projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
egl: Update headers from Khronos
[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 #include "glsl_parser_extras.h"
28
29
30 ir_rvalue::ir_rvalue(enum ir_node_type t)
31 : ir_instruction(t)
32 {
33 this->type = glsl_type::error_type;
34 }
35
36 bool ir_rvalue::is_zero() const
37 {
38 return false;
39 }
40
41 bool ir_rvalue::is_one() const
42 {
43 return false;
44 }
45
46 bool ir_rvalue::is_negative_one() const
47 {
48 return false;
49 }
50
51 /**
52 * Modify the swizzle make to move one component to another
53 *
54 * \param m IR swizzle to be modified
55 * \param from Component in the RHS that is to be swizzled
56 * \param to Desired swizzle location of \c from
57 */
58 static void
59 update_rhs_swizzle(ir_swizzle_mask &m, unsigned from, unsigned to)
60 {
61 switch (to) {
62 case 0: m.x = from; break;
63 case 1: m.y = from; break;
64 case 2: m.z = from; break;
65 case 3: m.w = from; break;
66 default: assert(!"Should not get here.");
67 }
68 }
69
70 void
71 ir_assignment::set_lhs(ir_rvalue *lhs)
72 {
73 void *mem_ctx = this;
74 bool swizzled = false;
75
76 while (lhs != NULL) {
77 ir_swizzle *swiz = lhs->as_swizzle();
78
79 if (swiz == NULL)
80 break;
81
82 unsigned write_mask = 0;
83 ir_swizzle_mask rhs_swiz = { 0, 0, 0, 0, 0, 0 };
84
85 for (unsigned i = 0; i < swiz->mask.num_components; i++) {
86 unsigned c = 0;
87
88 switch (i) {
89 case 0: c = swiz->mask.x; break;
90 case 1: c = swiz->mask.y; break;
91 case 2: c = swiz->mask.z; break;
92 case 3: c = swiz->mask.w; break;
93 default: assert(!"Should not get here.");
94 }
95
96 write_mask |= (((this->write_mask >> i) & 1) << c);
97 update_rhs_swizzle(rhs_swiz, i, c);
98 rhs_swiz.num_components = swiz->val->type->vector_elements;
99 }
100
101 this->write_mask = write_mask;
102 lhs = swiz->val;
103
104 this->rhs = new(mem_ctx) ir_swizzle(this->rhs, rhs_swiz);
105 swizzled = true;
106 }
107
108 if (swizzled) {
109 /* Now, RHS channels line up with the LHS writemask. Collapse it
110 * to just the channels that will be written.
111 */
112 ir_swizzle_mask rhs_swiz = { 0, 0, 0, 0, 0, 0 };
113 int rhs_chan = 0;
114 for (int i = 0; i < 4; i++) {
115 if (write_mask & (1 << i))
116 update_rhs_swizzle(rhs_swiz, i, rhs_chan++);
117 }
118 rhs_swiz.num_components = rhs_chan;
119 this->rhs = new(mem_ctx) ir_swizzle(this->rhs, rhs_swiz);
120 }
121
122 assert((lhs == NULL) || lhs->as_dereference());
123
124 this->lhs = (ir_dereference *) lhs;
125 }
126
127 ir_variable *
128 ir_assignment::whole_variable_written()
129 {
130 ir_variable *v = this->lhs->whole_variable_referenced();
131
132 if (v == NULL)
133 return NULL;
134
135 if (v->type->is_scalar())
136 return v;
137
138 if (v->type->is_vector()) {
139 const unsigned mask = (1U << v->type->vector_elements) - 1;
140
141 if (mask != this->write_mask)
142 return NULL;
143 }
144
145 /* Either all the vector components are assigned or the variable is some
146 * composite type (and the whole thing is assigned.
147 */
148 return v;
149 }
150
151 ir_assignment::ir_assignment(ir_dereference *lhs, ir_rvalue *rhs,
152 ir_rvalue *condition, unsigned write_mask)
153 : ir_instruction(ir_type_assignment)
154 {
155 this->condition = condition;
156 this->rhs = rhs;
157 this->lhs = lhs;
158 this->write_mask = write_mask;
159
160 if (lhs->type->is_scalar() || lhs->type->is_vector()) {
161 int lhs_components = 0;
162 for (int i = 0; i < 4; i++) {
163 if (write_mask & (1 << i))
164 lhs_components++;
165 }
166
167 assert(lhs_components == this->rhs->type->vector_elements);
168 }
169 }
170
171 ir_assignment::ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs,
172 ir_rvalue *condition)
173 : ir_instruction(ir_type_assignment)
174 {
175 this->condition = condition;
176 this->rhs = rhs;
177
178 /* If the RHS is a vector type, assume that all components of the vector
179 * type are being written to the LHS. The write mask comes from the RHS
180 * because we can have a case where the LHS is a vec4 and the RHS is a
181 * vec3. In that case, the assignment is:
182 *
183 * (assign (...) (xyz) (var_ref lhs) (var_ref rhs))
184 */
185 if (rhs->type->is_vector())
186 this->write_mask = (1U << rhs->type->vector_elements) - 1;
187 else if (rhs->type->is_scalar())
188 this->write_mask = 1;
189 else
190 this->write_mask = 0;
191
192 this->set_lhs(lhs);
193 }
194
195 ir_expression::ir_expression(int op, const struct glsl_type *type,
196 ir_rvalue *op0, ir_rvalue *op1,
197 ir_rvalue *op2, ir_rvalue *op3)
198 : ir_rvalue(ir_type_expression)
199 {
200 this->type = type;
201 this->operation = ir_expression_operation(op);
202 this->operands[0] = op0;
203 this->operands[1] = op1;
204 this->operands[2] = op2;
205 this->operands[3] = op3;
206 init_num_operands();
207
208 #ifndef NDEBUG
209 for (unsigned i = num_operands; i < 4; i++) {
210 assert(this->operands[i] == NULL);
211 }
212
213 for (unsigned i = 0; i < num_operands; i++) {
214 assert(this->operands[i] != NULL);
215 }
216 #endif
217 }
218
219 ir_expression::ir_expression(int op, ir_rvalue *op0)
220 : ir_rvalue(ir_type_expression)
221 {
222 this->operation = ir_expression_operation(op);
223 this->operands[0] = op0;
224 this->operands[1] = NULL;
225 this->operands[2] = NULL;
226 this->operands[3] = NULL;
227
228 assert(op <= ir_last_unop);
229 init_num_operands();
230 assert(num_operands == 1);
231 assert(this->operands[0]);
232
233 switch (this->operation) {
234 case ir_unop_bit_not:
235 case ir_unop_logic_not:
236 case ir_unop_neg:
237 case ir_unop_abs:
238 case ir_unop_sign:
239 case ir_unop_rcp:
240 case ir_unop_rsq:
241 case ir_unop_sqrt:
242 case ir_unop_exp:
243 case ir_unop_log:
244 case ir_unop_exp2:
245 case ir_unop_log2:
246 case ir_unop_trunc:
247 case ir_unop_ceil:
248 case ir_unop_floor:
249 case ir_unop_fract:
250 case ir_unop_round_even:
251 case ir_unop_sin:
252 case ir_unop_cos:
253 case ir_unop_dFdx:
254 case ir_unop_dFdx_coarse:
255 case ir_unop_dFdx_fine:
256 case ir_unop_dFdy:
257 case ir_unop_dFdy_coarse:
258 case ir_unop_dFdy_fine:
259 case ir_unop_bitfield_reverse:
260 case ir_unop_interpolate_at_centroid:
261 case ir_unop_saturate:
262 this->type = op0->type;
263 break;
264
265 case ir_unop_f2i:
266 case ir_unop_b2i:
267 case ir_unop_u2i:
268 case ir_unop_d2i:
269 case ir_unop_bitcast_f2i:
270 case ir_unop_bit_count:
271 case ir_unop_find_msb:
272 case ir_unop_find_lsb:
273 case ir_unop_subroutine_to_int:
274 case ir_unop_i642i:
275 case ir_unop_u642i:
276 this->type = glsl_type::get_instance(GLSL_TYPE_INT,
277 op0->type->vector_elements, 1);
278 break;
279
280 case ir_unop_b2f:
281 case ir_unop_i2f:
282 case ir_unop_u2f:
283 case ir_unop_d2f:
284 case ir_unop_bitcast_i2f:
285 case ir_unop_bitcast_u2f:
286 case ir_unop_i642f:
287 case ir_unop_u642f:
288 this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT,
289 op0->type->vector_elements, 1);
290 break;
291
292 case ir_unop_f2b:
293 case ir_unop_i2b:
294 case ir_unop_d2b:
295 case ir_unop_i642b:
