projects / mesa.git / blob
? search:


28511b5b81cff755b25c83d4a0cb6dc8a5ae3a2c
[mesa.git] / src / compiler / glsl / ir.cpp
1 /*
2 * Copyright © 2010 Intel Corporation
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 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * 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 NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
21 * DEALINGS IN THE SOFTWARE.
22 */
23 #include <string.h>
24 #include "main/core.h" /* for MAX2 */
25 #include "ir.h"
26 #include "compiler/glsl_types.h"
27
28 ir_rvalue::ir_rvalue(enum ir_node_type t)
29 : ir_instruction(t)
30 {
31 this->type = glsl_type::error_type;
32 }
33
34 bool ir_rvalue::is_zero() const
35 {
36 return false;
37 }
38
39 bool ir_rvalue::is_one() const
40 {
41 return false;
42 }
43
44 bool ir_rvalue::is_negative_one() const
45 {
46 return false;
47 }
48
49 /**
50 * Modify the swizzle make to move one component to another
51 *
52 * \param m IR swizzle to be modified
53 * \param from Component in the RHS that is to be swizzled
54 * \param to Desired swizzle location of \c from
55 */
56 static void
57 update_rhs_swizzle(ir_swizzle_mask &m, unsigned from, unsigned to)
58 {
59 switch (to) {
60 case 0: m.x = from; break;
61 case 1: m.y = from; break;
62 case 2: m.z = from; break;
63 case 3: m.w = from; break;
64 default: assert(!"Should not get here.");
65 }
66 }
67
68 void
69 ir_assignment::set_lhs(ir_rvalue *lhs)
70 {
71 void *mem_ctx = this;
72 bool swizzled = false;
73
74 while (lhs != NULL) {
75 ir_swizzle *swiz = lhs->as_swizzle();
76
77 if (swiz == NULL)
78 break;
79
80 unsigned write_mask = 0;
81 ir_swizzle_mask rhs_swiz = { 0, 0, 0, 0, 0, 0 };
82
83 for (unsigned i = 0; i < swiz->mask.num_components; i++) {
84 unsigned c = 0;
85
86 switch (i) {
87 case 0: c = swiz->mask.x; break;
88 case 1: c = swiz->mask.y; break;
89 case 2: c = swiz->mask.z; break;
90 case 3: c = swiz->mask.w; break;
91 default: assert(!"Should not get here.");
92 }
93
94 write_mask |= (((this->write_mask >> i) & 1) << c);
95 update_rhs_swizzle(rhs_swiz, i, c);
96 rhs_swiz.num_components = swiz->val->type->vector_elements;
97 }
98
99 this->write_mask = write_mask;
100 lhs = swiz->val;
101
102 this->rhs = new(mem_ctx) ir_swizzle(this->rhs, rhs_swiz);
103 swizzled = true;
104 }
105
106 if (swizzled) {
107 /* Now, RHS channels line up with the LHS writemask. Collapse it
108 * to just the channels that will be written.
109 */
110 ir_swizzle_mask rhs_swiz = { 0, 0, 0, 0, 0, 0 };
111 int rhs_chan = 0;
112 for (int i = 0; i < 4; i++) {
113 if (write_mask & (1 << i))
114 update_rhs_swizzle(rhs_swiz, i, rhs_chan++);
115 }
116 rhs_swiz.num_components = rhs_chan;
117 this->rhs = new(mem_ctx) ir_swizzle(this->rhs, rhs_swiz);
118 }
119
120 assert((lhs == NULL) || lhs->as_dereference());
121
122 this->lhs = (ir_dereference *) lhs;
123 }
124
125 ir_variable *
126 ir_assignment::whole_variable_written()
127 {
128 ir_variable *v = this->lhs->whole_variable_referenced();
129
130 if (v == NULL)
131 return NULL;
132
133 if (v->type->is_scalar())
134 return v;
135
136 if (v->type->is_vector()) {
137 const unsigned mask = (1U << v->type->vector_elements) - 1;
138
139 if (mask != this->write_mask)
140 return NULL;
141 }
142
143 /* Either all the vector components are assigned or the variable is some
144 * composite type (and the whole thing is assigned.
145 */
146 return v;
147 }
148
149 ir_assignment::ir_assignment(ir_dereference *lhs, ir_rvalue *rhs,
150 ir_rvalue *condition, unsigned write_mask)
151 : ir_instruction(ir_type_assignment)
152 {
153 this->condition = condition;
154 this->rhs = rhs;
155 this->lhs = lhs;
156 this->write_mask = write_mask;
157
158 if (lhs->type->is_scalar() || lhs->type->is_vector()) {
159 int lhs_components = 0;
160 for (int i = 0; i < 4; i++) {
161 if (write_mask & (1 << i))
162 lhs_components++;
163 }
164
165 assert(lhs_components == this->rhs->type->vector_elements);
166 }
167 }
168
169 ir_assignment::ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs,
170 ir_rvalue *condition)
171 : ir_instruction(ir_type_assignment)
172 {
173 this->condition = condition;
174 this->rhs = rhs;
175
176 /* If the RHS is a vector type, assume that all components of the vector
177 * type are being written to the LHS. The write mask comes from the RHS
178 * because we can have a case where the LHS is a vec4 and the RHS is a
179 * vec3. In that case, the assignment is:
180 *
181 * (assign (...) (xyz) (var_ref lhs) (var_ref rhs))
182 */
183 if (rhs->type->is_vector())
184 this->write_mask = (1U << rhs->type->vector_elements) - 1;
185 else if (rhs->type->is_scalar())
186 this->write_mask = 1;
187 else
188 this->write_mask = 0;
189
190 this->set_lhs(lhs);
191 }
192
193 ir_expression::ir_expression(int op, const struct glsl_type *type,
194 ir_rvalue *op0, ir_rvalue *op1,
195 ir_rvalue *op2, ir_rvalue *op3)
196 : ir_rvalue(ir_type_expression)
197 {
198 this->type = type;
199 this->operation = ir_expression_operation(op);
200 this->operands[0] = op0;
201 this->operands[1] = op1;
202 this->operands[2] = op2;
203 this->operands[3] = op3;
204 #ifndef NDEBUG
205 int num_operands = get_num_operands(this->operation);
206 for (int i = num_operands; i < 4; i++) {
207 assert(this->operands[i] == NULL);
208 }
209 #endif
210 }
211
212 ir_expression::ir_expression(int op, ir_rvalue *op0)
213 : ir_rvalue(ir_type_expression)
214 {
215 this->operation = ir_expression_operation(op);
216 this->operands[0] = op0;
217 this->operands[1] = NULL;
218 this->operands[2] = NULL;
219 this->operands[3] = NULL;
220
221 assert(op <= ir_last_unop);
222
223 switch (this->operation) {
224 case ir_unop_bit_not:
225 case ir_unop_logic_not:
226 case ir_unop_neg:
227 case ir_unop_abs:
228 case ir_unop_sign:
229 case ir_unop_rcp:
230 case ir_unop_rsq:
231 case ir_unop_sqrt:
232 case ir_unop_exp:
233 case ir_unop_log:
234 case ir_unop_exp2:
235 case ir_unop_log2:
236 case ir_unop_trunc:
237 case ir_unop_ceil:
238 case ir_unop_floor:
239 case ir_unop_fract:
240 case ir_unop_round_even:
241 case ir_unop_sin:
242 case ir_unop_cos:
243 case ir_unop_dFdx:
244 case ir_unop_dFdx_coarse:
245 case ir_unop_dFdx_fine:
246 case ir_unop_dFdy:
247 case ir_unop_dFdy_coarse:
248 case ir_unop_dFdy_fine:
249 case ir_unop_bitfield_reverse:
250 case ir_unop_interpolate_at_centroid:
251 case ir_unop_saturate:
252 this->type = op0->type;
253 break;
254
255 case ir_unop_f2i:
256 case ir_unop_b2i:
257 case ir_unop_u2i:
258 case ir_unop_d2i:
259 case ir_unop_bitcast_f2i:
260 case ir_unop_bit_count:
261 case ir_unop_find_msb:
262 case ir_unop_find_lsb:
263 case ir_unop_subroutine_to_int:
264 case ir_unop_i642i:
265 case ir_unop_u642i:
266 this->type = glsl_type::get_instance(GLSL_TYPE_INT,
267 op0->type->vector_elements, 1);
268 break;
269
270 case ir_unop_b2f:
271 case ir_unop_i2f:
272 case ir_unop_u2f:
273 case ir_unop_d2f:
274 case ir_unop_bitcast_i2f:
275 case ir_unop_bitcast_u2f:
276 case ir_unop_i642f:
277 case ir_unop_u642f:
278 this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT,
279 op0->type->vector_elements, 1);
280 break;
281
282 case ir_unop_f2b:
283 case ir_unop_i2b:
284 case ir_unop_d2b:
285 case ir_unop_i642b:
286 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL,
287 op0->type->vector_elements, 1);
288 break;
289
290 case ir_unop_f2d:
291 case ir_unop_i2d:
292 case ir_unop_u2d:
293 case ir_unop_i642d:
294 case ir_unop_u642d:
295 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE,
296 op0->type->vector_elements, 1);
297 break;
298
299 case ir_unop_i2u:
