projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
glsl: Add opcodes for atan and atan2
[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 case ir_unop_atan:
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 case ir_binop_atan2:
457 if (op0->type->is_scalar()) {
458 this->type = op1->type;
459 } else if (op1->type->is_scalar()) {
460 this->type = op0->type;
461 } else {
462 if (this->operation == ir_binop_mul) {
463 this->type = glsl_type::get_mul_type(op0->type, op1->type);
464 } else {
465 assert(op0->type == op1->type);
466 this->type = op0->type;
467 }
468 }
469 break;
470
471 case ir_binop_logic_and:
472 case ir_binop_logic_xor:
473 case ir_binop_logic_or:
474 case ir_binop_bit_and:
475 case ir_binop_bit_xor:
476 case ir_binop_bit_or:
477 assert(!op0->type->is_matrix());
478 assert(!op1->type->is_matrix());
479 if (op0->type->is_scalar()) {
480 this->type = op1->type;
481 } else if (op1->type->is_scalar()) {
482 this->type = op0->type;
483 } else {
484 assert(op0->type->vector_elements == op1->type->vector_elements);
485 this->type = op0->type;
486 }
487 break;
488
489 case ir_binop_equal:
490 case ir_binop_nequal:
491 case ir_binop_gequal:
492 case ir_binop_less:
493 assert(op0->type == op1->type);
494 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL,
495 op0->type->vector_elements, 1);
496 break;
497
498 case ir_binop_dot:
499 this->type = op0->type->get_base_type();
500 break;
501
502 case ir_binop_imul_high:
503 case ir_binop_carry:
504 case ir_binop_borrow:
505 case ir_binop_lshift:
506 case ir_binop_rshift:
507 case ir_binop_ldexp:
508 case ir_binop_interpolate_at_offset:
509 case ir_binop_interpolate_at_sample:
510 this->type = op0->type;
511 break;
512
513 case ir_binop_vector_extract:
514 this->type = op0->type->get_scalar_type();
515 break;
516
517 default:
518 assert(!"not reached: missing automatic type setup for ir_expression");
519 this->type = glsl_type::float_type;
520 }
521 }
522
523 ir_expression::ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1,
524 ir_rvalue *op2)
525 : ir_rvalue(ir_type_expression)
526 {
527 this->operation = ir_expression_operation(op);
528 this->operands[0] = op0;
529 this->operands[1] = op1;
530 this->operands[2] = op2;
531 this->operands[3] = NULL;
532
533 assert(op > ir_last_binop && op <= ir_last_triop);
534 init_num_operands();
535 assert(num_operands == 3);
536 for (unsigned i = 0; i < num_operands; i++) {
537 assert(this->operands[i] != NULL);
538 }
539
540 switch (this->operation) {
541 case ir_triop_fma:
542 case ir_triop_lrp:
543 case ir_triop_bitfield_extract:
544 case ir_triop_vector_insert:
545 this->type = op0->type;
546 break;
547
548 case ir_triop_csel:
549 this->type = op1->type;
550 break;
551
552 default:
553 assert(!"not reached: missing automatic type setup for ir_expression");
554 this->type = glsl_type::float_type;
555 }
556 }
557
558 /**
559 * This is only here for ir_reader to used for testing purposes. Please use
560 * the precomputed num_operands field if you need the number of operands.
561 */
562 unsigned
563 ir_expression::get_num_operands(ir_expression_operation op)
564 {
565 assert(op <= ir_last_opcode);
566
567 if (op <= ir_last_unop)
568 return 1;
569
570 if (op <= ir_last_binop)
571 return 2;
572
573 if (op <= ir_last_triop)
574 return 3;
575
576 if (op <= ir_last_quadop)
577 return 4;
578
579 unreachable("Could not calculate number of operands");
580 }
581
582 #include "ir_expression_operation_strings.h"
583
584 const char*
585 depth_layout_string(ir_depth_layout layout)
586 {
587 switch(layout) {
588 case ir_depth_layout_none: return "";
589 case ir_depth_layout_any: return "depth_any";
590 case ir_depth_layout_greater: return "depth_greater";
591 case ir_depth_layout_less: return "depth_less";
592 case ir_depth_layout_unchanged: return "depth_unchanged";
593
594 default:
595 assert(0);
596 return "";
597 }
598 }
599
600 ir_expression_operation
601 ir_expression::get_operator(const char *str)
602 {
603 for (int op = 0; op <= int(ir_last_opcode); op++) {
604 if (strcmp(str, ir_expression_operation_strings[op]) == 0)
605 return (ir_expression_operation) op;
606 }
607 return (ir_expression_operation) -1;
608 }
609
610 ir_variable *
611 ir_expression::variable_referenced() const
612 {
613 switch (operation) {
614 case ir_binop_vector_extract:
615 case ir_triop_vector_insert:
616 /* We get these for things like a[0] where a is a vector type. In these
617 * cases we want variable_referenced() to return the actual vector
618 * variable this is wrapping.
