Annotation of OpenXM_contrib2/asir2000/engine/nd.c, Revision 1.104
1.104 ! noro 1: /* $OpenXM: OpenXM_contrib2/asir2000/engine/nd.c,v 1.103 2004/09/15 06:06:42 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);
1.34 noro 919: else if ( !mod ) return nd_add_q(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:
964: int nm_destructive_add_q(NM *m1,NM *m2,NM *mr0,NM *mr)
965: {
966: NM s;
967: Q t;
968: int can;
969:
970: addq(CQ(*m1),CQ(*m2),&t);
971: s = *m1; *m1 = NEXT(*m1);
972: if ( t ) {
973: can = 1; NEXTNM2(*mr0,*mr,s); CQ(*mr) = (t);
974: } else {
975: can = 2; FREENM(s);
976: }
977: s = *m2; *m2 = NEXT(*m2); FREENM(s);
978: return can;
979: }
980:
981: ND nd_add_q(ND p1,ND p2)
982: {
983: int n,c,can;
984: ND r;
985: NM m1,m2,mr0,mr,s;
986: Q t;
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__)
997: can += nm_destructive_add_q(&m1,&m2,&mr0,&mr);
998: #else
999: addq(CQ(m1),CQ(m2),&t);
1000: s = m1; m1 = NEXT(m1);
1001: if ( t ) {
1002: can++; NEXTNM2(mr0,mr,s); CQ(mr) = (t);
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 */
1.69 noro 1080: int nd_nf(int mod,ND g,NDV *ps,int full,NDC 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;
1.69 noro 1089: Q cg,cred,gcd,tq,qq;
1.21 noro 1090: double hmag;
1.1 noro 1091:
1.69 noro 1092: if ( dn ) {
1093: if ( mod )
1094: dn->m = 1;
1095: else
1096: dn->z = ONE;
1097: }
1.1 noro 1098: if ( !g ) {
1099: *rp = 0;
1100: return 1;
1101: }
1.34 noro 1102: if ( !mod ) hmag = ((double)p_mag((P)HCQ(g)))*nd_scale;
1.21 noro 1103:
1.14 noro 1104: sugar0 = sugar = SG(g);
1.1 noro 1105: n = NV(g);
1.61 noro 1106: mul = (NM)ALLOCA(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT));
1.1 noro 1107: for ( d = 0; g; ) {
1.63 noro 1108: index = ndl_find_reducer(HDL(g));
1.6 noro 1109: if ( index >= 0 ) {
1.17 noro 1110: h = nd_psh[index];
1111: ndl_sub(HDL(g),DL(h),DL(mul));
1.14 noro 1112: if ( ndl_check_bound2(index,DL(mul)) ) {
1.6 noro 1113: nd_free(g); nd_free(d);
1114: return 0;
1115: }
1.77 noro 1116: if ( nd_demand )
1117: p = ndv_load(index);
1118: else
1119: p = ps[index];
1.71 noro 1120: if ( mod == -1 )
1121: CM(mul) = _mulsf(_invsf(HCM(p)),_chsgnsf(HCM(g)));
1122: else if ( mod ) {
1.19 noro 1123: c1 = invm(HCM(p),mod); c2 = mod-HCM(g);
1124: DMAR(c1,c2,0,mod,c); CM(mul) = c;
1.16 noro 1125: } else {
1.17 noro 1126: igcd_cofactor(HCQ(g),HCQ(p),&gcd,&cg,&cred);
1.16 noro 1127: chsgnq(cg,&CQ(mul));
1.20 noro 1128: nd_mul_c_q(d,cred); nd_mul_c_q(g,cred);
1.69 noro 1129: if ( dn ) {
1130: mulq(dn->z,cred,&tq); dn->z = tq;
1131: }
1.16 noro 1132: }
1.55 noro 1133: g = nd_add(mod,g,ndv_mul_nm(mod,mul,p));
1.34 noro 1134: sugar = MAX(sugar,SG(p)+TD(DL(mul)));
1.22 noro 1135: if ( !mod && hmag && g && ((double)(p_mag((P)HCQ(g))) > hmag) ) {
1.69 noro 1136: tq = HCQ(g);
1.21 noro 1137: nd_removecont2(d,g);
1.69 noro 1138: if ( dn ) {
1139: divq(tq,HCQ(g),&qq); divq(dn->z,qq,&tq); dn->z = tq;
1140: }
1.21 noro 1141: hmag = ((double)p_mag((P)HCQ(g)))*nd_scale;
1142: }
1.1 noro 1143: } else if ( !full ) {
1144: *rp = g;
1145: return 1;
1146: } else {
1147: m = BDY(g);
1148: if ( NEXT(m) ) {
1.34 noro 1149: BDY(g) = NEXT(m); NEXT(m) = 0; LEN(g)--;
1.1 noro 1150: } else {
1151: FREEND(g); g = 0;
1152: }
1153: if ( d ) {
1.34 noro 1154: NEXT(tail)=m; tail=m; LEN(d)++;
1.1 noro 1155: } else {
1.34 noro 1156: MKND(n,m,1,d); tail = BDY(d);
1.1 noro 1157: }
1158: }
1159: }
1.34 noro 1160: if ( d ) SG(d) = sugar;
1.1 noro 1161: *rp = d;
1162: return 1;
1163: }
1.28 noro 1164:
1.53 noro 1165: int nd_nf_pbucket(int mod,ND g,NDV *ps,int full,ND *rp)
1.25 noro 1166: {
1167: int hindex,index;
1168: NDV p;
1169: ND u,d,red;
1170: NODE l;
1.31 noro 1171: NM mul,m,mrd,tail;
1.25 noro 1172: int sugar,psugar,n,h_reducible;
1173: PGeoBucket bucket;
1174: int c,c1,c2;
1.26 noro 1175: Q cg,cred,gcd,zzz;
1.25 noro 1176: RHist h;
1.28 noro 1177: double hmag,gmag;
1.77 noro 1178: int count = 0;
1179: int hcount = 0;
1.25 noro 1180:
1181: if ( !g ) {
1182: *rp = 0;
1183: return 1;
1184: }
1185: sugar = SG(g);
1186: n = NV(g);
1.34 noro 1187: if ( !mod ) hmag = ((double)p_mag((P)HCQ(g)))*nd_scale;
1.25 noro 1188: bucket = create_pbucket();
1.31 noro 1189: add_pbucket(mod,bucket,g);
1.25 noro 1190: d = 0;
1.61 noro 1191: mul = (NM)ALLOCA(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT));
1.25 noro 1192: while ( 1 ) {
1.26 noro 1193: hindex = mod?head_pbucket(mod,bucket):head_pbucket_q(bucket);
1.25 noro 1194: if ( hindex < 0 ) {
1.77 noro 1195: if ( DP_Print > 3 ) printf("(%d %d)",count,hcount);
1.34 noro 1196: if ( d ) SG(d) = sugar;
1.25 noro 1197: *rp = d;
1198: return 1;
1199: }
1200: g = bucket->body[hindex];
1.63 noro 1201: index = ndl_find_reducer(HDL(g));
1.25 noro 1202: if ( index >= 0 ) {
1.77 noro 1203: count++;
1204: if ( !d ) hcount++;
1.25 noro 1205: h = nd_psh[index];
1206: ndl_sub(HDL(g),DL(h),DL(mul));
1207: if ( ndl_check_bound2(index,DL(mul)) ) {
1.26 noro 1208: nd_free(d);
1.25 noro 1209: free_pbucket(bucket);
1210: *rp = 0;
1211: return 0;
1212: }
1.53 noro 1213: p = ps[index];
1.71 noro 1214: if ( mod == -1 )
1215: CM(mul) = _mulsf(_invsf(HCM(p)),_chsgnsf(HCM(g)));
1216: else if ( mod ) {
1.25 noro 1217: c1 = invm(HCM(p),mod); c2 = mod-HCM(g);
1218: DMAR(c1,c2,0,mod,c); CM(mul) = c;
1219: } else {
1220: igcd_cofactor(HCQ(g),HCQ(p),&gcd,&cg,&cred);
1221: chsgnq(cg,&CQ(mul));
1.26 noro 1222: nd_mul_c_q(d,cred);
1223: mulq_pbucket(bucket,cred);
1224: g = bucket->body[hindex];
1.28 noro 1225: gmag = (double)p_mag((P)HCQ(g));
1.25 noro 1226: }
1.55 noro 1227: red = ndv_mul_nm(mod,mul,p);
1.25 noro 1228: bucket->body[hindex] = nd_remove_head(g);
1229: red = nd_remove_head(red);
1.31 noro 1230: add_pbucket(mod,bucket,red);
1.34 noro 1231: psugar = SG(p)+TD(DL(mul));
1232: sugar = MAX(sugar,psugar);
1.28 noro 1233: if ( !mod && hmag && (gmag > hmag) ) {
1234: g = normalize_pbucket(mod,bucket);
1235: if ( !g ) {
1.34 noro 1236: if ( d ) SG(d) = sugar;
1.28 noro 1237: *rp = d;
1238: return 1;
1239: }
1240: nd_removecont2(d,g);
1241: hmag = ((double)p_mag((P)HCQ(g)))*nd_scale;
1.31 noro 1242: add_pbucket(mod,bucket,g);
1.28 noro 1243: }
1.25 noro 1244: } else if ( !full ) {
1245: g = normalize_pbucket(mod,bucket);
1.34 noro 1246: if ( g ) SG(g) = sugar;
1.25 noro 1247: *rp = g;
1248: return 1;
1249: } else {
1250: m = BDY(g);
1251: if ( NEXT(m) ) {
1.34 noro 1252: BDY(g) = NEXT(m); NEXT(m) = 0; LEN(g)--;
1.25 noro 1253: } else {
1254: FREEND(g); g = 0;
1255: }
1256: bucket->body[hindex] = g;
1257: NEXT(m) = 0;
1258: if ( d ) {
1.34 noro 1259: NEXT(tail)=m; tail=m; LEN(d)++;
1.25 noro 1260: } else {
1.34 noro 1261: MKND(n,m,1,d); tail = BDY(d);
1.25 noro 1262: }
1263: }
1264: }
1265: }
1.27 noro 1266:
1.61 noro 1267: /* input : list of NDV, cand : list of NDV */
1.28 noro 1268:
1.61 noro 1269: int ndv_check_candidate(NODE input,int obpe,int oadv,EPOS oepos,NODE cand)
1.28 noro 1270: {
1271: int n,i,stat;
1272: ND nf,d;
1.61 noro 1273: NDV r;
1.45 noro 1274: NODE t,s;
1.69 noro 1275: union oNDC dn;
1.45 noro 1276:
1.61 noro 1277: ndv_setup(0,0,cand);
1.31 noro 1278: n = length(cand);
1.28 noro 1279:
1280: /* membercheck : list is a subset of Id(cand) ? */
1281: for ( t = input; t; t = NEXT(t) ) {
1.45 noro 1282: again:
1.61 noro 1283: if ( nd_bpe > obpe )
1284: r = ndv_dup_realloc((NDV)BDY(t),obpe,oadv,oepos);
1285: else
1286: r = (NDV)BDY(t);
1287: d = ndvtond(0,r);
1.69 noro 1288: stat = nd_nf(0,d,nd_ps,0,0,&nf);
1.45 noro 1289: if ( !stat ) {
1.103 noro 1290: nd_reconstruct(0,0);
1.45 noro 1291: goto again;
1292: } else if ( nf ) return 0;
1.71 noro 1293: if ( DP_Print ) { printf("."); fflush(stdout); }
1.28 noro 1294: }
1.71 noro 1295: if ( DP_Print ) { printf("\n"); }
1.28 noro 1296: /* gbcheck : cand is a GB of Id(cand) ? */
1.77 noro 1297: if ( !nd_gb(0,0,1) ) return 0;
1.28 noro 1298: /* XXX */
1.23 noro 1299: return 1;
1300: }
1.1 noro 1301:
1302: ND nd_remove_head(ND p)
1303: {
1304: NM m;
1305:
1306: m = BDY(p);
1307: if ( !NEXT(m) ) {
1.34 noro 1308: FREEND(p); p = 0;
1.31 noro 1309: } else {
1.34 noro 1310: BDY(p) = NEXT(m); LEN(p)--;
1.31 noro 1311: }
1.1 noro 1312: FREENM(m);
1313: return p;
1314: }
1315:
1.69 noro 1316: ND nd_separate_head(ND p,ND *head)
1317: {
1318: NM m,m0;
1319: ND r;
1320:
1321: m = BDY(p);
1322: if ( !NEXT(m) ) {
1323: *head = p; p = 0;
1324: } else {
1325: m0 = m;
1326: BDY(p) = NEXT(m); LEN(p)--;
1327: NEXT(m0) = 0;
1328: MKND(NV(p),m0,1,r);
1329: *head = r;
1330: }
1331: return p;
1332: }
1333:
1.1 noro 1334: PGeoBucket create_pbucket()
1335: {
1336: PGeoBucket g;
1337:
1338: g = CALLOC(1,sizeof(struct oPGeoBucket));
1339: g->m = -1;
1340: return g;
1341: }
1342:
1.25 noro 1343: void free_pbucket(PGeoBucket b) {
1344: int i;
1345:
1.26 noro 1346: for ( i = 0; i <= b->m; i++ )
1.25 noro 1347: if ( b->body[i] ) {
1348: nd_free(b->body[i]);
1349: b->body[i] = 0;
1350: }
1351: GC_free(b);
1352: }
1353:
1.63 noro 1354: void add_pbucket_symbolic(PGeoBucket g,ND d)
1355: {
1356: int l,i,k,m;
1357:
1358: if ( !d )
1359: return;
1360: l = LEN(d);
1361: for ( k = 0, m = 1; l > m; k++, m <<= 1 );
1362: /* 2^(k-1) < l <= 2^k (=m) */
1363: d = nd_merge(g->body[k],d);
1364: for ( ; d && LEN(d) > m; k++, m <<= 1 ) {
1365: g->body[k] = 0;
1366: d = nd_merge(g->body[k+1],d);
1367: }
1368: g->body[k] = d;
1369: g->m = MAX(g->m,k);
1370: }
1371:
1.31 noro 1372: void add_pbucket(int mod,PGeoBucket g,ND d)
1.1 noro 1373: {
1.31 noro 1374: int l,i,k,m;
1.1 noro 1375:
1.31 noro 1376: if ( !d )
1377: return;
1378: l = LEN(d);
1.29 noro 1379: for ( k = 0, m = 1; l > m; k++, m <<= 1 );
1380: /* 2^(k-1) < l <= 2^k (=m) */
1.31 noro 1381: d = nd_add(mod,g->body[k],d);
1382: for ( ; d && LEN(d) > m; k++, m <<= 1 ) {
1.1 noro 1383: g->body[k] = 0;
1.31 noro 1384: d = nd_add(mod,g->body[k+1],d);
1.1 noro 1385: }
1386: g->body[k] = d;
1387: g->m = MAX(g->m,k);
1388: }
1389:
1.26 noro 1390: void mulq_pbucket(PGeoBucket g,Q c)
1391: {
1392: int k;
1393:
1394: for ( k = 0; k <= g->m; k++ )
1395: nd_mul_c_q(g->body[k],c);
1396: }
1397:
1.63 noro 1398: NM remove_head_pbucket_symbolic(PGeoBucket g)
1399: {
1400: int j,i,k,c;
1401: NM head;
1402:
1403: k = g->m;
1404: j = -1;
1405: for ( i = 0; i <= k; i++ ) {
1406: if ( !g->body[i] ) continue;
1407: if ( j < 0 ) j = i;
1408: else {
1409: c = DL_COMPARE(HDL(g->body[i]),HDL(g->body[j]));
1410: if ( c > 0 )
1411: j = i;
1412: else if ( c == 0 )
1413: g->body[i] = nd_remove_head(g->body[i]);
1414: }
1415: }
1416: if ( j < 0 ) return 0;
1417: else {
1418: head = BDY(g->body[j]);
1419: if ( !NEXT(head) ) {
1420: FREEND(g->body[j]);
1421: g->body[j] = 0;
1422: } else {
1423: BDY(g->body[j]) = NEXT(head);
1424: LEN(g->body[j])--;
1425: }
1426: return head;
1427: }
1428: }
1429:
1.19 noro 1430: int head_pbucket(int mod,PGeoBucket g)
1.1 noro 1431: {
1432: int j,i,c,k,nv,sum;
1.61 noro 1433: UINT *di,*dj;
1.1 noro 1434: ND gi,gj;
1435:
1436: k = g->m;
1437: while ( 1 ) {
1438: j = -1;
1439: for ( i = 0; i <= k; i++ ) {
1440: if ( !(gi = g->body[i]) )
1441: continue;
1442: if ( j < 0 ) {
1443: j = i;
1444: gj = g->body[j];
1445: dj = HDL(gj);
1.14 noro 1446: sum = HCM(gj);
1.1 noro 1447: } else {
1.34 noro 1448: c = DL_COMPARE(HDL(gi),dj);
1.1 noro 1449: if ( c > 0 ) {
1.34 noro 1450: if ( sum ) HCM(gj) = sum;
1451: else g->body[j] = nd_remove_head(gj);
1.1 noro 1452: j = i;
1453: gj = g->body[j];
1454: dj = HDL(gj);
1.14 noro 1455: sum = HCM(gj);
1.1 noro 1456: } else if ( c == 0 ) {
1.71 noro 1457: if ( mod == -1 )
1458: sum = _addsf(sum,HCM(gi));
1459: else {
1460: sum = sum+HCM(gi)-mod;
1461: if ( sum < 0 ) sum += mod;
1462: }
1.1 noro 1463: g->body[i] = nd_remove_head(gi);
1464: }
1465: }
1466: }
1.34 noro 1467: if ( j < 0 ) return -1;
1.1 noro 1468: else if ( sum ) {
1.14 noro 1469: HCM(gj) = sum;
1.26 noro 1470: return j;
1.31 noro 1471: } else
1.26 noro 1472: g->body[j] = nd_remove_head(gj);
1473: }
1474: }
1475:
1476: int head_pbucket_q(PGeoBucket g)
1477: {
1478: int j,i,c,k,nv;
1479: Q sum,t;
1480: ND gi,gj;
1481:
1482: k = g->m;
1483: while ( 1 ) {
1484: j = -1;
1485: for ( i = 0; i <= k; i++ ) {
1.34 noro 1486: if ( !(gi = g->body[i]) ) continue;
1.26 noro 1487: if ( j < 0 ) {
1488: j = i;
1489: gj = g->body[j];
1490: sum = HCQ(gj);
1491: } else {
1492: nv = NV(gi);
1.34 noro 1493: c = DL_COMPARE(HDL(gi),HDL(gj));
1.26 noro 1494: if ( c > 0 ) {
1.34 noro 1495: if ( sum ) HCQ(gj) = sum;
1496: else g->body[j] = nd_remove_head(gj);
1.26 noro 1497: j = i;
1498: gj = g->body[j];
1499: sum = HCQ(gj);
1500: } else if ( c == 0 ) {
1501: addq(sum,HCQ(gi),&t);
1502: sum = t;
1503: g->body[i] = nd_remove_head(gi);
1504: }
1505: }
1506: }
1.34 noro 1507: if ( j < 0 ) return -1;
1.26 noro 1508: else if ( sum ) {
1509: HCQ(gj) = sum;
1.1 noro 1510: return j;
1.31 noro 1511: } else
1.1 noro 1512: g->body[j] = nd_remove_head(gj);
1513: }
1514: }
1515:
1.25 noro 1516: ND normalize_pbucket(int mod,PGeoBucket g)
1.1 noro 1517: {
1.31 noro 1518: int i;
1.1 noro 1519: ND r,t;
1520:
1521: r = 0;
1.28 noro 1522: for ( i = 0; i <= g->m; i++ ) {
1.31 noro 1523: r = nd_add(mod,r,g->body[i]);
1.28 noro 1524: g->body[i] = 0;
1525: }
1526: g->m = -1;
1.1 noro 1527: return r;
1528: }
1529:
1.92 noro 1530: void do_diagonalize(int sugar,int m)
1531: {
1532: int i,nh,stat;
1533: NODE r,g,t;
1534: ND h,nf,s,head;
1535: NDV nfv;
1536: Q q,num,den;
1537: union oNDC dn;
1538:
1539: for ( i = nd_psn-1; i >= 0 && SG(nd_psh[i]) == sugar; i-- ) {
1540: if ( nd_demand )
1541: nfv = ndv_load(i);
1542: else
1543: nfv = nd_ps[i];
1544: s = ndvtond(m,nfv);
1545: s = nd_separate_head(s,&head);
1546: nd_nf(m,s,nd_ps,1,&dn,&nf);