296 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL,
297 op0->type->vector_elements, 1);
298 break;
299
300 case ir_unop_f2d:
301 case ir_unop_i2d:
302 case ir_unop_u2d:
303 case ir_unop_i642d:
304 case ir_unop_u642d:
305 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE,
306 op0->type->vector_elements, 1);
307 break;
308
309 case ir_unop_i2u:
310 case ir_unop_f2u:
311 case ir_unop_d2u:
312 case ir_unop_bitcast_f2u:
313 case ir_unop_i642u:
314 case ir_unop_u642u:
315 this->type = glsl_type::get_instance(GLSL_TYPE_UINT,
316 op0->type->vector_elements, 1);
317 break;
318
319 case ir_unop_i2i64:
320 case ir_unop_u2i64:
321 case ir_unop_b2i64:
322 case ir_unop_f2i64:
323 case ir_unop_d2i64:
324 case ir_unop_u642i64:
325 this->type = glsl_type::get_instance(GLSL_TYPE_INT64,
326 op0->type->vector_elements, 1);
327 break;
328
329 case ir_unop_i2u64:
330 case ir_unop_u2u64:
331 case ir_unop_f2u64:
332 case ir_unop_d2u64:
333 case ir_unop_i642u64:
334 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64,
335 op0->type->vector_elements, 1);
336 break;
337 case ir_unop_noise:
338 this->type = glsl_type::float_type;
339 break;
340
341 case ir_unop_unpack_double_2x32:
342 case ir_unop_unpack_uint_2x32:
343 this->type = glsl_type::uvec2_type;
344 break;
345
346 case ir_unop_unpack_int_2x32:
347 this->type = glsl_type::ivec2_type;
348 break;
349
350 case ir_unop_pack_snorm_2x16:
351 case ir_unop_pack_snorm_4x8:
352 case ir_unop_pack_unorm_2x16:
353 case ir_unop_pack_unorm_4x8:
354 case ir_unop_pack_half_2x16:
355 this->type = glsl_type::uint_type;
356 break;
357
358 case ir_unop_pack_double_2x32:
359 this->type = glsl_type::double_type;
360 break;
361
362 case ir_unop_pack_int_2x32:
363 this->type = glsl_type::int64_t_type;
364 break;
365
366 case ir_unop_pack_uint_2x32:
367 this->type = glsl_type::uint64_t_type;
368 break;
369
370 case ir_unop_unpack_snorm_2x16:
371 case ir_unop_unpack_unorm_2x16:
372 case ir_unop_unpack_half_2x16:
373 this->type = glsl_type::vec2_type;
374 break;
375
376 case ir_unop_unpack_snorm_4x8:
377 case ir_unop_unpack_unorm_4x8:
378 this->type = glsl_type::vec4_type;
379 break;
380
381 case ir_unop_unpack_sampler_2x32:
382 case ir_unop_unpack_image_2x32:
383 this->type = glsl_type::uvec2_type;
384 break;
385
386 case ir_unop_pack_sampler_2x32:
387 case ir_unop_pack_image_2x32:
388 this->type = op0->type;
389 break;
390
391 case ir_unop_frexp_sig:
392 this->type = op0->type;
393 break;
394 case ir_unop_frexp_exp:
395 this->type = glsl_type::get_instance(GLSL_TYPE_INT,
396 op0->type->vector_elements, 1);
397 break;
398
399 case ir_unop_get_buffer_size:
400 case ir_unop_ssbo_unsized_array_length:
401 this->type = glsl_type::int_type;
402 break;
403
404 case ir_unop_bitcast_i642d:
405 case ir_unop_bitcast_u642d:
406 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE,
407 op0->type->vector_elements, 1);
408 break;
409
410 case ir_unop_bitcast_d2i64:
411 this->type = glsl_type::get_instance(GLSL_TYPE_INT64,
412 op0->type->vector_elements, 1);
413 break;
414 case ir_unop_bitcast_d2u64:
415 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64,
416 op0->type->vector_elements, 1);
417 break;
418
419 default:
420 assert(!"not reached: missing automatic type setup for ir_expression");
421 this->type = op0->type;
422 break;
423 }
424 }
425
426 ir_expression::ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1)
427 : ir_rvalue(ir_type_expression)
428 {
429 this->operation = ir_expression_operation(op);
430 this->operands[0] = op0;
431 this->operands[1] = op1;
432 this->operands[2] = NULL;
433 this->operands[3] = NULL;
434
435 assert(op > ir_last_unop);
436 init_num_operands();
437 assert(num_operands == 2);
438 for (unsigned i = 0; i < num_operands; i++) {
439 assert(this->operands[i] != NULL);
440 }
441
442 switch (this->operation) {
443 case ir_binop_all_equal:
444 case ir_binop_any_nequal:
445 this->type = glsl_type::bool_type;
446 break;
447
448 case ir_binop_add:
449 case ir_binop_sub:
450 case ir_binop_min:
451 case ir_binop_max:
452 case ir_binop_pow:
453 case ir_binop_mul:
454 case ir_binop_div:
455 case ir_binop_mod:
456 if (op0->type->is_scalar()) {
457 this->type = op1->type;
458 } else if (op1->type->is_scalar()) {
459 this->type = op0->type;
460 } else {
461 if (this->operation == ir_binop_mul) {
462 this->type = glsl_type::get_mul_type(op0->type, op1->type);
463 } else {
464 assert(op0->type == op1->type);
465 this->type = op0->type;
466 }
467 }
468 break;
469
470 case ir_binop_logic_and:
471 case ir_binop_logic_xor:
472 case ir_binop_logic_or:
473 case ir_binop_bit_and:
474 case ir_binop_bit_xor:
475 case ir_binop_bit_or:
476 assert(!op0->type->is_matrix());
477 assert(!op1->type->is_matrix());
478 if (op0->type->is_scalar()) {
479 this->type = op1->type;
480 } else if (op1->type->is_scalar()) {
481 this->type = op0->type;
482 } else {
483 assert(op0->type->vector_elements == op1->type->vector_elements);
484 this->type = op0->type;
485 }
486 break;
487
488 case ir_binop_equal:
489 case ir_binop_nequal:
490 case ir_binop_lequal:
491 case ir_binop_gequal:
492 case ir_binop_less:
493 case ir_binop_greater:
494 assert(op0->type == op1->type);
495 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL,
496 op0->type->vector_elements, 1);
497 break;
498
499 case ir_binop_dot:
500 this->type = op0->type->get_base_type();
501 break;
502
503 case ir_binop_imul_high:
504 case ir_binop_carry:
505 case ir_binop_borrow:
506 case ir_binop_lshift:
507 case ir_binop_rshift:
508 case ir_binop_ldexp:
509 case ir_binop_interpolate_at_offset:
510 case ir_binop_interpolate_at_sample:
511 this->type = op0->type;
512 break;
513
514 case ir_binop_vector_extract:
515 this->type = op0->type->get_scalar_type();
516 break;
517
518 default:
519 assert(!"not reached: missing automatic type setup for ir_expression");
520 this->type = glsl_type::float_type;
521 }
522 }
523
524 ir_expression::ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1,
525 ir_rvalue *op2)
526 : ir_rvalue(ir_type_expression)
527 {
528 this->operation = ir_expression_operation(op);
529 this->operands[0] = op0;
530 this->operands[1] = op1;
531 this->operands[2] = op2;
532 this->operands[3] = NULL;
533
534 assert(op > ir_last_binop && op <= ir_last_triop);
535 init_num_operands();
536 assert(num_operands == 3);
537 for (unsigned i = 0; i < num_operands; i++) {
538 assert(this->operands[i] != NULL);
539 }
540
541 switch (this->operation) {
542 case ir_triop_fma:
543 case ir_triop_lrp:
544 case ir_triop_bitfield_extract:
545 case ir_triop_vector_insert:
546 this->type = op0->type;
547 break;
548
549 case ir_triop_csel:
550 this->type = op1->type;
551 break;
552
553 default:
554 assert(!"not reached: missing automatic type setup for ir_expression");
555 this->type = glsl_type::float_type;
556 }
557 }
558
559 /**
560 * This is only here for ir_reader to used for testing purposes. Please use
561 * the precomputed num_operands field if you need the number of operands.
