Details | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
5563 | serge | 1 | /* |
2 | * Copyright 2013 Vadim Girlin |
||
3 | * |
||
4 | * Permission is hereby granted, free of charge, to any person obtaining a |
||
5 | * copy of this software and associated documentation files (the "Software"), |
||
6 | * to deal in the Software without restriction, including without limitation |
||
7 | * on the rights to use, copy, modify, merge, publish, distribute, sub |
||
8 | * license, and/or sell copies of the Software, and to permit persons to whom |
||
9 | * the Software is furnished to do so, subject to the following conditions: |
||
10 | * |
||
11 | * The above copyright notice and this permission notice (including the next |
||
12 | * paragraph) shall be included in all copies or substantial portions of the |
||
13 | * Software. |
||
14 | * |
||
15 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
||
16 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
||
17 | * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL |
||
18 | * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, |
||
19 | * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR |
||
20 | * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE |
||
21 | * USE OR OTHER DEALINGS IN THE SOFTWARE. |
||
22 | * |
||
23 | * Authors: |
||
24 | * Vadim Girlin |
||
25 | */ |
||
26 | |||
27 | #ifndef SB_PASS_H_ |
||
28 | #define SB_PASS_H_ |
||
29 | |||
30 | #include |
||
31 | |||
32 | namespace r600_sb { |
||
33 | |||
34 | class pass { |
||
35 | protected: |
||
36 | sb_context &ctx; |
||
37 | shader &sh; |
||
38 | |||
39 | public: |
||
40 | pass(shader &s); |
||
41 | |||
42 | virtual int run(); |
||
43 | |||
44 | virtual ~pass() {} |
||
45 | }; |
||
46 | |||
47 | class vpass : public pass { |
||
48 | |||
49 | public: |
||
50 | |||
51 | vpass(shader &s) : pass(s) {} |
||
52 | |||
53 | virtual int init(); |
||
54 | virtual int done(); |
||
55 | |||
56 | virtual int run(); |
||
57 | virtual void run_on(container_node &n); |
||
58 | |||
59 | virtual bool visit(node &n, bool enter); |
||
60 | virtual bool visit(container_node &n, bool enter); |
||
61 | virtual bool visit(alu_group_node &n, bool enter); |
||
62 | virtual bool visit(cf_node &n, bool enter); |
||
63 | virtual bool visit(alu_node &n, bool enter); |
||
64 | virtual bool visit(alu_packed_node &n, bool enter); |
||
65 | virtual bool visit(fetch_node &n, bool enter); |
||
66 | virtual bool visit(region_node &n, bool enter); |
||
67 | virtual bool visit(repeat_node &n, bool enter); |
||
68 | virtual bool visit(depart_node &n, bool enter); |
||
69 | virtual bool visit(if_node &n, bool enter); |
||
70 | virtual bool visit(bb_node &n, bool enter); |
||
71 | |||
72 | }; |
||
73 | |||
74 | class rev_vpass : public vpass { |
||
75 | |||
76 | public: |
||
77 | rev_vpass(shader &s) : vpass(s) {} |
||
78 | |||
79 | virtual void run_on(container_node &n); |
||
80 | }; |
||
81 | |||
82 | |||
83 | // =================== PASSES |
||
84 | |||
85 | class bytecode; |
||
86 | |||
87 | class bc_dump : public vpass { |
||
88 | using vpass::visit; |
||
89 | |||
90 | uint32_t *bc_data; |
||
91 | unsigned ndw; |
||
92 | |||
93 | unsigned id; |
||
94 | |||
95 | unsigned new_group, group_index; |
||
96 | |||
97 | public: |
||
98 | |||
