ( dn ) {
1.112 ! noro 1128: mulq(*dn,ztoq(cred),&tq); *dn = tq;
1.69 noro 1129: }
1.16 noro 1130: }
1.55 noro 1131: g = nd_add(mod,g,ndv_mul_nm(mod,mul,p));
1.112 ! noro 1132:
1.34 noro 1133: sugar = MAX(sugar,SG(p)+TD(DL(mul)));
1.112 ! noro 1134: if ( !mod && hmag && g && ((double)(z_mag(HCZ(g))) > hmag) ) {
! 1135: tq = ztoq(HCZ(g));
1.21 noro 1136: nd_removecont2(d,g);
1.69 noro 1137: if ( dn ) {
1.112 ! noro 1138: divq(tq,ztoq(HCZ(g)),&qq); divq(*dn,qq,&tq); *dn = tq;
1.69 noro 1139: }
1.112 ! noro 1140: hmag = ((double)z_mag((P)HCZ(g)))*nd_scale;
1.21 noro 1141: }
1.1 noro 1142: } else if ( !full ) {
1143: *rp = g;
1144: return 1;
1145: } else {
1146: m = BDY(g);
1147: if ( NEXT(m) ) {
1.34 noro 1148: BDY(g) = NEXT(m); NEXT(m) = 0; LEN(g)--;
1.1 noro 1149: } else {
1150: FREEND(g); g = 0;
1151: }
1152: if ( d ) {
1.34 noro 1153: NEXT(tail)=m; tail=m; LEN(d)++;
1.1 noro 1154: } else {
1.34 noro 1155: MKND(n,m,1,d); tail = BDY(d);
1.1 noro 1156: }
1157: }
1158: }
1.34 noro 1159: if ( d ) SG(d) = sugar;
1.1 noro 1160: *rp = d;
1161: return 1;
1162: }
1.28 noro 1163:
1.53 noro 1164: int nd_nf_pbucket(int mod,ND g,NDV *ps,int full,ND *rp)
1.25 noro 1165: {
1166: int hindex,index;
1167: NDV p;
1168: ND u,d,red;
1169: NODE l;
1.31 noro 1170: NM mul,m,mrd,tail;
1.25 noro 1171: int sugar,psugar,n,h_reducible;
1172: PGeoBucket bucket;
1173: int c,c1,c2;
1.112 ! noro 1174: Z cg,cred,gcd,zzz;
1.25 noro 1175: RHist h;
1.28 noro 1176: double hmag,gmag;
1.77 noro 1177: int count = 0;
1178: int hcount = 0;
1.25 noro 1179:
1180: if ( !g ) {
1181: *rp = 0;
1182: return 1;
1183: }
1184: sugar = SG(g);
1185: n = NV(g);
1.112 ! noro 1186: if ( !mod ) hmag = ((double)z_mag(HCZ(g)))*nd_scale;
1.25 noro 1187: bucket = create_pbucket();
1.31 noro 1188: add_pbucket(mod,bucket,g);
1.25 noro 1189: d = 0;
1.61 noro 1190: mul = (NM)ALLOCA(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT));
1.25 noro 1191: while ( 1 ) {
1.112 ! noro 1192: hindex = mod?head_pbucket(mod,bucket):head_pbucket_z(bucket);
1.25 noro 1193: if ( hindex < 0 ) {
1.77 noro 1194: if ( DP_Print > 3 ) printf("(%d %d)",count,hcount);
1.34 noro 1195: if ( d ) SG(d) = sugar;
1.25 noro 1196: *rp = d;
1197: return 1;
1198: }
1199: g = bucket->body[hindex];
1.63 noro 1200: index = ndl_find_reducer(HDL(g));
1.25 noro 1201: if ( index >= 0 ) {
1.77 noro 1202: count++;
1203: if ( !d ) hcount++;
1.25 noro 1204: h = nd_psh[index];
1205: ndl_sub(HDL(g),DL(h),DL(mul));
1206: if ( ndl_check_bound2(index,DL(mul)) ) {
1.26 noro 1207: nd_free(d);
1.25 noro 1208: free_pbucket(bucket);
1209: *rp = 0;
1210: return 0;
1211: }
1.53 noro 1212: p = ps[index];
1.71 noro 1213: if ( mod == -1 )
1214: CM(mul) = _mulsf(_invsf(HCM(p)),_chsgnsf(HCM(g)));
1215: else if ( mod ) {
1.25 noro 1216: c1 = invm(HCM(p),mod); c2 = mod-HCM(g);
1217: DMAR(c1,c2,0,mod,c); CM(mul) = c;
1218: } else {
1.112 ! noro 1219: gcd = gcdz_cofactor(HCZ(g),HCZ(p),&cg,&cred);
! 1220: CZ(mul) = chsgnz(cg);
! 1221: nd_mul_c_z(d,cred);
! 1222: mulz_pbucket(bucket,cred);
1.26 noro 1223: g = bucket->body[hindex];
1.112 ! noro 1224: gmag = (double)z_mag(HCZ(g));
1.25 noro 1225: }
1.55 noro 1226: red = ndv_mul_nm(mod,mul,p);
1.25 noro 1227: bucket->body[hindex] = nd_remove_head(g);
1228: red = nd_remove_head(red);
1.31 noro 1229: add_pbucket(mod,bucket,red);
1.34 noro 1230: psugar = SG(p)+TD(DL(mul));
1231: sugar = MAX(sugar,psugar);
1.28 noro 1232: if ( !mod && hmag && (gmag > hmag) ) {
1233: g = normalize_pbucket(mod,bucket);
1234: if ( !g ) {
1.34 noro 1235: if ( d ) SG(d) = sugar;
1.28 noro 1236: *rp = d;
1237: return 1;
1238: }
1239: nd_removecont2(d,g);
1.112 ! noro 1240: hmag = ((double)z_mag(HCZ(g)))*nd_scale;
1.31 noro 1241: add_pbucket(mod,bucket,g);
1.28 noro 1242: }
1.25 noro 1243: } else if ( !full ) {
1244: g = normalize_pbucket(mod,bucket);
1.34 noro 1245: if ( g ) SG(g) = sugar;
1.25 noro 1246: *rp = g;
1247: return 1;
1248: } else {
1249: m = BDY(g);
1250: if ( NEXT(m) ) {
1.34 noro 1251: BDY(g) = NEXT(m); NEXT(m) = 0; LEN(g)--;
1.25 noro 1252: } else {
1253: FREEND(g); g = 0;
1254: }
1255: bucket->body[hindex] = g;
1256: NEXT(m) = 0;
1257: if ( d ) {
1.34 noro 1258: NEXT(tail)=m; tail=m; LEN(d)++;
1.25 noro 1259: } else {
1.34 noro 1260: MKND(n,m,1,d); tail = BDY(d);
1.25 noro 1261: }
1262: }
1263: }
1264: }
1.27 noro 1265:
1.61 noro 1266: /* input : list of NDV, cand : list of NDV */
1.28 noro 1267:
1.61 noro 1268: int ndv_check_candidate(NODE input,int obpe,int oadv,EPOS oepos,NODE cand)
1.28 noro 1269: {
1270: int n,i,stat;
1271: ND nf,d;
1.61 noro 1272: NDV r;
1.45 noro 1273: NODE t,s;
1274:
1.61 noro 1275: ndv_setup(0,0,cand);
1.31 noro 1276: n = length(cand);
1.28 noro 1277:
1278: /* membercheck : list is a subset of Id(cand) ? */
1279: for ( t = input; t; t = NEXT(t) ) {
1.45 noro 1280: again:
1.61 noro 1281: if ( nd_bpe > obpe )
1282: r = ndv_dup_realloc((NDV)BDY(t),obpe,oadv,oepos);
1283: else
1284: r = (NDV)BDY(t);
1285: d = ndvtond(0,r);
1.69 noro 1286: stat = nd_nf(0,d,nd_ps,0,0,&nf);
1.45 noro 1287: if ( !stat ) {
1.103 noro 1288: nd_reconstruct(0,0);
1.45 noro 1289: goto again;
1290: } else if ( nf ) return 0;
1.71 noro 1291: if ( DP_Print ) { printf("."); fflush(stdout); }
1.28 noro 1292: }
1.71 noro 1293: if ( DP_Print ) { printf("\n"); }
1.28 noro 1294: /* gbcheck : cand is a GB of Id(cand) ? */
1.77 noro 1295: if ( !nd_gb(0,0,1) ) return 0;
1.28 noro 1296: /* XXX */
1.23 noro 1297: return 1;
1298: }
1.1 noro 1299:
1300: ND nd_remove_head(ND p)
1301: {
1302: NM m;
1303:
1304: m = BDY(p);
1305: if ( !NEXT(m) ) {
1.34 noro 1306: FREEND(p); p = 0;
1.31 noro 1307: } else {
1.34 noro 1308: BDY(p) = NEXT(m); LEN(p)--;
1.31 noro 1309: }
1.1 noro 1310: FREENM(m);
1311: return p;
1312: }
1313:
1.69 noro 1314: ND nd_separate_head(ND p,ND *head)
1315: {
1316: NM m,m0;
1317: ND r;
1318:
1319: m = BDY(p);
1320: if ( !NEXT(m) ) {
1321: *head = p; p = 0;
1322: } else {
1323: m0 = m;
1324: BDY(p) = NEXT(m); LEN(p)--;
1325: NEXT(m0) = 0;
1326: MKND(NV(p),m0,1,r);
1327: *head = r;
1328: }
1329: return p;
1330: }
1331:
1.1 noro 1332: PGeoBucket create_pbucket()
1333: {
1334: PGeoBucket g;
1335:
1336: g = CALLOC(1,sizeof(struct oPGeoBucket));
1337: g->m = -1;
1338: return g;
1339: }
1340:
1.25 noro 1341: void free_pbucket(PGeoBucket b) {
1342: int i;
1343:
1.26 noro 1344: for ( i = 0; i <= b->m; i++ )
1.25 noro 1345: if ( b->body[i] ) {
1346: nd_free(b->body[i]);
1347: b->body[i] = 0;
1348: }
1349: GC_free(b);
1350: }
1351:
1.63 noro 1352: void add_pbucket_symbolic(PGeoBucket g,ND d)
1353: {
1354: int l,i,k,m;
1355:
1356: if ( !d )
1357: return;
1358: l = LEN(d);
1359: for ( k = 0, m = 1; l > m; k++, m <<= 1 );
1360: /* 2^(k-1) < l <= 2^k (=m) */
1361: d = nd_merge(g->body[k],d);
1362: for ( ; d && LEN(d) > m; k++, m <<= 1 ) {
1363: g->body[k] = 0;
1364: d = nd_merge(g->body[k+1],d);
1365: }
1366: g->body[k] = d;
1367: g->m = MAX(g->m,k);
1368: }
1369:
1.31 noro 1370: void add_pbucket(int mod,PGeoBucket g,ND d)
1.1 noro 1371: {
1.31 noro 1372: int l,i,k,m;
1.1 noro 1373:
1.31 noro 1374: if ( !d )
1375: return;
1376: l = LEN(d);
1.29 noro 1377: for ( k = 0, m = 1; l > m; k++, m <<= 1 );
1378: /* 2^(k-1) < l <= 2^k (=m) */
1.31 noro 1379: d = nd_add(mod,g->body[k],d);
1380: for ( ; d && LEN(d) > m; k++, m <<= 1 ) {
1.1 noro 1381: g->body[k] = 0;
1.31 noro 1382: d = nd_add(mod,g->body[k+1],d);
1.1 noro 1383: }
1384: g->body[k] = d;
1385: g->m = MAX(g->m,k);
1386: }
1387:
1.112 ! noro 1388: void mulz_pbucket(PGeoBucket g,Z c)
1.26 noro 1389: {
1390: int k;
1391:
1392: for ( k = 0; k <= g->m; k++ )
1.112 ! noro 1393: nd_mul_c_z(g->body[k],c);
1.26 noro 1394: }
1395:
1.63 noro 1396: NM remove_head_pbucket_symbolic(PGeoBucket g)
1397: {
1398: int j,i,k,c;
1399: NM head;
1400:
1401: k = g->m;
1402: j = -1;
1403: for ( i = 0; i <= k; i++ ) {
1404: if ( !g->body[i] ) continue;
1405: if ( j < 0 ) j = i;
1406: else {
1407: c = DL_COMPARE(HDL(g->body[i]),HDL(g->body[j]));
1408: if ( c > 0 )
1409: j = i;
1410: else if ( c == 0 )
1411: g->body[i] = nd_remove_head(g->body[i]);
1412: }
1413: }
1414: if ( j < 0 ) return 0;
1415: else {
1416: head = BDY(g->body[j]);
1417: if ( !NEXT(head) ) {
1418: FREEND(g->body[j]);
1419: g->body[j] = 0;
1420: } else {
1421: BDY(g->body[j]) = NEXT(head);
1422: LEN(g->body[j])--;
1423: }
1424: return head;
1425: }
1426: }
1427:
1.19 noro 1428: int head_pbucket(int mod,PGeoBucket g)
1.1 noro 1429: {
1430: int j,i,c,k,nv,sum;
1.61 noro 1431: UINT *di,*dj;
1.1 noro 1432: ND gi,gj;
1433:
1434: k = g->m;
1435: while ( 1 ) {
1436: j = -1;
1437: for ( i = 0; i <= k; i++ ) {
1438: if ( !(gi = g->body[i]) )
1439: continue;
1440: if ( j < 0 ) {
1441: j = i;
1442: gj = g->body[j];
1443: dj = HDL(gj);
1.14 noro 1444: sum = HCM(gj);
1.1 noro 1445: } else {
1.34 noro 1446: c = DL_COMPARE(HDL(gi),dj);
1.1 noro 1447: if ( c > 0 ) {
1.34 noro 1448: if ( sum ) HCM(gj) = sum;
1449: else g->body[j] = nd_remove_head(gj);
1.1 noro 1450: j = i;
1451: gj = g->body[j];
1452: dj = HDL(gj);
1.14 noro 1453: sum = HCM(gj);
1.1 noro 1454: } else if ( c == 0 ) {
1.71 noro 1455: if ( mod == -1 )
1456: sum = _addsf(sum,HCM(gi));
1457: else {
1458: sum = sum+HCM(gi)-mod;
1459: if ( sum < 0 ) sum += mod;
1460: }
1.1 noro 1461: g->body[i] = nd_remove_head(gi);
1462: }
1463: }
1464: }
1.34 noro 1465: if ( j < 0 ) return -1;
1.1 noro 1466: else if ( sum ) {
1.14 noro 1467: HCM(gj) = sum;
1.26 noro 1468: return j;
1.31 noro 1469: } else
1.26 noro 1470: g->body[j] = nd_remove_head(gj);
1471: }
1472: }
1473:
1.112 ! noro 1474: int head_pbucket_z(PGeoBucket g)
1.26 noro 1475: {
1476: int j,i,c,k,nv;
1.112 ! noro 1477: Z sum,t;
1.26 noro 1478: ND gi,gj;
1479:
1480: k = g->m;
1481: while ( 1 ) {
1482: j = -1;
1483: for ( i = 0; i <= k; i++ ) {
1.34 noro 1484: if ( !(gi = g->body[i]) ) continue;
1.26 noro 1485: if ( j < 0 ) {
1486: j = i;
1487: gj = g->body[j];
1.112 ! noro 1488: sum = HCZ(gj);
1.26 noro 1489: } else {
1490: nv = NV(gi);
1.34 noro 1491: c = DL_COMPARE(HDL(gi),HDL(gj));
1.26 noro 1492: if ( c > 0 ) {
1.112 ! noro 1493: if ( sum ) HCZ(gj) = sum;
1.34 noro 1494: else g->body[j] = nd_remove_head(gj);
1.26 noro 1495: j = i;
1496: gj = g->body[j];
1.112 ! noro 1497: sum = HCZ(gj);
1.26 noro 1498: } else if ( c == 0 ) {
1.112 ! noro 1499: sum = addz(sum,HCZ(gi));
1.26 noro 1500: g->body[i] = nd_remove_head(gi);
1501: }
1502: }
1503: }
1.34 noro 1504: if ( j < 0 ) return -1;
1.26 noro 1505: else if ( sum ) {
1.112 ! noro 1506: HCZ(gj) = sum;
1.1 noro 1507: return j;
1.31 noro 1508: } else
1.1 noro 1509: g->body[j] = nd_remove_head(gj);
1510: }
1511: }
1512:
1.25 noro 1513: ND normalize_pbucket(int mod,PGeoBucket g)
1.1 noro 1514: {
1.31 noro 1515: int i;
1.1 noro 1516: ND r,t;
1517:
1518: r = 0;
1.28 noro 1519: for ( i = 0; i <= g->m; i++ ) {
1.31 noro 1520: r = nd_add(mod,r,g->body[i]);
1.28 noro 1521: g->body[i] = 0;
1522: }
1523: g->m = -1;
1.1 noro 1524: return r;
1525: }
1526:
1.92 noro 1527: void do_diagonalize(int sugar,int m)
1528: {
1529: int i,nh,stat;
1530: NODE r,g,t;
1531: ND h,nf,s,head;
1532: NDV nfv;
1.112 ! noro 1533: Z q,num,den;
! 1534: Q dn;
1.92 noro 1535:
1536: for ( i = nd_psn-1; i >= 0 && SG(nd_psh[i]) == sugar; i-- ) {
1537: if ( nd_demand )
1538: nfv = ndv_load(i);
1539: else
1540: nfv = nd_ps[i];
1541: s = ndvtond(m,nfv);
1542: s = nd_separate_head(s,&head);
1543: nd_nf(m,s,nd_ps,1,&dn,&nf);
1544: if ( !m ) {
1.112 ! noro 1545: HCZ(head) = mulz(HCZ(head),simpz((Z)NM(dn)));
! 1546: if ( SGN(dn) < 0 ) HCZ(head) = chsgnz(HCZ(head));
! 1547: if ( DN(dn) )
! 1548: nd_mul_c_z(nf,simpz((Z)DN(dn)));
1.92 noro 1549: }
1550: nf = nd_add(m,head,nf);
1551: ndv_free(nfv);
1552: nd_removecont(m,nf);
1553: nfv = ndtondv(m,nf);
1554: nd_free(nf);
1555: nd_bound[i] = ndv_compute_bound(nfv);
1556: if ( nd_demand ) {
1557: ndv_save(nfv,i);
1558: ndv_free(nfv);
1559: } else
1560: nd_ps[i] = nfv;
1561: }
1562: }
1563:
1.27 noro 1564: /* return value = 0 => input is not a GB */
1565:
1.77 noro 1566: NODE nd_gb(int m,int ishomo,int checkonly)
1.1 noro 1567: {
1568: int i,nh,sugar,stat;
1.23 noro 1569: NODE r,g,t;
1.1 noro 1570: ND_pairs d;
1571: ND_pairs l;
1.77 noro 1572: ND h,nf,s,head;
1.63 noro 1573: NDV nfv;
1.1 noro 1574:
1.23 noro 1575: g = 0; d = 0;
1576: for ( i = 0; i < nd_psn; i++ ) {
1.1 noro 1577: d = update_pairs(d,g,i);
1578: g = update_base(g,i);
1579: }
1580: sugar = 0;
1581: while ( d ) {
1582: again:
1583: l = nd_minp(d,&d);
1.14 noro 1584: if ( SG(l) != sugar ) {
1.92 noro 1585: if ( ishomo ) do_diagonalize(sugar,m);
1586:
1.14 noro 1587: sugar = SG(l);
1.71 noro 1588: if ( DP_Print ) fprintf(asir_out,"%d",sugar);
1.1 noro 1589: }
1.53 noro 1590: stat = nd_sp(m,0,l,&h);
1.1 noro 1591: if ( !stat ) {
1592: NEXT(l) = d; d = l;
1.103 noro 1593: d = nd_reconstruct(0,d);
1.1 noro 1594: goto again;
1595: }
1.41 noro 1596: #if USE_GEOBUCKET
1.69 noro 1597: stat = m?nd_nf_pbucket(m,h,nd_ps,!Top,&nf):nd_nf(m,h,nd_ps,!Top,0,&nf);
1.41 noro 1598: #else
1.69 noro 1599: stat = nd_nf(m,h,nd_ps,!Top,0,&nf);
1.41 noro 1600: #endif
1.1 noro 1601: if ( !stat ) {
1602: NEXT(l) = d; d = l;
1.103 noro 1603: d = nd_reconstruct(0,d);
1.1 noro 1604: goto again;
1605: } else if ( nf ) {
1.27 noro 1606: if ( checkonly ) return 0;
1.71 noro 1607: if ( DP_Print ) { printf("+"); fflush(stdout); }
1.63 noro 1608: nd_removecont(m,nf);
1609: nfv = ndtondv(m,nf); nd_free(nf);
1.77 noro 1610: nh = ndv_newps(m,nfv,0);
1.1 noro 1611: d = update_pairs(d,g,nh);
1612: g = update_base(g,nh);
1613: FREENDP(l);
1614: } else {
1.71 noro 1615: if ( DP_Print ) { printf("."); fflush(stdout); }
1.1 noro 1616: FREENDP(l);
1617: }
1618: }
1.77 noro 1619: if ( nd_demand )
1620: for ( t = g; t; t = NEXT(t) )
1621: BDY(t) = (pointer)ndv_load((int)BDY(t));
1622: else
1623: for ( t = g; t; t = NEXT(t) )
1624: BDY(t) = (pointer)nd_ps[(int)BDY(t)];
1.1 noro 1625: return g;
1626: }
1627:
1.92 noro 1628: void do_diagonalize_trace(int sugar,int m)
1.91 noro 1629: {
1630: int i,nh,stat;
1631: NODE r,g,t;
1632: ND h,nf,nfq,s,head;
1633: NDV nfv,nfqv;
1.112 ! noro 1634: Q dn;
1.91 noro 1635:
1636: for ( i = nd_psn-1; i >= 0 && SG(nd_psh[i]) == sugar; i-- ) {
1637: /* for nd_ps */
1638: s = ndvtond(m,nd_ps[i]);
1639: s = nd_separate_head(s,&head);
1640: nd_nf_pbucket(m,s,nd_ps,1,&nf);
1641: nf = nd_add(m,head,nf);
1642: ndv_free(nd_ps[i]);
1643: nd_ps[i] = ndtondv(m,nf);
1644: nd_free(nf);
1645:
1646: /* for nd_ps_trace */