1547: if ( !m ) {
1548: NTOQ(NM(dn.z),SGN(dn.z),num);
1549: mulq(HCQ(head),num,&q); HCQ(head) = q;
1550: if ( DN(dn.z) ) {
1551: NTOQ(DN(dn.z),1,den);
1552: nd_mul_c_q(nf,den);
1553: }
1554: }
1555: nf = nd_add(m,head,nf);
1556: ndv_free(nfv);
1557: nd_removecont(m,nf);
1558: nfv = ndtondv(m,nf);
1559: nd_free(nf);
1560: nd_bound[i] = ndv_compute_bound(nfv);
1561: if ( nd_demand ) {
1562: ndv_save(nfv,i);
1563: ndv_free(nfv);
1564: } else
1565: nd_ps[i] = nfv;
1566: }
1567: }
1568:
1.27 noro 1569: /* return value = 0 => input is not a GB */
1570:
1.77 noro 1571: NODE nd_gb(int m,int ishomo,int checkonly)
1.1 noro 1572: {
1573: int i,nh,sugar,stat;
1.23 noro 1574: NODE r,g,t;
1.1 noro 1575: ND_pairs d;
1576: ND_pairs l;
1.77 noro 1577: ND h,nf,s,head;
1.63 noro 1578: NDV nfv;
1.85 noro 1579: Q q,num,den;
1.69 noro 1580: union oNDC dn;
1.1 noro 1581:
1.23 noro 1582: g = 0; d = 0;
1583: for ( i = 0; i < nd_psn; i++ ) {
1.1 noro 1584: d = update_pairs(d,g,i);
1585: g = update_base(g,i);
1586: }
1587: sugar = 0;
1588: while ( d ) {
1589: again:
1590: l = nd_minp(d,&d);
1.14 noro 1591: if ( SG(l) != sugar ) {
1.92 noro 1592: if ( ishomo ) do_diagonalize(sugar,m);
1593:
1.14 noro 1594: sugar = SG(l);
1.71 noro 1595: if ( DP_Print ) fprintf(asir_out,"%d",sugar);
1.1 noro 1596: }
1.53 noro 1597: stat = nd_sp(m,0,l,&h);
1.1 noro 1598: if ( !stat ) {
1599: NEXT(l) = d; d = l;
1.103 noro 1600: d = nd_reconstruct(0,d);
1.1 noro 1601: goto again;
1602: }
1.41 noro 1603: #if USE_GEOBUCKET
1.69 noro 1604: stat = m?nd_nf_pbucket(m,h,nd_ps,!Top,&nf):nd_nf(m,h,nd_ps,!Top,0,&nf);
1.41 noro 1605: #else
1.69 noro 1606: stat = nd_nf(m,h,nd_ps,!Top,0,&nf);
1.41 noro 1607: #endif
1.1 noro 1608: if ( !stat ) {
1609: NEXT(l) = d; d = l;
1.103 noro 1610: d = nd_reconstruct(0,d);
1.1 noro 1611: goto again;
1612: } else if ( nf ) {
1.27 noro 1613: if ( checkonly ) return 0;
1.71 noro 1614: if ( DP_Print ) { printf("+"); fflush(stdout); }
1.63 noro 1615: nd_removecont(m,nf);
1616: nfv = ndtondv(m,nf); nd_free(nf);
1.77 noro 1617: nh = ndv_newps(m,nfv,0);
1.1 noro 1618: d = update_pairs(d,g,nh);
1619: g = update_base(g,nh);
1620: FREENDP(l);
1621: } else {
1.71 noro 1622: if ( DP_Print ) { printf("."); fflush(stdout); }
1.1 noro 1623: FREENDP(l);
1624: }
1625: }
1.77 noro 1626: if ( nd_demand )
1627: for ( t = g; t; t = NEXT(t) )
1628: BDY(t) = (pointer)ndv_load((int)BDY(t));
1629: else
1630: for ( t = g; t; t = NEXT(t) )
1631: BDY(t) = (pointer)nd_ps[(int)BDY(t)];
1.1 noro 1632: return g;
1633: }
1634:
1.92 noro 1635: void do_diagonalize_trace(int sugar,int m)
1.91 noro 1636: {
1637: int i,nh,stat;
1638: NODE r,g,t;
1639: ND h,nf,nfq,s,head;
1640: NDV nfv,nfqv;
1641: Q q,den,num;
1642: union oNDC dn;
1643:
1644: for ( i = nd_psn-1; i >= 0 && SG(nd_psh[i]) == sugar; i-- ) {
1645: /* for nd_ps */
1646: s = ndvtond(m,nd_ps[i]);
1647: s = nd_separate_head(s,&head);
1648: nd_nf_pbucket(m,s,nd_ps,1,&nf);
1649: nf = nd_add(m,head,nf);
1650: ndv_free(nd_ps[i]);
1651: nd_ps[i] = ndtondv(m,nf);
1652: nd_free(nf);
1653:
1654: /* for nd_ps_trace */
1655: if ( nd_demand )
1656: nfv = ndv_load(i);
1657: else
1658: nfv = nd_ps_trace[i];
1659: s = ndvtond(0,nfv);
1660: s = nd_separate_head(s,&head);
1661: nd_nf(0,s,nd_ps_trace,1,&dn,&nf);
1662: NTOQ(NM(dn.z),SGN(dn.z),num);
1663: mulq(HCQ(head),num,&q); HCQ(head) = q;
1664: if ( DN(dn.z) ) {
1665: NTOQ(DN(dn.z),1,den);
1666: nd_mul_c_q(nf,den);
1667: }
1668: nf = nd_add(0,head,nf);
1669: ndv_free(nfv);
1670: nd_removecont(0,nf);
1671: nfv = ndtondv(0,nf);
1672: nd_free(nf);
1673: nd_bound[i] = ndv_compute_bound(nfv);
1674: if ( nd_demand ) {
1675: ndv_save(nfv,i);
1676: ndv_free(nfv);
1677: } else
1678: nd_ps_trace[i] = nfv;
1679: }
1680: }
1681:
1.82 noro 1682: NODE nd_gb_trace(int m,int ishomo)
1.20 noro 1683: {
1684: int i,nh,sugar,stat;
1.23 noro 1685: NODE r,g,t;
1.20 noro 1686: ND_pairs d;
1687: ND_pairs l;
1.82 noro 1688: ND h,nf,nfq,s,head;
1.63 noro 1689: NDV nfv,nfqv;
1.85 noro 1690: Q q,den,num;
1.69 noro 1691: union oNDC dn;
1.20 noro 1692:
1.23 noro 1693: g = 0; d = 0;
1694: for ( i = 0; i < nd_psn; i++ ) {
1.20 noro 1695: d = update_pairs(d,g,i);
1696: g = update_base(g,i);
1697: }
1698: sugar = 0;
1699: while ( d ) {
1700: again:
1701: l = nd_minp(d,&d);
1702: if ( SG(l) != sugar ) {
1.93 noro 1703: if ( ishomo ) do_diagonalize_trace(sugar,m);
1.20 noro 1704: sugar = SG(l);
1.71 noro 1705: if ( DP_Print ) fprintf(asir_out,"%d",sugar);
1.20 noro 1706: }
1.53 noro 1707: stat = nd_sp(m,0,l,&h);
1.20 noro 1708: if ( !stat ) {
1709: NEXT(l) = d; d = l;
1.103 noro 1710: d = nd_reconstruct(1,d);
1.20 noro 1711: goto again;
1712: }
1.41 noro 1713: #if USE_GEOBUCKET
1.53 noro 1714: stat = nd_nf_pbucket(m,h,nd_ps,!Top,&nf);
1.41 noro 1715: #else
1.69 noro 1716: stat = nd_nf(m,h,nd_ps,!Top,0,&nf);
1.41 noro 1717: #endif
1.20 noro 1718: if ( !stat ) {
1719: NEXT(l) = d; d = l;
1.103 noro 1720: d = nd_reconstruct(1,d);
1.20 noro 1721: goto again;
1722: } else if ( nf ) {
1.77 noro 1723: if ( nd_demand ) {
1724: nfqv = ndv_load(nd_psn);
1725: nfq = ndvtond(0,nfqv);
1726: } else
1727: nfq = 0;
1728: if ( !nfq ) {
1729: if ( !nd_sp(0,1,l,&h) || !nd_nf(0,h,nd_ps_trace,!Top,0,&nfq) ) {
1730: NEXT(l) = d; d = l;
1.103 noro 1731: d = nd_reconstruct(1,d);
1.77 noro 1732: goto again;
1733: }
1.70 noro 1734: }
1.20 noro 1735: if ( nfq ) {
1.70 noro 1736: /* m|HC(nfq) => failure */
1.63 noro 1737: if ( !rem(NM(HCQ(nfq)),m) ) return 0;
1738:
1.71 noro 1739: if ( DP_Print ) { printf("+"); fflush(stdout); }
1.63 noro 1740: nd_removecont(m,nf); nfv = ndtondv(m,nf); nd_free(nf);
1741: nd_removecont(0,nfq); nfqv = ndtondv(0,nfq); nd_free(nfq);
1.77 noro 1742: nh = ndv_newps(0,nfv,nfqv);
1.20 noro 1743: d = update_pairs(d,g,nh);
1744: g = update_base(g,nh);
1745: } else {
1.71 noro 1746: if ( DP_Print ) { printf("*"); fflush(stdout); }
1.20 noro 1747: }
1748: } else {
1.71 noro 1749: if ( DP_Print ) { printf("."); fflush(stdout); }
1.20 noro 1750: }
1751: FREENDP(l);
1752: }
1.77 noro 1753: if ( nd_demand )
1754: for ( t = g; t; t = NEXT(t) )
1755: BDY(t) = (pointer)ndv_load((int)BDY(t));
1756: else
1757: for ( t = g; t; t = NEXT(t) )
1758: BDY(t) = (pointer)nd_ps_trace[(int)BDY(t)];
1.20 noro 1759: return g;
1760: }
1761:
1.23 noro 1762: int ndv_compare(NDV *p1,NDV *p2)
1763: {
1.34 noro 1764: return DL_COMPARE(HDL(*p1),HDL(*p2));
1.23 noro 1765: }
1766:
1767: int ndv_compare_rev(NDV *p1,NDV *p2)
1768: {
1.34 noro 1769: return -DL_COMPARE(HDL(*p1),HDL(*p2));
1.23 noro 1770: }
1771:
1.61 noro 1772: NODE ndv_reduceall(int m,NODE f)
1.23 noro 1773: {
1.69 noro 1774: int i,n,stat;
1775: ND nf,g,head;
1.23 noro 1776: NODE t,a0,a;
1.69 noro 1777: union oNDC dn;
1778: NDV *w;
1.85 noro 1779: Q q,num,den;
1.23 noro 1780:
1.69 noro 1781: n = length(f);
1782: #if 0
1783: w = (NDV *)ALLOCA(n*sizeof(NDV));
1784: for ( i = 0, t = f; i < n; i++, t = NEXT(t) ) w[i] = (NDV)BDY(t);
1785: qsort(w,n,sizeof(NDV),
1786: (int (*)(const void *,const void *))ndv_compare);
1787: for ( t = f, i = 0; t; i++, t = NEXT(t) ) BDY(t) = (pointer)w[i];
1788: #endif
1789: ndv_setup(m,0,f);
1790: for ( i = 0; i < n; ) {
1791: g = ndvtond(m,nd_ps[i]);
1792: g = nd_separate_head(g,&head);
1793: stat = nd_nf(m,g,nd_ps,1,&dn,&nf);
1.50 noro 1794: if ( !stat )
1.103 noro 1795: nd_reconstruct(0,0);
1.69 noro 1796: else {
1.71 noro 1797: if ( DP_Print ) { printf("."); fflush(stdout); }
1.85 noro 1798: if ( !m ) {
1799: NTOQ(NM(dn.z),SGN(dn.z),num);
1800: mulq(HCQ(head),num,&q); HCQ(head) = q;
1801: if ( DN(dn.z) ) {
1802: NTOQ(DN(dn.z),1,den);
1803: nd_mul_c_q(nf,den);
1804: }
1805: }
1.69 noro 1806: nf = nd_add(m,head,nf);
1807: ndv_free(nd_ps[i]);
1.24 noro 1808: nd_removecont(m,nf);
1.69 noro 1809: nd_ps[i] = ndtondv(m,nf); nd_free(nf);
1810: nd_bound[i] = ndv_compute_bound(nd_ps[i]);
1.50 noro 1811: i++;
1.23 noro 1812: }
1813: }
1.71 noro 1814: if ( DP_Print ) { printf("\n"); }
1.23 noro 1815: for ( a0 = 0, i = 0; i < n; i++ ) {
1816: NEXTNODE(a0,a);
1.69 noro 1817: BDY(a) = (pointer)nd_ps[i];
1.23 noro 1818: }
1819: NEXT(a) = 0;
1820: return a0;
1821: }
1822:
1.1 noro 1823: ND_pairs update_pairs( ND_pairs d, NODE /* of index */ g, int t)
1824: {
1825: ND_pairs d1,nd,cur,head,prev,remove;
1826:
1827: if ( !g ) return d;
1828: d = crit_B(d,t);
1829: d1 = nd_newpairs(g,t);
1830: d1 = crit_M(d1);
1831: d1 = crit_F(d1);
1.55 noro 1832: if ( do_weyl )
1833: head = d1;
1834: else {
1835: prev = 0; cur = head = d1;
1836: while ( cur ) {
1837: if ( crit_2( cur->i1,cur->i2 ) ) {
1838: remove = cur;
1839: if ( !prev ) head = cur = NEXT(cur);
1840: else cur = NEXT(prev) = NEXT(cur);
1841: FREENDP(remove);
1842: } else {
1843: prev = cur; cur = NEXT(cur);
1844: }
1.1 noro 1845: }
1846: }
1847: if ( !d )
1848: return head;
1849: else {
1850: nd = d;
1.34 noro 1851: while ( NEXT(nd) ) nd = NEXT(nd);
1.1 noro 1852: NEXT(nd) = head;
1853: return d;
1854: }
1855: }
1856:
1857: ND_pairs nd_newpairs( NODE g, int t )
1858: {
1859: NODE h;
1.61 noro 1860: UINT *dl;
1.34 noro 1861: int ts,s;
1.1 noro 1862: ND_pairs r,r0;
1863:
1.20 noro 1864: dl = DL(nd_psh[t]);
1.34 noro 1865: ts = SG(nd_psh[t]) - TD(dl);
1.1 noro 1866: for ( r0 = 0, h = g; h; h = NEXT(h) ) {
1867: NEXTND_pairs(r0,r);
1868: r->i1 = (int)BDY(h);
1869: r->i2 = t;
1.20 noro 1870: ndl_lcm(DL(nd_psh[r->i1]),dl,r->lcm);
1.34 noro 1871: s = SG(nd_psh[r->i1])-TD(DL(nd_psh[r->i1]));
1872: SG(r) = MAX(s,ts) + TD(LCM(r));
1.1 noro 1873: }
1874: NEXT(r) = 0;
1875: return r0;
1876: }
1877:
1878: ND_pairs crit_B( ND_pairs d, int s )
1879: {
1880: ND_pairs cur,head,prev,remove;
1.61 noro 1881: UINT *t,*tl,*lcm;
1.1 noro 1882: int td,tdl;
1883:
1884: if ( !d ) return 0;
1.20 noro 1885: t = DL(nd_psh[s]);
1.1 noro 1886: prev = 0;
1887: head = cur = d;
1.61 noro 1888: lcm = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
1.1 noro 1889: while ( cur ) {
1890: tl = cur->lcm;
1891: if ( ndl_reducible(tl,t)
1.20 noro 1892: && (ndl_lcm(DL(nd_psh[cur->i1]),t,lcm),!ndl_equal(lcm,tl))
1893: && (ndl_lcm(DL(nd_psh[cur->i2]),t,lcm),!ndl_equal(lcm,tl)) ) {
1.1 noro 1894: remove = cur;
1895: if ( !prev ) {
1896: head = cur = NEXT(cur);
1897: } else {
1898: cur = NEXT(prev) = NEXT(cur);
1899: }
1900: FREENDP(remove);
1901: } else {
1.34 noro 1902: prev = cur; cur = NEXT(cur);
1.1 noro 1903: }
1904: }
1905: return head;
1906: }
1907:
1908: ND_pairs crit_M( ND_pairs d1 )
1909: {
1910: ND_pairs e,d2,d3,dd,p;
1.61 noro 1911: UINT *id,*jd;
1.1 noro 1912:
1913: for ( dd = 0, e = d1; e; e = d3 ) {
1914: if ( !(d2 = NEXT(e)) ) {
1915: NEXT(e) = dd;
1916: return e;
1917: }
1.34 noro 1918: id = LCM(e);
1.1 noro 1919: for ( d3 = 0; d2; d2 = p ) {
1.34 noro 1920: p = NEXT(d2);
1921: jd = LCM(d2);
1922: if ( ndl_equal(jd,id) )
1923: ;
1924: else if ( TD(jd) > TD(id) )
1.1 noro 1925: if ( ndl_reducible(jd,id) ) continue;
1926: else ;
1.34 noro 1927: else if ( ndl_reducible(id,jd) ) goto delit;
1.1 noro 1928: NEXT(d2) = d3;
1929: d3 = d2;
1930: }
1931: NEXT(e) = dd;
1932: dd = e;
1933: continue;
1934: /**/
1935: delit: NEXT(d2) = d3;
1936: d3 = d2;
1937: for ( ; p; p = d2 ) {
1938: d2 = NEXT(p);
1939: NEXT(p) = d3;
1940: d3 = p;
1941: }
1942: FREENDP(e);
1943: }
1944: return dd;
1945: }
1946:
1947: ND_pairs crit_F( ND_pairs d1 )
1948: {
1949: ND_pairs rest, head,remove;
1950: ND_pairs last, p, r, w;
1951: int s;
1952:
1953: for ( head = last = 0, p = d1; NEXT(p); ) {
1954: r = w = equivalent_pairs(p,&rest);
1.14 noro 1955: s = SG(r);
1.1 noro 1956: w = NEXT(w);
1957: while ( w ) {
1958: if ( crit_2(w->i1,w->i2) ) {
1959: r = w;
1960: w = NEXT(w);
1961: while ( w ) {
1962: remove = w;
1963: w = NEXT(w);
1964: FREENDP(remove);
1965: }
1966: break;
1.14 noro 1967: } else if ( SG(w) < s ) {
1.1 noro 1968: FREENDP(r);
1969: r = w;
1.14 noro 1970: s = SG(r);
1.1 noro 1971: w = NEXT(w);
1972: } else {
1973: remove = w;
1974: w = NEXT(w);
1975: FREENDP(remove);
1976: }
1977: }
1978: if ( last ) NEXT(last) = r;
1979: else head = r;
1980: NEXT(last = r) = 0;
1981: p = rest;
1982: if ( !p ) return head;
1983: }
1984: if ( !last ) return p;
1985: NEXT(last) = p;
1986: return head;
1987: }
1988:
1989: int crit_2( int dp1, int dp2 )
1990: {
1.20 noro 1991: return ndl_disjoint(DL(nd_psh[dp1]),DL(nd_psh[dp2]));
1.1 noro 1992: }
1993:
1.40 noro 1994: ND_pairs equivalent_pairs( ND_pairs d1, ND_pairs *prest )
1.1 noro 1995: {
1996: ND_pairs w,p,r,s;
1.61 noro 1997: UINT *d;
1.1 noro 1998:
1999: w = d1;
1.34 noro 2000: d = LCM(w);
1.1 noro 2001: s = NEXT(w);
2002: NEXT(w) = 0;
2003: for ( r = 0; s; s = p ) {
2004: p = NEXT(s);
1.34 noro 2005: if ( ndl_equal(d,LCM(s)) ) {
1.39 noro 2006: NEXT(s) = w; w = s;
1.1 noro 2007: } else {
1.39 noro 2008: NEXT(s) = r; r = s;
1.1 noro 2009: }
2010: }
2011: *prest = r;
2012: return w;
2013: }
2014:
2015: NODE update_base(NODE nd,int ndp)
2016: {
1.61 noro 2017: UINT *dl, *dln;
1.1 noro 2018: NODE last, p, head;
2019:
1.20 noro 2020: dl = DL(nd_psh[ndp]);
1.1 noro 2021: for ( head = last = 0, p = nd; p; ) {
1.20 noro 2022: dln = DL(nd_psh[(int)BDY(p)]);
1.34 noro 2023: if ( ndl_reducible( dln, dl ) ) {
1.1 noro 2024: p = NEXT(p);
2025: if ( last ) NEXT(last) = p;
2026: } else {
2027: if ( !last ) head = p;
2028: p = NEXT(last = p);
2029: }
2030: }
2031: head = append_one(head,ndp);
2032: return head;
2033: }
2034:
2035: ND_pairs nd_minp( ND_pairs d, ND_pairs *prest )
2036: {
2037: ND_pairs m,ml,p,l;
1.61 noro 2038: UINT *lcm;
1.33 noro 2039: int s,td,len,tlen,c,c1;
1.1 noro 2040:
2041: if ( !(p = NEXT(m = d)) ) {
2042: *prest = p;
2043: NEXT(m) = 0;
2044: return m;
2045: }
1.14 noro 2046: s = SG(m);