99 | bc_dump(shader &s, bytecode *bc = NULL); |
||
100 | |||
101 | bc_dump(shader &s, uint32_t *bc_ptr, unsigned ndw) : |
||
102 | vpass(s), bc_data(bc_ptr), ndw(ndw), id(), new_group(), group_index() {} |
||
103 | |||
104 | virtual int init(); |
||
105 | virtual int done(); |
||
106 | |||
107 | virtual bool visit(cf_node &n, bool enter); |
||
108 | virtual bool visit(alu_node &n, bool enter); |
||
109 | virtual bool visit(fetch_node &n, bool enter); |
||
110 | |||
111 | void dump_dw(unsigned dw_id, unsigned count = 2); |
||
112 | |||
113 | void dump(cf_node& n); |
||
114 | void dump(alu_node& n); |
||
115 | void dump(fetch_node& n); |
||
116 | }; |
||
117 | |||
118 | |||
119 | class dce_cleanup : public vpass { |
||
120 | using vpass::visit; |
||
121 | |||
122 | public: |
||
123 | |||
124 | dce_cleanup(shader &s) : vpass(s) {} |
||
125 | |||
126 | virtual bool visit(node &n, bool enter); |
||
127 | virtual bool visit(alu_group_node &n, bool enter); |
||
128 | virtual bool visit(cf_node &n, bool enter); |
||
129 | virtual bool visit(alu_node &n, bool enter); |
||
130 | virtual bool visit(alu_packed_node &n, bool enter); |
||
131 | virtual bool visit(fetch_node &n, bool enter); |
||
132 | virtual bool visit(region_node &n, bool enter); |
||
133 | virtual bool visit(container_node &n, bool enter); |
||
134 | |||
135 | private: |
||
136 | |||
137 | void cleanup_dst(node &n); |
||
138 | void cleanup_dst_vec(vvec &vv); |
||
139 | |||
140 | }; |
||
141 | |||
142 | |||
143 | class def_use : public pass { |
||
144 | |||
145 | public: |
||
146 | |||
147 | def_use(shader &sh) : pass(sh) {} |
||
148 | |||
149 | virtual int run(); |
||
150 | void run_on(node *n, bool defs); |
||
151 | |||
152 | private: |
||
153 | |||
154 | void process_uses(node *n); |
||
155 | void process_defs(node *n, vvec &vv, bool arr_def); |
||
156 | void process_phi(container_node *c, bool defs, bool uses); |
||
157 | }; |
||
158 | |||
159 | |||
160 | |||
161 | class dump : public vpass { |
||
162 | using vpass::visit; |
||
163 | |||
164 | int level; |
||
165 | |||
166 | public: |
||
167 | |||
168 | dump(shader &s) : vpass(s), level(0) {} |
||
169 | |||
170 | virtual bool visit(node &n, bool enter); |
||
171 | virtual bool visit(container_node &n, bool enter); |
||
172 | virtual bool visit(alu_group_node &n, bool enter); |
||
173 | virtual bool visit(cf_node &n, bool enter); |
||
174 | virtual bool visit(alu_node &n, bool enter); |
||
175 | virtual bool visit(alu_packed_node &n, bool enter); |
||
176 | virtual bool visit(fetch_node &n, bool enter); |
||
177 | virtual bool visit(region_node &n, bool enter); |
||
178 | virtual bool visit(repeat_node &n, bool enter); |
||
179 | virtual bool visit(depart_node &n, bool enter); |
||
180 | virtual bool visit(if_node &n, bool enter); |
||
181 | virtual bool visit(bb_node &n, bool enter); |
||
182 | |||
183 | |||
184 | static void dump_op(node &n, const char *name); |
||
185 | static void dump_vec(const vvec & vv); |
||
186 | static void dump_set(shader &sh, val_set & v); |
||
187 | |||
188 | static void dump_rels(vvec & vv); |
||
189 | |||
190 | static void dump_val(value *v); |
||
191 | static void dump_op(node *n); |
||
192 | |||
193 | static void dump_op_list(container_node *c); |
||
194 | static void dump_queue(sched_queue &q); |
||
195 | |||