1647: if ( nd_demand )
1648: nfv = ndv_load(i);
1649: else
1650: nfv = nd_ps_trace[i];
1651: s = ndvtond(0,nfv);
1652: s = nd_separate_head(s,&head);
1653: nd_nf(0,s,nd_ps_trace,1,&dn,&nf);
1.112 ! noro 1654: HCZ(head) = mulz(HCZ(head),(Z)NM(dn));
! 1655: if ( SGN(dn) < 0 ) HCZ(head) = chsgnz(HCZ(head));
! 1656: if ( DN(dn) )
! 1657: nd_mul_c_z(nf,(Z)DN(dn));
1.91 noro 1658: nf = nd_add(0,head,nf);
1659: ndv_free(nfv);
1660: nd_removecont(0,nf);
1661: nfv = ndtondv(0,nf);
1662: nd_free(nf);
1663: nd_bound[i] = ndv_compute_bound(nfv);
1664: if ( nd_demand ) {
1665: ndv_save(nfv,i);
1666: ndv_free(nfv);
1667: } else
1668: nd_ps_trace[i] = nfv;
1669: }
1670: }
1671:
1.82 noro 1672: NODE nd_gb_trace(int m,int ishomo)
1.20 noro 1673: {
1674: int i,nh,sugar,stat;
1.23 noro 1675: NODE r,g,t;
1.20 noro 1676: ND_pairs d;
1677: ND_pairs l;
1.82 noro 1678: ND h,nf,nfq,s,head;
1.63 noro 1679: NDV nfv,nfqv;
1.20 noro 1680:
1.23 noro 1681: g = 0; d = 0;
1682: for ( i = 0; i < nd_psn; i++ ) {
1.20 noro 1683: d = update_pairs(d,g,i);
1684: g = update_base(g,i);
1685: }
1686: sugar = 0;
1687: while ( d ) {
1688: again:
1689: l = nd_minp(d,&d);
1690: if ( SG(l) != sugar ) {
1.93 noro 1691: if ( ishomo ) do_diagonalize_trace(sugar,m);
1.20 noro 1692: sugar = SG(l);
1.71 noro 1693: if ( DP_Print ) fprintf(asir_out,"%d",sugar);
1.20 noro 1694: }
1.53 noro 1695: stat = nd_sp(m,0,l,&h);
1.20 noro 1696: if ( !stat ) {
1697: NEXT(l) = d; d = l;
1.103 noro 1698: d = nd_reconstruct(1,d);
1.20 noro 1699: goto again;
1700: }
1.41 noro 1701: #if USE_GEOBUCKET
1.53 noro 1702: stat = nd_nf_pbucket(m,h,nd_ps,!Top,&nf);
1.41 noro 1703: #else
1.69 noro 1704: stat = nd_nf(m,h,nd_ps,!Top,0,&nf);
1.41 noro 1705: #endif
1.20 noro 1706: if ( !stat ) {
1707: NEXT(l) = d; d = l;
1.103 noro 1708: d = nd_reconstruct(1,d);
1.20 noro 1709: goto again;
1710: } else if ( nf ) {
1.77 noro 1711: if ( nd_demand ) {
1712: nfqv = ndv_load(nd_psn);
1713: nfq = ndvtond(0,nfqv);
1714: } else
1715: nfq = 0;
1716: if ( !nfq ) {
1717: if ( !nd_sp(0,1,l,&h) || !nd_nf(0,h,nd_ps_trace,!Top,0,&nfq) ) {
1718: NEXT(l) = d; d = l;
1.103 noro 1719: d = nd_reconstruct(1,d);
1.77 noro 1720: goto again;
1721: }
1.70 noro 1722: }
1.20 noro 1723: if ( nfq ) {
1.70 noro 1724: /* m|HC(nfq) => failure */
1.112 ! noro 1725: if ( !remzi(HCZ(nfq),m) ) return 0;
1.63 noro 1726:
1.71 noro 1727: if ( DP_Print ) { printf("+"); fflush(stdout); }
1.63 noro 1728: nd_removecont(m,nf); nfv = ndtondv(m,nf); nd_free(nf);
1729: nd_removecont(0,nfq); nfqv = ndtondv(0,nfq); nd_free(nfq);
1.77 noro 1730: nh = ndv_newps(0,nfv,nfqv);
1.20 noro 1731: d = update_pairs(d,g,nh);
1732: g = update_base(g,nh);
1733: } else {
1.71 noro 1734: if ( DP_Print ) { printf("*"); fflush(stdout); }
1.20 noro 1735: }
1736: } else {
1.71 noro 1737: if ( DP_Print ) { printf("."); fflush(stdout); }
1.20 noro 1738: }
1739: FREENDP(l);
1740: }
1.77 noro 1741: if ( nd_demand )
1742: for ( t = g; t; t = NEXT(t) )
1743: BDY(t) = (pointer)ndv_load((int)BDY(t));
1744: else
1745: for ( t = g; t; t = NEXT(t) )
1746: BDY(t) = (pointer)nd_ps_trace[(int)BDY(t)];
1.20 noro 1747: return g;
1748: }
1749:
1.23 noro 1750: int ndv_compare(NDV *p1,NDV *p2)
1751: {
1.34 noro 1752: return DL_COMPARE(HDL(*p1),HDL(*p2));
1.23 noro 1753: }
1754:
1755: int ndv_compare_rev(NDV *p1,NDV *p2)
1756: {
1.34 noro 1757: return -DL_COMPARE(HDL(*p1),HDL(*p2));
1.23 noro 1758: }
1759:
1.61 noro 1760: NODE ndv_reduceall(int m,NODE f)
1.23 noro 1761: {
1.69 noro 1762: int i,n,stat;
1763: ND nf,g,head;
1.23 noro 1764: NODE t,a0,a;
1.112 ! noro 1765: Q dn;
1.69 noro 1766: NDV *w;
1.23 noro 1767:
1.69 noro 1768: n = length(f);
1769: #if 0
1770: w = (NDV *)ALLOCA(n*sizeof(NDV));
1771: for ( i = 0, t = f; i < n; i++, t = NEXT(t) ) w[i] = (NDV)BDY(t);
1772: qsort(w,n,sizeof(NDV),
1773: (int (*)(const void *,const void *))ndv_compare);
1774: for ( t = f, i = 0; t; i++, t = NEXT(t) ) BDY(t) = (pointer)w[i];
1775: #endif
1776: ndv_setup(m,0,f);
1777: for ( i = 0; i < n; ) {
1778: g = ndvtond(m,nd_ps[i]);
1779: g = nd_separate_head(g,&head);
1780: stat = nd_nf(m,g,nd_ps,1,&dn,&nf);
1.50 noro 1781: if ( !stat )
1.103 noro 1782: nd_reconstruct(0,0);
1.69 noro 1783: else {
1.71 noro 1784: if ( DP_Print ) { printf("."); fflush(stdout); }
1.85 noro 1785: if ( !m ) {
1.112 ! noro 1786: HCZ(head) = mulz(HCZ(head),simpz((Z)NM(dn)));
! 1787: if ( SGN(dn) < 0 ) HCZ(head) = chsgnz(HCZ(head));
! 1788: if ( DN(dn) )
! 1789: nd_mul_c_z(nf,simpz((Z)DN(dn)));
1.85 noro 1790: }
1.69 noro 1791: nf = nd_add(m,head,nf);
1792: ndv_free(nd_ps[i]);
1.24 noro 1793: nd_removecont(m,nf);
1.69 noro 1794: nd_ps[i] = ndtondv(m,nf); nd_free(nf);
1795: nd_bound[i] = ndv_compute_bound(nd_ps[i]);
1.50 noro 1796: i++;
1.23 noro 1797: }
1798: }
1.71 noro 1799: if ( DP_Print ) { printf("\n"); }
1.23 noro 1800: for ( a0 = 0, i = 0; i < n; i++ ) {
1801: NEXTNODE(a0,a);
1.69 noro 1802: BDY(a) = (pointer)nd_ps[i];
1.23 noro 1803: }
1804: NEXT(a) = 0;
1805: return a0;
1806: }
1807:
1.1 noro 1808: ND_pairs update_pairs( ND_pairs d, NODE /* of index */ g, int t)
1809: {
1810: ND_pairs d1,nd,cur,head,prev,remove;
1811:
1812: if ( !g ) return d;
1813: d = crit_B(d,t);
1814: d1 = nd_newpairs(g,t);
1815: d1 = crit_M(d1);
1816: d1 = crit_F(d1);
1.55 noro 1817: if ( do_weyl )
1818: head = d1;
1819: else {
1820: prev = 0; cur = head = d1;
1821: while ( cur ) {
1822: if ( crit_2( cur->i1,cur->i2 ) ) {
1823: remove = cur;
1824: if ( !prev ) head = cur = NEXT(cur);
1825: else cur = NEXT(prev) = NEXT(cur);
1826: FREENDP(remove);
1827: } else {
1828: prev = cur; cur = NEXT(cur);
1829: }
1.1 noro 1830: }
1831: }
1832: if ( !d )
1833: return head;
1834: else {
1835: nd = d;
1.34 noro 1836: while ( NEXT(nd) ) nd = NEXT(nd);
1.1 noro 1837: NEXT(nd) = head;
1838: return d;
1839: }
1840: }
1841:
1842: ND_pairs nd_newpairs( NODE g, int t )
1843: {
1844: NODE h;
1.61 noro 1845: UINT *dl;
1.34 noro 1846: int ts,s;
1.1 noro 1847: ND_pairs r,r0;
1848:
1.20 noro 1849: dl = DL(nd_psh[t]);
1.34 noro 1850: ts = SG(nd_psh[t]) - TD(dl);
1.1 noro 1851: for ( r0 = 0, h = g; h; h = NEXT(h) ) {
1852: NEXTND_pairs(r0,r);
1853: r->i1 = (int)BDY(h);
1854: r->i2 = t;
1.20 noro 1855: ndl_lcm(DL(nd_psh[r->i1]),dl,r->lcm);
1.34 noro 1856: s = SG(nd_psh[r->i1])-TD(DL(nd_psh[r->i1]));
1857: SG(r) = MAX(s,ts) + TD(LCM(r));
1.1 noro 1858: }
1859: NEXT(r) = 0;
1860: return r0;
1861: }
1862:
1863: ND_pairs crit_B( ND_pairs d, int s )
1864: {
1865: ND_pairs cur,head,prev,remove;
1.61 noro 1866: UINT *t,*tl,*lcm;
1.1 noro 1867: int td,tdl;
1868:
1869: if ( !d ) return 0;
1.20 noro 1870: t = DL(nd_psh[s]);
1.1 noro 1871: prev = 0;
1872: head = cur = d;
1.61 noro 1873: lcm = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
1.1 noro 1874: while ( cur ) {
1875: tl = cur->lcm;
1876: if ( ndl_reducible(tl,t)
1.20 noro 1877: && (ndl_lcm(DL(nd_psh[cur->i1]),t,lcm),!ndl_equal(lcm,tl))
1878: && (ndl_lcm(DL(nd_psh[cur->i2]),t,lcm),!ndl_equal(lcm,tl)) ) {
1.1 noro 1879: remove = cur;
1880: if ( !prev ) {
1881: head = cur = NEXT(cur);
1882: } else {
1883: cur = NEXT(prev) = NEXT(cur);
1884: }
1885: FREENDP(remove);
1886: } else {
1.34 noro 1887: prev = cur; cur = NEXT(cur);
1.1 noro 1888: }
1889: }
1890: return head;
1891: }
1892:
1893: ND_pairs crit_M( ND_pairs d1 )
1894: {
1895: ND_pairs e,d2,d3,dd,p;
1.61 noro 1896: UINT *id,*jd;
1.1 noro 1897:
1898: for ( dd = 0, e = d1; e; e = d3 ) {
1899: if ( !(d2 = NEXT(e)) ) {
1900: NEXT(e) = dd;
1901: return e;
1902: }
1.34 noro 1903: id = LCM(e);
1.1 noro 1904: for ( d3 = 0; d2; d2 = p ) {
1.34 noro 1905: p = NEXT(d2);
1906: jd = LCM(d2);
1907: if ( ndl_equal(jd,id) )
1908: ;
1909: else if ( TD(jd) > TD(id) )
1.1 noro 1910: if ( ndl_reducible(jd,id) ) continue;
1911: else ;
1.34 noro 1912: else if ( ndl_reducible(id,jd) ) goto delit;
1.1 noro 1913: NEXT(d2) = d3;
1914: d3 = d2;
1915: }
1916: NEXT(e) = dd;
1917: dd = e;
1918: continue;
1919: /**/
1920: delit: NEXT(d2) = d3;
1921: d3 = d2;
1922: for ( ; p; p = d2 ) {
1923: d2 = NEXT(p);
1924: NEXT(p) = d3;
1925: d3 = p;
1926: }
1927: FREENDP(e);
1928: }
1929: return dd;
1930: }
1931:
1932: ND_pairs crit_F( ND_pairs d1 )
1933: {
1934: ND_pairs rest, head,remove;
1935: ND_pairs last, p, r, w;
1936: int s;
1937:
1938: for ( head = last = 0, p = d1; NEXT(p); ) {
1939: r = w = equivalent_pairs(p,&rest);
1.14 noro 1940: s = SG(r);
1.1 noro 1941: w = NEXT(w);
1942: while ( w ) {
1943: if ( crit_2(w->i1,w->i2) ) {
1944: r = w;
1945: w = NEXT(w);
1946: while ( w ) {
1947: remove = w;
1948: w = NEXT(w);
1949: FREENDP(remove);
1950: }
1951: break;
1.14 noro 1952: } else if ( SG(w) < s ) {
1.1 noro 1953: FREENDP(r);
1954: r = w;
1.14 noro 1955: s = SG(r);
1.1 noro 1956: w = NEXT(w);
1957: } else {
1958: remove = w;
1959: w = NEXT(w);
1960: FREENDP(remove);
1961: }
1962: }
1963: if ( last ) NEXT(last) = r;
1964: else head = r;
1965: NEXT(last = r) = 0;
1966: p = rest;
1967: if ( !p ) return head;
1968: }
1969: if ( !last ) return p;
1970: NEXT(last) = p;
1971: return head;
1972: }
1973:
1974: int crit_2( int dp1, int dp2 )
1975: {
1.20 noro 1976: return ndl_disjoint(DL(nd_psh[dp1]),DL(nd_psh[dp2]));
1.1 noro 1977: }
1978:
1.40 noro 1979: ND_pairs equivalent_pairs( ND_pairs d1, ND_pairs *prest )
1.1 noro 1980: {
1981: ND_pairs w,p,r,s;
1.61 noro 1982: UINT *d;
1.1 noro 1983:
1984: w = d1;
1.34 noro 1985: d = LCM(w);
1.1 noro 1986: s = NEXT(w);
1987: NEXT(w) = 0;
1988: for ( r = 0; s; s = p ) {
1989: p = NEXT(s);
1.34 noro 1990: if ( ndl_equal(d,LCM(s)) ) {
1.39 noro 1991: NEXT(s) = w; w = s;
1.1 noro 1992: } else {
1.39 noro 1993: NEXT(s) = r; r = s;
1.1 noro 1994: }
1995: }
1996: *prest = r;
1997: return w;
1998: }
1999:
2000: NODE update_base(NODE nd,int ndp)
2001: {
1.61 noro 2002: UINT *dl, *dln;
1.1 noro 2003: NODE last, p, head;
2004:
1.20 noro 2005: dl = DL(nd_psh[ndp]);
1.1 noro 2006: for ( head = last = 0, p = nd; p; ) {
1.20 noro 2007: dln = DL(nd_psh[(int)BDY(p)]);
1.34 noro 2008: if ( ndl_reducible( dln, dl ) ) {
1.1 noro 2009: p = NEXT(p);
2010: if ( last ) NEXT(last) = p;
2011: } else {
2012: if ( !last ) head = p;
2013: p = NEXT(last = p);
2014: }
2015: }
2016: head = append_one(head,ndp);
2017: return head;
2018: }
2019:
2020: ND_pairs nd_minp( ND_pairs d, ND_pairs *prest )
2021: {
2022: ND_pairs m,ml,p,l;
1.61 noro 2023: UINT *lcm;
1.33 noro 2024: int s,td,len,tlen,c,c1;
1.1 noro 2025:
2026: if ( !(p = NEXT(m = d)) ) {
2027: *prest = p;
2028: NEXT(m) = 0;
2029: return m;
2030: }
1.14 noro 2031: s = SG(m);
1.77 noro 2032: if ( !NoSugar ) {
2033: for ( ml = 0, l = m; p; p = NEXT(l = p) )
2034: if ( (SG(p) < s)
2035: || ((SG(p) == s) && (DL_COMPARE(LCM(p),LCM(m)) < 0)) ) {
2036: ml = l; m = p; s = SG(m);
2037: }
2038: } else {
2039: for ( ml = 0, l = m; p; p = NEXT(l = p) )
2040: if ( DL_COMPARE(LCM(p),LCM(m)) < 0 ) {
2041: ml = l; m = p; s = SG(m);
2042: }
2043: }
1.1 noro 2044: if ( !ml ) *prest = NEXT(m);
2045: else {
2046: NEXT(ml) = NEXT(m);
2047: *prest = d;
2048: }
2049: NEXT(m) = 0;
2050: return m;
2051: }
2052:
1.63 noro 2053: ND_pairs nd_minsugarp( ND_pairs d, ND_pairs *prest )
2054: {
1.69 noro 2055: int msugar,i;
1.63 noro 2056: ND_pairs t,dm0,dm,dr0,dr;
2057:
2058: for ( msugar = SG(d), t = NEXT(d); t; t = NEXT(t) )
2059: if ( SG(t) < msugar ) msugar = SG(t);
2060: dm0 = 0; dr0 = 0;
1.69 noro 2061: for ( i = 0, t = d; t; t = NEXT(t) )
2062: if ( i < nd_f4_nsp && SG(t) == msugar ) {
1.63 noro 2063: if ( dm0 ) NEXT(dm) = t;
2064: else dm0 = t;
2065: dm = t;
1.69 noro 2066: i++;
1.63 noro 2067: } else {
2068: if ( dr0 ) NEXT(dr) = t;
2069: else dr0 = t;
2070: dr = t;
2071: }
2072: NEXT(dm) = 0;
2073: if ( dr0 ) NEXT(dr) = 0;
2074: *prest = dr0;
2075: return dm0;
2076: }
2077:
1.77 noro 2078: int ndv_newps(int m,NDV a,NDV aq)
1.1 noro 2079: {
1.3 noro 2080: int len;
1.13 noro 2081: RHist r;
1.20 noro 2082: NDV b;
1.3 noro 2083:
1.1 noro 2084: if ( nd_psn == nd_pslen ) {
2085: nd_pslen *= 2;
1.11 noro 2086: nd_ps = (NDV *)REALLOC((char *)nd_ps,nd_pslen*sizeof(NDV));
1.53 noro 2087: nd_ps_trace = (NDV *)REALLOC((char *)nd_ps_trace,nd_pslen*sizeof(NDV));
1.13 noro 2088: nd_psh = (RHist *)REALLOC((char *)nd_psh,nd_pslen*sizeof(RHist));
1.61 noro 2089: nd_bound = (UINT **)
2090: REALLOC((char *)nd_bound,nd_pslen*sizeof(UINT *));
1.1 noro 2091: }
1.39 noro 2092: NEWRHist(r); nd_psh[nd_psn] = r;
1.63 noro 2093: nd_ps[nd_psn] = a;
1.39 noro 2094: if ( aq ) {
1.63 noro 2095: nd_ps_trace[nd_psn] = aq;
2096: nd_bound[nd_psn] = ndv_compute_bound(aq);
1.39 noro 2097: SG(r) = SG(aq); ndl_copy(HDL(aq),DL(r));
1.53 noro 2098: } else {
1.63 noro 2099: nd_bound[nd_psn] = ndv_compute_bound(a);
1.53 noro 2100: SG(r) = SG(a); ndl_copy(HDL(a),DL(r));
1.39 noro 2101: }
1.77 noro 2102: if ( nd_demand ) {
2103: if ( aq ) {
2104: ndv_save(nd_ps_trace[nd_psn],nd_psn);
2105: nd_ps_trace[nd_psn] = 0;
2106: } else {
2107: ndv_save(nd_ps[nd_psn],nd_psn);
2108: nd_ps[nd_psn] = 0;
2109: }
2110: }
1.1 noro 2111: return nd_psn++;
2112: }
2113:
1.61 noro 2114: void ndv_setup(int mod,int trace,NODE f)
1.1 noro 2115: {
1.5 noro 2116: int i,j,td,len,max;
1.1 noro 2117: NODE s,s0,f0;
1.61 noro 2118: UINT *d;
1.13 noro 2119: RHist r;
1.69 noro 2120: NDV *w;
1.61 noro 2121: NDV a,am;
1.11 noro 2122:
2123: nd_found = 0; nd_notfirst = 0; nd_create = 0;
1.1 noro 2124:
1.69 noro 2125: for ( nd_psn = 0, s = f; s; s = NEXT(s) ) if ( BDY(s) ) nd_psn++;
2126: w = (NDV *)ALLOCA(nd_psn*sizeof(NDV));
2127: for ( i = 0, s = f; s; s = NEXT(s) ) if ( BDY(s) ) w[i++] = BDY(s);
2128: qsort(w,nd_psn,sizeof(NDV),
2129: (int (*)(const void *,const void *))ndv_compare);
2130: nd_pslen = 2*nd_psn;
1.11 noro 2131: nd_ps = (NDV *)MALLOC(nd_pslen*sizeof(NDV));
1.53 noro 2132: nd_ps_trace = (NDV *)MALLOC(nd_pslen*sizeof(NDV));
1.13 noro 2133: nd_psh = (RHist *)MALLOC(nd_pslen*sizeof(RHist));
1.61 noro 2134: nd_bound = (UINT **)MALLOC(nd_pslen*sizeof(UINT *));
1.57 noro 2135:
1.11 noro 2136: if ( !nd_red )
1.13 noro 2137: nd_red = (RHist *)MALLOC(REDTAB_LEN*sizeof(RHist));
1.71 noro 2138: for ( i = 0; i < REDTAB_LEN; i++ ) nd_red[i] = 0;
1.69 noro 2139: for ( i = 0; i < nd_psn; i++ ) {
1.39 noro 2140: if ( trace ) {
1.69 noro 2141: a = nd_ps_trace[i] = ndv_dup(0,w[i]);
1.61 noro 2142: ndv_removecont(0,a);
2143: am = nd_ps[i] = ndv_dup(mod,a);
2144: ndv_mod(mod,am);
2145: ndv_removecont(mod,am);
2146: } else {
1.69 noro 2147: a = nd_ps[i] = ndv_dup(mod,w[i]);
