projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
glsl, glsl_to_tgsi: fix sampler/image constants
[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 "ir.h"
25 #include "compiler/glsl_types.h"
26 #include "glsl_parser_extras.h"
27
28
29 ir_rvalue::ir_rvalue(enum ir_node_type t)
30 : ir_instruction(t)
31 {
32 this->type = glsl_type::error_type;
33 }
34
35 bool ir_rvalue::is_zero() const
36 {
37 return false;
38 }
39
40 bool ir_rvalue::is_one() const
41 {
42 return false;
43 }
44
45 bool ir_rvalue::is_negative_one() const
46 {
47 return false;
48 }
49
50 /**
51 * Modify the swizzle make to move one component to another
52 *
53 * \param m IR swizzle to be modified
54 * \param from Component in the RHS that is to be swizzled
55 * \param to Desired swizzle location of \c from
56 */
57 static void
58 update_rhs_swizzle(ir_swizzle_mask &m, unsigned from, unsigned to)
59 {
60 switch (to) {
61 case 0: m.x = from; break;
62 case 1: m.y = from; break;
63 case 2: m.z = from; break;
64 case 3: m.w = from; break;
65 default: assert(!"Should not get here.");
66 }
67 }
68
69 void
70 ir_assignment::set_lhs(ir_rvalue *lhs)
71 {
72 void *mem_ctx = this;
73 bool swizzled = false;
74
75 while (lhs != NULL) {
76 ir_swizzle *swiz = lhs->as_swizzle();
77
78 if (swiz == NULL)
79 break;
80
81 unsigned write_mask = 0;
82 ir_swizzle_mask rhs_swiz = { 0, 0, 0, 0, 0, 0 };
83
84 for (unsigned i = 0; i < swiz->mask.num_components; i++) {
85 unsigned c = 0;
86
87 switch (i) {
88 case 0: c = swiz->mask.x; break;
89 case 1: c = swiz->mask.y; break;
90 case 2: c = swiz->mask.z; break;
91 case 3: c = swiz->mask.w; break;
92 default: assert(!"Should not get here.");
93 }
94
95 write_mask |= (((this->write_mask >> i) & 1) << c);
96 update_rhs_swizzle(rhs_swiz, i, c);
97 rhs_swiz.num_components = swiz->val->type->vector_elements;
98 }
99
100 this->write_mask = write_mask;
101 lhs = swiz->val;
102
103 this->rhs = new(mem_ctx) ir_swizzle(this->rhs, rhs_swiz);
104 swizzled = true;
105 }
106
107 if (swizzled) {
108 /* Now, RHS channels line up with the LHS writemask. Collapse it
109 * to just the channels that will be written.
110 */
111 ir_swizzle_mask rhs_swiz = { 0, 0, 0, 0, 0, 0 };
112 int rhs_chan = 0;
113 for (int i = 0; i < 4; i++) {
114 if (write_mask & (1 << i))
115 update_rhs_swizzle(rhs_swiz, i, rhs_chan++);
116 }
117 rhs_swiz.num_components = rhs_chan;
118 this->rhs = new(mem_ctx) ir_swizzle(this->rhs, rhs_swiz);
119 }
120
121 assert((lhs == NULL) || lhs->as_dereference());
122
123 this->lhs = (ir_dereference *) lhs;
124 }
125
126 ir_variable *
127 ir_assignment::whole_variable_written()
128 {
129 ir_variable *v = this->lhs->whole_variable_referenced();
130
131 if (v == NULL)
132 return NULL;
133
134 if (v->type->is_scalar())
135 return v;
136
137 if (v->type->is_vector()) {
138 const unsigned mask = (1U << v->type->vector_elements) - 1;
139
140 if (mask != this->write_mask)
141 return NULL;
142 }
143
144 /* Either all the vector components are assigned or the variable is some
145 * composite type (and the whole thing is assigned.
146 */
147 return v;
148 }
149
150 ir_assignment::ir_assignment(ir_dereference *lhs, ir_rvalue *rhs,
151 ir_rvalue *condition, unsigned write_mask)
152 : ir_instruction(ir_type_assignment)
153 {
154 this->condition = condition;
155 this->rhs = rhs;
156 this->lhs = lhs;
157 this->write_mask = write_mask;
158
159 if (lhs->type->is_scalar() || lhs->type->is_vector()) {
160 int lhs_components = 0;
161 for (int i = 0; i < 4; i++) {
162 if (write_mask & (1 << i))
163 lhs_components++;
164 }
165
166 assert(lhs_components == this->rhs->type->vector_elements);
167 }
168 }
169
170 ir_assignment::ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs,
171 ir_rvalue *condition)
172 : ir_instruction(ir_type_assignment)
173 {
174 this->condition = condition;
175 this->rhs = rhs;
176
177 /* If the RHS is a vector type, assume that all components of the vector
178 * type are being written to the LHS. The write mask comes from the RHS
179 * because we can have a case where the LHS is a vec4 and the RHS is a
180 * vec3. In that case, the assignment is:
181 *
182 * (assign (...) (xyz) (var_ref lhs) (var_ref rhs))
183 */
184 if (rhs->type->is_vector())
185 this->write_mask = (1U << rhs->type->vector_elements) - 1;
186 else if (rhs->type->is_scalar())
187 this->write_mask = 1;
188 else
189 this->write_mask = 0;
190
191 this->set_lhs(lhs);
192 }
193
194 ir_expression::ir_expression(int op, const struct glsl_type *type,
195 ir_rvalue *op0, ir_rvalue *op1,
196 ir_rvalue *op2, ir_rvalue *op3)
197 : ir_rvalue(ir_type_expression)
198 {
199 this->type = type;
200 this->operation = ir_expression_operation(op);
201 this->operands[0] = op0;
202 this->operands[1] = op1;
203 this->operands[2] = op2;
204 this->operands[3] = op3;
205 init_num_operands();
206
207 #ifndef NDEBUG
208 for (unsigned i = num_operands; i < 4; i++) {
209 assert(this->operands[i] == NULL);
210 }
211
212 for (unsigned i = 0; i < num_operands; i++) {
213 assert(this->operands[i] != NULL);
214 }
215 #endif
216 }
217
218 ir_expression::ir_expression(int op, ir_rvalue *op0)
219 : ir_rvalue(ir_type_expression)
220 {
221 this->operation = ir_expression_operation(op);
222 this->operands[0] = op0;
223 this->operands[1] = NULL;
224 this->operands[2] = NULL;
225 this->operands[3] = NULL;
226
227 assert(op <= ir_last_unop);
228 init_num_operands();
229 assert(num_operands == 1);
230 assert(this->operands[0]);
231
232 switch (this->operation) {
233 case ir_unop_bit_not:
234 case ir_unop_logic_not:
235 case ir_unop_neg:
236 case ir_unop_abs:
237 case ir_unop_sign:
238 case ir_unop_rcp:
239 case ir_unop_rsq:
240 case ir_unop_sqrt:
241 case ir_unop_exp:
242 case ir_unop_log:
243 case ir_unop_exp2:
244 case ir_unop_log2:
245 case ir_unop_trunc:
246 case ir_unop_ceil:
247 case ir_unop_floor:
248 case ir_unop_fract:
249 case ir_unop_round_even:
250 case ir_unop_sin:
251 case ir_unop_cos:
252 case ir_unop_dFdx:
253 case ir_unop_dFdx_coarse:
254 case ir_unop_dFdx_fine:
255 case ir_unop_dFdy:
256 case ir_unop_dFdy_coarse:
257 case ir_unop_dFdy_fine:
258 case ir_unop_bitfield_reverse:
259 case ir_unop_interpolate_at_centroid:
260 case ir_unop_saturate:
261 this->type = op0->type;
262 break;
263
264 case ir_unop_f2i:
265 case ir_unop_b2i:
266 case ir_unop_u2i:
267 case ir_unop_d2i:
268 case ir_unop_bitcast_f2i:
269 case ir_unop_bit_count:
270 case ir_unop_find_msb:
271 case ir_unop_find_lsb:
272 case ir_unop_subroutine_to_int:
273 case ir_unop_i642i:
274 case ir_unop_u642i:
275 this->type = glsl_type::get_instance(GLSL_TYPE_INT,
276 op0->type->vector_elements, 1);
277 break;
278
279 case ir_unop_b2f:
280 case ir_unop_i2f:
281 case ir_unop_u2f:
282 case ir_unop_d2f:
283 case ir_unop_bitcast_i2f:
284 case ir_unop_bitcast_u2f:
285 case ir_unop_i642f:
286 case ir_unop_u642f:
287 this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT,
288 op0->type->vector_elements, 1);
289 break;
290
291 case ir_unop_f2b:
292 case ir_unop_i2b:
293 case ir_unop_d2b:
294 case ir_unop_i642b:
295 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL,
296 op0->type->vector_elements, 1);
297 break;
298
299 case ir_unop_f2d:
300 case ir_unop_i2d:
301 case ir_unop_u2d:
302 case ir_unop_i642d:
303 case ir_unop_u642d:
304 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE,
305 op0->type->vector_elements, 1);
306 break;
307
308 case ir_unop_i2u:
309 case ir_unop_f2u:
310 case ir_unop_d2u:
311 case ir_unop_bitcast_f2u:
312 case ir_unop_i642u:
313 case ir_unop_u642u:
314 this->type = glsl_type::get_instance(GLSL_TYPE_UINT,
315 op0->type->vector_elements, 1);
316 break;
317
318 case ir_unop_i2i64:
319 case ir_unop_u2i64:
320 case ir_unop_b2i64:
321 case ir_unop_f2i64:
322 case ir_unop_d2i64:
323 case ir_unop_u642i64:
324 this->type = glsl_type::get_instance(GLSL_TYPE_INT64,
325 op0->type->vector_elements, 1);
326 break;
327
328 case ir_unop_i2u64:
329 case ir_unop_u2u64:
330 case ir_unop_f2u64:
331 case ir_unop_d2u64:
332 case ir_unop_i642u64:
333 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64,
334 op0->type->vector_elements, 1);
335 break;
336 case ir_unop_noise:
337 this->type = glsl_type::float_type;
338 break;
339
340 case ir_unop_unpack_double_2x32:
341 case ir_unop_unpack_uint_2x32:
342 this->type = glsl_type::uvec2_type;
343 break;
344
345 case ir_unop_unpack_int_2x32:
346 this->type = glsl_type::ivec2_type;
347 break;
348
349 case ir_unop_pack_snorm_2x16:
350 case ir_unop_pack_snorm_4x8:
351 case ir_unop_pack_unorm_2x16:
352 case ir_unop_pack_unorm_4x8:
353 case ir_unop_pack_half_2x16:
354 this->type = glsl_type::uint_type;
355 break;
356
357 case ir_unop_pack_double_2x32:
358 this->type = glsl_type::double_type;
359 break;
360
361 case ir_unop_pack_int_2x32:
362 this->type = glsl_type::int64_t_type;
363 break;
364
365 case ir_unop_pack_uint_2x32:
366 this->type = glsl_type::uint64_t_type;
367 break;
368
369 case ir_unop_unpack_snorm_2x16:
370 case ir_unop_unpack_unorm_2x16:
371 case ir_unop_unpack_half_2x16:
372 this->type = glsl_type::vec2_type;
373 break;
374
375 case ir_unop_unpack_snorm_4x8:
376 case ir_unop_unpack_unorm_4x8:
377 this->type = glsl_type::vec4_type;
378 break;
379
380 case ir_unop_unpack_sampler_2x32:
381 case ir_unop_unpack_image_2x32:
382 this->type = glsl_type::uvec2_type;
383 break;
384
385 case ir_unop_pack_sampler_2x32:
386 case ir_unop_pack_image_2x32:
387 this->type = op0->type;
388 break;
389
390 case ir_unop_frexp_sig:
391 this->type = op0->type;
392 break;
393 case ir_unop_frexp_exp:
394 this->type = glsl_type::get_instance(GLSL_TYPE_INT,
395 op0->type->vector_elements, 1);
396 break;
397
398 case ir_unop_get_buffer_size:
399 case ir_unop_ssbo_unsized_array_length:
400 this->type = glsl_type::int_type;
401 break;
402
403 case ir_unop_bitcast_i642d:
404 case ir_unop_bitcast_u642d:
405 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE,
406 op0->type->vector_elements, 1);
407 break;
408
409 case ir_unop_bitcast_d2i64:
410 this->type = glsl_type::get_instance(GLSL_TYPE_INT64,
411 op0->type->vector_elements, 1);
412 break;
413 case ir_unop_bitcast_d2u64:
414 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64,
415 op0->type->vector_elements, 1);
416 break;
417
418 default:
419 assert(!"not reached: missing automatic type setup for ir_expression");
420 this->type = op0->type;
421 break;
422 }
423 }
424
425 ir_expression::ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1)
426 : ir_rvalue(ir_type_expression)
427 {
428 this->operation = ir_expression_operation(op);
429 this->operands[0] = op0;
430 this->operands[1] = op1;
431 this->operands[2] = NULL;
432 this->operands[3] = NULL;
433
434 assert(op > ir_last_unop);
435 init_num_operands();
436 assert(num_operands == 2);
437 for (unsigned i = 0; i < num_operands; i++) {
438 assert(this->operands[i] != NULL);
439 }
440
441 switch (this->operation) {
442 case ir_binop_all_equal:
443 case ir_binop_any_nequal:
444 this->type = glsl_type::bool_type;
445 break;
446
447 case ir_binop_add:
448 case ir_binop_sub:
449 case ir_binop_min:
450 case ir_binop_max:
451 case ir_binop_pow:
452 case ir_binop_mul:
453 case ir_binop_div:
454 case ir_binop_mod:
455 if (op0->type->is_scalar()) {
456 this->type = op1->type;
457 } else if (op1->type->is_scalar()) {
458 this->type = op0->type;
459 } else {
460 if (this->operation == ir_binop_mul) {
461 this->type = glsl_type::get_mul_type(op0->type, op1->type);
462 } else {
463 assert(op0->type == op1->type);
464 this->type = op0->type;
465 }
466 }
467 break;
468
469 case ir_binop_logic_and:
470 case ir_binop_logic_xor:
471 case ir_binop_logic_or:
472 case ir_binop_bit_and:
473 case ir_binop_bit_xor:
474 case ir_binop_bit_or:
475 assert(!op0->type->is_matrix());
476 assert(!op1->type->is_matrix());
477 if (op0->type->is_scalar()) {
478 this->type = op1->type;
479 } else if (op1->type->is_scalar()) {
480 this->type = op0->type;
481 } else {
482 assert(op0->type->vector_elements == op1->type->vector_elements);
483 this->type = op0->type;
484 }
485 break;
486
487 case ir_binop_equal:
488 case ir_binop_nequal:
489 case ir_binop_gequal:
490 case ir_binop_less:
491 assert(op0->type == op1->type);
492 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL,
493 op0->type->vector_elements, 1);
494 break;
495
496 case ir_binop_dot:
497 this->type = op0->type->get_base_type();
498 break;
499
500 case ir_binop_imul_high:
501 case ir_binop_carry:
502 case ir_binop_borrow:
503 case ir_binop_lshift:
504 case ir_binop_rshift:
505 case ir_binop_ldexp:
506 case ir_binop_interpolate_at_offset:
507 case ir_binop_interpolate_at_sample:
508 this->type = op0->type;
509 break;
510
511 case ir_binop_vector_extract:
512 this->type = op0->type->get_scalar_type();
513 break;
514
515 default:
516 assert(!"not reached: missing automatic type setup for ir_expression");
517 this->type = glsl_type::float_type;
518 }
519 }
520
521 ir_expression::ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1,
522 ir_rvalue *op2)
523 : ir_rvalue(ir_type_expression)
524 {
525 this->operation = ir_expression_operation(op);
526 this->operands[0] = op0;
527 this->operands[1] = op1;
528 this->operands[2] = op2;
529 this->operands[3] = NULL;
530
531 assert(op > ir_last_binop && op <= ir_last_triop);
532 init_num_operands();
533 assert(num_operands == 3);
534 for (unsigned i = 0; i < num_operands; i++) {
535 assert(this->operands[i] != NULL);
536 }
537
538 switch (this->operation) {
539 case ir_triop_fma:
540 case ir_triop_lrp:
541 case ir_triop_bitfield_extract:
542 case ir_triop_vector_insert:
543 this->type = op0->type;
544 break;
545
546 case ir_triop_csel:
547 this->type = op1->type;
548 break;
549
550 default:
551 assert(!"not reached: missing automatic type setup for ir_expression");
552 this->type = glsl_type::float_type;
553 }
554 }
555
556 /**
557 * This is only here for ir_reader to used for testing purposes. Please use
558 * the precomputed num_operands field if you need the number of operands.