196 | static void dump_alu(alu_node *n); |
||
197 | |||
198 | private: |
||
199 | |||
200 | void indent(); |
||
201 | |||
202 | void dump_common(node &n); |
||
203 | void dump_flags(node &n); |
||
204 | |||
205 | void dump_live_values(container_node &n, bool before); |
||
206 | }; |
||
207 | |||
208 | |||
209 | // Global Code Motion |
||
210 | |||
211 | class gcm : public pass { |
||
212 | |||
213 | sched_queue bu_ready[SQ_NUM]; |
||
214 | sched_queue bu_ready_next[SQ_NUM]; |
||
215 | sched_queue bu_ready_early[SQ_NUM]; |
||
216 | sched_queue ready; |
||
217 | sched_queue ready_above; |
||
218 | |||
219 | container_node pending; |
||
220 | |||
221 | struct op_info { |
||
222 | bb_node* top_bb; |
||
223 | bb_node* bottom_bb; |
||
224 | op_info() : top_bb(), bottom_bb() {} |
||
225 | }; |
||
226 | |||
227 | typedef std::map |
||
228 | |||
229 | typedef std::map |
||
230 | |||
231 | op_info_map op_map; |
||
232 | nuc_map uses; |
||
233 | |||
234 | typedef std::vector |
||
235 | |||
236 | nuc_stack nuc_stk; |
||
237 | unsigned ucs_level; |
||
238 | |||
239 | bb_node * bu_bb; |
||
240 | |||
241 | vvec pending_defs; |
||
242 | |||
243 | node_list pending_nodes; |
||
244 | |||
245 | unsigned cur_sq; |
||
246 | |||
247 | // for register pressure tracking in bottom-up pass |
||
248 | val_set live; |
||
249 | int live_count; |
||
250 | |||
251 | static const int rp_threshold = 100; |
||
252 | |||
253 | bool pending_exec_mask_update; |
||
254 | |||
255 | public: |
||
256 | |||
257 | gcm(shader &sh) : pass(sh), |
||
258 | bu_ready(), bu_ready_next(), bu_ready_early(), |
||
259 | ready(), op_map(), uses(), nuc_stk(1), ucs_level(), |
||
260 | bu_bb(), pending_defs(), pending_nodes(), cur_sq(), |
||
261 | live(), live_count(), pending_exec_mask_update() {} |
||
262 | |||
263 | virtual int run(); |
||
264 | |||
265 | private: |
||
266 | |||
267 | void collect_instructions(container_node *c, bool early_pass); |
||
268 | |||
269 | void sched_early(container_node *n); |
||
270 | void td_sched_bb(bb_node *bb); |
||
271 | bool td_is_ready(node *n); |
||
272 | void td_release_uses(vvec &v); |
||
273 | void td_release_val(value *v); |
||
274 | void td_schedule(bb_node *bb, node *n); |
||
275 | |||
276 | void sched_late(container_node *n); |
||
277 | void bu_sched_bb(bb_node *bb); |
||
278 | void bu_release_defs(vvec &v, bool src); |
||
279 | void bu_release_phi_defs(container_node *p, unsigned op); |
||
280 | bool bu_is_ready(node *n); |
||
281 | void bu_release_val(value *v); |
||
282 | void bu_release_op(node * n); |
||
283 | void bu_find_best_bb(node *n, op_info &oi); |
||
284 | void bu_schedule(container_node *bb, node *n); |
||
285 | |||
286 | void push_uc_stack(); |
||
287 | void pop_uc_stack(); |
||
288 | |||
289 | void init_def_count(nuc_map &m, container_node &s); |
||
290 | void init_use_count(nuc_map &m, container_node &s); |
||
291 | unsigned get_uc_vec(vvec &vv); |
||
292 | unsigned get_dc_vec(vvec &vv, bool src); |
||
293 | |||
294 | void add_ready(node *n); |
||
295 | |||
296 | void dump_uc_stack(); |
||
297 | |||
298 | unsigned real_alu_count(sched_queue &q, unsigned max); |
||
299 | |||
300 | // check if we have not less than threshold ready alu instructions |
||
301 | bool check_alu_ready_count(unsigned threshold); |
||
302 | }; |
||