619 */
620 return operands[0]->variable_referenced();
621 default:
622 return ir_rvalue::variable_referenced();
623 }
624 }
625
626 ir_constant::ir_constant()
627 : ir_rvalue(ir_type_constant)
628 {
629 this->const_elements = NULL;
630 }
631
632 ir_constant::ir_constant(const struct glsl_type *type,
633 const ir_constant_data *data)
634 : ir_rvalue(ir_type_constant)
635 {
636 this->const_elements = NULL;
637
638 assert((type->base_type >= GLSL_TYPE_UINT)
639 && (type->base_type <= GLSL_TYPE_IMAGE));
640
641 this->type = type;
642 memcpy(& this->value, data, sizeof(this->value));
643 }
644
645 ir_constant::ir_constant(float f, unsigned vector_elements)
646 : ir_rvalue(ir_type_constant)
647 {
648 assert(vector_elements <= 4);
649 this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT, vector_elements, 1);
650 for (unsigned i = 0; i < vector_elements; i++) {
651 this->value.f[i] = f;
652 }
653 for (unsigned i = vector_elements; i < 16; i++) {
654 this->value.f[i] = 0;
655 }
656 }
657
658 ir_constant::ir_constant(double d, unsigned vector_elements)
659 : ir_rvalue(ir_type_constant)
660 {
661 assert(vector_elements <= 4);
662 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE, vector_elements, 1);
663 for (unsigned i = 0; i < vector_elements; i++) {
664 this->value.d[i] = d;
665 }
666 for (unsigned i = vector_elements; i < 16; i++) {
667 this->value.d[i] = 0.0;
668 }
669 }
670
671 ir_constant::ir_constant(unsigned int u, unsigned vector_elements)
672 : ir_rvalue(ir_type_constant)
673 {
674 assert(vector_elements <= 4);
675 this->type = glsl_type::get_instance(GLSL_TYPE_UINT, vector_elements, 1);
676 for (unsigned i = 0; i < vector_elements; i++) {
677 this->value.u[i] = u;
678 }
679 for (unsigned i = vector_elements; i < 16; i++) {
680 this->value.u[i] = 0;
681 }
682 }
683
684 ir_constant::ir_constant(int integer, unsigned vector_elements)
685 : ir_rvalue(ir_type_constant)
686 {
687 assert(vector_elements <= 4);
688 this->type = glsl_type::get_instance(GLSL_TYPE_INT, vector_elements, 1);
689 for (unsigned i = 0; i < vector_elements; i++) {
690 this->value.i[i] = integer;
691 }
692 for (unsigned i = vector_elements; i < 16; i++) {
693 this->value.i[i] = 0;
694 }
695 }
696
697 ir_constant::ir_constant(uint64_t u64, unsigned vector_elements)
698 : ir_rvalue(ir_type_constant)
699 {
700 assert(vector_elements <= 4);
701 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64, vector_elements, 1);
702 for (unsigned i = 0; i < vector_elements; i++) {
703 this->value.u64[i] = u64;
704 }
705 for (unsigned i = vector_elements; i < 16; i++) {
706 this->value.u64[i] = 0;
707 }
708 }
709
710 ir_constant::ir_constant(int64_t int64, unsigned vector_elements)
711 : ir_rvalue(ir_type_constant)
712 {
713 assert(vector_elements <= 4);
714 this->type = glsl_type::get_instance(GLSL_TYPE_INT64, vector_elements, 1);
715 for (unsigned i = 0; i < vector_elements; i++) {
716 this->value.i64[i] = int64;
717 }
718 for (unsigned i = vector_elements; i < 16; i++) {
719 this->value.i64[i] = 0;
720 }
721 }
722
723 ir_constant::ir_constant(bool b, unsigned vector_elements)
724 : ir_rvalue(ir_type_constant)
725 {
726 assert(vector_elements <= 4);
727 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL, vector_elements, 1);
728 for (unsigned i = 0; i < vector_elements; i++) {
729 this->value.b[i] = b;
730 }
731 for (unsigned i = vector_elements; i < 16; i++) {
732 this->value.b[i] = false;
733 }
734 }
735
736 ir_constant::ir_constant(const ir_constant *c, unsigned i)
737 : ir_rvalue(ir_type_constant)
738 {
739 this->const_elements = NULL;
740 this->type = c->type->get_base_type();
741
742 switch (this->type->base_type) {
743 case GLSL_TYPE_UINT: this->value.u[0] = c->value.u[i]; break;
744 case GLSL_TYPE_INT: this->value.i[0] = c->value.i[i]; break;
745 case GLSL_TYPE_FLOAT: this->value.f[0] = c->value.f[i]; break;
746 case GLSL_TYPE_BOOL: this->value.b[0] = c->value.b[i]; break;
747 case GLSL_TYPE_DOUBLE: this->value.d[0] = c->value.d[i]; break;
748 default: assert(!"Should not get here."); break;
749 }
750 }
751
752 ir_constant::ir_constant(const struct glsl_type *type, exec_list *value_list)
753 : ir_rvalue(ir_type_constant)
754 {
755 this->const_elements = NULL;
756 this->type = type;
757
758 assert(type->is_scalar() || type->is_vector() || type->is_matrix()
759 || type->is_struct() || type->is_array());
760
761 /* If the constant is a record, the types of each of the entries in
762 * value_list must be a 1-for-1 match with the structure components. Each
763 * entry must also be a constant. Just move the nodes from the value_list
764 * to the list in the ir_constant.
765 */
766 if (type->is_array() || type->is_struct()) {
767 this->const_elements = ralloc_array(this, ir_constant *, type->length);
768 unsigned i = 0;
769 foreach_in_list(ir_constant, value, value_list) {
770 assert(value->as_constant() != NULL);
771
772 this->const_elements[i++] = value;
773 }
774 return;
775 }
776
777 for (unsigned i = 0; i < 16; i++) {
778 this->value.u[i] = 0;
779 }
780
781 ir_constant *value = (ir_constant *) (value_list->get_head_raw());
782
783 /* Constructors with exactly one scalar argument are special for vectors
784 * and matrices. For vectors, the scalar value is replicated to fill all
785 * the components. For matrices, the scalar fills the components of the
786 * diagonal while the rest is filled with 0.
787 */
788 if (value->type->is_scalar() && value->next->is_tail_sentinel()) {
789 if (type->is_matrix()) {
790 /* Matrix - fill diagonal (rest is already set to 0) */
791 assert(type->is_float() || type->is_double());
792 for (unsigned i = 0; i < type->matrix_columns; i++) {
793 if (type->is_float())
794 this->value.f[i * type->vector_elements + i] =
795 value->value.f[0];
796 else
797 this->value.d[i * type->vector_elements + i] =
798 value->value.d[0];
799 }
800 } else {
801 /* Vector or scalar - fill all components */
802 switch (type->base_type) {
803 case GLSL_TYPE_UINT:
804 case GLSL_TYPE_INT:
805 for (unsigned i = 0; i < type->components(); i++)
806 this->value.u[i] = value->value.u[0];
807 break;
808 case GLSL_TYPE_FLOAT:
809 for (unsigned i = 0; i < type->components(); i++)
810 this->value.f[i] = value->value.f[0];
811 break;
812 case GLSL_TYPE_DOUBLE:
813 for (unsigned i = 0; i < type->components(); i++)
814 this->value.d[i] = value->value.d[0];
815 break;
816 case GLSL_TYPE_UINT64:
817 case GLSL_TYPE_INT64:
818 for (unsigned i = 0; i < type->components(); i++)
819 this->value.u64[i] = value->value.u64[0];
820 break;
821 case GLSL_TYPE_BOOL:
822 for (unsigned i = 0; i < type->components(); i++)
823 this->value.b[i] = value->value.b[0];
824 break;
825 case GLSL_TYPE_SAMPLER:
826 case GLSL_TYPE_IMAGE:
827 this->value.u64[0] = value->value.u64[0];
828 break;
829 default:
830 assert(!"Should not get here.");
831 break;
832 }
833 }
834 return;
835 }
836
837 if (type->is_matrix() && value->type->is_matrix()) {
838 assert(value->next->is_tail_sentinel());
839
840 /* From section 5.4.2 of the GLSL 1.20 spec:
841 * "If a matrix is constructed from a matrix, then each component
842 * (column i, row j) in the result that has a corresponding component
843 * (column i, row j) in the argument will be initialized from there."
844 */
845 unsigned cols = MIN2(type->matrix_columns, value->type->matrix_columns);
846 unsigned rows = MIN2(type->vector_elements, value->type->vector_elements);
847 for (unsigned i = 0; i < cols; i++) {
848 for (unsigned j = 0; j < rows; j++) {
849 const unsigned src = i * value->type->vector_elements + j;
850 const unsigned dst = i * type->vector_elements + j;
851 this->value.f[dst] = value->value.f[src];
852 }
853 }
854
855 /* "All other components will be initialized to the identity matrix." */
856 for (unsigned i = cols; i < type->matrix_columns; i++)
857 this->value.f[i * type->vector_elements + i] = 1.0;
858
859 return;
860 }
861
862 /* Use each component from each entry in the value_list to initialize one
863 * component of the constant being constructed.
