1 | // emacs edit mode for this file is -*- C++ -*- |
---|
2 | /**************************************** |
---|
3 | * Computer Algebra System SINGULAR * |
---|
4 | ****************************************/ |
---|
5 | /* |
---|
6 | * ABSTRACT: flint mpoly |
---|
7 | */ |
---|
8 | |
---|
9 | #include "misc/auxiliary.h" |
---|
10 | #include "flintconv.h" |
---|
11 | #include "flint_mpoly.h" |
---|
12 | |
---|
13 | #ifdef HAVE_FLINT |
---|
14 | #if __FLINT_RELEASE >= 20503 |
---|
15 | #include "coeffs/coeffs.h" |
---|
16 | #include "coeffs/longrat.h" |
---|
17 | #include "polys/monomials/p_polys.h" |
---|
18 | |
---|
19 | #include <vector> |
---|
20 | |
---|
21 | /****** ring conversion ******/ |
---|
22 | |
---|
23 | BOOLEAN convSingRFlintR(fmpq_mpoly_ctx_t ctx, const ring r) |
---|
24 | { |
---|
25 | if (rRing_ord_pure_dp(r)) |
---|
26 | { |
---|
27 | fmpq_mpoly_ctx_init(ctx,r->N,ORD_DEGREVLEX); |
---|
28 | return FALSE; |
---|
29 | } |
---|
30 | else if (rRing_ord_pure_Dp(r)) |
---|
31 | { |
---|
32 | fmpq_mpoly_ctx_init(ctx,r->N,ORD_DEGLEX); |
---|
33 | return FALSE; |
---|
34 | } |
---|
35 | else if (rRing_ord_pure_lp(r)) |
---|
36 | { |
---|
37 | fmpq_mpoly_ctx_init(ctx,r->N,ORD_LEX); |
---|
38 | return FALSE; |
---|
39 | } |
---|
40 | return TRUE; |
---|
41 | } |
---|
42 | |
---|
43 | BOOLEAN convSingRFlintR(nmod_mpoly_ctx_t ctx, const ring r) |
---|
44 | { |
---|
45 | if (rRing_ord_pure_dp(r)) |
---|
46 | { |
---|
47 | nmod_mpoly_ctx_init(ctx,r->N,ORD_DEGREVLEX,r->cf->ch); |
---|
48 | return FALSE; |
---|
49 | } |
---|
50 | else if (rRing_ord_pure_Dp(r)) |
---|
51 | { |
---|
52 | nmod_mpoly_ctx_init(ctx,r->N,ORD_DEGLEX,r->cf->ch); |
---|
53 | return FALSE; |
---|
54 | } |
---|
55 | else if (rRing_ord_pure_lp(r)) |
---|
56 | { |
---|
57 | nmod_mpoly_ctx_init(ctx,r->N,ORD_LEX,r->cf->ch); |
---|
58 | return FALSE; |
---|
59 | } |
---|
60 | return TRUE; |
---|
61 | } |
---|
62 | |
---|
63 | /******** polynomial conversion ***********/ |
---|
64 | |
---|
65 | #if 1 |
---|
66 | // malloc is not thread safe; singular polynomials must be constructed in serial |
---|
67 | |
---|
68 | void convSingPFlintMP(fmpq_mpoly_t res, fmpq_mpoly_ctx_t ctx, poly p, int lp, const ring r) |
---|
69 | { |
---|
70 | fmpq_mpoly_init2(res, lp, ctx); |
---|
71 | ulong* exp=(ulong*)omAlloc((r->N+1)*sizeof(ulong)); |
---|
72 | while(p!=NULL) |
---|
73 | { |
---|
74 | number n=pGetCoeff(p); |
---|
75 | fmpq_t c; |
---|
76 | convSingNFlintN_QQ(c,n); |
---|
77 | #if SIZEOF_LONG==8 |
---|
78 | p_GetExpVL(p,(int64*)exp,r); |
---|
79 | fmpq_mpoly_push_term_fmpq_ui(res, c, exp, ctx); |
---|
80 | #else |
---|
81 | p_GetExpV(p,(int*)exp,r); |
---|
82 | fmpq_mpoly_push_term_fmpq_ui(res, c, &(exp[1]), ctx); |
---|
83 | #endif |
---|
84 | fmpq_clear(c); |
---|
85 | pIter(p); |
---|
86 | } |
---|
87 | fmpq_mpoly_reduce(res, ctx); // extra step for QQ ensures res has content canonically factored out |
---|
88 | omFreeSize(exp,(r->N+1)*sizeof(ulong)); |
---|
89 | } |
---|
90 | |
---|
91 | poly convFlintMPSingP(fmpq_mpoly_t f, fmpq_mpoly_ctx_t ctx, const ring r) |
---|
92 | { |
---|
93 | int d=fmpq_mpoly_length(f,ctx)-1; |
---|
94 | poly p=NULL; |
---|
95 | ulong* exp=(ulong*)omAlloc0((r->N+1)*sizeof(ulong)); |
---|
96 | fmpq_t c; |
---|
97 | fmpq_init(c); |
---|
98 | for(int i=d; i>=0; i--) |
---|
99 | { |
---|
100 | fmpq_mpoly_get_term_coeff_fmpq(c,f,i,ctx); |
---|
101 | poly pp=p_Init(r); |
---|
102 | #if SIZEOF_LONG==8 |
---|
103 | fmpq_mpoly_get_term_exp_ui(exp,f,i,ctx); |
---|
104 | p_SetExpVL(pp,(int64*)exp,r); |
---|
105 | #else |
---|
106 | fmpq_mpoly_get_term_exp_ui(&(exp[1]),f,i,ctx); |
---|
107 | p_SetExpV(pp,(int*)exp,r); |
---|
108 | #endif |
---|
109 | p_Setm(pp,r); |
---|
110 | number n=convFlintNSingN_QQ(c,r->cf); |
---|
111 | pSetCoeff0(pp,n); |
---|
112 | pNext(pp)=p; |
---|
113 | p=pp; |
---|
114 | } |
---|
115 | fmpq_clear(c); |
---|
116 | omFreeSize(exp,(r->N+1)*sizeof(ulong)); |
---|
117 | p_Test(p,r); |
---|
118 | return p; |
