projects / mesa.git / blob
? search:


123de99005c442b7f69840e79f0e876968b86c90
[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 #ifndef NDEBUG
207 int num_operands = get_num_operands(this->operation);
208 for (int i = num_operands; i < 4; i++) {
209 assert(this->operands[i] == NULL);
210 }
211 #endif
212 }
213
214 ir_expression::ir_expression(int op, ir_rvalue *op0)
215 : ir_rvalue(ir_type_expression)
216 {
217 this->operation = ir_expression_operation(op);
218 this->operands[0] = op0;
219 this->operands[1] = NULL;
220 this->operands[2] = NULL;
221 this->operands[3] = NULL;
222
223 assert(op <= ir_last_unop);
224
225 switch (this->operation) {
226 case ir_unop_bit_not:
227 case ir_unop_logic_not:
228 case ir_unop_neg:
229 case ir_unop_abs:
230 case ir_unop_sign:
231 case ir_unop_rcp:
232 case ir_unop_rsq:
233 case ir_unop_sqrt:
234 case ir_unop_exp:
235 case ir_unop_log:
236 case ir_unop_exp2:
237 case ir_unop_log2:
238 case ir_unop_trunc:
239 case ir_unop_ceil:
240 case ir_unop_floor:
241 case ir_unop_fract:
242 case ir_unop_round_even:
243 case ir_unop_sin:
244 case ir_unop_cos:
245 case ir_unop_dFdx:
246 case ir_unop_dFdx_coarse:
247 case ir_unop_dFdx_fine:
248 case ir_unop_dFdy:
249 case ir_unop_dFdy_coarse:
250 case ir_unop_dFdy_fine:
251 case ir_unop_bitfield_reverse:
252 case ir_unop_interpolate_at_centroid:
253 case ir_unop_saturate:
254 this->type = op0->type;
255 break;
256
257 case ir_unop_f2i:
258 case ir_unop_b2i:
259 case ir_unop_u2i:
260 case ir_unop_d2i:
261 case ir_unop_bitcast_f2i:
262 case ir_unop_bit_count:
263 case ir_unop_find_msb:
264 case ir_unop_find_lsb:
265 case ir_unop_subroutine_to_int:
266 case ir_unop_i642i:
267 case ir_unop_u642i:
268 this->type = glsl_type::get_instance(GLSL_TYPE_INT,
269 op0->type->vector_elements, 1);
270 break;
271
272 case ir_unop_b2f:
273 case ir_unop_i2f:
274 case ir_unop_u2f:
275 case ir_unop_d2f:
276 case ir_unop_bitcast_i2f:
277 case ir_unop_bitcast_u2f:
278 case ir_unop_i642f:
279 case ir_unop_u642f:
280 this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT,
281 op0->type->vector_elements, 1);
282 break;
283
284 case ir_unop_f2b:
285 case ir_unop_i2b:
286 case ir_unop_d2b:
287 case ir_unop_i642b:
288 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL,
289 op0->type->vector_elements, 1);
290 break;
291
292 case ir_unop_f2d:
293 case ir_unop_i2d:
294 case ir_unop_u2d:
295 case ir_unop_i642d:
296 case ir_unop_u642d:
297 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE,
298 op0->type->vector_elements, 1);
299 break;
300
301 case ir_unop_i2u:
302 case ir_unop_f2u:
303 case ir_unop_d2u:
304 case ir_unop_bitcast_f2u:
305 case ir_unop_i642u:
306 case ir_unop_u642u:
307 this->type = glsl_type::get_instance(GLSL_TYPE_UINT,
308 op0->type->vector_elements, 1);
309 break;
310
311 case ir_unop_i2i64:
312 case ir_unop_u2i64:
313 case ir_unop_b2i64:
314 case ir_unop_f2i64:
315 case ir_unop_d2i64:
316 case ir_unop_u642i64:
317 this->type = glsl_type::get_instance(GLSL_TYPE_INT64,
318 op0->type->vector_elements, 1);
319 break;
320
321 case ir_unop_i2u64:
322 case ir_unop_u2u64:
323 case ir_unop_f2u64:
324 case ir_unop_d2u64:
325 case ir_unop_i642u64:
326 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64,
327 op0->type->vector_elements, 1);
328 break;
329 case ir_unop_noise:
330 this->type = glsl_type::float_type;
331 break;
332
333 case ir_unop_unpack_double_2x32:
334 case ir_unop_unpack_uint_2x32:
335 this->type = glsl_type::uvec2_type;
336 break;
337
338 case ir_unop_unpack_int_2x32:
339 this->type = glsl_type::ivec2_type;
340 break;
341
342 case ir_unop_pack_snorm_2x16:
343 case ir_unop_pack_snorm_4x8:
344 case ir_unop_pack_unorm_2x16:
345 case ir_unop_pack_unorm_4x8:
346 case ir_unop_pack_half_2x16:
347 this->type = glsl_type::uint_type;
348 break;
349
350 case ir_unop_pack_double_2x32:
351 this->type = glsl_type::double_type;
352 break;
353
354 case ir_unop_pack_int_2x32:
355 this->type = glsl_type::int64_t_type;
356 break;
357
358 case ir_unop_pack_uint_2x32:
359 this->type = glsl_type::uint64_t_type;
360 break;
361
362 case ir_unop_unpack_snorm_2x16:
363 case ir_unop_unpack_unorm_2x16:
364 case ir_unop_unpack_half_2x16:
365 this->type = glsl_type::vec2_type;
366 break;
367
368 case ir_unop_unpack_snorm_4x8:
369 case ir_unop_unpack_unorm_4x8:
370 this->type = glsl_type::vec4_type;
371 break;
372
373 case ir_unop_unpack_sampler_2x32:
374 case ir_unop_unpack_image_2x32:
375 this->type = glsl_type::uvec2_type;
376 break;
377
378 case ir_unop_pack_sampler_2x32:
379 case ir_unop_pack_image_2x32:
380 this->type = op0->type;
381 break;
382
383 case ir_unop_frexp_sig:
384 this->type = op0->type;
385 break;
386 case ir_unop_frexp_exp:
387 this->type = glsl_type::get_instance(GLSL_TYPE_INT,
388 op0->type->vector_elements, 1);
389 break;
390
391 case ir_unop_get_buffer_size:
392 case ir_unop_ssbo_unsized_array_length:
393 this->type = glsl_type::int_type;
394 break;
395
396 case ir_unop_bitcast_i642d:
397 case ir_unop_bitcast_u642d:
398 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE,
399 op0->type->vector_elements, 1);
400 break;
401
402 case ir_unop_bitcast_d2i64:
403 this->type = glsl_type::get_instance(GLSL_TYPE_INT64,
404 op0->type->vector_elements, 1);
405 break;
406 case ir_unop_bitcast_d2u64:
407 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64,
408 op0->type->vector_elements, 1);
409 break;
410
411 default:
412 assert(!"not reached: missing automatic type setup for ir_expression");
413 this->type = op0->type;
414 break;
415 }
416 }
417
418 ir_expression::ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1)
419 : ir_rvalue(ir_type_expression)
420 {
421 this->operation = ir_expression_operation(op);
422 this->operands[0] = op0;
423 this->operands[1] = op1;
424 this->operands[2] = NULL;
425 this->operands[3] = NULL;
426
427 assert(op > ir_last_unop);
428
429 switch (this->operation) {
430 case ir_binop_all_equal:
431 case ir_binop_any_nequal:
432 this->type = glsl_type::bool_type;
433 break;
434
435 case ir_binop_add:
436 case ir_binop_sub:
437 case ir_binop_min:
438 case ir_binop_max:
439 case ir_binop_pow:
440 case ir_binop_mul:
441 case ir_binop_div:
442 case ir_binop_mod:
443 if (op0->type->is_scalar()) {
444 this->type = op1->type;
445 } else if (op1->type->is_scalar()) {
446 this->type = op0->type;
447 } else {
448 if (this->operation == ir_binop_mul) {
449 this->type = glsl_type::get_mul_type(op0->type, op1->type);