864 */
865 unsigned i = 0;
866 for (;;) {
867 assert(value->as_constant() != NULL);
868 assert(!value->is_tail_sentinel());
869
870 for (unsigned j = 0; j < value->type->components(); j++) {
871 switch (type->base_type) {
872 case GLSL_TYPE_UINT:
873 this->value.u[i] = value->get_uint_component(j);
874 break;
875 case GLSL_TYPE_INT:
876 this->value.i[i] = value->get_int_component(j);
877 break;
878 case GLSL_TYPE_FLOAT:
879 this->value.f[i] = value->get_float_component(j);
880 break;
881 case GLSL_TYPE_BOOL:
882 this->value.b[i] = value->get_bool_component(j);
883 break;
884 case GLSL_TYPE_DOUBLE:
885 this->value.d[i] = value->get_double_component(j);
886 break;
887 case GLSL_TYPE_UINT64:
888 this->value.u64[i] = value->get_uint64_component(j);
889 break;
890 case GLSL_TYPE_INT64:
891 this->value.i64[i] = value->get_int64_component(j);
892 break;
893 default:
894 /* FINISHME: What to do? Exceptions are not the answer.
895 */
896 break;
897 }
898
899 i++;
900 if (i >= type->components())
901 break;
902 }
903
904 if (i >= type->components())
905 break; /* avoid downcasting a list sentinel */
906 value = (ir_constant *) value->next;
907 }
908 }
909
910 ir_constant *
911 ir_constant::zero(void *mem_ctx, const glsl_type *type)
912 {
913 assert(type->is_scalar() || type->is_vector() || type->is_matrix()
914 || type->is_struct() || type->is_array());
915
916 ir_constant *c = new(mem_ctx) ir_constant;
917 c->type = type;
918 memset(&c->value, 0, sizeof(c->value));
919
920 if (type->is_array()) {
921 c->const_elements = ralloc_array(c, ir_constant *, type->length);
922
923 for (unsigned i = 0; i < type->length; i++)
924 c->const_elements[i] = ir_constant::zero(c, type->fields.array);
925 }
926
927 if (type->is_struct()) {
928 c->const_elements = ralloc_array(c, ir_constant *, type->length);
929
930 for (unsigned i = 0; i < type->length; i++) {
931 c->const_elements[i] =
932 ir_constant::zero(mem_ctx, type->fields.structure[i].type);
933 }
934 }
935
936 return c;
937 }
938
939 bool
940 ir_constant::get_bool_component(unsigned i) const
941 {
942 switch (this->type->base_type) {
943 case GLSL_TYPE_UINT: return this->value.u[i] != 0;
944 case GLSL_TYPE_INT: return this->value.i[i] != 0;
945 case GLSL_TYPE_FLOAT: return ((int)this->value.f[i]) != 0;
946 case GLSL_TYPE_BOOL: return this->value.b[i];
947 case GLSL_TYPE_DOUBLE: return this->value.d[i] != 0.0;
948 case GLSL_TYPE_SAMPLER:
949 case GLSL_TYPE_IMAGE:
950 case GLSL_TYPE_UINT64: return this->value.u64[i] != 0;
951 case GLSL_TYPE_INT64: return this->value.i64[i] != 0;
952 default: assert(!"Should not get here."); break;
953 }
954
955 /* Must return something to make the compiler happy. This is clearly an
956 * error case.
957 */
958 return false;
959 }
960
961 float
962 ir_constant::get_float_component(unsigned i) const
963 {
964 switch (this->type->base_type) {
965 case GLSL_TYPE_UINT: return (float) this->value.u[i];
966 case GLSL_TYPE_INT: return (float) this->value.i[i];
967 case GLSL_TYPE_FLOAT: return this->value.f[i];
968 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1.0f : 0.0f;
969 case GLSL_TYPE_DOUBLE: return (float) this->value.d[i];
970 case GLSL_TYPE_SAMPLER:
971 case GLSL_TYPE_IMAGE:
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_SAMPLER:
993 case GLSL_TYPE_IMAGE:
994 case GLSL_TYPE_UINT64: return (double) this->value.u64[i];
995 case GLSL_TYPE_INT64: return (double) this->value.i64[i];
996 default: assert(!"Should not get here."); break;
997 }
998
999 /* Must return something to make the compiler happy. This is clearly an
1000 * error case.
1001 */
1002 return 0.0;
1003 }
1004
1005 int
1006 ir_constant::get_int_component(unsigned i) const
1007 {
1008 switch (this->type->base_type) {
1009 case GLSL_TYPE_UINT: return this->value.u[i];
1010 case GLSL_TYPE_INT: return this->value.i[i];
1011 case GLSL_TYPE_FLOAT: return (int) this->value.f[i];
1012 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1013 case GLSL_TYPE_DOUBLE: return (int) this->value.d[i];
1014 case GLSL_TYPE_SAMPLER:
1015 case GLSL_TYPE_IMAGE:
1016 case GLSL_TYPE_UINT64: return (int) this->value.u64[i];
1017 case GLSL_TYPE_INT64: return (int) this->value.i64[i];
1018 default: assert(!"Should not get here."); break;
1019 }
1020
1021 /* Must return something to make the compiler happy. This is clearly an
1022 * error case.
1023 */
1024 return 0;
1025 }
1026
1027 unsigned
1028 ir_constant::get_uint_component(unsigned i) const
1029 {
1030 switch (this->type->base_type) {
1031 case GLSL_TYPE_UINT: return this->value.u[i];
1032 case GLSL_TYPE_INT: return this->value.i[i];
1033 case GLSL_TYPE_FLOAT: return (unsigned) this->value.f[i];
1034 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1035 case GLSL_TYPE_DOUBLE: return (unsigned) this->value.d[i];
1036 case GLSL_TYPE_SAMPLER:
1037 case GLSL_TYPE_IMAGE:
1038 case GLSL_TYPE_UINT64: return (unsigned) this->value.u64[i];
1039 case GLSL_TYPE_INT64: return (unsigned) this->value.i64[i];
1040 default: assert(!"Should not get here."); break;
1041 }
1042
1043 /* Must return something to make the compiler happy. This is clearly an
1044 * error case.