303 | |||
304 | |||
305 | class gvn : public vpass { |
||
306 | using vpass::visit; |
||
307 | |||
308 | public: |
||
309 | |||
310 | gvn(shader &sh) : vpass(sh) {} |
||
311 | |||
312 | virtual bool visit(node &n, bool enter); |
||
313 | virtual bool visit(cf_node &n, bool enter); |
||
314 | virtual bool visit(alu_node &n, bool enter); |
||
315 | virtual bool visit(alu_packed_node &n, bool enter); |
||
316 | virtual bool visit(fetch_node &n, bool enter); |
||
317 | virtual bool visit(region_node &n, bool enter); |
||
318 | |||
319 | private: |
||
320 | |||
321 | void process_op(node &n, bool rewrite = true); |
||
322 | |||
323 | // returns true if the value was rewritten |
||
324 | bool process_src(value* &v, bool rewrite); |
||
325 | |||
326 | |||
327 | void process_alu_src_constants(node &n, value* &v); |
||
328 | }; |
||
329 | |||
330 | |||
331 | class if_conversion : public pass { |
||
332 | |||
333 | public: |
||
334 | |||
335 | if_conversion(shader &sh) : pass(sh) {} |
||
336 | |||
337 | virtual int run(); |
||
338 | |||
339 | bool run_on(region_node *r); |
||
340 | |||
341 | void convert_kill_instructions(region_node *r, value *em, bool branch, |
||
342 | container_node *c); |
||
343 | |||
344 | bool check_and_convert(region_node *r); |
||
345 | |||
346 | alu_node* convert_phi(value *select, node *phi); |
||
347 | |||
348 | }; |
||
349 | |||
350 | |||
351 | class liveness : public rev_vpass { |
||
352 | using vpass::visit; |
||
353 | |||
354 | val_set live; |
||
355 | bool live_changed; |
||
356 | |||
357 | public: |
||
358 | |||
359 | liveness(shader &s) : rev_vpass(s), live_changed(false) {} |
||
360 | |||
361 | virtual int init(); |
||
362 | |||
363 | virtual bool visit(node &n, bool enter); |
||
364 | virtual bool visit(bb_node &n, bool enter); |
||
365 | virtual bool visit(container_node &n, bool enter); |
||
366 | virtual bool visit(alu_group_node &n, bool enter); |
||
367 | virtual bool visit(cf_node &n, bool enter); |
||
368 | virtual bool visit(alu_node &n, bool enter); |
||
369 | virtual bool visit(alu_packed_node &n, bool enter); |
||
370 | virtual bool visit(fetch_node &n, bool enter); |
||
371 | virtual bool visit(region_node &n, bool enter); |
||
372 | virtual bool visit(repeat_node &n, bool enter); |
||
373 | virtual bool visit(depart_node &n, bool enter); |
||
374 | virtual bool visit(if_node &n, bool enter); |
||
375 | |||
376 | private: |
||
377 | |||
378 | void update_interferences(); |
||
379 | void process_op(node &n); |
||
380 | |||
381 | bool remove_val(value *v); |
||
382 | bool remove_vec(vvec &v); |
||
383 | bool process_outs(node& n); |
||
384 | void process_ins(node& n); |
||
385 | |||
386 | void process_phi_outs(container_node *phi); |
||
387 | void process_phi_branch(container_node *phi, unsigned id); |
||
388 | |||
389 | bool process_maydef(value *v); |
||
390 | |||
391 | bool add_vec(vvec &vv, bool src); |
||
392 | |||
393 | void update_src_vec(vvec &vv, bool src); |
||
394 | }; |
||
395 | |||
396 | |||
397 | struct bool_op_info { |
||
398 | bool invert; |
||
399 | unsigned int_cvt; |
||
400 | |||
401 | alu_node *n; |
||
402 | }; |
||
403 | |||
404 | class peephole : public pass { |
||
405 | |||
406 | public: |
||
407 | |||
408 | peephole(shader &sh) : pass(sh) {} |
||
409 | |||
410 | virtual int run(); |