450 } else {
451 assert(op0->type == op1->type);
452 this->type = op0->type;
453 }
454 }
455 break;
456
457 case ir_binop_logic_and:
458 case ir_binop_logic_xor:
459 case ir_binop_logic_or:
460 case ir_binop_bit_and:
461 case ir_binop_bit_xor:
462 case ir_binop_bit_or:
463 assert(!op0->type->is_matrix());
464 assert(!op1->type->is_matrix());
465 if (op0->type->is_scalar()) {
466 this->type = op1->type;
467 } else if (op1->type->is_scalar()) {
468 this->type = op0->type;
469 } else {
470 assert(op0->type->vector_elements == op1->type->vector_elements);
471 this->type = op0->type;
472 }
473 break;
474
475 case ir_binop_equal:
476 case ir_binop_nequal:
477 case ir_binop_lequal:
478 case ir_binop_gequal:
479 case ir_binop_less:
480 case ir_binop_greater:
481 assert(op0->type == op1->type);
482 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL,
483 op0->type->vector_elements, 1);
484 break;
485
486 case ir_binop_dot:
487 this->type = op0->type->get_base_type();
488 break;
489
490 case ir_binop_imul_high:
491 case ir_binop_carry:
492 case ir_binop_borrow:
493 case ir_binop_lshift:
494 case ir_binop_rshift:
495 case ir_binop_ldexp:
496 case ir_binop_interpolate_at_offset:
497 case ir_binop_interpolate_at_sample:
498 this->type = op0->type;
499 break;
500
501 case ir_binop_vector_extract:
502 this->type = op0->type->get_scalar_type();
503 break;
504
505 default:
506 assert(!"not reached: missing automatic type setup for ir_expression");
507 this->type = glsl_type::float_type;
508 }
509 }
510
511 ir_expression::ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1,
512 ir_rvalue *op2)
513 : ir_rvalue(ir_type_expression)
514 {
515 this->operation = ir_expression_operation(op);
516 this->operands[0] = op0;
517 this->operands[1] = op1;
518 this->operands[2] = op2;
519 this->operands[3] = NULL;
520
521 assert(op > ir_last_binop && op <= ir_last_triop);
522
523 switch (this->operation) {
524 case ir_triop_fma:
525 case ir_triop_lrp:
526 case ir_triop_bitfield_extract:
527 case ir_triop_vector_insert:
528 this->type = op0->type;
529 break;
530
531 case ir_triop_csel:
532 this->type = op1->type;
533 break;
534
535 default:
536 assert(!"not reached: missing automatic type setup for ir_expression");
537 this->type = glsl_type::float_type;
538 }
539 }
540
541 unsigned int
542 ir_expression::get_num_operands(ir_expression_operation op)
543 {
544 assert(op <= ir_last_opcode);
545
546 if (op <= ir_last_unop)
547 return 1;
548
549 if (op <= ir_last_binop)
550 return 2;
551
552 if (op <= ir_last_triop)
553 return 3;
554
555 if (op <= ir_last_quadop)
556 return 4;
557
558 assert(false);
559 return 0;
560 }
561
562 #include "ir_expression_operation_strings.h"
563
564 const char*
565 depth_layout_string(ir_depth_layout layout)
566 {
567 switch(layout) {
568 case ir_depth_layout_none: return "";
569 case ir_depth_layout_any: return "depth_any";
570 case ir_depth_layout_greater: return "depth_greater";
571 case ir_depth_layout_less: return "depth_less";
572 case ir_depth_layout_unchanged: return "depth_unchanged";
573
574 default:
575 assert(0);
576 return "";
577 }
578 }
579
580 ir_expression_operation
581 ir_expression::get_operator(const char *str)
582 {
583 for (int op = 0; op <= int(ir_last_opcode); op++) {
584 if (strcmp(str, ir_expression_operation_strings[op]) == 0)
585 return (ir_expression_operation) op;
586 }
587 return (ir_expression_operation) -1;
588 }
589
590 ir_variable *
591 ir_expression::variable_referenced() const
592 {
593 switch (operation) {
594 case ir_binop_vector_extract:
595 case ir_triop_vector_insert:
596 /* We get these for things like a[0] where a is a vector type. In these
597 * cases we want variable_referenced() to return the actual vector
598 * variable this is wrapping.
599 */
600 return operands[0]->variable_referenced();
601 default:
602 return ir_rvalue::variable_referenced();
603 }
604 }
605
606 ir_constant::ir_constant()
607 : ir_rvalue(ir_type_constant)
608 {
609 this->array_elements = NULL;
610 }
611
612 ir_constant::ir_constant(const struct glsl_type *type,
613 const ir_constant_data *data)
614 : ir_rvalue(ir_type_constant)
615 {
616 this->array_elements = NULL;
617
618 assert((type->base_type >= GLSL_TYPE_UINT)
619 && (type->base_type <= GLSL_TYPE_IMAGE));
620
621 this->type = type;
622 memcpy(& this->value, data, sizeof(this->value));
623 }
624
625 ir_constant::ir_constant(float f, unsigned vector_elements)
626 : ir_rvalue(ir_type_constant)
627 {
628 assert(vector_elements <= 4);
629 this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT, vector_elements, 1);
630 for (unsigned i = 0; i < vector_elements; i++) {
631 this->value.f[i] = f;
632 }
633 for (unsigned i = vector_elements; i < 16; i++) {
634 this->value.f[i] = 0;
635 }
636 }
637
638 ir_constant::ir_constant(double d, unsigned vector_elements)
639 : ir_rvalue(ir_type_constant)
640 {
641 assert(vector_elements <= 4);
642 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE, vector_elements, 1);
643 for (unsigned i = 0; i < vector_elements; i++) {
644 this->value.d[i] = d;
645 }
646 for (unsigned i = vector_elements; i < 16; i++) {
647 this->value.d[i] = 0.0;
648 }
649 }
650
651 ir_constant::ir_constant(unsigned int u, unsigned vector_elements)
652 : ir_rvalue(ir_type_constant)
653 {
654 assert(vector_elements <= 4);
655 this->type = glsl_type::get_instance(GLSL_TYPE_UINT, vector_elements, 1);
656 for (unsigned i = 0; i < vector_elements; i++) {
657 this->value.u[i] = u;
658 }
659 for (unsigned i = vector_elements; i < 16; i++) {
660 this->value.u[i] = 0;
661 }
662 }
663
664 ir_constant::ir_constant(int integer, unsigned vector_elements)
665 : ir_rvalue(ir_type_constant)
666 {
667 assert(vector_elements <= 4);
668 this->type = glsl_type::get_instance(GLSL_TYPE_INT, vector_elements, 1);
669 for (unsigned i = 0; i < vector_elements; i++) {
670 this->value.i[i] = integer;
671 }
672 for (unsigned i = vector_elements; i < 16; i++) {
673 this->value.i[i] = 0;
674 }
675 }
676
677 ir_constant::ir_constant(uint64_t u64, unsigned vector_elements)
678 : ir_rvalue(ir_type_constant)
679 {
680 assert(vector_elements <= 4);
681 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64, vector_elements, 1);
682 for (unsigned i = 0; i < vector_elements; i++) {
683 this->value.u64[i] = u64;
684 }
685 for (unsigned i = vector_elements; i < 16; i++) {
686 this->value.u64[i] = 0;
687 }
688 }
689
690 ir_constant::ir_constant(int64_t int64, unsigned vector_elements)