300 case ir_unop_f2u:
301 case ir_unop_d2u:
302 case ir_unop_bitcast_f2u:
303 case ir_unop_i642u:
304 case ir_unop_u642u:
305 this->type = glsl_type::get_instance(GLSL_TYPE_UINT,
306 op0->type->vector_elements, 1);
307 break;
308
309 case ir_unop_i2i64:
310 case ir_unop_u2i64:
311 case ir_unop_b2i64:
312 case ir_unop_f2i64:
313 case ir_unop_d2i64:
314 case ir_unop_u642i64:
315 this->type = glsl_type::get_instance(GLSL_TYPE_INT64,
316 op0->type->vector_elements, 1);
317 break;
318
319 case ir_unop_i2u64:
320 case ir_unop_u2u64:
321 case ir_unop_f2u64:
322 case ir_unop_d2u64:
323 case ir_unop_i642u64:
324 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64,
325 op0->type->vector_elements, 1);
326 break;
327 case ir_unop_noise:
328 this->type = glsl_type::float_type;
329 break;
330
331 case ir_unop_unpack_double_2x32:
332 case ir_unop_unpack_uint_2x32:
333 this->type = glsl_type::uvec2_type;
334 break;
335
336 case ir_unop_unpack_int_2x32:
337 this->type = glsl_type::ivec2_type;
338 break;
339
340 case ir_unop_pack_snorm_2x16:
341 case ir_unop_pack_snorm_4x8:
342 case ir_unop_pack_unorm_2x16:
343 case ir_unop_pack_unorm_4x8:
344 case ir_unop_pack_half_2x16:
345 this->type = glsl_type::uint_type;
346 break;
347
348 case ir_unop_pack_double_2x32:
349 this->type = glsl_type::double_type;
350 break;
351
352 case ir_unop_pack_int_2x32:
353 this->type = glsl_type::int64_t_type;
354 break;
355
356 case ir_unop_pack_uint_2x32:
357 this->type = glsl_type::uint64_t_type;
358 break;
359
360 case ir_unop_unpack_snorm_2x16:
361 case ir_unop_unpack_unorm_2x16:
362 case ir_unop_unpack_half_2x16:
363 this->type = glsl_type::vec2_type;
364 break;
365
366 case ir_unop_unpack_snorm_4x8:
367 case ir_unop_unpack_unorm_4x8:
368 this->type = glsl_type::vec4_type;
369 break;
370
371 case ir_unop_frexp_sig:
372 this->type = op0->type;
373 break;
374 case ir_unop_frexp_exp:
375 this->type = glsl_type::get_instance(GLSL_TYPE_INT,
376 op0->type->vector_elements, 1);
377 break;
378
379 case ir_unop_get_buffer_size:
380 case ir_unop_ssbo_unsized_array_length:
381 this->type = glsl_type::int_type;
382 break;
383
384 case ir_unop_vote_any:
385 case ir_unop_vote_all:
386 case ir_unop_vote_eq:
387 this->type = glsl_type::bool_type;
388 break;
389
390 case ir_unop_bitcast_i642d:
391 case ir_unop_bitcast_u642d:
392 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE,
393 op0->type->vector_elements, 1);
394 break;
395
396 case ir_unop_bitcast_d2i64:
397 this->type = glsl_type::get_instance(GLSL_TYPE_INT64,
398 op0->type->vector_elements, 1);
399 break;
400 case ir_unop_bitcast_d2u64:
401 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64,
402 op0->type->vector_elements, 1);
403 break;
404
405 default:
406 assert(!"not reached: missing automatic type setup for ir_expression");
407 this->type = op0->type;
408 break;
409 }
410 }
411
412 ir_expression::ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1)
413 : ir_rvalue(ir_type_expression)
414 {
415 this->operation = ir_expression_operation(op);
416 this->operands[0] = op0;
417 this->operands[1] = op1;
418 this->operands[2] = NULL;
419 this->operands[3] = NULL;
420
421 assert(op > ir_last_unop);
422
423 switch (this->operation) {
424 case ir_binop_all_equal:
425 case ir_binop_any_nequal:
426 this->type = glsl_type::bool_type;
427 break;
428
429 case ir_binop_add:
430 case ir_binop_sub:
431 case ir_binop_min:
432 case ir_binop_max:
433 case ir_binop_pow:
434 case ir_binop_mul:
435 case ir_binop_div:
436 case ir_binop_mod:
437 if (op0->type->is_scalar()) {
438 this->type = op1->type;
439 } else if (op1->type->is_scalar()) {
440 this->type = op0->type;
441 } else {
442 if (this->operation == ir_binop_mul) {
443 this->type = glsl_type::get_mul_type(op0->type, op1->type);
444 } else {
445 assert(op0->type == op1->type);
446 this->type = op0->type;
447 }
448 }
449 break;
450
451 case ir_binop_logic_and:
452 case ir_binop_logic_xor:
453 case ir_binop_logic_or:
454 case ir_binop_bit_and:
455 case ir_binop_bit_xor:
456 case ir_binop_bit_or:
457 assert(!op0->type->is_matrix());
458 assert(!op1->type->is_matrix());
459 if (op0->type->is_scalar()) {
460 this->type = op1->type;
461 } else if (op1->type->is_scalar()) {
462 this->type = op0->type;
463 } else {
464 assert(op0->type->vector_elements == op1->type->vector_elements);
465 this->type = op0->type;
466 }
467 break;
468
469 case ir_binop_equal:
470 case ir_binop_nequal:
471 case ir_binop_lequal:
472 case ir_binop_gequal:
473 case ir_binop_less:
474 case ir_binop_greater:
475 assert(op0->type == op1->type);
476 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL,
477 op0->type->vector_elements, 1);
478 break;
479
480 case ir_binop_dot:
481 this->type = op0->type->get_base_type();
482 break;
483
484 case ir_binop_imul_high:
485 case ir_binop_carry:
486 case ir_binop_borrow:
487 case ir_binop_lshift:
488 case ir_binop_rshift:
489 case ir_binop_ldexp:
490 case ir_binop_interpolate_at_offset:
491 case ir_binop_interpolate_at_sample:
492 this->type = op0->type;
493 break;
494
495 case ir_binop_vector_extract:
496 this->type = op0->type->get_scalar_type();
497 break;
498
499 default:
500 assert(!"not reached: missing automatic type setup for ir_expression");
501 this->type = glsl_type::float_type;
502 }
503 }
504
505 ir_expression::ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1,
506 ir_rvalue *op2)
507 : ir_rvalue(ir_type_expression)
508 {
509 this->operation = ir_expression_operation(op);
510 this->operands[0] = op0;
511 this->operands[1] = op1;
512 this->operands[2] = op2;
513 this->operands[3] = NULL;
514
515 assert(op > ir_last_binop && op <= ir_last_triop);
516
517 switch (this->operation) {
518 case ir_triop_fma:
519 case ir_triop_lrp:
520 case ir_triop_bitfield_extract:
521 case ir_triop_vector_insert:
522 this->type = op0->type;
523 break;
524
525 case ir_triop_csel:
526 this->type = op1->type;
527 break;
528
529 default:
530 assert(!"not reached: missing automatic type setup for ir_expression");
531 this->type = glsl_type::float_type;
532 }
533 }
534
535 unsigned int
536 ir_expression::get_num_operands(ir_expression_operation op)
537 {
538 assert(op <= ir_last_opcode);
539
540 if (op <= ir_last_unop)
541 return 1;
542
543 if (op <= ir_last_binop)
544 return 2;
545
546 if (op <= ir_last_triop)
547 return 3;
548
549 if (op <= ir_last_quadop)
550 return 4;
551
552 assert(false);
553 return 0;
554 }
555
556 #include "ir_expression_operation_strings.h"
557
558 const char*
559 depth_layout_string(ir_depth_layout layout)
560 {
561 switch(layout) {
562 case ir_depth_layout_none: return "";
563 case ir_depth_layout_any: return "depth_any";
564 case ir_depth_layout_greater: return "depth_greater";
565 case ir_depth_layout_less: return "depth_less";
566 case ir_depth_layout_unchanged: return "depth_unchanged";
567
568 default:
569 assert(0);
570 return "";
571 }
572 }
573
574 ir_expression_operation
575 ir_expression::get_operator(const char *str)
576 {
577 for (int op = 0; op <= int(ir_last_opcode); op++) {
578 if (strcmp(str, ir_expression_operation_strings[op]) == 0)
579 return (ir_expression_operation) op;
580 }
581 return (ir_expression_operation) -1;
582 }
583
584 ir_variable *
585 ir_expression::variable_referenced() const
586 {
587 switch (operation) {
588 case ir_binop_vector_extract:
589 case ir_triop_vector_insert:
590 /* We get these for things like a[0] where a is a vector type. In these
591 * cases we want variable_referenced() to return the actual vector
592 * variable this is wrapping.