559 */
560 unsigned
561 ir_expression::get_num_operands(ir_expression_operation op)
562 {
563 assert(op <= ir_last_opcode);
564
565 if (op <= ir_last_unop)
566 return 1;
567
568 if (op <= ir_last_binop)
569 return 2;
570
571 if (op <= ir_last_triop)
572 return 3;
573
574 if (op <= ir_last_quadop)
575 return 4;
576
577 unreachable("Could not calculate number of operands");
578 }
579
580 #include "ir_expression_operation_strings.h"
581
582 const char*
583 depth_layout_string(ir_depth_layout layout)
584 {
585 switch(layout) {
586 case ir_depth_layout_none: return "";
587 case ir_depth_layout_any: return "depth_any";
588 case ir_depth_layout_greater: return "depth_greater";
589 case ir_depth_layout_less: return "depth_less";
590 case ir_depth_layout_unchanged: return "depth_unchanged";
591
592 default:
593 assert(0);
594 return "";
595 }
596 }
597
598 ir_expression_operation
599 ir_expression::get_operator(const char *str)
600 {
601 for (int op = 0; op <= int(ir_last_opcode); op++) {
602 if (strcmp(str, ir_expression_operation_strings[op]) == 0)
603 return (ir_expression_operation) op;
604 }
605 return (ir_expression_operation) -1;
606 }
607
608 ir_variable *
609 ir_expression::variable_referenced() const
610 {
611 switch (operation) {
612 case ir_binop_vector_extract:
613 case ir_triop_vector_insert:
614 /* We get these for things like a[0] where a is a vector type. In these
615 * cases we want variable_referenced() to return the actual vector
616 * variable this is wrapping.
617 */
618 return operands[0]->variable_referenced();
619 default:
620 return ir_rvalue::variable_referenced();
621 }
622 }
623
624 ir_constant::ir_constant()
625 : ir_rvalue(ir_type_constant)
626 {
627 this->const_elements = NULL;
628 }
629
630 ir_constant::ir_constant(const struct glsl_type *type,
631 const ir_constant_data *data)
632 : ir_rvalue(ir_type_constant)
633 {
634 this->const_elements = NULL;
635
636 assert((type->base_type >= GLSL_TYPE_UINT)
637 && (type->base_type <= GLSL_TYPE_IMAGE));
638
639 this->type = type;
640 memcpy(& this->value, data, sizeof(this->value));
641 }
642
643 ir_constant::ir_constant(float f, unsigned vector_elements)
644 : ir_rvalue(ir_type_constant)
645 {
646 assert(vector_elements <= 4);
647 this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT, vector_elements, 1);
648 for (unsigned i = 0; i < vector_elements; i++) {
649 this->value.f[i] = f;
650 }
651 for (unsigned i = vector_elements; i < 16; i++) {
652 this->value.f[i] = 0;
653 }
654 }
655
656 ir_constant::ir_constant(double d, unsigned vector_elements)
657 : ir_rvalue(ir_type_constant)
658 {
659 assert(vector_elements <= 4);
660 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE, vector_elements, 1);
661 for (unsigned i = 0; i < vector_elements; i++) {
662 this->value.d[i] = d;
663 }
664 for (unsigned i = vector_elements; i < 16; i++) {
665 this->value.d[i] = 0.0;
666 }
667 }
668
669 ir_constant::ir_constant(unsigned int u, unsigned vector_elements)
670 : ir_rvalue(ir_type_constant)
671 {
672 assert(vector_elements <= 4);
673 this->type = glsl_type::get_instance(GLSL_TYPE_UINT, vector_elements, 1);
674 for (unsigned i = 0; i < vector_elements; i++) {
675 this->value.u[i] = u;
676 }
677 for (unsigned i = vector_elements; i < 16; i++) {
678 this->value.u[i] = 0;
679 }
680 }
681
682 ir_constant::ir_constant(int integer, unsigned vector_elements)
683 : ir_rvalue(ir_type_constant)
684 {
685 assert(vector_elements <= 4);
686 this->type = glsl_type::get_instance(GLSL_TYPE_INT, vector_elements, 1);
687 for (unsigned i = 0; i < vector_elements; i++) {
688 this->value.i[i] = integer;
689 }
690 for (unsigned i = vector_elements; i < 16; i++) {
691 this->value.i[i] = 0;
692 }
693 }
694
695 ir_constant::ir_constant(uint64_t u64, unsigned vector_elements)
696 : ir_rvalue(ir_type_constant)
697 {
698 assert(vector_elements <= 4);
699 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64, vector_elements, 1);
700 for (unsigned i = 0; i < vector_elements; i++) {
701 this->value.u64[i] = u64;
702 }
703 for (unsigned i = vector_elements; i < 16; i++) {
704 this->value.u64[i] = 0;
705 }
706 }
707
708 ir_constant::ir_constant(int64_t int64, unsigned vector_elements)
709 : ir_rvalue(ir_type_constant)
710 {
711 assert(vector_elements <= 4);
712 this->type = glsl_type::get_instance(GLSL_TYPE_INT64, vector_elements, 1);
713 for (unsigned i = 0; i < vector_elements; i++) {
714 this->value.i64[i] = int64;
715 }
716 for (unsigned i = vector_elements; i < 16; i++) {
717 this->value.i64[i] = 0;
718 }
719 }
720
721 ir_constant::ir_constant(bool b, unsigned vector_elements)
722 : ir_rvalue(ir_type_constant)
723 {
724 assert(vector_elements <= 4);
725 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL, vector_elements, 1);
726 for (unsigned i = 0; i < vector_elements; i++) {
727 this->value.b[i] = b;
728 }
729 for (unsigned i = vector_elements; i < 16; i++) {
730 this->value.b[i] = false;
731 }
732 }
733
734 ir_constant::ir_constant(const ir_constant *c, unsigned i)
735 : ir_rvalue(ir_type_constant)
736 {
737 this->const_elements = NULL;
738 this->type = c->type->get_base_type();
739
740 switch (this->type->base_type) {
741 case GLSL_TYPE_UINT: this->value.u[0] = c->value.u[i]; break;
742 case GLSL_TYPE_INT: this->value.i[0] = c->value.i[i]; break;
743 case GLSL_TYPE_FLOAT: this->value.f[0] = c->value.f[i]; break;
744 case GLSL_TYPE_BOOL: this->value.b[0] = c->value.b[i]; break;
745 case GLSL_TYPE_DOUBLE: this->value.d[0] = c->value.d[i]; break;
746 default: assert(!"Should not get here."); break;
747 }
748 }
749
750 ir_constant::ir_constant(const struct glsl_type *type, exec_list *value_list)
751 : ir_rvalue(ir_type_constant)
752 {
753 this->const_elements = NULL;
754 this->type = type;
755
756 assert(type->is_scalar() || type->is_vector() || type->is_matrix()
757 || type->is_record() || type->is_array());
758
759 /* If the constant is a record, the types of each of the entries in
760 * value_list must be a 1-for-1 match with the structure components. Each
761 * entry must also be a constant. Just move the nodes from the value_list
762 * to the list in the ir_constant.
763 */
764 if (type->is_array() || type->is_record()) {
765 this->const_elements = ralloc_array(this, ir_constant *, type->length);
766 unsigned i = 0;
767 foreach_in_list(ir_constant, value, value_list) {
768 assert(value->as_constant() != NULL);
769
770 this->const_elements[i++] = value;
771 }
772 return;
773 }
774
775 for (unsigned i = 0; i < 16; i++) {
776 this->value.u[i] = 0;
777 }
778
779 ir_constant *value = (ir_constant *) (value_list->get_head_raw());
780
781 /* Constructors with exactly one scalar argument are special for vectors
782 * and matrices. For vectors, the scalar value is replicated to fill all
783 * the components. For matrices, the scalar fills the components of the
784 * diagonal while the rest is filled with 0.