691 : ir_rvalue(ir_type_constant)
692 {
693 assert(vector_elements <= 4);
694 this->type = glsl_type::get_instance(GLSL_TYPE_INT64, vector_elements, 1);
695 for (unsigned i = 0; i < vector_elements; i++) {
696 this->value.i64[i] = int64;
697 }
698 for (unsigned i = vector_elements; i < 16; i++) {
699 this->value.i64[i] = 0;
700 }
701 }
702
703 ir_constant::ir_constant(bool b, unsigned vector_elements)
704 : ir_rvalue(ir_type_constant)
705 {
706 assert(vector_elements <= 4);
707 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL, vector_elements, 1);
708 for (unsigned i = 0; i < vector_elements; i++) {
709 this->value.b[i] = b;
710 }
711 for (unsigned i = vector_elements; i < 16; i++) {
712 this->value.b[i] = false;
713 }
714 }
715
716 ir_constant::ir_constant(const ir_constant *c, unsigned i)
717 : ir_rvalue(ir_type_constant)
718 {
719 this->array_elements = NULL;
720 this->type = c->type->get_base_type();
721
722 switch (this->type->base_type) {
723 case GLSL_TYPE_UINT: this->value.u[0] = c->value.u[i]; break;
724 case GLSL_TYPE_INT: this->value.i[0] = c->value.i[i]; break;
725 case GLSL_TYPE_FLOAT: this->value.f[0] = c->value.f[i]; break;
726 case GLSL_TYPE_BOOL: this->value.b[0] = c->value.b[i]; break;
727 case GLSL_TYPE_DOUBLE: this->value.d[0] = c->value.d[i]; break;
728 default: assert(!"Should not get here."); break;
729 }
730 }
731
732 ir_constant::ir_constant(const struct glsl_type *type, exec_list *value_list)
733 : ir_rvalue(ir_type_constant)
734 {
735 this->array_elements = NULL;
736 this->type = type;
737
738 assert(type->is_scalar() || type->is_vector() || type->is_matrix()
739 || type->is_record() || type->is_array());
740
741 if (type->is_array()) {
742 this->array_elements = ralloc_array(this, ir_constant *, type->length);
743 unsigned i = 0;
744 foreach_in_list(ir_constant, value, value_list) {
745 assert(value->as_constant() != NULL);
746
747 this->array_elements[i++] = value;
748 }
749 return;
750 }
751
752 /* If the constant is a record, the types of each of the entries in
753 * value_list must be a 1-for-1 match with the structure components. Each
754 * entry must also be a constant. Just move the nodes from the value_list
755 * to the list in the ir_constant.
756 */
757 /* FINISHME: Should there be some type checking and / or assertions here? */
758 /* FINISHME: Should the new constant take ownership of the nodes from
759 * FINISHME: value_list, or should it make copies?
760 */
761 if (type->is_record()) {
762 value_list->move_nodes_to(& this->components);
763 return;
764 }
765
766 for (unsigned i = 0; i < 16; i++) {
767 this->value.u[i] = 0;
768 }
769
770 ir_constant *value = (ir_constant *) (value_list->get_head_raw());
771
772 /* Constructors with exactly one scalar argument are special for vectors
773 * and matrices. For vectors, the scalar value is replicated to fill all
774 * the components. For matrices, the scalar fills the components of the
775 * diagonal while the rest is filled with 0.
776 */
777 if (value->type->is_scalar() && value->next->is_tail_sentinel()) {
778 if (type->is_matrix()) {
779 /* Matrix - fill diagonal (rest is already set to 0) */
780 assert(type->is_float() || type->is_double());
781 for (unsigned i = 0; i < type->matrix_columns; i++) {
782 if (type->is_float())
783 this->value.f[i * type->vector_elements + i] =
784 value->value.f[0];
785 else
786 this->value.d[i * type->vector_elements + i] =
787 value->value.d[0];
788 }
789 } else {
790 /* Vector or scalar - fill all components */
791 switch (type->base_type) {
792 case GLSL_TYPE_UINT:
793 case GLSL_TYPE_INT:
794 for (unsigned i = 0; i < type->components(); i++)
795 this->value.u[i] = value->value.u[0];
796 break;
797 case GLSL_TYPE_FLOAT:
798 for (unsigned i = 0; i < type->components(); i++)
799 this->value.f[i] = value->value.f[0];
800 break;
801 case GLSL_TYPE_DOUBLE:
802 for (unsigned i = 0; i < type->components(); i++)
803 this->value.d[i] = value->value.d[0];
804 break;
805 case GLSL_TYPE_UINT64:
806 case GLSL_TYPE_INT64:
807 for (unsigned i = 0; i < type->components(); i++)
808 this->value.u64[i] = value->value.u64[0];
809 break;
810 case GLSL_TYPE_BOOL:
811 for (unsigned i = 0; i < type->components(); i++)
812 this->value.b[i] = value->value.b[0];
813 break;
814 default:
815 assert(!"Should not get here.");
816 break;
817 }
818 }
819 return;
820 }
821
822 if (type->is_matrix() && value->type->is_matrix()) {
823 assert(value->next->is_tail_sentinel());
824
825 /* From section 5.4.2 of the GLSL 1.20 spec:
826 * "If a matrix is constructed from a matrix, then each component
827 * (column i, row j) in the result that has a corresponding component
828 * (column i, row j) in the argument will be initialized from there."
829 */
830 unsigned cols = MIN2(type->matrix_columns, value->type->matrix_columns);
831 unsigned rows = MIN2(type->vector_elements, value->type->vector_elements);
832 for (unsigned i = 0; i < cols; i++) {
833 for (unsigned j = 0; j < rows; j++) {
834 const unsigned src = i * value->type->vector_elements + j;
835 const unsigned dst = i * type->vector_elements + j;
836 this->value.f[dst] = value->value.f[src];
837 }
838 }
839
840 /* "All other components will be initialized to the identity matrix." */
841 for (unsigned i = cols; i < type->matrix_columns; i++)
842 this->value.f[i * type->vector_elements + i] = 1.0;
843
844 return;
845 }
846
847 /* Use each component from each entry in the value_list to initialize one
848 * component of the constant being constructed.
849 */
850 unsigned i = 0;
851 for (;;) {
852 assert(value->as_constant() != NULL);
853 assert(!value->is_tail_sentinel());
854
855 for (unsigned j = 0; j < value->type->components(); j++) {
856 switch (type->base_type) {
857 case GLSL_TYPE_UINT:
858 this->value.u[i] = value->get_uint_component(j);
859 break;
860 case GLSL_TYPE_INT:
861 this->value.i[i] = value->get_int_component(j);
862 break;
863 case GLSL_TYPE_FLOAT:
864 this->value.f[i] = value->get_float_component(j);
865 break;
866 case GLSL_TYPE_BOOL:
867 this->value.b[i] = value->get_bool_component(j);
868 break;
869 case GLSL_TYPE_DOUBLE:
870 this->value.d[i] = value->get_double_component(j);
871 break;
872 case GLSL_TYPE_UINT64:
873 this->value.u64[i] = value->get_uint64_component(j);
874 break;
875 case GLSL_TYPE_INT64:
876 this->value.i64[i] = value->get_int64_component(j);
877 break;
878 default:
879 /* FINISHME: What to do? Exceptions are not the answer.