562 */
563 unsigned
564 ir_expression::get_num_operands(ir_expression_operation op)
565 {
566 assert(op <= ir_last_opcode);
567
568 if (op <= ir_last_unop)
569 return 1;
570
571 if (op <= ir_last_binop)
572 return 2;
573
574 if (op <= ir_last_triop)
575 return 3;
576
577 if (op <= ir_last_quadop)
578 return 4;
579
580 unreachable("Could not calculate number of operands");
581 }
582
583 #include "ir_expression_operation_strings.h"
584
585 const char*
586 depth_layout_string(ir_depth_layout layout)
587 {
588 switch(layout) {
589 case ir_depth_layout_none: return "";
590 case ir_depth_layout_any: return "depth_any";
591 case ir_depth_layout_greater: return "depth_greater";
592 case ir_depth_layout_less: return "depth_less";
593 case ir_depth_layout_unchanged: return "depth_unchanged";
594
595 default:
596 assert(0);
597 return "";
598 }
599 }
600
601 ir_expression_operation
602 ir_expression::get_operator(const char *str)
603 {
604 for (int op = 0; op <= int(ir_last_opcode); op++) {
605 if (strcmp(str, ir_expression_operation_strings[op]) == 0)
606 return (ir_expression_operation) op;
607 }
608 return (ir_expression_operation) -1;
609 }
610
611 ir_variable *
612 ir_expression::variable_referenced() const
613 {
614 switch (operation) {
615 case ir_binop_vector_extract:
616 case ir_triop_vector_insert:
617 /* We get these for things like a[0] where a is a vector type. In these
618 * cases we want variable_referenced() to return the actual vector
619 * variable this is wrapping.
620 */
621 return operands[0]->variable_referenced();
622 default:
623 return ir_rvalue::variable_referenced();
624 }
625 }
626
627 ir_constant::ir_constant()
628 : ir_rvalue(ir_type_constant)
629 {
630 this->array_elements = NULL;
631 }
632
633 ir_constant::ir_constant(const struct glsl_type *type,
634 const ir_constant_data *data)
635 : ir_rvalue(ir_type_constant)
636 {
637 this->array_elements = NULL;
638
639 assert((type->base_type >= GLSL_TYPE_UINT)
640 && (type->base_type <= GLSL_TYPE_IMAGE));
641
642 this->type = type;
643 memcpy(& this->value, data, sizeof(this->value));
644 }
645
646 ir_constant::ir_constant(float f, unsigned vector_elements)
647 : ir_rvalue(ir_type_constant)
648 {
649 assert(vector_elements <= 4);
650 this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT, vector_elements, 1);
651 for (unsigned i = 0; i < vector_elements; i++) {
652 this->value.f[i] = f;
653 }
654 for (unsigned i = vector_elements; i < 16; i++) {
655 this->value.f[i] = 0;
656 }
657 }
658
659 ir_constant::ir_constant(double d, unsigned vector_elements)
660 : ir_rvalue(ir_type_constant)
661 {
662 assert(vector_elements <= 4);
663 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE, vector_elements, 1);
664 for (unsigned i = 0; i < vector_elements; i++) {
665 this->value.d[i] = d;
666 }
667 for (unsigned i = vector_elements; i < 16; i++) {
668 this->value.d[i] = 0.0;
669 }
670 }
671
672 ir_constant::ir_constant(unsigned int u, unsigned vector_elements)
673 : ir_rvalue(ir_type_constant)
674 {
675 assert(vector_elements <= 4);
676 this->type = glsl_type::get_instance(GLSL_TYPE_UINT, vector_elements, 1);
677 for (unsigned i = 0; i < vector_elements; i++) {
678 this->value.u[i] = u;
679 }
680 for (unsigned i = vector_elements; i < 16; i++) {
681 this->value.u[i] = 0;
682 }
683 }
684
685 ir_constant::ir_constant(int integer, unsigned vector_elements)
686 : ir_rvalue(ir_type_constant)
687 {
688 assert(vector_elements <= 4);
689 this->type = glsl_type::get_instance(GLSL_TYPE_INT, vector_elements, 1);
690 for (unsigned i = 0; i < vector_elements; i++) {
691 this->value.i[i] = integer;
692 }
693 for (unsigned i = vector_elements; i < 16; i++) {
694 this->value.i[i] = 0;
695 }
696 }
697
698 ir_constant::ir_constant(uint64_t u64, unsigned vector_elements)
699 : ir_rvalue(ir_type_constant)
700 {
701 assert(vector_elements <= 4);
702 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64, vector_elements, 1);
703 for (unsigned i = 0; i < vector_elements; i++) {
704 this->value.u64[i] = u64;
705 }
706 for (unsigned i = vector_elements; i < 16; i++) {
707 this->value.u64[i] = 0;
708 }
709 }
710
711 ir_constant::ir_constant(int64_t int64, unsigned vector_elements)
712 : ir_rvalue(ir_type_constant)
713 {
714 assert(vector_elements <= 4);
715 this->type = glsl_type::get_instance(GLSL_TYPE_INT64, vector_elements, 1);
716 for (unsigned i = 0; i < vector_elements; i++) {
717 this->value.i64[i] = int64;
718 }
719 for (unsigned i = vector_elements; i < 16; i++) {
720 this->value.i64[i] = 0;
721 }
722 }
723
724 ir_constant::ir_constant(bool b, unsigned vector_elements)
725 : ir_rvalue(ir_type_constant)
726 {
727 assert(vector_elements <= 4);
728 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL, vector_elements, 1);
729 for (unsigned i = 0; i < vector_elements; i++) {
730 this->value.b[i] = b;
731 }
732 for (unsigned i = vector_elements; i < 16; i++) {
733 this->value.b[i] = false;
734 }
735 }
736
737 ir_constant::ir_constant(const ir_constant *c, unsigned i)
738 : ir_rvalue(ir_type_constant)
739 {
740 this->array_elements = NULL;
741 this->type = c->type->get_base_type();
742
743 switch (this->type->base_type) {
744 case GLSL_TYPE_UINT: this->value.u[0] = c->value.u[i]; break;
745 case GLSL_TYPE_INT: this->value.i[0] = c->value.i[i]; break;
746 case GLSL_TYPE_FLOAT: this->value.f[0] = c->value.f[i]; break;
747 case GLSL_TYPE_BOOL: this->value.b[0] = c->value.b[i]; break;
748 case GLSL_TYPE_DOUBLE: this->value.d[0] = c->value.d[i]; break;
749 default: assert(!"Should not get here."); break;
750 }
751 }
752
753 ir_constant::ir_constant(const struct glsl_type *type, exec_list *value_list)
754 : ir_rvalue(ir_type_constant)
755 {
756 this->array_elements = NULL;
757 this->type = type;
758
759 assert(type->is_scalar() || type->is_vector() || type->is_matrix()
760 || type->is_record() || type->is_array());
761
762 if (type->is_array()) {
763 this->array_elements = ralloc_array(this, ir_constant *, type->length);
764 unsigned i = 0;
765 foreach_in_list(ir_constant, value, value_list) {
766 assert(value->as_constant() != NULL);
767
768 this->array_elements[i++] = value;
769 }
770 return;
771 }
772
773 /* If the constant is a record, the types of each of the entries in
774 * value_list must be a 1-for-1 match with the structure components. Each
775 * entry must also be a constant. Just move the nodes from the value_list
776 * to the list in the ir_constant.
777 */
778 /* FINISHME: Should there be some type checking and / or assertions here? */
779 /* FINISHME: Should the new constant take ownership of the nodes from
780 * FINISHME: value_list, or should it make copies?
781 */
782 if (type->is_record()) {
783 value_list->move_nodes_to(& this->components);
784 return;
785 }
786
787 for (unsigned i = 0; i < 16; i++) {
788 this->value.u[i] = 0;
789 }
790
791 ir_constant *value = (ir_constant *) (value_list->get_head_raw());
792
793 /* Constructors with exactly one scalar argument are special for vectors
794 * and matrices. For vectors, the scalar value is replicated to fill all
795 * the components. For matrices, the scalar fills the components of the
796 * diagonal while the rest is filled with 0.