593 */
594 return operands[0]->variable_referenced();
595 default:
596 return ir_rvalue::variable_referenced();
597 }
598 }
599
600 ir_constant::ir_constant()
601 : ir_rvalue(ir_type_constant)
602 {
603 this->array_elements = NULL;
604 }
605
606 ir_constant::ir_constant(const struct glsl_type *type,
607 const ir_constant_data *data)
608 : ir_rvalue(ir_type_constant)
609 {
610 this->array_elements = NULL;
611
612 assert((type->base_type >= GLSL_TYPE_UINT)
613 && (type->base_type <= GLSL_TYPE_BOOL));
614
615 this->type = type;
616 memcpy(& this->value, data, sizeof(this->value));
617 }
618
619 ir_constant::ir_constant(float f, unsigned vector_elements)
620 : ir_rvalue(ir_type_constant)
621 {
622 assert(vector_elements <= 4);
623 this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT, vector_elements, 1);
624 for (unsigned i = 0; i < vector_elements; i++) {
625 this->value.f[i] = f;
626 }
627 for (unsigned i = vector_elements; i < 16; i++) {
628 this->value.f[i] = 0;
629 }
630 }
631
632 ir_constant::ir_constant(double d, unsigned vector_elements)
633 : ir_rvalue(ir_type_constant)
634 {
635 assert(vector_elements <= 4);
636 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE, vector_elements, 1);
637 for (unsigned i = 0; i < vector_elements; i++) {
638 this->value.d[i] = d;
639 }
640 for (unsigned i = vector_elements; i < 16; i++) {
641 this->value.d[i] = 0.0;
642 }
643 }
644
645 ir_constant::ir_constant(unsigned int u, unsigned vector_elements)
646 : ir_rvalue(ir_type_constant)
647 {
648 assert(vector_elements <= 4);
649 this->type = glsl_type::get_instance(GLSL_TYPE_UINT, vector_elements, 1);
650 for (unsigned i = 0; i < vector_elements; i++) {
651 this->value.u[i] = u;
652 }
653 for (unsigned i = vector_elements; i < 16; i++) {
654 this->value.u[i] = 0;
655 }
656 }
657
658 ir_constant::ir_constant(int integer, unsigned vector_elements)
659 : ir_rvalue(ir_type_constant)
660 {
661 assert(vector_elements <= 4);
662 this->type = glsl_type::get_instance(GLSL_TYPE_INT, vector_elements, 1);
663 for (unsigned i = 0; i < vector_elements; i++) {
664 this->value.i[i] = integer;
665 }
666 for (unsigned i = vector_elements; i < 16; i++) {
667 this->value.i[i] = 0;
668 }
669 }
670
671 ir_constant::ir_constant(uint64_t u64, unsigned vector_elements)
672 : ir_rvalue(ir_type_constant)
673 {
674 assert(vector_elements <= 4);
675 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64, vector_elements, 1);
676 for (unsigned i = 0; i < vector_elements; i++) {
677 this->value.u64[i] = u64;
678 }
679 for (unsigned i = vector_elements; i < 16; i++) {
680 this->value.u64[i] = 0;
681 }
682 }
683
684 ir_constant::ir_constant(int64_t int64, unsigned vector_elements)
685 : ir_rvalue(ir_type_constant)
686 {
687 assert(vector_elements <= 4);
688 this->type = glsl_type::get_instance(GLSL_TYPE_INT64, vector_elements, 1);
689 for (unsigned i = 0; i < vector_elements; i++) {
690 this->value.i64[i] = int64;
691 }
692 for (unsigned i = vector_elements; i < 16; i++) {
693 this->value.i64[i] = 0;
694 }
695 }
696
697 ir_constant::ir_constant(bool b, unsigned vector_elements)
698 : ir_rvalue(ir_type_constant)
699 {
700 assert(vector_elements <= 4);
701 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL, vector_elements, 1);
702 for (unsigned i = 0; i < vector_elements; i++) {
703 this->value.b[i] = b;
704 }
705 for (unsigned i = vector_elements; i < 16; i++) {
706 this->value.b[i] = false;
707 }
708 }
709
710 ir_constant::ir_constant(const ir_constant *c, unsigned i)
711 : ir_rvalue(ir_type_constant)
712 {
713 this->array_elements = NULL;
714 this->type = c->type->get_base_type();
715
716 switch (this->type->base_type) {
717 case GLSL_TYPE_UINT: this->value.u[0] = c->value.u[i]; break;
718 case GLSL_TYPE_INT: this->value.i[0] = c->value.i[i]; break;
719 case GLSL_TYPE_FLOAT: this->value.f[0] = c->value.f[i]; break;
720 case GLSL_TYPE_BOOL: this->value.b[0] = c->value.b[i]; break;
721 case GLSL_TYPE_DOUBLE: this->value.d[0] = c->value.d[i]; break;
722 default: assert(!"Should not get here."); break;
723 }
724 }
725
726 ir_constant::ir_constant(const struct glsl_type *type, exec_list *value_list)
727 : ir_rvalue(ir_type_constant)
728 {
729 this->array_elements = NULL;
730 this->type = type;
731
732 assert(type->is_scalar() || type->is_vector() || type->is_matrix()
733 || type->is_record() || type->is_array());
734
735 if (type->is_array()) {
736 this->array_elements = ralloc_array(this, ir_constant *, type->length);
737 unsigned i = 0;
738 foreach_in_list(ir_constant, value, value_list) {
739 assert(value->as_constant() != NULL);
740
741 this->array_elements[i++] = value;
742 }
743 return;
744 }
745
746 /* If the constant is a record, the types of each of the entries in
747 * value_list must be a 1-for-1 match with the structure components. Each
748 * entry must also be a constant. Just move the nodes from the value_list
749 * to the list in the ir_constant.
750 */
751 /* FINISHME: Should there be some type checking and / or assertions here? */
752 /* FINISHME: Should the new constant take ownership of the nodes from
753 * FINISHME: value_list, or should it make copies?
754 */
755 if (type->is_record()) {
756 value_list->move_nodes_to(& this->components);
757 return;
758 }
759
760 for (unsigned i = 0; i < 16; i++) {
761 this->value.u[i] = 0;
762 }
763
764 ir_constant *value = (ir_constant *) (value_list->get_head_raw());
765
766 /* Constructors with exactly one scalar argument are special for vectors
767 * and matrices. For vectors, the scalar value is replicated to fill all
768 * the components. For matrices, the scalar fills the components of the
769 * diagonal while the rest is filled with 0.
770 */
771 if (value->type->is_scalar() && value->next->is_tail_sentinel()) {
772 if (type->is_matrix()) {
773 /* Matrix - fill diagonal (rest is already set to 0) */
774 assert(type->base_type == GLSL_TYPE_FLOAT ||
775 type->base_type == GLSL_TYPE_DOUBLE);
776 for (unsigned i = 0; i < type->matrix_columns; i++) {
777 if (type->base_type == GLSL_TYPE_FLOAT)
778 this->value.f[i * type->vector_elements + i] =
779 value->value.f[0];
780 else
781 this->value.d[i * type->vector_elements + i] =
782 value->value.d[0];
783 }
784 } else {
785 /* Vector or scalar - fill all components */
786 switch (type->base_type) {
787 case GLSL_TYPE_UINT:
788 case GLSL_TYPE_INT:
789 for (unsigned i = 0; i < type->components(); i++)
790 this->value.u[i] = value->value.u[0];
791 break;
792 case GLSL_TYPE_FLOAT:
793 for (unsigned i = 0; i < type->components(); i++)
794 this->value.f[i] = value->value.f[0];
795 break;
796 case GLSL_TYPE_DOUBLE:
797 for (unsigned i = 0; i < type->components(); i++)
798 this->value.d[i] = value->value.d[0];
799 break;
800 case GLSL_TYPE_UINT64:
801 case GLSL_TYPE_INT64:
802 for (unsigned i = 0; i < type->components(); i++)
803 this->value.u64[i] = value->value.u64[0];
804 break;
805 case GLSL_TYPE_BOOL:
806 for (unsigned i = 0; i < type->components(); i++)
807 this->value.b[i] = value->value.b[0];
808 break;
809 default:
810 assert(!"Should not get here.");
811 break;
812 }
813 }
814 return;
815 }
816
817 if (type->is_matrix() && value->type->is_matrix()) {
818 assert(value->next->is_tail_sentinel());
819
820 /* From section 5.4.2 of the GLSL 1.20 spec:
821 * "If a matrix is constructed from a matrix, then each component
822 * (column i, row j) in the result that has a corresponding component
823 * (column i, row j) in the argument will be initialized from there."