880 */
881 break;
882 }
883
884 i++;
885 if (i >= type->components())
886 break;
887 }
888
889 if (i >= type->components())
890 break; /* avoid downcasting a list sentinel */
891 value = (ir_constant *) value->next;
892 }
893 }
894
895 ir_constant *
896 ir_constant::zero(void *mem_ctx, const glsl_type *type)
897 {
898 assert(type->is_scalar() || type->is_vector() || type->is_matrix()
899 || type->is_record() || type->is_array());
900
901 ir_constant *c = new(mem_ctx) ir_constant;
902 c->type = type;
903 memset(&c->value, 0, sizeof(c->value));
904
905 if (type->is_array()) {
906 c->array_elements = ralloc_array(c, ir_constant *, type->length);
907
908 for (unsigned i = 0; i < type->length; i++)
909 c->array_elements[i] = ir_constant::zero(c, type->fields.array);
910 }
911
912 if (type->is_record()) {
913 for (unsigned i = 0; i < type->length; i++) {
914 ir_constant *comp = ir_constant::zero(mem_ctx, type->fields.structure[i].type);
915 c->components.push_tail(comp);
916 }
917 }
918
919 return c;
920 }
921
922 bool
923 ir_constant::get_bool_component(unsigned i) const
924 {
925 switch (this->type->base_type) {
926 case GLSL_TYPE_UINT: return this->value.u[i] != 0;
927 case GLSL_TYPE_INT: return this->value.i[i] != 0;
928 case GLSL_TYPE_FLOAT: return ((int)this->value.f[i]) != 0;
929 case GLSL_TYPE_BOOL: return this->value.b[i];
930 case GLSL_TYPE_DOUBLE: return this->value.d[i] != 0.0;
931 case GLSL_TYPE_UINT64: return this->value.u64[i] != 0;
932 case GLSL_TYPE_INT64: return this->value.i64[i] != 0;
933 default: assert(!"Should not get here."); break;
934 }
935
936 /* Must return something to make the compiler happy. This is clearly an
937 * error case.
938 */
939 return false;
940 }
941
942 float
943 ir_constant::get_float_component(unsigned i) const
944 {
945 switch (this->type->base_type) {
946 case GLSL_TYPE_UINT: return (float) this->value.u[i];
947 case GLSL_TYPE_INT: return (float) this->value.i[i];
948 case GLSL_TYPE_FLOAT: return this->value.f[i];
949 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1.0f : 0.0f;
950 case GLSL_TYPE_DOUBLE: return (float) this->value.d[i];
951 case GLSL_TYPE_UINT64: return (float) this->value.u64[i];
952 case GLSL_TYPE_INT64: return (float) this->value.i64[i];
953 default: assert(!"Should not get here."); break;
954 }
955
956 /* Must return something to make the compiler happy. This is clearly an
957 * error case.
958 */
959 return 0.0;
960 }
961
962 double
963 ir_constant::get_double_component(unsigned i) const
964 {
965 switch (this->type->base_type) {
966 case GLSL_TYPE_UINT: return (double) this->value.u[i];
967 case GLSL_TYPE_INT: return (double) this->value.i[i];
968 case GLSL_TYPE_FLOAT: return (double) this->value.f[i];
969 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1.0 : 0.0;
970 case GLSL_TYPE_DOUBLE: return this->value.d[i];
971 case GLSL_TYPE_UINT64: return (double) this->value.u64[i];
972 case GLSL_TYPE_INT64: return (double) this->value.i64[i];
973 default: assert(!"Should not get here."); break;
974 }
975
976 /* Must return something to make the compiler happy. This is clearly an
977 * error case.
978 */
979 return 0.0;
980 }
981
982 int
983 ir_constant::get_int_component(unsigned i) const
984 {
985 switch (this->type->base_type) {
986 case GLSL_TYPE_UINT: return this->value.u[i];
987 case GLSL_TYPE_INT: return this->value.i[i];
988 case GLSL_TYPE_FLOAT: return (int) this->value.f[i];
989 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
990 case GLSL_TYPE_DOUBLE: return (int) this->value.d[i];
991 case GLSL_TYPE_UINT64: return (int) this->value.u64[i];
992 case GLSL_TYPE_INT64: return (int) this->value.i64[i];
993 default: assert(!"Should not get here."); break;
994 }
995
996 /* Must return something to make the compiler happy. This is clearly an
997 * error case.
998 */
999 return 0;
1000 }
1001
1002 unsigned
1003 ir_constant::get_uint_component(unsigned i) const
1004 {
1005 switch (this->type->base_type) {
1006 case GLSL_TYPE_UINT: return this->value.u[i];
1007 case GLSL_TYPE_INT: return this->value.i[i];
1008 case GLSL_TYPE_FLOAT: return (unsigned) this->value.f[i];
1009 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1010 case GLSL_TYPE_DOUBLE: return (unsigned) this->value.d[i];
1011 case GLSL_TYPE_UINT64: return (unsigned) this->value.u64[i];
1012 case GLSL_TYPE_INT64: return (unsigned) this->value.i64[i];
1013 default: assert(!"Should not get here."); break;
1014 }
1015
1016 /* Must return something to make the compiler happy. This is clearly an
1017 * error case.
1018 */
1019 return 0;
1020 }
1021
1022 int64_t
1023 ir_constant::get_int64_component(unsigned i) const
1024 {
1025 switch (this->type->base_type) {
1026 case GLSL_TYPE_UINT: return this->value.u[i];
1027 case GLSL_TYPE_INT: return this->value.i[i];
1028 case GLSL_TYPE_FLOAT: return (int64_t) this->value.f[i];
1029 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1030 case GLSL_TYPE_DOUBLE: return (int64_t) this->value.d[i];
1031 case GLSL_TYPE_UINT64: return (int64_t) this->value.u64[i];
1032 case GLSL_TYPE_INT64: return this->value.i64[i];
1033 default: assert(!"Should not get here."); break;
1034 }
1035
1036 /* Must return something to make the compiler happy. This is clearly an
1037 * error case.
1038 */
1039 return 0;
1040 }
1041
1042 uint64_t
1043 ir_constant::get_uint64_component(unsigned i) const
1044 {
1045 switch (this->type->base_type) {
1046 case GLSL_TYPE_UINT: return this->value.u[i];
1047 case GLSL_TYPE_INT: return this->value.i[i];
1048 case GLSL_TYPE_FLOAT: return (uint64_t) this->value.f[i];
1049 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1050 case GLSL_TYPE_DOUBLE: return (uint64_t) this->value.d[i];
1051 case GLSL_TYPE_UINT64: return this->value.u64[i];
1052 case GLSL_TYPE_INT64: return (uint64_t) this->value.i64[i];
1053 default: assert(!"Should not get here."); break;
1054 }
1055
1056 /* Must return something to make the compiler happy. This is clearly an
1057 * error case.
1058 */
1059 return 0;
1060 }
1061
1062 ir_constant *
1063 ir_constant::get_array_element(unsigned i) const
1064 {
1065 assert(this->type->is_array());
1066
1067 /* From page 35 (page 41 of the PDF) of the GLSL 1.20 spec:
1068 *
1069 * "Behavior is undefined if a shader subscripts an array with an index
1070 * less than 0 or greater than or equal to the size the array was
1071 * declared with."
1072 *
1073 * Most out-of-bounds accesses are removed before things could get this far.
1074 * There are cases where non-constant array index values can get constant
1075 * folded.