1.77 noro 2047: if ( !NoSugar ) {
2048: for ( ml = 0, l = m; p; p = NEXT(l = p) )
2049: if ( (SG(p) < s)
2050: || ((SG(p) == s) && (DL_COMPARE(LCM(p),LCM(m)) < 0)) ) {
2051: ml = l; m = p; s = SG(m);
2052: }
2053: } else {
2054: for ( ml = 0, l = m; p; p = NEXT(l = p) )
2055: if ( DL_COMPARE(LCM(p),LCM(m)) < 0 ) {
2056: ml = l; m = p; s = SG(m);
2057: }
2058: }
1.1 noro 2059: if ( !ml ) *prest = NEXT(m);
2060: else {
2061: NEXT(ml) = NEXT(m);
2062: *prest = d;
2063: }
2064: NEXT(m) = 0;
2065: return m;
2066: }
2067:
1.63 noro 2068: ND_pairs nd_minsugarp( ND_pairs d, ND_pairs *prest )
2069: {
1.69 noro 2070: int msugar,i;
1.63 noro 2071: ND_pairs t,dm0,dm,dr0,dr;
2072:
2073: for ( msugar = SG(d), t = NEXT(d); t; t = NEXT(t) )
2074: if ( SG(t) < msugar ) msugar = SG(t);
2075: dm0 = 0; dr0 = 0;
1.69 noro 2076: for ( i = 0, t = d; t; t = NEXT(t) )
2077: if ( i < nd_f4_nsp && SG(t) == msugar ) {
1.63 noro 2078: if ( dm0 ) NEXT(dm) = t;
2079: else dm0 = t;
2080: dm = t;
1.69 noro 2081: i++;
1.63 noro 2082: } else {
2083: if ( dr0 ) NEXT(dr) = t;
2084: else dr0 = t;
2085: dr = t;
2086: }
2087: NEXT(dm) = 0;
2088: if ( dr0 ) NEXT(dr) = 0;
2089: *prest = dr0;
2090: return dm0;
2091: }
2092:
1.77 noro 2093: int ndv_newps(int m,NDV a,NDV aq)
1.1 noro 2094: {
1.3 noro 2095: int len;
1.13 noro 2096: RHist r;
1.20 noro 2097: NDV b;
1.3 noro 2098:
1.1 noro 2099: if ( nd_psn == nd_pslen ) {
2100: nd_pslen *= 2;
1.11 noro 2101: nd_ps = (NDV *)REALLOC((char *)nd_ps,nd_pslen*sizeof(NDV));
1.53 noro 2102: nd_ps_trace = (NDV *)REALLOC((char *)nd_ps_trace,nd_pslen*sizeof(NDV));
1.13 noro 2103: nd_psh = (RHist *)REALLOC((char *)nd_psh,nd_pslen*sizeof(RHist));
1.61 noro 2104: nd_bound = (UINT **)
2105: REALLOC((char *)nd_bound,nd_pslen*sizeof(UINT *));
1.1 noro 2106: }
1.39 noro 2107: NEWRHist(r); nd_psh[nd_psn] = r;
1.63 noro 2108: nd_ps[nd_psn] = a;
1.39 noro 2109: if ( aq ) {
1.63 noro 2110: nd_ps_trace[nd_psn] = aq;
2111: nd_bound[nd_psn] = ndv_compute_bound(aq);
1.39 noro 2112: SG(r) = SG(aq); ndl_copy(HDL(aq),DL(r));
1.53 noro 2113: } else {
1.63 noro 2114: nd_bound[nd_psn] = ndv_compute_bound(a);
1.53 noro 2115: SG(r) = SG(a); ndl_copy(HDL(a),DL(r));
1.39 noro 2116: }
1.77 noro 2117: if ( nd_demand ) {
2118: if ( aq ) {
2119: ndv_save(nd_ps_trace[nd_psn],nd_psn);
2120: nd_ps_trace[nd_psn] = 0;
2121: } else {
2122: ndv_save(nd_ps[nd_psn],nd_psn);
2123: nd_ps[nd_psn] = 0;
2124: }
2125: }
1.1 noro 2126: return nd_psn++;
2127: }
2128:
1.61 noro 2129: void ndv_setup(int mod,int trace,NODE f)
1.1 noro 2130: {
1.5 noro 2131: int i,j,td,len,max;
1.1 noro 2132: NODE s,s0,f0;
1.61 noro 2133: UINT *d;
1.13 noro 2134: RHist r;
1.69 noro 2135: NDV *w;
1.61 noro 2136: NDV a,am;
1.11 noro 2137:
2138: nd_found = 0; nd_notfirst = 0; nd_create = 0;
1.1 noro 2139:
1.69 noro 2140: for ( nd_psn = 0, s = f; s; s = NEXT(s) ) if ( BDY(s) ) nd_psn++;
2141: w = (NDV *)ALLOCA(nd_psn*sizeof(NDV));
2142: for ( i = 0, s = f; s; s = NEXT(s) ) if ( BDY(s) ) w[i++] = BDY(s);
2143: qsort(w,nd_psn,sizeof(NDV),
2144: (int (*)(const void *,const void *))ndv_compare);
2145: nd_pslen = 2*nd_psn;
1.11 noro 2146: nd_ps = (NDV *)MALLOC(nd_pslen*sizeof(NDV));
1.53 noro 2147: nd_ps_trace = (NDV *)MALLOC(nd_pslen*sizeof(NDV));
1.13 noro 2148: nd_psh = (RHist *)MALLOC(nd_pslen*sizeof(RHist));
1.61 noro 2149: nd_bound = (UINT **)MALLOC(nd_pslen*sizeof(UINT *));
1.57 noro 2150:
1.11 noro 2151: if ( !nd_red )
1.13 noro 2152: nd_red = (RHist *)MALLOC(REDTAB_LEN*sizeof(RHist));
1.71 noro 2153: for ( i = 0; i < REDTAB_LEN; i++ ) nd_red[i] = 0;
1.69 noro 2154: for ( i = 0; i < nd_psn; i++ ) {
1.39 noro 2155: if ( trace ) {
1.69 noro 2156: a = nd_ps_trace[i] = ndv_dup(0,w[i]);
1.61 noro 2157: ndv_removecont(0,a);
2158: am = nd_ps[i] = ndv_dup(mod,a);
2159: ndv_mod(mod,am);
2160: ndv_removecont(mod,am);
2161: } else {
1.69 noro 2162: a = nd_ps[i] = ndv_dup(mod,w[i]);
1.61 noro 2163: ndv_removecont(mod,a);
1.39 noro 2164: }
1.61 noro 2165: NEWRHist(r); SG(r) = HTD(a); ndl_copy(HDL(a),DL(r));
1.57 noro 2166: nd_bound[i] = ndv_compute_bound(a);
1.20 noro 2167: nd_psh[i] = r;
1.77 noro 2168: if ( nd_demand ) {
2169: if ( trace ) {
2170: ndv_save(nd_ps_trace[i],i);
2171: nd_ps_trace[i] = 0;
2172: } else {
2173: ndv_save(nd_ps[i],i);
2174: nd_ps[i] = 0;
2175: }
2176: }
1.20 noro 2177: }
2178: }
2179:
1.63 noro 2180: void nd_gr(LIST f,LIST v,int m,int f4,struct order_spec *ord,LIST *rp)
1.1 noro 2181: {
1.61 noro 2182: VL tv,fv,vv,vc;
1.1 noro 2183: NODE fd,fd0,r,r0,t,x,s,xx;
1.61 noro 2184: int e,max,nvar;
1.69 noro 2185: NDV b;
1.77 noro 2186: int ishomo;
1.1 noro 2187:
1.78 noro 2188: if ( !m && Demand ) nd_demand = 1;
2189: else nd_demand = 0;
2190:
1.103 noro 2191: #if 0
1.74 noro 2192: ndv_alloc = 0;
1.103 noro 2193: #endif
1.1 noro 2194: get_vars((Obj)f,&fv); pltovl(v,&vv);
1.74 noro 2195: for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++ );
1.88 noro 2196: switch ( ord->id ) {
2197: case 1:
2198: if ( ord->nv != nvar )
2199: error("nd_{gr,f4} : invalid order specification");
2200: break;
2201: default:
2202: break;
2203: }
1.32 noro 2204: nd_init_ord(ord);
1.61 noro 2205: for ( t = BDY(f), max = 0; t; t = NEXT(t) )
2206: for ( tv = vv; tv; tv = NEXT(tv) ) {
2207: e = getdeg(tv->v,(P)BDY(t));
2208: max = MAX(e,max);
2209: }
2210: nd_setup_parameters(nvar,max);
1.77 noro 2211: ishomo = 1;
1.1 noro 2212: for ( fd0 = 0, t = BDY(f); t; t = NEXT(t) ) {
1.61 noro 2213: b = (pointer)ptondv(CO,vv,(P)BDY(t));
1.77 noro 2214: if ( ishomo )
2215: ishomo = ishomo && ndv_ishomo(b);
1.69 noro 2216: if ( m ) ndv_mod(m,b);
1.61 noro 2217: if ( b ) { NEXTNODE(fd0,fd); BDY(fd) = (pointer)b; }
1.1 noro 2218: }
2219: if ( fd0 ) NEXT(fd) = 0;
1.61 noro 2220: ndv_setup(m,0,fd0);
1.77 noro 2221: x = f4?nd_f4(m):nd_gb(m,ishomo,0);
1.82 noro 2222: nd_demand = 0;
1.61 noro 2223: x = ndv_reducebase(x);
2224: x = ndv_reduceall(m,x);
1.23 noro 2225: for ( r0 = 0, t = x; t; t = NEXT(t) ) {
1.1 noro 2226: NEXTNODE(r0,r);
1.61 noro 2227: BDY(r) = ndvtop(m,CO,vv,BDY(t));
1.20 noro 2228: }
2229: if ( r0 ) NEXT(r) = 0;
2230: MKLIST(*rp,r0);
1.103 noro 2231: #if 0
1.74 noro 2232: fprintf(asir_out,"ndv_alloc=%d\n",ndv_alloc);
1.103 noro 2233: #endif
1.20 noro 2234: }
2235:
1.52 noro 2236: void nd_gr_trace(LIST f,LIST v,int trace,int homo,struct order_spec *ord,LIST *rp)
1.20 noro 2237: {
1.89 noro 2238: struct order_spec *ord1;
1.61 noro 2239: VL tv,fv,vv,vc;
1.27 noro 2240: NODE fd,fd0,in0,in,r,r0,t,s,cand;
1.61 noro 2241: int m,nocheck,nvar,mindex,e,max;
2242: NDV c;
2243: NMV a;
1.27 noro 2244: P p;
1.61 noro 2245: EPOS oepos;
1.82 noro 2246: int obpe,oadv,wmax,i,len,cbpe,ishomo;
1.20 noro 2247:
2248: get_vars((Obj)f,&fv); pltovl(v,&vv);
1.74 noro 2249: for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++ );
1.88 noro 2250: switch ( ord->id ) {
2251: case 1:
2252: if ( ord->nv != nvar )
2253: error("nd_gr_trace : invalid order specification");
2254: break;
2255: default:
2256: break;
2257: }
1.52 noro 2258: nocheck = 0;
2259: mindex = 0;
2260:
1.78 noro 2261: if ( Demand ) nd_demand = 1;
2262: else nd_demand = 0;
2263:
1.52 noro 2264: /* setup modulus */
2265: if ( trace < 0 ) {
2266: trace = -trace;
2267: nocheck = 1;
2268: }
2269: m = trace > 1 ? trace : get_lprime(mindex);
1.61 noro 2270: for ( t = BDY(f), max = 0; t; t = NEXT(t) )
2271: for ( tv = vv; tv; tv = NEXT(tv) ) {
2272: e = getdeg(tv->v,(P)BDY(t));
2273: max = MAX(e,max);
1.23 noro 2274: }
1.61 noro 2275: nd_init_ord(ord);
2276: nd_setup_parameters(nvar,max);
2277: obpe = nd_bpe; oadv = nmv_adv; oepos = nd_epos;
1.82 noro 2278: ishomo = 1;
1.61 noro 2279: for ( in0 = 0, fd0 = 0, t = BDY(f); t; t = NEXT(t) ) {
2280: c = ptondv(CO,vv,(P)BDY(t));
1.82 noro 2281: if ( ishomo )
2282: ishomo = ishomo && ndv_ishomo(c);
1.61 noro 2283: if ( c ) {
2284: NEXTNODE(in0,in); BDY(in) = (pointer)c;
2285: NEXTNODE(fd0,fd); BDY(fd) = (pointer)ndv_dup(0,c);
1.23 noro 2286: }
1.61 noro 2287: }
2288: if ( in0 ) NEXT(in) = 0;
2289: if ( fd0 ) NEXT(fd) = 0;
1.82 noro 2290: if ( !ishomo && homo ) {
1.61 noro 2291: for ( t = in0, wmax = 0; t; t = NEXT(t) ) {
2292: c = (NDV)BDY(t); len = LEN(c);
2293: for ( a = BDY(c), i = 0; i < len; i++, NMV_ADV(a) )
2294: wmax = MAX(TD(DL(a)),wmax);
2295: }
2296: homogenize_order(ord,nvar,&ord1);
1.89 noro 2297: nd_init_ord(ord1);
1.61 noro 2298: nd_setup_parameters(nvar+1,wmax);
2299: for ( t = fd0; t; t = NEXT(t) )
2300: ndv_homogenize((NDV)BDY(t),obpe,oadv,oepos);
1.27 noro 2301: }
1.52 noro 2302: while ( 1 ) {
1.77 noro 2303: if ( Demand )
2304: nd_demand = 1;
1.61 noro 2305: ndv_setup(m,1,fd0);
1.82 noro 2306: cand = nd_gb_trace(m,ishomo || homo);
1.52 noro 2307: if ( !cand ) {
2308: /* failure */
1.61 noro 2309: if ( trace > 1 ) { *rp = 0; return; }
2310: else m = get_lprime(++mindex);
1.52 noro 2311: continue;
2312: }
1.82 noro 2313: if ( !ishomo && homo ) {
1.27 noro 2314: /* dehomogenization */
1.61 noro 2315: for ( t = cand; t; t = NEXT(t) ) ndv_dehomogenize((NDV)BDY(t),ord);
1.45 noro 2316: nd_init_ord(ord);
1.61 noro 2317: nd_setup_parameters(nvar,0);
1.27 noro 2318: }
1.77 noro 2319: nd_demand = 0;
1.61 noro 2320: cand = ndv_reducebase(cand);
2321: cand = ndv_reduceall(0,cand);
1.78 noro 2322: cbpe = nd_bpe;
1.61 noro 2323: if ( nocheck )
2324: break;
2325: if ( ndv_check_candidate(in0,obpe,oadv,oepos,cand) )
1.52 noro 2326: /* success */
2327: break;
2328: else if ( trace > 1 ) {
2329: /* failure */
2330: *rp = 0; return;
1.61 noro 2331: } else {
1.52 noro 2332: /* try the next modulus */
2333: m = get_lprime(++mindex);
1.61 noro 2334: /* reset the parameters */
1.82 noro 2335: if ( !ishomo && homo ) {
1.89 noro 2336: nd_init_ord(ord1);
1.61 noro 2337: nd_setup_parameters(nvar+1,wmax);
2338: } else {
2339: nd_init_ord(ord);
2340: nd_setup_parameters(nvar,max);
2341: }
2342: }
1.52 noro 2343: }
1.27 noro 2344: /* dp->p */
1.61 noro 2345: nd_bpe = cbpe;
1.73 noro 2346: nd_setup_parameters(nd_nvar,0);
1.61 noro 2347: for ( r = cand; r; r = NEXT(r) ) BDY(r) = (pointer)ndvtop(0,CO,vv,BDY(r));
1.27 noro 2348: MKLIST(*rp,cand);
1.1 noro 2349: }
2350:
1.61 noro 2351: void dltondl(int n,DL dl,UINT *r)
1.1 noro 2352: {
1.61 noro 2353: UINT *d;
1.57 noro 2354: int i,j,l,s,ord_l;
1.43 noro 2355: struct order_pair *op;
1.1 noro 2356:
2357: d = dl->d;
1.41 noro 2358: for ( i = 0; i < nd_wpd; i++ ) r[i] = 0;
1.43 noro 2359: if ( nd_blockmask ) {
2360: l = nd_blockmask->n;
2361: op = nd_blockmask->order_pair;
2362: for ( j = 0, s = 0; j < l; j++ ) {
2363: ord_l = op[j].length;
1.57 noro 2364: for ( i = 0; i < ord_l; i++, s++ ) PUT_EXP(r,s,d[s]);
1.43 noro 2365: }
2366: TD(r) = ndl_weight(r);
1.61 noro 2367: ndl_weight_mask(r);
1.43 noro 2368: } else {
1.56 noro 2369: for ( i = 0; i < n; i++ ) PUT_EXP(r,i,d[i]);
1.43 noro 2370: TD(r) = ndl_weight(r);
2371: }
1.1 noro 2372: }
2373:
1.61 noro 2374: DL ndltodl(int n,UINT *ndl)
1.1 noro 2375: {
2376: DL dl;
2377: int *d;
1.57 noro 2378: int i,j,l,s,ord_l;
1.43 noro 2379: struct order_pair *op;
1.1 noro 2380:
2381: NEWDL(dl,n);
1.34 noro 2382: dl->td = TD(ndl);
1.1 noro 2383: d = dl->d;
1.43 noro 2384: if ( nd_blockmask ) {
2385: l = nd_blockmask->n;
2386: op = nd_blockmask->order_pair;
2387: for ( j = 0, s = 0; j < l; j++ ) {
2388: ord_l = op[j].length;
1.57 noro 2389: for ( i = 0; i < ord_l; i++, s++ ) d[s] = GET_EXP(ndl,s);
1.43 noro 2390: }
2391: } else {
1.56 noro 2392: for ( i = 0; i < n; i++ ) d[i] = GET_EXP(ndl,i);
1.43 noro 2393: }
1.1 noro 2394: return dl;
2395: }
2396:
1.61 noro 2397: void ndl_print(UINT *dl)
1.1 noro 2398: {
2399: int n;
1.57 noro 2400: int i,j,l,ord_l,s,s0;
1.43 noro 2401: struct order_pair *op;
1.1 noro 2402:
2403: n = nd_nvar;
2404: printf("<<");
1.43 noro 2405: if ( nd_blockmask ) {
2406: l = nd_blockmask->n;
2407: op = nd_blockmask->order_pair;
2408: for ( j = 0, s = s0 = 0; j < l; j++ ) {
2409: ord_l = op[j].length;
1.57 noro 2410: for ( i = 0; i < ord_l; i++, s++ )
2411: printf(s==n-1?"%d":"%d,",GET_EXP(dl,s));
1.43 noro 2412: }
2413: } else {
1.56 noro 2414: for ( i = 0; i < n; i++ ) printf(i==n-1?"%d":"%d,",GET_EXP(dl,i));
1.43 noro 2415: }
1.1 noro 2416: printf(">>");
2417: }
2418:
2419: void nd_print(ND p)
2420: {
2421: NM m;
2422:
2423: if ( !p )
2424: printf("0\n");
2425: else {
2426: for ( m = BDY(p); m; m = NEXT(m) ) {
1.71 noro 2427: if ( CM(m) & 0x80000000 ) printf("+@_%d*",IFTOF(CM(m)));
2428: else printf("+%d*",CM(m));
1.14 noro 2429: ndl_print(DL(m));
1.1 noro 2430: }
2431: printf("\n");
2432: }
2433: }
2434:
1.16 noro 2435: void nd_print_q(ND p)
2436: {
2437: NM m;
2438:
2439: if ( !p )
2440: printf("0\n");
2441: else {
2442: for ( m = BDY(p); m; m = NEXT(m) ) {
2443: printf("+");
1.74 noro 2444: printexpr(CO,(Obj)CQ(m));
1.16 noro 2445: printf("*");
2446: ndl_print(DL(m));
2447: }
2448: printf("\n");
2449: }
2450: }
2451:
1.1 noro 2452: void ndp_print(ND_pairs d)
2453: {
2454: ND_pairs t;
2455:
1.34 noro 2456: for ( t = d; t; t = NEXT(t) ) printf("%d,%d ",t->i1,t->i2);
1.1 noro 2457: printf("\n");
2458: }
2459:
1.20 noro 2460: void nd_removecont(int mod,ND p)
1.16 noro 2461: {
2462: int i,n;
2463: Q *w;
2464: Q dvr,t;
2465: NM m;
1.21 noro 2466: struct oVECT v;
2467: N q,r;
1.16 noro 2468:
1.71 noro 2469: if ( mod == -1 ) nd_mul_c(mod,p,_invsf(HCM(p)));
2470: else if ( mod ) nd_mul_c(mod,p,invm(HCM(p),mod));
1.20 noro 2471: else {
2472: for ( m = BDY(p), n = 0; m; m = NEXT(m), n++ );
2473: w = (Q *)ALLOCA(n*sizeof(Q));
1.21 noro 2474: v.len = n;
2475: v.body = (pointer *)w;
1.34 noro 2476: for ( m = BDY(p), i = 0; i < n; m = NEXT(m), i++ ) w[i] = CQ(m);
1.21 noro 2477: removecont_array(w,n);