785 */
786 if (value->type->is_scalar() && value->next->is_tail_sentinel()) {
787 if (type->is_matrix()) {
788 /* Matrix - fill diagonal (rest is already set to 0) */
789 assert(type->is_float() || type->is_double());
790 for (unsigned i = 0; i < type->matrix_columns; i++) {
791 if (type->is_float())
792 this->value.f[i * type->vector_elements + i] =
793 value->value.f[0];
794 else
795 this->value.d[i * type->vector_elements + i] =
796 value->value.d[0];
797 }
798 } else {
799 /* Vector or scalar - fill all components */
800 switch (type->base_type) {
801 case GLSL_TYPE_UINT:
802 case GLSL_TYPE_INT:
803 for (unsigned i = 0; i < type->components(); i++)
804 this->value.u[i] = value->value.u[0];
805 break;
806 case GLSL_TYPE_FLOAT:
807 for (unsigned i = 0; i < type->components(); i++)
808 this->value.f[i] = value->value.f[0];
809 break;
810 case GLSL_TYPE_DOUBLE:
811 for (unsigned i = 0; i < type->components(); i++)
812 this->value.d[i] = value->value.d[0];
813 break;
814 case GLSL_TYPE_UINT64:
815 case GLSL_TYPE_INT64:
816 for (unsigned i = 0; i < type->components(); i++)
817 this->value.u64[i] = value->value.u64[0];
818 break;
819 case GLSL_TYPE_BOOL:
820 for (unsigned i = 0; i < type->components(); i++)
821 this->value.b[i] = value->value.b[0];
822 break;
823 case GLSL_TYPE_SAMPLER:
824 case GLSL_TYPE_IMAGE:
825 this->value.u64[0] = value->value.u64[0];
826 break;
827 default:
828 assert(!"Should not get here.");
829 break;
830 }
831 }
832 return;
833 }
834
835 if (type->is_matrix() && value->type->is_matrix()) {
836 assert(value->next->is_tail_sentinel());
837
838 /* From section 5.4.2 of the GLSL 1.20 spec:
839 * "If a matrix is constructed from a matrix, then each component
840 * (column i, row j) in the result that has a corresponding component
841 * (column i, row j) in the argument will be initialized from there."
842 */
843 unsigned cols = MIN2(type->matrix_columns, value->type->matrix_columns);
844 unsigned rows = MIN2(type->vector_elements, value->type->vector_elements);
845 for (unsigned i = 0; i < cols; i++) {
846 for (unsigned j = 0; j < rows; j++) {
847 const unsigned src = i * value->type->vector_elements + j;
848 const unsigned dst = i * type->vector_elements + j;
849 this->value.f[dst] = value->value.f[src];
850 }
851 }
852
853 /* "All other components will be initialized to the identity matrix." */
854 for (unsigned i = cols; i < type->matrix_columns; i++)
855 this->value.f[i * type->vector_elements + i] = 1.0;
856
857 return;
858 }
859
860 /* Use each component from each entry in the value_list to initialize one
861 * component of the constant being constructed.
862 */
863 unsigned i = 0;
864 for (;;) {
865 assert(value->as_constant() != NULL);
866 assert(!value->is_tail_sentinel());
867
868 for (unsigned j = 0; j < value->type->components(); j++) {
869 switch (type->base_type) {
870 case GLSL_TYPE_UINT:
871 this->value.u[i] = value->get_uint_component(j);
872 break;
873 case GLSL_TYPE_INT:
874 this->value.i[i] = value->get_int_component(j);
875 break;
876 case GLSL_TYPE_FLOAT:
877 this->value.f[i] = value->get_float_component(j);
878 break;
879 case GLSL_TYPE_BOOL:
880 this->value.b[i] = value->get_bool_component(j);
881 break;
882 case GLSL_TYPE_DOUBLE:
883 this->value.d[i] = value->get_double_component(j);
884 break;
885 case GLSL_TYPE_UINT64:
886 this->value.u64[i] = value->get_uint64_component(j);
887 break;
888 case GLSL_TYPE_INT64:
889 this->value.i64[i] = value->get_int64_component(j);
890 break;
891 default:
892 /* FINISHME: What to do? Exceptions are not the answer.
893 */
894 break;
895 }
896
897 i++;
898 if (i >= type->components())
899 break;
900 }
901
902 if (i >= type->components())
903 break; /* avoid downcasting a list sentinel */
904 value = (ir_constant *) value->next;
905 }
906 }
907
908 ir_constant *
909 ir_constant::zero(void *mem_ctx, const glsl_type *type)
910 {
911 assert(type->is_scalar() || type->is_vector() || type->is_matrix()
912 || type->is_record() || type->is_array());
913
914 ir_constant *c = new(mem_ctx) ir_constant;
915 c->type = type;
916 memset(&c->value, 0, sizeof(c->value));
917
918 if (type->is_array()) {
919 c->const_elements = ralloc_array(c, ir_constant *, type->length);
920
921 for (unsigned i = 0; i < type->length; i++)
922 c->const_elements[i] = ir_constant::zero(c, type->fields.array);
923 }
924
925 if (type->is_record()) {
926 c->const_elements = ralloc_array(c, ir_constant *, type->length);
927
928 for (unsigned i = 0; i < type->length; i++) {
929 c->const_elements[i] =
930 ir_constant::zero(mem_ctx, type->fields.structure[i].type);
931 }
932 }
933
934 return c;
935 }
936
937 bool
938 ir_constant::get_bool_component(unsigned i) const
939 {
940 switch (this->type->base_type) {
941 case GLSL_TYPE_UINT: return this->value.u[i] != 0;
942 case GLSL_TYPE_INT: return this->value.i[i] != 0;
943 case GLSL_TYPE_FLOAT: return ((int)this->value.f[i]) != 0;
944 case GLSL_TYPE_BOOL: return this->value.b[i];
945 case GLSL_TYPE_DOUBLE: return this->value.d[i] != 0.0;
946 case GLSL_TYPE_SAMPLER:
947 case GLSL_TYPE_IMAGE:
948 case GLSL_TYPE_UINT64: return this->value.u64[i] != 0;
949 case GLSL_TYPE_INT64: return this->value.i64[i] != 0;
950 default: assert(!"Should not get here."); break;
951 }
952
953 /* Must return something to make the compiler happy. This is clearly an
954 * error case.
955 */
956 return false;
957 }
958
959 float
960 ir_constant::get_float_component(unsigned i) const
961 {
962 switch (this->type->base_type) {
963 case GLSL_TYPE_UINT: return (float) this->value.u[i];
964 case GLSL_TYPE_INT: return (float) this->value.i[i];
965 case GLSL_TYPE_FLOAT: return this->value.f[i];
966 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1.0f : 0.0f;
967 case GLSL_TYPE_DOUBLE: return (float) this->value.d[i];
968 case GLSL_TYPE_SAMPLER:
969 case GLSL_TYPE_IMAGE:
970 case GLSL_TYPE_UINT64: return (float) this->value.u64[i];
971 case GLSL_TYPE_INT64: return (float) this->value.i64[i];
972 default: assert(!"Should not get here."); break;
973 }
974
975 /* Must return something to make the compiler happy. This is clearly an
976 * error case.
977 */
978 return 0.0;
979 }
980
981 double
982 ir_constant::get_double_component(unsigned i) const
983 {
984 switch (this->type->base_type) {
985 case GLSL_TYPE_UINT: return (double) this->value.u[i];
986 case GLSL_TYPE_INT: return (double) this->value.i[i];
987 case GLSL_TYPE_FLOAT: return (double) this->value.f[i];
988 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1.0 : 0.0;
989 case GLSL_TYPE_DOUBLE: return this->value.d[i];
990 case GLSL_TYPE_SAMPLER:
991 case GLSL_TYPE_IMAGE:
992 case GLSL_TYPE_UINT64: return (double) this->value.u64[i];
993 case GLSL_TYPE_INT64: return (double) this->value.i64[i];
994 default: assert(!"Should not get here."); break;
995 }
996
997 /* Must return something to make the compiler happy. This is clearly an
998 * error case.