797 */
798 if (value->type->is_scalar() && value->next->is_tail_sentinel()) {
799 if (type->is_matrix()) {
800 /* Matrix - fill diagonal (rest is already set to 0) */
801 assert(type->is_float() || type->is_double());
802 for (unsigned i = 0; i < type->matrix_columns; i++) {
803 if (type->is_float())
804 this->value.f[i * type->vector_elements + i] =
805 value->value.f[0];
806 else
807 this->value.d[i * type->vector_elements + i] =
808 value->value.d[0];
809 }
810 } else {
811 /* Vector or scalar - fill all components */
812 switch (type->base_type) {
813 case GLSL_TYPE_UINT:
814 case GLSL_TYPE_INT:
815 for (unsigned i = 0; i < type->components(); i++)
816 this->value.u[i] = value->value.u[0];
817 break;
818 case GLSL_TYPE_FLOAT:
819 for (unsigned i = 0; i < type->components(); i++)
820 this->value.f[i] = value->value.f[0];
821 break;
822 case GLSL_TYPE_DOUBLE:
823 for (unsigned i = 0; i < type->components(); i++)
824 this->value.d[i] = value->value.d[0];
825 break;
826 case GLSL_TYPE_UINT64:
827 case GLSL_TYPE_INT64:
828 for (unsigned i = 0; i < type->components(); i++)
829 this->value.u64[i] = value->value.u64[0];
830 break;
831 case GLSL_TYPE_BOOL:
832 for (unsigned i = 0; i < type->components(); i++)
833 this->value.b[i] = value->value.b[0];
834 break;
835 default:
836 assert(!"Should not get here.");
837 break;
838 }
839 }
840 return;
841 }
842
843 if (type->is_matrix() && value->type->is_matrix()) {
844 assert(value->next->is_tail_sentinel());
845
846 /* From section 5.4.2 of the GLSL 1.20 spec:
847 * "If a matrix is constructed from a matrix, then each component
848 * (column i, row j) in the result that has a corresponding component
849 * (column i, row j) in the argument will be initialized from there."
850 */
851 unsigned cols = MIN2(type->matrix_columns, value->type->matrix_columns);
852 unsigned rows = MIN2(type->vector_elements, value->type->vector_elements);
853 for (unsigned i = 0; i < cols; i++) {
854 for (unsigned j = 0; j < rows; j++) {
855 const unsigned src = i * value->type->vector_elements + j;
856 const unsigned dst = i * type->vector_elements + j;
857 this->value.f[dst] = value->value.f[src];
858 }
859 }
860
861 /* "All other components will be initialized to the identity matrix." */
862 for (unsigned i = cols; i < type->matrix_columns; i++)
863 this->value.f[i * type->vector_elements + i] = 1.0;
864
865 return;
866 }
867
868 /* Use each component from each entry in the value_list to initialize one
869 * component of the constant being constructed.
870 */
871 unsigned i = 0;
872 for (;;) {
873 assert(value->as_constant() != NULL);
874 assert(!value->is_tail_sentinel());
875
876 for (unsigned j = 0; j < value->type->components(); j++) {
877 switch (type->base_type) {
878 case GLSL_TYPE_UINT:
879 this->value.u[i] = value->get_uint_component(j);
880 break;
881 case GLSL_TYPE_INT:
882 this->value.i[i] = value->get_int_component(j);
883 break;
884 case GLSL_TYPE_FLOAT:
885 this->value.f[i] = value->get_float_component(j);
886 break;
887 case GLSL_TYPE_BOOL:
888 this->value.b[i] = value->get_bool_component(j);
889 break;
890 case GLSL_TYPE_DOUBLE:
891 this->value.d[i] = value->get_double_component(j);
892 break;
893 case GLSL_TYPE_UINT64:
894 this->value.u64[i] = value->get_uint64_component(j);
895 break;
896 case GLSL_TYPE_INT64:
897 this->value.i64[i] = value->get_int64_component(j);
898 break;
899 default:
900 /* FINISHME: What to do? Exceptions are not the answer.
901 */
902 break;
903 }
904
905 i++;
906 if (i >= type->components())
907 break;
908 }
909
910 if (i >= type->components())
911 break; /* avoid downcasting a list sentinel */
912 value = (ir_constant *) value->next;
913 }
914 }
915
916 ir_constant *
917 ir_constant::zero(void *mem_ctx, const glsl_type *type)
918 {
919 assert(type->is_scalar() || type->is_vector() || type->is_matrix()
920 || type->is_record() || type->is_array());
921
922 ir_constant *c = new(mem_ctx) ir_constant;
923 c->type = type;
924 memset(&c->value, 0, sizeof(c->value));
925
926 if (type->is_array()) {
927 c->array_elements = ralloc_array(c, ir_constant *, type->length);
928
929 for (unsigned i = 0; i < type->length; i++)
930 c->array_elements[i] = ir_constant::zero(c, type->fields.array);
931 }
932
933 if (type->is_record()) {
934 for (unsigned i = 0; i < type->length; i++) {
935 ir_constant *comp = ir_constant::zero(mem_ctx, type->fields.structure[i].type);
936 c->components.push_tail(comp);
937 }
938 }
939
940 return c;
941 }
942
943 bool
944 ir_constant::get_bool_component(unsigned i) const
945 {
946 switch (this->type->base_type) {
947 case GLSL_TYPE_UINT: return this->value.u[i] != 0;
948 case GLSL_TYPE_INT: return this->value.i[i] != 0;
949 case GLSL_TYPE_FLOAT: return ((int)this->value.f[i]) != 0;
950 case GLSL_TYPE_BOOL: return this->value.b[i];
951 case GLSL_TYPE_DOUBLE: return this->value.d[i] != 0.0;
952 case GLSL_TYPE_UINT64: return this->value.u64[i] != 0;
953 case GLSL_TYPE_INT64: return this->value.i64[i] != 0;
954 default: assert(!"Should not get here."); break;
955 }
956
957 /* Must return something to make the compiler happy. This is clearly an
958 * error case.
959 */
960 return false;
961 }
962
963 float
964 ir_constant::get_float_component(unsigned i) const
965 {
966 switch (this->type->base_type) {
967 case GLSL_TYPE_UINT: return (float) this->value.u[i];
968 case GLSL_TYPE_INT: return (float) this->value.i[i];
969 case GLSL_TYPE_FLOAT: return this->value.f[i];
970 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1.0f : 0.0f;
971 case GLSL_TYPE_DOUBLE: return (float) this->value.d[i];
972 case GLSL_TYPE_UINT64: return (float) this->value.u64[i];
973 case GLSL_TYPE_INT64: return (float) this->value.i64[i];
974 default: assert(!"Should not get here."); break;
975 }
976
977 /* Must return something to make the compiler happy. This is clearly an
978 * error case.
979 */
980 return 0.0;
981 }
982
983 double
984 ir_constant::get_double_component(unsigned i) const
985 {
986 switch (this->type->base_type) {
987 case GLSL_TYPE_UINT: return (double) this->value.u[i];
988 case GLSL_TYPE_INT: return (double) this->value.i[i];
989 case GLSL_TYPE_FLOAT: return (double) this->value.f[i];
990 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1.0 : 0.0;
991 case GLSL_TYPE_DOUBLE: return this->value.d[i];
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_UINT64: return (int) this->value.u64[i];
1013 case GLSL_TYPE_INT64: return (int) this->value.i64[i];
1014 default: assert(!"Should not get here."); break;
1015 }
1016
1017 /* Must return something to make the compiler happy. This is clearly an
1018 * error case.
1019 */
1020 return 0;
1021 }
1022
1023 unsigned
1024 ir_constant::get_uint_component(unsigned i) const
1025 {
1026 switch (this->type->base_type) {
1027 case GLSL_TYPE_UINT: return this->value.u[i];
1028 case GLSL_TYPE_INT: return this->value.i[i];
1029 case GLSL_TYPE_FLOAT: return (unsigned) this->value.f[i];
1030 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1031 case GLSL_TYPE_DOUBLE: return (unsigned) this->value.d[i];
1032 case GLSL_TYPE_UINT64: return (unsigned) this->value.u64[i];
1033 case GLSL_TYPE_INT64: return (unsigned) this->value.i64[i];
1034 default: assert(!"Should not get here."); break;
1035 }
1036
1037 /* Must return something to make the compiler happy. This is clearly an
1038 * error case.
1039 */
1040 return 0;
1041 }
1042
1043 int64_t
1044 ir_constant::get_int64_component(unsigned i) const
1045 {
1046 switch (this->type->base_type) {
1047 case GLSL_TYPE_UINT: return this->value.u[i];
1048 case GLSL_TYPE_INT: return this->value.i[i];
1049 case GLSL_TYPE_FLOAT: return (int64_t) this->value.f[i];
1050 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1051 case GLSL_TYPE_DOUBLE: return (int64_t) this->value.d[i];
1052 case GLSL_TYPE_UINT64: return (int64_t) this->value.u64[i];
1053 case GLSL_TYPE_INT64: return this->value.i64[i];
1054 default: assert(!"Should not get here."); break;
1055 }
1056
1057 /* Must return something to make the compiler happy. This is clearly an
1058 * error case.
1059 */
1060 return 0;
1061 }
1062
1063 uint64_t
1064 ir_constant::get_uint64_component(unsigned i) const
1065 {
1066 switch (this->type->base_type) {
1067 case GLSL_TYPE_UINT: return this->value.u[i];
1068 case GLSL_TYPE_INT: return this->value.i[i];
1069 case GLSL_TYPE_FLOAT: return (uint64_t) this->value.f[i];
1070 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1071 case GLSL_TYPE_DOUBLE: return (uint64_t) this->value.d[i];
1072 case GLSL_TYPE_UINT64: return this->value.u64[i];
1073 case GLSL_TYPE_INT64: return (uint64_t) this->value.i64[i];
1074 default: assert(!"Should not get here."); break;
1075 }
1076
1077 /* Must return something to make the compiler happy. This is clearly an
1078 * error case.