1045 */
1046 return 0;
1047 }
1048
1049 int64_t
1050 ir_constant::get_int64_component(unsigned i) const
1051 {
1052 switch (this->type->base_type) {
1053 case GLSL_TYPE_UINT: return this->value.u[i];
1054 case GLSL_TYPE_INT: return this->value.i[i];
1055 case GLSL_TYPE_FLOAT: return (int64_t) this->value.f[i];
1056 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1057 case GLSL_TYPE_DOUBLE: return (int64_t) this->value.d[i];
1058 case GLSL_TYPE_SAMPLER:
1059 case GLSL_TYPE_IMAGE:
1060 case GLSL_TYPE_UINT64: return (int64_t) this->value.u64[i];
1061 case GLSL_TYPE_INT64: return this->value.i64[i];
1062 default: assert(!"Should not get here."); break;
1063 }
1064
1065 /* Must return something to make the compiler happy. This is clearly an
1066 * error case.
1067 */
1068 return 0;
1069 }
1070
1071 uint64_t
1072 ir_constant::get_uint64_component(unsigned i) const
1073 {
1074 switch (this->type->base_type) {
1075 case GLSL_TYPE_UINT: return this->value.u[i];
1076 case GLSL_TYPE_INT: return this->value.i[i];
1077 case GLSL_TYPE_FLOAT: return (uint64_t) this->value.f[i];
1078 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1079 case GLSL_TYPE_DOUBLE: return (uint64_t) this->value.d[i];
1080 case GLSL_TYPE_SAMPLER:
1081 case GLSL_TYPE_IMAGE:
1082 case GLSL_TYPE_UINT64: return this->value.u64[i];
1083 case GLSL_TYPE_INT64: return (uint64_t) this->value.i64[i];
1084 default: assert(!"Should not get here."); break;
1085 }
1086
1087 /* Must return something to make the compiler happy. This is clearly an
1088 * error case.
1089 */
1090 return 0;
1091 }
1092
1093 ir_constant *
1094 ir_constant::get_array_element(unsigned i) const
1095 {
1096 assert(this->type->is_array());
1097
1098 /* From page 35 (page 41 of the PDF) of the GLSL 1.20 spec:
1099 *
1100 * "Behavior is undefined if a shader subscripts an array with an index
1101 * less than 0 or greater than or equal to the size the array was
1102 * declared with."
1103 *
1104 * Most out-of-bounds accesses are removed before things could get this far.
1105 * There are cases where non-constant array index values can get constant
1106 * folded.
1107 */
1108 if (int(i) < 0)
1109 i = 0;
1110 else if (i >= this->type->length)
1111 i = this->type->length - 1;
1112
1113 return const_elements[i];
1114 }
1115
1116 ir_constant *
1117 ir_constant::get_record_field(int idx)
1118 {
1119 assert(this->type->is_struct());
1120 assert(idx >= 0 && (unsigned) idx < this->type->length);
1121
1122 return const_elements[idx];
1123 }
1124
1125 void
1126 ir_constant::copy_offset(ir_constant *src, int offset)
1127 {
1128 switch (this->type->base_type) {
1129 case GLSL_TYPE_UINT:
1130 case GLSL_TYPE_INT:
1131 case GLSL_TYPE_FLOAT:
1132 case GLSL_TYPE_DOUBLE:
1133 case GLSL_TYPE_SAMPLER:
1134 case GLSL_TYPE_IMAGE:
1135 case GLSL_TYPE_UINT64:
1136 case GLSL_TYPE_INT64:
1137 case GLSL_TYPE_BOOL: {
1138 unsigned int size = src->type->components();
1139 assert (size <= this->type->components() - offset);
1140 for (unsigned int i=0; i<size; i++) {
1141 switch (this->type->base_type) {
1142 case GLSL_TYPE_UINT:
1143 value.u[i+offset] = src->get_uint_component(i);
1144 break;
1145 case GLSL_TYPE_INT:
1146 value.i[i+offset] = src->get_int_component(i);
1147 break;
1148 case GLSL_TYPE_FLOAT:
1149 value.f[i+offset] = src->get_float_component(i);
1150 break;
1151 case GLSL_TYPE_BOOL:
1152 value.b[i+offset] = src->get_bool_component(i);
1153 break;
1154 case GLSL_TYPE_DOUBLE:
1155 value.d[i+offset] = src->get_double_component(i);
1156 break;
1157 case GLSL_TYPE_SAMPLER:
1158 case GLSL_TYPE_IMAGE:
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 case GLSL_TYPE_ARRAY: {
1174 assert (src->type == this->type);
1175 for (unsigned i = 0; i < this->type->length; i++) {
1176 this->const_elements[i] = src->const_elements[i]->clone(this, NULL);
1177 }
1178 break;
1179 }
1180
1181 default:
1182 assert(!"Should not get here.");
1183 break;
1184 }
1185 }
1186
1187 void
1188 ir_constant::copy_masked_offset(ir_constant *src, int offset, unsigned int mask)
1189 {
1190 assert (!type->is_array() && !type->is_struct());
1191
1192 if (!type->is_vector() && !type->is_matrix()) {
1193 offset = 0;
1194 mask = 1;
1195 }
1196
1197 int id = 0;
1198 for (int i=0; i<4; i++) {
1199 if (mask & (1 << i)) {
1200 switch (this->type->base_type) {
1201 case GLSL_TYPE_UINT:
1202 value.u[i+offset] = src->get_uint_component(id++);
1203 break;
1204 case GLSL_TYPE_INT:
1205 value.i[i+offset] = src->get_int_component(id++);
1206 break;
1207 case GLSL_TYPE_FLOAT:
1208 value.f[i+offset] = src->get_float_component(id++);
1209 break;
1210 case GLSL_TYPE_BOOL:
1211 value.b[i+offset] = src->get_bool_component(id++);
1212 break;
1213 case GLSL_TYPE_DOUBLE:
1214 value.d[i+offset] = src->get_double_component(id++);
1215 break;
1216 case GLSL_TYPE_SAMPLER:
1217 case GLSL_TYPE_IMAGE:
1218 case GLSL_TYPE_UINT64:
1219 value.u64[i+offset] = src->get_uint64_component(id++);
1220 break;
1221 case GLSL_TYPE_INT64:
1222 value.i64[i+offset] = src->get_int64_component(id++);
1223 break;
1224 default:
1225 assert(!"Should not get here.");
1226 return;
1227 }
1228 }
1229 }
1230 }
1231
1232 bool
1233 ir_constant::has_value(const ir_constant *c) const
1234 {
1235 if (this->type != c->type)
1236 return false;
1237
1238 if (this->type->is_array() || this->type->is_struct()) {
1239 for (unsigned i = 0; i < this->type->length; i++) {
1240 if (!this->const_elements[i]->has_value(c->const_elements[i]))
1241 return false;
1242 }
1243 return true;
1244 }
1245
1246 for (unsigned i = 0; i < this->type->components(); i++) {
1247 switch (this->type->base_type) {
1248 case GLSL_TYPE_UINT:
1249 if (this->value.u[i] != c->value.u[i])
1250 return false;
1251 break;