824 */
825 unsigned cols = MIN2(type->matrix_columns, value->type->matrix_columns);
826 unsigned rows = MIN2(type->vector_elements, value->type->vector_elements);
827 for (unsigned i = 0; i < cols; i++) {
828 for (unsigned j = 0; j < rows; j++) {
829 const unsigned src = i * value->type->vector_elements + j;
830 const unsigned dst = i * type->vector_elements + j;
831 this->value.f[dst] = value->value.f[src];
832 }
833 }
834
835 /* "All other components will be initialized to the identity matrix." */
836 for (unsigned i = cols; i < type->matrix_columns; i++)
837 this->value.f[i * type->vector_elements + i] = 1.0;
838
839 return;
840 }
841
842 /* Use each component from each entry in the value_list to initialize one
843 * component of the constant being constructed.
844 */
845 unsigned i = 0;
846 for (;;) {
847 assert(value->as_constant() != NULL);
848 assert(!value->is_tail_sentinel());
849
850 for (unsigned j = 0; j < value->type->components(); j++) {
851 switch (type->base_type) {
852 case GLSL_TYPE_UINT:
853 this->value.u[i] = value->get_uint_component(j);
854 break;
855 case GLSL_TYPE_INT:
856 this->value.i[i] = value->get_int_component(j);
857 break;
858 case GLSL_TYPE_FLOAT:
859 this->value.f[i] = value->get_float_component(j);
860 break;
861 case GLSL_TYPE_BOOL:
862 this->value.b[i] = value->get_bool_component(j);
863 break;
864 case GLSL_TYPE_DOUBLE:
865 this->value.d[i] = value->get_double_component(j);
866 break;
867 case GLSL_TYPE_UINT64:
868 this->value.u64[i] = value->get_uint64_component(j);
869 break;
870 case GLSL_TYPE_INT64:
871 this->value.i64[i] = value->get_int64_component(j);
872 break;
873 default:
874 /* FINISHME: What to do? Exceptions are not the answer.
875 */
876 break;
877 }
878
879 i++;
880 if (i >= type->components())
881 break;
882 }
883
884 if (i >= type->components())
885 break; /* avoid downcasting a list sentinel */
886 value = (ir_constant *) value->next;
887 }
888 }
889
890 ir_constant *
891 ir_constant::zero(void *mem_ctx, const glsl_type *type)
892 {
893 assert(type->is_scalar() || type->is_vector() || type->is_matrix()
894 || type->is_record() || type->is_array());
895
896 ir_constant *c = new(mem_ctx) ir_constant;
897 c->type = type;
898 memset(&c->value, 0, sizeof(c->value));
899
900 if (type->is_array()) {
901 c->array_elements = ralloc_array(c, ir_constant *, type->length);
902
903 for (unsigned i = 0; i < type->length; i++)
904 c->array_elements[i] = ir_constant::zero(c, type->fields.array);
905 }
906
907 if (type->is_record()) {
908 for (unsigned i = 0; i < type->length; i++) {
909 ir_constant *comp = ir_constant::zero(mem_ctx, type->fields.structure[i].type);
910 c->components.push_tail(comp);
911 }
912 }
913
914 return c;
915 }
916
917 bool
918 ir_constant::get_bool_component(unsigned i) const
919 {
920 switch (this->type->base_type) {
921 case GLSL_TYPE_UINT: return this->value.u[i] != 0;
922 case GLSL_TYPE_INT: return this->value.i[i] != 0;
923 case GLSL_TYPE_FLOAT: return ((int)this->value.f[i]) != 0;
924 case GLSL_TYPE_BOOL: return this->value.b[i];
925 case GLSL_TYPE_DOUBLE: return this->value.d[i] != 0.0;
926 case GLSL_TYPE_UINT64: return this->value.u64[i] != 0;
927 case GLSL_TYPE_INT64: return this->value.i64[i] != 0;
928 default: assert(!"Should not get here."); break;
929 }
930
931 /* Must return something to make the compiler happy. This is clearly an
932 * error case.
933 */
934 return false;
935 }
936
937 float
938 ir_constant::get_float_component(unsigned i) const
939 {
940 switch (this->type->base_type) {
941 case GLSL_TYPE_UINT: return (float) this->value.u[i];
942 case GLSL_TYPE_INT: return (float) this->value.i[i];
943 case GLSL_TYPE_FLOAT: return this->value.f[i];
944 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1.0f : 0.0f;
945 case GLSL_TYPE_DOUBLE: return (float) this->value.d[i];
946 case GLSL_TYPE_UINT64: return (float) this->value.u64[i];
947 case GLSL_TYPE_INT64: return (float) this->value.i64[i];
948 default: assert(!"Should not get here."); break;
949 }
950
951 /* Must return something to make the compiler happy. This is clearly an
952 * error case.
953 */
954 return 0.0;
955 }
956
957 double
958 ir_constant::get_double_component(unsigned i) const
959 {
960 switch (this->type->base_type) {
961 case GLSL_TYPE_UINT: return (double) this->value.u[i];
962 case GLSL_TYPE_INT: return (double) this->value.i[i];
963 case GLSL_TYPE_FLOAT: return (double) this->value.f[i];
964 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1.0 : 0.0;
965 case GLSL_TYPE_DOUBLE: return this->value.d[i];
966 case GLSL_TYPE_UINT64: return (double) this->value.u64[i];
967 case GLSL_TYPE_INT64: return (double) this->value.i64[i];
968 default: assert(!"Should not get here."); break;
969 }
970
971 /* Must return something to make the compiler happy. This is clearly an
972 * error case.
973 */
974 return 0.0;
975 }
976
977 int
978 ir_constant::get_int_component(unsigned i) const
979 {
980 switch (this->type->base_type) {
981 case GLSL_TYPE_UINT: return this->value.u[i];
982 case GLSL_TYPE_INT: return this->value.i[i];
983 case GLSL_TYPE_FLOAT: return (int) this->value.f[i];
984 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
985 case GLSL_TYPE_DOUBLE: return (int) this->value.d[i];
986 case GLSL_TYPE_UINT64: return (int) this->value.u64[i];
987 case GLSL_TYPE_INT64: return (int) this->value.i64[i];
988 default: assert(!"Should not get here."); break;
989 }
990
991 /* Must return something to make the compiler happy. This is clearly an
992 * error case.
993 */
994 return 0;
995 }
996
997 unsigned
998 ir_constant::get_uint_component(unsigned i) const
999 {
1000 switch (this->type->base_type) {
1001 case GLSL_TYPE_UINT: return this->value.u[i];
1002 case GLSL_TYPE_INT: return this->value.i[i];
1003 case GLSL_TYPE_FLOAT: return (unsigned) this->value.f[i];
1004 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1005 case GLSL_TYPE_DOUBLE: return (unsigned) this->value.d[i];
1006 case GLSL_TYPE_UINT64: return (unsigned) this->value.u64[i];
1007 case GLSL_TYPE_INT64: return (unsigned) this->value.i64[i];
1008 default: assert(!"Should not get here."); break;
1009 }
1010
1011 /* Must return something to make the compiler happy. This is clearly an
1012 * error case.
1013 */
1014 return 0;
1015 }
1016
1017 int64_t
1018 ir_constant::get_int64_component(unsigned i) const
1019 {
1020 switch (this->type->base_type) {
1021 case GLSL_TYPE_UINT: return this->value.u[i];
1022 case GLSL_TYPE_INT: return this->value.i[i];
1023 case GLSL_TYPE_FLOAT: return (int64_t) this->value.f[i];
1024 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1025 case GLSL_TYPE_DOUBLE: return (int64_t) this->value.d[i];
1026 case GLSL_TYPE_UINT64: return (int64_t) this->value.u64[i];
1027 case GLSL_TYPE_INT64: return this->value.i64[i];
1028 default: assert(!"Should not get here."); break;
1029 }
1030
1031 /* Must return something to make the compiler happy. This is clearly an
1032 * error case.