||
411 | |||
412 | void run_on(container_node *c); |
||
413 | |||
414 | void optimize_cc_op(alu_node *a); |
||
415 | |||
416 | void optimize_cc_op2(alu_node *a); |
||
417 | void optimize_CNDcc_op(alu_node *a); |
||
418 | |||
419 | bool get_bool_op_info(value *b, bool_op_info& bop); |
||
420 | bool get_bool_flt_to_int_source(alu_node* &a); |
||
421 | void convert_float_setcc(alu_node *f2i, alu_node *s); |
||
422 | }; |
||
423 | |||
424 | |||
425 | class psi_ops : public rev_vpass { |
||
426 | using rev_vpass::visit; |
||
427 | |||
428 | public: |
||
429 | |||
430 | psi_ops(shader &s) : rev_vpass(s) {} |
||
431 | |||
432 | virtual bool visit(node &n, bool enter); |
||
433 | virtual bool visit(alu_node &n, bool enter); |
||
434 | |||
435 | bool try_inline(node &n); |
||
436 | bool try_reduce(node &n); |
||
437 | bool eliminate(node &n); |
||
438 | |||
439 | void unpredicate(node *n); |
||
440 | }; |
||
441 | |||
442 | |||
443 | // check correctness of the generated code, e.g.: |
||
444 | // - expected source operand value is the last value written to its gpr, |
||
445 | // - all arguments of phi node should be allocated to the same gpr, |
||
446 | // TODO other tests |
||
447 | class ra_checker : public pass { |
||
448 | |||
449 | typedef std::map |
||
450 | |||
451 | typedef std::vector |
||
452 | |||
453 | regmap_stack rm_stack; |
||
454 | unsigned rm_stk_level; |
||
455 | |||
456 | value* prev_dst[5]; |
||
457 | |||
458 | public: |
||
459 | |||
460 | ra_checker(shader &sh) : pass(sh), rm_stk_level(0), prev_dst() {} |
||
461 | |||
462 | virtual int run(); |
||
463 | |||
464 | void run_on(container_node *c); |
||
465 | |||
466 | void dump_error(const error_info &e); |
||
467 | void dump_all_errors(); |
||
468 | |||
469 | private: |
||
470 | |||
471 | reg_value_map& rmap() { return rm_stack[rm_stk_level]; } |
||
472 | |||
473 | void push_stack(); |
||
474 | void pop_stack(); |
||
475 | |||
476 | // when going out of the alu clause, values in the clause temporary gprs, |
||
477 | // AR, predicate values, PS/PV are destroyed |
||
478 | void kill_alu_only_regs(); |
||
479 | void error(node *n, unsigned id, std::string msg); |
||
480 | |||
481 | void check_phi_src(container_node *p, unsigned id); |
||
482 | void process_phi_dst(container_node *p); |
||
483 | void check_alu_group(alu_group_node *g); |
||
484 | void process_op_dst(node *n); |
||
485 | void check_op_src(node *n); |
||
486 | void check_src_vec(node *n, unsigned id, vvec &vv, bool src); |
||
487 | void check_value_gpr(node *n, unsigned id, value *v); |
||
488 | }; |
||
489 | |||
490 | // ======================================= |
||
491 | |||
492 | |||
493 | class ra_coalesce : public pass { |
||
494 | |||
495 | public: |
||
496 | |||
497 | ra_coalesce(shader &sh) : pass(sh) {} |
||
498 | |||
499 | virtual int run(); |
||
500 | }; |
||
501 | |||
502 | |||
503 | |||
504 | // ======================================= |
||
505 | |||
506 | class ra_init : public pass { |
||
507 | |||
508 | public: |
||
509 | |||
510 | ra_init(shader &sh) : pass(sh), prev_chans() { |
||
511 | |||
512 | // The parameter below affects register channels distribution. |
||
513 | // For cayman (VLIW-4) we're trying to distribute the channels |
||
514 | // uniformly, this means significantly better alu slots utilization |