1252 case GLSL_TYPE_INT:
1253 if (this->value.i[i] != c->value.i[i])
1254 return false;
1255 break;
1256 case GLSL_TYPE_FLOAT:
1257 if (this->value.f[i] != c->value.f[i])
1258 return false;
1259 break;
1260 case GLSL_TYPE_BOOL:
1261 if (this->value.b[i] != c->value.b[i])
1262 return false;
1263 break;
1264 case GLSL_TYPE_DOUBLE:
1265 if (this->value.d[i] != c->value.d[i])
1266 return false;
1267 break;
1268 case GLSL_TYPE_SAMPLER:
1269 case GLSL_TYPE_IMAGE:
1270 case GLSL_TYPE_UINT64:
1271 if (this->value.u64[i] != c->value.u64[i])
1272 return false;
1273 break;
1274 case GLSL_TYPE_INT64:
1275 if (this->value.i64[i] != c->value.i64[i])
1276 return false;
1277 break;
1278 default:
1279 assert(!"Should not get here.");
1280 return false;
1281 }
1282 }
1283
1284 return true;
1285 }
1286
1287 bool
1288 ir_constant::is_value(float f, int i) const
1289 {
1290 if (!this->type->is_scalar() && !this->type->is_vector())
1291 return false;
1292
1293 /* Only accept boolean values for 0/1. */
1294 if (int(bool(i)) != i && this->type->is_boolean())
1295 return false;
1296
1297 for (unsigned c = 0; c < this->type->vector_elements; c++) {
1298 switch (this->type->base_type) {
1299 case GLSL_TYPE_FLOAT:
1300 if (this->value.f[c] != f)
1301 return false;
1302 break;
1303 case GLSL_TYPE_INT:
1304 if (this->value.i[c] != i)
1305 return false;
1306 break;
1307 case GLSL_TYPE_UINT:
1308 if (this->value.u[c] != unsigned(i))
1309 return false;
1310 break;
1311 case GLSL_TYPE_BOOL:
1312 if (this->value.b[c] != bool(i))
1313 return false;
1314 break;
1315 case GLSL_TYPE_DOUBLE:
1316 if (this->value.d[c] != double(f))
1317 return false;
1318 break;
1319 case GLSL_TYPE_SAMPLER:
1320 case GLSL_TYPE_IMAGE:
1321 case GLSL_TYPE_UINT64:
1322 if (this->value.u64[c] != uint64_t(i))
1323 return false;
1324 break;
1325 case GLSL_TYPE_INT64:
1326 if (this->value.i64[c] != i)
1327 return false;
1328 break;
1329 default:
1330 /* The only other base types are structures, arrays, and samplers.
1331 * Samplers cannot be constants, and the others should have been
1332 * filtered out above.
1333 */
1334 assert(!"Should not get here.");
1335 return false;
1336 }
1337 }
1338
1339 return true;
1340 }
1341
1342 bool
1343 ir_constant::is_zero() const
1344 {
1345 return is_value(0.0, 0);
1346 }
1347
1348 bool
1349 ir_constant::is_one() const
1350 {
1351 return is_value(1.0, 1);
1352 }
1353
1354 bool
1355 ir_constant::is_negative_one() const
1356 {
1357 return is_value(-1.0, -1);
1358 }
1359
1360 bool
1361 ir_constant::is_uint16_constant() const
1362 {
1363 if (!type->is_integer_32())
1364 return false;
1365
1366 return value.u[0] < (1 << 16);
1367 }
1368
1369 ir_loop::ir_loop()
1370 : ir_instruction(ir_type_loop)
1371 {
1372 }
1373
1374
1375 ir_dereference_variable::ir_dereference_variable(ir_variable *var)
1376 : ir_dereference(ir_type_dereference_variable)
1377 {
1378 assert(var != NULL);
1379
1380 this->var = var;
1381 this->type = var->type;
1382 }
1383
1384
1385 ir_dereference_array::ir_dereference_array(ir_rvalue *value,
1386 ir_rvalue *array_index)
1387 : ir_dereference(ir_type_dereference_array)
1388 {
1389 this->array_index = array_index;
1390 this->set_array(value);
1391 }
1392
1393
1394 ir_dereference_array::ir_dereference_array(ir_variable *var,
1395 ir_rvalue *array_index)
1396 : ir_dereference(ir_type_dereference_array)
1397 {
1398 void *ctx = ralloc_parent(var);
1399
1400 this->array_index = array_index;
1401 this->set_array(new(ctx) ir_dereference_variable(var));
1402 }
1403
1404
1405 void
1406 ir_dereference_array::set_array(ir_rvalue *value)
1407 {
1408 assert(value != NULL);
1409
1410 this->array = value;
1411
1412 const glsl_type *const vt = this->array->type;
1413
1414 if (vt->is_array()) {
1415 type = vt->fields.array;
1416 } else if (vt->is_matrix()) {
1417 type = vt->column_type();
1418 } else if (vt->is_vector()) {
1419 type = vt->get_base_type();
1420 }
1421 }
1422
1423
1424 ir_dereference_record::ir_dereference_record(ir_rvalue *value,
1425 const char *field)
1426 : ir_dereference(ir_type_dereference_record)
1427 {
1428 assert(value != NULL);
1429
1430 this->record = value;
1431 this->type = this->record->type->field_type(field);
1432 this->field_idx = this->record->type->field_index(field);
1433 }
1434
1435
1436 ir_dereference_record::ir_dereference_record(ir_variable *var,
1437 const char *field)
1438 : ir_dereference(ir_type_dereference_record)
1439 {
1440 void *ctx = ralloc_parent(var);
1441
1442 this->record = new(ctx) ir_dereference_variable(var);
1443 this->type = this->record->type->field_type(field);
1444 this->field_idx = this->record->type->field_index(field);
1445 }
1446
1447 bool
1448 ir_dereference::is_lvalue(const struct _mesa_glsl_parse_state *state) const
1449 {
1450 ir_variable *var = this->variable_referenced();
1451
1452 /* Every l-value derference chain eventually ends in a variable.
1453 */
1454 if ((var == NULL) || var->data.read_only)
1455 return false;
1456
1457 /* From section 4.1.7 of the ARB_bindless_texture spec:
1458 *
1459 * "Samplers can be used as l-values, so can be assigned into and used as
1460 * "out" and "inout" function parameters."
1461 *
1462 * From section 4.1.X of the ARB_bindless_texture spec:
1463 *
1464 * "Images can be used as l-values, so can be assigned into and used as
1465 * "out" and "inout" function parameters."
1466 */
1467 if ((!state || state->has_bindless()) &&
1468 (this->type->contains_sampler() || this->type->contains_image()))
1469 return true;
1470
1471 /* From section 4.1.7 of the GLSL 4.40 spec:
1472 *
1473 * "Opaque variables cannot be treated as l-values; hence cannot
1474 * be used as out or inout function parameters, nor can they be
1475 * assigned into."