999 */
1000 return 0.0;
1001 }
1002
1003 int
1004 ir_constant::get_int_component(unsigned i) const
1005 {
1006 switch (this->type->base_type) {
1007 case GLSL_TYPE_UINT: return this->value.u[i];
1008 case GLSL_TYPE_INT: return this->value.i[i];
1009 case GLSL_TYPE_FLOAT: return (int) this->value.f[i];
1010 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1011 case GLSL_TYPE_DOUBLE: return (int) this->value.d[i];
1012 case GLSL_TYPE_SAMPLER:
1013 case GLSL_TYPE_IMAGE:
1014 case GLSL_TYPE_UINT64: return (int) this->value.u64[i];
1015 case GLSL_TYPE_INT64: return (int) this->value.i64[i];
1016 default: assert(!"Should not get here."); break;
1017 }
1018
1019 /* Must return something to make the compiler happy. This is clearly an
1020 * error case.
1021 */
1022 return 0;
1023 }
1024
1025 unsigned
1026 ir_constant::get_uint_component(unsigned i) const
1027 {
1028 switch (this->type->base_type) {
1029 case GLSL_TYPE_UINT: return this->value.u[i];
1030 case GLSL_TYPE_INT: return this->value.i[i];
1031 case GLSL_TYPE_FLOAT: return (unsigned) this->value.f[i];
1032 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1033 case GLSL_TYPE_DOUBLE: return (unsigned) this->value.d[i];
1034 case GLSL_TYPE_SAMPLER:
1035 case GLSL_TYPE_IMAGE:
1036 case GLSL_TYPE_UINT64: return (unsigned) this->value.u64[i];
1037 case GLSL_TYPE_INT64: return (unsigned) this->value.i64[i];
1038 default: assert(!"Should not get here."); break;
1039 }
1040
1041 /* Must return something to make the compiler happy. This is clearly an
1042 * error case.
1043 */
1044 return 0;
1045 }
1046
1047 int64_t
1048 ir_constant::get_int64_component(unsigned i) const
1049 {
1050 switch (this->type->base_type) {
1051 case GLSL_TYPE_UINT: return this->value.u[i];
1052 case GLSL_TYPE_INT: return this->value.i[i];
1053 case GLSL_TYPE_FLOAT: return (int64_t) this->value.f[i];
1054 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1055 case GLSL_TYPE_DOUBLE: return (int64_t) this->value.d[i];
1056 case GLSL_TYPE_SAMPLER:
1057 case GLSL_TYPE_IMAGE:
1058 case GLSL_TYPE_UINT64: return (int64_t) this->value.u64[i];
1059 case GLSL_TYPE_INT64: return this->value.i64[i];
1060 default: assert(!"Should not get here."); break;
1061 }
1062
1063 /* Must return something to make the compiler happy. This is clearly an
1064 * error case.
1065 */
1066 return 0;
1067 }
1068
1069 uint64_t
1070 ir_constant::get_uint64_component(unsigned i) const
1071 {
1072 switch (this->type->base_type) {
1073 case GLSL_TYPE_UINT: return this->value.u[i];
1074 case GLSL_TYPE_INT: return this->value.i[i];
1075 case GLSL_TYPE_FLOAT: return (uint64_t) this->value.f[i];
1076 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1077 case GLSL_TYPE_DOUBLE: return (uint64_t) this->value.d[i];
1078 case GLSL_TYPE_SAMPLER:
1079 case GLSL_TYPE_IMAGE:
1080 case GLSL_TYPE_UINT64: return this->value.u64[i];
1081 case GLSL_TYPE_INT64: return (uint64_t) this->value.i64[i];
1082 default: assert(!"Should not get here."); break;
1083 }
1084
1085 /* Must return something to make the compiler happy. This is clearly an
1086 * error case.
1087 */
1088 return 0;
1089 }
1090
1091 ir_constant *
1092 ir_constant::get_array_element(unsigned i) const
1093 {
1094 assert(this->type->is_array());
1095
1096 /* From page 35 (page 41 of the PDF) of the GLSL 1.20 spec:
1097 *
1098 * "Behavior is undefined if a shader subscripts an array with an index
1099 * less than 0 or greater than or equal to the size the array was
1100 * declared with."
1101 *
1102 * Most out-of-bounds accesses are removed before things could get this far.
1103 * There are cases where non-constant array index values can get constant
1104 * folded.
1105 */
1106 if (int(i) < 0)
1107 i = 0;
1108 else if (i >= this->type->length)
1109 i = this->type->length - 1;
1110
1111 return const_elements[i];
1112 }
1113
1114 ir_constant *
1115 ir_constant::get_record_field(int idx)
1116 {
1117 assert(this->type->is_record());
1118 assert(idx >= 0 && (unsigned) idx < this->type->length);
1119
1120 return const_elements[idx];
1121 }
1122
1123 void
1124 ir_constant::copy_offset(ir_constant *src, int offset)
1125 {
1126 switch (this->type->base_type) {
1127 case GLSL_TYPE_UINT:
1128 case GLSL_TYPE_INT:
1129 case GLSL_TYPE_FLOAT:
1130 case GLSL_TYPE_DOUBLE:
1131 case GLSL_TYPE_SAMPLER:
1132 case GLSL_TYPE_IMAGE:
1133 case GLSL_TYPE_UINT64:
1134 case GLSL_TYPE_INT64:
1135 case GLSL_TYPE_BOOL: {
1136 unsigned int size = src->type->components();
1137 assert (size <= this->type->components() - offset);
1138 for (unsigned int i=0; i<size; i++) {
1139 switch (this->type->base_type) {
1140 case GLSL_TYPE_UINT:
1141 value.u[i+offset] = src->get_uint_component(i);
1142 break;
1143 case GLSL_TYPE_INT:
1144 value.i[i+offset] = src->get_int_component(i);
1145 break;
1146 case GLSL_TYPE_FLOAT:
1147 value.f[i+offset] = src->get_float_component(i);
1148 break;
1149 case GLSL_TYPE_BOOL:
1150 value.b[i+offset] = src->get_bool_component(i);
1151 break;
1152 case GLSL_TYPE_DOUBLE:
1153 value.d[i+offset] = src->get_double_component(i);
1154 break;
1155 case GLSL_TYPE_SAMPLER:
1156 case GLSL_TYPE_IMAGE:
1157 case GLSL_TYPE_UINT64:
1158 value.u64[i+offset] = src->get_uint64_component(i);
1159 break;
1160 case GLSL_TYPE_INT64:
1161 value.i64[i+offset] = src->get_int64_component(i);
1162 break;
1163 default: // Shut up the compiler
1164 break;
1165 }
1166 }
1167 break;
1168 }
1169
1170 case GLSL_TYPE_STRUCT:
1171 case GLSL_TYPE_ARRAY: {
1172 assert (src->type == this->type);
1173 for (unsigned i = 0; i < this->type->length; i++) {
1174 this->const_elements[i] = src->const_elements[i]->clone(this, NULL);
1175 }
1176 break;
1177 }
1178
1179 default:
1180 assert(!"Should not get here.");
1181 break;
1182 }
1183 }
1184
1185 void
1186 ir_constant::copy_masked_offset(ir_constant *src, int offset, unsigned int mask)
1187 {
1188 assert (!type->is_array() && !type->is_record());
1189
1190 if (!type->is_vector() && !type->is_matrix()) {
1191 offset = 0;
1192 mask = 1;
1193 }
1194
1195 int id = 0;
1196 for (int i=0; i<4; i++) {
1197 if (mask & (1 << i)) {
1198 switch (this->type->base_type) {
1199 case GLSL_TYPE_UINT:
1200 value.u[i+offset] = src->get_uint_component(id++);
1201 break;
1202 case GLSL_TYPE_INT:
1203 value.i[i+offset] = src->get_int_component(id++);
1204 break;
1205 case GLSL_TYPE_FLOAT:
1206 value.f[i+offset] = src->get_float_component(id++);
1207 break;
1208 case GLSL_TYPE_BOOL:
1209 value.b[i+offset] = src->get_bool_component(id++);
1210 break;
1211 case GLSL_TYPE_DOUBLE:
1212 value.d[i+offset] = src->get_double_component(id++);
1213 break;
1214 case GLSL_TYPE_SAMPLER:
1215 case GLSL_TYPE_IMAGE:
1216 case GLSL_TYPE_UINT64:
1217 value.u64[i+offset] = src->get_uint64_component(id++);
1218 break;
1219 case GLSL_TYPE_INT64:
1220 value.i64[i+offset] = src->get_int64_component(id++);
1221 break;
1222 default:
1223 assert(!"Should not get here.");
1224 return;
1225 }
1226 }
1227 }
1228 }
1229
1230 bool
1231 ir_constant::has_value(const ir_constant *c) const
1232 {
1233 if (this->type != c->type)
1234 return false;
1235
1236 if (this->type->is_array() || this->type->is_record()) {
1237 for (unsigned i = 0; i < this->type->length; i++) {
1238 if (!this->const_elements[i]->has_value(c->const_elements[i]))