||
515 | // at the expense of higher gpr usage. Hopefully this will improve |
||
516 | // performance, though it has to be proven with real benchmarks yet. |
||
517 | // For VLIW-5 this method could also slightly improve slots |
||
518 | // utilization, but increased register pressure seems more significant |
||
519 | // and overall performance effect is negative according to some |
||
520 | // benchmarks, so it's not used currently. Basically, VLIW-5 doesn't |
||
521 | // really need it because trans slot (unrestricted by register write |
||
522 | // channel) allows to consume most deviations from uniform channel |
||
523 | // distribution. |
||
524 | // Value 3 means that for new allocation we'll use channel that differs |
||
525 | // from 3 last used channels. 0 for VLIW-5 effectively turns this off. |
||
526 | |||
527 | ra_tune = sh.get_ctx().is_cayman() ? 3 : 0; |
||
528 | } |
||
529 | |||
530 | virtual int run(); |
||
531 | |||
532 | private: |
||
533 | |||
534 | unsigned prev_chans; |
||
535 | unsigned ra_tune; |
||
536 | |||
537 | void add_prev_chan(unsigned chan); |
||
538 | unsigned get_preferable_chan_mask(); |
||
539 | |||
540 | void ra_node(container_node *c); |
||
541 | void process_op(node *n); |
||
542 | |||
543 | void color(value *v); |
||
544 | |||
545 | void color_bs_constraint(ra_constraint *c); |
||
546 | |||
547 | void assign_color(value *v, sel_chan c); |
||
548 | void alloc_arrays(); |
||
549 | }; |
||
550 | |||
551 | // ======================================= |
||
552 | |||
553 | class ra_split : public pass { |
||
554 | |||
555 | public: |
||
556 | |||
557 | ra_split(shader &sh) : pass(sh) {} |
||
558 | |||
559 | virtual int run(); |
||
560 | |||
561 | void split(container_node *n); |
||
562 | void split_op(node *n); |
||
563 | void split_alu_packed(alu_packed_node *n); |
||
564 | void split_vector_inst(node *n); |
||
565 | |||
566 | void split_packed_ins(alu_packed_node *n); |
||
567 | |||
568 | #if 0 |
||
569 | void split_pinned_outs(node *n); |
||
570 | #endif |
||
571 | |||
572 | void split_vec(vvec &vv, vvec &v1, vvec &v2, bool allow_swz); |
||
573 | |||
574 | void split_phi_src(container_node *loc, container_node *c, unsigned id, |
||
575 | bool loop); |
||
576 | void split_phi_dst(node *loc, container_node *c, bool loop); |
||
577 | void init_phi_constraints(container_node *c); |
||
578 | }; |
||
579 | |||
580 | |||
581 | |||
582 | class ssa_prepare : public vpass { |
||
583 | using vpass::visit; |
||
584 | |||
585 | typedef std::vector |
||
586 | vd_stk stk; |
||
587 | |||
588 | unsigned level; |
||
589 | |||
590 | public: |
||
591 | ssa_prepare(shader &s) : vpass(s), level(0) {} |
||
592 | |||
593 | virtual bool visit(cf_node &n, bool enter); |
||
594 | virtual bool visit(alu_node &n, bool enter); |
||
595 | virtual bool visit(fetch_node &n, bool enter); |
||
596 | virtual bool visit(region_node &n, bool enter); |
||
597 | virtual bool visit(repeat_node &n, bool enter); |
||
598 | virtual bool visit(depart_node &n, bool enter); |
||
599 | |||
600 | private: |
||
601 | |||
602 | void push_stk() { |
||
603 | ++level; |
||
604 | if (level + 1 > stk.size()) |
||
605 | stk.resize(level+1); |
||
606 | else |
||
607 | stk[level].clear(); |
||
608 | } |
||
609 | void pop_stk() { |
||
610 | assert(level); |