1476 */
1477 if (this->type->contains_opaque())
1478 return false;
1479
1480 return true;
1481 }
1482
1483
1484 static const char * const tex_opcode_strs[] = { "tex", "txb", "txl", "txd", "txf", "txf_ms", "txs", "lod", "tg4", "query_levels", "texture_samples", "samples_identical" };
1485
1486 const char *ir_texture::opcode_string()
1487 {
1488 assert((unsigned int) op < ARRAY_SIZE(tex_opcode_strs));
1489 return tex_opcode_strs[op];
1490 }
1491
1492 ir_texture_opcode
1493 ir_texture::get_opcode(const char *str)
1494 {
1495 const int count = sizeof(tex_opcode_strs) / sizeof(tex_opcode_strs[0]);
1496 for (int op = 0; op < count; op++) {
1497 if (strcmp(str, tex_opcode_strs[op]) == 0)
1498 return (ir_texture_opcode) op;
1499 }
1500 return (ir_texture_opcode) -1;
1501 }
1502
1503
1504 void
1505 ir_texture::set_sampler(ir_dereference *sampler, const glsl_type *type)
1506 {
1507 assert(sampler != NULL);
1508 assert(type != NULL);
1509 this->sampler = sampler;
1510 this->type = type;
1511
1512 if (this->op == ir_txs || this->op == ir_query_levels ||
1513 this->op == ir_texture_samples) {
1514 assert(type->base_type == GLSL_TYPE_INT);
1515 } else if (this->op == ir_lod) {
1516 assert(type->vector_elements == 2);
1517 assert(type->is_float());
1518 } else if (this->op == ir_samples_identical) {
1519 assert(type == glsl_type::bool_type);
1520 assert(sampler->type->is_sampler());
1521 assert(sampler->type->sampler_dimensionality == GLSL_SAMPLER_DIM_MS);
1522 } else {
1523 assert(sampler->type->sampled_type == (int) type->base_type);
1524 if (sampler->type->sampler_shadow)
1525 assert(type->vector_elements == 4 || type->vector_elements == 1);
1526 else
1527 assert(type->vector_elements == 4);
1528 }
1529 }
1530
1531
1532 void
1533 ir_swizzle::init_mask(const unsigned *comp, unsigned count)
1534 {
1535 assert((count >= 1) && (count <= 4));
1536
1537 memset(&this->mask, 0, sizeof(this->mask));
1538 this->mask.num_components = count;
1539
1540 unsigned dup_mask = 0;
1541 switch (count) {
1542 case 4:
1543 assert(comp[3] <= 3);
1544 dup_mask |= (1U << comp[3])
1545 & ((1U << comp[0]) | (1U << comp[1]) | (1U << comp[2]));
1546 this->mask.w = comp[3];
1547
1548 case 3:
1549 assert(comp[2] <= 3);
1550 dup_mask |= (1U << comp[2])
1551 & ((1U << comp[0]) | (1U << comp[1]));
1552 this->mask.z = comp[2];
1553
1554 case 2:
1555 assert(comp[1] <= 3);
1556 dup_mask |= (1U << comp[1])
1557 & ((1U << comp[0]));
1558 this->mask.y = comp[1];
1559
1560 case 1:
1561 assert(comp[0] <= 3);
1562 this->mask.x = comp[0];
1563 }
1564
1565 this->mask.has_duplicates = dup_mask != 0;
1566
1567 /* Based on the number of elements in the swizzle and the base type
1568 * (i.e., float, int, unsigned, or bool) of the vector being swizzled,
1569 * generate the type of the resulting value.
1570 */
1571 type = glsl_type::get_instance(val->type->base_type, mask.num_components, 1);
1572 }
1573
1574 ir_swizzle::ir_swizzle(ir_rvalue *val, unsigned x, unsigned y, unsigned z,
1575 unsigned w, unsigned count)
1576 : ir_rvalue(ir_type_swizzle), val(val)
1577 {
1578 const unsigned components[4] = { x, y, z, w };
1579 this->init_mask(components, count);
1580 }
1581
1582 ir_swizzle::ir_swizzle(ir_rvalue *val, const unsigned *comp,
1583 unsigned count)
1584 : ir_rvalue(ir_type_swizzle), val(val)
1585 {
1586 this->init_mask(comp, count);
1587 }
1588
1589 ir_swizzle::ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask)
1590 : ir_rvalue(ir_type_swizzle), val(val), mask(mask)
1591 {
1592 this->type = glsl_type::get_instance(val->type->base_type,
1593 mask.num_components, 1);
1594 }
1595
1596 #define X 1
1597 #define R 5
1598 #define S 9
1599 #define I 13
1600
1601 ir_swizzle *
1602 ir_swizzle::create(ir_rvalue *val, const char *str, unsigned vector_length)
1603 {
1604 void *ctx = ralloc_parent(val);
1605
1606 /* For each possible swizzle character, this table encodes the value in
1607 * \c idx_map that represents the 0th element of the vector. For invalid
1608 * swizzle characters (e.g., 'k'), a special value is used that will allow
1609 * detection of errors.
1610 */
1611 static const unsigned char base_idx[26] = {
1612 /* a b c d e f g h i j k l m */
1613 R, R, I, I, I, I, R, I, I, I, I, I, I,
1614 /* n o p q r s t u v w x y z */
1615 I, I, S, S, R, S, S, I, I, X, X, X, X
1616 };
1617
1618 /* Each valid swizzle character has an entry in the previous table. This
1619 * table encodes the base index encoded in the previous table plus the actual
1620 * index of the swizzle character. When processing swizzles, the first
1621 * character in the string is indexed in the previous table. Each character
1622 * in the string is indexed in this table, and the value found there has the
1623 * value form the first table subtracted. The result must be on the range
1624 * [0,3].
1625 *
1626 * For example, the string "wzyx" will get X from the first table. Each of
1627 * the charcaters will get X+3, X+2, X+1, and X+0 from this table. After
1628 * subtraction, the swizzle values are { 3, 2, 1, 0 }.
1629 *
1630 * The string "wzrg" will get X from the first table. Each of the characters
1631 * will get X+3, X+2, R+0, and R+1 from this table. After subtraction, the
1632 * swizzle values are { 3, 2, 4, 5 }. Since 4 and 5 are outside the range
1633 * [0,3], the error is detected.
1634 */
1635 static const unsigned char idx_map[26] = {
1636 /* a b c d e f g h i j k l m */
1637 R+3, R+2, 0, 0, 0, 0, R+1, 0, 0, 0, 0, 0, 0,
1638 /* n o p q r s t u v w x y z */
1639 0, 0, S+2, S+3, R+0, S+0, S+1, 0, 0, X+3, X+0, X+1, X+2
1640 };
1641
1642 int swiz_idx[4] = { 0, 0, 0, 0 };
1643 unsigned i;
1644
1645
1646 /* Validate the first character in the swizzle string and look up the base
1647 * index value as described above.