2478: for ( m = BDY(p), i = 0; i < n; m = NEXT(m), i++ ) CQ(m) = w[i];
1.16 noro 2479: }
2480: }
2481:
1.21 noro 2482: void nd_removecont2(ND p1,ND p2)
2483: {
2484: int i,n1,n2,n;
2485: Q *w;
2486: Q dvr,t;
2487: NM m;
2488: struct oVECT v;
2489: N q,r;
2490:
2491: if ( !p1 ) {
2492: nd_removecont(0,p2); return;
2493: } else if ( !p2 ) {
2494: nd_removecont(0,p1); return;
2495: }
2496: n1 = nd_length(p1);
2497: n2 = nd_length(p2);
2498: n = n1+n2;
2499: w = (Q *)ALLOCA(n*sizeof(Q));
2500: v.len = n;
2501: v.body = (pointer *)w;
1.34 noro 2502: for ( m = BDY(p1), i = 0; i < n1; m = NEXT(m), i++ ) w[i] = CQ(m);
2503: for ( m = BDY(p2); i < n; m = NEXT(m), i++ ) w[i] = CQ(m);
1.21 noro 2504: removecont_array(w,n);
2505: for ( m = BDY(p1), i = 0; i < n1; m = NEXT(m), i++ ) CQ(m) = w[i];
2506: for ( m = BDY(p2); i < n; m = NEXT(m), i++ ) CQ(m) = w[i];
2507: }
2508:
1.20 noro 2509: void ndv_removecont(int mod,NDV p)
1.16 noro 2510: {
2511: int i,len;
2512: Q *w;
2513: Q dvr,t;
2514: NMV m;
2515:
1.71 noro 2516: if ( mod == -1 )
2517: ndv_mul_c(mod,p,_invsf(HCM(p)));
2518: else if ( mod )
1.20 noro 2519: ndv_mul_c(mod,p,invm(HCM(p),mod));
2520: else {
2521: len = p->len;
2522: w = (Q *)ALLOCA(len*sizeof(Q));
1.34 noro 2523: for ( m = BDY(p), i = 0; i < len; NMV_ADV(m), i++ ) w[i] = CQ(m);
1.20 noro 2524: sortbynm(w,len);
2525: qltozl(w,len,&dvr);
2526: for ( m = BDY(p), i = 0; i < len; NMV_ADV(m), i++ ) {
2527: divq(CQ(m),dvr,&t); CQ(m) = t;
2528: }
1.16 noro 2529: }
1.21 noro 2530: }
2531:
1.61 noro 2532: void ndv_homogenize(NDV p,int obpe,int oadv,EPOS oepos)
2533: {
2534: int len,i,max;
2535: NMV m,mr0,mr,t;
2536:
2537: len = p->len;
2538: for ( m = BDY(p), i = 0, max = 0; i < len; NMV_OADV(m), i++ )
2539: max = MAX(max,TD(DL(m)));
2540: mr0 = nmv_adv>oadv?(NMV)REALLOC(BDY(p),len*nmv_adv):BDY(p);
2541: m = (NMV)((char *)mr0+(len-1)*oadv);
2542: mr = (NMV)((char *)mr0+(len-1)*nmv_adv);
2543: t = (NMV)ALLOCA(nmv_adv);
2544: for ( i = 0; i < len; i++, NMV_OPREV(m), NMV_PREV(mr) ) {
2545: ndl_homogenize(DL(m),DL(t),obpe,oepos,max);
2546: CQ(mr) = CQ(m);
2547: ndl_copy(DL(t),DL(mr));
2548: }
2549: NV(p)++;
2550: BDY(p) = mr0;
2551: }
2552:
1.45 noro 2553: void ndv_dehomogenize(NDV p,struct order_spec *ord)
1.23 noro 2554: {
1.45 noro 2555: int i,j,adj,len,newnvar,newwpd,newadv,newexporigin;
1.23 noro 2556: Q *w;
2557: Q dvr,t;
2558: NMV m,r;
2559:
2560: len = p->len;
2561: newnvar = nd_nvar-1;
1.48 noro 2562: newexporigin = nd_get_exporigin(ord);
1.45 noro 2563: newwpd = newnvar/nd_epw+(newnvar%nd_epw?1:0)+newexporigin;
1.23 noro 2564: for ( m = BDY(p), i = 0; i < len; NMV_ADV(m), i++ )
1.34 noro 2565: ndl_dehomogenize(DL(m));
1.23 noro 2566: if ( newwpd != nd_wpd ) {
1.90 noro 2567: newadv = ROUND_FOR_ALIGN(sizeof(struct oNMV)+(newwpd-1)*sizeof(UINT));
1.61 noro 2568: for ( m = r = BDY(p), i = 0; i < len; NMV_ADV(m), NDV_NADV(r), i++ ) {
1.45 noro 2569: CQ(r) = CQ(m);
2570: for ( j = 0; j < newexporigin; j++ ) DL(r)[j] = DL(m)[j];
2571: adj = nd_exporigin-newexporigin;
2572: for ( ; j < newwpd; j++ ) DL(r)[j] = DL(m)[j+adj];
1.23 noro 2573: }
2574: }
2575: NV(p)--;
2576: }
2577:
1.21 noro 2578: void removecont_array(Q *c,int n)
2579: {
2580: struct oVECT v;
2581: Q d0,d1,a,u,u1,gcd;
2582: int i;
2583: N qn,rn,gn;
2584: Q *q,*r;
2585:
2586: q = (Q *)ALLOCA(n*sizeof(Q));
2587: r = (Q *)ALLOCA(n*sizeof(Q));
2588: v.id = O_VECT; v.len = n; v.body = (pointer *)c;
2589: igcdv_estimate(&v,&d0);
2590: for ( i = 0; i < n; i++ ) {
2591: divn(NM(c[i]),NM(d0),&qn,&rn);
2592: NTOQ(qn,SGN(c[i])*SGN(d0),q[i]);
2593: NTOQ(rn,SGN(c[i]),r[i]);
2594: }
1.34 noro 2595: for ( i = 0; i < n; i++ ) if ( r[i] ) break;
1.21 noro 2596: if ( i < n ) {
2597: v.id = O_VECT; v.len = n; v.body = (pointer *)r;
2598: igcdv(&v,&d1);
2599: gcdn(NM(d0),NM(d1),&gn); NTOQ(gn,1,gcd);
2600: divsn(NM(d0),gn,&qn); NTOQ(qn,1,a);
2601: for ( i = 0; i < n; i++ ) {
2602: mulq(a,q[i],&u);
2603: if ( r[i] ) {
2604: divsn(NM(r[i]),gn,&qn); NTOQ(qn,SGN(r[i]),u1);
2605: addq(u,u1,&q[i]);
2606: } else
2607: q[i] = u;
2608: }
2609: }
1.34 noro 2610: for ( i = 0; i < n; i++ ) c[i] = q[i];
1.16 noro 2611: }
2612:
1.19 noro 2613: void nd_mul_c(int mod,ND p,int mul)
1.1 noro 2614: {
2615: NM m;
2616: int c,c1;
2617:
1.34 noro 2618: if ( !p ) return;
1.71 noro 2619: if ( mod == -1 )
2620: for ( m = BDY(p); m; m = NEXT(m) )
2621: CM(m) = _mulsf(CM(m),mul);
2622: else
2623: for ( m = BDY(p); m; m = NEXT(m) ) {
2624: c1 = CM(m); DMAR(c1,mul,0,mod,c); CM(m) = c;
2625: }
1.1 noro 2626: }
2627:
1.16 noro 2628: void nd_mul_c_q(ND p,Q mul)
2629: {
2630: NM m;
2631: Q c;
2632:
1.34 noro 2633: if ( !p ) return;
1.16 noro 2634: for ( m = BDY(p); m; m = NEXT(m) ) {
2635: mulq(CQ(m),mul,&c); CQ(m) = c;
2636: }
2637: }
2638:
1.61 noro 2639: void nd_mul_c_p(VL vl,ND p,P mul)
2640: {
2641: NM m;
2642: P c;
2643:
2644: if ( !p ) return;
2645: for ( m = BDY(p); m; m = NEXT(m) ) {
2646: mulp(vl,CP(m),mul,&c); CP(m) = c;
2647: }
2648: }
2649:
1.1 noro 2650: void nd_free(ND p)
2651: {
2652: NM t,s;
2653:
1.34 noro 2654: if ( !p ) return;
1.1 noro 2655: t = BDY(p);
2656: while ( t ) {
2657: s = NEXT(t);
2658: FREENM(t);
2659: t = s;
2660: }
2661: FREEND(p);
2662: }
2663:
1.23 noro 2664: void ndv_free(NDV p)
2665: {
2666: GC_free(BDY(p));
2667: }
2668:
1.61 noro 2669: void nd_append_red(UINT *d,int i)
1.1 noro 2670: {
1.13 noro 2671: RHist m,m0;
1.1 noro 2672: int h;
2673:
1.13 noro 2674: NEWRHist(m);
1.34 noro 2675: h = ndl_hash_value(d);
1.13 noro 2676: m->index = i;
1.14 noro 2677: ndl_copy(d,DL(m));
1.1 noro 2678: NEXT(m) = nd_red[h];
2679: nd_red[h] = m;
2680: }
2681:
1.61 noro 2682: UINT *ndv_compute_bound(NDV p)
1.1 noro 2683: {
1.61 noro 2684: UINT *d1,*d2,*t;
2685: UINT u;
1.57 noro 2686: int i,j,k,l,len,ind;
1.45 noro 2687: NMV m;
1.1 noro 2688:
2689: if ( !p )
2690: return 0;
1.61 noro 2691: d1 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
2692: d2 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
1.45 noro 2693: len = LEN(p);
2694: m = BDY(p); ndl_copy(DL(m),d1); NMV_ADV(m);
2695: for ( i = 1; i < len; i++, NMV_ADV(m) ) {
1.14 noro 2696: ndl_lcm(DL(m),d1,d2);
1.1 noro 2697: t = d1; d1 = d2; d2 = t;
2698: }
1.12 noro 2699: l = nd_nvar+31;
1.61 noro 2700: t = (UINT *)MALLOC_ATOMIC(l*sizeof(UINT));
1.57 noro 2701: for ( i = nd_exporigin, ind = 0; i < nd_wpd; i++ ) {
2702: u = d1[i];
2703: k = (nd_epw-1)*nd_bpe;
2704: for ( j = 0; j < nd_epw; j++, k -= nd_bpe, ind++ )
2705: t[ind] = (u>>k)&nd_mask0;
2706: }
2707: for ( ; ind < l; ind++ ) t[ind] = 0;
1.1 noro 2708: return t;
2709: }
2710:
1.99 noro 2711: UINT *nd_compute_bound(ND p)
2712: {
2713: UINT *d1,*d2,*t;
2714: UINT u;
2715: int i,j,k,l,len,ind;
2716: NM m;
2717:
2718: if ( !p )
2719: return 0;
2720: d1 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
2721: d2 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
2722: len = LEN(p);
2723: m = BDY(p); ndl_copy(DL(m),d1); m = NEXT(m);
2724: for ( m = NEXT(m); m; m = NEXT(m) ) {
2725: ndl_lcm(DL(m),d1,d2);
2726: t = d1; d1 = d2; d2 = t;
2727: }
2728: l = nd_nvar+31;
2729: t = (UINT *)MALLOC_ATOMIC(l*sizeof(UINT));
2730: for ( i = nd_exporigin, ind = 0; i < nd_wpd; i++ ) {
2731: u = d1[i];
2732: k = (nd_epw-1)*nd_bpe;
2733: for ( j = 0; j < nd_epw; j++, k -= nd_bpe, ind++ )
2734: t[ind] = (u>>k)&nd_mask0;
2735: }
2736: for ( ; ind < l; ind++ ) t[ind] = 0;
2737: return t;
2738: }
2739:
1.48 noro 2740: int nd_get_exporigin(struct order_spec *ord)
2741: {
1.51 noro 2742: switch ( ord->id ) {
1.96 noro 2743: case 0: case 2:
1.48 noro 2744: return 1;
1.41 noro 2745: case 1:
2746: /* block order */
1.43 noro 2747: /* d[0]:weight d[1]:w0,...,d[nd_exporigin-1]:w(n-1) */
1.48 noro 2748: return ord->ord.block.length+1;
1.96 noro 2749: case 3:
2750: error("nd_get_exporigin : composite order is not supported yet.");
1.41 noro 2751: }
1.48 noro 2752: }
2753:
1.61 noro 2754: void nd_setup_parameters(int nvar,int max) {
1.79 noro 2755: int i,j,n,elen,ord_o,ord_l,l,s,wpd;
1.57 noro 2756: struct order_pair *op;
1.48 noro 2757:
1.73 noro 2758: nd_nvar = nvar;
1.79 noro 2759: if ( max ) {
1.83 noro 2760: /* XXX */
2761: if ( do_weyl ) nd_bpe = 32;
2762: else if ( max < 2 ) nd_bpe = 1;
1.79 noro 2763: else if ( max < 4 ) nd_bpe = 2;
2764: else if ( max < 8 ) nd_bpe = 3;
2765: else if ( max < 16 ) nd_bpe = 4;
2766: else if ( max < 32 ) nd_bpe = 5;
2767: else if ( max < 64 ) nd_bpe = 6;
2768: else if ( max < 256 ) nd_bpe = 8;
2769: else if ( max < 1024 ) nd_bpe = 10;
2770: else if ( max < 65536 ) nd_bpe = 16;
2771: else nd_bpe = 32;
2772: }
1.61 noro 2773: nd_epw = (sizeof(UINT)*8)/nd_bpe;
1.48 noro 2774: elen = nd_nvar/nd_epw+(nd_nvar%nd_epw?1:0);
2775: nd_exporigin = nd_get_exporigin(nd_ord);
1.79 noro 2776: wpd = nd_exporigin+elen;
2777: if ( wpd != nd_wpd ) {
2778: nd_free_private_storage();
2779: nd_wpd = wpd;
2780: }
1.1 noro 2781: if ( nd_bpe < 32 ) {
2782: nd_mask0 = (1<<nd_bpe)-1;
2783: } else {
2784: nd_mask0 = 0xffffffff;
2785: }
2786: bzero(nd_mask,sizeof(nd_mask));
2787: nd_mask1 = 0;
2788: for ( i = 0; i < nd_epw; i++ ) {
2789: nd_mask[nd_epw-i-1] = (nd_mask0<<(i*nd_bpe));
2790: nd_mask1 |= (1<<(nd_bpe-1))<<(i*nd_bpe);
2791: }
1.90 noro 2792: nmv_adv = ROUND_FOR_ALIGN(sizeof(struct oNMV)+(nd_wpd-1)*sizeof(UINT));
1.57 noro 2793: nd_epos = nd_create_epos(nd_ord);
1.43 noro 2794: nd_blockmask = nd_create_blockmask(nd_ord);
1.96 noro 2795: nd_work_vector = (int *)REALLOC(nd_work_vector,nd_nvar*sizeof(int));
1.1 noro 2796: }
2797:
1.103 noro 2798: ND_pairs nd_reconstruct(int trace,ND_pairs d)
1.1 noro 2799: {
1.37 noro 2800: int i,obpe,oadv,h;
1.71 noro 2801: static NM prev_nm_free_list;
2802: static ND_pairs prev_ndp_free_list;
1.13 noro 2803: RHist mr0,mr;
2804: RHist r;
1.37 noro 2805: RHist *old_red;
1.71 noro 2806: ND_pairs s0,s,t;
1.43 noro 2807: EPOS oepos;
1.15 noro 2808:
1.1 noro 2809: obpe = nd_bpe;
1.11 noro 2810: oadv = nmv_adv;
1.43 noro 2811: oepos = nd_epos;
1.63 noro 2812: if ( obpe < 2 ) nd_bpe = 2;
2813: else if ( obpe < 3 ) nd_bpe = 3;
1.62 noro 2814: else if ( obpe < 4 ) nd_bpe = 4;
1.63 noro 2815: else if ( obpe < 5 ) nd_bpe = 5;
1.34 noro 2816: else if ( obpe < 6 ) nd_bpe = 6;
2817: else if ( obpe < 8 ) nd_bpe = 8;
1.63 noro 2818: else if ( obpe < 10 ) nd_bpe = 10;
1.34 noro 2819: else if ( obpe < 16 ) nd_bpe = 16;
2820: else if ( obpe < 32 ) nd_bpe = 32;
2821: else error("nd_reconstruct : exponent too large");
1.5 noro 2822:
1.73 noro 2823: nd_setup_parameters(nd_nvar,0);
1.1 noro 2824: prev_nm_free_list = _nm_free_list;
2825: prev_ndp_free_list = _ndp_free_list;
2826: _nm_free_list = 0;
2827: _ndp_free_list = 0;
1.53 noro 2828: for ( i = nd_psn-1; i >= 0; i-- ) ndv_realloc(nd_ps[i],obpe,oadv,oepos);
2829: if ( trace )
2830: for ( i = nd_psn-1; i >= 0; i-- )
2831: ndv_realloc(nd_ps_trace[i],obpe,oadv,oepos);
1.1 noro 2832: s0 = 0;
2833: for ( t = d; t; t = NEXT(t) ) {
2834: NEXTND_pairs(s0,s);
2835: s->i1 = t->i1;
2836: s->i2 = t->i2;
1.14 noro 2837: SG(s) = SG(t);
1.61 noro 2838: ndl_reconstruct(LCM(t),LCM(s),obpe,oepos);
1.1 noro 2839: }
1.37 noro 2840:
2841: old_red = (RHist *)ALLOCA(REDTAB_LEN*sizeof(RHist));
1.6 noro 2842: for ( i = 0; i < REDTAB_LEN; i++ ) {
1.37 noro 2843: old_red[i] = nd_red[i];
2844: nd_red[i] = 0;
2845: }
2846: for ( i = 0; i < REDTAB_LEN; i++ )
2847: for ( r = old_red[i]; r; r = NEXT(r) ) {
2848: NEWRHist(mr);
1.13 noro 2849: mr->index = r->index;
1.20 noro 2850: SG(mr) = SG(r);
1.61 noro 2851: ndl_reconstruct(DL(r),DL(mr),obpe,oepos);
1.37 noro 2852: h = ndl_hash_value(DL(mr));
2853: NEXT(mr) = nd_red[h];
2854: nd_red[h] = mr;
1.6 noro 2855: }
1.37 noro 2856: for ( i = 0; i < REDTAB_LEN; i++ ) old_red[i] = 0;
2857: old_red = 0;
1.11 noro 2858: for ( i = 0; i < nd_psn; i++ ) {
1.20 noro 2859: NEWRHist(r); SG(r) = SG(nd_psh[i]);
1.61 noro 2860: ndl_reconstruct(DL(nd_psh[i]),DL(r),obpe,oepos);
1.13 noro 2861: nd_psh[i] = r;
1.11 noro 2862: }
1.1 noro 2863: if ( s0 ) NEXT(s) = 0;
2864: prev_nm_free_list = 0;
2865: prev_ndp_free_list = 0;
1.71 noro 2866: #if 0
1.1 noro 2867: GC_gcollect();
1.71 noro 2868: #endif
1.1 noro 2869: return s0;
2870: }
2871:
1.61 noro 2872: void ndl_reconstruct(UINT *d,UINT *r,int obpe,EPOS oepos)
1.1 noro 2873: {
1.57 noro 2874: int n,i,ei,oepw,omask0,j,s,ord_l,l;
1.43 noro 2875: struct order_pair *op;
1.1 noro 2876:
2877: n = nd_nvar;
1.61 noro 2878: oepw = (sizeof(UINT)*8)/obpe;
1.43 noro 2879: omask0 = (1<<obpe)-1;
1.34 noro 2880: TD(r) = TD(d);
1.41 noro 2881: for ( i = nd_exporigin; i < nd_wpd; i++ ) r[i] = 0;
1.43 noro 2882: if ( nd_blockmask ) {
2883: l = nd_blockmask->n;
2884: op = nd_blockmask->order_pair;
2885: for ( i = 1; i < nd_exporigin; i++ )
2886: r[i] = d[i];
2887: for ( j = 0, s = 0; j < l; j++ ) {
2888: ord_l = op[j].length;
1.57 noro 2889: for ( i = 0; i < ord_l; i++, s++ ) {
2890: ei = GET_EXP_OLD(d,s);
2891: PUT_EXP(r,s,ei);
2892: }
1.1 noro 2893: }
1.43 noro 2894: } else {
1.56 noro 2895: for ( i = 0; i < n; i++ ) {
2896: ei = GET_EXP_OLD(d,i);
2897: PUT_EXP(r,i,ei);
2898: }
1.1 noro 2899: }
2900: }
1.3 noro 2901:
1.6 noro 2902: ND nd_copy(ND p)
2903: {
2904: NM m,mr,mr0;
1.41 noro 2905: int c,n;
1.6 noro 2906: ND r;
2907:
2908: if ( !p )
2909: return 0;
2910: else {
2911: for ( mr0 = 0, m = BDY(p); m; m = NEXT(m) ) {
2912: NEXTNM(mr0,mr);
1.14 noro 2913: CM(mr) = CM(m);
2914: ndl_copy(DL(m),DL(mr));
1.6 noro 2915: }
2916: NEXT(mr) = 0;
1.31 noro 2917: MKND(NV(p),mr0,LEN(p),r);
1.14 noro 2918: SG(r) = SG(p);
1.6 noro 2919: return r;
2920: }
2921: }
2922:
1.53 noro 2923: int nd_sp(int mod,int trace,ND_pairs p,ND *rp)
1.11 noro 2924: {
2925: NM m;
2926: NDV p1,p2;