1079 */
1080 return 0;
1081 }
1082
1083 ir_constant *
1084 ir_constant::get_array_element(unsigned i) const
1085 {
1086 assert(this->type->is_array());
1087
1088 /* From page 35 (page 41 of the PDF) of the GLSL 1.20 spec:
1089 *
1090 * "Behavior is undefined if a shader subscripts an array with an index
1091 * less than 0 or greater than or equal to the size the array was
1092 * declared with."
1093 *
1094 * Most out-of-bounds accesses are removed before things could get this far.
1095 * There are cases where non-constant array index values can get constant
1096 * folded.
1097 */
1098 if (int(i) < 0)
1099 i = 0;
1100 else if (i >= this->type->length)
1101 i = this->type->length - 1;
1102
1103 return array_elements[i];
1104 }
1105
1106 ir_constant *
1107 ir_constant::get_record_field(int idx)
1108 {
1109 if (idx < 0)
1110 return NULL;
1111
1112 if (this->components.is_empty())
1113 return NULL;
1114
1115 exec_node *node = this->components.get_head_raw();
1116 for (int i = 0; i < idx; i++) {
1117 node = node->next;
1118
1119 /* If the end of the list is encountered before the element matching the
1120 * requested field is found, return NULL.
1121 */
1122 if (node->is_tail_sentinel())
1123 return NULL;
1124 }
1125
1126 return (ir_constant *) node;
1127 }
1128
1129 void
1130 ir_constant::copy_offset(ir_constant *src, int offset)
1131 {
1132 switch (this->type->base_type) {
1133 case GLSL_TYPE_UINT:
1134 case GLSL_TYPE_INT:
1135 case GLSL_TYPE_FLOAT:
1136 case GLSL_TYPE_DOUBLE:
1137 case GLSL_TYPE_UINT64:
1138 case GLSL_TYPE_INT64:
1139 case GLSL_TYPE_BOOL: {
1140 unsigned int size = src->type->components();
1141 assert (size <= this->type->components() - offset);
1142 for (unsigned int i=0; i<size; i++) {
1143 switch (this->type->base_type) {
1144 case GLSL_TYPE_UINT:
1145 value.u[i+offset] = src->get_uint_component(i);
1146 break;
1147 case GLSL_TYPE_INT:
1148 value.i[i+offset] = src->get_int_component(i);
1149 break;
1150 case GLSL_TYPE_FLOAT:
1151 value.f[i+offset] = src->get_float_component(i);
1152 break;
1153 case GLSL_TYPE_BOOL:
1154 value.b[i+offset] = src->get_bool_component(i);
1155 break;
1156 case GLSL_TYPE_DOUBLE:
1157 value.d[i+offset] = src->get_double_component(i);
1158 break;
1159 case GLSL_TYPE_UINT64:
1160 value.u64[i+offset] = src->get_uint64_component(i);
1161 break;
1162 case GLSL_TYPE_INT64:
1163 value.i64[i+offset] = src->get_int64_component(i);
1164 break;
1165 default: // Shut up the compiler
1166 break;
1167 }
1168 }
1169 break;
1170 }
1171
1172 case GLSL_TYPE_STRUCT: {
1173 assert (src->type == this->type);
1174 this->components.make_empty();
1175 foreach_in_list(ir_constant, orig, &src->components) {
1176 this->components.push_tail(orig->clone(this, NULL));
1177 }
1178 break;
1179 }
1180
1181 case GLSL_TYPE_ARRAY: {
1182 assert (src->type == this->type);
1183 for (unsigned i = 0; i < this->type->length; i++) {
1184 this->array_elements[i] = src->array_elements[i]->clone(this, NULL);
1185 }
1186 break;
1187 }
1188
1189 default:
1190 assert(!"Should not get here.");
1191 break;
1192 }
1193 }
1194
1195 void
1196 ir_constant::copy_masked_offset(ir_constant *src, int offset, unsigned int mask)
1197 {
1198 assert (!type->is_array() && !type->is_record());
1199
1200 if (!type->is_vector() && !type->is_matrix()) {
1201 offset = 0;
1202 mask = 1;
1203 }
1204
1205 int id = 0;
1206 for (int i=0; i<4; i++) {
1207 if (mask & (1 << i)) {
1208 switch (this->type->base_type) {
1209 case GLSL_TYPE_UINT:
1210 value.u[i+offset] = src->get_uint_component(id++);
1211 break;
1212 case GLSL_TYPE_INT:
1213 value.i[i+offset] = src->get_int_component(id++);
1214 break;
1215 case GLSL_TYPE_FLOAT:
1216 value.f[i+offset] = src->get_float_component(id++);
1217 break;
1218 case GLSL_TYPE_BOOL:
1219 value.b[i+offset] = src->get_bool_component(id++);
1220 break;
1221 case GLSL_TYPE_DOUBLE:
1222 value.d[i+offset] = src->get_double_component(id++);
1223 break;
1224 case GLSL_TYPE_UINT64:
1225 value.u64[i+offset] = src->get_uint64_component(id++);
1226 break;
1227 case GLSL_TYPE_INT64:
1228 value.i64[i+offset] = src->get_int64_component(id++);
1229 break;
1230 default:
1231 assert(!"Should not get here.");
1232 return;
1233 }
1234 }
1235 }
1236 }
1237
1238 bool
1239 ir_constant::has_value(const ir_constant *c) const
1240 {
1241 if (this->type != c->type)
1242 return false;
1243
1244 if (this->type->is_array()) {
1245 for (unsigned i = 0; i < this->type->length; i++) {
1246 if (!this->array_elements[i]->has_value(c->array_elements[i]))
1247 return false;
1248 }
1249 return true;
1250 }
1251
1252 if (this->type->is_record()) {
1253 const exec_node *a_node = this->components.get_head_raw();
1254 const exec_node *b_node = c->components.get_head_raw();
1255
1256 while (!a_node->is_tail_sentinel()) {
1257 assert(!b_node->is_tail_sentinel());
1258
1259 const ir_constant *const a_field = (ir_constant *) a_node;
1260 const ir_constant *const b_field = (ir_constant *) b_node;
1261
1262 if (!a_field->has_value(b_field))
1263 return false;
1264
1265 a_node = a_node->next;
1266 b_node = b_node->next;
1267 }
1268
1269 return true;
1270 }
1271
1272 for (unsigned i = 0; i < this->type->components(); i++) {
1273 switch (this->type->base_type) {
1274 case GLSL_TYPE_UINT:
1275 if (this->value.u[i] != c->value.u[i])
1276 return false;
1277 break;
1278 case GLSL_TYPE_INT:
1279 if (this->value.i[i] != c->value.i[i])
1280 return false;
1281 break;
1282 case GLSL_TYPE_FLOAT:
1283 if (this->value.f[i] != c->value.f[i])
1284 return false;
1285 break;
1286 case GLSL_TYPE_BOOL:
1287 if (this->value.b[i] != c->value.b[i])
1288 return false;
1289 break;
1290 case GLSL_TYPE_DOUBLE:
1291 if (this->value.d[i] != c->value.d[i])
1292 return false;
1293 break;
1294 case GLSL_TYPE_UINT64:
1295 if (this->value.u64[i] != c->value.u64[i])
1296 return false;
1297 break;
1298 case GLSL_TYPE_INT64:
1299 if (this->value.i64[i] != c->value.i64[i])
1300 return false;
1301 break;
1302 default:
1303 assert(!"Should not get here.");
1304 return false;
1305 }
1306 }
1307
1308 return true;
1309 }
1310
1311 bool
1312 ir_constant::is_value(float f, int i) const
1313 {
1314 if (!this->type->is_scalar() && !this->type->is_vector())
1315 return false;
1316
1317 /* Only accept boolean values for 0/1. */
1318 if (int(bool(i)) != i && this->type->is_boolean())
1319 return false;
1320
1321 for (unsigned c = 0; c < this->type->vector_elements; c++) {
1322 switch (this->type->base_type) {
1323 case GLSL_TYPE_FLOAT:
1324 if (this->value.f[c] != f)
1325 return false;
1326 break;
1327 case GLSL_TYPE_INT:
1328 if (this->value.i[c] != i)
1329 return false;
1330 break;
1331 case GLSL_TYPE_UINT:
1332 if (this->value.u[c] != unsigned(i))
1333 return false;
1334 break;
1335 case GLSL_TYPE_BOOL:
1336 if (this->value.b[c] != bool(i))
1337 return false;
1338 break;
1339 case GLSL_TYPE_DOUBLE:
1340 if (this->value.d[c] != double(f))
1341 return false;
1342 break;
1343 case GLSL_TYPE_UINT64:
1344 if (this->value.u64[c] != uint64_t(i))
1345 return false;
1346 break;
1347 case GLSL_TYPE_INT64:
1348 if (this->value.i64[c] != i)
1349 return false;
1350 break;
1351 default:
1352 /* The only other base types are structures, arrays, and samplers.