1076 */
1077 if (int(i) < 0)
1078 i = 0;
1079 else if (i >= this->type->length)
1080 i = this->type->length - 1;
1081
1082 return array_elements[i];
1083 }
1084
1085 ir_constant *
1086 ir_constant::get_record_field(const char *name)
1087 {
1088 int idx = this->type->field_index(name);
1089
1090 if (idx < 0)
1091 return NULL;
1092
1093 if (this->components.is_empty())
1094 return NULL;
1095
1096 exec_node *node = this->components.get_head_raw();
1097 for (int i = 0; i < idx; i++) {
1098 node = node->next;
1099
1100 /* If the end of the list is encountered before the element matching the
1101 * requested field is found, return NULL.
1102 */
1103 if (node->is_tail_sentinel())
1104 return NULL;
1105 }
1106
1107 return (ir_constant *) node;
1108 }
1109
1110 void
1111 ir_constant::copy_offset(ir_constant *src, int offset)
1112 {
1113 switch (this->type->base_type) {
1114 case GLSL_TYPE_UINT:
1115 case GLSL_TYPE_INT:
1116 case GLSL_TYPE_FLOAT:
1117 case GLSL_TYPE_DOUBLE:
1118 case GLSL_TYPE_UINT64:
1119 case GLSL_TYPE_INT64:
1120 case GLSL_TYPE_BOOL: {
1121 unsigned int size = src->type->components();
1122 assert (size <= this->type->components() - offset);
1123 for (unsigned int i=0; i<size; i++) {
1124 switch (this->type->base_type) {
1125 case GLSL_TYPE_UINT:
1126 value.u[i+offset] = src->get_uint_component(i);
1127 break;
1128 case GLSL_TYPE_INT:
1129 value.i[i+offset] = src->get_int_component(i);
1130 break;
1131 case GLSL_TYPE_FLOAT:
1132 value.f[i+offset] = src->get_float_component(i);
1133 break;
1134 case GLSL_TYPE_BOOL:
1135 value.b[i+offset] = src->get_bool_component(i);
1136 break;
1137 case GLSL_TYPE_DOUBLE:
1138 value.d[i+offset] = src->get_double_component(i);
1139 break;
1140 case GLSL_TYPE_UINT64:
1141 value.u64[i+offset] = src->get_uint64_component(i);
1142 break;
1143 case GLSL_TYPE_INT64:
1144 value.i64[i+offset] = src->get_int64_component(i);
1145 break;
1146 default: // Shut up the compiler
1147 break;
1148 }
1149 }
1150 break;
1151 }
1152
1153 case GLSL_TYPE_STRUCT: {
1154 assert (src->type == this->type);
1155 this->components.make_empty();
1156 foreach_in_list(ir_constant, orig, &src->components) {
1157 this->components.push_tail(orig->clone(this, NULL));
1158 }
1159 break;
1160 }
1161
1162 case GLSL_TYPE_ARRAY: {
1163 assert (src->type == this->type);
1164 for (unsigned i = 0; i < this->type->length; i++) {
1165 this->array_elements[i] = src->array_elements[i]->clone(this, NULL);
1166 }
1167 break;
1168 }
1169
1170 default:
1171 assert(!"Should not get here.");
1172 break;
1173 }
1174 }
1175
1176 void
1177 ir_constant::copy_masked_offset(ir_constant *src, int offset, unsigned int mask)
1178 {
1179 assert (!type->is_array() && !type->is_record());
1180
1181 if (!type->is_vector() && !type->is_matrix()) {
1182 offset = 0;
1183 mask = 1;
1184 }
1185
1186 int id = 0;
1187 for (int i=0; i<4; i++) {
1188 if (mask & (1 << i)) {
1189 switch (this->type->base_type) {
1190 case GLSL_TYPE_UINT:
1191 value.u[i+offset] = src->get_uint_component(id++);
1192 break;
1193 case GLSL_TYPE_INT:
1194 value.i[i+offset] = src->get_int_component(id++);
1195 break;
1196 case GLSL_TYPE_FLOAT:
1197 value.f[i+offset] = src->get_float_component(id++);
1198 break;
1199 case GLSL_TYPE_BOOL:
1200 value.b[i+offset] = src->get_bool_component(id++);
1201 break;
1202 case GLSL_TYPE_DOUBLE:
1203 value.d[i+offset] = src->get_double_component(id++);
1204 break;
1205 case GLSL_TYPE_UINT64:
1206 value.u64[i+offset] = src->get_uint64_component(id++);
1207 break;
1208 case GLSL_TYPE_INT64:
1209 value.i64[i+offset] = src->get_int64_component(id++);
1210 break;
1211 default:
1212 assert(!"Should not get here.");
1213 return;
1214 }
1215 }
1216 }
1217 }
1218
1219 bool
1220 ir_constant::has_value(const ir_constant *c) const
1221 {
1222 if (this->type != c->type)
1223 return false;
1224
1225 if (this->type->is_array()) {
1226 for (unsigned i = 0; i < this->type->length; i++) {
1227 if (!this->array_elements[i]->has_value(c->array_elements[i]))
1228 return false;
1229 }
1230 return true;
1231 }
1232
1233 if (this->type->is_record()) {
1234 const exec_node *a_node = this->components.get_head_raw();
1235 const exec_node *b_node = c->components.get_head_raw();
1236
1237 while (!a_node->is_tail_sentinel()) {
1238 assert(!b_node->is_tail_sentinel());
1239
1240 const ir_constant *const a_field = (ir_constant *) a_node;
1241 const ir_constant *const b_field = (ir_constant *) b_node;
1242
1243 if (!a_field->has_value(b_field))
1244 return false;
1245
1246 a_node = a_node->next;
1247 b_node = b_node->next;
1248 }
1249
1250 return true;
1251 }
1252
1253 for (unsigned i = 0; i < this->type->components(); i++) {
1254 switch (this->type->base_type) {
1255 case GLSL_TYPE_UINT:
1256 if (this->value.u[i] != c->value.u[i])
1257 return false;
1258 break;
1259 case GLSL_TYPE_INT:
1260 if (this->value.i[i] != c->value.i[i])
1261 return false;
1262 break;
1263 case GLSL_TYPE_FLOAT:
1264 if (this->value.f[i] != c->value.f[i])
1265 return false;
1266 break;
1267 case GLSL_TYPE_BOOL:
1268 if (this->value.b[i] != c->value.b[i])
1269 return false;
1270 break;
1271 case GLSL_TYPE_DOUBLE:
1272 if (this->value.d[i] != c->value.d[i])
1273 return false;
1274 break;
1275 case GLSL_TYPE_UINT64:
1276 if (this->value.u64[i] != c->value.u64[i])
1277 return false;
1278 break;
1279 case GLSL_TYPE_INT64:
1280 if (this->value.i64[i] != c->value.i64[i])
1281 return false;
1282 break;
1283 default:
1284 assert(!"Should not get here.");
1285 return false;
1286 }
1287 }
1288
1289 return true;
1290 }
1291
1292 bool
1293 ir_constant::is_value(float f, int i) const
1294 {
1295 if (!this->type->is_scalar() && !this->type->is_vector())
1296 return false;
1297
1298 /* Only accept boolean values for 0/1. */
1299 if (int(bool(i)) != i && this->type->is_boolean())
1300 return false;
1301
1302 for (unsigned c = 0; c < this->type->vector_elements; c++) {
1303 switch (this->type->base_type) {
1304 case GLSL_TYPE_FLOAT:
1305 if (this->value.f[c] != f)
1306 return false;
1307 break;
1308 case GLSL_TYPE_INT:
1309 if (this->value.i[c] != i)
1310 return false;
1311 break;
1312 case GLSL_TYPE_UINT:
1313 if (this->value.u[c] != unsigned(i))
1314 return false;
1315 break;
1316 case GLSL_TYPE_BOOL:
1317 if (this->value.b[c] != bool(i))
1318 return false;
1319 break;
1320 case GLSL_TYPE_DOUBLE:
1321 if (this->value.d[c] != double(f))
1322 return false;
1323 break;
1324 case GLSL_TYPE_UINT64:
1325 if (this->value.u64[c] != uint64_t(i))
1326 return false;
1327 break;
1328 case GLSL_TYPE_INT64:
1329 if (this->value.i64[c] != i)
1330 return false;
1331 break;
1332 default:
1333 /* The only other base types are structures, arrays, and samplers.