2927: ND t1,t2;
1.61 noro 2928: UINT *lcm;
1.31 noro 2929: int td;
1.11 noro 2930:
1.77 noro 2931: if ( !mod && nd_demand ) {
2932: p1 = ndv_load(p->i1); p2 = ndv_load(p->i2);
1.53 noro 2933: } else {
1.77 noro 2934: if ( trace ) {
2935: p1 = nd_ps_trace[p->i1]; p2 = nd_ps_trace[p->i2];
2936: } else {
2937: p1 = nd_ps[p->i1]; p2 = nd_ps[p->i2];
2938: }
1.20 noro 2939: }
1.34 noro 2940: lcm = LCM(p);
1.11 noro 2941: NEWNM(m);
1.20 noro 2942: CQ(m) = HCQ(p2);
1.34 noro 2943: ndl_sub(lcm,HDL(p1),DL(m));
1.56 noro 2944: if ( ndl_check_bound2(p->i1,DL(m)) )
2945: return 0;
1.55 noro 2946: t1 = ndv_mul_nm(mod,m,p1);
1.71 noro 2947: if ( mod == -1 ) CM(m) = _chsgnsf(HCM(p1));
2948: else if ( mod ) CM(m) = mod-HCM(p1);
1.34 noro 2949: else chsgnq(HCQ(p1),&CQ(m));
2950: ndl_sub(lcm,HDL(p2),DL(m));
1.14 noro 2951: if ( ndl_check_bound2(p->i2,DL(m)) ) {
1.11 noro 2952: nd_free(t1);
2953: return 0;
2954: }
1.55 noro 2955: t2 = ndv_mul_nm(mod,m,p2);
1.31 noro 2956: *rp = nd_add(mod,t1,t2);
1.11 noro 2957: FREENM(m);
2958: return 1;
2959: }
2960:
1.19 noro 2961: void ndv_mul_c(int mod,NDV p,int mul)
1.11 noro 2962: {
2963: NMV m;
2964: int c,c1,len,i;
2965:
1.34 noro 2966: if ( !p ) return;
1.14 noro 2967: len = LEN(p);
1.71 noro 2968: if ( mod == -1 )
2969: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) )
2970: CM(m) = _mulsf(CM(m),mul);
2971: else
2972: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
2973: c1 = CM(m); DMAR(c1,mul,0,mod,c); CM(m) = c;
2974: }
1.11 noro 2975: }
2976:
1.16 noro 2977: void ndv_mul_c_q(NDV p,Q mul)
2978: {
2979: NMV m;
2980: Q c;
2981: int len,i;
2982:
1.34 noro 2983: if ( !p ) return;
1.16 noro 2984: len = LEN(p);
2985: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
2986: mulq(CQ(m),mul,&c); CQ(m) = c;
2987: }
2988: }
2989:
1.55 noro 2990: ND weyl_ndv_mul_nm(int mod,NM m0,NDV p) {
2991: int n2,i,j,l,n,tlen;
1.61 noro 2992: UINT *d0;
1.55 noro 2993: NM *tab,*psum;
2994: ND s,r;
2995: NM t;
2996: NMV m1;
2997:
2998: if ( !p ) return 0;
2999: n = NV(p); n2 = n>>1;
3000: d0 = DL(m0);
3001: l = LEN(p);
3002: for ( i = 0, tlen = 1; i < n2; i++ ) tlen *= (GET_EXP(d0,n2+i)+1);
3003: tab = (NM *)ALLOCA(tlen*sizeof(NM));
3004: psum = (NM *)ALLOCA(tlen*sizeof(NM));
3005: for ( i = 0; i < tlen; i++ ) psum[i] = 0;
1.56 noro 3006: m1 = (NMV)(((char *)BDY(p))+nmv_adv*(l-1));
3007: for ( i = l-1; i >= 0; i--, NMV_PREV(m1) ) {
1.55 noro 3008: /* m0(NM) * m1(NMV) => tab(NM) */
1.56 noro 3009: weyl_mul_nm_nmv(n,mod,m0,m1,tab,tlen);
1.55 noro 3010: for ( j = 0; j < tlen; j++ ) {
3011: if ( tab[j] ) {
3012: NEXT(tab[j]) = psum[j]; psum[j] = tab[j];
3013: }
3014: }
3015: }
3016: for ( i = tlen-1, r = 0; i >= 0; i-- )
3017: if ( psum[i] ) {
3018: for ( j = 0, t = psum[i]; t; t = NEXT(t), j++ );
3019: MKND(n,psum[i],j,s);
3020: r = nd_add(mod,r,s);
3021: }
1.56 noro 3022: if ( r ) SG(r) = SG(p)+TD(d0);
3023: return r;
1.55 noro 3024: }
3025:
1.56 noro 3026: /* product of monomials */
3027: /* XXX block order is not handled correctly */
3028:
1.55 noro 3029: void weyl_mul_nm_nmv(int n,int mod,NM m0,NMV m1,NM *tab,int tlen)
3030: {
1.56 noro 3031: int i,n2,j,s,curlen,homo,h,a,b,k,l,u,min;
1.61 noro 3032: UINT *d0,*d1,*d,*dt,*ctab;
1.56 noro 3033: Q *ctab_q;
3034: Q q,q1;
1.61 noro 3035: UINT c0,c1,c;
1.55 noro 3036: NM *p;
3037: NM m,t;
3038:
3039: for ( i = 0; i < tlen; i++ ) tab[i] = 0;
3040: if ( !m0 || !m1 ) return;
3041: d0 = DL(m0); d1 = DL(m1); n2 = n>>1;
3042: NEWNM(m); d = DL(m);
1.56 noro 3043: if ( mod ) {
3044: c0 = CM(m0); c1 = CM(m1); DMAR(c0,c1,0,mod,c); CM(m) = c;
3045: } else
3046: mulq(CQ(m0),CQ(m1),&CQ(m));
1.55 noro 3047: for ( i = 0; i < nd_wpd; i++ ) d[i] = 0;
3048: homo = n&1 ? 1 : 0;
3049: if ( homo ) {
3050: /* offset of h-degree */
3051: h = GET_EXP(d0,n-1)+GET_EXP(d1,n-1);
3052: PUT_EXP(DL(m),n-1,h);
3053: TD(DL(m)) = h;
1.61 noro 3054: if ( nd_blockmask ) ndl_weight_mask(DL(m));
1.55 noro 3055: }
3056: tab[0] = m;
3057: NEWNM(m); d = DL(m);
1.57 noro 3058: for ( i = 0, curlen = 1; i < n2; i++ ) {
1.55 noro 3059: a = GET_EXP(d0,i); b = GET_EXP(d1,n2+i);
3060: k = GET_EXP(d0,n2+i); l = GET_EXP(d1,i);
3061: /* xi^a*(Di^k*xi^l)*Di^b */
3062: a += l; b += k;
1.56 noro 3063: s = MUL_WEIGHT(a,i)+MUL_WEIGHT(b,n2+i);
1.55 noro 3064: if ( !k || !l ) {
3065: for ( j = 0; j < curlen; j++ )
1.56 noro 3066: if ( t = tab[j] ) {
3067: dt = DL(t);
3068: PUT_EXP(dt,i,a); PUT_EXP(dt,n2+i,b); TD(dt) += s;
1.61 noro 3069: if ( nd_blockmask ) ndl_weight_mask(dt);
1.55 noro 3070: }
3071: curlen *= k+1;
3072: continue;
3073: }
3074: min = MIN(k,l);
1.56 noro 3075: if ( mod ) {
1.61 noro 3076: ctab = (UINT *)ALLOCA((min+1)*sizeof(UINT));
1.56 noro 3077: mkwcm(k,l,mod,ctab);
3078: } else {
3079: ctab_q = (Q *)ALLOCA((min+1)*sizeof(Q));
3080: mkwc(k,l,ctab_q);
3081: }
1.57 noro 3082: for ( j = min; j >= 0; j-- ) {
1.56 noro 3083: for ( u = 0; u < nd_wpd; u++ ) d[u] = 0;
1.55 noro 3084: PUT_EXP(d,i,a-j); PUT_EXP(d,n2+i,b-j);
1.56 noro 3085: h = MUL_WEIGHT(a-j,i)+MUL_WEIGHT(b-j,n2+i);
1.55 noro 3086: if ( homo ) {
3087: TD(d) = s;
1.56 noro 3088: PUT_EXP(d,n-1,s-h);
1.55 noro 3089: } else TD(d) = h;
1.61 noro 3090: if ( nd_blockmask ) ndl_weight_mask(d);
1.56 noro 3091: if ( mod ) c = ctab[j];
3092: else q = ctab_q[j];
1.57 noro 3093: p = tab+curlen*j;
3094: if ( j == 0 ) {
3095: for ( u = 0; u < curlen; u++, p++ ) {
3096: if ( tab[u] ) {
3097: ndl_addto(DL(tab[u]),d);
3098: if ( mod ) {
3099: c0 = CM(tab[u]); DMAR(c0,c,0,mod,c1); CM(tab[u]) = c1;
3100: } else {
3101: mulq(CQ(tab[u]),q,&q1); CQ(tab[u]) = q1;
3102: }
3103: }
1.56 noro 3104: }
1.57 noro 3105: } else {
3106: for ( u = 0; u < curlen; u++, p++ ) {
3107: if ( tab[u] ) {
3108: NEWNM(t);
3109: ndl_add(DL(tab[u]),d,DL(t));
3110: if ( mod ) {
3111: c0 = CM(tab[u]); DMAR(c0,c,0,mod,c1); CM(t) = c1;
3112: } else
3113: mulq(CQ(tab[u]),q,&CQ(t));
3114: *p = t;
3115: }
1.55 noro 3116: }
3117: }
3118: }
3119: curlen *= k+1;
3120: }
3121: FREENM(m);
3122: }
3123:
1.63 noro 3124: ND ndv_mul_nm_symbolic(NM m0,NDV p)
3125: {
3126: NM mr,mr0;
3127: NMV m;
3128: UINT *d,*dt,*dm;
3129: int c,n,td,i,c1,c2,len;
3130: Q q;
3131: ND r;
3132:
3133: if ( !p ) return 0;
3134: else {
3135: n = NV(p); m = BDY(p);
3136: d = DL(m0);
3137: len = LEN(p);
3138: mr0 = 0;
3139: td = TD(d);
3140: c = CM(m0);
3141: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
3142: NEXTNM(mr0,mr);
3143: CM(mr) = 1;
3144: ndl_add(DL(m),d,DL(mr));
3145: }
3146: NEXT(mr) = 0;
3147: MKND(NV(p),mr0,len,r);
3148: SG(r) = SG(p) + TD(d);
3149: return r;
3150: }
3151: }
3152:
1.55 noro 3153: ND ndv_mul_nm(int mod,NM m0,NDV p)
1.9 noro 3154: {
3155: NM mr,mr0;
3156: NMV m;
1.61 noro 3157: UINT *d,*dt,*dm;
1.9 noro 3158: int c,n,td,i,c1,c2,len;
1.16 noro 3159: Q q;
1.9 noro 3160: ND r;
3161:
1.34 noro 3162: if ( !p ) return 0;
1.55 noro 3163: else if ( do_weyl )
1.71 noro 3164: if ( mod == -1 )
3165: error("ndv_mul_nm : not implemented (weyl)");
3166: else
3167: return weyl_ndv_mul_nm(mod,m0,p);
1.9 noro 3168: else {
3169: n = NV(p); m = BDY(p);
1.34 noro 3170: d = DL(m0);
1.14 noro 3171: len = LEN(p);
1.9 noro 3172: mr0 = 0;
1.34 noro 3173: td = TD(d);
1.71 noro 3174: if ( mod == -1 ) {
3175: c = CM(m0);
3176: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
3177: NEXTNM(mr0,mr);
3178: CM(mr) = _mulsf(CM(m),c);
3179: ndl_add(DL(m),d,DL(mr));
3180: }
3181: } else if ( mod ) {
1.16 noro 3182: c = CM(m0);
3183: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
3184: NEXTNM(mr0,mr);
3185: c1 = CM(m);
1.19 noro 3186: DMAR(c1,c,0,mod,c2);
1.16 noro 3187: CM(mr) = c2;
3188: ndl_add(DL(m),d,DL(mr));
3189: }
3190: } else {
3191: q = CQ(m0);
3192: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
3193: NEXTNM(mr0,mr);
3194: mulq(CQ(m),q,&CQ(mr));
3195: ndl_add(DL(m),d,DL(mr));
3196: }
1.4 noro 3197: }
1.9 noro 3198: NEXT(mr) = 0;
1.31 noro 3199: MKND(NV(p),mr0,len,r);
1.34 noro 3200: SG(r) = SG(p) + TD(d);
1.9 noro 3201: return r;
1.4 noro 3202: }
3203: }
3204:
1.104 ! noro 3205: ND nd_quo(int mod,PGeoBucket bucket,NDV d)
1.99 noro 3206: {
3207: NM mq0,mq;
1.102 noro 3208: NMV tm;
1.99 noro 3209: Q q;
1.104 ! noro 3210: int i,nv,sg,c,c1,c2,hindex;
! 3211: ND p,t,r;
! 3212: N tnm;
! 3213:
1.99 noro 3214: if ( !p ) return 0;
3215: else {
1.104 ! noro 3216: nv = NV(d);
1.99 noro 3217: mq0 = 0;
1.102 noro 3218: tm = (NMV)ALLOCA(nmv_adv);
1.104 ! noro 3219: while ( 1 ) {
! 3220: hindex = mod?head_pbucket(mod,bucket):head_pbucket_q(bucket);
! 3221: if ( hindex < 0 ) break;
! 3222: p = bucket->body[hindex];
1.99 noro 3223: NEXTNM(mq0,mq);
3224: ndl_sub(HDL(p),HDL(d),DL(mq));
1.102 noro 3225: ndl_copy(DL(mq),DL(tm));
3226: if ( mod ) {
3227: c1 = invm(HCM(d),mod); c2 = HCM(p);
3228: DMAR(c1,c2,0,mod,c); CM(mq) = c;
3229: CM(tm) = mod-c;
3230: } else {
1.104 ! noro 3231: divsn(NM(HCQ(p)),NM(HCQ(d)),&tnm);
! 3232: NTOQ(tnm,SGN(HCQ(p))*SGN(HCQ(d)),CQ(mq));
1.102 noro 3233: chsgnq(CQ(mq),&CQ(tm));
3234: }
3235: t = ndv_mul_nmv_trunc(mod,tm,d,HDL(d));
1.104 ! noro 3236: bucket->body[hindex] = nd_remove_head(p);
! 3237: t = nd_remove_head(t);
! 3238: add_pbucket(mod,bucket,t);
! 3239: }
! 3240: if ( !mq0 )
! 3241: r = 0;
! 3242: else {
! 3243: NEXT(mq) = 0;
! 3244: for ( i = 0, mq = mq0; mq; mq = NEXT(mq), i++ );
! 3245: MKND(nv,mq0,i,r);
! 3246: /* XXX */
! 3247: SG(r) = HTD(r);
1.99 noro 3248: }
3249: return r;
3250: }
3251: }
3252:
1.43 noro 3253: void ndv_realloc(NDV p,int obpe,int oadv,EPOS oepos)
1.11 noro 3254: {
1.13 noro 3255: NMV m,mr,mr0,t;
3256: int len,i,k;
1.11 noro 3257:
1.61 noro 3258: if ( !p ) return;
3259: m = BDY(p); len = LEN(p);
3260: mr0 = nmv_adv>oadv?(NMV)REALLOC(BDY(p),len*nmv_adv):BDY(p);
3261: m = (NMV)((char *)mr0+(len-1)*oadv);
3262: mr = (NMV)((char *)mr0+(len-1)*nmv_adv);
3263: t = (NMV)ALLOCA(nmv_adv);
3264: for ( i = 0; i < len; i++, NMV_OPREV(m), NMV_PREV(mr) ) {
3265: CQ(t) = CQ(m);
3266: for ( k = 0; k < nd_wpd; k++ ) DL(t)[k] = 0;
3267: ndl_reconstruct(DL(m),DL(t),obpe,oepos);
3268: CQ(mr) = CQ(t);
3269: ndl_copy(DL(t),DL(mr));
3270: }
3271: BDY(p) = mr0;
3272: }
3273:
3274: NDV ndv_dup_realloc(NDV p,int obpe,int oadv,EPOS oepos)
3275: {
3276: NMV m,mr,mr0;
3277: int len,i;
3278: NDV r;
1.11 noro 3279:
1.61 noro 3280: if ( !p ) return 0;
3281: m = BDY(p); len = LEN(p);
3282: mr0 = mr = (NMV)MALLOC(len*nmv_adv);
3283: for ( i = 0; i < len; i++, NMV_OADV(m), NMV_ADV(mr) ) {
3284: ndl_zero(DL(mr));
3285: ndl_reconstruct(DL(m),DL(mr),obpe,oepos);
3286: CQ(mr) = CQ(m);
1.11 noro 3287: }
1.61 noro 3288: MKNDV(NV(p),mr0,len,r);
3289: SG(r) = SG(p);
3290: return r;
1.11 noro 3291: }
3292:
1.61 noro 3293: /* duplicate p */
3294:
3295: NDV ndv_dup(int mod,NDV p)
1.3 noro 3296: {
3297: NDV d;
1.61 noro 3298: NMV t,m,m0;
1.3 noro 3299: int i,len;
3300:
1.34 noro 3301: if ( !p ) return 0;
1.31 noro 3302: len = LEN(p);
1.34 noro 3303: m0 = m = (NMV)(mod?MALLOC_ATOMIC(len*nmv_adv):MALLOC(len*nmv_adv));
1.61 noro 3304: for ( t = BDY(p), i = 0; i < len; i++, NMV_ADV(t), NMV_ADV(m) ) {
1.14 noro 3305: ndl_copy(DL(t),DL(m));
1.16 noro 3306: CQ(m) = CQ(t);
1.3 noro 3307: }
3308: MKNDV(NV(p),m0,len,d);
1.23 noro 3309: SG(d) = SG(p);
3310: return d;
3311: }
3312:
1.63 noro 3313: ND nd_dup(ND p)
3314: {
3315: ND d;
3316: NM t,m,m0;
3317:
3318: if ( !p ) return 0;
3319: for ( m0 = 0, t = BDY(p); t; t = NEXT(t) ) {
3320: NEXTNM(m0,m);
3321: ndl_copy(DL(t),DL(m));
3322: CQ(m) = CQ(t);
3323: }
3324: if ( m0 ) NEXT(m) = 0;
3325: MKND(NV(p),m0,LEN(p),d);
3326: SG(d) = SG(p);
3327: return d;
3328: }
3329:
1.61 noro 3330: /* XXX if p->len == 0 then it represents 0 */
3331:
3332: void ndv_mod(int mod,NDV p)
3333: {
3334: NMV t,d;
3335: int r;
3336: int i,len,dlen;
1.71 noro 3337: Obj gfs;
1.61 noro 3338:
3339: if ( !p ) return;
3340: len = LEN(p);
3341: dlen = 0;
1.71 noro 3342: if ( mod == -1 )
3343: for ( t = d = BDY(p), i = 0; i < len; i++, NMV_ADV(t) ) {
3344: simp_ff((Obj)CP(t),&gfs);
3345: r = FTOIF(CONT((GFS)gfs));
1.61 noro 3346: CM(d) = r;
3347: ndl_copy(DL(t),DL(d));
3348: NMV_ADV(d);
3349: dlen++;
3350: }
1.71 noro 3351: else
3352: for ( t = d = BDY(p), i = 0; i < len; i++, NMV_ADV(t) ) {
3353: r = rem(NM(CQ(t)),mod);
3354: if ( r ) {
3355: if ( SGN(CQ(t)) < 0 )
3356: r = mod-r;
3357: CM(d) = r;
3358: ndl_copy(DL(t),DL(d));
3359: NMV_ADV(d);
3360: dlen++;
3361: }
3362: }
1.61 noro 3363: LEN(p) = dlen;
3364: }
3365:
3366: NDV ptondv(VL vl,VL dvl,P p)
3367: {
3368: ND nd;
3369:
3370: nd = ptond(vl,dvl,p);
3371: return ndtondv(0,nd);
3372: }
3373:
3374: ND ptond(VL vl,VL dvl,P p)
1.23 noro 3375: {
1.61 noro 3376: int n,i,j,k,e;
3377: VL tvl;
3378: V v;
3379: DCP dc;
3380: DCP *w;
3381: ND r,s,t,u;
3382: P x;
3383: int c;
3384: UINT *d;
1.23 noro 3385: NM m,m0;
1.61 noro 3386:
3387: if ( !p )
3388: return 0;
3389: else if ( NUM(p) ) {
3390: NEWNM(m);
3391: ndl_zero(DL(m));
3392: CQ(m) = (Q)p;
3393: NEXT(m) = 0;
3394: MKND(nd_nvar,m,1,r);
3395: SG(r) = 0;
3396: return r;
3397: } else {
3398: for ( dc = DC(p), k = 0; dc; dc = NEXT(dc), k++ );
3399: w = (DCP *)ALLOCA(k*sizeof(DCP));
3400: for ( dc = DC(p), j = 0; j < k; dc = NEXT(dc), j++ ) w[j] = dc;
3401: for ( i = 0, tvl = dvl, v = VR(p);
1.98 noro 3402: tvl && tvl->v != v; tvl = NEXT(tvl), i++ );
1.61 noro 3403: if ( !tvl ) {
3404: for ( j = k-1, s = 0, MKV(v,x); j >= 0; j-- ) {
3405: t = ptond(vl,dvl,COEF(w[j]));
3406: pwrp(vl,x,DEG(w[j]),&p);
3407: nd_mul_c_p(CO,t,p); s = nd_add(0,s,t);
3408: }
3409: return s;
3410: } else {
3411: NEWNM(m0); d = DL(m0);
3412: for ( j = k-1, s = 0; j >= 0; j-- ) {
3413: ndl_zero(d); e = QTOS(DEG(w[j])); PUT_EXP(d,i,e);
3414: TD(d) = MUL_WEIGHT(e,i);
3415: if ( nd_blockmask) ndl_weight_mask(d);