1648 */
1649 if ((str[0] < 'a') || (str[0] > 'z'))
1650 return NULL;
1651
1652 const unsigned base = base_idx[str[0] - 'a'];
1653
1654
1655 for (i = 0; (i < 4) && (str[i] != '\0'); i++) {
1656 /* Validate the next character, and, as described above, convert it to a
1657 * swizzle index.
1658 */
1659 if ((str[i] < 'a') || (str[i] > 'z'))
1660 return NULL;
1661
1662 swiz_idx[i] = idx_map[str[i] - 'a'] - base;
1663 if ((swiz_idx[i] < 0) || (swiz_idx[i] >= (int) vector_length))
1664 return NULL;
1665 }
1666
1667 if (str[i] != '\0')
1668 return NULL;
1669
1670 return new(ctx) ir_swizzle(val, swiz_idx[0], swiz_idx[1], swiz_idx[2],
1671 swiz_idx[3], i);
1672 }
1673
1674 #undef X
1675 #undef R
1676 #undef S
1677 #undef I
1678
1679 ir_variable *
1680 ir_swizzle::variable_referenced() const
1681 {
1682 return this->val->variable_referenced();
1683 }
1684
1685
1686 bool ir_variable::temporaries_allocate_names = false;
1687
1688 const char ir_variable::tmp_name[] = "compiler_temp";
1689
1690 ir_variable::ir_variable(const struct glsl_type *type, const char *name,
1691 ir_variable_mode mode)
1692 : ir_instruction(ir_type_variable)
1693 {
1694 this->type = type;
1695
1696 if (mode == ir_var_temporary && !ir_variable::temporaries_allocate_names)
1697 name = NULL;
1698
1699 /* The ir_variable clone method may call this constructor with name set to
1700 * tmp_name.
1701 */
1702 assert(name != NULL
1703 || mode == ir_var_temporary
1704 || mode == ir_var_function_in
1705 || mode == ir_var_function_out
1706 || mode == ir_var_function_inout);
1707 assert(name != ir_variable::tmp_name
1708 || mode == ir_var_temporary);
1709 if (mode == ir_var_temporary
1710 && (name == NULL || name == ir_variable::tmp_name)) {
1711 this->name = ir_variable::tmp_name;
1712 } else if (name == NULL ||
1713 strlen(name) < ARRAY_SIZE(this->name_storage)) {
1714 strcpy(this->name_storage, name ? name : "");
1715 this->name = this->name_storage;
1716 } else {
1717 this->name = ralloc_strdup(this, name);
1718 }
1719
1720 this->u.max_ifc_array_access = NULL;
1721
1722 this->data.explicit_location = false;
1723 this->data.has_initializer = false;
1724 this->data.location = -1;
1725 this->data.location_frac = 0;
1726 this->data.binding = 0;
1727 this->data.warn_extension_index = 0;
1728 this->constant_value = NULL;
1729 this->constant_initializer = NULL;
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.explicit_invariant = false;
1738 this->data.invariant = false;
1739 this->data.how_declared = ir_var_declared_normally;
1740 this->data.mode = mode;
1741 this->data.interpolation = INTERP_MODE_NONE;
1742 this->data.max_array_access = -1;
1743 this->data.offset = 0;
1744 this->data.precision = GLSL_PRECISION_NONE;
1745 this->data.memory_read_only = false;
1746 this->data.memory_write_only = false;
1747 this->data.memory_coherent = false;
1748 this->data.memory_volatile = false;
1749 this->data.memory_restrict = false;
1750 this->data.from_ssbo_unsized_array = false;
1751 this->data.fb_fetch_output = false;
1752 this->data.bindless = false;
1753 this->data.bound = false;
1754
1755 if (type != NULL) {
1756 if (type->is_interface())
1757 this->init_interface_type(type);
1758 else if (type->without_array()->is_interface())
1759 this->init_interface_type(type->without_array());
1760 }
1761 }
1762
1763
1764 const char *
1765 interpolation_string(unsigned interpolation)
1766 {
1767 switch (interpolation) {
1768 case INTERP_MODE_NONE: return "no";
1769 case INTERP_MODE_SMOOTH: return "smooth";
1770 case INTERP_MODE_FLAT: return "flat";
1771 case INTERP_MODE_NOPERSPECTIVE: return "noperspective";
1772 }
1773
1774 assert(!"Should not get here.");
1775 return "";
1776 }
1777
1778 const char *const ir_variable::warn_extension_table[] = {
1779 "",
1780 "GL_ARB_shader_stencil_export",
1781 "GL_AMD_shader_stencil_export",
1782 };
1783
1784 void
1785 ir_variable::enable_extension_warning(const char *extension)
1786 {
1787 for (unsigned i = 0; i < ARRAY_SIZE(warn_extension_table); i++) {
1788 if (strcmp(warn_extension_table[i], extension) == 0) {
1789 this->data.warn_extension_index = i;
1790 return;
1791 }
1792 }
1793
1794 assert(!"Should not get here.");
1795 this->data.warn_extension_index = 0;
1796 }
1797
1798 const char *
1799 ir_variable::get_extension_warning() const
1800 {
1801 return this->data.warn_extension_index == 0
1802 ? NULL : warn_extension_table[this->data.warn_extension_index];
1803 }
1804
1805 ir_function_signature::ir_function_signature(const glsl_type *return_type,
1806 builtin_available_predicate b)
1807 : ir_instruction(ir_type_function_signature),
1808 return_type(return_type), is_defined(false),
1809 return_precision(GLSL_PRECISION_NONE),
1810 intrinsic_id(ir_intrinsic_invalid), builtin_avail(b), _function(NULL)
1811 {
1812 this->origin = NULL;
1813 }
1814
1815
1816 bool
1817 ir_function_signature::is_builtin() const
1818 {
1819 return builtin_avail != NULL;
1820 }
1821
1822
1823 bool
1824 ir_function_signature::is_builtin_available(const _mesa_glsl_parse_state *state) const
1825 {
1826 /* We can't call the predicate without a state pointer, so just say that
1827 * the signature is available. At compile time, we need the filtering,
1828 * but also receive a valid state pointer. At link time, we're resolving
1829 * imported built-in prototypes to their definitions, which will always
1830 * be an exact match. So we can skip the filtering.