1353 * Samplers cannot be constants, and the others should have been
1354 * filtered out above.
1355 */
1356 assert(!"Should not get here.");
1357 return false;
1358 }
1359 }
1360
1361 return true;
1362 }
1363
1364 bool
1365 ir_constant::is_zero() const
1366 {
1367 return is_value(0.0, 0);
1368 }
1369
1370 bool
1371 ir_constant::is_one() const
1372 {
1373 return is_value(1.0, 1);
1374 }
1375
1376 bool
1377 ir_constant::is_negative_one() const
1378 {
1379 return is_value(-1.0, -1);
1380 }
1381
1382 bool
1383 ir_constant::is_uint16_constant() const
1384 {
1385 if (!type->is_integer())
1386 return false;
1387
1388 return value.u[0] < (1 << 16);
1389 }
1390
1391 ir_loop::ir_loop()
1392 : ir_instruction(ir_type_loop)
1393 {
1394 }
1395
1396
1397 ir_dereference_variable::ir_dereference_variable(ir_variable *var)
1398 : ir_dereference(ir_type_dereference_variable)
1399 {
1400 assert(var != NULL);
1401
1402 this->var = var;
1403 this->type = var->type;
1404 }
1405
1406
1407 ir_dereference_array::ir_dereference_array(ir_rvalue *value,
1408 ir_rvalue *array_index)
1409 : ir_dereference(ir_type_dereference_array)
1410 {
1411 this->array_index = array_index;
1412 this->set_array(value);
1413 }
1414
1415
1416 ir_dereference_array::ir_dereference_array(ir_variable *var,
1417 ir_rvalue *array_index)
1418 : ir_dereference(ir_type_dereference_array)
1419 {
1420 void *ctx = ralloc_parent(var);
1421
1422 this->array_index = array_index;
1423 this->set_array(new(ctx) ir_dereference_variable(var));
1424 }
1425
1426
1427 void
1428 ir_dereference_array::set_array(ir_rvalue *value)
1429 {
1430 assert(value != NULL);
1431
1432 this->array = value;
1433
1434 const glsl_type *const vt = this->array->type;
1435
1436 if (vt->is_array()) {
1437 type = vt->fields.array;
1438 } else if (vt->is_matrix()) {
1439 type = vt->column_type();
1440 } else if (vt->is_vector()) {
1441 type = vt->get_base_type();
1442 }
1443 }
1444
1445
1446 ir_dereference_record::ir_dereference_record(ir_rvalue *value,
1447 const char *field)
1448 : ir_dereference(ir_type_dereference_record)
1449 {
1450 assert(value != NULL);
1451
1452 this->record = value;
1453 this->type = this->record->type->field_type(field);
1454 this->field_idx = this->record->type->field_index(field);
1455 }
1456
1457
1458 ir_dereference_record::ir_dereference_record(ir_variable *var,
1459 const char *field)
1460 : ir_dereference(ir_type_dereference_record)
1461 {
1462 void *ctx = ralloc_parent(var);
1463
1464 this->record = new(ctx) ir_dereference_variable(var);
1465 this->type = this->record->type->field_type(field);
1466 this->field_idx = this->record->type->field_index(field);
1467 }
1468
1469 bool
1470 ir_dereference::is_lvalue(const struct _mesa_glsl_parse_state *state) const
1471 {
1472 ir_variable *var = this->variable_referenced();
1473
1474 /* Every l-value derference chain eventually ends in a variable.
1475 */
1476 if ((var == NULL) || var->data.read_only)
1477 return false;
1478
1479 /* From section 4.1.7 of the ARB_bindless_texture spec:
1480 *
1481 * "Samplers can be used as l-values, so can be assigned into and used as
1482 * "out" and "inout" function parameters."
1483 *
1484 * From section 4.1.X of the ARB_bindless_texture spec:
1485 *
1486 * "Images can be used as l-values, so can be assigned into and used as
1487 * "out" and "inout" function parameters."
1488 */
1489 if ((!state || state->has_bindless()) &&
1490 (this->type->contains_sampler() || this->type->contains_image()))
1491 return true;
1492
1493 /* From section 4.1.7 of the GLSL 4.40 spec:
1494 *
1495 * "Opaque variables cannot be treated as l-values; hence cannot
1496 * be used as out or inout function parameters, nor can they be
1497 * assigned into."
1498 */
1499 if (this->type->contains_opaque())
1500 return false;
1501
1502 return true;
1503 }
1504
1505
1506 static const char * const tex_opcode_strs[] = { "tex", "txb", "txl", "txd", "txf", "txf_ms", "txs", "lod", "tg4", "query_levels", "texture_samples", "samples_identical" };
1507
1508 const char *ir_texture::opcode_string()
1509 {
1510 assert((unsigned int) op < ARRAY_SIZE(tex_opcode_strs));
1511 return tex_opcode_strs[op];
1512 }
1513
1514 ir_texture_opcode
1515 ir_texture::get_opcode(const char *str)
1516 {
1517 const int count = sizeof(tex_opcode_strs) / sizeof(tex_opcode_strs[0]);
1518 for (int op = 0; op < count; op++) {
1519 if (strcmp(str, tex_opcode_strs[op]) == 0)
1520 return (ir_texture_opcode) op;
1521 }
1522 return (ir_texture_opcode) -1;
1523 }
1524
1525
1526 void
1527 ir_texture::set_sampler(ir_dereference *sampler, const glsl_type *type)
1528 {
1529 assert(sampler != NULL);
1530 assert(type != NULL);
1531 this->sampler = sampler;
1532 this->type = type;
1533
1534 if (this->op == ir_txs || this->op == ir_query_levels ||
1535 this->op == ir_texture_samples) {
1536 assert(type->base_type == GLSL_TYPE_INT);
1537 } else if (this->op == ir_lod) {
1538 assert(type->vector_elements == 2);
1539 assert(type->is_float());
1540 } else if (this->op == ir_samples_identical) {
1541 assert(type == glsl_type::bool_type);
1542 assert(sampler->type->is_sampler());
1543 assert(sampler->type->sampler_dimensionality == GLSL_SAMPLER_DIM_MS);
1544 } else {
1545 assert(sampler->type->sampled_type == (int) type->base_type);
1546 if (sampler->type->sampler_shadow)
1547 assert(type->vector_elements == 4 || type->vector_elements == 1);
1548 else
1549 assert(type->vector_elements == 4);
1550 }
1551 }
1552
1553
1554 void
1555 ir_swizzle::init_mask(const unsigned *comp, unsigned count)
1556 {
1557 assert((count >= 1) && (count <= 4));
1558
1559 memset(&this->mask, 0, sizeof(this->mask));
1560 this->mask.num_components = count;
1561
1562 unsigned dup_mask = 0;
1563 switch (count) {
1564 case 4:
1565 assert(comp[3] <= 3);
1566 dup_mask |= (1U << comp[3])
1567 & ((1U << comp[0]) | (1U << comp[1]) | (1U << comp[2]));
1568 this->mask.w = comp[3];
1569
1570 case 3:
1571 assert(comp[2] <= 3);
1572 dup_mask |= (1U << comp[2])
1573 & ((1U << comp[0]) | (1U << comp[1]));
1574 this->mask.z = comp[2];
1575
1576 case 2:
1577 assert(comp[1] <= 3);
1578 dup_mask |= (1U << comp[1])
1579 & ((1U << comp[0]));
1580 this->mask.y = comp[1];
1581
1582 case 1:
1583 assert(comp[0] <= 3);
1584 this->mask.x = comp[0];
1585 }
1586
1587 this->mask.has_duplicates = dup_mask != 0;
1588
1589 /* Based on the number of elements in the swizzle and the base type
1590 * (i.e., float, int, unsigned, or bool) of the vector being swizzled,
1591 * generate the type of the resulting value.