3416: t = ptond(vl,dvl,COEF(w[j]));
3417: for ( m = BDY(t); m; m = NEXT(m) )
3418: ndl_addto(DL(m),d);
3419: SG(t) += TD(d);
3420: s = nd_add(0,s,t);
3421: }
3422: FREENM(m0);
3423: return s;
3424: }
3425: }
3426: }
3427:
3428: P ndvtop(int mod,VL vl,VL dvl,NDV p)
3429: {
3430: VL tvl;
3431: int len,n,j,i,e;
3432: NMV m;
3433: Q q;
3434: P c;
3435: UINT *d;
3436: P s,r,u,t,w;
1.71 noro 3437: GFS gfs;
1.23 noro 3438:
1.34 noro 3439: if ( !p ) return 0;
1.61 noro 3440: else {
3441: len = LEN(p);
3442: n = NV(p);
3443: m = (NMV)(((char *)BDY(p))+nmv_adv*(len-1));
3444: for ( j = len-1, s = 0; j >= 0; j--, NMV_PREV(m) ) {
1.71 noro 3445: if ( mod == -1 ) {
3446: e = IFTOF(CM(m)); MKGFS(e,gfs); c = (P)gfs;
3447: } else if ( mod ) {
1.61 noro 3448: STOQ(CM(m),q); c = (P)q;
3449: } else
3450: c = CP(m);
3451: d = DL(m);
3452: for ( i = 0, t = c, tvl = dvl; i < n; tvl = NEXT(tvl), i++ ) {
3453: MKV(tvl->v,r); e = GET_EXP(d,i); STOQ(e,q);
3454: pwrp(vl,r,q,&u); mulp(vl,t,u,&w); t = w;
3455: }
3456: addp(vl,s,t,&u); s = u;
3457: }
3458: return s;
1.23 noro 3459: }
1.3 noro 3460: }
3461:
1.61 noro 3462: NDV ndtondv(int mod,ND p)
1.11 noro 3463: {
3464: NDV d;
1.61 noro 3465: NMV m,m0;
3466: NM t;
3467: int i,len;
1.11 noro 3468:
1.34 noro 3469: if ( !p ) return 0;
1.61 noro 3470: len = LEN(p);
1.74 noro 3471: if ( mod )
3472: m0 = m = (NMV)GC_malloc_atomic_ignore_off_page(len*nmv_adv);
3473: else
3474: m0 = m = MALLOC(len*nmv_adv);
1.103 noro 3475: #if 0
1.74 noro 3476: ndv_alloc += nmv_adv*len;
1.103 noro 3477: #endif
1.61 noro 3478: for ( t = BDY(p), i = 0; t; t = NEXT(t), i++, NMV_ADV(m) ) {
3479: ndl_copy(DL(t),DL(m));
3480: CQ(m) = CQ(t);
1.11 noro 3481: }
1.61 noro 3482: MKNDV(NV(p),m0,len,d);
1.14 noro 3483: SG(d) = SG(p);
1.11 noro 3484: return d;
3485: }
3486:
1.61 noro 3487: ND ndvtond(int mod,NDV p)
1.11 noro 3488: {
1.61 noro 3489: ND d;
3490: NM m,m0;
1.11 noro 3491: NMV t;
1.61 noro 3492: int i,len;
1.11 noro 3493:
1.34 noro 3494: if ( !p ) return 0;
1.11 noro 3495: m0 = 0;
1.61 noro 3496: len = p->len;
3497: for ( t = BDY(p), i = 0; i < len; NMV_ADV(t), i++ ) {
3498: NEXTNM(m0,m);
3499: ndl_copy(DL(t),DL(m));
3500: CQ(m) = CQ(t);
1.11 noro 3501: }
3502: NEXT(m) = 0;
1.61 noro 3503: MKND(NV(p),m0,len,d);
1.14 noro 3504: SG(d) = SG(p);
1.11 noro 3505: return d;
3506: }
3507:
1.3 noro 3508: void ndv_print(NDV p)
3509: {
3510: NMV m;
3511: int i,len;
3512:
1.34 noro 3513: if ( !p ) printf("0\n");
1.3 noro 3514: else {
1.14 noro 3515: len = LEN(p);
1.3 noro 3516: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
1.71 noro 3517: if ( CM(m) & 0x80000000 ) printf("+@_%d*",IFTOF(CM(m)));
3518: else printf("+%d*",CM(m));
1.16 noro 3519: ndl_print(DL(m));
3520: }
3521: printf("\n");
3522: }
3523: }
3524:
3525: void ndv_print_q(NDV p)
3526: {
3527: NMV m;
3528: int i,len;
3529:
1.34 noro 3530: if ( !p ) printf("0\n");
1.16 noro 3531: else {
3532: len = LEN(p);
3533: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
3534: printf("+");
1.74 noro 3535: printexpr(CO,(Obj)CQ(m));
1.16 noro 3536: printf("*");
1.14 noro 3537: ndl_print(DL(m));
1.3 noro 3538: }
3539: printf("\n");
3540: }
1.25 noro 3541: }
3542:
1.61 noro 3543: NODE ndv_reducebase(NODE x)
1.27 noro 3544: {
3545: int len,i,j;
3546: NDV *w;
3547: NODE t,t0;
3548:
3549: len = length(x);
3550: w = (NDV *)ALLOCA(len*sizeof(NDV));
3551: for ( i = 0, t = x; i < len; i++, t = NEXT(t) ) w[i] = BDY(t);
3552: for ( i = 0; i < len; i++ ) {
3553: for ( j = 0; j < i; j++ ) {
3554: if ( w[i] && w[j] )
3555: if ( ndl_reducible(HDL(w[i]),HDL(w[j])) ) w[i] = 0;
3556: else if ( ndl_reducible(HDL(w[j]),HDL(w[i])) ) w[j] = 0;
3557: }
3558: }
3559: for ( i = len-1, t0 = 0; i >= 0; i-- ) {
3560: if ( w[i] ) { NEXTNODE(t0,t); BDY(t) = (pointer)w[i]; }
3561: }
3562: NEXT(t) = 0; x = t0;
3563: return x;
1.11 noro 3564: }
1.32 noro 3565:
1.43 noro 3566: /* XXX incomplete */
3567:
1.32 noro 3568: void nd_init_ord(struct order_spec *ord)
3569: {
1.43 noro 3570: switch ( ord->id ) {
1.32 noro 3571: case 0:
1.43 noro 3572: switch ( ord->ord.simple ) {
3573: case 0:
3574: nd_dcomp = 1;
3575: nd_isrlex = 1;
3576: break;
3577: case 1:
3578: nd_dcomp = 1;
3579: nd_isrlex = 0;
3580: break;
3581: case 2:
3582: nd_dcomp = 0;
3583: nd_isrlex = 0;
1.45 noro 3584: ndl_compare_function = ndl_lex_compare;
1.58 noro 3585: break;
3586: case 11:
3587: /* XXX */
3588: nd_dcomp = 0;
3589: nd_isrlex = 1;
3590: ndl_compare_function = ndl_ww_lex_compare;
1.43 noro 3591: break;
3592: default:
3593: error("nd_gr : unsupported order");
3594: }
1.32 noro 3595: break;
3596: case 1:
1.96 noro 3597: /* block order */
1.43 noro 3598: /* XXX */
3599: nd_dcomp = -1;
1.32 noro 3600: nd_isrlex = 0;
1.45 noro 3601: ndl_compare_function = ndl_block_compare;
1.34 noro 3602: break;
1.43 noro 3603: case 2:
1.96 noro 3604: /* matrix order */
3605: /* XXX */
3606: nd_dcomp = -1;
3607: nd_isrlex = 0;
3608: nd_matrix_len = ord->ord.matrix.row;
3609: nd_matrix = ord->ord.matrix.matrix;
3610: ndl_compare_function = ndl_matrix_compare;
3611: break;
3612: case 3:
1.97 noro 3613: /* composite order */
3614: nd_dcomp = -1;
3615: nd_isrlex = 0;
3616: nd_worb_len = ord->ord.composite.length;
3617: nd_worb = ord->ord.composite.w_or_b;
3618: ndl_compare_function = ndl_composite_compare;
1.32 noro 3619: break;
3620: }
1.41 noro 3621: nd_ord = ord;
1.32 noro 3622: }
3623:
1.43 noro 3624: BlockMask nd_create_blockmask(struct order_spec *ord)
3625: {
3626: int n,i,j,s,l;
1.61 noro 3627: UINT *t;
1.43 noro 3628: BlockMask bm;
3629:
1.96 noro 3630: /* we only create mask table for block order */
3631: if ( ord->id != 1 )
1.43 noro 3632: return 0;
3633: n = ord->ord.block.length;
3634: bm = (BlockMask)MALLOC(sizeof(struct oBlockMask));
3635: bm->n = n;
3636: bm->order_pair = ord->ord.block.order_pair;
1.61 noro 3637: bm->mask = (UINT **)MALLOC(n*sizeof(UINT *));
1.43 noro 3638: for ( i = 0, s = 0; i < n; i++ ) {
1.61 noro 3639: bm->mask[i] = t = (UINT *)MALLOC_ATOMIC(nd_wpd*sizeof(UINT));
1.43 noro 3640: for ( j = 0; j < nd_wpd; j++ ) t[j] = 0;
3641: l = bm->order_pair[i].length;
3642: for ( j = 0; j < l; j++, s++ ) PUT_EXP(t,s,nd_mask0);
3643: }
3644: return bm;
1.57 noro 3645: }
3646:
3647: EPOS nd_create_epos(struct order_spec *ord)
3648: {
3649: int i,j,l,s,ord_l,ord_o;
3650: EPOS epos;
3651: struct order_pair *op;
3652:
3653: epos = (EPOS)MALLOC_ATOMIC(nd_nvar*sizeof(struct oEPOS));
3654: switch ( ord->id ) {
3655: case 0:
3656: if ( nd_isrlex ) {
3657: for ( i = 0; i < nd_nvar; i++ ) {
3658: epos[i].i = nd_exporigin + (nd_nvar-1-i)/nd_epw;
3659: epos[i].s = (nd_epw-((nd_nvar-1-i)%nd_epw)-1)*nd_bpe;
3660: }
3661: } else {
3662: for ( i = 0; i < nd_nvar; i++ ) {
3663: epos[i].i = nd_exporigin + i/nd_epw;
3664: epos[i].s = (nd_epw-(i%nd_epw)-1)*nd_bpe;
3665: }
3666: }
3667: break;
3668: case 1:
3669: /* block order */
3670: l = ord->ord.block.length;
3671: op = ord->ord.block.order_pair;
3672: for ( j = 0, s = 0; j < l; j++ ) {
3673: ord_o = op[j].order;
3674: ord_l = op[j].length;
3675: if ( !ord_o )
3676: for ( i = 0; i < ord_l; i++ ) {
3677: epos[s+i].i = nd_exporigin + (s+ord_l-i-1)/nd_epw;
3678: epos[s+i].s = (nd_epw-((s+ord_l-i-1)%nd_epw)-1)*nd_bpe;
3679: }
3680: else
3681: for ( i = 0; i < ord_l; i++ ) {
3682: epos[s+i].i = nd_exporigin + (s+i)/nd_epw;
3683: epos[s+i].s = (nd_epw-((s+i)%nd_epw)-1)*nd_bpe;
3684: }
3685: s += ord_l;
3686: }
3687: break;
3688: case 2:
1.96 noro 3689: /* matrix order */
3690: case 3:
3691: /* composite order */
3692: for ( i = 0; i < nd_nvar; i++ ) {
3693: epos[i].i = nd_exporigin + i/nd_epw;
3694: epos[i].s = (nd_epw-(i%nd_epw)-1)*nd_bpe;
3695: }
3696: break;
1.57 noro 3697: }
3698: return epos;
1.43 noro 3699: }
1.59 noro 3700:
3701: /* external interface */
3702:
3703: void nd_nf_p(P f,LIST g,LIST v,int m,struct order_spec *ord,P *rp)
3704: {
1.61 noro 3705: NODE t,in0,in;
1.59 noro 3706: ND nd,nf;
1.61 noro 3707: NDV ndv;
3708: VL vv,tv;
3709: int stat,nvar,max,e;
1.69 noro 3710: union oNDC dn;
1.59 noro 3711:
3712: pltovl(v,&vv);
1.74 noro 3713: for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++ );
1.61 noro 3714:
3715: /* get the degree bound */
3716: for ( t = BDY(g), max = 0; t; t = NEXT(t) )
3717: for ( tv = vv; tv; tv = NEXT(tv) ) {
3718: e = getdeg(tv->v,(P)BDY(t));
3719: max = MAX(e,max);
3720: }
3721: for ( tv = vv; tv; tv = NEXT(tv) ) {
3722: e = getdeg(tv->v,f);
3723: max = MAX(e,max);
3724: }
3725:
1.59 noro 3726: nd_init_ord(ord);
1.61 noro 3727: nd_setup_parameters(nvar,max);
3728:
3729: /* conversion to ndv */
3730: for ( in0 = 0, t = BDY(g); t; t = NEXT(t) ) {
3731: NEXTNODE(in0,in);
3732: BDY(in) = (pointer)ptondv(CO,vv,(P)BDY(t));
1.69 noro 3733: if ( m ) ndv_mod(m,(NDV)BDY(in));
1.61 noro 3734: }
3735: NEXTNODE(in0,in);
3736: BDY(in) = (pointer)ptondv(CO,vv,f);
1.69 noro 3737: if ( m ) ndv_mod(m,(NDV)BDY(in));
1.61 noro 3738: NEXT(in) = 0;
3739:
3740: ndv_setup(m,0,in0);
1.59 noro 3741: nd_psn--;
3742: nd_scale=2;
3743: while ( 1 ) {
3744: nd = (pointer)ndvtond(m,nd_ps[nd_psn]);
1.69 noro 3745: stat = nd_nf(m,nd,nd_ps,1,0,&nf);
1.59 noro 3746: if ( !stat ) {
3747: nd_psn++;
1.103 noro 3748: nd_reconstruct(0,0);
1.59 noro 3749: nd_psn--;
3750: } else
3751: break;
3752: }
1.61 noro 3753: *rp = ndvtop(m,CO,vv,ndtondv(m,nf));
1.63 noro 3754: }
3755:
3756: int nd_to_vect(int mod,UINT *s0,int n,ND d,UINT *r)
3757: {
3758: NM m;
3759: UINT *t,*s;
3760: int i;
3761:
3762: for ( i = 0; i < n; i++ ) r[i] = 0;
3763: for ( i = 0, s = s0, m = BDY(d); m; m = NEXT(m) ) {
3764: t = DL(m);
3765: for ( ; !ndl_equal(t,s); s += nd_wpd, i++ );
3766: r[i] = CM(m);
3767: }
3768: for ( i = 0; !r[i]; i++ );
3769: return i;
3770: }
3771:
1.74 noro 3772: int ndv_to_vect(int mod,UINT *s0,int n,NDV d,UINT *r)
3773: {
3774: NMV m;
3775: UINT *t,*s;
3776: int i,j,len;
3777:
3778: for ( i = 0; i < n; i++ ) r[i] = 0;
3779: m = BDY(d);
3780: len = LEN(d);
3781: for ( i = j = 0, s = s0; j < len; j++, NMV_ADV(m)) {
3782: t = DL(m);
3783: for ( ; !ndl_equal(t,s); s += nd_wpd, i++ );
3784: r[i] = CM(m);
3785: }
3786: for ( i = 0; !r[i]; i++ );
3787: return i;
3788: }
3789:
1.63 noro 3790: int nm_ind_pair_to_vect(int mod,UINT *s0,int n,NM_ind_pair pair,UINT *r)
3791: {
3792: NM m;
3793: NMV mr;
3794: UINT *d,*t,*s;
3795: NDV p;
3796: int i,j,len;
3797:
3798: m = pair->mul;
3799: d = DL(m);
3800: p = nd_ps[pair->index];
3801: t = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
3802: for ( i = 0; i < n; i++ ) r[i] = 0;
3803: len = LEN(p);
3804: for ( i = j = 0, s = s0, mr = BDY(p); j < len; j++, NMV_ADV(mr) ) {
3805: ndl_add(d,DL(mr),t);
3806: for ( ; !ndl_equal(t,s); s += nd_wpd, i++ );
3807: r[i] = CM(mr);
3808: }
3809: for ( i = 0; !r[i]; i++ );
3810: return i;
3811: }
3812:
1.67 noro 3813: IndArray nm_ind_pair_to_vect_compress(int mod,UINT *s0,int n,NM_ind_pair pair)
1.64 noro 3814: {
3815: NM m;
3816: NMV mr;
3817: UINT *d,*t,*s;
3818: NDV p;
1.67 noro 3819: unsigned char *ivc;
3820: unsigned short *ivs;
1.81 noro 3821: UINT *v,*ivi,*s0v;
1.67 noro 3822: int i,j,len,prev,diff,cdiff;
3823: IndArray r;
1.64 noro 3824:
3825: m = pair->mul;
3826: d = DL(m);
3827: p = nd_ps[pair->index];
3828: len = LEN(p);
3829: t = (UINT *)ALLOCA(nd_wpd*sizeof(UINT));
1.67 noro 3830: v = (unsigned int *)ALLOCA(len*sizeof(unsigned int));
1.64 noro 3831: for ( i = j = 0, s = s0, mr = BDY(p); j < len; j++, NMV_ADV(mr) ) {
3832: ndl_add(d,DL(mr),t);
3833: for ( ; !ndl_equal(t,s); s += nd_wpd, i++ );
1.67 noro 3834: v[j] = i;
3835: }
1.81 noro 3836: r = (IndArray)MALLOC(sizeof(struct oIndArray));
1.67 noro 3837: r->head = v[0];
3838: diff = 0;
3839: for ( i = 1; i < len; i++ ) {
3840: cdiff = v[i]-v[i-1]; diff = MAX(cdiff,diff);
3841: }
3842: if ( diff < 256 ) {
3843: r->width = 1;
3844: ivc = (unsigned char *)MALLOC_ATOMIC(len*sizeof(unsigned char));
3845: r->index.c = ivc;
3846: for ( i = 1, ivc[0] = 0; i < len; i++ ) ivc[i] = v[i]-v[i-1];
3847: } else if ( diff < 65536 ) {
3848: r->width = 2;
3849: ivs = (unsigned short *)MALLOC_ATOMIC(len*sizeof(unsigned short));
3850: r->index.s = ivs;
3851: for ( i = 1, ivs[0] = 0; i < len; i++ ) ivs[i] = v[i]-v[i-1];
3852: } else {
3853: r->width = 4;
3854: ivi = (unsigned int *)MALLOC_ATOMIC(len*sizeof(unsigned int));
3855: r->index.i = ivi;
3856: for ( i = 1, ivi[0] = 0; i < len; i++ ) ivi[i] = v[i]-v[i-1];
1.64 noro 3857: }
1.67 noro 3858: return r;
1.64 noro 3859: }
3860:
1.65 noro 3861:
1.76 noro 3862: int ndv_reduce_vect(int m,UINT *svect,int col,IndArray *imat,NM_ind_pair *rp0,int nred)
1.65 noro 3863: {
1.67 noro 3864: int i,j,k,len,pos,prev;
1.66 noro 3865: UINT c,c1,c2,c3,up,lo,dmy;
1.67 noro 3866: IndArray ivect;
3867: unsigned char *ivc;
3868: unsigned short *ivs;
3869: unsigned int *ivi;
1.65 noro 3870: NDV redv;
1.67 noro 3871: NMV mr;
1.65 noro 3872: NODE rp;
1.76 noro 3873: int maxrs;
1.65 noro 3874:
1.76 noro 3875: maxrs = 0;
1.74 noro 3876: for ( i = 0; i < nred; i++ ) {
1.65 noro 3877: ivect = imat[i];
1.67 noro 3878: k = ivect->head; svect[k] %= m;
1.65 noro 3879: if ( c = svect[k] ) {
1.76 noro 3880: maxrs = MAX(maxrs,rp0[i]->sugar);
1.74 noro 3881: c = m-c; redv = nd_ps[rp0[i]->index];
1.67 noro 3882: len = LEN(redv); mr = BDY(redv);
3883: svect[k] = 0; prev = k;
3884: switch ( ivect->width ) {
3885: case 1:
3886: ivc = ivect->index.c;
3887: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3888: pos = prev+ivc[j]; c1 = CM(mr); c2 = svect[pos];
3889: prev = pos;
3890: DMA(c1,c,c2,up,lo);
3891: if ( up ) { DSAB(m,up,lo,dmy,c3); svect[pos] = c3;
3892: } else svect[pos] = lo;
3893: }
3894: break;
3895: case 2:
3896: ivs = ivect->index.s;
3897: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3898: pos = prev+ivs[j]; c1 = CM(mr); c2 = svect[pos];
3899: prev = pos;
3900: DMA(c1,c,c2,up,lo);
3901: if ( up ) { DSAB(m,up,lo,dmy,c3); svect[pos] = c3;
3902: } else svect[pos] = lo;
3903: }
3904: break;
3905: case 4:
3906: ivi = ivect->index.i;
3907: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3908: pos = prev+ivi[j]; c1 = CM(mr); c2 = svect[pos];
3909: prev = pos;
3910: DMA(c1,c,c2,up,lo);
3911: if ( up ) { DSAB(m,up,lo,dmy,c3); svect[pos] = c3;
3912: } else svect[pos] = lo;
3913: }
3914: break;
1.65 noro 3915: }
3916: }
3917: }
1.66 noro 3918: for ( i = 0; i < col; i++ )
3919: if ( svect[i] >= (UINT)m ) svect[i] %= m;
1.76 noro 3920: return maxrs;
1.65 noro 3921: }
3922:
1.76 noro 3923: int ndv_reduce_vect_sf(int m,UINT *svect,int col,IndArray *imat,NM_ind_pair *rp0,int nred)
1.72 noro 3924: {
3925: int i,j,k,len,pos,prev;
3926: UINT c,c1,c2,c3,up,lo,dmy;
3927: IndArray ivect;
3928: unsigned char *ivc;
3929: unsigned short *ivs;
3930: unsigned int *ivi;
3931: NDV redv;
3932: NMV mr;
3933: NODE rp;
1.76 noro 3934: int maxrs;
1.72 noro 3935:
1.76 noro 3936: maxrs = 0;
1.74 noro 3937: for ( i = 0; i < nred; i++ ) {
1.72 noro 3938: ivect = imat[i];
3939: k = ivect->head; svect[k] %= m;
3940: if ( c = svect[k] ) {
1.76 noro 3941: maxrs = MAX(maxrs,rp0[i]->sugar);
1.74 noro 3942: c = _chsgnsf(c); redv = nd_ps[rp0[i]->index];
1.72 noro 3943: len = LEN(redv); mr = BDY(redv);
3944: svect[k] = 0; prev = k;
3945: switch ( ivect->width ) {
3946: case 1:
3947: ivc = ivect->index.c;
3948: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3949: pos = prev+ivc[j]; prev = pos;
3950: svect[pos] = _addsf(_mulsf(CM(mr),c),svect[pos]);
3951: }
3952: break;
3953: case 2:
3954: ivs = ivect->index.s;
3955: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3956: pos = prev+ivs[j]; prev = pos;
3957: svect[pos] = _addsf(_mulsf(CM(mr),c),svect[pos]);
3958: }
3959: break;
3960: case 4:
3961: ivi = ivect->index.i;
3962: for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) {
3963: pos = prev+ivi[j]; prev = pos;
3964: svect[pos] = _addsf(_mulsf(CM(mr),c),svect[pos]);
3965: }
3966: break;
3967: }
3968: }
3969: }
1.76 noro 3970: return maxrs;
1.72 noro 3971: }
3972:
1.65 noro 3973: NDV vect_to_ndv(UINT *vect,int spcol,int col,int *rhead,UINT *s0vect)
3974: {
3975: int j,k,len;
3976: UINT *p;
3977: UINT c;
3978: NDV r;
3979: NMV mr0,mr;
3980:
3981: for ( j = 0, len = 0; j < spcol; j++ ) if ( vect[j] ) len++;
3982: if ( !len ) return 0;
3983: else {
1.74 noro 3984: mr0 = (NMV)GC_malloc_atomic_ignore_off_page(nmv_adv*len);
1.103 noro 3985: #if 0
1.74 noro 3986: ndv_alloc += nmv_adv*len;
1.103 noro 3987: #endif
1.65 noro 3988: mr = mr0;
3989: p = s0vect;
3990: for ( j = k = 0; j < col; j++, p += nd_wpd )
3991: if ( !rhead[j] ) {
3992: if ( c = vect[k++] ) {
3993: ndl_copy(p,DL(mr)); CM(mr) = c; NMV_ADV(mr);
3994: }
3995: }
3996: MKNDV(nd_nvar,mr0,len,r);
3997: return r;
3998: }
3999: }
4000:
1.74 noro 4001: int nd_sp_f4(int m,ND_pairs l,PGeoBucket bucket)
1.65 noro 4002: {
4003: ND_pairs t;
4004: NODE sp0,sp;
4005: int stat;
4006: ND spol;
4007:
4008: for ( t = l; t; t = NEXT(t) ) {
4009: stat = nd_sp(m,0,t,&spol);
4010: if ( !stat ) return 0;
4011: if ( spol ) {
4012: add_pbucket_symbolic(bucket,spol);
4013: }
4014: }
1.68 noro 4015: return 1;
1.65 noro 4016: }
4017:
4018: int nd_symbolic_preproc(PGeoBucket bucket,UINT **s0vect,NODE *r)
4019: {
4020: NODE rp0,rp;
4021: NM mul,head,s0,s;
1.76 noro 4022: int index,col,i,sugar;
1.65 noro 4023: RHist h;
4024: UINT *s0v,*p;
4025: NM_ind_pair pair;
4026: ND red;
4027:
4028: s0 = 0; rp0 = 0; col = 0;
4029: while ( 1 ) {
4030: head = remove_head_pbucket_symbolic(bucket);
4031: if ( !head ) break;
4032: if ( !s0 ) s0 = head;
4033: else NEXT(s) = head;
4034: s = head;
4035: index = ndl_find_reducer(DL(head));
4036: if ( index >= 0 ) {
4037: h = nd_psh[index];
4038: NEWNM(mul);
4039: ndl_sub(DL(head),DL(h),DL(mul));
4040: if ( ndl_check_bound2(index,DL(mul)) ) return 0;
1.76 noro 4041: sugar = TD(DL(mul))+SG(nd_ps[index]);
4042: MKNM_ind_pair(pair,mul,index,sugar);
1.65 noro 4043: red = ndv_mul_nm_symbolic(mul,nd_ps[index]);
4044: add_pbucket_symbolic(bucket,nd_remove_head(red));
4045: NEXTNODE(rp0,rp); BDY(rp) = (pointer)pair;
4046: }
4047: col++;
4048: }
1.72 noro 4049: if ( rp0 ) NEXT(rp) = 0;
4050: NEXT(s) = 0;
1.65 noro 4051: s0v = (UINT *)MALLOC_ATOMIC(col*nd_wpd*sizeof(UINT));
4052: for ( i = 0, p = s0v, s = s0; i < col;
4053: i++, p += nd_wpd, s = NEXT(s) ) ndl_copy(DL(s),p);
4054: *s0vect = s0v;
4055: *r = rp0;
4056: return col;
4057: }
4058:
1.69 noro 4059: NODE nd_f4(int m)
4060: {
4061: int i,nh,stat,index;
4062: NODE r,g;
4063: ND_pairs d,l,t;
4064: ND spol,red;
4065: NDV nf,redv;
4066: NM s0,s;
1.84 noro 4067: NODE rp0,srp0,nflist;
1.69 noro 4068: int nsp,nred,col,rank,len,k,j,a;
4069: UINT c;
1.74 noro 4070: UINT **spmat;
1.69 noro 4071: UINT *s0vect,*svect,*p,*v;
4072: int *colstat;
4073: IndArray *imat;
4074: int *rhead;
4075: int spcol,sprow;
4076: int sugar;
4077: PGeoBucket bucket;
4078: struct oEGT eg0,eg1,eg_f4;
4079:
4080: if ( !m )
4081: error("nd_f4 : not implemented");
1.103 noro 4082: #if 0
1.74 noro 4083: ndv_alloc = 0;
1.103 noro 4084: #endif
1.69 noro 4085: g = 0; d = 0;
4086: for ( i = 0; i < nd_psn; i++ ) {
4087: d = update_pairs(d,g,i);
4088: g = update_base(g,i);
4089: }
4090: while ( d ) {
4091: get_eg(&eg0);
4092: l = nd_minsugarp(d,&d);
4093: sugar = SG(l);
4094: bucket = create_pbucket();
1.74 noro 4095: stat = nd_sp_f4(m,l,bucket);
1.69 noro 4096: if ( !stat ) {
4097: for ( t = l; NEXT(t); t = NEXT(t) );
4098: NEXT(t) = d; d = l;
1.103 noro 4099: d = nd_reconstruct(0,d);
1.69 noro 4100: continue;
4101: }
1.86 noro 4102: if ( bucket->m < 0 ) continue;
1.69 noro 4103: col = nd_symbolic_preproc(bucket,&s0vect,&rp0);
4104: if ( !col ) {
4105: for ( t = l; NEXT(t); t = NEXT(t) );
4106: NEXT(t) = d; d = l;
1.103 noro 4107: d = nd_reconstruct(0,d);
1.69 noro 4108: continue;
4109: }
4110: get_eg(&eg1); init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg1);
1.74 noro 4111: if ( DP_Print )
4112: fprintf(asir_out,"sugar=%d,symb=%fsec,",
4113: sugar,eg_f4.exectime+eg_f4.gctime);
4114: if ( 1 )
4115: nflist = nd_f4_red(m,l,s0vect,col,rp0);
4116: else
4117: nflist = nd_f4_red_dist(m,l,s0vect,col,rp0);
1.69 noro 4118: /* adding new bases */
1.74 noro 4119: for ( r = nflist; r; r = NEXT(r) ) {
4120: nf = (NDV)BDY(r);
1.69 noro 4121: ndv_removecont(m,nf);
1.77 noro 4122: nh = ndv_newps(m,nf,0);
1.69 noro 4123: d = update_pairs(d,g,nh);
4124: g = update_base(g,nh);
4125: }
4126: }
4127: for ( r = g; r; r = NEXT(r) ) BDY(r) = (pointer)nd_ps[(int)BDY(r)];
1.103 noro 4128: #if 0
1.74 noro 4129: fprintf(asir_out,"ndv_alloc=%d\n",ndv_alloc);
1.103 noro 4130: #endif
1.69 noro 4131: return g;
4132: }
1.74 noro 4133:
4134: NODE nd_f4_red(int m,ND_pairs sp0,UINT *s0vect,int col,NODE rp0)
1.63 noro 4135: {
1.67 noro 4136: IndArray *imat;
1.74 noro 4137: int nsp,nred,spcol,sprow,a;
1.65 noro 4138: int *rhead;
1.74 noro 4139: int i,j,k,l,rank;
4140: NODE rp,r0,r;
4141: ND_pairs sp;
4142: ND spol;
4143: int **spmat;
4144: UINT *svect,*v;
4145: int *colstat;
1.80 noro 4146: struct oEGT eg0,eg1,eg2,eg_f4,eg_f4_1,eg_f4_2;
1.74 noro 4147: NM_ind_pair *rvect;
1.76 noro 4148: int maxrs;
4149: int *spsugar;
1.63 noro 4150:
1.74 noro 4151: get_eg(&eg0);
4152: for ( sp = sp0, nsp = 0; sp; sp = NEXT(sp), nsp++ );
4153: nred = length(rp0); spcol = col-nred;
4154: imat = (IndArray *)ALLOCA(nred*sizeof(IndArray));
4155: rhead = (int *)ALLOCA(col*sizeof(int));
4156: for ( i = 0; i < col; i++ ) rhead[i] = 0;
1.63 noro 4157:
1.74 noro 4158: /* construction of index arrays */
4159: rvect = (NM_ind_pair *)ALLOCA(nred*sizeof(NM_ind_pair));
4160: for ( rp = rp0, i = 0; rp; i++, rp = NEXT(rp) ) {
4161: rvect[i] = (NM_ind_pair)BDY(rp);
4162: imat[i] = nm_ind_pair_to_vect_compress(m,s0vect,col,rvect[i]);
4163: rhead[imat[i]->head] = 1;
4164: }
4165:
4166: /* elimination (1st step) */
4167: spmat = (int **)ALLOCA(nsp*sizeof(UINT *));
4168: svect = (UINT *)ALLOCA(col*sizeof(UINT));
1.76 noro 4169: spsugar = (int *)ALLOCA(nsp*sizeof(UINT));
1.74 noro 4170: for ( a = sprow = 0, sp = sp0; a < nsp; a++, sp = NEXT(sp) ) {
4171: nd_sp(m,0,sp,&spol);
1.75 noro 4172: if ( !spol ) continue;
1.74 noro 4173: nd_to_vect(m,s0vect,col,spol,svect);
1.76 noro 4174: if ( m == -1 )
4175: maxrs = ndv_reduce_vect_sf(m,svect,col,imat,rvect,nred);
4176: else
4177: maxrs = ndv_reduce_vect(m,svect,col,imat,rvect,nred);
1.74 noro 4178: for ( i = 0; i < col; i++ ) if ( svect[i] ) break;
4179: if ( i < col ) {
4180: spmat[sprow] = v = (UINT *)MALLOC_ATOMIC(spcol*sizeof(UINT));
4181: for ( j = k = 0; j < col; j++ )
4182: if ( !rhead[j] ) v[k++] = svect[j];
1.76 noro 4183: spsugar[sprow] = MAX(maxrs,SG(spol));
1.74 noro 4184: sprow++;
4185: }
1.76 noro 4186: nd_free(spol);
1.74 noro 4187: }
1.80 noro 4188: get_eg(&eg1); init_eg(&eg_f4_1); add_eg(&eg_f4_1,&eg0,&eg1);
4189: if ( DP_Print ) {
4190: fprintf(asir_out,"elim1=%fsec,",eg_f4_1.exectime+eg_f4_1.gctime);
4191: fflush(asir_out);
4192: }
1.74 noro 4193: /* free index arrays */
4194: for ( i = 0; i < nred; i++ ) GC_free(imat[i]->index.c);
4195:
4196: /* elimination (2nd step) */
4197: colstat = (int *)ALLOCA(spcol*sizeof(int));
4198: if ( m == -1 )
1.76 noro 4199: rank = nd_gauss_elim_sf(spmat,spsugar,sprow,spcol,m,colstat);
1.74 noro 4200: else
1.76 noro 4201: rank = nd_gauss_elim_mod(spmat,spsugar,sprow,spcol,m,colstat);
1.74 noro 4202: r0 = 0;
4203: for ( i = 0; i < rank; i++ ) {
4204: NEXTNODE(r0,r); BDY(r) =
4205: (pointer)vect_to_ndv(spmat[i],spcol,col,rhead,s0vect);
1.76 noro 4206: SG((NDV)BDY(r)) = spsugar[i];
1.74 noro 4207: GC_free(spmat[i]);
4208: }
4209: for ( ; i < sprow; i++ ) GC_free(spmat[i]);
1.80 noro 4210: get_eg(&eg2); init_eg(&eg_f4_2); add_eg(&eg_f4_2,&eg1,&eg2);
4211: init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg2);
1.74 noro 4212: if ( DP_Print ) {
1.80 noro 4213: fprintf(asir_out,"elim2=%fsec\n",eg_f4_2.exectime+eg_f4_2.gctime);
1.74 noro 4214: fprintf(asir_out,"nsp=%d,nred=%d,spmat=(%d,%d),rank=%d ",
4215: nsp,nred,sprow,spcol,rank);
4216: fprintf(asir_out,"%fsec\n",eg_f4.exectime+eg_f4.gctime);
1.63 noro 4217: }
1.74 noro 4218: return r0;
4219: }
4220:
4221: FILE *nd_write,*nd_read;
4222:
4223: void nd_send_int(int a) {
4224: write_int(nd_write,&a);
4225: }
4226:
4227: void nd_send_intarray(int *p,int len) {
4228: write_intarray(nd_write,p,len);
4229: }
4230:
4231: int nd_recv_int() {
4232: int a;
4233:
4234: read_int(nd_read,&a);
4235: return a;
4236: }
4237:
4238: void nd_recv_intarray(int *p,int len) {
4239: read_intarray(nd_read,p,len);
4240: }
4241:
4242: void nd_send_ndv(NDV p) {
4243: int len,i;
4244: NMV m;
4245:
4246: if ( !p ) nd_send_int(0);
4247: else {
4248: len = LEN(p);
4249: nd_send_int(len);
4250: m = BDY(p);
4251: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
4252: nd_send_int(CM(m));
4253: nd_send_intarray(DL(m),nd_wpd);
1.68 noro 4254: }
1.74 noro 4255: }
4256: }
4257:
4258: void nd_send_nd(ND p) {
4259: int len,i;
4260: NM m;
4261:
4262: if ( !p ) nd_send_int(0);
4263: else {
4264: len = LEN(p);
4265: nd_send_int(len);
4266: m = BDY(p);
4267: for ( i = 0; i < len; i++, m = NEXT(m) ) {
4268: nd_send_int(CM(m));
4269: nd_send_intarray(DL(m),nd_wpd);
1.65 noro 4270: }
1.74 noro 4271: }
4272: }
1.65 noro 4273:
1.74 noro 4274: NDV nd_recv_ndv()
4275: {
4276: int len,i;
4277: NMV m,m0;
4278: NDV r;
1.65 noro 4279:
1.74 noro 4280: len = nd_recv_int();
4281: if ( !len ) return 0;
4282: else {
4283: m0 = m = (NMV)GC_malloc_atomic_ignore_off_page(nmv_adv*len);
1.103 noro 4284: #if 0
1.74 noro 4285: ndv_alloc += len*nmv_adv;
1.103 noro 4286: #endif
1.74 noro 4287: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
4288: CM(m) = nd_recv_int();
4289: nd_recv_intarray(DL(m),nd_wpd);
1.71 noro 4290: }
1.74 noro 4291: MKNDV(nd_nvar,m0,len,r);
4292: return r;
4293: }
4294: }
1.65 noro 4295:
1.74 noro 4296: int ox_exec_f4_red(Q proc)
4297: {
4298: Obj obj;
4299: STRING fname;
4300: NODE arg;
4301: int s;
4302: extern int ox_need_conv,ox_file_io;
4303:
4304: MKSTR(fname,"nd_exec_f4_red");
4305: arg = mknode(2,proc,fname);
4306: Pox_cmo_rpc(arg,&obj);
4307: s = get_ox_server_id(QTOS(proc));
4308: nd_write = iofp[s].out;
4309: nd_read = iofp[s].in;
4310: ox_need_conv = ox_file_io = 0;
4311: return s;
4312: }
4313:
4314: NODE nd_f4_red_dist(int m,ND_pairs sp0,UINT *s0vect,int col,NODE rp0)
4315: {
4316: int nsp,nred;
4317: int i,rank,s;
4318: NODE rp,r0,r;
4319: ND_pairs sp;
4320: NM_ind_pair pair;
4321: NMV nmv;
4322: NM nm;
4323: NDV nf;
4324: Obj proc,dmy;
4325:
4326: ox_launch_main(0,0,&proc);
4327: s = ox_exec_f4_red((Q)proc);
4328:
4329: nd_send_int(m);
4330: nd_send_int(nd_nvar);
4331: nd_send_int(nd_bpe);
4332: nd_send_int(nd_wpd);
4333: nd_send_int(nmv_adv);
4334:
1.89 noro 4335: saveobj(nd_write,dp_current_spec->obj); fflush(nd_write);
1.74 noro 4336:
4337: nd_send_int(nd_psn);
4338: for ( i = 0; i < nd_psn; i++ ) nd_send_ndv(nd_ps[i]);
4339:
4340: for ( sp = sp0, nsp = 0; sp; sp = NEXT(sp), nsp++ );
4341: nd_send_int(nsp);
4342: for ( i = 0, sp = sp0; i < nsp; i++, sp = NEXT(sp) ) {
4343: nd_send_int(sp->i1); nd_send_int(sp->i2);
4344: }
4345:
4346: nd_send_int(col); nd_send_intarray(s0vect,col*nd_wpd);
4347:
4348: nred = length(rp0); nd_send_int(nred);
4349: for ( i = 0, rp = rp0; i < nred; i++, rp = NEXT(rp) ) {
4350: pair = (NM_ind_pair)BDY(rp);
4351: nd_send_int(pair->index);
4352: nd_send_intarray(pair->mul->dl,nd_wpd);