1.61 noro 2148: ndv_removecont(mod,a);
1.39 noro 2149: }
1.61 noro 2150: NEWRHist(r); SG(r) = HTD(a); ndl_copy(HDL(a),DL(r));
1.57 noro 2151: nd_bound[i] = ndv_compute_bound(a);
1.20 noro 2152: nd_psh[i] = r;
1.77 noro 2153: if ( nd_demand ) {
2154: if ( trace ) {
2155: ndv_save(nd_ps_trace[i],i);
2156: nd_ps_trace[i] = 0;
2157: } else {
2158: ndv_save(nd_ps[i],i);
2159: nd_ps[i] = 0;
2160: }
2161: }
1.20 noro 2162: }
2163: }
2164:
1.63 noro 2165: void nd_gr(LIST f,LIST v,int m,int f4,struct order_spec *ord,LIST *rp)
1.1 noro 2166: {
1.61 noro 2167: VL tv,fv,vv,vc;
1.1 noro 2168: NODE fd,fd0,r,r0,t,x,s,xx;
1.61 noro 2169: int e,max,nvar;
1.69 noro 2170: NDV b;
1.77 noro 2171: int ishomo;
1.1 noro 2172:
1.78 noro 2173: if ( !m && Demand ) nd_demand = 1;
2174: else nd_demand = 0;
2175:
1.103 noro 2176: #if 0
1.74 noro 2177: ndv_alloc = 0;
1.103 noro 2178: #endif
1.1 noro 2179: get_vars((Obj)f,&fv); pltovl(v,&vv);
1.74 noro 2180: for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++ );
1.88 noro 2181: switch ( ord->id ) {
2182: case 1:
2183: if ( ord->nv != nvar )
2184: error("nd_{gr,f4} : invalid order specification");
2185: break;
2186: default:
2187: break;
2188: }
1.32 noro 2189: nd_init_ord(ord);
1.61 noro 2190: for ( t = BDY(f), max = 0; t; t = NEXT(t) )
2191: for ( tv = vv; tv; tv = NEXT(tv) ) {
2192: e = getdeg(tv->v,(P)BDY(t));
2193: max = MAX(e,max);
2194: }
2195: nd_setup_parameters(nvar,max);
1.77 noro 2196: ishomo = 1;
1.1 noro 2197: for ( fd0 = 0, t = BDY(f); t; t = NEXT(t) ) {
1.61 noro 2198: b = (pointer)ptondv(CO,vv,(P)BDY(t));
1.77 noro 2199: if ( ishomo )
2200: ishomo = ishomo && ndv_ishomo(b);
1.69 noro 2201: if ( m ) ndv_mod(m,b);
1.61 noro 2202: if ( b ) { NEXTNODE(fd0,fd); BDY(fd) = (pointer)b; }
1.1 noro 2203: }
2204: if ( fd0 ) NEXT(fd) = 0;
1.61 noro 2205: ndv_setup(m,0,fd0);
1.77 noro 2206: x = f4?nd_f4(m):nd_gb(m,ishomo,0);
1.82 noro 2207: nd_demand = 0;
1.61 noro 2208: x = ndv_reducebase(x);
2209: x = ndv_reduceall(m,x);
1.23 noro 2210: for ( r0 = 0, t = x; t; t = NEXT(t) ) {
1.1 noro 2211: NEXTNODE(r0,r);
1.61 noro 2212: BDY(r) = ndvtop(m,CO,vv,BDY(t));
1.20 noro 2213: }
2214: if ( r0 ) NEXT(r) = 0;
2215: MKLIST(*rp,r0);
1.103 noro 2216: #if 0
1.74 noro 2217: fprintf(asir_out,"ndv_alloc=%d\n",ndv_alloc);
1.103 noro 2218: #endif
1.20 noro 2219: }
2220:
1.52 noro 2221: void nd_gr_trace(LIST f,LIST v,int trace,int homo,struct order_spec *ord,LIST *rp)
1.20 noro 2222: {
1.89 noro 2223: struct order_spec *ord1;
1.61 noro 2224: VL tv,fv,vv,vc;
1.27 noro 2225: NODE fd,fd0,in0,in,r,r0,t,s,cand;
1.61 noro 2226: int m,nocheck,nvar,mindex,e,max;
2227: NDV c;
2228: NMV a;
1.27 noro 2229: P p;
1.61 noro 2230: EPOS oepos;
1.82 noro 2231: int obpe,oadv,wmax,i,len,cbpe,ishomo;
1.20 noro 2232:
2233: get_vars((Obj)f,&fv); pltovl(v,&vv);
1.74 noro 2234: for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++ );
1.88 noro 2235: switch ( ord->id ) {
2236: case 1:
2237: if ( ord->nv != nvar )
2238: error("nd_gr_trace : invalid order specification");
2239: break;
2240: default:
2241: break;
2242: }
1.52 noro 2243: nocheck = 0;
2244: mindex = 0;
2245:
1.78 noro 2246: if ( Demand ) nd_demand = 1;
2247: else nd_demand = 0;
2248:
1.52 noro 2249: /* setup modulus */
2250: if ( trace < 0 ) {
2251: trace = -trace;
2252: nocheck = 1;
2253: }
2254: m = trace > 1 ? trace : get_lprime(mindex);
1.61 noro 2255: for ( t = BDY(f), max = 0; t; t = NEXT(t) )
2256: for ( tv = vv; tv; tv = NEXT(tv) ) {
2257: e = getdeg(tv->v,(P)BDY(t));
2258: max = MAX(e,max);
1.23 noro 2259: }
1.61 noro 2260: nd_init_ord(ord);
2261: nd_setup_parameters(nvar,max);
2262: obpe = nd_bpe; oadv = nmv_adv; oepos = nd_epos;
1.82 noro 2263: ishomo = 1;
1.61 noro 2264: for ( in0 = 0, fd0 = 0, t = BDY(f); t; t = NEXT(t) ) {
2265: c = ptondv(CO,vv,(P)BDY(t));
1.82 noro 2266: if ( ishomo )
2267: ishomo = ishomo && ndv_ishomo(c);
1.61 noro 2268: if ( c ) {
2269: NEXTNODE(in0,in); BDY(in) = (pointer)c;
2270: NEXTNODE(fd0,fd); BDY(fd) = (pointer)ndv_dup(0,c);
1.23 noro 2271: }
1.61 noro 2272: }
2273: if ( in0 ) NEXT(in) = 0;
2274: if ( fd0 ) NEXT(fd) = 0;
1.82 noro 2275: if ( !ishomo && homo ) {
1.61 noro 2276: for ( t = in0, wmax = 0; t; t = NEXT(t) ) {
2277: c = (NDV)BDY(t); len = LEN(c);
2278: for ( a = BDY(c), i = 0; i < len; i++, NMV_ADV(a) )
2279: wmax = MAX(TD(DL(a)),wmax);
2280: }
2281: homogenize_order(ord,nvar,&ord1);
1.89 noro 2282: nd_init_ord(ord1);
1.61 noro 2283: nd_setup_parameters(nvar+1,wmax);
2284: for ( t = fd0; t; t = NEXT(t) )
2285: ndv_homogenize((NDV)BDY(t),obpe,oadv,oepos);
1.27 noro 2286: }
1.52 noro 2287: while ( 1 ) {
1.77 noro 2288: if ( Demand )
2289: nd_demand = 1;
1.61 noro 2290: ndv_setup(m,1,fd0);
1.82 noro 2291: cand = nd_gb_trace(m,ishomo || homo);
1.52 noro 2292: if ( !cand ) {
2293: /* failure */
1.61 noro 2294: if ( trace > 1 ) { *rp = 0; return; }
2295: else m = get_lprime(++mindex);
1.52 noro 2296: continue;
2297: }
1.82 noro 2298: if ( !ishomo && homo ) {
1.27 noro 2299: /* dehomogenization */
1.61 noro 2300: for ( t = cand; t; t = NEXT(t) ) ndv_dehomogenize((NDV)BDY(t),ord);
1.45 noro 2301: nd_init_ord(ord);
1.61 noro 2302: nd_setup_parameters(nvar,0);
1.27 noro 2303: }
1.77 noro 2304: nd_demand = 0;
1.61 noro 2305: cand = ndv_reducebase(cand);
2306: cand = ndv_reduceall(0,cand);
1.78 noro 2307: cbpe = nd_bpe;
1.61 noro 2308: if ( nocheck )
2309: break;
2310: if ( ndv_check_candidate(in0,obpe,oadv,oepos,cand) )
1.52 noro 2311: /* success */
2312: break;
2313: else if ( trace > 1 ) {
2314: /* failure */
2315: *rp = 0; return;
1.61 noro 2316: } else {
1.52 noro 2317: /* try the next modulus */
2318: m = get_lprime(++mindex);
1.61 noro 2319: /* reset the parameters */
1.82 noro 2320: if ( !ishomo && homo ) {
1.89 noro 2321: nd_init_ord(ord1);
1.61 noro 2322: nd_setup_parameters(nvar+1,wmax);
2323: } else {
2324: nd_init_ord(ord);
2325: nd_setup_parameters(nvar,max);
2326: }
2327: }
1.52 noro 2328: }
1.27 noro 2329: /* dp->p */
1.61 noro 2330: nd_bpe = cbpe;
1.73 noro 2331: nd_setup_parameters(nd_nvar,0);
1.61 noro 2332: for ( r = cand; r; r = NEXT(r) ) BDY(r) = (pointer)ndvtop(0,CO,vv,BDY(r));
1.27 noro 2333: MKLIST(*rp,cand);
1.1 noro 2334: }
2335:
1.61 noro 2336: void dltondl(int n,DL dl,UINT *r)
1.1 noro 2337: {
1.61 noro 2338: UINT *d;
1.57 noro 2339: int i,j,l,s,ord_l;
1.43 noro 2340: struct order_pair *op;
1.1 noro 2341:
2342: d = dl->d;
1.41 noro 2343: for ( i = 0; i < nd_wpd; i++ ) r[i] = 0;
1.43 noro 2344: if ( nd_blockmask ) {
2345: l = nd_blockmask->n;
2346: op = nd_blockmask->order_pair;
2347: for ( j = 0, s = 0; j < l; j++ ) {
2348: ord_l = op[j].length;
1.57 noro 2349: for ( i = 0; i < ord_l; i++, s++ ) PUT_EXP(r,s,d[s]);
1.43 noro 2350: }
2351: TD(r) = ndl_weight(r);
1.61 noro 2352: ndl_weight_mask(r);
1.43 noro 2353: } else {
1.56 noro 2354: for ( i = 0; i < n; i++ ) PUT_EXP(r,i,d[i]);
1.43 noro 2355: TD(r) = ndl_weight(r);
2356: }
1.1 noro 2357: }
2358:
1.61 noro 2359: DL ndltodl(int n,UINT *ndl)
1.1 noro 2360: {
2361: DL dl;
2362: int *d;
1.57 noro 2363: int i,j,l,s,ord_l;
1.43 noro 2364: struct order_pair *op;
1.1 noro 2365:
2366: NEWDL(dl,n);
1.34 noro 2367: dl->td = TD(ndl);
1.1 noro 2368: d = dl->d;
1.43 noro 2369: if ( nd_blockmask ) {
2370: l = nd_blockmask->n;
2371: op = nd_blockmask->order_pair;
2372: for ( j = 0, s = 0; j < l; j++ ) {
2373: ord_l = op[j].length;
1.57 noro 2374: for ( i = 0; i < ord_l; i++, s++ ) d[s] = GET_EXP(ndl,s);
1.43 noro 2375: }
2376: } else {
1.56 noro 2377: for ( i = 0; i < n; i++ ) d[i] = GET_EXP(ndl,i);
1.43 noro 2378: }
1.1 noro 2379: return dl;
2380: }
2381:
1.61 noro 2382: void ndl_print(UINT *dl)
1.1 noro 2383: {
2384: int n;
1.57 noro 2385: int i,j,l,ord_l,s,s0;
1.43 noro 2386: struct order_pair *op;
1.1 noro 2387:
2388: n = nd_nvar;
2389: printf("<<");
1.43 noro 2390: if ( nd_blockmask ) {
2391: l = nd_blockmask->n;
2392: op = nd_blockmask->order_pair;
2393: for ( j = 0, s = s0 = 0; j < l; j++ ) {
2394: ord_l = op[j].length;
1.57 noro 2395: for ( i = 0; i < ord_l; i++, s++ )
2396: printf(s==n-1?"%d":"%d,",GET_EXP(dl,s));
1.43 noro 2397: }
2398: } else {
1.56 noro 2399: for ( i = 0; i < n; i++ ) printf(i==n-1?"%d":"%d,",GET_EXP(dl,i));
1.43 noro 2400: }
1.1 noro 2401: printf(">>");
2402: }
2403:
2404: void nd_print(ND p)
2405: {
2406: NM m;
2407:
2408: if ( !p )
2409: printf("0\n");
2410: else {
2411: for ( m = BDY(p); m; m = NEXT(m) ) {
1.71 noro 2412: if ( CM(m) & 0x80000000 ) printf("+@_%d*",IFTOF(CM(m)));
2413: else printf("+%d*",CM(m));
1.14 noro 2414: ndl_print(DL(m));
1.1 noro 2415: }
2416: printf("\n");
2417: }
2418: }
2419:
1.112 ! noro 2420: void nd_print_z(ND p)
1.16 noro 2421: {
2422: NM m;
2423:
2424: if ( !p )
2425: printf("0\n");
2426: else {
2427: for ( m = BDY(p); m; m = NEXT(m) ) {
2428: printf("+");
1.112 ! noro 2429: printz(CZ(m));
1.16 noro 2430: printf("*");
2431: ndl_print(DL(m));
2432: }
2433: printf("\n");
2434: }
2435: }
2436:
1.1 noro 2437: void ndp_print(ND_pairs d)
2438: {
2439: ND_pairs t;
2440:
1.34 noro 2441: for ( t = d; t; t = NEXT(t) ) printf("%d,%d ",t->i1,t->i2);
1.1 noro 2442: printf("\n");
2443: }
2444:
1.20 noro 2445: void nd_removecont(int mod,ND p)
1.16 noro 2446: {
2447: int i,n;
1.112 ! noro 2448: Z *w;
! 2449: Z dvr,t;
1.16 noro 2450: NM m;
1.21 noro 2451: struct oVECT v;
1.16 noro 2452:
1.71 noro 2453: if ( mod == -1 ) nd_mul_c(mod,p,_invsf(HCM(p)));
2454: else if ( mod ) nd_mul_c(mod,p,invm(HCM(p),mod));
1.20 noro 2455: else {
2456: for ( m = BDY(p), n = 0; m; m = NEXT(m), n++ );
1.112 ! noro 2457: w = (Z *)ALLOCA(n*sizeof(Z));
1.21 noro 2458: v.len = n;
2459: v.body = (pointer *)w;
1.112 ! noro 2460: for ( m = BDY(p), i = 0; i < n; m = NEXT(m), i++ ) w[i] = CZ(m);
1.21 noro 2461: removecont_array(w,n);
1.112 ! noro 2462: for ( m = BDY(p), i = 0; i < n; m = NEXT(m), i++ ) CZ(m) = w[i];
1.16 noro 2463: }
2464: }
2465:
1.21 noro 2466: void nd_removecont2(ND p1,ND p2)
2467: {
2468: int i,n1,n2,n;
1.112 ! noro 2469: Z *w;
1.21 noro 2470: NM m;
2471: struct oVECT v;
2472:
2473: if ( !p1 ) {
2474: nd_removecont(0,p2); return;
2475: } else if ( !p2 ) {
2476: nd_removecont(0,p1); return;
2477: }
2478: n1 = nd_length(p1);
2479: n2 = nd_length(p2);
2480: n = n1+n2;
1.112 ! noro 2481: w = (Z *)ALLOCA(n*sizeof(Z));
1.21 noro 2482: v.len = n;
2483: v.body = (pointer *)w;
1.112 ! noro 2484: for ( m = BDY(p1), i = 0; i < n1; m = NEXT(m), i++ ) w[i] = CZ(m);
! 2485: for ( m = BDY(p2); i < n; m = NEXT(m), i++ ) w[i] = CZ(m);
1.21 noro 2486: removecont_array(w,n);
1.112 ! noro 2487: for ( m = BDY(p1), i = 0; i < n1; m = NEXT(m), i++ ) CZ(m) = w[i];
! 2488: for ( m = BDY(p2); i < n; m = NEXT(m), i++ ) CZ(m) = w[i];
1.21 noro 2489: }
2490:
1.20 noro 2491: void ndv_removecont(int mod,NDV p)
1.16 noro 2492: {
2493: int i,len;
1.112 ! noro 2494: Z *w;
! 2495: Z dvr,t;
1.16 noro 2496: NMV m;
2497:
1.71 noro 2498: if ( mod == -1 )
2499: ndv_mul_c(mod,p,_invsf(HCM(p)));
2500: else if ( mod )
1.20 noro 2501: ndv_mul_c(mod,p,invm(HCM(p),mod));
2502: else {
2503: len = p->len;
1.112 ! noro 2504: w = (Z *)ALLOCA(len*sizeof(Z));
! 2505: for ( m = BDY(p), i = 0; i < len; NMV_ADV(m), i++ ) w[i] = CZ(m);
! 2506: removecont_array(w,len);
! 2507: for ( m = BDY(p), i = 0; i < len; NMV_ADV(m), i++ ) CZ(m) = w[i];
1.16 noro 2508: }
1.21 noro 2509: }
2510:
1.61 noro 2511: void ndv_homogenize(NDV p,int obpe,int oadv,EPOS oepos)
2512: {
2513: int len,i,max;
2514: NMV m,mr0,mr,t;
2515:
2516: len = p->len;
2517: for ( m = BDY(p), i = 0, max = 0; i < len; NMV_OADV(m), i++ )
2518: max = MAX(max,TD(DL(m)));
2519: mr0 = nmv_adv>oadv?(NMV)REALLOC(BDY(p),len*nmv_adv):BDY(p);
2520: m = (NMV)((char *)mr0+(len-1)*oadv);
2521: mr = (NMV)((char *)mr0+(len-1)*nmv_adv);
2522: t = (NMV)ALLOCA(nmv_adv);
2523: for ( i = 0; i < len; i++, NMV_OPREV(m), NMV_PREV(mr) ) {
2524: ndl_homogenize(DL(m),DL(t),obpe,oepos,max);
1.112 ! noro 2525: CZ(mr) = CZ(m);
1.61 noro 2526: ndl_copy(DL(t),DL(mr));
2527: }
2528: NV(p)++;
2529: BDY(p) = mr0;
2530: }
2531:
1.45 noro 2532: void ndv_dehomogenize(NDV p,struct order_spec *ord)
1.23 noro 2533: {
1.45 noro 2534: int i,j,adj,len,newnvar,newwpd,newadv,newexporigin;
1.112 ! noro 2535: Z *w;
! 2536: Z dvr,t;
1.23 noro 2537: NMV m,r;
2538:
2539: len = p->len;
2540: newnvar = nd_nvar-1;
1.48 noro 2541: newexporigin = nd_get_exporigin(ord);
1.45 noro 2542: newwpd = newnvar/nd_epw+(newnvar%nd_epw?1:0)+newexporigin;
1.23 noro 2543: for ( m = BDY(p), i = 0; i < len; NMV_ADV(m), i++ )
1.34 noro 2544: ndl_dehomogenize(DL(m));
1.23 noro 2545: if ( newwpd != nd_wpd ) {
1.90 noro 2546: newadv = ROUND_FOR_ALIGN(sizeof(struct oNMV)+(newwpd-1)*sizeof(UINT));
1.61 noro 2547: for ( m = r = BDY(p), i = 0; i < len; NMV_ADV(m), NDV_NADV(r), i++ ) {
1.112 ! noro 2548: CZ(r) = CZ(m);
1.45 noro 2549: for ( j = 0; j < newexporigin; j++ ) DL(r)[j] = DL(m)[j];
2550: adj = nd_exporigin-newexporigin;
2551: for ( ; j < newwpd; j++ ) DL(r)[j] = DL(m)[j+adj];
1.23 noro 2552: }
2553: }
2554: NV(p)--;
2555: }
2556:
1.112 ! noro 2557: int comp_z(Z *a,Z *b)
! 2558: {
! 2559: return cmpz(*a,*b);
! 2560: }
! 2561:
! 2562: Z zgcdv(VECT a)
! 2563: {
! 2564: int i,j,n,nz;
! 2565: Z g,gt,q,r;
! 2566: Z *c;
! 2567:
! 2568: n = a->len;
! 2569: c = (Z *)ALLOCA(n*sizeof(Z));
! 2570: for ( i = 0; i < n; i++ )
! 2571: c[i] = (Z)a->body[i];
! 2572: qsort(c,n,sizeof(Z),(int (*) (const void *,const void *))comp_z);
! 2573: for ( ; n && ! *c; n--, c++ );
! 2574:
! 2575: if ( !n )
! 2576: return 0;
! 2577: else if ( n == 1 )
! 2578: return absz(c[0]);
! 2579:
! 2580: g = gcdz(c[0],c[1]);
! 2581: for ( i = 2; i < n; i++ ) {
! 2582: q = divz(c[i],g,&r);
! 2583: gt = gcdz(g,r);
! 2584: if ( !cmpz(g,gt) ) {
! 2585: for ( j = i+1; j < n; j++ ) {
! 2586: q = divz(c[j],g,&r); c[j] = r;
! 2587: }
! 2588: } else
! 2589: g = gt;
! 2590: }
! 2591: return absz(g);
! 2592: }
! 2593:
! 2594: Z zgcdv_estimate(VECT a)
! 2595: {
! 2596: int n,i,m;
! 2597: Z s,t,u,g;
! 2598: Z *q;
! 2599:
! 2600: n = a->len; q = (Z *)a->body;
! 2601: if ( n == 1 )
! 2602: return absz(q[0]);
! 2603:
! 2604: m = n/2;
! 2605: for ( i = 0 , s = 0; i < m; i++ )
! 2606: s = sgnz(q[i])>=0 ? addz(s,q[i]) : subz(s,q[i]);
! 2607: for ( t = 0; i < n; i++ )
! 2608: t = sgnz(q[i])>=0 ? addz(t,q[i]) : subz(t,q[i]);
! 2609: g = gcdz(s,t);
! 2610: return absz(g);
! 2611: }
! 2612:
! 2613: void removecont_array(Z *c,int n)
1.21 noro 2614: {
2615: struct oVECT v;
1.112 ! noro 2616: Z d0,d1,a,u,u1,gcd;