1033 */
1034 return 0;
1035 }
1036
1037 uint64_t
1038 ir_constant::get_uint64_component(unsigned i) const
1039 {
1040 switch (this->type->base_type) {
1041 case GLSL_TYPE_UINT: return this->value.u[i];
1042 case GLSL_TYPE_INT: return this->value.i[i];
1043 case GLSL_TYPE_FLOAT: return (uint64_t) this->value.f[i];
1044 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1045 case GLSL_TYPE_DOUBLE: return (uint64_t) this->value.d[i];
1046 case GLSL_TYPE_UINT64: return this->value.u64[i];
1047 case GLSL_TYPE_INT64: return (uint64_t) this->value.i64[i];
1048 default: assert(!"Should not get here."); break;
1049 }
1050
1051 /* Must return something to make the compiler happy. This is clearly an
1052 * error case.
1053 */
1054 return 0;
1055 }
1056
1057 ir_constant *
1058 ir_constant::get_array_element(unsigned i) const
1059 {
1060 assert(this->type->is_array());
1061
1062 /* From page 35 (page 41 of the PDF) of the GLSL 1.20 spec:
1063 *
1064 * "Behavior is undefined if a shader subscripts an array with an index
1065 * less than 0 or greater than or equal to the size the array was
1066 * declared with."
1067 *
1068 * Most out-of-bounds accesses are removed before things could get this far.
1069 * There are cases where non-constant array index values can get constant
1070 * folded.
1071 */
1072 if (int(i) < 0)
1073 i = 0;
1074 else if (i >= this->type->length)
1075 i = this->type->length - 1;
1076
1077 return array_elements[i];
1078 }
1079
1080 ir_constant *
1081 ir_constant::get_record_field(const char *name)
1082 {
1083 int idx = this->type->field_index(name);
1084
1085 if (idx < 0)
1086 return NULL;
1087
1088 if (this->components.is_empty())
1089 return NULL;
1090
1091 exec_node *node = this->components.get_head_raw();
1092 for (int i = 0; i < idx; i++) {
1093 node = node->next;
1094
1095 /* If the end of the list is encountered before the element matching the
1096 * requested field is found, return NULL.
1097 */
1098 if (node->is_tail_sentinel())
1099 return NULL;
1100 }
1101
1102 return (ir_constant *) node;
1103 }
1104
1105 void
1106 ir_constant::copy_offset(ir_constant *src, int offset)
1107 {
1108 switch (this->type->base_type) {
1109 case GLSL_TYPE_UINT:
1110 case GLSL_TYPE_INT:
1111 case GLSL_TYPE_FLOAT:
1112 case GLSL_TYPE_DOUBLE:
1113 case GLSL_TYPE_UINT64:
1114 case GLSL_TYPE_INT64:
1115 case GLSL_TYPE_BOOL: {
1116 unsigned int size = src->type->components();
1117 assert (size <= this->type->components() - offset);
1118 for (unsigned int i=0; i<size; i++) {
1119 switch (this->type->base_type) {
1120 case GLSL_TYPE_UINT:
1121 value.u[i+offset] = src->get_uint_component(i);
1122 break;
1123 case GLSL_TYPE_INT:
1124 value.i[i+offset] = src->get_int_component(i);
1125 break;
1126 case GLSL_TYPE_FLOAT:
1127 value.f[i+offset] = src->get_float_component(i);
1128 break;
1129 case GLSL_TYPE_BOOL:
1130 value.b[i+offset] = src->get_bool_component(i);
1131 break;
1132 case GLSL_TYPE_DOUBLE:
1133 value.d[i+offset] = src->get_double_component(i);
1134 break;
1135 case GLSL_TYPE_UINT64:
1136 value.u64[i+offset] = src->get_uint64_component(i);
1137 break;
1138 case GLSL_TYPE_INT64:
1139 value.i64[i+offset] = src->get_int64_component(i);
1140 break;
1141 default: // Shut up the compiler
1142 break;
1143 }
1144 }
1145 break;
1146 }
1147
1148 case GLSL_TYPE_STRUCT: {
1149 assert (src->type == this->type);
1150 this->components.make_empty();
1151 foreach_in_list(ir_constant, orig, &src->components) {
1152 this->components.push_tail(orig->clone(this, NULL));
1153 }
1154 break;
1155 }
1156
1157 case GLSL_TYPE_ARRAY: {
1158 assert (src->type == this->type);
1159 for (unsigned i = 0; i < this->type->length; i++) {
1160 this->array_elements[i] = src->array_elements[i]->clone(this, NULL);
1161 }
1162 break;
1163 }
1164
1165 default:
1166 assert(!"Should not get here.");
1167 break;
1168 }
1169 }
1170
1171 void
1172 ir_constant::copy_masked_offset(ir_constant *src, int offset, unsigned int mask)
1173 {
1174 assert (!type->is_array() && !type->is_record());
1175
1176 if (!type->is_vector() && !type->is_matrix()) {
1177 offset = 0;
1178 mask = 1;
1179 }
1180
1181 int id = 0;
1182 for (int i=0; i<4; i++) {
1183 if (mask & (1 << i)) {
1184 switch (this->type->base_type) {
1185 case GLSL_TYPE_UINT:
1186 value.u[i+offset] = src->get_uint_component(id++);
1187 break;
1188 case GLSL_TYPE_INT:
1189 value.i[i+offset] = src->get_int_component(id++);
1190 break;
1191 case GLSL_TYPE_FLOAT:
1192 value.f[i+offset] = src->get_float_component(id++);
1193 break;
1194 case GLSL_TYPE_BOOL:
1195 value.b[i+offset] = src->get_bool_component(id++);
1196 break;
1197 case GLSL_TYPE_DOUBLE:
1198 value.d[i+offset] = src->get_double_component(id++);
1199 break;
1200 case GLSL_TYPE_UINT64:
1201 value.u64[i+offset] = src->get_uint64_component(id++);
1202 break;
1203 case GLSL_TYPE_INT64:
1204 value.i64[i+offset] = src->get_int64_component(id++);
1205 break;
1206 default:
1207 assert(!"Should not get here.");
1208 return;
1209 }
1210 }
1211 }
1212 }
1213
1214 bool
1215 ir_constant::has_value(const ir_constant *c) const
1216 {
1217 if (this->type != c->type)
1218 return false;
1219
1220 if (this->type->is_array()) {
1221 for (unsigned i = 0; i < this->type->length; i++) {
1222 if (!this->array_elements[i]->has_value(c->array_elements[i]))
1223 return false;
1224 }
1225 return true;
1226 }
1227
1228 if (this->type->base_type == GLSL_TYPE_STRUCT) {
1229 const exec_node *a_node = this->components.get_head_raw();
1230 const exec_node *b_node = c->components.get_head_raw();
1231
1232 while (!a_node->is_tail_sentinel()) {
1233 assert(!b_node->is_tail_sentinel());
1234
1235 const ir_constant *const a_field = (ir_constant *) a_node;
1236 const ir_constant *const b_field = (ir_constant *) b_node;
1237
1238 if (!a_field->has_value(b_field))
1239 return false;
1240
1241 a_node = a_node->next;
1242 b_node = b_node->next;
1243 }
1244
1245 return true;
1246 }
1247
1248 for (unsigned i = 0; i < this->type->components(); i++) {
1249 switch (this->type->base_type) {
1250 case GLSL_TYPE_UINT:
1251 if (this->value.u[i] != c->value.u[i])
1252 return false;
1253 break;
1254 case GLSL_TYPE_INT:
1255 if (this->value.i[i] != c->value.i[i])
1256 return false;
1257 break;
1258 case GLSL_TYPE_FLOAT:
1259 if (this->value.f[i] != c->value.f[i])
1260 return false;
1261 break;
1262 case GLSL_TYPE_BOOL:
1263 if (this->value.b[i] != c->value.b[i])
1264 return false;
1265 break;
1266 case GLSL_TYPE_DOUBLE:
1267 if (this->value.d[i] != c->value.d[i])
1268 return false;
1269 break;
1270 case GLSL_TYPE_UINT64:
1271 if (this->value.u64[i] != c->value.u64[i])
1272 return false;
1273 break;
1274 case GLSL_TYPE_INT64:
1275 if (this->value.i64[i] != c->value.i64[i])
1276 return false;
1277 break;
1278 default:
1279 assert(!"Should not get here.");
1280 return false;
1281 }
1282 }
1283
1284 return true;
1285 }
1286
1287 bool
1288 ir_constant::is_value(float f, int i) const
1289 {
1290 if (!this->type->is_scalar() && !this->type->is_vector())
1291 return false;
1292
1293 /* Only accept boolean values for 0/1. */
1294 if (int(bool(i)) != i && this->type->is_boolean())
1295 return false;
1296
1297 for (unsigned c = 0; c < this->type->vector_elements; c++) {
1298 switch (this->type->base_type) {
1299 case GLSL_TYPE_FLOAT:
1300 if (this->value.f[c] != f)
1301 return false;
1302 break;
1303 case GLSL_TYPE_INT:
1304 if (this->value.i[c] != i)
1305 return false;
1306 break;
1307 case GLSL_TYPE_UINT:
1308 if (this->value.u[c] != unsigned(i))
1309 return false;
1310 break;
1311 case GLSL_TYPE_BOOL:
1312 if (this->value.b[c] != bool(i))
1313 return false;
1314 break;
1315 case GLSL_TYPE_DOUBLE:
1316 if (this->value.d[c] != double(f))
1317 return false;
1318 break;
1319 case GLSL_TYPE_UINT64:
1320 if (this->value.u64[c] != uint64_t(i))
1321 return false;
1322 break;
1323 case GLSL_TYPE_INT64:
1324 if (this->value.i64[c] != i)
1325 return false;
1326 break;
1327 default:
1328 /* The only other base types are structures, arrays, and samplers.