4353: }
4354: fflush(nd_write);
4355: rank = nd_recv_int();
4356: fprintf(asir_out,"rank=%d\n",rank);
4357: r0 = 0;
4358: for ( i = 0; i < rank; i++ ) {
4359: nf = nd_recv_ndv();
4360: NEXTNODE(r0,r); BDY(r) = (pointer)nf;
4361: }
4362: Pox_shutdown(mknode(1,proc),&dmy);
4363: return r0;
4364: }
4365:
4366: /* server side */
4367:
4368: void nd_exec_f4_red_dist()
4369: {
4370: int m,i,nsp,col,s0size,nred,spcol,j,k;
4371: NM_ind_pair *rp0;
4372: NDV nf;
4373: UINT *s0vect;
4374: IndArray *imat;
4375: int *rhead;
4376: int **spmat;
4377: UINT *svect,*v;
4378: ND_pairs *sp0;
4379: int *colstat;
4380: int a,sprow,rank;
1.89 noro 4381: struct order_spec *ord;
1.74 noro 4382: Obj ordspec;
4383: ND spol;
1.76 noro 4384: int maxrs;
4385: int *spsugar;
1.74 noro 4386:
4387: nd_read = iofp[0].in;
4388: nd_write = iofp[0].out;
4389: m = nd_recv_int();
4390: nd_nvar = nd_recv_int();
4391: nd_bpe = nd_recv_int();
4392: nd_wpd = nd_recv_int();
4393: nmv_adv = nd_recv_int();
4394:
4395: loadobj(nd_read,&ordspec);
1.89 noro 4396: create_order_spec(0,ordspec,&ord);
4397: nd_init_ord(ord);
1.74 noro 4398: nd_setup_parameters(nd_nvar,0);
4399:
4400: nd_psn = nd_recv_int();
4401: nd_ps = (NDV *)MALLOC(nd_psn*sizeof(NDV));
4402: nd_bound = (UINT **)MALLOC(nd_psn*sizeof(UINT *));
4403: for ( i = 0; i < nd_psn; i++ ) {
4404: nd_ps[i] = nd_recv_ndv();
4405: nd_bound[i] = ndv_compute_bound(nd_ps[i]);
4406: }
4407:
4408: nsp = nd_recv_int();
4409: sp0 = (ND_pairs *)MALLOC(nsp*sizeof(ND_pairs));
4410: for ( i = 0; i < nsp; i++ ) {
4411: NEWND_pairs(sp0[i]);
4412: sp0[i]->i1 = nd_recv_int(); sp0[i]->i2 = nd_recv_int();
4413: ndl_lcm(HDL(nd_ps[sp0[i]->i1]),HDL(nd_ps[sp0[i]->i2]),LCM(sp0[i]));
4414: }
4415:
4416: col = nd_recv_int();
4417: s0size = col*nd_wpd;
4418: s0vect = (UINT *)MALLOC(s0size*sizeof(UINT));
4419: nd_recv_intarray(s0vect,s0size);
4420:
4421: nred = nd_recv_int();
4422: rp0 = (NM_ind_pair *)MALLOC(nred*sizeof(NM_ind_pair));
4423: for ( i = 0; i < nred; i++ ) {
4424: rp0[i] = (NM_ind_pair)MALLOC(sizeof(struct oNM_ind_pair));
4425: rp0[i]->index = nd_recv_int();
4426: rp0[i]->mul = (NM)MALLOC(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT));
4427: nd_recv_intarray(rp0[i]->mul->dl,nd_wpd);
4428: }
4429:
4430: spcol = col-nred;
4431: imat = (IndArray *)MALLOC(nred*sizeof(IndArray));
4432: rhead = (int *)MALLOC(col*sizeof(int));
4433: for ( i = 0; i < col; i++ ) rhead[i] = 0;
4434:
4435: /* construction of index arrays */
4436: for ( i = 0; i < nred; i++ ) {
4437: imat[i] = nm_ind_pair_to_vect_compress(m,s0vect,col,rp0[i]);
4438: rhead[imat[i]->head] = 1;
4439: }
4440:
4441: /* elimination (1st step) */
4442: spmat = (int **)MALLOC(nsp*sizeof(UINT *));
4443: svect = (UINT *)MALLOC(col*sizeof(UINT));
1.76 noro 4444: spsugar = (int *)ALLOCA(nsp*sizeof(UINT));
1.74 noro 4445: for ( a = sprow = 0; a < nsp; a++ ) {
4446: nd_sp(m,0,sp0[a],&spol);
1.75 noro 4447: if ( !spol ) continue;
1.74 noro 4448: nd_to_vect(m,s0vect,col,spol,svect);
1.76 noro 4449: if ( m == -1 )
4450: maxrs = ndv_reduce_vect_sf(m,svect,col,imat,rp0,nred);
4451: else
4452: maxrs = ndv_reduce_vect(m,svect,col,imat,rp0,nred);
1.74 noro 4453: for ( i = 0; i < col; i++ ) if ( svect[i] ) break;
4454: if ( i < col ) {
4455: spmat[sprow] = v = (UINT *)MALLOC(spcol*sizeof(UINT));
4456: for ( j = k = 0; j < col; j++ )
4457: if ( !rhead[j] ) v[k++] = svect[j];
1.76 noro 4458: spsugar[sprow] = MAX(maxrs,SG(spol));
1.74 noro 4459: sprow++;
4460: }
1.76 noro 4461: nd_free(spol);
1.74 noro 4462: }
4463: /* elimination (2nd step) */
4464: colstat = (int *)ALLOCA(spcol*sizeof(int));
4465: if ( m == -1 )
1.76 noro 4466: rank = nd_gauss_elim_sf(spmat,spsugar,sprow,spcol,m,colstat);
1.74 noro 4467: else
1.76 noro 4468: rank = nd_gauss_elim_mod(spmat,spsugar,sprow,spcol,m,colstat);
1.74 noro 4469: nd_send_int(rank);
4470: for ( i = 0; i < rank; i++ ) {
4471: nf = vect_to_ndv(spmat[i],spcol,col,rhead,s0vect);
4472: nd_send_ndv(nf);
1.63 noro 4473: }
1.74 noro 4474: fflush(nd_write);
1.76 noro 4475: }
4476:
4477: int nd_gauss_elim_mod(int **mat0,int *sugar,int row,int col,int md,int *colstat)
4478: {
4479: int i,j,k,l,inv,a,rank,s;
4480: unsigned int *t,*pivot,*pk;
4481: unsigned int **mat;
4482:
4483: mat = (unsigned int **)mat0;
4484: for ( rank = 0, j = 0; j < col; j++ ) {
4485: for ( i = rank; i < row; i++ )
4486: mat[i][j] %= md;
4487: for ( i = rank; i < row; i++ )
4488: if ( mat[i][j] )
4489: break;
4490: if ( i == row ) {
4491: colstat[j] = 0;
4492: continue;
4493: } else
4494: colstat[j] = 1;
4495: if ( i != rank ) {
4496: t = mat[i]; mat[i] = mat[rank]; mat[rank] = t;
4497: s = sugar[i]; sugar[i] = sugar[rank]; sugar[rank] = s;
4498: }
4499: pivot = mat[rank];
4500: s = sugar[rank];
4501: inv = invm(pivot[j],md);
4502: for ( k = j, pk = pivot+k; k < col; k++, pk++ )
4503: if ( *pk ) {
4504: if ( *pk >= (unsigned int)md )
4505: *pk %= md;
4506: DMAR(*pk,inv,0,md,*pk)
4507: }
4508: for ( i = rank+1; i < row; i++ ) {
4509: t = mat[i];
4510: if ( a = t[j] ) {
4511: sugar[i] = MAX(sugar[i],s);
4512: red_by_vect(md,t+j,pivot+j,md-a,col-j);
4513: }
4514: }
4515: rank++;
4516: }
4517: for ( j = col-1, l = rank-1; j >= 0; j-- )
4518: if ( colstat[j] ) {
4519: pivot = mat[l];
4520: s = sugar[l];
4521: for ( i = 0; i < l; i++ ) {
4522: t = mat[i];
4523: t[j] %= md;
4524: if ( a = t[j] ) {
4525: sugar[i] = MAX(sugar[i],s);
4526: red_by_vect(md,t+j,pivot+j,md-a,col-j);
4527: }
4528: }
4529: l--;
4530: }
4531: for ( j = 0, l = 0; l < rank; j++ )
4532: if ( colstat[j] ) {
4533: t = mat[l];
4534: for ( k = j; k < col; k++ )
4535: if ( t[k] >= (unsigned int)md )
4536: t[k] %= md;
4537: l++;
4538: }
4539: return rank;
4540: }
4541:
4542: int nd_gauss_elim_sf(int **mat0,int *sugar,int row,int col,int md,int *colstat)
4543: {
4544: int i,j,k,l,inv,a,rank,s;
4545: unsigned int *t,*pivot,*pk;
4546: unsigned int **mat;
4547:
4548: mat = (unsigned int **)mat0;
4549: for ( rank = 0, j = 0; j < col; j++ ) {
4550: for ( i = rank; i < row; i++ )
4551: if ( mat[i][j] )
4552: break;
4553: if ( i == row ) {
4554: colstat[j] = 0;
4555: continue;
4556: } else
4557: colstat[j] = 1;
4558: if ( i != rank ) {
4559: t = mat[i]; mat[i] = mat[rank]; mat[rank] = t;
4560: s = sugar[i]; sugar[i] = sugar[rank]; sugar[rank] = s;
4561: }
4562: pivot = mat[rank];
4563: s = sugar[rank];
4564: inv = _invsf(pivot[j]);
4565: for ( k = j, pk = pivot+k; k < col; k++, pk++ )
4566: if ( *pk )
4567: *pk = _mulsf(*pk,inv);
4568: for ( i = rank+1; i < row; i++ ) {
4569: t = mat[i];
4570: if ( a = t[j] ) {
4571: sugar[i] = MAX(sugar[i],s);
4572: red_by_vect_sf(md,t+j,pivot+j,_chsgnsf(a),col-j);
4573: }
4574: }
4575: rank++;
4576: }
4577: for ( j = col-1, l = rank-1; j >= 0; j-- )
4578: if ( colstat[j] ) {
4579: pivot = mat[l];
4580: s = sugar[l];
4581: for ( i = 0; i < l; i++ ) {
4582: t = mat[i];
4583: if ( a = t[j] ) {
4584: sugar[i] = MAX(sugar[i],s);
4585: red_by_vect_sf(md,t+j,pivot+j,_chsgnsf(a),col-j);
4586: }
4587: }
4588: l--;
4589: }
4590: return rank;
1.77 noro 4591: }
4592:
4593: int ndv_ishomo(NDV p)
4594: {
4595: NMV m;
4596: int len,h;
4597:
4598: if ( !p ) return 1;
4599: len = LEN(p);
4600: m = BDY(p);
4601: h = TD(DL(m));
4602: NMV_ADV(m);
4603: for ( len--; len; len--, NMV_ADV(m) )
4604: if ( TD(DL(m)) != h ) return 0;
4605: return 1;
4606: }
4607:
4608: void ndv_save(NDV p,int index)
4609: {
4610: FILE *s;
4611: char name[BUFSIZ];
4612: short id;
4613: int nv,sugar,len,n,i,td,e,j;
4614: NMV m;
4615: unsigned int *dl;
4616:
4617: sprintf(name,"%s/%d",Demand,index);
4618: s = fopen(name,"w");
4619: savevl(s,0);
4620: if ( !p ) {
4621: saveobj(s,0);
4622: return;
4623: }
4624: id = O_DP;
4625: nv = NV(p);
4626: sugar = SG(p);
4627: len = LEN(p);
4628: write_short(s,&id); write_int(s,&nv); write_int(s,&sugar);
4629: write_int(s,&len);
4630:
4631: for ( m = BDY(p), i = 0; i < len; i++, NMV_ADV(m) ) {
4632: saveobj(s,(Obj)CQ(m));
4633: dl = DL(m);
4634: td = TD(dl);
4635: write_int(s,&td);
4636: for ( j = 0; j < nv; j++ ) {
4637: e = GET_EXP(dl,j);
4638: write_int(s,&e);
4639: }
4640: }
4641: fclose(s);
4642: }
4643:
4644: NDV ndv_load(int index)
4645: {
4646: FILE *s;
4647: char name[BUFSIZ];
4648: short id;
4649: int nv,sugar,len,n,i,td,e,j;
4650: NDV d;
4651: NMV m0,m;
4652: unsigned int *dl;
4653: Obj obj;
4654:
4655: sprintf(name,"%s/%d",Demand,index);
4656: s = fopen(name,"r");
4657: if ( !s ) return 0;
4658:
4659: skipvl(s);
4660: read_short(s,&id);
4661: if ( !id ) return 0;
4662: read_int(s,&nv);
4663: read_int(s,&sugar);
4664: read_int(s,&len);
4665:
4666: m0 = m = MALLOC(len*nmv_adv);
4667: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
4668: loadobj(s,&obj); CQ(m) = (Q)obj;
4669: dl = DL(m);
4670: ndl_zero(dl);
4671: read_int(s,&td); TD(dl) = td;
4672: for ( j = 0; j < nv; j++ ) {
4673: read_int(s,&e);
4674: PUT_EXP(dl,j,e);
4675: }
4676: if ( nd_blockmask ) ndl_weight_mask(dl);
4677: }
4678: fclose(s);
4679: MKNDV(nv,m0,len,d);
4680: SG(d) = sugar;
4681: return d;
1.99 noro 4682: }
4683:
1.102 noro 4684: void nd_det(int mod,MAT f,P *rp)
1.99 noro 4685: {
4686: VL fv,tv;
1.102 noro 4687: int n,i,j,max,e,nvar,sgn,k0,l0,len0,len,k,l,a;
1.99 noro 4688: pointer **m;
4689: Q mone;
1.102 noro 4690: NDV **dm;
4691: NDV *t,*mi,*mj;
4692: NDV d,s,mij,mjj;
4693: ND u;
4694: NMV nmv;
1.99 noro 4695: PGeoBucket bucket;
4696: struct order_spec *ord;
4697:
4698: create_order_spec(0,0,&ord);
4699: nd_init_ord(ord);
4700: get_vars((Obj)f,&fv);
4701: if ( f->row != f->col )
4702: error("nd_det : non-square matrix");
4703: n = f->row;
4704: for ( nvar = 0, tv = fv; tv; tv = NEXT(tv), nvar++ );
4705: m = f->body;
4706: for ( i = 0, max = 0; i < n; i++ )
4707: for ( j = 0; j < n; j++ )
4708: for ( tv = fv; tv; tv = NEXT(tv) ) {
4709: e = getdeg(tv->v,(P)m[i][j]);
4710: max = MAX(e,max);
4711: }
4712: nd_setup_parameters(nvar,1024);
1.102 noro 4713: dm = (NDV **)almat_pointer(n,n);
1.99 noro 4714: for ( i = 0, max = 0; i < n; i++ )
1.102 noro 4715: for ( j = 0; j < n; j++ ) {
4716: dm[i][j] = ptondv(CO,fv,m[i][j]);
4717: if ( mod ) ndv_mod(mod,dm[i][j]);
4718: if ( dm[i][j] && !LEN(dm[i][j]) ) dm[i][j] = 0;
4719: }
4720: d = ptondv(CO,fv,(P)ONE);
4721: if ( mod ) ndv_mod(mod,d);
1.99 noro 4722: chsgnq(ONE,&mone);
4723: for ( j = 0, sgn = 1; j < n; j++ ) {
1.104 ! noro 4724: if ( DP_Print ) fprintf(stderr,"j=%d\n",j);
1.99 noro 4725: for ( i = j; i < n && !dm[i][j]; i++ );
4726: if ( i == n ) {
4727: *rp = 0;
4728: return;
4729: }
1.102 noro 4730: k0 = i; l0 = j; len0 = LEN(dm[k0][l0]);
1.99 noro 4731: for ( k = j; k < n; k++ )
4732: for ( l = j; l < n; l++ )
1.102 noro 4733: if ( dm[k][l] && LEN(dm[k][l]) < len0 ) {
4734: k0 = k; l0 = l; len0 = LEN(dm[k][l]);
1.99 noro 4735: }
4736: if ( k0 != j ) {
4737: t = dm[j]; dm[j] = dm[k0]; dm[k0] = t;
4738: sgn = -sgn;
4739: }
4740: if ( l0 != j ) {
4741: for ( k = j; k < n; k++ ) {
4742: s = dm[k][j]; dm[k][j] = dm[k][l0]; dm[k][l0] = s;
4743: }
4744: sgn = -sgn;
4745: }
4746: for ( i = j+1, mj = dm[j], mjj = mj[j]; i < n; i++ ) {
1.104 ! noro 4747: if ( DP_Print ) fprintf(stderr," i=%d\n ",i);
1.99 noro 4748: mi = dm[i]; mij = mi[j];
1.102 noro 4749: if ( mod )
4750: ndv_mul_c(mod,mij,mod-1);
4751: else
4752: ndv_mul_c_q(mij,mone);
1.99 noro 4753: for ( k = j+1; k < n; k++ ) {
1.104 ! noro 4754: if ( DP_Print ) fprintf(stderr,"k=%d ",k);
1.99 noro 4755: bucket = create_pbucket();
1.104 ! noro 4756: if ( mi[k] ) {
1.102 noro 4757: nmv = BDY(mjj); len = LEN(mjj);
4758: for ( a = 0; a < len; a++, NMV_ADV(nmv) ) {
4759: u = ndv_mul_nmv_trunc(mod,nmv,mi[k],DL(BDY(d)));
4760: add_pbucket(mod,bucket,u);
1.99 noro 4761: }
1.104 ! noro 4762: }
1.99 noro 4763: if ( mj[k] && mij ) {
1.102 noro 4764: nmv = BDY(mij); len = LEN(mij);
4765: for ( a = 0; a < len; a++, NMV_ADV(nmv) ) {
4766: u = ndv_mul_nmv_trunc(mod,nmv,mj[k],DL(BDY(d)));
4767: add_pbucket(mod,bucket,u);
1.99 noro 4768: }
4769: }
1.104 ! noro 4770: u = nd_quo(mod,bucket,d);
1.102 noro 4771: mi[k] = ndtondv(mod,u);
1.99 noro 4772: }
1.104 ! noro 4773: if ( DP_Print ) fprintf(stderr,"\n",k);
1.99 noro 4774: }
4775: d = mjj;
4776: }
1.100 noro 4777: if ( sgn < 0 )
1.102 noro 4778: if ( mod )
4779: ndv_mul_c(mod,d,mod-1);
4780: else
4781: ndv_mul_c_q(d,mone);
4782: *rp = ndvtop(mod,CO,fv,d);
1.99 noro 4783: }
4784:
1.102 noro 4785: ND ndv_mul_nmv_trunc(int mod,NMV m0,NDV p,UINT *d)
1.99 noro 4786: {
4787: NM mr,mr0;
1.102 noro 4788: NM tnm;
4789: NMV m;
1.99 noro 4790: UINT *d0,*dt,*dm;
4791: int c,n,td,i,c1,c2,len;
4792: Q q;
4793: ND r;
4794:
4795: if ( !p ) return 0;
4796: else {
1.102 noro 4797: n = NV(p); m = BDY(p); len = LEN(p);
1.99 noro 4798: d0 = DL(m0);
1.102 noro 4799: td = TD(d);
1.99 noro 4800: mr0 = 0;
4801: NEWNM(tnm);
1.102 noro 4802: if ( mod ) {
4803: c = CM(m0);
4804: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
4805: ndl_add(DL(m),d0,DL(tnm));
4806: if ( ndl_reducible(DL(tnm),d) ) {
4807: NEXTNM(mr0,mr);
4808: c1 = CM(m); DMAR(c1,c,0,mod,c2); CM(mr) = c2;
1.104 ! noro 4809: ndl_copy(DL(tnm),DL(mr));
1.102 noro 4810: }
4811: }
4812: } else {
4813: q = CQ(m0);
4814: for ( i = 0; i < len; i++, NMV_ADV(m) ) {
4815: ndl_add(DL(m),d0,DL(tnm));
4816: if ( ndl_reducible(DL(tnm),d) ) {
4817: NEXTNM(mr0,mr);
4818: mulq(CQ(m),q,&CQ(mr));
1.104 ! noro 4819: ndl_copy(DL(tnm),DL(mr));
1.102 noro 4820: }
1.99 noro 4821: }
4822: }
4823: if ( !mr0 )
4824: return 0;
4825: else {
4826: NEXT(mr) = 0;
4827: MKND(NV(p),mr0,len,r);
4828: SG(r) = SG(p) + TD(d0);
4829: return r;
4830: }
4831: }
1.59 noro 4832: }
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>