1.110 noro 2617: int i,j;
1.112 ! noro 2618: Z q0,q1;
! 2619: Z *q,*r;
1.21 noro 2620:
1.112 ! noro 2621: q = (Z *)ALLOCA(n*sizeof(Z));
! 2622: r = (Z *)ALLOCA(n*sizeof(Z));
1.21 noro 2623: v.id = O_VECT; v.len = n; v.body = (pointer *)c;
1.112 ! noro 2624: d0 = zgcdv_estimate(&v);
! 2625: /* c[i] = d0*q[i]+r[i] */
! 2626: for ( i = 0; i < n; i++ )
! 2627: q[i] = divz(c[i],d0,&r[i]);
1.34 noro 2628: for ( i = 0; i < n; i++ ) if ( r[i] ) break;
1.21 noro 2629: if ( i < n ) {
2630: v.id = O_VECT; v.len = n; v.body = (pointer *)r;
1.112 ! noro 2631: /* d1 = GCD(r[0],...,r[n-1]) */
! 2632: d1 = zgcdv(&v);
! 2633: /* q0 = d0/GCD(d0,d1) */
! 2634: gcd = gcdz_cofactor(d0,d1,&q0,&q1);
1.21 noro 2635: for ( i = 0; i < n; i++ ) {
1.112 ! noro 2636: /* q[i] = d0/gcd*q[i]+r[i]/gcd */
! 2637: q[i] = addz(mulz(q0,q[i]),divsz(r[i],gcd));
1.21 noro 2638: }
2639: }
1.34 noro 2640: for ( i = 0; i < n; i++ ) c[i] = q[i];
1.16 noro 2641: }
2642:
1.19 noro 2643: void nd_mul_c(int mod,ND p,int mul)
1.1 noro 2644: {
2645: NM m;
2646: int c,c1;
2647:
1.34 noro 2648: if ( !p ) return;
1.71 noro 2649: if ( mod == -1 )
2650: for ( m = BDY(p); m; m = NEXT(m) )
2651: CM(m) = _mulsf(CM(m),mul);
2652: else
2653: for ( m = BDY(p); m; m = NEXT(m) ) {
2654: c1 = CM(m); DMAR(c1,mul,0,mod,c); CM(m) = c;
2655: }
1.1 noro 2656: }
2657:
1.112 ! noro 2658: void nd_mul_c_z(ND p,Z mul)
1.16 noro 2659: {
2660: NM m;
1.112 ! noro 2661: Z c;
1.16 noro 2662:
1.112 ! noro 2663: if ( !p || uniz(mul) ) return;
1.16 noro 2664: for ( m = BDY(p); m; m = NEXT(m) ) {
1.112 ! noro 2665: CZ(m) = mulz(CZ(m),mul);
1.16 noro 2666: }
2667: }
2668:
1.61 noro 2669: void nd_mul_c_p(VL vl,ND p,P mul)
2670: {
2671: NM m;
2672: P c;
2673:
2674: if ( !p ) return;
2675: for ( m = BDY(p); m; m = NEXT(m) ) {
2676: mulp(vl,CP(m),mul,&c); CP(m) = c;
2677: }
2678: }
2679:
1.1 noro 2680: void nd_free(ND p)
2681: {
2682: NM t,s;
2683:
1.34 noro 2684: if ( !p ) return;
1.1 noro 2685: t = BDY(p);
2686: while ( t ) {
2687: s = NEXT(t);
2688: FREENM(t);
2689: t = s;
2690: }
2691: FREEND(p);
2692: }
2693:
1.23 noro 2694: void ndv_free(NDV p)
2695: {
2696: GC_free(BDY(p));
2697: }
2698:
1.61 noro 2699: void nd_append_red(UINT *d,int i)
1.1 noro 2700: {
1.13 noro 2701: RHist m,m0;
1.1 noro 2702: int h;
2703:
1.13 noro 2704: NEWRHist(m);
1.34 noro 2705: h = ndl_hash_value(d);
1.13 noro 2706: m->index = i;
1.14 noro 2707: ndl_copy(d,DL(m));
1.1 noro 2708: NEXT(m) = nd_red[h];
2709: nd_red[h] = m;
2710: }
2711:
1.61 noro 2712: UINT *ndv_compute_bound(NDV p)
1.1 noro 2713: {
1.61 noro 2714: UINT *d1,*d2,*t;
2715: UINT u;
1.57 noro 2716: int i,j,k,l,len,ind;
1.45 noro 2717: NMV m;
1.1 noro 2718:
2719: if ( !p )
2720: return 0;
1.61 noro 2721: d1 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
2722: d2 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
1.45 noro 2723: len = LEN(p);
2724: m = BDY(p); ndl_copy(DL(m),d1); NMV_ADV(m);
2725: for ( i = 1; i < len; i++, NMV_ADV(m) ) {
1.14 noro 2726: ndl_lcm(DL(m),d1,d2);
1.1 noro 2727: t = d1; d1 = d2; d2 = t;
2728: }
1.12 noro 2729: l = nd_nvar+31;
1.61 noro 2730: t = (UINT *)MALLOC_ATOMIC(l*sizeof(UINT));
1.57 noro 2731: for ( i = nd_exporigin, ind = 0; i < nd_wpd; i++ ) {
2732: u = d1[i];
2733: k = (nd_epw-1)*nd_bpe;
2734: for ( j = 0; j < nd_epw; j++, k -= nd_bpe, ind++ )
2735: t[ind] = (u>>k)&nd_mask0;
2736: }
2737: for ( ; ind < l; ind++ ) t[ind] = 0;
1.1 noro 2738: return t;
2739: }
2740:
1.99 noro 2741: UINT *nd_compute_bound(ND p)
2742: {
2743: UINT *d1,*d2,*t;
2744: UINT u;
2745: int i,j,k,l,len,ind;
2746: NM m;
2747:
2748: if ( !p )
2749: return 0;
2750: d1 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
2751: d2 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
2752: len = LEN(p);
2753: m = BDY(p); ndl_copy(DL(m),d1); m = NEXT(m);
2754: for ( m = NEXT(m); m; m = NEXT(m) ) {
2755: ndl_lcm(DL(m),d1,d2);
2756: t = d1; d1 = d2; d2 = t;
2757: }
2758: l = nd_nvar+31;
2759: t = (UINT *)MALLOC_ATOMIC(l*sizeof(UINT));
2760: for ( i = nd_exporigin, ind = 0; i < nd_wpd; i++ ) {
2761: u = d1[i];
2762: k = (nd_epw-1)*nd_bpe;
2763: for ( j = 0; j < nd_epw; j++, k -= nd_bpe, ind++ )
2764: t[ind] = (u>>k)&nd_mask0;
2765: }
2766: for ( ; ind < l; ind++ ) t[ind] = 0;
2767: return t;
2768: }
2769:
1.48 noro 2770: int nd_get_exporigin(struct order_spec *ord)
2771: {
1.51 noro 2772: switch ( ord->id ) {
1.96 noro 2773: case 0: case 2:
1.48 noro 2774: return 1;
1.41 noro 2775: case 1:
2776: /* block order */
1.43 noro 2777: /* d[0]:weight d[1]:w0,...,d[nd_exporigin-1]:w(n-1) */
1.48 noro 2778: return ord->ord.block.length+1;
1.96 noro 2779: case 3:
2780: error("nd_get_exporigin : composite order is not supported yet.");
1.41 noro 2781: }
1.48 noro 2782: }
2783:
1.61 noro 2784: void nd_setup_parameters(int nvar,int max) {
1.79 noro 2785: int i,j,n,elen,ord_o,ord_l,l,s,wpd;
1.57 noro 2786: struct order_pair *op;
1.48 noro 2787:
1.73 noro 2788: nd_nvar = nvar;
1.79 noro 2789: if ( max ) {
1.83 noro 2790: /* XXX */
2791: if ( do_weyl ) nd_bpe = 32;
2792: else if ( max < 2 ) nd_bpe = 1;
1.79 noro 2793: else if ( max < 4 ) nd_bpe = 2;
2794: else if ( max < 8 ) nd_bpe = 3;
2795: else if ( max < 16 ) nd_bpe = 4;
2796: else if ( max < 32 ) nd_bpe = 5;
2797: else if ( max < 64 ) nd_bpe = 6;
2798: else if ( max < 256 ) nd_bpe = 8;
2799: else if ( max < 1024 ) nd_bpe = 10;
2800: else if ( max < 65536 ) nd_bpe = 16;
2801: else nd_bpe = 32;
2802: }
1.61 noro 2803: nd_epw = (sizeof(UINT)*8)/nd_bpe;
1.48 noro 2804: elen = nd_nvar/nd_epw+(nd_nvar%nd_epw?1:0);
2805: nd_exporigin = nd_get_exporigin(nd_ord);
1.79 noro 2806: wpd = nd_exporigin+elen;
2807: if ( wpd != nd_wpd ) {
2808: nd_free_private_storage();
2809: nd_wpd = wpd;
2810: }
1.1 noro 2811: if ( nd_bpe < 32 ) {
2812: nd_mask0 = (1<<nd_bpe)-1;
2813: } else {
2814: nd_mask0 = 0xffffffff;
2815: }
2816: bzero(nd_mask,sizeof(nd_mask));
2817: nd_mask1 = 0;
2818: for ( i = 0; i < nd_epw; i++ ) {
2819: nd_mask[nd_epw-i-1] = (nd_mask0<<(i*nd_bpe));
2820: nd_mask1 |= (1<<(nd_bpe-1))<<(i*nd_bpe);
2821: }
1.90 noro 2822: nmv_adv = ROUND_FOR_ALIGN(sizeof(struct oNMV)+(nd_wpd-1)*sizeof(UINT));
1.57 noro 2823: nd_epos = nd_create_epos(nd_ord);
1.43 noro 2824: nd_blockmask = nd_create_blockmask(nd_ord);
1.96 noro 2825: nd_work_vector = (int *)REALLOC(nd_work_vector,nd_nvar*sizeof(int));
1.1 noro 2826: }
2827:
1.103 noro 2828: ND_pairs nd_reconstruct(int trace,ND_pairs d)
1.1 noro 2829: {
1.37 noro 2830: int i,obpe,oadv,h;
1.71 noro 2831: static NM prev_nm_free_list;
2832: static ND_pairs prev_ndp_free_list;
1.13 noro 2833: RHist mr0,mr;
2834: RHist r;
1.37 noro 2835: RHist *old_red;
1.71 noro 2836: ND_pairs s0,s,t;
1.43 noro 2837: EPOS oepos;
1.15 noro 2838:
1.1 noro 2839: obpe = nd_bpe;
1.11 noro 2840: oadv = nmv_adv;
1.43 noro 2841: oepos = nd_epos;
1.63 noro 2842: if ( obpe < 2 ) nd_bpe = 2;
2843: else if ( obpe < 3 ) nd_bpe = 3;
1.62 noro 2844: else if ( obpe < 4 ) nd_bpe = 4;
1.63 noro 2845: else if ( obpe < 5 ) nd_bpe = 5;
1.34 noro 2846: else if ( obpe < 6 ) nd_bpe = 6;
2847: else if ( obpe < 8 ) nd_bpe = 8;
1.63 noro 2848: else if ( obpe < 10 ) nd_bpe = 10;
1.34 noro 2849: else if ( obpe < 16 ) nd_bpe = 16;
2850: else if ( obpe < 32 ) nd_bpe = 32;
2851: else error("nd_reconstruct : exponent too large");
1.5 noro 2852:
1.73 noro 2853: nd_setup_parameters(nd_nvar,0);
1.1 noro 2854: prev_nm_free_list = _nm_free_list;
2855: prev_ndp_free_list = _ndp_free_list;
2856: _nm_free_list = 0;
2857: _ndp_free_list = 0;
1.53 noro 2858: for ( i = nd_psn-1; i >= 0; i-- ) ndv_realloc(nd_ps[i],obpe,oadv,oepos);
2859: if ( trace )
2860: for ( i = nd_psn-1; i >= 0; i-- )
2861: ndv_realloc(nd_ps_trace[i],obpe,oadv,oepos);
1.1 noro 2862: s0 = 0;
2863: for ( t = d; t; t = NEXT(t) ) {
2864: NEXTND_pairs(s0,s);
2865: s->i1 = t->i1;
2866: s->i2 = t->i2;
1.14 noro 2867: SG(s) = SG(t);
1.61 noro 2868: ndl_reconstruct(LCM(t),LCM(s),obpe,oepos);
1.1 noro 2869: }
1.37 noro 2870:
2871: old_red = (RHist *)ALLOCA(REDTAB_LEN*sizeof(RHist));
1.6 noro 2872: for ( i = 0; i < REDTAB_LEN; i++ ) {
1.37 noro 2873: old_red[i] = nd_red[i];
2874: nd_red[i] = 0;
2875: }
2876: for ( i = 0; i < REDTAB_LEN; i++ )
2877: for ( r = old_red[i]; r; r = NEXT(r) ) {
2878: NEWRHist(mr);
1.13 noro 2879: mr->index = r->index;
1.20 noro 2880: SG(mr) = SG(r);
1.61 noro 2881: ndl_reconstruct(DL(r),DL(mr),obpe,oepos);
1.37 noro 2882: h = ndl_hash_value(DL(mr));
2883: NEXT(mr) = nd_red[h];
2884: nd_red[h] = mr;
1.6 noro 2885: }
1.37 noro 2886: for ( i = 0; i < REDTAB_LEN; i++ ) old_red[i] = 0;
2887: old_red = 0;
1.11 noro 2888: for ( i = 0; i < nd_psn; i++ ) {
1.20 noro 2889: NEWRHist(r); SG(r) = SG(nd_psh[i]);
1.61 noro 2890: ndl_reconstruct(DL(nd_psh[i]),DL(r),obpe,oepos);
1.13 noro 2891: nd_psh[i] = r;
1.11 noro 2892: }
1.1 noro 2893: if ( s0 ) NEXT(s) = 0;
2894: prev_nm_free_list = 0;
2895: prev_ndp_free_list = 0;
1.71 noro 2896: #if 0
1.1 noro 2897: GC_gcollect();
1.71 noro 2898: #endif
1.1 noro 2899: return s0;
2900: }
2901:
1.61 noro 2902: void ndl_reconstruct(UINT *d,UINT *r,int obpe,EPOS oepos)
1.1 noro 2903: {
1.57 noro 2904: int n,i,ei,oepw,omask0,j,s,ord_l,l;
1.43 noro 2905: struct order_pair *op;
1.1 noro 2906:
2907: n = nd_nvar;
1.61 noro 2908: oepw = (sizeof(UINT)*8)/obpe;
1.43 noro 2909: omask0 = (1<<obpe)-1;
1.34 noro 2910: TD(r) = TD(d);
1.41 noro 2911: for ( i = nd_exporigin; i < nd_wpd; i++ ) r[i] = 0;
1.43 noro 2912: if ( nd_blockmask ) {
2913: l = nd_blockmask->n;
2914: op = nd_blockmask->order_pair;
2915: for ( i = 1; i < nd_exporigin; i++ )
2916: r[i] = d[i];
2917: for ( j = 0, s = 0; j < l; j++ ) {
2918: ord_l = op[j].length;
1.57 noro 2919: for ( i = 0; i < ord_l; i++, s++ ) {
2920: ei = GET_EXP_OLD(d,s);
2921: PUT_EXP(r,s,ei);
2922: }
1.1 noro 2923: }
1.43 noro 2924: } else {
1.56 noro 2925: for ( i = 0; i < n; i++ ) {
2926: ei = GET_EXP_OLD(d,i);
2927: PUT_EXP(r,i,ei);
2928: }
1.1 noro 2929: }
2930: }
1.3 noro 2931:
1.6 noro 2932: ND nd_copy(ND p)
2933: {
2934: NM m,mr,mr0;
1.41 noro 2935: int c,n;
1.6 noro 2936: ND r;
2937:
2938: if ( !p )
2939: return 0;
2940: else {
2941: for ( mr0 = 0, m = BDY(p); m; m = NEXT(m) ) {
2942: NEXTNM(mr0,mr);
1.14 noro 2943: CM(mr) = CM(m);
2944: ndl_copy(DL(m),DL(mr));
1.6 noro 2945: }
2946: NEXT(mr) = 0;
1.31 noro 2947: MKND(NV(p),mr0,LEN(p),r);
1.14 noro 2948: SG(r) = SG(p);
1.6 noro 2949: return r;
2950: }
2951: }
2952:
1.53 noro 2953: int nd_sp(int mod,int trace,ND_pairs p,ND *rp)
1.11 noro 2954: {
2955: NM m;
2956: NDV p1,p2;
2957: ND t1,t2;
1.61 noro 2958: UINT *lcm;
1.31 noro 2959: int td;
1.11 noro 2960:
1.77 noro 2961: if ( !mod && nd_demand ) {
2962: p1 = ndv_load(p->i1); p2 = ndv_load(p->i2);
1.53 noro 2963: } else {
1.77 noro 2964: if ( trace ) {
2965: p1 = nd_ps_trace[p->i1]; p2 = nd_ps_trace[p->i2];
2966: } else {
2967: p1 = nd_ps[p->i1]; p2 = nd_ps[p->i2];
2968: }
1.20 noro 2969: }
1.34 noro 2970: lcm = LCM(p);
1.11 noro 2971: NEWNM(m);
1.112 ! noro 2972: CZ(m) = HCZ(p2);
1.34 noro 2973: ndl_sub(lcm,HDL(p1),DL(m));
1.56 noro 2974: if ( ndl_check_bound2(p->i1,DL(m)) )
2975: return 0;
1.55 noro 2976: t1 = ndv_mul_nm(mod,m,p1);
1.71 noro 2977: if ( mod == -1 ) CM(m) = _chsgnsf(HCM(p1));
2978: else if ( mod ) CM(m) = mod-HCM(p1);
1.112 ! noro 2979: else CZ(m) = chsgnz(HCZ(p1));
1.34 noro 2980: ndl_sub(lcm,HDL(p2),DL(m));
1.14 noro 2981: if ( ndl_check_bound2(p->i2,DL(m)) ) {
1.11 noro 2982: nd_free(t1);
2983: return 0;
2984: }
1.55 noro 2985: t2 = ndv_mul_nm(mod,m,p2);
1.31 noro 2986: *rp = nd_add(mod,t1,t2);
1.11 noro 2987: FREENM(m);
2988: return 1;
2989: }
2990:
1.19 noro 2991: void ndv_mul_c(int mod,NDV p,int mul)
1.11 noro 2992: {
2993: NMV m;
2994: int c,c1,len,i;
2995:
1.34 noro 2996: if ( !p ) return;
1.14 noro 2997: len = LEN(p);
1.71 noro 2998: if ( mod == -1 )
2999: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) )
3000: CM(m) = _mulsf(CM(m),mul);
3001: else
3002: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
3003: c1 = CM(m); DMAR(c1,mul,0,mod,c); CM(m) = c;
3004: }
1.11 noro 3005: }
3006:
1.112 ! noro 3007: void ndv_mul_c_z(NDV p,Z mul)
1.16 noro 3008: {
3009: NMV m;
1.112 ! noro 3010: Z c;
1.16 noro 3011: int len,i;
3012:
1.34 noro 3013: if ( !p ) return;
1.16 noro 3014: len = LEN(p);
3015: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
1.112 ! noro 3016: CZ(m) = mulz(CZ(m),mul);
1.16 noro 3017: }
3018: }
3019:
1.55 noro 3020: ND weyl_ndv_mul_nm(int mod,NM m0,NDV p) {
3021: int n2,i,j,l,n,tlen;
1.61 noro 3022: UINT *d0;
1.55 noro 3023: NM *tab,*psum;
3024: ND s,r;
3025: NM t;
3026: NMV m1;
3027:
3028: if ( !p ) return 0;
3029: n = NV(p); n2 = n>>1;
3030: d0 = DL(m0);
3031: l = LEN(p);
3032: for ( i = 0, tlen = 1; i < n2; i++ ) tlen *= (GET_EXP(d0,n2+i)+1);
3033: tab = (NM *)ALLOCA(tlen*sizeof(NM));
3034: psum = (NM *)ALLOCA(tlen*sizeof(NM));
3035: for ( i = 0; i < tlen; i++ ) psum[i] = 0;
1.56 noro 3036: m1 = (NMV)(((char *)BDY(p))+nmv_adv*(l-1));
3037: for ( i = l-1; i >= 0; i--, NMV_PREV(m1) ) {
1.55 noro 3038: /* m0(NM) * m1(NMV) => tab(NM) */
1.56 noro 3039: weyl_mul_nm_nmv(n,mod,m0,m1,tab,tlen);
1.55 noro 3040: for ( j = 0; j < tlen; j++ ) {
3041: if ( tab[j] ) {
3042: NEXT(tab[j]) = psum[j]; psum[j] = tab[j];
3043: }
3044: }
3045: }
3046: for ( i = tlen-1, r = 0; i >= 0; i-- )
3047: if ( psum[i] ) {
3048: for ( j = 0, t = psum[i]; t; t = NEXT(t), j++ );
3049: MKND(n,psum[i],j,s);
3050: r = nd_add(mod,r,s);
3051: }
1.56 noro 3052: if ( r ) SG(r) = SG(p)+TD(d0);
3053: return r;
1.55 noro 3054: }
3055:
1.56 noro 3056: /* product of monomials */
3057: /* XXX block order is not handled correctly */
3058:
1.55 noro 3059: void weyl_mul_nm_nmv(int n,int mod,NM m0,NMV m1,NM *tab,int tlen)
3060: {
1.56 noro 3061: int i,n2,j,s,curlen,homo,h,a,b,k,l,u,min;
1.61 noro 3062: UINT *d0,*d1,*d,*dt,*ctab;
1.112 ! noro 3063: Z *ctab_z;
! 3064: Z q,q1;
1.61 noro 3065: UINT c0,c1,c;
1.55 noro 3066: NM *p;
3067: NM m,t;
3068:
3069: for ( i = 0; i < tlen; i++ ) tab[i] = 0;
3070: if ( !m0 || !m1 ) return;
3071: d0 = DL(m0); d1 = DL(m1); n2 = n>>1;
3072: NEWNM(m); d = DL(m);
1.56 noro 3073: if ( mod ) {
3074: c0 = CM(m0); c1 = CM(m1); DMAR(c0,c1,0,mod,c); CM(m) = c;
3075: } else
1.112 ! noro 3076: CZ(m) = mulz(CZ(m0),CZ(m1));
1.55 noro 3077: for ( i = 0; i < nd_wpd; i++ ) d[i] = 0;
3078: homo = n&1 ? 1 : 0;
3079: if ( homo ) {
3080: /* offset of h-degree */
3081: h = GET_EXP(d0,n-1)+GET_EXP(d1,n-1);
3082: PUT_EXP(DL(m),n-1,h);