---|
119 | } |
---|
120 | |
---|
121 | poly convFlintMPSingP(nmod_mpoly_t f, nmod_mpoly_ctx_t ctx, const ring r) |
---|
122 | { |
---|
123 | int d=nmod_mpoly_length(f,ctx)-1; |
---|
124 | poly p=NULL; |
---|
125 | ulong* exp=(ulong*)omAlloc0((r->N+1)*sizeof(ulong)); |
---|
126 | for(int i=d; i>=0; i--) |
---|
127 | { |
---|
128 | ulong c=nmod_mpoly_get_term_coeff_ui(f,i,ctx); |
---|
129 | poly pp=p_Init(r); |
---|
130 | #if SIZEOF_LONG==8 |
---|
131 | nmod_mpoly_get_term_exp_ui(exp,f,i,ctx); |
---|
132 | p_SetExpVL(pp,(int64*)exp,r); |
---|
133 | #else |
---|
134 | nmod_mpoly_get_term_exp_ui(&(exp[1]),f,i,ctx); |
---|
135 | p_SetExpV(pp,(int*)exp,r); |
---|
136 | #endif |
---|
137 | p_Setm(pp,r); |
---|
138 | pSetCoeff0(pp,(number)c); |
---|
139 | pNext(pp)=p; |
---|
140 | p=pp; |
---|
141 | } |
---|
142 | omFreeSize(exp,(r->N+1)*sizeof(ulong)); |
---|
143 | p_Test(p,r); |
---|
144 | return p; |
---|
145 | } |
---|
146 | |
---|
147 | void convSingPFlintMP(nmod_mpoly_t res, nmod_mpoly_ctx_t ctx, poly p, int lp,const ring r) |
---|
148 | { |
---|
149 | nmod_mpoly_init2(res, lp, ctx); |
---|
150 | ulong* exp=(ulong*)omAlloc((r->N+1)*sizeof(ulong)); |
---|
151 | while(p!=NULL) |
---|
152 | { |
---|
153 | number n=pGetCoeff(p); |
---|
154 | #if SIZEOF_LONG==8 |
---|
155 | p_GetExpVL(p,(int64*)exp,r); |
---|
156 | nmod_mpoly_push_term_ui_ui(res, (ulong)n, exp, ctx); |
---|
157 | #else |
---|
158 | p_GetExpV(p,(int*)exp,r); |
---|
159 | nmod_mpoly_push_term_ui_ui(res, (ulong)n, &(exp[1]), ctx); |
---|
160 | #endif |
---|
161 | pIter(p); |
---|
162 | } |
---|
163 | omFreeSize(exp,(r->N+1)*sizeof(ulong)); |
---|
164 | } |
---|
165 | |
---|
166 | #else |
---|
167 | // malloc is thread safe; singular polynomials may be constructed in serial |
---|
168 | |
---|
169 | // like convSingNFlintN_QQ but it takes an initialized fmpq_t f |
---|
170 | static void my_convSingNFlintN_QQ(fmpq_t f, number n) |
---|
171 | { |
---|
172 | if (SR_HDL(n)&SR_INT) |
---|
173 | { |
---|
174 | fmpq_set_si(f,SR_TO_INT(n),1); |
---|
175 | } |
---|
176 | else if (n->s<3) |
---|
177 | { |
---|
178 | fmpz_set_mpz(fmpq_numref(f), n->z); |
---|
179 | fmpz_set_mpz(fmpq_denref(f), n->n); |
---|
180 | } |
---|
181 | else |
---|
182 | { |
---|
183 | fmpz_set_mpz(fmpq_numref(f), n->z); |
---|
184 | fmpz_one(fmpq_denref(f)); |
---|
185 | } |
---|
186 | } |
---|
187 | |
---|
188 | /* singular -> fmpq_mpoly conversion */ |
---|
189 | |
---|
190 | class convert_sing_to_fmpq_mpoly_base |
---|
191 | { |
---|
192 | public: |
---|
193 | slong num_threads; |
---|
194 | slong length; |
---|
195 | fmpq_mpoly_struct * res; |
---|
196 | const fmpq_mpoly_ctx_struct * ctx; |
---|
197 | std::vector<poly> markers; |
---|
198 | ring r; |
---|
199 | |
---|
200 | convert_sing_to_fmpq_mpoly_base(slong num_threads_, fmpq_mpoly_struct * res_, |
---|
201 | const fmpq_mpoly_ctx_struct * ctx_, const ring r_) |
---|
202 | : num_threads(num_threads_), |
---|
203 | res(res_), |
---|
204 | ctx(ctx_), |
---|
205 | r(r_) |
---|
206 | { |
---|
207 | } |
---|
208 | |
---|
209 | void init_poly(poly p) |
---|
210 | { |
---|
211 | length = 0; |
---|
212 | while (p != NULL) |
---|
213 | { |
---|
214 | markers.push_back(p); |
---|
215 | for (int i = 4096; i > 0; i--) |
---|
216 | { |
---|
217 | pIter(p); |
---|
218 | length++; |
---|
219 | if (p == NULL) |
---|
220 | break; |
---|
221 | } |
---|
222 | } |
---|
223 | |
---|
224 | fmpq_mpoly_init3(res, length, SI_LOG2(r->bitmask), ctx); |
---|
225 | } |
---|
226 | }; |
---|
227 | |
---|
228 | class convert_sing_to_fmpq_mpoly_arg |
---|
229 | { |
---|
230 | public: |
---|
231 | fmpq_t content; |
---|
232 | slong start_idx, end_idx; |
---|
233 | convert_sing_to_fmpq_mpoly_base* base; |
---|
234 | |
---|
235 | convert_sing_to_fmpq_mpoly_arg() {fmpq_init(content);} |
---|
236 | ~convert_sing_to_fmpq_mpoly_arg() {fmpq_clear(content);} |
---|
237 | }; |
---|
238 | |
---|
239 | static void convert_sing_to_fmpq_mpoly_content_worker(void * varg) |
---|