1239 return false;
1240 }
1241 return true;
1242 }
1243
1244 for (unsigned i = 0; i < this->type->components(); i++) {
1245 switch (this->type->base_type) {
1246 case GLSL_TYPE_UINT:
1247 if (this->value.u[i] != c->value.u[i])
1248 return false;
1249 break;
1250 case GLSL_TYPE_INT:
1251 if (this->value.i[i] != c->value.i[i])
1252 return false;
1253 break;
1254 case GLSL_TYPE_FLOAT:
1255 if (this->value.f[i] != c->value.f[i])
1256 return false;
1257 break;
1258 case GLSL_TYPE_BOOL:
1259 if (this->value.b[i] != c->value.b[i])
1260 return false;
1261 break;
1262 case GLSL_TYPE_DOUBLE:
1263 if (this->value.d[i] != c->value.d[i])
1264 return false;
1265 break;
1266 case GLSL_TYPE_SAMPLER:
1267 case GLSL_TYPE_IMAGE:
1268 case GLSL_TYPE_UINT64:
1269 if (this->value.u64[i] != c->value.u64[i])
1270 return false;
1271 break;
1272 case GLSL_TYPE_INT64:
1273 if (this->value.i64[i] != c->value.i64[i])
1274 return false;
1275 break;
1276 default:
1277 assert(!"Should not get here.");
1278 return false;
1279 }
1280 }
1281
1282 return true;
1283 }
1284
1285 bool
1286 ir_constant::is_value(float f, int i) const
1287 {
1288 if (!this->type->is_scalar() && !this->type->is_vector())
1289 return false;
1290
1291 /* Only accept boolean values for 0/1. */
1292 if (int(bool(i)) != i && this->type->is_boolean())
1293 return false;
1294
1295 for (unsigned c = 0; c < this->type->vector_elements; c++) {
1296 switch (this->type->base_type) {
1297 case GLSL_TYPE_FLOAT:
1298 if (this->value.f[c] != f)
1299 return false;
1300 break;
1301 case GLSL_TYPE_INT:
1302 if (this->value.i[c] != i)
1303 return false;
1304 break;
1305 case GLSL_TYPE_UINT:
1306 if (this->value.u[c] != unsigned(i))
1307 return false;
1308 break;
1309 case GLSL_TYPE_BOOL:
1310 if (this->value.b[c] != bool(i))
1311 return false;
1312 break;
1313 case GLSL_TYPE_DOUBLE:
1314 if (this->value.d[c] != double(f))
1315 return false;
1316 break;
1317 case GLSL_TYPE_SAMPLER:
1318 case GLSL_TYPE_IMAGE:
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->type = this->record->type->field_type(field);
1430 this->field_idx = this->record->type->field_index(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->type = this->record->type->field_type(field);
1442 this->field_idx = this->record->type->field_index(field);
1443 }
1444
1445 bool
1446 ir_dereference::is_lvalue(const struct _mesa_glsl_parse_state *state) 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 ARB_bindless_texture spec:
1456 *
1457 * "Samplers can be used as l-values, so can be assigned into and used as
1458 * "out" and "inout" function parameters."
1459 *
1460 * From section 4.1.X of the ARB_bindless_texture spec:
1461 *
1462 * "Images can be used as l-values, so can be assigned into and used as
1463 * "out" and "inout" function parameters."
1464 */
1465 if ((!state || state->has_bindless()) &&
1466 (this->type->contains_sampler() || this->type->contains_image()))
1467 return true;
1468
1469 /* From section 4.1.7 of the GLSL 4.40 spec:
1470 *
1471 * "Opaque variables cannot be treated as l-values; hence cannot
1472 * be used as out or inout function parameters, nor can they be
1473 * assigned into."
1474 */
1475 if (this->type->contains_opaque())
1476 return false;
1477
1478 return true;
1479 }
1480
1481
1482 static const char * const tex_opcode_strs[] = { "tex", "txb", "txl", "txd", "txf", "txf_ms", "txs", "lod", "tg4", "query_levels", "texture_samples", "samples_identical" };
1483
1484 const char *ir_texture::opcode_string()
1485 {
1486 assert((unsigned int) op < ARRAY_SIZE(tex_opcode_strs));
1487 return tex_opcode_strs[op];
1488 }
1489
1490 ir_texture_opcode
1491 ir_texture::get_opcode(const char *str)
1492 {
1493 const int count = sizeof(tex_opcode_strs) / sizeof(tex_opcode_strs[0]);
1494 for (int op = 0; op < count; op++) {
1495 if (strcmp(str, tex_opcode_strs[op]) == 0)
1496 return (ir_texture_opcode) op;
1497 }
1498 return (ir_texture_opcode) -1;
1499 }
1500
1501
1502 void
1503 ir_texture::set_sampler(ir_dereference *sampler, const glsl_type *type)
1504 {
1505 assert(sampler != NULL);
1506 assert(type != NULL);
1507 this->sampler = sampler;
1508 this->type = type;
1509
1510 if (this->op == ir_txs || this->op == ir_query_levels ||
1511 this->op == ir_texture_samples) {
1512 assert(type->base_type == GLSL_TYPE_INT);
1513 } else if (this->op == ir_lod) {
1514 assert(type->vector_elements == 2);
1515 assert(type->is_float());
1516 } else if (this->op == ir_samples_identical) {
1517 assert(type == glsl_type::bool_type);
1518 assert(sampler->type->is_sampler());
1519 assert(sampler->type->sampler_dimensionality == GLSL_SAMPLER_DIM_MS);
1520 } else {
1521 assert(sampler->type->sampled_type == (int) type->base_type);
1522 if (sampler->type->sampler_shadow)
1523 assert(type->vector_elements == 4 || type->vector_elements == 1);
1524 else
1525 assert(type->vector_elements == 4);
1526 }
1527 }
1528
1529
1530 void
1531 ir_swizzle::init_mask(const unsigned *comp, unsigned count)
1532 {
1533 assert((count >= 1) && (count <= 4));
1534
1535 memset(&this->mask, 0, sizeof(this->mask));
1536 this->mask.num_components = count;
1537
1538 unsigned dup_mask = 0;
1539 switch (count) {
1540 case 4:
1541 assert(comp[3] <= 3);
1542 dup_mask |= (1U << comp[3])
1543 & ((1U << comp[0]) | (1U << comp[1]) | (1U << comp[2]));
1544 this->mask.w = comp[3];
1545
1546 case 3:
1547 assert(comp[2] <= 3);
1548 dup_mask |= (1U << comp[2])
1549 & ((1U << comp[0]) | (1U << comp[1]));
1550 this->mask.z = comp[2];
1551
1552 case 2:
1553 assert(comp[1] <= 3);
1554 dup_mask |= (1U << comp[1])
1555 & ((1U << comp[0]));
1556 this->mask.y = comp[1];
1557
1558 case 1:
1559 assert(comp[0] <= 3);
1560 this->mask.x = comp[0];
1561 }
1562
1563 this->mask.has_duplicates = dup_mask != 0;
1564
1565 /* Based on the number of elements in the swizzle and the base type
1566 * (i.e., float, int, unsigned, or bool) of the vector being swizzled,
1567 * generate the type of the resulting value.
1568 */
1569 type = glsl_type::get_instance(val->type->base_type, mask.num_components, 1);
1570 }
1571
1572 ir_swizzle::ir_swizzle(ir_rvalue *val, unsigned x, unsigned y, unsigned z,
1573 unsigned w, unsigned count)
1574 : ir_rvalue(ir_type_swizzle), val(val)
1575 {
1576 const unsigned components[4] = { x, y, z, w };
1577 this->init_mask(components, count);
1578 }
1579
1580 ir_swizzle::ir_swizzle(ir_rvalue *val, const unsigned *comp,
1581 unsigned count)
1582 : ir_rvalue(ir_type_swizzle), val(val)
1583 {
1584 this->init_mask(comp, count);
1585 }
1586
1587 ir_swizzle::ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask)
1588 : ir_rvalue(ir_type_swizzle), val(val), mask(mask)
1589 {
1590 this->type = glsl_type::get_instance(val->type->base_type,
1591 mask.num_components, 1);
1592 }
1593
1594 #define X 1
1595 #define R 5
1596 #define S 9
1597 #define I 13
1598
1599 ir_swizzle *
1600 ir_swizzle::create(ir_rvalue *val, const char *str, unsigned vector_length)
1601 {
1602 void *ctx = ralloc_parent(val);
1603
1604 /* For each possible swizzle character, this table encodes the value in
1605 * \c idx_map that represents the 0th element of the vector. For invalid
1606 * swizzle characters (e.g., 'k'), a special value is used that will allow
1607 * detection of errors.