1329 * Samplers cannot be constants, and the others should have been
1330 * filtered out above.
1331 */
1332 assert(!"Should not get here.");
1333 return false;
1334 }
1335 }
1336
1337 return true;
1338 }
1339
1340 bool
1341 ir_constant::is_zero() const
1342 {
1343 return is_value(0.0, 0);
1344 }
1345
1346 bool
1347 ir_constant::is_one() const
1348 {
1349 return is_value(1.0, 1);
1350 }
1351
1352 bool
1353 ir_constant::is_negative_one() const
1354 {
1355 return is_value(-1.0, -1);
1356 }
1357
1358 bool
1359 ir_constant::is_uint16_constant() const
1360 {
1361 if (!type->is_integer())
1362 return false;
1363
1364 return value.u[0] < (1 << 16);
1365 }
1366
1367 ir_loop::ir_loop()
1368 : ir_instruction(ir_type_loop)
1369 {
1370 }
1371
1372
1373 ir_dereference_variable::ir_dereference_variable(ir_variable *var)
1374 : ir_dereference(ir_type_dereference_variable)
1375 {
1376 assert(var != NULL);
1377
1378 this->var = var;
1379 this->type = var->type;
1380 }
1381
1382
1383 ir_dereference_array::ir_dereference_array(ir_rvalue *value,
1384 ir_rvalue *array_index)
1385 : ir_dereference(ir_type_dereference_array)
1386 {
1387 this->array_index = array_index;
1388 this->set_array(value);
1389 }
1390
1391
1392 ir_dereference_array::ir_dereference_array(ir_variable *var,
1393 ir_rvalue *array_index)
1394 : ir_dereference(ir_type_dereference_array)
1395 {
1396 void *ctx = ralloc_parent(var);
1397
1398 this->array_index = array_index;
1399 this->set_array(new(ctx) ir_dereference_variable(var));
1400 }
1401
1402
1403 void
1404 ir_dereference_array::set_array(ir_rvalue *value)
1405 {
1406 assert(value != NULL);
1407
1408 this->array = value;
1409
1410 const glsl_type *const vt = this->array->type;
1411
1412 if (vt->is_array()) {
1413 type = vt->fields.array;
1414 } else if (vt->is_matrix()) {
1415 type = vt->column_type();
1416 } else if (vt->is_vector()) {
1417 type = vt->get_base_type();
1418 }
1419 }
1420
1421
1422 ir_dereference_record::ir_dereference_record(ir_rvalue *value,
1423 const char *field)
1424 : ir_dereference(ir_type_dereference_record)
1425 {
1426 assert(value != NULL);
1427
1428 this->record = value;
1429 this->field = ralloc_strdup(this, field);
1430 this->type = this->record->type->field_type(field);
1431 }
1432
1433
1434 ir_dereference_record::ir_dereference_record(ir_variable *var,
1435 const char *field)
1436 : ir_dereference(ir_type_dereference_record)
1437 {
1438 void *ctx = ralloc_parent(var);
1439
1440 this->record = new(ctx) ir_dereference_variable(var);
1441 this->field = ralloc_strdup(this, field);
1442 this->type = this->record->type->field_type(field);
1443 }
1444
1445 bool
1446 ir_dereference::is_lvalue() const
1447 {
1448 ir_variable *var = this->variable_referenced();
1449
1450 /* Every l-value derference chain eventually ends in a variable.
1451 */
1452 if ((var == NULL) || var->data.read_only)
1453 return false;
1454
1455 /* From section 4.1.7 of the GLSL 4.40 spec:
1456 *
1457 * "Opaque variables cannot be treated as l-values; hence cannot
1458 * be used as out or inout function parameters, nor can they be
1459 * assigned into."
1460 */
1461 if (this->type->contains_opaque())
1462 return false;
1463
1464 return true;
1465 }
1466
1467
1468 static const char * const tex_opcode_strs[] = { "tex", "txb", "txl", "txd", "txf", "txf_ms", "txs", "lod", "tg4", "query_levels", "texture_samples", "samples_identical" };
1469
1470 const char *ir_texture::opcode_string()
1471 {
1472 assert((unsigned int) op < ARRAY_SIZE(tex_opcode_strs));
1473 return tex_opcode_strs[op];
1474 }
1475
1476 ir_texture_opcode
1477 ir_texture::get_opcode(const char *str)
1478 {
1479 const int count = sizeof(tex_opcode_strs) / sizeof(tex_opcode_strs[0]);
1480 for (int op = 0; op < count; op++) {
1481 if (strcmp(str, tex_opcode_strs[op]) == 0)
1482 return (ir_texture_opcode) op;
1483 }
1484 return (ir_texture_opcode) -1;
1485 }
1486
1487
1488 void
1489 ir_texture::set_sampler(ir_dereference *sampler, const glsl_type *type)
1490 {
1491 assert(sampler != NULL);
1492 assert(type != NULL);
1493 this->sampler = sampler;
1494 this->type = type;
1495
1496 if (this->op == ir_txs || this->op == ir_query_levels ||
1497 this->op == ir_texture_samples) {
1498 assert(type->base_type == GLSL_TYPE_INT);
1499 } else if (this->op == ir_lod) {
1500 assert(type->vector_elements == 2);
1501 assert(type->base_type == GLSL_TYPE_FLOAT);
1502 } else if (this->op == ir_samples_identical) {
1503 assert(type == glsl_type::bool_type);
1504 assert(sampler->type->base_type == GLSL_TYPE_SAMPLER);
1505 assert(sampler->type->sampler_dimensionality == GLSL_SAMPLER_DIM_MS);
1506 } else {
1507 assert(sampler->type->sampled_type == (int) type->base_type);
1508 if (sampler->type->sampler_shadow)
1509 assert(type->vector_elements == 4 || type->vector_elements == 1);
1510 else
1511 assert(type->vector_elements == 4);
1512 }
1513 }
1514
1515
1516 void
1517 ir_swizzle::init_mask(const unsigned *comp, unsigned count)
1518 {
1519 assert((count >= 1) && (count <= 4));
1520
1521 memset(&this->mask, 0, sizeof(this->mask));
1522 this->mask.num_components = count;
1523
1524 unsigned dup_mask = 0;
1525 switch (count) {
1526 case 4:
1527 assert(comp[3] <= 3);
1528 dup_mask |= (1U << comp[3])
1529 & ((1U << comp[0]) | (1U << comp[1]) | (1U << comp[2]));
1530 this->mask.w = comp[3];
1531
1532 case 3:
1533 assert(comp[2] <= 3);
1534 dup_mask |= (1U << comp[2])
1535 & ((1U << comp[0]) | (1U << comp[1]));
1536 this->mask.z = comp[2];
1537
1538 case 2:
1539 assert(comp[1] <= 3);
1540 dup_mask |= (1U << comp[1])
1541 & ((1U << comp[0]));
1542 this->mask.y = comp[1];
1543
1544 case 1:
1545 assert(comp[0] <= 3);
1546 this->mask.x = comp[0];
1547 }
1548
1549 this->mask.has_duplicates = dup_mask != 0;
1550
1551 /* Based on the number of elements in the swizzle and the base type
1552 * (i.e., float, int, unsigned, or bool) of the vector being swizzled,
1553 * generate the type of the resulting value.