240 | { |
---|
241 | convert_sing_to_fmpq_mpoly_arg* arg = (convert_sing_to_fmpq_mpoly_arg*) varg; |
---|
242 | fmpq_t c; |
---|
243 | fmpq_init(c); |
---|
244 | |
---|
245 | slong idx = arg->start_idx/4096; |
---|
246 | poly p = arg->base->markers[idx]; |
---|
247 | idx *= 4096; |
---|
248 | while (idx < arg->start_idx) |
---|
249 | { |
---|
250 | pIter(p); |
---|
251 | idx++; |
---|
252 | } |
---|
253 | |
---|
254 | /* first find gcd of coeffs */ |
---|
255 | fmpq_zero(arg->content); |
---|
256 | while (idx < arg->end_idx) |
---|
257 | { |
---|
258 | my_convSingNFlintN_QQ(c, number(pGetCoeff(p))); |
---|
259 | fmpq_gcd(arg->content, arg->content, c); |
---|
260 | pIter(p); |
---|
261 | idx++; |
---|
262 | } |
---|
263 | |
---|
264 | fmpq_clear(c); |
---|
265 | } |
---|
266 | |
---|
267 | static void convert_sing_to_fmpq_mpoly_zpoly_worker(void * varg) |
---|
268 | { |
---|
269 | convert_sing_to_fmpq_mpoly_arg * arg = (convert_sing_to_fmpq_mpoly_arg *) varg; |
---|
270 | convert_sing_to_fmpq_mpoly_base * base = arg->base; |
---|
271 | ulong * exp = (ulong*) flint_malloc((base->r->N + 1)*sizeof(ulong)); |
---|
272 | fmpq_t c, t; |
---|
273 | fmpq_init(c); |
---|
274 | fmpq_init(t); |
---|
275 | |
---|
276 | slong idx = arg->start_idx/4096; |
---|
277 | poly p = base->markers[idx]; |
---|
278 | idx *= 4096; |
---|
279 | while (idx < arg->start_idx) |
---|
280 | { |
---|
281 | pIter(p); |
---|
282 | idx++; |
---|
283 | } |
---|
284 | |
---|
285 | while (idx < arg->end_idx) |
---|
286 | { |
---|
287 | my_convSingNFlintN_QQ(c, number(pGetCoeff(p))); |
---|
288 | FLINT_ASSERT(!fmpq_is_zero(base->res->content)); |
---|
289 | fmpq_div(t, c, base->res->content); |
---|
290 | FLINT_ASSERT(fmpz_is_one(fmpq_denref(t))); |
---|
291 | |
---|
292 | slong N = mpoly_words_per_exp(base->res->zpoly->bits, base->ctx->zctx->minfo); |
---|
293 | fmpz_swap(base->res->zpoly->coeffs + idx, fmpq_numref(t)); |
---|
294 | #if SIZEOF_LONG==8 |
---|
295 | p_GetExpVL(p, (int64*)exp, base->r); |
---|
296 | mpoly_set_monomial_ui(base->res->zpoly->exps + N*idx, exp, base->res->zpoly->bits, base->ctx->zctx->minfo); |
---|
297 | #else |
---|
298 | p_GetExpV(p, (int*)exp, base->r); |
---|
299 | mpoly_set_monomial_ui(base->res->zpoly->exps + N*idx, &(exp[1]), base->res->zpoly->bits, base->ctx->minfo); |
---|
300 | #endif |
---|
301 | |
---|
302 | pIter(p); |
---|
303 | idx++; |
---|
304 | } |
---|
305 | |
---|
306 | flint_free(exp); |
---|
307 | fmpq_clear(c); |
---|
308 | fmpq_clear(t); |
---|
309 | } |
---|
310 | |
---|
311 | /* res comes in unitialized!!!! */ |
---|
312 | void convSingPFlintMP(fmpq_mpoly_t res, fmpq_mpoly_ctx_t ctx, poly p, int lp, const ring r) |
---|
313 | { |
---|
314 | thread_pool_handle * handles; |
---|
315 | slong num_handles; |
---|
316 | slong thread_limit = 1000; |
---|
317 | |
---|
318 | /* get workers */ |
---|
319 | handles = NULL; |
---|
320 | num_handles = 0; |
---|
321 | if (thread_limit > 1 && global_thread_pool_initialized) |
---|
322 | { |
---|
323 | slong max_num_handles = thread_pool_get_size(global_thread_pool); |
---|
324 | max_num_handles = FLINT_MIN(thread_limit - 1, max_num_handles); |
---|
325 | if (max_num_handles > 0) |
---|
326 | { |
---|
327 | handles = (thread_pool_handle *) flint_malloc(max_num_handles*sizeof(thread_pool_handle)); |
---|
328 | num_handles = thread_pool_request(global_thread_pool, handles, max_num_handles); |
---|
329 | } |
---|
330 | } |
---|
331 | |
---|
332 | convert_sing_to_fmpq_mpoly_base base(num_handles + 1, res, ctx, r); |
---|
333 | base.init_poly(p); |
---|
334 | |
---|
335 | convert_sing_to_fmpq_mpoly_arg * args = new convert_sing_to_fmpq_mpoly_arg[base.num_threads]; |
---|
336 | slong cur_idx = 0; |
---|
337 | for (slong i = 0; i < base.num_threads; i++) |
---|
338 | { |
---|
339 | slong next_idx = i + 1 < base.num_threads ? (i + 1)*base.length/base.num_threads : base.length; |
---|
340 | next_idx = FLINT_MAX(next_idx, cur_idx); |
---|