1608 */
1609 static const unsigned char base_idx[26] = {
1610 /* a b c d e f g h i j k l m */
1611 R, R, I, I, I, I, R, I, I, I, I, I, I,
1612 /* n o p q r s t u v w x y z */
1613 I, I, S, S, R, S, S, I, I, X, X, X, X
1614 };
1615
1616 /* Each valid swizzle character has an entry in the previous table. This
1617 * table encodes the base index encoded in the previous table plus the actual
1618 * index of the swizzle character. When processing swizzles, the first
1619 * character in the string is indexed in the previous table. Each character
1620 * in the string is indexed in this table, and the value found there has the
1621 * value form the first table subtracted. The result must be on the range
1622 * [0,3].
1623 *
1624 * For example, the string "wzyx" will get X from the first table. Each of
1625 * the charcaters will get X+3, X+2, X+1, and X+0 from this table. After
1626 * subtraction, the swizzle values are { 3, 2, 1, 0 }.
1627 *
1628 * The string "wzrg" will get X from the first table. Each of the characters
1629 * will get X+3, X+2, R+0, and R+1 from this table. After subtraction, the
1630 * swizzle values are { 3, 2, 4, 5 }. Since 4 and 5 are outside the range
1631 * [0,3], the error is detected.
1632 */
1633 static const unsigned char idx_map[26] = {
1634 /* a b c d e f g h i j k l m */
1635 R+3, R+2, 0, 0, 0, 0, R+1, 0, 0, 0, 0, 0, 0,
1636 /* n o p q r s t u v w x y z */
1637 0, 0, S+2, S+3, R+0, S+0, S+1, 0, 0, X+3, X+0, X+1, X+2
1638 };
1639
1640 int swiz_idx[4] = { 0, 0, 0, 0 };
1641 unsigned i;
1642
1643
1644 /* Validate the first character in the swizzle string and look up the base
1645 * index value as described above.
1646 */
1647 if ((str[0] < 'a') || (str[0] > 'z'))
1648 return NULL;
1649
1650 const unsigned base = base_idx[str[0] - 'a'];
1651
1652
1653 for (i = 0; (i < 4) && (str[i] != '\0'); i++) {
1654 /* Validate the next character, and, as described above, convert it to a
1655 * swizzle index.
1656 */
1657 if ((str[i] < 'a') || (str[i] > 'z'))
1658 return NULL;
1659
1660 swiz_idx[i] = idx_map[str[i] - 'a'] - base;
1661 if ((swiz_idx[i] < 0) || (swiz_idx[i] >= (int) vector_length))
1662 return NULL;
1663 }
1664
1665 if (str[i] != '\0')
1666 return NULL;
1667
1668 return new(ctx) ir_swizzle(val, swiz_idx[0], swiz_idx[1], swiz_idx[2],
1669 swiz_idx[3], i);
1670 }
1671
1672 #undef X
1673 #undef R
1674 #undef S
1675 #undef I
1676
1677 ir_variable *
1678 ir_swizzle::variable_referenced() const
1679 {
1680 return this->val->variable_referenced();
1681 }
1682
1683
1684 bool ir_variable::temporaries_allocate_names = false;
1685
1686 const char ir_variable::tmp_name[] = "compiler_temp";
1687
1688 ir_variable::ir_variable(const struct glsl_type *type, const char *name,
1689 ir_variable_mode mode)
1690 : ir_instruction(ir_type_variable)
1691 {
1692 this->type = type;
1693
1694 if (mode == ir_var_temporary && !ir_variable::temporaries_allocate_names)
1695 name = NULL;
1696
1697 /* The ir_variable clone method may call this constructor with name set to
1698 * tmp_name.
1699 */
1700 assert(name != NULL
1701 || mode == ir_var_temporary
1702 || mode == ir_var_function_in
1703 || mode == ir_var_function_out
1704 || mode == ir_var_function_inout);
1705 assert(name != ir_variable::tmp_name
1706 || mode == ir_var_temporary);
1707 if (mode == ir_var_temporary
1708 && (name == NULL || name == ir_variable::tmp_name)) {
1709 this->name = ir_variable::tmp_name;
1710 } else if (name == NULL ||
1711 strlen(name) < ARRAY_SIZE(this->name_storage)) {
1712 strcpy(this->name_storage, name ? name : "");
1713 this->name = this->name_storage;
1714 } else {
1715 this->name = ralloc_strdup(this, name);
1716 }
1717
1718 this->u.max_ifc_array_access = NULL;
1719
1720 this->data.explicit_location = false;
1721 this->data.has_initializer = false;
1722 this->data.location = -1;
1723 this->data.location_frac = 0;
1724 this->data.binding = 0;
1725 this->data.warn_extension_index = 0;
1726 this->constant_value = NULL;
1727 this->constant_initializer = NULL;
1728 this->data.origin_upper_left = false;
1729 this->data.pixel_center_integer = false;
1730 this->data.depth_layout = ir_depth_layout_none;
1731 this->data.used = false;
1732 this->data.always_active_io = false;
1733 this->data.read_only = false;
1734 this->data.centroid = false;
1735 this->data.sample = false;
1736 this->data.patch = false;
1737 this->data.invariant = false;
1738 this->data.how_declared = ir_var_declared_normally;
1739 this->data.mode = mode;
1740 this->data.interpolation = INTERP_MODE_NONE;
1741 this->data.max_array_access = -1;
1742 this->data.offset = 0;
1743 this->data.precision = GLSL_PRECISION_NONE;
1744 this->data.memory_read_only = false;
1745 this->data.memory_write_only = false;
1746 this->data.memory_coherent = false;
1747 this->data.memory_volatile = false;
1748 this->data.memory_restrict = false;
1749 this->data.from_ssbo_unsized_array = false;
1750 this->data.fb_fetch_output = false;
1751 this->data.bindless = false;
1752 this->data.bound = false;
1753
1754 if (type != NULL) {
1755 if (type->is_interface())
1756 this->init_interface_type(type);
1757 else if (type->without_array()->is_interface())
1758 this->init_interface_type(type->without_array());
1759 }
1760 }
1761
1762
1763 const char *
1764 interpolation_string(unsigned interpolation)
1765 {
1766 switch (interpolation) {
1767 case INTERP_MODE_NONE: return "no";
1768 case INTERP_MODE_SMOOTH: return "smooth";
1769 case INTERP_MODE_FLAT: return "flat";
1770 case INTERP_MODE_NOPERSPECTIVE: return "noperspective";
1771 }
1772
1773 assert(!"Should not get here.");
1774 return "";
1775 }
1776
1777 const char *const ir_variable::warn_extension_table[] = {
1778 "",
1779 "GL_ARB_shader_stencil_export",
1780 "GL_AMD_shader_stencil_export",
1781 };
1782
1783 void
1784 ir_variable::enable_extension_warning(const char *extension)
1785 {
1786 for (unsigned i = 0; i < ARRAY_SIZE(warn_extension_table); i++) {
1787 if (strcmp(warn_extension_table[i], extension) == 0) {
1788 this->data.warn_extension_index = i;
1789 return;
1790 }
1791 }
1792
1793 assert(!"Should not get here.");
1794 this->data.warn_extension_index = 0;
1795 }
1796
1797 const char *
1798 ir_variable::get_extension_warning() const
1799 {
1800 return this->data.warn_extension_index == 0
1801 ? NULL : warn_extension_table[this->data.warn_extension_index];
1802 }
1803
1804 ir_function_signature::ir_function_signature(const glsl_type *return_type,
1805 builtin_available_predicate b)
1806 : ir_instruction(ir_type_function_signature),
1807 return_type(return_type), is_defined(false),
1808 intrinsic_id(ir_intrinsic_invalid), builtin_avail(b), _function(NULL)
1809 {
1810 this->origin = NULL;
1811 }
1812
1813
1814 bool
1815 ir_function_signature::is_builtin() const
1816 {
1817 return builtin_avail != NULL;
1818 }
1819
1820
1821 bool
1822 ir_function_signature::is_builtin_available(const _mesa_glsl_parse_state *state) const
1823 {
1824 /* We can't call the predicate without a state pointer, so just say that
1825 * the signature is available. At compile time, we need the filtering,
1826 * but also receive a valid state pointer. At link time, we're resolving
1827 * imported built-in prototypes to their definitions, which will always
1828 * be an exact match. So we can skip the filtering.