1554 */
1555 type = glsl_type::get_instance(val->type->base_type, mask.num_components, 1);
1556 }
1557
1558 ir_swizzle::ir_swizzle(ir_rvalue *val, unsigned x, unsigned y, unsigned z,
1559 unsigned w, unsigned count)
1560 : ir_rvalue(ir_type_swizzle), val(val)
1561 {
1562 const unsigned components[4] = { x, y, z, w };
1563 this->init_mask(components, count);
1564 }
1565
1566 ir_swizzle::ir_swizzle(ir_rvalue *val, const unsigned *comp,
1567 unsigned count)
1568 : ir_rvalue(ir_type_swizzle), val(val)
1569 {
1570 this->init_mask(comp, count);
1571 }
1572
1573 ir_swizzle::ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask)
1574 : ir_rvalue(ir_type_swizzle)
1575 {
1576 this->val = val;
1577 this->mask = mask;
1578 this->type = glsl_type::get_instance(val->type->base_type,
1579 mask.num_components, 1);
1580 }
1581
1582 #define X 1
1583 #define R 5
1584 #define S 9
1585 #define I 13
1586
1587 ir_swizzle *
1588 ir_swizzle::create(ir_rvalue *val, const char *str, unsigned vector_length)
1589 {
1590 void *ctx = ralloc_parent(val);
1591
1592 /* For each possible swizzle character, this table encodes the value in
1593 * \c idx_map that represents the 0th element of the vector. For invalid
1594 * swizzle characters (e.g., 'k'), a special value is used that will allow
1595 * detection of errors.
1596 */
1597 static const unsigned char base_idx[26] = {
1598 /* a b c d e f g h i j k l m */
1599 R, R, I, I, I, I, R, I, I, I, I, I, I,
1600 /* n o p q r s t u v w x y z */
1601 I, I, S, S, R, S, S, I, I, X, X, X, X
1602 };
1603
1604 /* Each valid swizzle character has an entry in the previous table. This
1605 * table encodes the base index encoded in the previous table plus the actual
1606 * index of the swizzle character. When processing swizzles, the first
1607 * character in the string is indexed in the previous table. Each character
1608 * in the string is indexed in this table, and the value found there has the
1609 * value form the first table subtracted. The result must be on the range
1610 * [0,3].
1611 *
1612 * For example, the string "wzyx" will get X from the first table. Each of
1613 * the charcaters will get X+3, X+2, X+1, and X+0 from this table. After
1614 * subtraction, the swizzle values are { 3, 2, 1, 0 }.
1615 *
1616 * The string "wzrg" will get X from the first table. Each of the characters
1617 * will get X+3, X+2, R+0, and R+1 from this table. After subtraction, the
1618 * swizzle values are { 3, 2, 4, 5 }. Since 4 and 5 are outside the range
1619 * [0,3], the error is detected.
1620 */
1621 static const unsigned char idx_map[26] = {
1622 /* a b c d e f g h i j k l m */
1623 R+3, R+2, 0, 0, 0, 0, R+1, 0, 0, 0, 0, 0, 0,
1624 /* n o p q r s t u v w x y z */
1625 0, 0, S+2, S+3, R+0, S+0, S+1, 0, 0, X+3, X+0, X+1, X+2
1626 };
1627
1628 int swiz_idx[4] = { 0, 0, 0, 0 };
1629 unsigned i;
1630
1631
1632 /* Validate the first character in the swizzle string and look up the base
1633 * index value as described above.
1634 */
1635 if ((str[0] < 'a') || (str[0] > 'z'))
1636 return NULL;
1637
1638 const unsigned base = base_idx[str[0] - 'a'];
1639
1640
1641 for (i = 0; (i < 4) && (str[i] != '\0'); i++) {
1642 /* Validate the next character, and, as described above, convert it to a
1643 * swizzle index.
1644 */
1645 if ((str[i] < 'a') || (str[i] > 'z'))
1646 return NULL;
1647
1648 swiz_idx[i] = idx_map[str[i] - 'a'] - base;
1649 if ((swiz_idx[i] < 0) || (swiz_idx[i] >= (int) vector_length))
1650 return NULL;
1651 }
1652
1653 if (str[i] != '\0')
1654 return NULL;
1655
1656 return new(ctx) ir_swizzle(val, swiz_idx[0], swiz_idx[1], swiz_idx[2],
1657 swiz_idx[3], i);
1658 }
1659
1660 #undef X
1661 #undef R
1662 #undef S
1663 #undef I
1664
1665 ir_variable *
1666 ir_swizzle::variable_referenced() const
1667 {
1668 return this->val->variable_referenced();
1669 }
1670
1671
1672 bool ir_variable::temporaries_allocate_names = false;
1673
1674 const char ir_variable::tmp_name[] = "compiler_temp";
1675
1676 ir_variable::ir_variable(const struct glsl_type *type, const char *name,
1677 ir_variable_mode mode)
1678 : ir_instruction(ir_type_variable)
1679 {
1680 this->type = type;
1681
1682 if (mode == ir_var_temporary && !ir_variable::temporaries_allocate_names)
1683 name = NULL;
1684
1685 /* The ir_variable clone method may call this constructor with name set to
1686 * tmp_name.
1687 */
1688 assert(name != NULL
1689 || mode == ir_var_temporary
1690 || mode == ir_var_function_in
1691 || mode == ir_var_function_out
1692 || mode == ir_var_function_inout);
1693 assert(name != ir_variable::tmp_name
1694 || mode == ir_var_temporary);
1695 if (mode == ir_var_temporary
1696 && (name == NULL || name == ir_variable::tmp_name)) {
1697 this->name = ir_variable::tmp_name;
1698 } else if (name == NULL ||
1699 strlen(name) < ARRAY_SIZE(this->name_storage)) {
1700 strcpy(this->name_storage, name ? name : "");
1701 this->name = this->name_storage;
1702 } else {
1703 this->name = ralloc_strdup(this, name);
1704 }
1705
1706 this->u.max_ifc_array_access = NULL;
1707
1708 this->data.explicit_location = false;
1709 this->data.has_initializer = false;
1710 this->data.location = -1;
1711 this->data.location_frac = 0;
1712 this->data.binding = 0;
1713 this->data.warn_extension_index = 0;
1714 this->constant_value = NULL;
1715 this->constant_initializer = NULL;
1716 this->data.origin_upper_left = false;
1717 this->data.pixel_center_integer = false;
1718 this->data.depth_layout = ir_depth_layout_none;
1719 this->data.used = false;
1720 this->data.always_active_io = false;
1721 this->data.read_only = false;
1722 this->data.centroid = false;
1723 this->data.sample = false;
1724 this->data.patch = false;
1725 this->data.invariant = false;
1726 this->data.how_declared = ir_var_declared_normally;
1727 this->data.mode = mode;
1728 this->data.interpolation = INTERP_MODE_NONE;
1729 this->data.max_array_access = -1;
1730 this->data.offset = 0;
1731 this->data.precision = GLSL_PRECISION_NONE;
1732 this->data.image_read_only = false;
1733 this->data.image_write_only = false;
1734 this->data.image_coherent = false;
1735 this->data.image_volatile = false;
1736 this->data.image_restrict = false;
1737 this->data.from_ssbo_unsized_array = false;
1738 this->data.fb_fetch_output = false;
1739
1740 if (type != NULL) {
1741 if (type->base_type == GLSL_TYPE_SAMPLER)
1742 this->data.read_only = true;
1743
1744 if (type->is_interface())
1745 this->init_interface_type(type);
1746 else if (type->without_array()->is_interface())
1747 this->init_interface_type(type->without_array());
1748 }
1749 }
1750
1751
1752 const char *
1753 interpolation_string(unsigned interpolation)
1754 {
1755 switch (interpolation) {
1756 case INTERP_MODE_NONE: return "no";
1757 case INTERP_MODE_SMOOTH: return "smooth";
1758 case INTERP_MODE_FLAT: return "flat";
1759 case INTERP_MODE_NOPERSPECTIVE: return "noperspective";
1760 }
1761
1762 assert(!"Should not get here.");
1763 return "";
1764 }
1765
1766 const char *const ir_variable::warn_extension_table[] = {
1767 "",
1768 "GL_ARB_shader_stencil_export",
1769 "GL_AMD_shader_stencil_export",
1770 };
1771
1772 void
1773 ir_variable::enable_extension_warning(const char *extension)
1774 {
1775 for (unsigned i = 0; i < ARRAY_SIZE(warn_extension_table); i++) {
1776 if (strcmp(warn_extension_table[i], extension) == 0) {
1777 this->data.warn_extension_index = i;
1778 return;
1779 }
1780 }
1781
1782 assert(!"Should not get here.");
1783 this->data.warn_extension_index = 0;
1784 }
1785
1786 const char *
1787 ir_variable::get_extension_warning() const
1788 {
1789 return this->data.warn_extension_index == 0
1790 ? NULL : warn_extension_table[this->data.warn_extension_index];
1791 }
1792
1793 ir_function_signature::ir_function_signature(const glsl_type *return_type,
1794 builtin_available_predicate b)
1795 : ir_instruction(ir_type_function_signature),
1796 return_type(return_type), is_defined(false),
1797 intrinsic_id(ir_intrinsic_invalid), builtin_avail(b), _function(NULL)
1798 {
1799 this->origin = NULL;
1800 }
1801
1802
1803 bool
1804 ir_function_signature::is_builtin() const
1805 {
1806 return builtin_avail != NULL;
1807 }
1808
1809
1810 bool
1811 ir_function_signature::is_builtin_available(const _mesa_glsl_parse_state *state) const
1812 {
1813 /* We can't call the predicate without a state pointer, so just say that
1814 * the signature is available. At compile time, we need the filtering,
1815 * but also receive a valid state pointer. At link time, we're resolving
1816 * imported built-in prototypes to their definitions, which will always
1817 * be an exact match. So we can skip the filtering.