341 | next_idx = FLINT_MIN(next_idx, base.length); |
---|
342 | args[i].base = &base; |
---|
343 | args[i].start_idx = cur_idx; |
---|
344 | args[i].end_idx = next_idx; |
---|
345 | cur_idx = next_idx; |
---|
346 | } |
---|
347 | |
---|
348 | /* get content */ |
---|
349 | for (slong i = 0; i < num_handles; i++) |
---|
350 | thread_pool_wake(global_thread_pool, handles[i], convert_sing_to_fmpq_mpoly_content_worker, args + i); |
---|
351 | convert_sing_to_fmpq_mpoly_content_worker(args + num_handles); |
---|
352 | for (slong i = 0; i < num_handles; i++) |
---|
353 | { |
---|
354 | thread_pool_wait(global_thread_pool, handles[i]); |
---|
355 | } |
---|
356 | |
---|
357 | /* sign of content should match sign of first coeff */ |
---|
358 | fmpq_zero(base.res->content); |
---|
359 | for (slong i = 0; i < base.num_threads; i++) |
---|
360 | fmpq_gcd(base.res->content, base.res->content, args[i].content); |
---|
361 | if (p != NULL) |
---|
362 | { |
---|
363 | fmpq_t c; |
---|
364 | fmpq_init(c); |
---|
365 | my_convSingNFlintN_QQ(c, number(pGetCoeff(p))); |
---|
366 | if (fmpq_sgn(c) < 0) |
---|
367 | fmpq_neg(base.res->content, base.res->content); |
---|
368 | fmpq_clear(c); |
---|
369 | } |
---|
370 | |
---|
371 | /* fill in res->zpoly */ |
---|
372 | for (slong i = 0; i < num_handles; i++) |
---|
373 | thread_pool_wake(global_thread_pool, handles[i], convert_sing_to_fmpq_mpoly_zpoly_worker, args + i); |
---|
374 | convert_sing_to_fmpq_mpoly_zpoly_worker(args + num_handles); |
---|
375 | for (slong i = 0; i < num_handles; i++) |
---|
376 | thread_pool_wait(global_thread_pool, handles[i]); |
---|
377 | |
---|
378 | base.res->zpoly->length = base.length; |
---|
379 | |
---|
380 | delete[] args; |
---|
381 | |
---|
382 | for (slong i = 0; i < num_handles; i++) |
---|
383 | thread_pool_give_back(global_thread_pool, handles[i]); |
---|
384 | if (handles) |
---|
385 | flint_free(handles); |
---|
386 | } |
---|
387 | |
---|
388 | /* fmpq_mpoly -> singular conversion */ |
---|
389 | |
---|
390 | class convert_fmpq_mpoly_to_sing_base |
---|
391 | { |
---|
392 | public: |
---|
393 | slong num_threads; |
---|
394 | fmpq_mpoly_struct * f; |
---|
395 | const fmpq_mpoly_ctx_struct * ctx; |
---|
396 | ring r; |
---|
397 | |
---|
398 | convert_fmpq_mpoly_to_sing_base(slong num_threads_, fmpq_mpoly_struct * f_, |
---|
399 | const fmpq_mpoly_ctx_struct * ctx_, const ring r_) |
---|
400 | : num_threads(num_threads_), |
---|
401 | f(f_), |
---|
402 | ctx(ctx_), |
---|
403 | r(r_) |
---|
404 | { |
---|
405 | } |
---|
406 | }; |
---|
407 | |
---|
408 | class convert_fmpq_mpoly_to_sing_arg |
---|
409 | { |
---|
410 | public: |
---|
411 | poly poly_start, poly_end; |
---|
412 | slong start_idx, end_idx; |
---|
413 | convert_fmpq_mpoly_to_sing_base* base; |
---|
414 | }; |
---|
415 | |
---|
416 | static void convert_fmpq_mpoly_to_sing_worker(void * varg) |
---|
417 | { |
---|
418 | convert_fmpq_mpoly_to_sing_arg * arg = (convert_fmpq_mpoly_to_sing_arg *) varg; |
---|
419 | convert_fmpq_mpoly_to_sing_base * base = arg->base; |
---|
420 | ulong* exp = (ulong*) flint_calloc(base->r->N + 1, sizeof(ulong)); |
---|
421 | fmpq_t c; |
---|
422 | fmpq_init(c); |
---|
423 | |
---|
424 | for (slong idx = arg->end_idx - 1; idx >= arg->start_idx; idx--) |
---|
425 | { |
---|
426 | poly pp = p_Init(base->r); |
---|
427 | |
---|
428 | #if SIZEOF_LONG==8 |
---|
429 | fmpq_mpoly_get_term_exp_ui(exp, base->f, idx, base->ctx); |
---|
430 | p_SetExpVL(pp, (int64*)exp, base->r); |
---|
431 | #else |
---|
432 | fmpq_mpoly_get_term_exp_ui(&(exp[1]), base->f, idx, base->ctx); |
---|
433 | p_SetExpV(pp, (int*)exp, base->r); |
---|
434 | #endif |
---|
435 | p_Setm(pp, base->r); |
---|
436 | |
---|
437 | fmpq_mpoly_get_term_coeff_fmpq(c, base->f, idx, base->ctx); |
---|
438 | pSetCoeff0(pp, number(convFlintNSingN_QQ(c, base->r->cf))); |
---|
439 | |
---|
440 | if (idx == arg->end_idx - 1) |
---|
441 | arg->poly_end = pp; |
---|
442 | |
---|
443 | pNext(pp) = arg->poly_start; |
---|