1829 */
1830 if (state == NULL)
1831 return true;
1832
1833 assert(builtin_avail != NULL);
1834 return builtin_avail(state);
1835 }
1836
1837
1838 static bool
1839 modes_match(unsigned a, unsigned b)
1840 {
1841 if (a == b)
1842 return true;
1843
1844 /* Accept "in" vs. "const in" */
1845 if ((a == ir_var_const_in && b == ir_var_function_in) ||
1846 (b == ir_var_const_in && a == ir_var_function_in))
1847 return true;
1848
1849 return false;
1850 }
1851
1852
1853 const char *
1854 ir_function_signature::qualifiers_match(exec_list *params)
1855 {
1856 /* check that the qualifiers match. */
1857 foreach_two_lists(a_node, &this->parameters, b_node, params) {
1858 ir_variable *a = (ir_variable *) a_node;
1859 ir_variable *b = (ir_variable *) b_node;
1860
1861 if (a->data.read_only != b->data.read_only ||
1862 !modes_match(a->data.mode, b->data.mode) ||
1863 a->data.interpolation != b->data.interpolation ||
1864 a->data.centroid != b->data.centroid ||
1865 a->data.sample != b->data.sample ||
1866 a->data.patch != b->data.patch ||
1867 a->data.memory_read_only != b->data.memory_read_only ||
1868 a->data.memory_write_only != b->data.memory_write_only ||
1869 a->data.memory_coherent != b->data.memory_coherent ||
1870 a->data.memory_volatile != b->data.memory_volatile ||
1871 a->data.memory_restrict != b->data.memory_restrict) {
1872
1873 /* parameter a's qualifiers don't match */
1874 return a->name;
1875 }
1876 }
1877 return NULL;
1878 }
1879
1880
1881 void
1882 ir_function_signature::replace_parameters(exec_list *new_params)
1883 {
1884 /* Destroy all of the previous parameter information. If the previous
1885 * parameter information comes from the function prototype, it may either
1886 * specify incorrect parameter names or not have names at all.
1887 */
1888 new_params->move_nodes_to(&parameters);
1889 }
1890
1891
1892 ir_function::ir_function(const char *name)
1893 : ir_instruction(ir_type_function)
1894 {
1895 this->subroutine_index = -1;
1896 this->name = ralloc_strdup(this, name);
1897 }
1898
1899
1900 bool
1901 ir_function::has_user_signature()
1902 {
1903 foreach_in_list(ir_function_signature, sig, &this->signatures) {
1904 if (!sig->is_builtin())
1905 return true;
1906 }
1907 return false;
1908 }
1909
1910
1911 ir_rvalue *
1912 ir_rvalue::error_value(void *mem_ctx)
1913 {
1914 ir_rvalue *v = new(mem_ctx) ir_rvalue(ir_type_unset);
1915
1916 v->type = glsl_type::error_type;
1917 return v;
1918 }
1919
1920
1921 void
1922 visit_exec_list(exec_list *list, ir_visitor *visitor)
1923 {
1924 foreach_in_list_safe(ir_instruction, node, list) {
1925 node->accept(visitor);
1926 }
1927 }
1928
1929
1930 static void
1931 steal_memory(ir_instruction *ir, void *new_ctx)
1932 {
1933 ir_variable *var = ir->as_variable();
1934 ir_function *fn = ir->as_function();
1935 ir_constant *constant = ir->as_constant();
1936 if (var != NULL && var->constant_value != NULL)
1937 steal_memory(var->constant_value, ir);
1938
1939 if (var != NULL && var->constant_initializer != NULL)
1940 steal_memory(var->constant_initializer, ir);
1941
1942 if (fn != NULL && fn->subroutine_types)
1943 ralloc_steal(new_ctx, fn->subroutine_types);
1944
1945 /* The components of aggregate constants are not visited by the normal
1946 * visitor, so steal their values by hand.
1947 */
1948 if (constant != NULL &&
1949 (constant->type->is_array() || constant->type->is_record())) {
1950 for (unsigned int i = 0; i < constant->type->length; i++) {
1951 steal_memory(constant->const_elements[i], ir);
1952 }
1953 }
1954
1955 ralloc_steal(new_ctx, ir);
1956 }
1957
1958
1959 void
1960 reparent_ir(exec_list *list, void *mem_ctx)
1961 {
1962 foreach_in_list(ir_instruction, node, list) {
1963 visit_tree(node, steal_memory, mem_ctx);
1964 }
1965 }
1966
1967
1968 static ir_rvalue *
1969 try_min_one(ir_rvalue *ir)
1970 {
1971 ir_expression *expr = ir->as_expression();
1972
1973 if (!expr || expr->operation != ir_binop_min)
1974 return NULL;
1975
1976 if (expr->operands[0]->is_one())
1977 return expr->operands[1];
1978
1979 if (expr->operands[1]->is_one())
1980 return expr->operands[0];
1981
1982 return NULL;
1983 }
1984
1985 static ir_rvalue *
1986 try_max_zero(ir_rvalue *ir)
1987 {
1988 ir_expression *expr = ir->as_expression();
1989
1990 if (!expr || expr->operation != ir_binop_max)
1991 return NULL;
1992
1993 if (expr->operands[0]->is_zero())
1994 return expr->operands[1];
1995
1996 if (expr->operands[1]->is_zero())
1997 return expr->operands[0];
1998
1999 return NULL;
2000 }
2001
2002 ir_rvalue *
2003 ir_rvalue::as_rvalue_to_saturate()
2004 {
2005 ir_expression *expr = this->as_expression();
2006
2007 if (!expr)
2008 return NULL;
2009
2010 ir_rvalue *max_zero = try_max_zero(expr);
2011 if (max_zero) {
2012 return try_min_one(max_zero);
2013 } else {
2014 ir_rvalue *min_one = try_min_one(expr);
2015 if (min_one) {
2016 return try_max_zero(min_one);
2017 }
2018 }
2019
2020 return NULL;
2021 }
2022
2023
2024 unsigned
2025 vertices_per_prim(GLenum prim)
2026 {
2027 switch (prim) {
2028 case GL_POINTS:
2029 return 1;
2030 case GL_LINES:
2031 return 2;
2032 case GL_TRIANGLES:
2033 return 3;
2034 case GL_LINES_ADJACENCY:
2035 return 4;
2036 case GL_TRIANGLES_ADJACENCY:
2037 return 6;
2038 default:
2039 assert(!"Bad primitive");
2040 return 3;
2041 }
2042 }
2043
2044 /**
2045 * Generate a string describing the mode of a variable
2046 */
2047 const char *
2048 mode_string(const ir_variable *var)
2049 {
2050 switch (var->data.mode) {
2051 case ir_var_auto:
2052 return (var->data.read_only) ? "global constant" : "global variable";
2053
2054 case ir_var_uniform:
2055 return "uniform";
2056
2057 case ir_var_shader_storage:
2058 return "buffer";
2059
2060 case ir_var_shader_in:
2061 return "shader input";
2062
2063 case ir_var_shader_out:
2064 return "shader output";
2065
2066 case ir_var_function_in:
2067 case ir_var_const_in:
2068 return "function input";
2069
2070 case ir_var_function_out:
2071 return "function output";
2072
2073 case ir_var_function_inout:
2074 return "function inout";
2075
2076 case ir_var_system_value:
2077 return "shader input";
2078
2079 case ir_var_temporary:
2080 return "compiler temporary";
2081
2082 case ir_var_mode_count:
2083 break;
2084 }
2085
2086 assert(!"Should not get here.");
2087 return "invalid variable";
2088 }