1818 */
1819 if (state == NULL)
1820 return true;
1821
1822 assert(builtin_avail != NULL);
1823 return builtin_avail(state);
1824 }
1825
1826
1827 static bool
1828 modes_match(unsigned a, unsigned b)
1829 {
1830 if (a == b)
1831 return true;
1832
1833 /* Accept "in" vs. "const in" */
1834 if ((a == ir_var_const_in && b == ir_var_function_in) ||
1835 (b == ir_var_const_in && a == ir_var_function_in))
1836 return true;
1837
1838 return false;
1839 }
1840
1841
1842 const char *
1843 ir_function_signature::qualifiers_match(exec_list *params)
1844 {
1845 /* check that the qualifiers match. */
1846 foreach_two_lists(a_node, &this->parameters, b_node, params) {
1847 ir_variable *a = (ir_variable *) a_node;
1848 ir_variable *b = (ir_variable *) b_node;
1849
1850 if (a->data.read_only != b->data.read_only ||
1851 !modes_match(a->data.mode, b->data.mode) ||
1852 a->data.interpolation != b->data.interpolation ||
1853 a->data.centroid != b->data.centroid ||
1854 a->data.sample != b->data.sample ||
1855 a->data.patch != b->data.patch ||
1856 a->data.image_read_only != b->data.image_read_only ||
1857 a->data.image_write_only != b->data.image_write_only ||
1858 a->data.image_coherent != b->data.image_coherent ||
1859 a->data.image_volatile != b->data.image_volatile ||
1860 a->data.image_restrict != b->data.image_restrict) {
1861
1862 /* parameter a's qualifiers don't match */
1863 return a->name;
1864 }
1865 }
1866 return NULL;
1867 }
1868
1869
1870 void
1871 ir_function_signature::replace_parameters(exec_list *new_params)
1872 {
1873 /* Destroy all of the previous parameter information. If the previous
1874 * parameter information comes from the function prototype, it may either
1875 * specify incorrect parameter names or not have names at all.
1876 */
1877 new_params->move_nodes_to(&parameters);
1878 }
1879
1880
1881 ir_function::ir_function(const char *name)
1882 : ir_instruction(ir_type_function)
1883 {
1884 this->subroutine_index = -1;
1885 this->name = ralloc_strdup(this, name);
1886 }
1887
1888
1889 bool
1890 ir_function::has_user_signature()
1891 {
1892 foreach_in_list(ir_function_signature, sig, &this->signatures) {
1893 if (!sig->is_builtin())
1894 return true;
1895 }
1896 return false;
1897 }
1898
1899
1900 ir_rvalue *
1901 ir_rvalue::error_value(void *mem_ctx)
1902 {
1903 ir_rvalue *v = new(mem_ctx) ir_rvalue(ir_type_unset);
1904
1905 v->type = glsl_type::error_type;
1906 return v;
1907 }
1908
1909
1910 void
1911 visit_exec_list(exec_list *list, ir_visitor *visitor)
1912 {
1913 foreach_in_list_safe(ir_instruction, node, list) {
1914 node->accept(visitor);
1915 }
1916 }
1917
1918
1919 static void
1920 steal_memory(ir_instruction *ir, void *new_ctx)
1921 {
1922 ir_variable *var = ir->as_variable();
1923 ir_function *fn = ir->as_function();
1924 ir_constant *constant = ir->as_constant();
1925 if (var != NULL && var->constant_value != NULL)
1926 steal_memory(var->constant_value, ir);
1927
1928 if (var != NULL && var->constant_initializer != NULL)
1929 steal_memory(var->constant_initializer, ir);
1930
1931 if (fn != NULL && fn->subroutine_types)
1932 ralloc_steal(new_ctx, fn->subroutine_types);
1933
1934 /* The components of aggregate constants are not visited by the normal
1935 * visitor, so steal their values by hand.
1936 */
1937 if (constant != NULL) {
1938 if (constant->type->is_record()) {
1939 foreach_in_list(ir_constant, field, &constant->components) {
1940 steal_memory(field, ir);
1941 }
1942 } else if (constant->type->is_array()) {
1943 for (unsigned int i = 0; i < constant->type->length; i++) {
1944 steal_memory(constant->array_elements[i], ir);
1945 }
1946 }
1947 }
1948
1949 ralloc_steal(new_ctx, ir);
1950 }
1951
1952
1953 void
1954 reparent_ir(exec_list *list, void *mem_ctx)
1955 {
1956 foreach_in_list(ir_instruction, node, list) {
1957 visit_tree(node, steal_memory, mem_ctx);
1958 }
1959 }
1960
1961
1962 static ir_rvalue *
1963 try_min_one(ir_rvalue *ir)
1964 {
1965 ir_expression *expr = ir->as_expression();
1966
1967 if (!expr || expr->operation != ir_binop_min)
1968 return NULL;
1969
1970 if (expr->operands[0]->is_one())
1971 return expr->operands[1];
1972
1973 if (expr->operands[1]->is_one())
1974 return expr->operands[0];
1975
1976 return NULL;
1977 }
1978
1979 static ir_rvalue *
1980 try_max_zero(ir_rvalue *ir)
1981 {
1982 ir_expression *expr = ir->as_expression();
1983
1984 if (!expr || expr->operation != ir_binop_max)
1985 return NULL;
1986
1987 if (expr->operands[0]->is_zero())
1988 return expr->operands[1];
1989
1990 if (expr->operands[1]->is_zero())
1991 return expr->operands[0];
1992
1993 return NULL;
1994 }
1995
1996 ir_rvalue *
1997 ir_rvalue::as_rvalue_to_saturate()
1998 {
1999 ir_expression *expr = this->as_expression();
2000
2001 if (!expr)
2002 return NULL;
2003
2004 ir_rvalue *max_zero = try_max_zero(expr);
2005 if (max_zero) {
2006 return try_min_one(max_zero);
2007 } else {
2008 ir_rvalue *min_one = try_min_one(expr);
2009 if (min_one) {
2010 return try_max_zero(min_one);
2011 }
2012 }
2013
2014 return NULL;
2015 }
2016
2017
2018 unsigned
2019 vertices_per_prim(GLenum prim)
2020 {
2021 switch (prim) {
2022 case GL_POINTS:
2023 return 1;
2024 case GL_LINES:
2025 return 2;
2026 case GL_TRIANGLES:
2027 return 3;
2028 case GL_LINES_ADJACENCY:
2029 return 4;
2030 case GL_TRIANGLES_ADJACENCY:
2031 return 6;
2032 default:
2033 assert(!"Bad primitive");
2034 return 3;
2035 }
2036 }
2037
2038 /**
2039 * Generate a string describing the mode of a variable
2040 */
2041 const char *
2042 mode_string(const ir_variable *var)
2043 {
2044 switch (var->data.mode) {
2045 case ir_var_auto:
2046 return (var->data.read_only) ? "global constant" : "global variable";
2047
2048 case ir_var_uniform:
2049 return "uniform";
2050
2051 case ir_var_shader_storage:
2052 return "buffer";
2053
2054 case ir_var_shader_in:
2055 return "shader input";
2056
2057 case ir_var_shader_out:
2058 return "shader output";
2059
2060 case ir_var_function_in:
2061 case ir_var_const_in:
2062 return "function input";
2063
2064 case ir_var_function_out:
2065 return "function output";
2066
2067 case ir_var_function_inout:
2068 return "function inout";
2069
2070 case ir_var_system_value:
2071 return "shader input";
2072
2073 case ir_var_temporary:
2074 return "compiler temporary";
2075
2076 case ir_var_mode_count:
2077 break;
2078 }
2079
2080 assert(!"Should not get here.");
2081 return "invalid variable";
2082 }