444 | arg->poly_start = pp; |
---|
445 | } |
---|
446 | |
---|
447 | flint_free(exp); |
---|
448 | fmpq_clear(c); |
---|
449 | } |
---|
450 | |
---|
451 | poly convFlintMPSingP(fmpq_mpoly_t f, fmpq_mpoly_ctx_t ctx, const ring r) |
---|
452 | { |
---|
453 | thread_pool_handle * handles; |
---|
454 | slong num_handles; |
---|
455 | slong thread_limit = 1000; |
---|
456 | |
---|
457 | /* get workers */ |
---|
458 | handles = NULL; |
---|
459 | num_handles = 0; |
---|
460 | if (thread_limit > 1 && global_thread_pool_initialized) |
---|
461 | { |
---|
462 | slong max_num_handles = thread_pool_get_size(global_thread_pool); |
---|
463 | max_num_handles = FLINT_MIN(thread_limit - 1, max_num_handles); |
---|
464 | if (max_num_handles > 0) |
---|
465 | { |
---|
466 | handles = (thread_pool_handle *) flint_malloc(max_num_handles*sizeof(thread_pool_handle)); |
---|
467 | num_handles = thread_pool_request(global_thread_pool, handles, max_num_handles); |
---|
468 | } |
---|
469 | } |
---|
470 | |
---|
471 | convert_fmpq_mpoly_to_sing_base base(num_handles + 1, f, ctx, r); |
---|
472 | |
---|
473 | convert_fmpq_mpoly_to_sing_arg * args = new convert_fmpq_mpoly_to_sing_arg[base.num_threads]; |
---|
474 | slong cur_idx = 0; |
---|
475 | for (slong i = 0; i < base.num_threads; i++) |
---|
476 | { |
---|
477 | slong next_idx = i + 1 < base.num_threads ? (i + 1)*base.f->zpoly->length/base.num_threads |
---|
478 | : base.f->zpoly->length; |
---|
479 | next_idx = FLINT_MAX(next_idx, cur_idx); |
---|
480 | next_idx = FLINT_MIN(next_idx, base.f->zpoly->length); |
---|
481 | args[i].base = &base; |
---|
482 | args[i].start_idx = cur_idx; |
---|
483 | args[i].end_idx = next_idx; |
---|
484 | cur_idx = next_idx; |
---|
485 | args[i].poly_end = NULL; |
---|
486 | args[i].poly_start = NULL; |
---|
487 | } |
---|
488 | |
---|
489 | /* construct pieces */ |
---|
490 | for (slong i = 0; i < num_handles; i++) |
---|
491 | thread_pool_wake(global_thread_pool, handles[i], convert_fmpq_mpoly_to_sing_worker, args + i); |
---|
492 | convert_fmpq_mpoly_to_sing_worker(args + num_handles); |
---|
493 | for (slong i = 0; i < num_handles; i++) |
---|
494 | thread_pool_wait(global_thread_pool, handles[i]); |
---|
495 | |
---|
496 | /* join pieces */ |
---|
497 | poly p = NULL; |
---|
498 | for (slong i = num_handles; i >= 0; i--) |
---|
499 | { |
---|
500 | if (args[i].poly_end != NULL) |
---|
501 | { |
---|
502 | pNext(args[i].poly_end) = p; |
---|
503 | p = args[i].poly_start; |
---|
504 | } |
---|
505 | } |
---|
506 | |
---|
507 | for (slong i = 0; i < num_handles; i++) |
---|
508 | thread_pool_give_back(global_thread_pool, handles[i]); |
---|
509 | if (handles) |
---|
510 | flint_free(handles); |
---|
511 | |
---|
512 | p_Test(p,r); |
---|
513 | return p; |
---|
514 | } |
---|
515 | |
---|
516 | |
---|
517 | /* singular -> nmod_mpoly conversion */ |
---|
518 | |
---|
519 | class convert_sing_to_nmod_mpoly_base |
---|
520 | { |
---|
521 | public: |
---|
522 | slong num_threads; |
---|
523 | slong length; |
---|
524 | nmod_mpoly_struct * res; |
---|
525 | const nmod_mpoly_ctx_struct * ctx; |
---|
526 | std::vector<poly> markers; |
---|
527 | ring r; |
---|
528 | |
---|
529 | convert_sing_to_nmod_mpoly_base(slong num_threads_, nmod_mpoly_struct * res_, |
---|
530 | const nmod_mpoly_ctx_struct * ctx_, const ring r_) |
---|
531 | : num_threads(num_threads_), |
---|
532 | res(res_), |
---|
533 | ctx(ctx_), |
---|
534 | r(r_) |
---|
535 | { |
---|
536 | } |
---|
537 | |
---|
538 | void init_poly(poly p) |
---|
539 | { |
---|
540 | length = 0; |
---|
541 | while (p != NULL) |
---|
542 | { |
---|
543 | markers.push_back(p); |
---|
544 | for (int i = 4096; i > 0; i--) |
---|
545 | { |
---|
546 | pIter(p); |
---|
547 | length++; |
---|
548 | if (p == NULL) |
---|
549 | break; |
---|
550 | } |
---|
551 | } |
---|
552 | |
---|
553 | nmod_mpoly_init3(res, length, SI_LOG2(r->bitmask), ctx); |
---|
554 | } |
---|
555 | }; |
---|
556 | |
---|
557 | class convert_sing_to_nmod_mpoly_arg |