||
611 | --level; |
||
612 | stk[level].add_set(stk[level + 1]); |
||
613 | } |
||
614 | |||
615 | void add_defs(node &n); |
||
616 | |||
617 | val_set & cur_set() { return stk[level]; } |
||
618 | |||
619 | container_node* create_phi_nodes(int count); |
||
620 | }; |
||
621 | |||
622 | class ssa_rename : public vpass { |
||
623 | using vpass::visit; |
||
624 | |||
625 | typedef sb_map |
||
626 | |||
627 | def_map def_count; |
||
628 | std::stack |
||
629 | |||
630 | typedef std::map |
||
631 | val_map values; |
||
632 | |||
633 | public: |
||
634 | |||
635 | ssa_rename(shader &s) : vpass(s) {} |
||
636 | |||
637 | virtual int init(); |
||
638 | |||
639 | virtual bool visit(container_node &n, bool enter); |
||
640 | virtual bool visit(node &n, bool enter); |
||
641 | virtual bool visit(alu_group_node &n, bool enter); |
||
642 | virtual bool visit(cf_node &n, bool enter); |
||
643 | virtual bool visit(alu_node &n, bool enter); |
||
644 | virtual bool visit(alu_packed_node &n, bool enter); |
||
645 | virtual bool visit(fetch_node &n, bool enter); |
||
646 | virtual bool visit(region_node &n, bool enter); |
||
647 | virtual bool visit(repeat_node &n, bool enter); |
||
648 | virtual bool visit(depart_node &n, bool enter); |
||
649 | virtual bool visit(if_node &n, bool enter); |
||
650 | |||
651 | private: |
||
652 | |||
653 | void push(node *phi); |
||
654 | void pop(); |
||
655 | |||
656 | unsigned get_index(def_map& m, value* v); |
||
657 | void set_index(def_map& m, value* v, unsigned index); |
||
658 | unsigned new_index(def_map& m, value* v); |
||
659 | |||
660 | value* rename_use(node *n, value* v); |
||
661 | value* rename_def(node *def, value* v); |
||
662 | |||
663 | void rename_src_vec(node *n, vvec &vv, bool src); |
||
664 | void rename_dst_vec(node *def, vvec &vv, bool set_def); |
||
665 | |||
666 | void rename_src(node *n); |
||
667 | void rename_dst(node *n); |
||
668 | |||
669 | void rename_phi_args(container_node *phi, unsigned op, bool def); |
||
670 | |||
671 | void rename_virt(node *n); |
||
672 | void rename_virt_val(node *n, value *v); |
||
673 | }; |
||
674 | |||
675 | class bc_finalizer : public pass { |
||
676 | |||
677 | cf_node *last_export[EXP_TYPE_COUNT]; |
||
678 | cf_node *last_cf; |
||
679 | |||
680 | unsigned ngpr; |
||
681 | unsigned nstack; |
||
682 | |||
683 | public: |
||
684 | |||
685 | bc_finalizer(shader &sh) : pass(sh), last_export(), last_cf(), ngpr(), |
||
686 | nstack() {} |
||
687 | |||
688 | virtual int run(); |
||
689 | |||
690 | void finalize_loop(region_node *r); |
||
691 | void finalize_if(region_node *r); |
||
692 | |||
693 | void run_on(container_node *c); |
||
694 | |||
695 | void finalize_alu_group(alu_group_node *g); |
||
696 | void finalize_alu_src(alu_group_node *g, alu_node *a); |
||
697 | |||
698 | void emit_set_grad(fetch_node* f); |
||
699 | void finalize_fetch(fetch_node *f); |
||
700 | |||
701 | void finalize_cf(cf_node *c); |
||
702 | |||
703 | sel_chan translate_kcache(cf_node *alu, value *v); |
||
704 | |||
705 | void update_ngpr(unsigned gpr); |
||
706 | void update_nstack(region_node *r, unsigned add = 0); |
||
707 | |||
708 | void cf_peephole(); |
||
709 | |||
710 | }; |
||
711 | |||
712 | |||
713 | } // namespace r600_sb |
||
714 | |||
715 | #endif /* SB_PASS_H_ */ |