3083: TD(DL(m)) = h;
1.61 noro 3084: if ( nd_blockmask ) ndl_weight_mask(DL(m));
1.55 noro 3085: }
3086: tab[0] = m;
3087: NEWNM(m); d = DL(m);
1.57 noro 3088: for ( i = 0, curlen = 1; i < n2; i++ ) {
1.55 noro 3089: a = GET_EXP(d0,i); b = GET_EXP(d1,n2+i);
3090: k = GET_EXP(d0,n2+i); l = GET_EXP(d1,i);
3091: /* xi^a*(Di^k*xi^l)*Di^b */
3092: a += l; b += k;
1.56 noro 3093: s = MUL_WEIGHT(a,i)+MUL_WEIGHT(b,n2+i);
1.55 noro 3094: if ( !k || !l ) {
3095: for ( j = 0; j < curlen; j++ )
1.56 noro 3096: if ( t = tab[j] ) {
3097: dt = DL(t);
3098: PUT_EXP(dt,i,a); PUT_EXP(dt,n2+i,b); TD(dt) += s;
1.61 noro 3099: if ( nd_blockmask ) ndl_weight_mask(dt);
1.55 noro 3100: }
3101: curlen *= k+1;
3102: continue;
3103: }
3104: min = MIN(k,l);
1.56 noro 3105: if ( mod ) {
1.61 noro 3106: ctab = (UINT *)ALLOCA((min+1)*sizeof(UINT));
1.56 noro 3107: mkwcm(k,l,mod,ctab);
3108: } else {
1.112 ! noro 3109: ctab_z = (Z *)ALLOCA((min+1)*sizeof(Z));
! 3110: mkwcz(k,l,ctab_z);
1.56 noro 3111: }
1.57 noro 3112: for ( j = min; j >= 0; j-- ) {
1.56 noro 3113: for ( u = 0; u < nd_wpd; u++ ) d[u] = 0;
1.55 noro 3114: PUT_EXP(d,i,a-j); PUT_EXP(d,n2+i,b-j);
1.56 noro 3115: h = MUL_WEIGHT(a-j,i)+MUL_WEIGHT(b-j,n2+i);
1.55 noro 3116: if ( homo ) {
3117: TD(d) = s;
1.56 noro 3118: PUT_EXP(d,n-1,s-h);
1.55 noro 3119: } else TD(d) = h;
1.61 noro 3120: if ( nd_blockmask ) ndl_weight_mask(d);
1.56 noro 3121: if ( mod ) c = ctab[j];
1.112 ! noro 3122: else q = ctab_z[j];
1.57 noro 3123: p = tab+curlen*j;
3124: if ( j == 0 ) {
3125: for ( u = 0; u < curlen; u++, p++ ) {
3126: if ( tab[u] ) {
3127: ndl_addto(DL(tab[u]),d);
3128: if ( mod ) {
3129: c0 = CM(tab[u]); DMAR(c0,c,0,mod,c1); CM(tab[u]) = c1;
3130: } else {
1.112 ! noro 3131: CZ(tab[u]) = mulz(CZ(tab[u]),q);
1.57 noro 3132: }
3133: }
1.56 noro 3134: }
1.57 noro 3135: } else {
3136: for ( u = 0; u < curlen; u++, p++ ) {
3137: if ( tab[u] ) {
3138: NEWNM(t);
3139: ndl_add(DL(tab[u]),d,DL(t));
3140: if ( mod ) {
3141: c0 = CM(tab[u]); DMAR(c0,c,0,mod,c1); CM(t) = c1;
3142: } else
1.112 ! noro 3143: CZ(t) = mulz(CZ(tab[u]),q);
1.57 noro 3144: *p = t;
3145: }
1.55 noro 3146: }
3147: }
3148: }
3149: curlen *= k+1;
3150: }
3151: FREENM(m);
3152: }
3153:
1.63 noro 3154: ND ndv_mul_nm_symbolic(NM m0,NDV p)
3155: {
3156: NM mr,mr0;
3157: NMV m;
3158: UINT *d,*dt,*dm;
3159: int c,n,td,i,c1,c2,len;
3160: ND r;
3161:
3162: if ( !p ) return 0;
3163: else {
3164: n = NV(p); m = BDY(p);
3165: d = DL(m0);
3166: len = LEN(p);
3167: mr0 = 0;
3168: td = TD(d);
3169: c = CM(m0);
3170: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
3171: NEXTNM(mr0,mr);
3172: CM(mr) = 1;
3173: ndl_add(DL(m),d,DL(mr));
3174: }
3175: NEXT(mr) = 0;
3176: MKND(NV(p),mr0,len,r);
3177: SG(r) = SG(p) + TD(d);
3178: return r;
3179: }
3180: }
3181:
1.55 noro 3182: ND ndv_mul_nm(int mod,NM m0,NDV p)
1.9 noro 3183: {
3184: NM mr,mr0;
3185: NMV m;
1.61 noro 3186: UINT *d,*dt,*dm;
1.9 noro 3187: int c,n,td,i,c1,c2,len;
1.112 ! noro 3188: Z q;
1.9 noro 3189: ND r;
3190:
1.34 noro 3191: if ( !p ) return 0;
1.55 noro 3192: else if ( do_weyl )
1.71 noro 3193: if ( mod == -1 )
3194: error("ndv_mul_nm : not implemented (weyl)");
3195: else
3196: return weyl_ndv_mul_nm(mod,m0,p);
1.9 noro 3197: else {
3198: n = NV(p); m = BDY(p);
1.34 noro 3199: d = DL(m0);
1.14 noro 3200: len = LEN(p);
1.9 noro 3201: mr0 = 0;
1.34 noro 3202: td = TD(d);
1.71 noro 3203: if ( mod == -1 ) {
3204: c = CM(m0);
3205: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
3206: NEXTNM(mr0,mr);
3207: CM(mr) = _mulsf(CM(m),c);
3208: ndl_add(DL(m),d,DL(mr));
3209: }
3210: } else if ( mod ) {
1.16 noro 3211: c = CM(m0);
3212: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
3213: NEXTNM(mr0,mr);
3214: c1 = CM(m);
1.19 noro 3215: DMAR(c1,c,0,mod,c2);
1.16 noro 3216: CM(mr) = c2;
3217: ndl_add(DL(m),d,DL(mr));
3218: }
3219: } else {
1.112 ! noro 3220: q = CZ(m0);
1.16 noro 3221: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
3222: NEXTNM(mr0,mr);
1.112 ! noro 3223: CZ(mr) = mulz(CZ(m),q);
1.16 noro 3224: ndl_add(DL(m),d,DL(mr));
3225: }
1.4 noro 3226: }
1.9 noro 3227: NEXT(mr) = 0;
1.31 noro 3228: MKND(NV(p),mr0,len,r);
1.34 noro 3229: SG(r) = SG(p) + TD(d);
1.9 noro 3230: return r;
1.4 noro 3231: }
3232: }
3233:
1.104 noro 3234: ND nd_quo(int mod,PGeoBucket bucket,NDV d)
1.99 noro 3235: {
3236: NM mq0,mq;
1.102 noro 3237: NMV tm;
1.104 noro 3238: int i,nv,sg,c,c1,c2,hindex;
3239: ND p,t,r;
3240: N tnm;
3241:
1.99 noro 3242: if ( !p ) return 0;
3243: else {
1.104 noro 3244: nv = NV(d);
1.99 noro 3245: mq0 = 0;
1.102 noro 3246: tm = (NMV)ALLOCA(nmv_adv);
1.104 noro 3247: while ( 1 ) {
1.112 ! noro 3248: hindex = mod?head_pbucket(mod,bucket):head_pbucket_z(bucket);
1.104 noro 3249: if ( hindex < 0 ) break;
3250: p = bucket->body[hindex];
1.99 noro 3251: NEXTNM(mq0,mq);
3252: ndl_sub(HDL(p),HDL(d),DL(mq));
1.102 noro 3253: ndl_copy(DL(mq),DL(tm));
3254: if ( mod ) {
3255: c1 = invm(HCM(d),mod); c2 = HCM(p);
3256: DMAR(c1,c2,0,mod,c); CM(mq) = c;
3257: CM(tm) = mod-c;
3258: } else {
1.112 ! noro 3259: CZ(mq) = divsz(HCZ(p),HCZ(d));
! 3260: CZ(tm) = chsgnz(CZ(mq));
1.102 noro 3261: }
3262: t = ndv_mul_nmv_trunc(mod,tm,d,HDL(d));
1.104 noro 3263: bucket->body[hindex] = nd_remove_head(p);
3264: t = nd_remove_head(t);
3265: add_pbucket(mod,bucket,t);
3266: }
3267: if ( !mq0 )
3268: r = 0;
3269: else {
3270: NEXT(mq) = 0;
3271: for ( i = 0, mq = mq0; mq; mq = NEXT(mq), i++ );
3272: MKND(nv,mq0,i,r);
3273: /* XXX */
3274: SG(r) = HTD(r);
1.99 noro 3275: }
3276: return r;
3277: }
3278: }
3279:
1.43 noro 3280: void ndv_realloc(NDV p,int obpe,int oadv,EPOS oepos)
1.11 noro 3281: {
1.13 noro 3282: NMV m,mr,mr0,t;
3283: int len,i,k;
1.11 noro 3284:
1.61 noro 3285: if ( !p ) return;
3286: m = BDY(p); len = LEN(p);
3287: mr0 = nmv_adv>oadv?(NMV)REALLOC(BDY(p),len*nmv_adv):BDY(p);
3288: m = (NMV)((char *)mr0+(len-1)*oadv);
3289: mr = (NMV)((char *)mr0+(len-1)*nmv_adv);
3290: t = (NMV)ALLOCA(nmv_adv);
3291: for ( i = 0; i < len; i++, NMV_OPREV(m), NMV_PREV(mr) ) {
1.112 ! noro 3292: CZ(t) = CZ(m);
1.61 noro 3293: for ( k = 0; k < nd_wpd; k++ ) DL(t)[k] = 0;
3294: ndl_reconstruct(DL(m),DL(t),obpe,oepos);
1.112 ! noro 3295: CZ(mr) = CZ(t);
1.61 noro 3296: ndl_copy(DL(t),DL(mr));
3297: }
3298: BDY(p) = mr0;
3299: }
3300:
3301: NDV ndv_dup_realloc(NDV p,int obpe,int oadv,EPOS oepos)
3302: {
3303: NMV m,mr,mr0;
3304: int len,i;
3305: NDV r;
1.11 noro 3306:
1.61 noro 3307: if ( !p ) return 0;
3308: m = BDY(p); len = LEN(p);
3309: mr0 = mr = (NMV)MALLOC(len*nmv_adv);
3310: for ( i = 0; i < len; i++, NMV_OADV(m), NMV_ADV(mr) ) {
3311: ndl_zero(DL(mr));
3312: ndl_reconstruct(DL(m),DL(mr),obpe,oepos);
1.112 ! noro 3313: CZ(mr) = CZ(m);
1.11 noro 3314: }
1.61 noro 3315: MKNDV(NV(p),mr0,len,r);
3316: SG(r) = SG(p);
3317: return r;
1.11 noro 3318: }
3319:
1.61 noro 3320: /* duplicate p */
3321:
3322: NDV ndv_dup(int mod,NDV p)
1.3 noro 3323: {
3324: NDV d;
1.61 noro 3325: NMV t,m,m0;
1.3 noro 3326: int i,len;
3327:
1.34 noro 3328: if ( !p ) return 0;
1.31 noro 3329: len = LEN(p);
1.34 noro 3330: m0 = m = (NMV)(mod?MALLOC_ATOMIC(len*nmv_adv):MALLOC(len*nmv_adv));
1.61 noro 3331: for ( t = BDY(p), i = 0; i < len; i++, NMV_ADV(t), NMV_ADV(m) ) {
1.14 noro 3332: ndl_copy(DL(t),DL(m));
1.112 ! noro 3333: CZ(m) = CZ(t);
1.3 noro 3334: }
3335: MKNDV(NV(p),m0,len,d);
1.23 noro 3336: SG(d) = SG(p);
3337: return d;
3338: }
3339:
1.63 noro 3340: ND nd_dup(ND p)
3341: {
3342: ND d;
3343: NM t,m,m0;
3344:
3345: if ( !p ) return 0;
3346: for ( m0 = 0, t = BDY(p); t; t = NEXT(t) ) {
3347: NEXTNM(m0,m);
3348: ndl_copy(DL(t),DL(m));
1.112 ! noro 3349: CZ(m) = CZ(t);
1.63 noro 3350: }
3351: if ( m0 ) NEXT(m) = 0;
3352: MKND(NV(p),m0,LEN(p),d);
3353: SG(d) = SG(p);
3354: return d;
3355: }
3356:
1.61 noro 3357: /* XXX if p->len == 0 then it represents 0 */
3358:
3359: void ndv_mod(int mod,NDV p)
3360: {
3361: NMV t,d;
3362: int r;
3363: int i,len,dlen;
1.71 noro 3364: Obj gfs;
1.61 noro 3365:
3366: if ( !p ) return;
3367: len = LEN(p);
3368: dlen = 0;
1.71 noro 3369: if ( mod == -1 )
3370: for ( t = d = BDY(p), i = 0; i < len; i++, NMV_ADV(t) ) {
3371: simp_ff((Obj)CP(t),&gfs);
3372: r = FTOIF(CONT((GFS)gfs));
1.61 noro 3373: CM(d) = r;
3374: ndl_copy(DL(t),DL(d));
3375: NMV_ADV(d);
3376: dlen++;
3377: }
1.71 noro 3378: else
3379: for ( t = d = BDY(p), i = 0; i < len; i++, NMV_ADV(t) ) {
1.112 ! noro 3380: r = remzi(CZ(t),mod);
1.71 noro 3381: if ( r ) {
3382: CM(d) = r;
3383: ndl_copy(DL(t),DL(d));
3384: NMV_ADV(d);
3385: dlen++;
3386: }
3387: }
1.61 noro 3388: LEN(p) = dlen;
3389: }
3390:
3391: NDV ptondv(VL vl,VL dvl,P p)
3392: {
3393: ND nd;
3394:
3395: nd = ptond(vl,dvl,p);
3396: return ndtondv(0,nd);
3397: }
3398:
3399: ND ptond(VL vl,VL dvl,P p)
1.23 noro 3400: {
1.61 noro 3401: int n,i,j,k,e;
3402: VL tvl;
3403: V v;
3404: DCP dc;
3405: DCP *w;
3406: ND r,s,t,u;
3407: P x;
3408: int c;
3409: UINT *d;
1.23 noro 3410: NM m,m0;
1.61 noro 3411:
3412: if ( !p )
3413: return 0;
3414: else if ( NUM(p) ) {
3415: NEWNM(m);
3416: ndl_zero(DL(m));
1.112 ! noro 3417: CZ(m) = qtoz((Q)p);
1.61 noro 3418: NEXT(m) = 0;
3419: MKND(nd_nvar,m,1,r);
3420: SG(r) = 0;
3421: return r;
3422: } else {
3423: for ( dc = DC(p), k = 0; dc; dc = NEXT(dc), k++ );
3424: w = (DCP *)ALLOCA(k*sizeof(DCP));
3425: for ( dc = DC(p), j = 0; j < k; dc = NEXT(dc), j++ ) w[j] = dc;
3426: for ( i = 0, tvl = dvl, v = VR(p);
1.98 noro 3427: tvl && tvl->v != v; tvl = NEXT(tvl), i++ );
1.61 noro 3428: if ( !tvl ) {
3429: for ( j = k-1, s = 0, MKV(v,x); j >= 0; j-- ) {
3430: t = ptond(vl,dvl,COEF(w[j]));
3431: pwrp(vl,x,DEG(w[j]),&p);
3432: nd_mul_c_p(CO,t,p); s = nd_add(0,s,t);
3433: }
3434: return s;
3435: } else {
3436: NEWNM(m0); d = DL(m0);
3437: for ( j = k-1, s = 0; j >= 0; j-- ) {
3438: ndl_zero(d); e = QTOS(DEG(w[j])); PUT_EXP(d,i,e);
3439: TD(d) = MUL_WEIGHT(e,i);
3440: if ( nd_blockmask) ndl_weight_mask(d);
3441: t = ptond(vl,dvl,COEF(w[j]));
3442: for ( m = BDY(t); m; m = NEXT(m) )
3443: ndl_addto(DL(m),d);
3444: SG(t) += TD(d);
3445: s = nd_add(0,s,t);
3446: }
3447: FREENM(m0);
3448: return s;
3449: }
3450: }
3451: }
3452:
3453: P ndvtop(int mod,VL vl,VL dvl,NDV p)
3454: {
3455: VL tvl;
3456: int len,n,j,i,e;
3457: NMV m;
3458: Q q;
3459: P c;
3460: UINT *d;
3461: P s,r,u,t,w;
1.71 noro 3462: GFS gfs;
1.23 noro 3463:
1.34 noro 3464: if ( !p ) return 0;
1.61 noro 3465: else {
3466: len = LEN(p);
3467: n = NV(p);
3468: m = (NMV)(((char *)BDY(p))+nmv_adv*(len-1));
3469: for ( j = len-1, s = 0; j >= 0; j--, NMV_PREV(m) ) {
1.71 noro 3470: if ( mod == -1 ) {
3471: e = IFTOF(CM(m)); MKGFS(e,gfs); c = (P)gfs;
3472: } else if ( mod ) {
1.61 noro 3473: STOQ(CM(m),q); c = (P)q;
3474: } else
1.112 ! noro 3475: c = (P)ztoq(CZ(m));
1.61 noro 3476: d = DL(m);
3477: for ( i = 0, t = c, tvl = dvl; i < n; tvl = NEXT(tvl), i++ ) {
3478: MKV(tvl->v,r); e = GET_EXP(d,i); STOQ(e,q);
3479: pwrp(vl,r,q,&u); mulp(vl,t,u,&w); t = w;
3480: }
3481: addp(vl,s,t,&u); s = u;
3482: }
3483: return s;
1.23 noro 3484: }
1.3 noro 3485: }
3486:
1.61 noro 3487: NDV ndtondv(int mod,ND p)
1.11 noro 3488: {
3489: NDV d;
1.61 noro 3490: NMV m,m0;
3491: NM t;
3492: int i,len;
1.11 noro 3493:
1.34 noro 3494: if ( !p ) return 0;
1.61 noro 3495: len = LEN(p);
1.74 noro 3496: if ( mod )
3497: m0 = m = (NMV)GC_malloc_atomic_ignore_off_page(len*nmv_adv);
3498: else
3499: m0 = m = MALLOC(len*nmv_adv);
1.103 noro 3500: #if 0
1.74 noro 3501: ndv_alloc += nmv_adv*len;
1.103 noro 3502: #endif
1.61 noro 3503: for ( t = BDY(p), i = 0; t; t = NEXT(t), i++, NMV_ADV(m) ) {
3504: ndl_copy(DL(t),DL(m));
1.112 ! noro 3505: CZ(m) = CZ(t);
1.11 noro 3506: }
1.61 noro 3507: MKNDV(NV(p),m0,len,d);
1.14 noro 3508: SG(d) = SG(p);
1.11 noro 3509: return d;
3510: }
3511:
1.61 noro 3512: ND ndvtond(int mod,NDV p)
1.11 noro 3513: {
1.61 noro 3514: ND d;
3515: NM m,m0;
1.11 noro 3516: NMV t;
1.61 noro 3517: int i,len;
1.11 noro 3518:
1.34 noro 3519: if ( !p ) return 0;
1.11 noro 3520: m0 = 0;
1.61 noro 3521: len = p->len;
3522: for ( t = BDY(p), i = 0; i < len; NMV_ADV(t), i++ ) {
3523: NEXTNM(m0,m);
3524: ndl_copy(DL(t),DL(m));
1.112 ! noro 3525: CZ(m) = CZ(t);
1.11 noro 3526: }
3527: NEXT(m) = 0;
1.61 noro 3528: MKND(NV(p),m0,len,d);
1.14 noro 3529: SG(d) = SG(p);
1.11 noro 3530: return d;
3531: }
3532:
1.3 noro 3533: void ndv_print(NDV p)
3534: {
3535: NMV m;
3536: int i,len;
3537:
1.34 noro 3538: if ( !p ) printf("0\n");
1.3 noro 3539: else {
1.14 noro 3540: len = LEN(p);
1.3 noro 3541: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
1.71 noro 3542: if ( CM(m) & 0x80000000 ) printf("+@_%d*",IFTOF(CM(m)));
3543: else printf("+%d*",CM(m));
1.16 noro 3544: ndl_print(DL(m));
3545: }
3546: printf("\n");
3547: }
3548: }
3549:
1.112 ! noro 3550: void ndv_print_z(NDV p)
1.16 noro 3551: {
3552: NMV m;
3553: int i,len;
3554:
1.34 noro 3555: if ( !p ) printf("0\n");
1.16 noro 3556: else {
3557: len = LEN(p);
3558: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
3559: printf("+");
1.112 ! noro 3560: printz(CZ(m));
1.16 noro 3561: printf("*");
1.14 noro 3562: ndl_print(DL(m));
1.3 noro 3563: }
3564: printf("\n");
3565: }
1.25 noro 3566: }
3567:
1.61 noro 3568: NODE ndv_reducebase(NODE x)
1.27 noro 3569: {
3570: int len,i,j;
3571: NDV *w;
3572: NODE t,t0;
3573:
3574: len = length(x);
3575: w = (NDV *)ALLOCA(len*sizeof(NDV));
3576: for ( i = 0, t = x; i < len; i++, t = NEXT(t) ) w[i] = BDY(t);
3577: for ( i = 0; i < len; i++ ) {
3578: for ( j = 0; j < i; j++ ) {
3579: if ( w[i] && w[j] )
3580: if ( ndl_reducible(HDL(w[i]),HDL(w[j])) ) w[i] = 0;
3581: else if ( ndl_reducible(HDL(w[j]),HDL(w[i])) ) w[j] = 0;
3582: }
3583: }
3584: for ( i = len-1, t0 = 0; i >= 0; i-- ) {
3585: if ( w[i] ) { NEXTNODE(t0,t); BDY(t) = (pointer)w[i]; }
3586: }
3587: NEXT(t) = 0; x = t0;
3588: return x;
1.11 noro 3589: }
1.32 noro 3590:
1.43 noro 3591: /* XXX incomplete */
3592:
1.32 noro 3593: void nd_init_ord(struct order_spec *ord)
3594: {
1.43 noro 3595: switch ( ord->id ) {
1.32 noro 3596: case 0:
1.43 noro 3597: switch ( ord->ord.simple ) {
3598: case 0:
3599: nd_dcomp = 1;
3600: nd_isrlex = 1;
3601: break;
3602: case 1:
3603: nd_dcomp = 1;
3604: nd_isrlex = 0;
3605: break;
3606: case 2:
3607: nd_dcomp = 0;
3608: nd_isrlex = 0;
1.45 noro 3609: ndl_compare_function = ndl_lex_compare;
1.58 noro 3610: break;
3611: case 11:
3612: /* XXX */
3613: nd_dcomp = 0;
3614: nd_isrlex = 1;