---|
558 | { |
---|
559 | public: |
---|
560 | slong start_idx, end_idx; |
---|
561 | convert_sing_to_nmod_mpoly_base* base; |
---|
562 | }; |
---|
563 | |
---|
564 | static void convert_sing_to_nmod_mpoly_worker(void * varg) |
---|
565 | { |
---|
566 | convert_sing_to_nmod_mpoly_arg * arg = (convert_sing_to_nmod_mpoly_arg *) varg; |
---|
567 | convert_sing_to_nmod_mpoly_base * base = arg->base; |
---|
568 | ulong * exp = (ulong*) flint_malloc((base->r->N + 1)*sizeof(ulong)); |
---|
569 | |
---|
570 | slong idx = arg->start_idx/4096; |
---|
571 | poly p = base->markers[idx]; |
---|
572 | idx *= 4096; |
---|
573 | while (idx < arg->start_idx) |
---|
574 | { |
---|
575 | pIter(p); |
---|
576 | idx++; |
---|
577 | } |
---|
578 | |
---|
579 | while (idx < arg->end_idx) |
---|
580 | { |
---|
581 | slong N = mpoly_words_per_exp(base->res->bits, base->ctx->minfo); |
---|
582 | #if SIZEOF_LONG==8 |
---|
583 | p_GetExpVL(p, (int64*)exp, base->r); |
---|
584 | mpoly_set_monomial_ui(base->res->exps + N*idx, exp, base->res->bits, base->ctx->minfo); |
---|
585 | #else |
---|
586 | p_GetExpV(p, (int*)exp, base->r); |
---|
587 | mpoly_set_monomial_ui(base->res->exps + N*idx, &(exp[1]), base->res->bits, base->ctx->minfo); |
---|
588 | #endif |
---|
589 | |
---|
590 | base->res->coeffs[idx] = (ulong)(number(pGetCoeff(p))); |
---|
591 | |
---|
592 | pIter(p); |
---|
593 | idx++; |
---|
594 | } |
---|
595 | |
---|
596 | flint_free(exp); |
---|
597 | } |
---|
598 | |
---|
599 | /* res comes in unitialized!!!! */ |
---|
600 | void convSingPFlintMP(nmod_mpoly_t res, nmod_mpoly_ctx_t ctx, poly p, int lp, const ring r) |
---|
601 | { |
---|
602 | thread_pool_handle * handles; |
---|
603 | slong num_handles; |
---|
604 | slong thread_limit = 1000; |
---|
605 | |
---|
606 | /* get workers */ |
---|
607 | handles = NULL; |
---|
608 | num_handles = 0; |
---|
609 | if (thread_limit > 1 && global_thread_pool_initialized) |
---|
610 | { |
---|
611 | slong max_num_handles = thread_pool_get_size(global_thread_pool); |
---|
612 | max_num_handles = FLINT_MIN(thread_limit - 1, max_num_handles); |
---|
613 | if (max_num_handles > 0) |
---|
614 | { |
---|
615 | handles = (thread_pool_handle *) flint_malloc(max_num_handles*sizeof(thread_pool_handle)); |
---|
616 | num_handles = thread_pool_request(global_thread_pool, handles, max_num_handles); |
---|
617 | } |
---|
618 | } |
---|
619 | |
---|
620 | convert_sing_to_nmod_mpoly_base base(num_handles + 1, res, ctx, r); |
---|
621 | base.init_poly(p); |
---|
622 | |
---|
623 | convert_sing_to_nmod_mpoly_arg * args = new convert_sing_to_nmod_mpoly_arg[base.num_threads]; |
---|
624 | slong cur_idx = 0; |
---|
625 | for (slong i = 0; i < base.num_threads; i++) |
---|
626 | { |
---|
627 | slong next_idx = i + 1 < base.num_threads ? (i + 1)*base.length/base.num_threads |
---|
628 | : base.length; |
---|
629 | next_idx = FLINT_MAX(next_idx, cur_idx); |
---|
630 | next_idx = FLINT_MIN(next_idx, base.length); |
---|
631 | args[i].base = &base; |
---|
632 | args[i].start_idx = cur_idx; |
---|
633 | args[i].end_idx = next_idx; |
---|
634 | cur_idx = next_idx; |
---|
635 | } |
---|
636 | |
---|
637 | /* fill in res */ |
---|
638 | for (slong i = 0; i < num_handles; i++) |
---|
639 | thread_pool_wake(global_thread_pool, handles[i], convert_sing_to_nmod_mpoly_worker, args + i); |
---|
640 | convert_sing_to_nmod_mpoly_worker(args + num_handles); |
---|
641 | for (slong i = 0; i < num_handles; i++) |
---|
642 | thread_pool_wait(global_thread_pool, handles[i]); |
---|
643 | |
---|
644 | base.res->length = base.length; |
---|
645 | |
---|
646 | delete[] args; |
---|
647 | |
---|
648 | for (slong i = 0; i < num_handles; i++) |
---|
649 | thread_pool_give_back(global_thread_pool, handles[i]); |
---|
650 | if (handles) |
---|
651 | flint_free(handles); |
---|
652 | } |
---|
653 | |
---|
654 | |
---|
655 | /* nmod_mpoly -> singular conversion */ |
---|
656 | |
---|
657 | class convert_nmod_mpoly_to_sing_base |