1334 * Samplers cannot be constants, and the others should have been
1335 * filtered out above.
1336 */
1337 assert(!"Should not get here.");
1338 return false;
1339 }
1340 }
1341
1342 return true;
1343 }
1344
1345 bool
1346 ir_constant::is_zero() const
1347 {
1348 return is_value(0.0, 0);
1349 }
1350
1351 bool
1352 ir_constant::is_one() const
1353 {
1354 return is_value(1.0, 1);
1355 }
1356
1357 bool
1358 ir_constant::is_negative_one() const
1359 {
1360 return is_value(-1.0, -1);
1361 }
1362
1363 bool
1364 ir_constant::is_uint16_constant() const
1365 {
1366 if (!type->is_integer())
1367 return false;
1368
1369 return value.u[0] < (1 << 16);
1370 }
1371
1372 ir_loop::ir_loop()
1373 : ir_instruction(ir_type_loop)
1374 {
1375 }
1376
1377
1378 ir_dereference_variable::ir_dereference_variable(ir_variable *var)
1379 : ir_dereference(ir_type_dereference_variable)
1380 {
1381 assert(var != NULL);
1382
1383 this->var = var;
1384 this->type = var->type;
1385 }
1386
1387
1388 ir_dereference_array::ir_dereference_array(ir_rvalue *value,
1389 ir_rvalue *array_index)
1390 : ir_dereference(ir_type_dereference_array)
1391 {
1392 this->array_index = array_index;
1393 this->set_array(value);
1394 }
1395
1396
1397 ir_dereference_array::ir_dereference_array(ir_variable *var,
1398 ir_rvalue *array_index)
1399 : ir_dereference(ir_type_dereference_array)
1400 {
1401 void *ctx = ralloc_parent(var);
1402
1403 this->array_index = array_index;
1404 this->set_array(new(ctx) ir_dereference_variable(var));
1405 }
1406
1407
1408 void
1409 ir_dereference_array::set_array(ir_rvalue *value)
1410 {
1411 assert(value != NULL);
1412
1413 this->array = value;
1414
1415 const glsl_type *const vt = this->array->type;
1416
1417 if (vt->is_array()) {
1418 type = vt->fields.array;
1419 } else if (vt->is_matrix()) {
1420 type = vt->column_type();
1421 } else if (vt->is_vector()) {
1422 type = vt->get_base_type();
1423 }
1424 }
1425
1426
1427 ir_dereference_record::ir_dereference_record(ir_rvalue *value,
1428 const char *field)
1429 : ir_dereference(ir_type_dereference_record)
1430 {
1431 assert(value != NULL);
1432
1433 this->record = value;
1434 this->field = ralloc_strdup(this, field);
1435 this->type = this->record->type->field_type(field);
1436 }
1437
1438
1439 ir_dereference_record::ir_dereference_record(ir_variable *var,
1440 const char *field)
1441 : ir_dereference(ir_type_dereference_record)
1442 {
1443 void *ctx = ralloc_parent(var);
1444
1445 this->record = new(ctx) ir_dereference_variable(var);
1446 this->field = ralloc_strdup(this, field);
1447 this->type = this->record->type->field_type(field);
1448 }
1449
1450 bool
1451 ir_dereference::is_lvalue(const struct _mesa_glsl_parse_state *state) const
1452 {
1453 ir_variable *var = this->variable_referenced();
1454
1455 /* Every l-value derference chain eventually ends in a variable.
1456 */
1457 if ((var == NULL) || var->data.read_only)
1458 return false;
1459
1460 /* From section 4.1.7 of the ARB_bindless_texture spec:
1461 *
1462 * "Samplers can be used as l-values, so can be assigned into and used as
1463 * "out" and "inout" function parameters."
1464 *
1465 * From section 4.1.X of the ARB_bindless_texture spec:
1466 *
1467 * "Images can be used as l-values, so can be assigned into and used as
1468 * "out" and "inout" function parameters."
1469 */
1470 if ((!state || state->has_bindless()) &&
1471 (this->type->contains_sampler() || this->type->contains_image()))
1472 return true;
1473
1474 /* From section 4.1.7 of the GLSL 4.40 spec:
1475 *
1476 * "Opaque variables cannot be treated as l-values; hence cannot
1477 * be used as out or inout function parameters, nor can they be
1478 * assigned into."
1479 */
1480 if (this->type->contains_opaque())
1481 return false;
1482
1483 return true;
1484 }
1485
1486
1487 static const char * const tex_opcode_strs[] = { "tex", "txb", "txl", "txd", "txf", "txf_ms", "txs", "lod", "tg4", "query_levels", "texture_samples", "samples_identical" };
1488
1489 const char *ir_texture::opcode_string()
1490 {
1491 assert((unsigned int) op < ARRAY_SIZE(tex_opcode_strs));
1492 return tex_opcode_strs[op];
1493 }
1494
1495 ir_texture_opcode
1496 ir_texture::get_opcode(const char *str)
1497 {
1498 const int count = sizeof(tex_opcode_strs) / sizeof(tex_opcode_strs[0]);
1499 for (int op = 0; op < count; op++) {
1500 if (strcmp(str, tex_opcode_strs[op]) == 0)
1501 return (ir_texture_opcode) op;
1502 }
1503 return (ir_texture_opcode) -1;
1504 }
1505
1506
1507 void
1508 ir_texture::set_sampler(ir_dereference *sampler, const glsl_type *type)
1509 {
1510 assert(sampler != NULL);
1511 assert(type != NULL);
1512 this->sampler = sampler;
1513 this->type = type;
1514
1515 if (this->op == ir_txs || this->op == ir_query_levels ||
1516 this->op == ir_texture_samples) {
1517 assert(type->base_type == GLSL_TYPE_INT);
1518 } else if (this->op == ir_lod) {
1519 assert(type->vector_elements == 2);
1520 assert(type->is_float());
1521 } else if (this->op == ir_samples_identical) {
1522 assert(type == glsl_type::bool_type);
1523 assert(sampler->type->is_sampler());
1524 assert(sampler->type->sampler_dimensionality == GLSL_SAMPLER_DIM_MS);
1525 } else {
1526 assert(sampler->type->sampled_type == (int) type->base_type);
1527 if (sampler->type->sampler_shadow)
1528 assert(type->vector_elements == 4 || type->vector_elements == 1);
1529 else
1530 assert(type->vector_elements == 4);
1531 }
1532 }
1533
1534
1535 void
1536 ir_swizzle::init_mask(const unsigned *comp, unsigned count)
1537 {
1538 assert((count >= 1) && (count <= 4));
1539
1540 memset(&this->mask, 0, sizeof(this->mask));
1541 this->mask.num_components = count;
1542
1543 unsigned dup_mask = 0;
1544 switch (count) {
1545 case 4:
1546 assert(comp[3] <= 3);
1547 dup_mask |= (1U << comp[3])
1548 & ((1U << comp[0]) | (1U << comp[1]) | (1U << comp[2]));
1549 this->mask.w = comp[3];
1550
1551 case 3:
1552 assert(comp[2] <= 3);
1553 dup_mask |= (1U << comp[2])
1554 & ((1U << comp[0]) | (1U << comp[1]));
1555 this->mask.z = comp[2];
1556
1557 case 2:
1558 assert(comp[1] <= 3);
1559 dup_mask |= (1U << comp[1])
1560 & ((1U << comp[0]));
1561 this->mask.y = comp[1];
1562
1563 case 1:
1564 assert(comp[0] <= 3);
1565 this->mask.x = comp[0];
1566 }
1567
1568 this->mask.has_duplicates = dup_mask != 0;
1569
1570 /* Based on the number of elements in the swizzle and the base type
1571 * (i.e., float, int, unsigned, or bool) of the vector being swizzled,
1572 * generate the type of the resulting value.