1592 */
1593 type = glsl_type::get_instance(val->type->base_type, mask.num_components, 1);
1594 }
1595
1596 ir_swizzle::ir_swizzle(ir_rvalue *val, unsigned x, unsigned y, unsigned z,
1597 unsigned w, unsigned count)
1598 : ir_rvalue(ir_type_swizzle), val(val)
1599 {
1600 const unsigned components[4] = { x, y, z, w };
1601 this->init_mask(components, count);
1602 }
1603
1604 ir_swizzle::ir_swizzle(ir_rvalue *val, const unsigned *comp,
1605 unsigned count)
1606 : ir_rvalue(ir_type_swizzle), val(val)
1607 {
1608 this->init_mask(comp, count);
1609 }
1610
1611 ir_swizzle::ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask)
1612 : ir_rvalue(ir_type_swizzle), val(val), mask(mask)
1613 {
1614 this->type = glsl_type::get_instance(val->type->base_type,
1615 mask.num_components, 1);
1616 }
1617
1618 #define X 1
1619 #define R 5
1620 #define S 9
1621 #define I 13
1622
1623 ir_swizzle *
1624 ir_swizzle::create(ir_rvalue *val, const char *str, unsigned vector_length)
1625 {
1626 void *ctx = ralloc_parent(val);
1627
1628 /* For each possible swizzle character, this table encodes the value in
1629 * \c idx_map that represents the 0th element of the vector. For invalid
1630 * swizzle characters (e.g., 'k'), a special value is used that will allow
1631 * detection of errors.
1632 */
1633 static const unsigned char base_idx[26] = {
1634 /* a b c d e f g h i j k l m */
1635 R, R, I, I, I, I, R, I, I, I, I, I, I,
1636 /* n o p q r s t u v w x y z */
1637 I, I, S, S, R, S, S, I, I, X, X, X, X
1638 };
1639
1640 /* Each valid swizzle character has an entry in the previous table. This
1641 * table encodes the base index encoded in the previous table plus the actual
1642 * index of the swizzle character. When processing swizzles, the first
1643 * character in the string is indexed in the previous table. Each character
1644 * in the string is indexed in this table, and the value found there has the
1645 * value form the first table subtracted. The result must be on the range
1646 * [0,3].
1647 *
1648 * For example, the string "wzyx" will get X from the first table. Each of
1649 * the charcaters will get X+3, X+2, X+1, and X+0 from this table. After
1650 * subtraction, the swizzle values are { 3, 2, 1, 0 }.
1651 *
1652 * The string "wzrg" will get X from the first table. Each of the characters
1653 * will get X+3, X+2, R+0, and R+1 from this table. After subtraction, the
1654 * swizzle values are { 3, 2, 4, 5 }. Since 4 and 5 are outside the range
1655 * [0,3], the error is detected.
1656 */
1657 static const unsigned char idx_map[26] = {
1658 /* a b c d e f g h i j k l m */
1659 R+3, R+2, 0, 0, 0, 0, R+1, 0, 0, 0, 0, 0, 0,
1660 /* n o p q r s t u v w x y z */
1661 0, 0, S+2, S+3, R+0, S+0, S+1, 0, 0, X+3, X+0, X+1, X+2
1662 };
1663
1664 int swiz_idx[4] = { 0, 0, 0, 0 };
1665 unsigned i;
1666
1667
1668 /* Validate the first character in the swizzle string and look up the base
1669 * index value as described above.
1670 */
1671 if ((str[0] < 'a') || (str[0] > 'z'))
1672 return NULL;
1673
1674 const unsigned base = base_idx[str[0] - 'a'];
1675
1676
1677 for (i = 0; (i < 4) && (str[i] != '\0'); i++) {
1678 /* Validate the next character, and, as described above, convert it to a
1679 * swizzle index.
1680 */
1681 if ((str[i] < 'a') || (str[i] > 'z'))
1682 return NULL;
1683
1684 swiz_idx[i] = idx_map[str[i] - 'a'] - base;
1685 if ((swiz_idx[i] < 0) || (swiz_idx[i] >= (int) vector_length))
1686 return NULL;
1687 }
1688
1689 if (str[i] != '\0')
1690 return NULL;
1691
1692 return new(ctx) ir_swizzle(val, swiz_idx[0], swiz_idx[1], swiz_idx[2],
1693 swiz_idx[3], i);
1694 }
1695
1696 #undef X
1697 #undef R
1698 #undef S
1699 #undef I
1700
1701 ir_variable *
1702 ir_swizzle::variable_referenced() const
1703 {
1704 return this->val->variable_referenced();
1705 }
1706
1707
1708 bool ir_variable::temporaries_allocate_names = false;
1709
1710 const char ir_variable::tmp_name[] = "compiler_temp";
1711
1712 ir_variable::ir_variable(const struct glsl_type *type, const char *name,
1713 ir_variable_mode mode)
1714 : ir_instruction(ir_type_variable)
1715 {
1716 this->type = type;
1717
1718 if (mode == ir_var_temporary && !ir_variable::temporaries_allocate_names)
1719 name = NULL;
1720
1721 /* The ir_variable clone method may call this constructor with name set to
1722 * tmp_name.
1723 */
1724 assert(name != NULL
1725 || mode == ir_var_temporary
1726 || mode == ir_var_function_in
1727 || mode == ir_var_function_out
1728 || mode == ir_var_function_inout);
1729 assert(name != ir_variable::tmp_name
1730 || mode == ir_var_temporary);
1731 if (mode == ir_var_temporary
1732 && (name == NULL || name == ir_variable::tmp_name)) {
1733 this->name = ir_variable::tmp_name;
1734 } else if (name == NULL ||
1735 strlen(name) < ARRAY_SIZE(this->name_storage)) {
1736 strcpy(this->name_storage, name ? name : "");
1737 this->name = this->name_storage;
1738 } else {
1739 this->name = ralloc_strdup(this, name);
1740 }
1741
1742 this->u.max_ifc_array_access = NULL;
1743
1744 this->data.explicit_location = false;
1745 this->data.has_initializer = false;
1746 this->data.location = -1;
1747 this->data.location_frac = 0;
1748 this->data.binding = 0;
1749 this->data.warn_extension_index = 0;
1750 this->constant_value = NULL;
1751 this->constant_initializer = NULL;
1752 this->data.origin_upper_left = false;
1753 this->data.pixel_center_integer = false;
1754 this->data.depth_layout = ir_depth_layout_none;
1755 this->data.used = false;
1756 this->data.always_active_io = false;
1757 this->data.read_only = false;
1758 this->data.centroid = false;
1759 this->data.sample = false;
1760 this->data.patch = false;
1761 this->data.invariant = false;
1762 this->data.how_declared = ir_var_declared_normally;
1763 this->data.mode = mode;
1764 this->data.interpolation = INTERP_MODE_NONE;
1765 this->data.max_array_access = -1;
1766 this->data.offset = 0;
1767 this->data.precision = GLSL_PRECISION_NONE;
1768 this->data.memory_read_only = false;
1769 this->data.memory_write_only = false;
1770 this->data.memory_coherent = false;
1771 this->data.memory_volatile = false;
1772 this->data.memory_restrict = false;
1773 this->data.from_ssbo_unsized_array = false;
1774 this->data.fb_fetch_output = false;
1775 this->data.bindless = false;
1776 this->data.bound = false;
1777
1778 if (type != NULL) {
1779 if (type->is_interface())
1780 this->init_interface_type(type);
1781 else if (type->without_array()->is_interface())
1782 this->init_interface_type(type->without_array());
1783 }
1784 }
1785
1786
1787 const char *
1788 interpolation_string(unsigned interpolation)
1789 {
1790 switch (interpolation) {
1791 case INTERP_MODE_NONE: return "no";
1792 case INTERP_MODE_SMOOTH: return "smooth";
1793 case INTERP_MODE_FLAT: return "flat";
1794 case INTERP_MODE_NOPERSPECTIVE: return "noperspective";
1795 }
1796
1797 assert(!"Should not get here.");
1798 return "";
1799 }
1800
1801 const char *const ir_variable::warn_extension_table[] = {
1802 "",
1803 "GL_ARB_shader_stencil_export",
1804 "GL_AMD_shader_stencil_export",
1805 };
1806
1807 void
1808 ir_variable::enable_extension_warning(const char *extension)
1809 {
1810 for (unsigned i = 0; i < ARRAY_SIZE(warn_extension_table); i++) {
1811 if (strcmp(warn_extension_table[i], extension) == 0) {
1812 this->data.warn_extension_index = i;
1813 return;
1814 }
1815 }
1816
1817 assert(!"Should not get here.");
1818 this->data.warn_extension_index = 0;
1819 }
1820
1821 const char *
1822 ir_variable::get_extension_warning() const
1823 {
1824 return this->data.warn_extension_index == 0
1825 ? NULL : warn_extension_table[this->data.warn_extension_index];
1826 }
1827
1828 ir_function_signature::ir_function_signature(const glsl_type *return_type,
1829 builtin_available_predicate b)
1830 : ir_instruction(ir_type_function_signature),
1831 return_type(return_type), is_defined(false),
1832 intrinsic_id(ir_intrinsic_invalid), builtin_avail(b), _function(NULL)
1833 {
1834 this->origin = NULL;
1835 }
1836
1837
1838 bool
1839 ir_function_signature::is_builtin() const
1840 {
1841 return builtin_avail != NULL;
1842 }
1843
1844
1845 bool
1846 ir_function_signature::is_builtin_available(const _mesa_glsl_parse_state *state) const
1847 {
1848 /* We can't call the predicate without a state pointer, so just say that
1849 * the signature is available. At compile time, we need the filtering,
1850 * but also receive a valid state pointer. At link time, we're resolving
1851 * imported built-in prototypes to their definitions, which will always
1852 * be an exact match. So we can skip the filtering.