---|
658 | { |
---|
659 | public: |
---|
660 | slong num_threads; |
---|
661 | nmod_mpoly_struct * f; |
---|
662 | const nmod_mpoly_ctx_struct * ctx; |
---|
663 | ring r; |
---|
664 | |
---|
665 | convert_nmod_mpoly_to_sing_base(slong num_threads_, nmod_mpoly_struct * f_, |
---|
666 | const nmod_mpoly_ctx_struct * ctx_, const ring r_) |
---|
667 | : num_threads(num_threads_), |
---|
668 | f(f_), |
---|
669 | ctx(ctx_), |
---|
670 | r(r_) |
---|
671 | { |
---|
672 | } |
---|
673 | }; |
---|
674 | |
---|
675 | class convert_nmod_mpoly_to_sing_arg |
---|
676 | { |
---|
677 | public: |
---|
678 | poly poly_start, poly_end; |
---|
679 | slong start_idx, end_idx; |
---|
680 | convert_nmod_mpoly_to_sing_base* base; |
---|
681 | }; |
---|
682 | |
---|
683 | static void convert_nmod_mpoly_to_sing_worker(void * varg) |
---|
684 | { |
---|
685 | convert_nmod_mpoly_to_sing_arg * arg = (convert_nmod_mpoly_to_sing_arg *) varg; |
---|
686 | convert_nmod_mpoly_to_sing_base * base = arg->base; |
---|
687 | ulong* exp = (ulong*) flint_calloc(base->r->N + 1, sizeof(ulong)); |
---|
688 | |
---|
689 | for (slong idx = arg->end_idx - 1; idx >= arg->start_idx; idx--) |
---|
690 | { |
---|
691 | poly pp = p_Init(base->r); |
---|
692 | |
---|
693 | #if SIZEOF_LONG==8 |
---|
694 | nmod_mpoly_get_term_exp_ui(exp, base->f, idx, base->ctx); |
---|
695 | p_SetExpVL(pp, (int64*)exp, base->r); |
---|
696 | #else |
---|
697 | nmod_mpoly_get_term_exp_ui(&(exp[1]), base->f, idx, base->ctx); |
---|
698 | p_SetExpV(pp, (int*)exp, base->r); |
---|
699 | #endif |
---|
700 | p_Setm(pp, base->r); |
---|
701 | |
---|
702 | pSetCoeff0(pp, number(nmod_mpoly_get_term_coeff_ui(base->f, idx, base->ctx))); |
---|
703 | |
---|
704 | if (idx == arg->end_idx - 1) |
---|
705 | arg->poly_end = pp; |
---|
706 | |
---|
707 | pNext(pp) = arg->poly_start; |
---|
708 | arg->poly_start = pp; |
---|
709 | } |
---|
710 | |
---|
711 | flint_free(exp); |
---|
712 | } |
---|
713 | |
---|
714 | poly convFlintMPSingP(nmod_mpoly_t f, nmod_mpoly_ctx_t ctx, const ring r) |
---|
715 | { |
---|
716 | thread_pool_handle * handles; |
---|
717 | slong num_handles; |
---|
718 | slong thread_limit = 1000; |
---|
719 | |
---|
720 | /* get workers */ |
---|
721 | handles = NULL; |
---|
722 | num_handles = 0; |
---|
723 | if (thread_limit > 1 && global_thread_pool_initialized) |
---|
724 | { |
---|
725 | slong max_num_handles = thread_pool_get_size(global_thread_pool); |
---|
726 | max_num_handles = FLINT_MIN(thread_limit - 1, max_num_handles); |
---|
727 | if (max_num_handles > 0) |
---|
728 | { |
---|
729 | handles = (thread_pool_handle *) flint_malloc(max_num_handles*sizeof(thread_pool_handle)); |
---|
730 | num_handles = thread_pool_request(global_thread_pool, handles, max_num_handles); |
---|
731 | } |
---|
732 | } |
---|
733 | |
---|
734 | convert_nmod_mpoly_to_sing_base base(num_handles + 1, f, ctx, r); |
---|
735 | |
---|
736 | convert_nmod_mpoly_to_sing_arg * args = new convert_nmod_mpoly_to_sing_arg[base.num_threads]; |
---|
737 | slong cur_idx = 0; |
---|
738 | for (slong i = 0; i < base.num_threads; i++) |
---|
739 | { |
---|
740 | slong next_idx = i + 1 < base.num_threads ? (i + 1)*base.f->length/base.num_threads |
---|
741 | : base.f->length; |
---|
742 | next_idx = FLINT_MAX(next_idx, cur_idx); |
---|
743 | next_idx = FLINT_MIN(next_idx, base.f->length); |
---|
744 | args[i].base = &base; |
---|
745 | args[i].start_idx = cur_idx; |
---|
746 | args[i].end_idx = next_idx; |
---|
747 | cur_idx = next_idx; |
---|
748 | args[i].poly_end = NULL; |
---|
749 | args[i].poly_start = NULL; |
---|
750 | } |
---|
751 | |
---|
752 | /* construct pieces */ |
---|
753 | for (slong i = 0; i < num_handles; i++) |
---|
754 | thread_pool_wake(global_thread_pool, handles[i], convert_nmod_mpoly_to_sing_worker, args + i); |
---|
755 | convert_nmod_mpoly_to_sing_worker(args + num_handles); |
---|
756 | for (slong i = 0; i < num_handles; i++) |
---|
757 | thread_pool_wait(global_thread_pool, handles[i]); |