1573 */
1574 type = glsl_type::get_instance(val->type->base_type, mask.num_components, 1);
1575 }
1576
1577 ir_swizzle::ir_swizzle(ir_rvalue *val, unsigned x, unsigned y, unsigned z,
1578 unsigned w, unsigned count)
1579 : ir_rvalue(ir_type_swizzle), val(val)
1580 {
1581 const unsigned components[4] = { x, y, z, w };
1582 this->init_mask(components, count);
1583 }
1584
1585 ir_swizzle::ir_swizzle(ir_rvalue *val, const unsigned *comp,
1586 unsigned count)
1587 : ir_rvalue(ir_type_swizzle), val(val)
1588 {
1589 this->init_mask(comp, count);
1590 }
1591
1592 ir_swizzle::ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask)
1593 : ir_rvalue(ir_type_swizzle)
1594 {
1595 this->val = val;
1596 this->mask = mask;
1597 this->type = glsl_type::get_instance(val->type->base_type,
1598 mask.num_components, 1);
1599 }
1600
1601 #define X 1
1602 #define R 5
1603 #define S 9
1604 #define I 13
1605
1606 ir_swizzle *
1607 ir_swizzle::create(ir_rvalue *val, const char *str, unsigned vector_length)
1608 {
1609 void *ctx = ralloc_parent(val);
1610
1611 /* For each possible swizzle character, this table encodes the value in
1612 * \c idx_map that represents the 0th element of the vector. For invalid
1613 * swizzle characters (e.g., 'k'), a special value is used that will allow
1614 * detection of errors.
1615 */
1616 static const unsigned char base_idx[26] = {
1617 /* a b c d e f g h i j k l m */
1618 R, R, I, I, I, I, R, I, I, I, I, I, I,
1619 /* n o p q r s t u v w x y z */
1620 I, I, S, S, R, S, S, I, I, X, X, X, X
1621 };
1622
1623 /* Each valid swizzle character has an entry in the previous table. This
1624 * table encodes the base index encoded in the previous table plus the actual
1625 * index of the swizzle character. When processing swizzles, the first
1626 * character in the string is indexed in the previous table. Each character
1627 * in the string is indexed in this table, and the value found there has the
1628 * value form the first table subtracted. The result must be on the range
1629 * [0,3].
1630 *
1631 * For example, the string "wzyx" will get X from the first table. Each of
1632 * the charcaters will get X+3, X+2, X+1, and X+0 from this table. After
1633 * subtraction, the swizzle values are { 3, 2, 1, 0 }.
1634 *
1635 * The string "wzrg" will get X from the first table. Each of the characters
1636 * will get X+3, X+2, R+0, and R+1 from this table. After subtraction, the
1637 * swizzle values are { 3, 2, 4, 5 }. Since 4 and 5 are outside the range
1638 * [0,3], the error is detected.
1639 */
1640 static const unsigned char idx_map[26] = {
1641 /* a b c d e f g h i j k l m */
1642 R+3, R+2, 0, 0, 0, 0, R+1, 0, 0, 0, 0, 0, 0,
1643 /* n o p q r s t u v w x y z */
1644 0, 0, S+2, S+3, R+0, S+0, S+1, 0, 0, X+3, X+0, X+1, X+2
1645 };
1646
1647 int swiz_idx[4] = { 0, 0, 0, 0 };
1648 unsigned i;
1649
1650
1651 /* Validate the first character in the swizzle string and look up the base
1652 * index value as described above.
1653 */
1654 if ((str[0] < 'a') || (str[0] > 'z'))
1655 return NULL;
1656
1657 const unsigned base = base_idx[str[0] - 'a'];
1658
1659
1660 for (i = 0; (i < 4) && (str[i] != '\0'); i++) {
1661 /* Validate the next character, and, as described above, convert it to a
1662 * swizzle index.
1663 */
1664 if ((str[i] < 'a') || (str[i] > 'z'))
1665 return NULL;
1666
1667 swiz_idx[i] = idx_map[str[i] - 'a'] - base;
1668 if ((swiz_idx[i] < 0) || (swiz_idx[i] >= (int) vector_length))
1669 return NULL;
1670 }
1671
1672 if (str[i] != '\0')
1673 return NULL;
1674
1675 return new(ctx) ir_swizzle(val, swiz_idx[0], swiz_idx[1], swiz_idx[2],
1676 swiz_idx[3], i);
1677 }
1678
1679 #undef X
1680 #undef R
1681 #undef S
1682 #undef I
1683
1684 ir_variable *
1685 ir_swizzle::variable_referenced() const
1686 {
1687 return this->val->variable_referenced();
1688 }
1689
1690
1691 bool ir_variable::temporaries_allocate_names = false;
1692
1693 const char ir_variable::tmp_name[] = "compiler_temp";
1694
1695 ir_variable::ir_variable(const struct glsl_type *type, const char *name,
1696 ir_variable_mode mode)
1697 : ir_instruction(ir_type_variable)
1698 {
1699 this->type = type;
1700
1701 if (mode == ir_var_temporary && !ir_variable::temporaries_allocate_names)
1702 name = NULL;
1703
1704 /* The ir_variable clone method may call this constructor with name set to
1705 * tmp_name.