3615: ndl_compare_function = ndl_ww_lex_compare;
1.43 noro 3616: break;
3617: default:
3618: error("nd_gr : unsupported order");
3619: }
1.32 noro 3620: break;
3621: case 1:
1.96 noro 3622: /* block order */
1.43 noro 3623: /* XXX */
3624: nd_dcomp = -1;
1.32 noro 3625: nd_isrlex = 0;
1.45 noro 3626: ndl_compare_function = ndl_block_compare;
1.34 noro 3627: break;
1.43 noro 3628: case 2:
1.96 noro 3629: /* matrix order */
3630: /* XXX */
3631: nd_dcomp = -1;
3632: nd_isrlex = 0;
3633: nd_matrix_len = ord->ord.matrix.row;
3634: nd_matrix = ord->ord.matrix.matrix;
3635: ndl_compare_function = ndl_matrix_compare;
3636: break;
3637: case 3:
1.97 noro 3638: /* composite order */
3639: nd_dcomp = -1;
3640: nd_isrlex = 0;
3641: nd_worb_len = ord->ord.composite.length;
3642: nd_worb = ord->ord.composite.w_or_b;
3643: ndl_compare_function = ndl_composite_compare;
1.32 noro 3644: break;
3645: }
1.41 noro 3646: nd_ord = ord;
1.32 noro 3647: }
3648:
1.43 noro 3649: BlockMask nd_create_blockmask(struct order_spec *ord)
3650: {
3651: int n,i,j,s,l;
1.61 noro 3652: UINT *t;
1.43 noro 3653: BlockMask bm;
3654:
1.96 noro 3655: /* we only create mask table for block order */
3656: if ( ord->id != 1 )
1.43 noro 3657: return 0;
3658: n = ord->ord.block.length;
3659: bm = (BlockMask)MALLOC(sizeof(struct oBlockMask));
3660: bm->n = n;
3661: bm->order_pair = ord->ord.block.order_pair;
1.61 noro 3662: bm->mask = (UINT **)MALLOC(n*sizeof(UINT *));
1.43 noro 3663: for ( i = 0, s = 0; i < n; i++ ) {
1.61 noro 3664: bm->mask[i] = t = (UINT *)MALLOC_ATOMIC(nd_wpd*sizeof(UINT));
1.43 noro 3665: for ( j = 0; j < nd_wpd; j++ ) t[j] = 0;
3666: l = bm->order_pair[i].length;
3667: for ( j = 0; j < l; j++, s++ ) PUT_EXP(t,s,nd_mask0);
3668: }
3669: return bm;
1.57 noro 3670: }
3671:
3672: EPOS nd_create_epos(struct order_spec *ord)
3673: {
3674: int i,j,l,s,ord_l,ord_o;
3675: EPOS epos;
3676: struct order_pair *op;
3677:
3678: epos = (EPOS)MALLOC_ATOMIC(nd_nvar*sizeof(struct oEPOS));
3679: switch ( ord->id ) {
3680: case 0:
3681: if ( nd_isrlex ) {
3682: for ( i = 0; i < nd_nvar; i++ ) {
3683: epos[i].i = nd_exporigin + (nd_nvar-1-i)/nd_epw;
3684: epos[i].s = (nd_epw-((nd_nvar-1-i)%nd_epw)-1)*nd_bpe;
3685: }
3686: } else {
3687: for ( i = 0; i < nd_nvar; i++ ) {
3688: epos[i].i = nd_exporigin + i/nd_epw;
3689: epos[i].s = (nd_epw-(i%nd_epw)-1)*nd_bpe;
3690: }
3691: }
3692: break;
3693: case 1:
3694: /* block order */
3695: l = ord->ord.block.length;
3696: op = ord->ord.block.order_pair;
3697: for ( j = 0, s = 0; j < l; j++ ) {
3698: ord_o = op[j].order;
3699: ord_l = op[j].length;
3700: if ( !ord_o )
3701: for ( i = 0; i < ord_l; i++ ) {
3702: epos[s+i].i = nd_exporigin + (s+ord_l-i-1)/nd_epw;
3703: epos[s+i].s = (nd_epw-((s+ord_l-i-1)%nd_epw)-1)*nd_bpe;
3704: }
3705: else
3706: for ( i = 0; i < ord_l; i++ ) {
3707: epos[s+i].i = nd_exporigin + (s+i)/nd_epw;
3708: epos[s+i].s = (nd_epw-((s+i)%nd_epw)-1)*nd_bpe;
3709: }
3710: s += ord_l;
3711: }
3712: break;
3713: case 2:
1.96 noro 3714: /* matrix order */
3715: case 3:
3716: /* composite order */
3717: for ( i = 0; i < nd_nvar; i++ ) {
3718: epos[i].i = nd_exporigin + i/nd_epw;
3719: epos[i].s = (nd_epw-(i%nd_epw)-1)*nd_bpe;
3720: }
3721: break;
1.57 noro 3722: }
3723: return epos;
1.43 noro 3724: }
1.59 noro 3725:
3726: /* external interface */
3727:
3728: void nd_nf_p(P f,LIST g,LIST v,int m,struct order_spec *ord,P *rp)
3729: {
1.61 noro 3730: NODE t,in0,in;
1.59 noro 3731: ND nd,nf;
1.61 noro 3732: NDV ndv;
3733: VL vv,tv;
3734: int stat,nvar,max,e;
1.59 noro 3735:
3736: pltovl(v,&vv);
1.74 noro 3737: for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++ );
1.61 noro 3738:
3739: /* get the degree bound */
3740: for ( t = BDY(g), max = 0; t; t = NEXT(t) )
3741: for ( tv = vv; tv; tv = NEXT(tv) ) {
3742: e = getdeg(tv->v,(P)BDY(t));
3743: max = MAX(e,max);
3744: }
3745: for ( tv = vv; tv; tv = NEXT(tv) ) {
3746: e = getdeg(tv->v,f);
3747: max = MAX(e,max);
3748: }
3749:
1.59 noro 3750: nd_init_ord(ord);
1.61 noro 3751: nd_setup_parameters(nvar,max);
3752:
3753: /* conversion to ndv */
3754: for ( in0 = 0, t = BDY(g); t; t = NEXT(t) ) {
3755: NEXTNODE(in0,in);
3756: BDY(in) = (pointer)ptondv(CO,vv,(P)BDY(t));
1.69 noro 3757: if ( m ) ndv_mod(m,(NDV)BDY(in));
1.61 noro 3758: }
3759: NEXTNODE(in0,in);
3760: BDY(in) = (pointer)ptondv(CO,vv,f);
1.69 noro 3761: if ( m ) ndv_mod(m,(NDV)BDY(in));
1.61 noro 3762: NEXT(in) = 0;
3763:
3764: ndv_setup(m,0,in0);
1.59 noro 3765: nd_psn--;
3766: nd_scale=2;
3767: while ( 1 ) {
3768: nd = (pointer)ndvtond(m,nd_ps[nd_psn]);
1.69 noro 3769: stat = nd_nf(m,nd,nd_ps,1,0,&nf);
1.59 noro 3770: if ( !stat ) {
3771: nd_psn++;
1.103 noro 3772: nd_reconstruct(0,0);
1.59 noro 3773: nd_psn--;
3774: } else
3775: break;
3776: }
1.61 noro 3777: *rp = ndvtop(m,CO,vv,ndtondv(m,nf));
1.63 noro 3778: }
3779:
3780: int nd_to_vect(int mod,UINT *s0,int n,ND d,UINT *r)
3781: {
3782: NM m;
3783: UINT *t,*s;
3784: int i;
3785:
3786: for ( i = 0; i < n; i++ ) r[i] = 0;
3787: for ( i = 0, s = s0, m = BDY(d); m; m = NEXT(m) ) {
3788: t = DL(m);
3789: for ( ; !ndl_equal(t,s); s += nd_wpd, i++ );
3790: r[i] = CM(m);
3791: }
3792: for ( i = 0; !r[i]; i++ );
3793: return i;
3794: }
3795:
1.112 ! noro 3796: int nd_to_vect_z(UINT *s0,int n,ND d,Z *r)
1.74 noro 3797: {
1.107 noro 3798: NM m;
1.74 noro 3799: UINT *t,*s;
1.107 noro 3800: int i;
1.74 noro 3801:
3802: for ( i = 0; i < n; i++ ) r[i] = 0;
1.107 noro 3803: for ( i = 0, s = s0, m = BDY(d); m; m = NEXT(m) ) {
1.74 noro 3804: t = DL(m);
3805: for ( ; !ndl_equal(t,s); s += nd_wpd, i++ );
1.112 ! noro 3806: r[i] = CZ(m);
1.74 noro 3807: }
3808: for ( i = 0; !r[i]; i++ );
3809: return i;
3810: }
3811:
1.67 noro 3812: IndArray nm_ind_pair_to_vect_compress(int mod,UINT *s0,int n,NM_ind_pair pair)
1.64 noro 3813: {
3814: NM m;
3815: NMV mr;
3816: UINT *d,*t,*s;
3817: NDV p;
1.67 noro 3818: unsigned char *ivc;
3819: unsigned short *ivs;
1.81 noro 3820: UINT *v,*ivi,*s0v;
1.67 noro 3821: int i,j,len,prev,diff,cdiff;
3822: IndArray r;
1.64 noro 3823:
3824: m = pair->mul;
3825: d = DL(m);
3826: p = nd_ps[pair->index];
3827: len = LEN(p);
3828: t = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
1.67 noro 3829: v = (unsigned int *)ALLOCA(len*sizeof(unsigned int));
1.64 noro 3830: for ( i = j = 0, s = s0, mr = BDY(p); j < len; j++, NMV_ADV(mr) ) {
3831: ndl_add(d,DL(mr),t);
3832: for ( ; !ndl_equal(t,s); s += nd_wpd, i++ );
1.67 noro 3833: v[j] = i;
3834: }
1.81 noro 3835: r = (IndArray)MALLOC(sizeof(struct oIndArray));
1.67 noro 3836: r->head = v[0];
3837: diff = 0;
3838: for ( i = 1; i < len; i++ ) {
3839: cdiff = v[i]-v[i-1]; diff = MAX(cdiff,diff);
3840: }
3841: if ( diff < 256 ) {
3842: r->width = 1;
3843: ivc = (unsigned char *)MALLOC_ATOMIC(len*sizeof(unsigned char));
3844: r->index.c = ivc;
3845: for ( i = 1, ivc[0] = 0; i < len; i++ ) ivc[i] = v[i]-v[i-1];
3846: } else if ( diff < 65536 ) {
3847: r->width = 2;
3848: ivs = (unsigned short *)MALLOC_ATOMIC(len*sizeof(unsigned short));
3849: r->index.s = ivs;
3850: for ( i = 1, ivs[0] = 0; i < len; i++ ) ivs[i] = v[i]-v[i-1];
3851: } else {
3852: r->width = 4;
3853: ivi = (unsigned int *)MALLOC_ATOMIC(len*sizeof(unsigned int));
3854: r->index.i = ivi;
3855: for ( i = 1, ivi[0] = 0; i < len; i++ ) ivi[i] = v[i]-v[i-1];
1.64 noro 3856: }
1.67 noro 3857: return r;
1.64 noro 3858: }
3859:
1.65 noro 3860:
1.112 ! noro 3861: int ndv_reduce_vect_z(Z *svect,int col,IndArray *imat,NM_ind_pair *rp0,int nred)
1.107 noro 3862: {
3863: int i,j,k,len,pos,prev;
1.112 ! noro 3864: Z cs,mcs,c1,c2,cr,gcd,t;
1.107 noro 3865: IndArray ivect;
3866: unsigned char *ivc;
3867: unsigned short *ivs;
3868: unsigned int *ivi;
3869: NDV redv;
3870: NMV mr;
3871: NODE rp;
3872: int maxrs;
3873:
3874: maxrs = 0;
3875: for ( i = 0; i < nred; i++ ) {
3876: ivect = imat[i];
3877: k = ivect->head;
3878: if ( svect[k] ) {
3879: maxrs = MAX(maxrs,rp0[i]->sugar);
3880: redv = nd_ps[rp0[i]->index];
3881: len = LEN(redv); mr = BDY(redv);
1.112 ! noro 3882: gcd = gcdz_cofactor(svect[k],CZ(mr),&cs,&cr);
! 3883: mcs = chsgnz(cs);
! 3884: if ( !uniz(cr) )
! 3885: for ( j = 0; j < col; j++ )
! 3886: svect[j] = mulz(svect[j],cr);
1.107 noro 3887: svect[k] = 0; prev = k;
3888: switch ( ivect->width ) {
3889: case 1:
3890: ivc = ivect->index.c;
3891: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3892: pos = prev+ivc[j]; prev = pos;
1.112 ! noro 3893: svect[pos] = addz(svect[pos],mulz(CZ(mr),mcs));
1.107 noro 3894: }
3895: break;
3896: case 2:
3897: ivs = ivect->index.s;
3898: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3899: pos = prev+ivs[j]; prev = pos;
1.112 ! noro 3900: svect[pos] = addz(svect[pos],mulz(CZ(mr),mcs));
1.107 noro 3901: }
3902: break;
3903: case 4:
3904: ivi = ivect->index.i;
3905: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3906: pos = prev+ivi[j]; prev = pos;
1.112 ! noro 3907: svect[pos] = addz(svect[pos],mulz(CZ(mr),mcs));
1.107 noro 3908: }
3909: break;
3910: }
3911: }
3912: }
3913: return maxrs;
3914: }
3915:
1.76 noro 3916: int ndv_reduce_vect(int m,UINT *svect,int col,IndArray *imat,NM_ind_pair *rp0,int nred)
1.65 noro 3917: {
1.67 noro 3918: int i,j,k,len,pos,prev;
1.66 noro 3919: UINT c,c1,c2,c3,up,lo,dmy;
1.67 noro 3920: IndArray ivect;
3921: unsigned char *ivc;
3922: unsigned short *ivs;
3923: unsigned int *ivi;
1.65 noro 3924: NDV redv;
1.67 noro 3925: NMV mr;
1.65 noro 3926: NODE rp;
1.76 noro 3927: int maxrs;
1.65 noro 3928:
1.76 noro 3929: maxrs = 0;
1.74 noro 3930: for ( i = 0; i < nred; i++ ) {
1.65 noro 3931: ivect = imat[i];
1.67 noro 3932: k = ivect->head; svect[k] %= m;
1.65 noro 3933: if ( c = svect[k] ) {
1.76 noro 3934: maxrs = MAX(maxrs,rp0[i]->sugar);
1.74 noro 3935: c = m-c; redv = nd_ps[rp0[i]->index];
1.67 noro 3936: len = LEN(redv); mr = BDY(redv);
3937: svect[k] = 0; prev = k;
3938: switch ( ivect->width ) {
3939: case 1:
3940: ivc = ivect->index.c;
3941: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3942: pos = prev+ivc[j]; c1 = CM(mr); c2 = svect[pos];
3943: prev = pos;
3944: DMA(c1,c,c2,up,lo);
3945: if ( up ) { DSAB(m,up,lo,dmy,c3); svect[pos] = c3;
3946: } else svect[pos] = lo;
3947: }
3948: break;
3949: case 2:
3950: ivs = ivect->index.s;
3951: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3952: pos = prev+ivs[j]; c1 = CM(mr); c2 = svect[pos];
3953: prev = pos;
3954: DMA(c1,c,c2,up,lo);
3955: if ( up ) { DSAB(m,up,lo,dmy,c3); svect[pos] = c3;
3956: } else svect[pos] = lo;
3957: }
3958: break;
3959: case 4:
3960: ivi = ivect->index.i;
3961: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3962: pos = prev+ivi[j]; c1 = CM(mr); c2 = svect[pos];
3963: prev = pos;
3964: DMA(c1,c,c2,up,lo);
3965: if ( up ) { DSAB(m,up,lo,dmy,c3); svect[pos] = c3;
3966: } else svect[pos] = lo;
3967: }
3968: break;
1.65 noro 3969: }
3970: }
3971: }
1.66 noro 3972: for ( i = 0; i < col; i++ )
3973: if ( svect[i] >= (UINT)m ) svect[i] %= m;
1.76 noro 3974: return maxrs;
1.65 noro 3975: }
3976:
1.76 noro 3977: int ndv_reduce_vect_sf(int m,UINT *svect,int col,IndArray *imat,NM_ind_pair *rp0,int nred)
1.72 noro 3978: {
3979: int i,j,k,len,pos,prev;
3980: UINT c,c1,c2,c3,up,lo,dmy;
3981: IndArray ivect;
3982: unsigned char *ivc;
3983: unsigned short *ivs;
3984: unsigned int *ivi;
3985: NDV redv;
3986: NMV mr;
3987: NODE rp;
1.76 noro 3988: int maxrs;
1.72 noro 3989:
1.76 noro 3990: maxrs = 0;
1.74 noro 3991: for ( i = 0; i < nred; i++ ) {
1.72 noro 3992: ivect = imat[i];
3993: k = ivect->head; svect[k] %= m;
3994: if ( c = svect[k] ) {
1.76 noro 3995: maxrs = MAX(maxrs,rp0[i]->sugar);
1.74 noro 3996: c = _chsgnsf(c); redv = nd_ps[rp0[i]->index];
1.72 noro 3997: len = LEN(redv); mr = BDY(redv);
3998: svect[k] = 0; prev = k;
3999: switch ( ivect->width ) {
4000: case 1:
4001: ivc = ivect->index.c;
4002: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
4003: pos = prev+ivc[j]; prev = pos;
4004: svect[pos] = _addsf(_mulsf(CM(mr),c),svect[pos]);
4005: }
4006: break;
4007: case 2:
4008: ivs = ivect->index.s;
4009: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
4010: pos = prev+ivs[j]; prev = pos;
4011: svect[pos] = _addsf(_mulsf(CM(mr),c),svect[pos]);
4012: }
4013: break;
4014: case 4:
4015: ivi = ivect->index.i;
4016: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
4017: pos = prev+ivi[j]; prev = pos;
4018: svect[pos] = _addsf(_mulsf(CM(mr),c),svect[pos]);
4019: }
4020: break;
4021: }
4022: }
4023: }
1.76 noro 4024: return maxrs;
1.72 noro 4025: }
4026:
1.65 noro 4027: NDV vect_to_ndv(UINT *vect,int spcol,int col,int *rhead,UINT *s0vect)
4028: {
4029: int j,k,len;
4030: UINT *p;
4031: UINT c;
4032: NDV r;
4033: NMV mr0,mr;
4034:
4035: for ( j = 0, len = 0; j < spcol; j++ ) if ( vect[j] ) len++;
4036: if ( !len ) return 0;
4037: else {
1.74 noro 4038: mr0 = (NMV)GC_malloc_atomic_ignore_off_page(nmv_adv*len);
1.103 noro 4039: #if 0
1.74 noro 4040: ndv_alloc += nmv_adv*len;
1.103 noro 4041: #endif
1.65 noro 4042: mr = mr0;
4043: p = s0vect;
4044: for ( j = k = 0; j < col; j++, p += nd_wpd )
4045: if ( !rhead[j] ) {
4046: if ( c = vect[k++] ) {
4047: ndl_copy(p,DL(mr)); CM(mr) = c; NMV_ADV(mr);
4048: }
4049: }
4050: MKNDV(nd_nvar,mr0,len,r);
4051: return r;
4052: }
4053: }
4054:
1.112 ! noro 4055: NDV vect_to_ndv_z(Z *vect,int spcol,int col,int *rhead,UINT *s0vect)
1.107 noro 4056: {
4057: int j,k,len;
4058: UINT *p;
1.112 ! noro 4059: Z c;
1.107 noro 4060: NDV r;
4061: NMV mr0,mr;
4062:
4063: for ( j = 0, len = 0; j < spcol; j++ ) if ( vect[j] ) len++;
4064: if ( !len ) return 0;
4065: else {
1.109 noro 4066: mr0 = (NMV)GC_malloc(nmv_adv*len);
1.107 noro 4067: #if 0
4068: ndv_alloc += nmv_adv*len;
4069: #endif
4070: mr = mr0;
4071: p = s0vect;
4072: for ( j = k = 0; j < col; j++, p += nd_wpd )
4073: if ( !rhead[j] ) {
4074: if ( c = vect[k++] ) {
1.112 ! noro 4075: ndl_copy(p,DL(mr)); CZ(mr) = c; NMV_ADV(mr);
1.107 noro 4076: }
4077: }
4078: MKNDV(nd_nvar,mr0,len,r);
4079: return r;
4080: }
4081: }
4082:
1.74 noro 4083: int nd_sp_f4(int m,ND_pairs l,PGeoBucket bucket)
1.65 noro 4084: {
4085: ND_pairs t;
4086: NODE sp0,sp;
4087: int stat;
4088: ND spol;
4089:
4090: for ( t = l; t; t = NEXT(t) ) {
4091: stat = nd_sp(m,0,t,&spol);
4092: if ( !stat ) return 0;
4093: if ( spol ) {
4094: add_pbucket_symbolic(bucket,spol);
4095: }
4096: }
1.68 noro 4097: return 1;
1.65 noro 4098: }
4099:
4100: int nd_symbolic_preproc(PGeoBucket bucket,UINT **s0vect,NODE *r)
4101: {
4102: NODE rp0,rp;
4103: NM mul,head,s0,s;
1.76 noro 4104: int index,col,i,sugar;
1.65 noro 4105: RHist h;
4106: UINT *s0v,*p;
4107: NM_ind_pair pair;
4108: ND red;
4109:
4110: s0 = 0; rp0 = 0; col = 0;
4111: while ( 1 ) {
4112: head = remove_head_pbucket_symbolic(bucket);
4113: if ( !head ) break;
4114: if ( !s0 ) s0 = head;
4115: else NEXT(s) = head;
4116: s = head;
4117: index = ndl_find_reducer(DL(head));
4118: if ( index >= 0 ) {
4119: h = nd_psh[index];
4120: NEWNM(mul);
4121: ndl_sub(DL(head),DL(h),DL(mul));
4122: if ( ndl_check_bound2(index,DL(mul)) ) return 0;
1.76 noro 4123: sugar = TD(DL(mul))+SG(nd_ps[index]);
4124: MKNM_ind_pair(pair,mul,index,sugar);
1.65 noro 4125: red = ndv_mul_nm_symbolic(mul,nd_ps[index]);
4126: add_pbucket_symbolic(bucket,nd_remove_head(red));
4127: NEXTNODE(rp0,rp); BDY(rp) = (pointer)pair;
4128: }
4129: col++;
4130: }
1.72 noro 4131: if ( rp0 ) NEXT(rp) = 0;
4132: NEXT(s) = 0;
1.65 noro 4133: s0v = (UINT *)MALLOC_ATOMIC(col*nd_wpd*sizeof(UINT));
4134: for ( i = 0, p = s0v, s = s0; i < col;
4135: i++, p += nd_wpd, s = NEXT(s) ) ndl_copy(DL(s),p);
4136: *s0vect = s0v;
4137: *r = rp0;
4138: return col;
4139: }
4140:
1.69 noro 4141: NODE nd_f4(int m)
4142: {
4143: int i,nh,stat,index;
4144: NODE r,g;
4145: ND_pairs d,l,t;
4146: ND spol,red;
4147: NDV nf,redv;
4148: NM s0,s;
1.84 noro 4149: NODE rp0,srp0,nflist;
1.69 noro 4150: int nsp,nred,col,rank,len,k,j,a;
4151: UINT c;
1.74 noro 4152: UINT **spmat;
1.69 noro 4153: UINT *s0vect,*svect,*p,*v;
4154: int *colstat;
4155: IndArray *imat;
4156: int *rhead;
4157: int spcol,sprow;
4158: int sugar;
4159: PGeoBucket bucket;
4160: struct oEGT eg0,eg1,eg_f4;
4161:
1.103 noro 4162: #if 0
1.74 noro 4163: ndv_alloc = 0;
1.103 noro 4164: #endif
1.69 noro 4165: g = 0; d = 0;
4166: for ( i = 0; i < nd_psn; i++ ) {
4167: d = update_pairs(d,g,i);
4168: g = update_base(g,i);
4169: }
4170: while ( d ) {
4171: get_eg(&eg0);
4172: l = nd_minsugarp(d,&d);
4173: sugar = SG(l);
4174: bucket = create_pbucket();
1.74 noro 4175: stat = nd_sp_f4(m,l,bucket);
1.69 noro 4176: if ( !stat ) {
4177: for ( t = l; NEXT(t); t = NEXT(t) );
4178: NEXT(t) = d; d = l;
1.103 noro 4179: d = nd_reconstruct(0,d);
1.69 noro 4180: continue;
4181: }
1.86 noro 4182: if ( bucket->m < 0 ) continue;
1.69 noro 4183: col = nd_symbolic_preproc(bucket,&s0vect,&rp0);
4184: if ( !col ) {
4185: for ( t = l; NEXT(t); t = NEXT(t) );
4186: NEXT(t) = d; d = l;
1.103 noro 4187: d = nd_reconstruct(0,d);
1.69 noro 4188: continue;
4189: }
4190: get_eg(&eg1); init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg1);
1.74 noro 4191: if ( DP_Print )
4192: fprintf(asir_out,"sugar=%d,symb=%fsec,",
4193: sugar,eg_f4.exectime+eg_f4.gctime);
4194: if ( 1 )
4195: nflist = nd_f4_red(m,l,s0vect,col,rp0);
4196: else
4197: nflist = nd_f4_red_dist(m,l,s0vect,col,rp0);
1.69 noro 4198: /* adding new bases */
1.74 noro 4199: for ( r = nflist; r; r = NEXT(r) ) {
4200: nf = (NDV)BDY(r);
1.69 noro 4201: ndv_removecont(m,nf);
1.77 noro 4202: nh = ndv_newps(m,nf,0);
1.69 noro 4203: d = update_pairs(d,g,nh);
4204: g = update_base(g,nh);
4205: }
4206: }
4207: for ( r = g; r; r = NEXT(r) ) BDY(r) = (pointer)nd_ps[(int)BDY(r)];
1.103 noro 4208: #if 0
1.74 noro 4209: fprintf(asir_out,"ndv_alloc=%d\n",ndv_alloc);
1.103 noro 4210: #endif
1.69 noro 4211: return g;
4212: }
1.74 noro 4213:
4214: NODE nd_f4_red(int m,ND_pairs sp0,UINT *s0vect,int col,NODE rp0)
1.63 noro 4215: {
1.67 noro 4216: IndArray *imat;
1.106 noro 4217: int nsp,nred,i;
1.65 noro 4218: int *rhead;
1.106 noro 4219: NODE r0,rp;
1.74 noro 4220: ND_pairs sp;
4221: NM_ind_pair *rvect;
1.63 noro 4222:
1.74 noro 4223: for ( sp = sp0, nsp = 0; sp; sp = NEXT(sp), nsp++ );
1.106 noro 4224: nred = length(rp0);
1.74 noro 4225: imat = (IndArray *)ALLOCA(nred*sizeof(IndArray));
4226: rhead = (int *)ALLOCA(col*sizeof(int));
4227: for ( i = 0; i < col; i++ ) rhead[i] = 0;
1.63 noro 4228:
1.74 noro 4229: /* construction of index arrays */
4230: rvect = (NM_ind_pair *)ALLOCA(nred*sizeof(NM_ind_pair));
4231: for ( rp = rp0, i = 0; rp; i++, rp = NEXT(rp) ) {
4232: rvect[i] = (NM_ind_pair)BDY(rp);
4233: imat[i] = nm_ind_pair_to_vect_compress(m,s0vect,col,rvect[i]);
4234: rhead[imat[i]->head] = 1;
4235: }
1.107 noro 4236: if ( m )
4237: r0 = nd_f4_red_main(m,sp0,nsp,s0vect,col,rvect,rhead,imat,nred);
4238: else
1.112 ! noro 4239: r0 = nd_f4_red_z_main(sp0,nsp,s0vect,col,rvect,rhead,imat,nred);
1.106 noro 4240: return r0;
4241: }
1.74 noro 4242:
1.106 noro 4243: NODE nd_f4_red_main(int m,ND_pairs sp0,int nsp,UINT *s0vect,int col,
4244: NM_ind_pair *rvect,int *rhead,IndArray *imat,int nred)
4245: {
4246: int spcol,sprow,a;
4247: int i,j,k,l,rank;
4248: NODE r0,r;
4249: ND_pairs sp;
4250: ND spol;
4251: int **spmat;
4252: UINT *svect,*v;
4253: int *colstat;
4254: struct oEGT eg0,eg1,eg2,eg_f4,eg_f4_1,eg_f4_2;
4255: int maxrs;
4256: int *spsugar;
4257:
4258: spcol = col-nred;
4259: get_eg(&eg0);
1.74 noro 4260: /* elimination (1st step) */
4261: spmat = (int **)ALLOCA(nsp*sizeof(UINT *));
4262: svect = (UINT *)ALLOCA(col*sizeof(UINT));
1.76 noro 4263: spsugar = (int *)ALLOCA(nsp*sizeof(UINT));
1.74 noro 4264: for ( a = sprow = 0, sp = sp0; a < nsp; a++, sp = NEXT(sp) ) {
4265: nd_sp(m,0,sp,&spol);
1.75 noro 4266: if ( !spol ) continue;
1.74 noro 4267: nd_to_vect(m,s0vect,col,spol,svect);
1.76 noro 4268: if ( m == -1 )
4269: maxrs = ndv_reduce_vect_sf(m,svect,col,imat,rvect,nred);
4270: else
4271: maxrs = ndv_reduce_vect(m,svect,col,imat,rvect,nred);
1.74 noro 4272: for ( i = 0; i < col; i++ ) if ( svect[i] ) break;
4273: if ( i < col ) {
4274: spmat[sprow] = v = (UINT *)MALLOC_ATOMIC(spcol*sizeof(UINT));
4275: for ( j = k = 0; j < col; j++ )
4276: if ( !rhead[j] ) v[k++] = svect[j];
1.76 noro 4277: spsugar[sprow] = MAX(maxrs,SG(spol));
1.74 noro 4278: sprow++;
4279: }
1.76 noro 4280: nd_free(spol);
1.74 noro 4281: }
1.80 noro 4282: get_eg(&eg1); init_eg(&eg_f4_1); add_eg(&eg_f4_1,&eg0,&eg1);
4283: if ( DP_Print ) {
4284: fprintf(asir_out,"elim1=%fsec,",eg_f4_1.exectime+eg_f4_1.gctime);
4285: fflush(asir_out);
4286: }
1.74 noro 4287: /* free index arrays */
4288: for ( i = 0; i < nred; i++ ) GC_free(imat[i]->index.c);
4289:
4290: /* elimination (2nd step) */
4291: colstat = (int *)ALLOCA(spcol*sizeof(int));
4292: if ( m == -1 )
1.76 noro 4293: rank = nd_gauss_elim_sf(spmat,spsugar,sprow,spcol,m,colstat);
1.74 noro 4294: else
1.76 noro 4295: rank = nd_gauss_elim_mod(spmat,spsugar,sprow,spcol,m,colstat);
1.74 noro 4296: r0 = 0;
4297: for ( i = 0; i < rank; i++ ) {
4298: NEXTNODE(r0,r); BDY(r) =
4299: (pointer)vect_to_ndv(spmat[i],spcol,col,rhead,s0vect);
1.76 noro 4300: SG((NDV)BDY(r)) = spsugar[i];
1.74 noro 4301: GC_free(spmat[i]);
4302: }
1.109 noro 4303: if ( r0 ) NEXT(r) = 0;
1.74 noro 4304: for ( ; i < sprow; i++ ) GC_free(spmat[i]);
1.80 noro 4305: get_eg(&eg2); init_eg(&eg_f4_2); add_eg(&eg_f4_2,&eg1,&eg2);
4306: init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg2);
1.74 noro 4307: if ( DP_Print ) {
1.80 noro 4308: fprintf(asir_out,"elim2=%fsec\n",eg_f4_2.exectime+eg_f4_2.gctime);
1.74 noro 4309: fprintf(asir_out,"nsp=%d,nred=%d,spmat=(%d,%d),rank=%d ",
4310: nsp,nred,sprow,spcol,rank);
4311: fprintf(asir_out,"%fsec\n",eg_f4.exectime+eg_f4.gctime);
1.63 noro 4312: }
1.74 noro 4313: return r0;
4314: }
4315:
1.112 ! noro 4316: NODE nd_f4_red_z_main(ND_pairs sp0,int nsp,UINT *s0vect,int col,
1.107 noro 4317: NM_ind_pair *rvect,int *rhead,IndArray *imat,int nred)
4318: {
4319: int spcol,sprow,a;
4320: int i,j,k,l,rank;
4321: NODE r0,r;
4322: ND_pairs sp;
4323: ND spol;
1.112 ! noro 4324: Z **spmat;
! 4325: Z *svect,*v;
1.107 noro 4326: int *colstat;
4327: struct oEGT eg0,eg1,eg2,eg_f4,eg_f4_1,eg_f4_2;
4328: int maxrs;
4329: int *spsugar;
4330:
4331: spcol = col-nred;
4332: get_eg(&eg0);
4333: /* elimination (1st step) */
1.112 ! noro 4334: spmat = (Z **)ALLOCA(nsp*sizeof(Z *));
! 4335: svect = (Z *)ALLOCA(col*sizeof(Z));
! 4336: spsugar = (int *)ALLOCA(nsp*sizeof(Z));
1.107 noro 4337: for ( a = sprow = 0, sp = sp0; a < nsp; a++, sp = NEXT(sp) ) {
4338: nd_sp(0,0,sp,&spol);
4339: if ( !spol ) continue;
1.112 ! noro 4340: nd_to_vect_z(s0vect,col,spol,svect);
! 4341: maxrs = ndv_reduce_vect_z(svect,col,imat,rvect,nred);
1.107 noro 4342: for ( i = 0; i < col; i++ ) if ( svect[i] ) break;
4343: if ( i < col ) {
1.112 ! noro 4344: spmat[sprow] = v = (Z *)MALLOC(spcol*sizeof(Z));
1.107 noro 4345: for ( j = k = 0; j < col; j++ )
4346: if ( !rhead[j] ) v[k++] = svect[j];
4347: spsugar[sprow] = MAX(maxrs,SG(spol));
4348: sprow++;
4349: }
1.109 noro 4350: /* nd_free(spol); */
1.107 noro 4351: }
4352: get_eg(&eg1); init_eg(&eg_f4_1); add_eg(&eg_f4_1,&eg0,&eg1);
4353: if ( DP_Print ) {
4354: fprintf(asir_out,"elim1=%fsec,",eg_f4_1.exectime+eg_f4_1.gctime);
4355: fflush(asir_out);
4356: }
4357: /* free index arrays */
1.109 noro 4358: /* for ( i = 0; i < nred; i++ ) GC_free(imat[i]->index.c); */
1.107 noro 4359:
4360: /* elimination (2nd step) */
4361: colstat = (int *)ALLOCA(spcol*sizeof(int));
1.112 ! noro 4362: rank = nd_gauss_elim_z(spmat,spsugar,sprow,spcol,colstat);
1.107 noro 4363: r0 = 0;
4364: for ( i = 0; i < rank; i++ ) {
4365: NEXTNODE(r0,r); BDY(r) =
1.112 ! noro 4366: (pointer)vect_to_ndv_z(spmat[i],spcol,col,rhead,s0vect);
1.107 noro 4367: SG((NDV)BDY(r)) = spsugar[i];
1.109 noro 4368: /* GC_free(spmat[i]); */
1.107 noro 4369: }
1.109 noro 4370: if ( r0 ) NEXT(r) = 0;
4371:
4372: /* for ( ; i < sprow; i++ ) GC_free(spmat[i]); */
1.107 noro 4373: get_eg(&eg2); init_eg(&eg_f4_2); add_eg(&eg_f4_2,&eg1,&eg2);
4374: init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg2);
4375: if ( DP_Print ) {
4376: fprintf(asir_out,"elim2=%fsec\n",eg_f4_2.exectime+eg_f4_2.gctime);
4377: fprintf(asir_out,"nsp=%d,nred=%d,spmat=(%d,%d),rank=%d ",
4378: nsp,nred,sprow,spcol,rank);
4379: fprintf(asir_out,"%fsec\n",eg_f4.exectime+eg_f4.gctime);
4380: }
4381: return r0;
4382: }
4383:
1.74 noro 4384: FILE *nd_write,*nd_read;
4385:
4386: void nd_send_int(int a) {
4387: write_int(nd_write,&a);
4388: }
4389:
4390: void nd_send_intarray(int *p,int len) {
4391: write_intarray(nd_write,p,len);
4392: }
4393:
4394: int nd_recv_int() {
4395: int a;
4396:
4397: read_int(nd_read,&a);
4398: return a;
4399: }
4400:
4401: void nd_recv_intarray(int *p,int len) {
4402: read_intarray(nd_read,p,len);
4403: }
4404:
4405: void nd_send_ndv(NDV p) {
4406: int len,i;
4407: NMV m;
4408:
4409: if ( !p ) nd_send_int(0);
4410: else {
4411: len = LEN(p);
4412: nd_send_int(len);
4413: m = BDY(p);
4414: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
4415: nd_send_int(CM(m));
4416: nd_send_intarray(DL(m),nd_wpd);
1.68 noro 4417: }
1.74 noro 4418: }
4419: }
4420:
4421: void nd_send_nd(ND p) {
4422: int len,i;
4423: NM m;
4424:
4425: if ( !p ) nd_send_int(0);
4426: else {
4427: len = LEN(p);
4428: nd_send_int(len);
4429: m = BDY(p);
4430: for ( i = 0; i < len; i++, m = NEXT(m) ) {
4431: nd_send_int(CM(m));
4432: nd_send_intarray(DL(m),nd_wpd);
1.65 noro 4433: }
1.74 noro 4434: }
4435: }
1.65 noro 4436:
1.74 noro 4437: NDV nd_recv_ndv()
4438: {
4439: int len,i;
4440: NMV m,m0;
4441: NDV r;
1.65 noro 4442:
1.74 noro 4443: len = nd_recv_int();
4444: if ( !len ) return 0;
4445: else {
4446: m0 = m = (NMV)GC_malloc_atomic_ignore_off_page(nmv_adv*len);
1.103 noro 4447: #if 0
1.74 noro 4448: ndv_alloc += len*nmv_adv;
1.103 noro 4449: #endif
1.74 noro 4450: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
4451: CM(m) = nd_recv_int();
4452: nd_recv_intarray(DL(m),nd_wpd);
1.71 noro 4453: }
1.74 noro 4454: MKNDV(nd_nvar,m0,len,r);
4455: return r;
4456: }
4457: }
1.65 noro 4458:
1.74 noro 4459: int ox_exec_f4_red(Q proc)
4460: {
4461: Obj obj;
4462: STRING fname;
4463: NODE arg;
4464: int s;
4465: extern int ox_need_conv,ox_file_io;
4466:
4467: MKSTR(fname,"nd_exec_f4_red");
4468: arg = mknode(2,proc,fname);
4469: Pox_cmo_rpc(arg,&obj);
4470: s = get_ox_server_id(QTOS(proc));
4471: nd_write = iofp[s].out;
4472: nd_read = iofp[s].in;
4473: ox_need_conv = ox_file_io = 0;
4474: return s;
4475: }
4476:
4477: NODE nd_f4_red_dist(int m,ND_pairs sp0,UINT *s0vect,int col,NODE rp0)
4478: {
4479: int nsp,nred;
4480: int i,rank,s;
4481: NODE rp,r0,r;
4482: ND_pairs sp;
4483: NM_ind_pair pair;
4484: NMV nmv;
4485: NM nm;
4486: NDV nf;
4487: Obj proc,dmy;
4488:
4489: ox_launch_main(0,0,&proc);
4490: s = ox_exec_f4_red((Q)proc);
4491:
4492: nd_send_int(m);
4493: nd_send_int(nd_nvar);
4494: nd_send_int(nd_bpe);
4495: nd_send_int(nd_wpd);
4496: nd_send_int(nmv_adv);
4497:
1.89 noro 4498: saveobj(nd_write,dp_current_spec->obj); fflush(nd_write);
1.74 noro 4499:
4500: nd_send_int(nd_psn);
4501: for ( i = 0; i < nd_psn; i++ ) nd_send_ndv(nd_ps[i]);
4502:
4503: for ( sp = sp0, nsp = 0; sp; sp = NEXT(sp), nsp++ );
4504: nd_send_int(nsp);
4505: for ( i = 0, sp = sp0; i < nsp; i++, sp = NEXT(sp) ) {
4506: nd_send_int(sp->i1); nd_send_int(sp->i2);
4507: }
4508:
4509: nd_send_int(col); nd_send_intarray(s0vect,col*nd_wpd);
4510:
4511: nred = length(rp0); nd_send_int(nred);
4512: for ( i = 0, rp = rp0; i < nred; i++, rp = NEXT(rp) ) {
4513: pair = (NM_ind_pair)BDY(rp);
4514: nd_send_int(pair->index);
4515: nd_send_intarray(pair->mul->dl,nd_wpd);
4516: }
4517: fflush(nd_write);
4518: rank = nd_recv_int();
4519: fprintf(asir_out,"rank=%d\n",rank);
4520: r0 = 0;
4521: for ( i = 0; i < rank; i++ ) {
4522: nf = nd_recv_ndv();
4523: NEXTNODE(r0,r); BDY(r) = (pointer)nf;
4524: }
4525: Pox_shutdown(mknode(1,proc),&dmy);
4526: return r0;
4527: }
4528:
4529: /* server side */
4530:
4531: void nd_exec_f4_red_dist()
4532: {
4533: int m,i,nsp,col,s0size,nred,spcol,j,k;
4534: NM_ind_pair *rp0;
4535: NDV nf;
4536: UINT *s0vect;
4537: IndArray *imat;
4538: int *rhead;
4539: int **spmat;
4540: UINT *svect,*v;
4541: ND_pairs *sp0;
4542: int *colstat;
4543: int a,sprow,rank;
1.89 noro 4544: struct order_spec *ord;
1.74 noro 4545: Obj ordspec;
4546: ND spol;
1.76 noro 4547: int maxrs;
4548: int *spsugar;
1.74 noro 4549:
4550: nd_read = iofp[0].in;
4551: nd_write = iofp[0].out;
4552: m = nd_recv_int();
4553: nd_nvar = nd_recv_int();
4554: nd_bpe = nd_recv_int();
4555: nd_wpd = nd_recv_int();
4556: nmv_adv = nd_recv_int();
4557:
4558: loadobj(nd_read,&ordspec);
1.89 noro 4559: create_order_spec(0,ordspec,&ord);
4560: nd_init_ord(ord);
1.74 noro 4561: nd_setup_parameters(nd_nvar,0);