1831 */
1832 if (state == NULL)
1833 return true;
1834
1835 assert(builtin_avail != NULL);
1836 return builtin_avail(state);
1837 }
1838
1839
1840 static bool
1841 modes_match(unsigned a, unsigned b)
1842 {
1843 if (a == b)
1844 return true;
1845
1846 /* Accept "in" vs. "const in" */
1847 if ((a == ir_var_const_in && b == ir_var_function_in) ||
1848 (b == ir_var_const_in && a == ir_var_function_in))
1849 return true;
1850
1851 return false;
1852 }
1853
1854
1855 const char *
1856 ir_function_signature::qualifiers_match(exec_list *params)
1857 {
1858 /* check that the qualifiers match. */
1859 foreach_two_lists(a_node, &this->parameters, b_node, params) {
1860 ir_variable *a = (ir_variable *) a_node;
1861 ir_variable *b = (ir_variable *) b_node;
1862
1863 if (a->data.read_only != b->data.read_only ||
1864 !modes_match(a->data.mode, b->data.mode) ||
1865 a->data.interpolation != b->data.interpolation ||
1866 a->data.centroid != b->data.centroid ||
1867 a->data.sample != b->data.sample ||
1868 a->data.patch != b->data.patch ||
1869 a->data.memory_read_only != b->data.memory_read_only ||
1870 a->data.memory_write_only != b->data.memory_write_only ||
1871 a->data.memory_coherent != b->data.memory_coherent ||
1872 a->data.memory_volatile != b->data.memory_volatile ||
1873 a->data.memory_restrict != b->data.memory_restrict) {
1874
1875 /* parameter a's qualifiers don't match */
1876 return a->name;
1877 }
1878 }
1879 return NULL;
1880 }
1881
1882
1883 void
1884 ir_function_signature::replace_parameters(exec_list *new_params)
1885 {
1886 /* Destroy all of the previous parameter information. If the previous
1887 * parameter information comes from the function prototype, it may either
1888 * specify incorrect parameter names or not have names at all.
1889 */
1890 new_params->move_nodes_to(&parameters);
1891 }
1892
1893
1894 ir_function::ir_function(const char *name)
1895 : ir_instruction(ir_type_function)
1896 {
1897 this->subroutine_index = -1;
1898 this->name = ralloc_strdup(this, name);
1899 }
1900
1901
1902 bool
1903 ir_function::has_user_signature()
1904 {
1905 foreach_in_list(ir_function_signature, sig, &this->signatures) {
1906 if (!sig->is_builtin())
1907 return true;
1908 }
1909 return false;
1910 }
1911
1912
1913 ir_rvalue *
1914 ir_rvalue::error_value(void *mem_ctx)
1915 {
1916 ir_rvalue *v = new(mem_ctx) ir_rvalue(ir_type_unset);
1917
1918 v->type = glsl_type::error_type;
1919 return v;
1920 }
1921
1922
1923 void
1924 visit_exec_list(exec_list *list, ir_visitor *visitor)
1925 {
1926 foreach_in_list_safe(ir_instruction, node, list) {
1927 node->accept(visitor);
1928 }
1929 }
1930
1931
1932 static void
1933 steal_memory(ir_instruction *ir, void *new_ctx)
1934 {
1935 ir_variable *var = ir->as_variable();
1936 ir_function *fn = ir->as_function();
1937 ir_constant *constant = ir->as_constant();
1938 if (var != NULL && var->constant_value != NULL)
1939 steal_memory(var->constant_value, ir);
1940
1941 if (var != NULL && var->constant_initializer != NULL)
1942 steal_memory(var->constant_initializer, ir);
1943
1944 if (fn != NULL && fn->subroutine_types)
1945 ralloc_steal(new_ctx, fn->subroutine_types);
1946
1947 /* The components of aggregate constants are not visited by the normal
1948 * visitor, so steal their values by hand.
1949 */
1950 if (constant != NULL &&
1951 (constant->type->is_array() || constant->type->is_struct())) {
1952 for (unsigned int i = 0; i < constant->type->length; i++) {
1953 steal_memory(constant->const_elements[i], ir);
1954 }
1955 }
1956
1957 ralloc_steal(new_ctx, ir);
1958 }
1959
1960
1961 void
1962 reparent_ir(exec_list *list, void *mem_ctx)
1963 {
1964 foreach_in_list(ir_instruction, node, list) {
1965 visit_tree(node, steal_memory, mem_ctx);
1966 }
1967 }
1968
1969
1970 static ir_rvalue *
1971 try_min_one(ir_rvalue *ir)
1972 {
1973 ir_expression *expr = ir->as_expression();
1974
1975 if (!expr || expr->operation != ir_binop_min)
1976 return NULL;
1977
1978 if (expr->operands[0]->is_one())
1979 return expr->operands[1];
1980
1981 if (expr->operands[1]->is_one())
1982 return expr->operands[0];
1983
1984 return NULL;
1985 }
1986
1987 static ir_rvalue *
1988 try_max_zero(ir_rvalue *ir)
1989 {
1990 ir_expression *expr = ir->as_expression();
1991
1992 if (!expr || expr->operation != ir_binop_max)
1993 return NULL;
1994
1995 if (expr->operands[0]->is_zero())
1996 return expr->operands[1];
1997
1998 if (expr->operands[1]->is_zero())
1999 return expr->operands[0];
2000
2001 return NULL;
2002 }
2003
2004 ir_rvalue *
2005 ir_rvalue::as_rvalue_to_saturate()
2006 {
2007 ir_expression *expr = this->as_expression();
2008
2009 if (!expr)
2010 return NULL;
2011
2012 ir_rvalue *max_zero = try_max_zero(expr);
2013 if (max_zero) {
2014 return try_min_one(max_zero);
2015 } else {
2016 ir_rvalue *min_one = try_min_one(expr);
2017 if (min_one) {
2018 return try_max_zero(min_one);
2019 }
2020 }
2021
2022 return NULL;
2023 }
2024
2025
2026 unsigned
2027 vertices_per_prim(GLenum prim)
2028 {
2029 switch (prim) {
2030 case GL_POINTS:
2031 return 1;
2032 case GL_LINES:
2033 return 2;
2034 case GL_TRIANGLES:
2035 return 3;
2036 case GL_LINES_ADJACENCY:
2037 return 4;
2038 case GL_TRIANGLES_ADJACENCY:
2039 return 6;
2040 default:
2041 assert(!"Bad primitive");
2042 return 3;
2043 }
2044 }
2045
2046 /**
2047 * Generate a string describing the mode of a variable
2048 */
2049 const char *
2050 mode_string(const ir_variable *var)
2051 {
2052 switch (var->data.mode) {
2053 case ir_var_auto:
2054 return (var->data.read_only) ? "global constant" : "global variable";
2055
2056 case ir_var_uniform:
2057 return "uniform";
2058
2059 case ir_var_shader_storage:
2060 return "buffer";
2061
2062 case ir_var_shader_in:
2063 return "shader input";
2064
2065 case ir_var_shader_out:
2066 return "shader output";
2067
2068 case ir_var_function_in:
2069 case ir_var_const_in:
2070 return "function input";
2071
2072 case ir_var_function_out:
2073 return "function output";
2074
2075 case ir_var_function_inout:
2076 return "function inout";
2077
2078 case ir_var_system_value:
2079 return "shader input";
2080
2081 case ir_var_temporary:
2082 return "compiler temporary";
2083
2084 case ir_var_mode_count:
2085 break;
2086 }
2087
2088 assert(!"Should not get here.");
2089 return "invalid variable";
2090 }