---|
758 | |
---|
759 | /* join pieces */ |
---|
760 | poly p = NULL; |
---|
761 | for (slong i = num_handles; i >= 0; i--) |
---|
762 | { |
---|
763 | if (args[i].poly_end != NULL) |
---|
764 | { |
---|
765 | pNext(args[i].poly_end) = p; |
---|
766 | p = args[i].poly_start; |
---|
767 | } |
---|
768 | } |
---|
769 | |
---|
770 | for (slong i = 0; i < num_handles; i++) |
---|
771 | thread_pool_give_back(global_thread_pool, handles[i]); |
---|
772 | if (handles) |
---|
773 | flint_free(handles); |
---|
774 | |
---|
775 | p_Test(p,r); |
---|
776 | return p; |
---|
777 | } |
---|
778 | |
---|
779 | #endif |
---|
780 | |
---|
781 | /****** polynomial operations ***********/ |
---|
782 | |
---|
783 | poly Flint_Mult_MP(poly p,int lp, poly q, int lq, fmpq_mpoly_ctx_t ctx, const ring r) |
---|
784 | { |
---|
785 | fmpq_mpoly_t pp,qq,res; |
---|
786 | convSingPFlintMP(pp,ctx,p,lp,r); |
---|
787 | convSingPFlintMP(qq,ctx,q,lq,r); |
---|
788 | fmpq_mpoly_init(res,ctx); |
---|
789 | fmpq_mpoly_mul(res,pp,qq,ctx); |
---|
790 | poly pres=convFlintMPSingP(res,ctx,r); |
---|
791 | fmpq_mpoly_clear(res,ctx); |
---|
792 | fmpq_mpoly_clear(pp,ctx); |
---|
793 | fmpq_mpoly_clear(qq,ctx); |
---|
794 | fmpq_mpoly_ctx_clear(ctx); |
---|
795 | p_Test(pres,r); |
---|
796 | return pres; |
---|
797 | } |
---|
798 | |
---|
799 | poly Flint_Mult_MP(poly p,int lp, poly q, int lq, nmod_mpoly_ctx_t ctx, const ring r) |
---|
800 | { |
---|
801 | nmod_mpoly_t pp,qq,res; |
---|
802 | convSingPFlintMP(pp,ctx,p,lp,r); |
---|
803 | convSingPFlintMP(qq,ctx,q,lq,r); |
---|
804 | nmod_mpoly_init(res,ctx); |
---|
805 | nmod_mpoly_mul(res,pp,qq,ctx); |
---|
806 | poly pres=convFlintMPSingP(res,ctx,r); |
---|
807 | nmod_mpoly_clear(res,ctx); |
---|
808 | nmod_mpoly_clear(pp,ctx); |
---|
809 | nmod_mpoly_clear(qq,ctx); |
---|
810 | nmod_mpoly_ctx_clear(ctx); |
---|
811 | p_Test(pres,r); |
---|
812 | return pres; |
---|
813 | } |
---|
814 | |
---|
815 | poly Flint_GCD_MP(poly p,int lp,poly q,int lq,nmod_mpoly_ctx_t ctx,const ring r) |
---|
816 | { |
---|
817 | nmod_mpoly_t pp,qq,res; |
---|
818 | convSingPFlintMP(pp,ctx,p,lp,r); |
---|
819 | convSingPFlintMP(qq,ctx,q,lq,r); |
---|
820 | nmod_mpoly_init(res,ctx); |
---|
821 | int ok=nmod_mpoly_gcd(res,pp,qq,ctx); |
---|
822 | poly pres; |
---|
823 | if (ok) |
---|
824 | { |
---|
825 | pres=convFlintMPSingP(res,ctx,r); |
---|
826 | p_Test(pres,r); |
---|
827 | } |
---|
828 | else |
---|
829 | { |
---|
830 | pres=p_One(r); |
---|
831 | } |
---|
832 | nmod_mpoly_clear(res,ctx); |
---|
833 | nmod_mpoly_clear(pp,ctx); |
---|
834 | nmod_mpoly_clear(qq,ctx); |
---|
835 | nmod_mpoly_ctx_clear(ctx); |
---|
836 | return pres; |
---|
837 | } |
---|
838 | |
---|
839 | poly Flint_GCD_MP(poly p,int lp,poly q,int lq,fmpq_mpoly_ctx_t ctx,const ring r) |
---|
840 | { |
---|
841 | fmpq_mpoly_t pp,qq,res; |
---|
842 | convSingPFlintMP(pp,ctx,p,lp,r); |
---|
843 | convSingPFlintMP(qq,ctx,q,lq,r); |
---|
844 | fmpq_mpoly_init(res,ctx); |
---|
845 | int ok=fmpq_mpoly_gcd(res,pp,qq,ctx); |
---|
846 | poly pres; |
---|
847 | if (ok) |
---|
848 | { |
---|
849 | // Flint normalizes the gcd to be monic. |
---|
850 | // Singular wants a gcd defined over ZZ that is primitive and has a positive leading coeff. |
---|
851 | if (!fmpq_mpoly_is_zero(res, ctx)) |
---|
852 | { |
---|
853 | fmpq_t content; |
---|
854 | fmpq_init(content); |
---|
855 | fmpq_mpoly_content(content, res, ctx); |
---|
856 | fmpq_mpoly_scalar_div_fmpq(res, res, content, ctx); |
---|
857 | fmpq_clear(content); |
---|
858 | } |
---|
859 | pres=convFlintMPSingP(res,ctx,r); |
---|
860 | p_Test(pres,r); |
---|
861 | } |
---|
862 | else |
---|
863 | { |
---|
864 | pres=p_One(r); |
---|
865 | } |
---|
866 | fmpq_mpoly_clear(res,ctx); |
---|
867 | fmpq_mpoly_clear(pp,ctx); |
---|
868 | fmpq_mpoly_clear(qq,ctx); |
---|
869 | fmpq_mpoly_ctx_clear(ctx); |
---|
870 | return pres; |
---|
871 | } |
---|
872 | |
---|
873 | #endif |
---|
874 | #endif |
---|