4562:
4563: nd_psn = nd_recv_int();
4564: nd_ps = (NDV *)MALLOC(nd_psn*sizeof(NDV));
4565: nd_bound = (UINT **)MALLOC(nd_psn*sizeof(UINT *));
4566: for ( i = 0; i < nd_psn; i++ ) {
4567: nd_ps[i] = nd_recv_ndv();
4568: nd_bound[i] = ndv_compute_bound(nd_ps[i]);
4569: }
4570:
4571: nsp = nd_recv_int();
4572: sp0 = (ND_pairs *)MALLOC(nsp*sizeof(ND_pairs));
4573: for ( i = 0; i < nsp; i++ ) {
4574: NEWND_pairs(sp0[i]);
4575: sp0[i]->i1 = nd_recv_int(); sp0[i]->i2 = nd_recv_int();
4576: ndl_lcm(HDL(nd_ps[sp0[i]->i1]),HDL(nd_ps[sp0[i]->i2]),LCM(sp0[i]));
4577: }
4578:
4579: col = nd_recv_int();
4580: s0size = col*nd_wpd;
4581: s0vect = (UINT *)MALLOC(s0size*sizeof(UINT));
4582: nd_recv_intarray(s0vect,s0size);
4583:
4584: nred = nd_recv_int();
4585: rp0 = (NM_ind_pair *)MALLOC(nred*sizeof(NM_ind_pair));
4586: for ( i = 0; i < nred; i++ ) {
4587: rp0[i] = (NM_ind_pair)MALLOC(sizeof(struct oNM_ind_pair));
4588: rp0[i]->index = nd_recv_int();
4589: rp0[i]->mul = (NM)MALLOC(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT));
4590: nd_recv_intarray(rp0[i]->mul->dl,nd_wpd);
4591: }
4592:
4593: spcol = col-nred;
4594: imat = (IndArray *)MALLOC(nred*sizeof(IndArray));
4595: rhead = (int *)MALLOC(col*sizeof(int));
4596: for ( i = 0; i < col; i++ ) rhead[i] = 0;
4597:
4598: /* construction of index arrays */
4599: for ( i = 0; i < nred; i++ ) {
4600: imat[i] = nm_ind_pair_to_vect_compress(m,s0vect,col,rp0[i]);
4601: rhead[imat[i]->head] = 1;
4602: }
4603:
4604: /* elimination (1st step) */
4605: spmat = (int **)MALLOC(nsp*sizeof(UINT *));
4606: svect = (UINT *)MALLOC(col*sizeof(UINT));
1.76 noro 4607: spsugar = (int *)ALLOCA(nsp*sizeof(UINT));
1.74 noro 4608: for ( a = sprow = 0; a < nsp; a++ ) {
4609: nd_sp(m,0,sp0[a],&spol);
1.75 noro 4610: if ( !spol ) continue;
1.74 noro 4611: nd_to_vect(m,s0vect,col,spol,svect);
1.76 noro 4612: if ( m == -1 )
4613: maxrs = ndv_reduce_vect_sf(m,svect,col,imat,rp0,nred);
4614: else
4615: maxrs = ndv_reduce_vect(m,svect,col,imat,rp0,nred);
1.74 noro 4616: for ( i = 0; i < col; i++ ) if ( svect[i] ) break;
4617: if ( i < col ) {
4618: spmat[sprow] = v = (UINT *)MALLOC(spcol*sizeof(UINT));
4619: for ( j = k = 0; j < col; j++ )
4620: if ( !rhead[j] ) v[k++] = svect[j];
1.76 noro 4621: spsugar[sprow] = MAX(maxrs,SG(spol));
1.74 noro 4622: sprow++;
4623: }
1.76 noro 4624: nd_free(spol);
1.74 noro 4625: }
4626: /* elimination (2nd step) */
4627: colstat = (int *)ALLOCA(spcol*sizeof(int));
4628: if ( m == -1 )
1.76 noro 4629: rank = nd_gauss_elim_sf(spmat,spsugar,sprow,spcol,m,colstat);
1.74 noro 4630: else
1.76 noro 4631: rank = nd_gauss_elim_mod(spmat,spsugar,sprow,spcol,m,colstat);
1.74 noro 4632: nd_send_int(rank);
4633: for ( i = 0; i < rank; i++ ) {
4634: nf = vect_to_ndv(spmat[i],spcol,col,rhead,s0vect);
4635: nd_send_ndv(nf);
1.63 noro 4636: }
1.74 noro 4637: fflush(nd_write);
1.107 noro 4638: }
4639:
1.112 ! noro 4640: /* XXX */
! 4641: int generic_gauss_elim_z(MAT m,MAT *nm,Z *dn,int **ri,int **ci)
! 4642: {}
! 4643:
! 4644: int nd_gauss_elim_z(Z **mat0,int *sugar,int row,int col,int *colstat)
1.107 noro 4645: {
1.109 noro 4646: int mod,i,j,t,c,rank,rank0,inv;
4647: int *ci,*ri;
1.112 ! noro 4648: Z dn;
1.109 noro 4649: MAT m,nm;
1.108 noro 4650: int **wmat;
4651:
4652: /* XXX */
4653: mod = 99999989;
4654: wmat = (int **)ALLOCA(row*sizeof(int *));
4655: for ( i = 0; i < row; i++ ) {
4656: wmat[i] = (int *)ALLOCA(col*sizeof(int));
4657: for ( j = 0; j < col; j++ ) {
1.112 ! noro 4658: if ( mat0[i][j] )
! 4659: wmat[i][j] = remzi(mat0[i][j],mod);
! 4660: else
1.108 noro 4661: wmat[i][j] = 0;
4662: }
4663: }
1.109 noro 4664: rank0 = nd_gauss_elim_mod(wmat,sugar,row,col,mod,colstat);
4665: NEWMAT(m); m->row = row; m->col = col; m->body = (pointer **)mat0;
1.112 ! noro 4666: rank = generic_gauss_elim_z(m,&nm,&dn,&ri,&ci);
1.109 noro 4667: if ( rank != rank0 )
4668: error("afo");
4669: for ( i = 0; i < row; i++ )
4670: for ( j = 0; j < col; j++ )
4671: mat0[i][j] = 0;
4672: c = col-rank;
4673: for ( i = 0; i < rank; i++ ) {
1.112 ! noro 4674: mat0[i][ri[i]] = dn;
1.109 noro 4675: for ( j = 0; j < c; j++ )
1.112 ! noro 4676: mat0[i][ci[j]] = (Z)BDY(nm)[i][j];
1.109 noro 4677: }
1.112 ! noro 4678: inv = invm(remzi(dn,mod),mod);
1.109 noro 4679: for ( i = 0; i < row; i++ )
4680: for ( j = 0; j < col; j++ ) {
1.112 ! noro 4681: if ( mat0[i][j] )
! 4682: t = remzi(mat0[i][j],mod);
! 4683: else
1.109 noro 4684: t = 0;
4685: c = dmar(t,inv,0,mod);
4686: if ( wmat[i][j] != c )
4687: error("afo");
4688: }
4689: return rank;
1.76 noro 4690: }
4691:
4692: int nd_gauss_elim_mod(int **mat0,int *sugar,int row,int col,int md,int *colstat)
4693: {
4694: int i,j,k,l,inv,a,rank,s;
4695: unsigned int *t,*pivot,*pk;
4696: unsigned int **mat;
4697:
4698: mat = (unsigned int **)mat0;
4699: for ( rank = 0, j = 0; j < col; j++ ) {
4700: for ( i = rank; i < row; i++ )
4701: mat[i][j] %= md;
4702: for ( i = rank; i < row; i++ )
4703: if ( mat[i][j] )
4704: break;
4705: if ( i == row ) {
4706: colstat[j] = 0;
4707: continue;
4708: } else
4709: colstat[j] = 1;
4710: if ( i != rank ) {
4711: t = mat[i]; mat[i] = mat[rank]; mat[rank] = t;
4712: s = sugar[i]; sugar[i] = sugar[rank]; sugar[rank] = s;
4713: }
4714: pivot = mat[rank];
4715: s = sugar[rank];
4716: inv = invm(pivot[j],md);
4717: for ( k = j, pk = pivot+k; k < col; k++, pk++ )
4718: if ( *pk ) {
4719: if ( *pk >= (unsigned int)md )
4720: *pk %= md;
4721: DMAR(*pk,inv,0,md,*pk)
4722: }
4723: for ( i = rank+1; i < row; i++ ) {
4724: t = mat[i];
4725: if ( a = t[j] ) {
4726: sugar[i] = MAX(sugar[i],s);
4727: red_by_vect(md,t+j,pivot+j,md-a,col-j);
4728: }
4729: }
4730: rank++;
4731: }
4732: for ( j = col-1, l = rank-1; j >= 0; j-- )
4733: if ( colstat[j] ) {
4734: pivot = mat[l];
4735: s = sugar[l];
4736: for ( i = 0; i < l; i++ ) {
4737: t = mat[i];
4738: t[j] %= md;
4739: if ( a = t[j] ) {
4740: sugar[i] = MAX(sugar[i],s);
4741: red_by_vect(md,t+j,pivot+j,md-a,col-j);
4742: }
4743: }
4744: l--;
4745: }
4746: for ( j = 0, l = 0; l < rank; j++ )
4747: if ( colstat[j] ) {
4748: t = mat[l];
4749: for ( k = j; k < col; k++ )
4750: if ( t[k] >= (unsigned int)md )
4751: t[k] %= md;
4752: l++;
4753: }
4754: return rank;
4755: }
4756:
4757: int nd_gauss_elim_sf(int **mat0,int *sugar,int row,int col,int md,int *colstat)
4758: {
4759: int i,j,k,l,inv,a,rank,s;
4760: unsigned int *t,*pivot,*pk;
4761: unsigned int **mat;
4762:
4763: mat = (unsigned int **)mat0;
4764: for ( rank = 0, j = 0; j < col; j++ ) {
4765: for ( i = rank; i < row; i++ )
4766: if ( mat[i][j] )
4767: break;
4768: if ( i == row ) {
4769: colstat[j] = 0;
4770: continue;
4771: } else
4772: colstat[j] = 1;
4773: if ( i != rank ) {
4774: t = mat[i]; mat[i] = mat[rank]; mat[rank] = t;
4775: s = sugar[i]; sugar[i] = sugar[rank]; sugar[rank] = s;
4776: }
4777: pivot = mat[rank];
4778: s = sugar[rank];
4779: inv = _invsf(pivot[j]);
4780: for ( k = j, pk = pivot+k; k < col; k++, pk++ )
4781: if ( *pk )
4782: *pk = _mulsf(*pk,inv);
4783: for ( i = rank+1; i < row; i++ ) {
4784: t = mat[i];
4785: if ( a = t[j] ) {
4786: sugar[i] = MAX(sugar[i],s);
4787: red_by_vect_sf(md,t+j,pivot+j,_chsgnsf(a),col-j);
4788: }
4789: }
4790: rank++;
4791: }
4792: for ( j = col-1, l = rank-1; j >= 0; j-- )
4793: if ( colstat[j] ) {
4794: pivot = mat[l];
4795: s = sugar[l];
4796: for ( i = 0; i < l; i++ ) {
4797: t = mat[i];
4798: if ( a = t[j] ) {
4799: sugar[i] = MAX(sugar[i],s);
4800: red_by_vect_sf(md,t+j,pivot+j,_chsgnsf(a),col-j);
4801: }
4802: }
4803: l--;
4804: }
4805: return rank;
1.77 noro 4806: }
4807:
4808: int ndv_ishomo(NDV p)
4809: {
4810: NMV m;
4811: int len,h;
4812:
4813: if ( !p ) return 1;
4814: len = LEN(p);
4815: m = BDY(p);
4816: h = TD(DL(m));
4817: NMV_ADV(m);
4818: for ( len--; len; len--, NMV_ADV(m) )
4819: if ( TD(DL(m)) != h ) return 0;
4820: return 1;
4821: }
4822:
4823: void ndv_save(NDV p,int index)
4824: {
4825: FILE *s;
4826: char name[BUFSIZ];
4827: short id;
4828: int nv,sugar,len,n,i,td,e,j;
4829: NMV m;
4830: unsigned int *dl;
4831:
4832: sprintf(name,"%s/%d",Demand,index);
4833: s = fopen(name,"w");
4834: savevl(s,0);
4835: if ( !p ) {
4836: saveobj(s,0);
4837: return;
4838: }
4839: id = O_DP;
4840: nv = NV(p);
4841: sugar = SG(p);
4842: len = LEN(p);
4843: write_short(s,&id); write_int(s,&nv); write_int(s,&sugar);
4844: write_int(s,&len);
4845:
4846: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
1.112 ! noro 4847: saveobj(s,(Obj)ztoq(CZ(m)));
1.77 noro 4848: dl = DL(m);
4849: td = TD(dl);
4850: write_int(s,&td);
4851: for ( j = 0; j < nv; j++ ) {
4852: e = GET_EXP(dl,j);
4853: write_int(s,&e);
4854: }
4855: }
4856: fclose(s);
4857: }
4858:
4859: NDV ndv_load(int index)
4860: {
4861: FILE *s;
4862: char name[BUFSIZ];
4863: short id;
4864: int nv,sugar,len,n,i,td,e,j;
4865: NDV d;
4866: NMV m0,m;
4867: unsigned int *dl;
4868: Obj obj;
4869:
4870: sprintf(name,"%s/%d",Demand,index);
4871: s = fopen(name,"r");
4872: if ( !s ) return 0;
4873:
4874: skipvl(s);
4875: read_short(s,&id);
4876: if ( !id ) return 0;
4877: read_int(s,&nv);
4878: read_int(s,&sugar);
4879: read_int(s,&len);
4880:
4881: m0 = m = MALLOC(len*nmv_adv);
4882: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
1.112 ! noro 4883: loadobj(s,&obj); CZ(m) = qtoz((Q)obj);
1.77 noro 4884: dl = DL(m);
4885: ndl_zero(dl);
4886: read_int(s,&td); TD(dl) = td;
4887: for ( j = 0; j < nv; j++ ) {
4888: read_int(s,&e);
4889: PUT_EXP(dl,j,e);
4890: }
4891: if ( nd_blockmask ) ndl_weight_mask(dl);
4892: }
4893: fclose(s);
4894: MKNDV(nv,m0,len,d);
4895: SG(d) = sugar;
4896: return d;
1.99 noro 4897: }
4898:
1.102 noro 4899: void nd_det(int mod,MAT f,P *rp)
1.99 noro 4900: {
4901: VL fv,tv;
1.102 noro 4902: int n,i,j,max,e,nvar,sgn,k0,l0,len0,len,k,l,a;
1.99 noro 4903: pointer **m;
1.112 ! noro 4904: Z mone;
1.102 noro 4905: NDV **dm;
4906: NDV *t,*mi,*mj;
4907: NDV d,s,mij,mjj;
4908: ND u;
4909: NMV nmv;
1.99 noro 4910: PGeoBucket bucket;
4911: struct order_spec *ord;
4912:
4913: create_order_spec(0,0,&ord);
4914: nd_init_ord(ord);
4915: get_vars((Obj)f,&fv);
4916: if ( f->row != f->col )
4917: error("nd_det : non-square matrix");
4918: n = f->row;
4919: for ( nvar = 0, tv = fv; tv; tv = NEXT(tv), nvar++ );
4920: m = f->body;
4921: for ( i = 0, max = 0; i < n; i++ )
4922: for ( j = 0; j < n; j++ )
4923: for ( tv = fv; tv; tv = NEXT(tv) ) {
4924: e = getdeg(tv->v,(P)m[i][j]);
4925: max = MAX(e,max);
4926: }
4927: nd_setup_parameters(nvar,1024);
1.102 noro 4928: dm = (NDV **)almat_pointer(n,n);
1.99 noro 4929: for ( i = 0, max = 0; i < n; i++ )
1.102 noro 4930: for ( j = 0; j < n; j++ ) {
4931: dm[i][j] = ptondv(CO,fv,m[i][j]);
4932: if ( mod ) ndv_mod(mod,dm[i][j]);
4933: if ( dm[i][j] && !LEN(dm[i][j]) ) dm[i][j] = 0;
4934: }
4935: d = ptondv(CO,fv,(P)ONE);
4936: if ( mod ) ndv_mod(mod,d);
1.112 ! noro 4937: mone = chsgnz(qtoz(ONE));
1.99 noro 4938: for ( j = 0, sgn = 1; j < n; j++ ) {
1.104 noro 4939: if ( DP_Print ) fprintf(stderr,"j=%d\n",j);
1.99 noro 4940: for ( i = j; i < n && !dm[i][j]; i++ );
4941: if ( i == n ) {
4942: *rp = 0;
4943: return;
4944: }
1.102 noro 4945: k0 = i; l0 = j; len0 = LEN(dm[k0][l0]);
1.99 noro 4946: for ( k = j; k < n; k++ )
4947: for ( l = j; l < n; l++ )
1.102 noro 4948: if ( dm[k][l] && LEN(dm[k][l]) < len0 ) {
4949: k0 = k; l0 = l; len0 = LEN(dm[k][l]);
1.99 noro 4950: }
4951: if ( k0 != j ) {
4952: t = dm[j]; dm[j] = dm[k0]; dm[k0] = t;
4953: sgn = -sgn;
4954: }
4955: if ( l0 != j ) {
4956: for ( k = j; k < n; k++ ) {
4957: s = dm[k][j]; dm[k][j] = dm[k][l0]; dm[k][l0] = s;
4958: }
4959: sgn = -sgn;
4960: }
4961: for ( i = j+1, mj = dm[j], mjj = mj[j]; i < n; i++ ) {
1.104 noro 4962: if ( DP_Print ) fprintf(stderr," i=%d\n ",i);
1.99 noro 4963: mi = dm[i]; mij = mi[j];
1.102 noro 4964: if ( mod )
4965: ndv_mul_c(mod,mij,mod-1);
4966: else
1.112 ! noro 4967: ndv_mul_c_z(mij,mone);
1.99 noro 4968: for ( k = j+1; k < n; k++ ) {
1.104 noro 4969: if ( DP_Print ) fprintf(stderr,"k=%d ",k);
1.99 noro 4970: bucket = create_pbucket();
1.104 noro 4971: if ( mi[k] ) {
1.102 noro 4972: nmv = BDY(mjj); len = LEN(mjj);
4973: for ( a = 0; a < len; a++, NMV_ADV(nmv) ) {
4974: u = ndv_mul_nmv_trunc(mod,nmv,mi[k],DL(BDY(d)));
4975: add_pbucket(mod,bucket,u);
1.99 noro 4976: }
1.104 noro 4977: }
1.99 noro 4978: if ( mj[k] && mij ) {
1.102 noro 4979: nmv = BDY(mij); len = LEN(mij);
4980: for ( a = 0; a < len; a++, NMV_ADV(nmv) ) {
4981: u = ndv_mul_nmv_trunc(mod,nmv,mj[k],DL(BDY(d)));
4982: add_pbucket(mod,bucket,u);
1.99 noro 4983: }
4984: }
1.104 noro 4985: u = nd_quo(mod,bucket,d);
1.102 noro 4986: mi[k] = ndtondv(mod,u);
1.99 noro 4987: }
1.104 noro 4988: if ( DP_Print ) fprintf(stderr,"\n",k);
1.99 noro 4989: }
4990: d = mjj;
4991: }
1.100 noro 4992: if ( sgn < 0 )
1.102 noro 4993: if ( mod )
4994: ndv_mul_c(mod,d,mod-1);
4995: else
1.112 ! noro 4996: ndv_mul_c_z(d,mone);
1.102 noro 4997: *rp = ndvtop(mod,CO,fv,d);
1.99 noro 4998: }
4999:
1.102 noro 5000: ND ndv_mul_nmv_trunc(int mod,NMV m0,NDV p,UINT *d)
1.99 noro 5001: {
5002: NM mr,mr0;
1.102 noro 5003: NM tnm;
5004: NMV m;
1.99 noro 5005: UINT *d0,*dt,*dm;
5006: int c,n,td,i,c1,c2,len;
1.112 ! noro 5007: Z q;
1.99 noro 5008: ND r;
5009:
5010: if ( !p ) return 0;
5011: else {
1.102 noro 5012: n = NV(p); m = BDY(p); len = LEN(p);
1.99 noro 5013: d0 = DL(m0);
1.102 noro 5014: td = TD(d);
1.99 noro 5015: mr0 = 0;
5016: NEWNM(tnm);
1.102 noro 5017: if ( mod ) {
5018: c = CM(m0);
5019: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
5020: ndl_add(DL(m),d0,DL(tnm));
5021: if ( ndl_reducible(DL(tnm),d) ) {
5022: NEXTNM(mr0,mr);
5023: c1 = CM(m); DMAR(c1,c,0,mod,c2); CM(mr) = c2;
1.104 noro 5024: ndl_copy(DL(tnm),DL(mr));
1.102 noro 5025: }
5026: }
5027: } else {
1.112 ! noro 5028: q = CZ(m0);
1.102 noro 5029: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
5030: ndl_add(DL(m),d0,DL(tnm));
5031: if ( ndl_reducible(DL(tnm),d) ) {
5032: NEXTNM(mr0,mr);
1.112 ! noro 5033: CZ(mr) = mulz(CZ(m),q);
1.104 noro 5034: ndl_copy(DL(tnm),DL(mr));
1.102 noro 5035: }
1.99 noro 5036: }
5037: }
5038: if ( !mr0 )
5039: return 0;
5040: else {
5041: NEXT(mr) = 0;
1.105 noro 5042: for ( len = 0, mr = mr0; mr; mr = NEXT(mr), len++ );
1.99 noro 5043: MKND(NV(p),mr0,len,r);
5044: SG(r) = SG(p) + TD(d0);
5045: return r;
5046: }
5047: }
1.59 noro 5048: }
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>