1853 */
1854 if (state == NULL)
1855 return true;
1856
1857 assert(builtin_avail != NULL);
1858 return builtin_avail(state);
1859 }
1860
1861
1862 static bool
1863 modes_match(unsigned a, unsigned b)
1864 {
1865 if (a == b)
1866 return true;
1867
1868 /* Accept "in" vs. "const in" */
1869 if ((a == ir_var_const_in && b == ir_var_function_in) ||
1870 (b == ir_var_const_in && a == ir_var_function_in))
1871 return true;
1872
1873 return false;
1874 }
1875
1876
1877 const char *
1878 ir_function_signature::qualifiers_match(exec_list *params)
1879 {
1880 /* check that the qualifiers match. */
1881 foreach_two_lists(a_node, &this->parameters, b_node, params) {
1882 ir_variable *a = (ir_variable *) a_node;
1883 ir_variable *b = (ir_variable *) b_node;
1884
1885 if (a->data.read_only != b->data.read_only ||
1886 !modes_match(a->data.mode, b->data.mode) ||
1887 a->data.interpolation != b->data.interpolation ||
1888 a->data.centroid != b->data.centroid ||
1889 a->data.sample != b->data.sample ||
1890 a->data.patch != b->data.patch ||
1891 a->data.memory_read_only != b->data.memory_read_only ||
1892 a->data.memory_write_only != b->data.memory_write_only ||
1893 a->data.memory_coherent != b->data.memory_coherent ||
1894 a->data.memory_volatile != b->data.memory_volatile ||
1895 a->data.memory_restrict != b->data.memory_restrict) {
1896
1897 /* parameter a's qualifiers don't match */
1898 return a->name;
1899 }
1900 }
1901 return NULL;
1902 }
1903
1904
1905 void
1906 ir_function_signature::replace_parameters(exec_list *new_params)
1907 {
1908 /* Destroy all of the previous parameter information. If the previous
1909 * parameter information comes from the function prototype, it may either
1910 * specify incorrect parameter names or not have names at all.
1911 */
1912 new_params->move_nodes_to(&parameters);
1913 }
1914
1915
1916 ir_function::ir_function(const char *name)
1917 : ir_instruction(ir_type_function)
1918 {
1919 this->subroutine_index = -1;
1920 this->name = ralloc_strdup(this, name);
1921 }
1922
1923
1924 bool
1925 ir_function::has_user_signature()
1926 {
1927 foreach_in_list(ir_function_signature, sig, &this->signatures) {
1928 if (!sig->is_builtin())
1929 return true;
1930 }
1931 return false;
1932 }
1933
1934
1935 ir_rvalue *
1936 ir_rvalue::error_value(void *mem_ctx)
1937 {
1938 ir_rvalue *v = new(mem_ctx) ir_rvalue(ir_type_unset);
1939
1940 v->type = glsl_type::error_type;
1941 return v;
1942 }
1943
1944
1945 void
1946 visit_exec_list(exec_list *list, ir_visitor *visitor)
1947 {
1948 foreach_in_list_safe(ir_instruction, node, list) {
1949 node->accept(visitor);
1950 }
1951 }
1952
1953
1954 static void
1955 steal_memory(ir_instruction *ir, void *new_ctx)
1956 {
1957 ir_variable *var = ir->as_variable();
1958 ir_function *fn = ir->as_function();
1959 ir_constant *constant = ir->as_constant();
1960 if (var != NULL && var->constant_value != NULL)
1961 steal_memory(var->constant_value, ir);
1962
1963 if (var != NULL && var->constant_initializer != NULL)
1964 steal_memory(var->constant_initializer, ir);
1965
1966 if (fn != NULL && fn->subroutine_types)
1967 ralloc_steal(new_ctx, fn->subroutine_types);
1968
1969 /* The components of aggregate constants are not visited by the normal
1970 * visitor, so steal their values by hand.
1971 */
1972 if (constant != NULL) {
1973 if (constant->type->is_record()) {
1974 foreach_in_list(ir_constant, field, &constant->components) {
1975 steal_memory(field, ir);
1976 }
1977 } else if (constant->type->is_array()) {
1978 for (unsigned int i = 0; i < constant->type->length; i++) {
1979 steal_memory(constant->array_elements[i], ir);
1980 }
1981 }
1982 }
1983
1984 ralloc_steal(new_ctx, ir);
1985 }
1986
1987
1988 void
1989 reparent_ir(exec_list *list, void *mem_ctx)
1990 {
1991 foreach_in_list(ir_instruction, node, list) {
1992 visit_tree(node, steal_memory, mem_ctx);
1993 }
1994 }
1995
1996
1997 static ir_rvalue *
1998 try_min_one(ir_rvalue *ir)
1999 {
2000 ir_expression *expr = ir->as_expression();
2001
2002 if (!expr || expr->operation != ir_binop_min)
2003 return NULL;
2004
2005 if (expr->operands[0]->is_one())
2006 return expr->operands[1];
2007
2008 if (expr->operands[1]->is_one())
2009 return expr->operands[0];
2010
2011 return NULL;
2012 }
2013
2014 static ir_rvalue *
2015 try_max_zero(ir_rvalue *ir)
2016 {
2017 ir_expression *expr = ir->as_expression();
2018
2019 if (!expr || expr->operation != ir_binop_max)
2020 return NULL;
2021
2022 if (expr->operands[0]->is_zero())
2023 return expr->operands[1];
2024
2025 if (expr->operands[1]->is_zero())
2026 return expr->operands[0];
2027
2028 return NULL;
2029 }
2030
2031 ir_rvalue *
2032 ir_rvalue::as_rvalue_to_saturate()
2033 {
2034 ir_expression *expr = this->as_expression();
2035
2036 if (!expr)
2037 return NULL;
2038
2039 ir_rvalue *max_zero = try_max_zero(expr);
2040 if (max_zero) {
2041 return try_min_one(max_zero);
2042 } else {
2043 ir_rvalue *min_one = try_min_one(expr);
2044 if (min_one) {
2045 return try_max_zero(min_one);
2046 }
2047 }
2048
2049 return NULL;
2050 }
2051
2052
2053 unsigned
2054 vertices_per_prim(GLenum prim)
2055 {
2056 switch (prim) {
2057 case GL_POINTS:
2058 return 1;
2059 case GL_LINES:
2060 return 2;
2061 case GL_TRIANGLES:
2062 return 3;
2063 case GL_LINES_ADJACENCY:
2064 return 4;
2065 case GL_TRIANGLES_ADJACENCY:
2066 return 6;
2067 default:
2068 assert(!"Bad primitive");
2069 return 3;
2070 }
2071 }
2072
2073 /**
2074 * Generate a string describing the mode of a variable
2075 */
2076 const char *
2077 mode_string(const ir_variable *var)
2078 {
2079 switch (var->data.mode) {
2080 case ir_var_auto:
2081 return (var->data.read_only) ? "global constant" : "global variable";
2082
2083 case ir_var_uniform:
2084 return "uniform";
2085
2086 case ir_var_shader_storage:
2087 return "buffer";
2088
2089 case ir_var_shader_in:
2090 return "shader input";
2091
2092 case ir_var_shader_out:
2093 return "shader output";
2094
2095 case ir_var_function_in:
2096 case ir_var_const_in:
2097 return "function input";
2098
2099 case ir_var_function_out:
2100 return "function output";
2101
2102 case ir_var_function_inout:
2103 return "function inout";
2104
2105 case ir_var_system_value:
2106 return "shader input";
2107
2108 case ir_var_temporary:
2109 return "compiler temporary";
2110
2111 case ir_var_mode_count:
2112 break;
2113 }
2114
2115 assert(!"Should not get here.");
2116 return "invalid variable";
2117 }