1706 */
1707 assert(name != NULL
1708 || mode == ir_var_temporary
1709 || mode == ir_var_function_in
1710 || mode == ir_var_function_out
1711 || mode == ir_var_function_inout);
1712 assert(name != ir_variable::tmp_name
1713 || mode == ir_var_temporary);
1714 if (mode == ir_var_temporary
1715 && (name == NULL || name == ir_variable::tmp_name)) {
1716 this->name = ir_variable::tmp_name;
1717 } else if (name == NULL ||
1718 strlen(name) < ARRAY_SIZE(this->name_storage)) {
1719 strcpy(this->name_storage, name ? name : "");
1720 this->name = this->name_storage;
1721 } else {
1722 this->name = ralloc_strdup(this, name);
1723 }
1724
1725 this->u.max_ifc_array_access = NULL;
1726
1727 this->data.explicit_location = false;
1728 this->data.has_initializer = false;
1729 this->data.location = -1;
1730 this->data.location_frac = 0;
1731 this->data.binding = 0;
1732 this->data.warn_extension_index = 0;
1733 this->constant_value = NULL;
1734 this->constant_initializer = NULL;
1735 this->data.origin_upper_left = false;
1736 this->data.pixel_center_integer = false;
1737 this->data.depth_layout = ir_depth_layout_none;
1738 this->data.used = false;
1739 this->data.always_active_io = false;
1740 this->data.read_only = false;
1741 this->data.centroid = false;
1742 this->data.sample = false;
1743 this->data.patch = false;
1744 this->data.invariant = false;
1745 this->data.how_declared = ir_var_declared_normally;
1746 this->data.mode = mode;
1747 this->data.interpolation = INTERP_MODE_NONE;
1748 this->data.max_array_access = -1;
1749 this->data.offset = 0;
1750 this->data.precision = GLSL_PRECISION_NONE;
1751 this->data.memory_read_only = false;
1752 this->data.memory_write_only = false;
1753 this->data.memory_coherent = false;
1754 this->data.memory_volatile = false;
1755 this->data.memory_restrict = false;
1756 this->data.from_ssbo_unsized_array = false;
1757 this->data.fb_fetch_output = false;
1758 this->data.bindless = false;
1759 this->data.bound = false;
1760
1761 if (type != NULL) {
1762 if (type->is_interface())
1763 this->init_interface_type(type);
1764 else if (type->without_array()->is_interface())
1765 this->init_interface_type(type->without_array());
1766 }
1767 }
1768
1769
1770 const char *
1771 interpolation_string(unsigned interpolation)
1772 {
1773 switch (interpolation) {
1774 case INTERP_MODE_NONE: return "no";
1775 case INTERP_MODE_SMOOTH: return "smooth";
1776 case INTERP_MODE_FLAT: return "flat";
1777 case INTERP_MODE_NOPERSPECTIVE: return "noperspective";
1778 }
1779
1780 assert(!"Should not get here.");
1781 return "";
1782 }
1783
1784 const char *const ir_variable::warn_extension_table[] = {
1785 "",
1786 "GL_ARB_shader_stencil_export",
1787 "GL_AMD_shader_stencil_export",
1788 };
1789
1790 void
1791 ir_variable::enable_extension_warning(const char *extension)
1792 {
1793 for (unsigned i = 0; i < ARRAY_SIZE(warn_extension_table); i++) {
1794 if (strcmp(warn_extension_table[i], extension) == 0) {
1795 this->data.warn_extension_index = i;
1796 return;
1797 }
1798 }
1799
1800 assert(!"Should not get here.");
1801 this->data.warn_extension_index = 0;
1802 }
1803
1804 const char *
1805 ir_variable::get_extension_warning() const
1806 {
1807 return this->data.warn_extension_index == 0
1808 ? NULL : warn_extension_table[this->data.warn_extension_index];
1809 }
1810
1811 ir_function_signature::ir_function_signature(const glsl_type *return_type,
1812 builtin_available_predicate b)
1813 : ir_instruction(ir_type_function_signature),
1814 return_type(return_type), is_defined(false),
1815 intrinsic_id(ir_intrinsic_invalid), builtin_avail(b), _function(NULL)
1816 {
1817 this->origin = NULL;
1818 }
1819
1820
1821 bool
1822 ir_function_signature::is_builtin() const
1823 {
1824 return builtin_avail != NULL;
1825 }
1826
1827
1828 bool
1829 ir_function_signature::is_builtin_available(const _mesa_glsl_parse_state *state) const
1830 {
1831 /* We can't call the predicate without a state pointer, so just say that
1832 * the signature is available. At compile time, we need the filtering,
1833 * but also receive a valid state pointer. At link time, we're resolving
1834 * imported built-in prototypes to their definitions, which will always
1835 * be an exact match. So we can skip the filtering.
1836 */
1837 if (state == NULL)
1838 return true;
1839
1840 assert(builtin_avail != NULL);
1841 return builtin_avail(state);
1842 }
1843
1844
1845 static bool
1846 modes_match(unsigned a, unsigned b)
1847 {
1848 if (a == b)
1849 return true;
1850
1851 /* Accept "in" vs. "const in" */
1852 if ((a == ir_var_const_in && b == ir_var_function_in) ||
1853 (b == ir_var_const_in && a == ir_var_function_in))
1854 return true;
1855
1856 return false;
1857 }
1858
1859
1860 const char *
1861 ir_function_signature::qualifiers_match(exec_list *params)
1862 {
1863 /* check that the qualifiers match. */
1864 foreach_two_lists(a_node, &this->parameters, b_node, params) {
1865 ir_variable *a = (ir_variable *) a_node;
1866 ir_variable *b = (ir_variable *) b_node;
1867
1868 if (a->data.read_only != b->data.read_only ||
1869 !modes_match(a->data.mode, b->data.mode) ||
1870 a->data.interpolation != b->data.interpolation ||
1871 a->data.centroid != b->data.centroid ||
1872 a->data.sample != b->data.sample ||
1873 a->data.patch != b->data.patch ||
1874 a->data.memory_read_only != b->data.memory_read_only ||
1875 a->data.memory_write_only != b->data.memory_write_only ||
1876 a->data.memory_coherent != b->data.memory_coherent ||
1877 a->data.memory_volatile != b->data.memory_volatile ||
1878 a->data.memory_restrict != b->data.memory_restrict) {
1879
1880 /* parameter a's qualifiers don't match */
1881 return a->name;
1882 }
1883 }
1884 return NULL;
1885 }
1886
1887
1888 void
1889 ir_function_signature::replace_parameters(exec_list *new_params)
1890 {
1891 /* Destroy all of the previous parameter information. If the previous
1892 * parameter information comes from the function prototype, it may either
1893 * specify incorrect parameter names or not have names at all.
1894 */
1895 new_params->move_nodes_to(&parameters);
1896 }
1897
1898
1899 ir_function::ir_function(const char *name)
1900 : ir_instruction(ir_type_function)
1901 {
1902 this->subroutine_index = -1;
1903 this->name = ralloc_strdup(this, name);
1904 }
1905
1906
1907 bool
1908 ir_function::has_user_signature()
1909 {
1910 foreach_in_list(ir_function_signature, sig, &this->signatures) {
1911 if (!sig->is_builtin())
1912 return true;
1913 }
1914 return false;
1915 }
1916
1917
1918 ir_rvalue *
1919 ir_rvalue::error_value(void *mem_ctx)
1920 {
1921 ir_rvalue *v = new(mem_ctx) ir_rvalue(ir_type_unset);
1922
1923 v->type = glsl_type::error_type;
1924 return v;
1925 }
1926
1927
1928 void
1929 visit_exec_list(exec_list *list, ir_visitor *visitor)
1930 {
1931 foreach_in_list_safe(ir_instruction, node, list) {
1932 node->accept(visitor);
1933 }
1934 }
1935
1936
1937 static void
1938 steal_memory(ir_instruction *ir, void *new_ctx)
1939 {
1940 ir_variable *var = ir->as_variable();
1941 ir_function *fn = ir->as_function();
1942 ir_constant *constant = ir->as_constant();
1943 if (var != NULL && var->constant_value != NULL)
1944 steal_memory(var->constant_value, ir);
1945
1946 if (var != NULL && var->constant_initializer != NULL)
1947 steal_memory(var->constant_initializer, ir);
1948
1949 if (fn != NULL && fn->subroutine_types)
1950 ralloc_steal(new_ctx, fn->subroutine_types);
1951
1952 /* The components of aggregate constants are not visited by the normal
1953 * visitor, so steal their values by hand.
1954 */
1955 if (constant != NULL) {
1956 if (constant->type->is_record()) {
1957 foreach_in_list(ir_constant, field, &constant->components) {
1958 steal_memory(field, ir);
1959 }
1960 } else if (constant->type->is_array()) {
1961 for (unsigned int i = 0; i < constant->type->length; i++) {
1962 steal_memory(constant->array_elements[i], ir);
1963 }
1964 }
1965 }
1966
1967 ralloc_steal(new_ctx, ir);
1968 }
1969
1970
1971 void
1972 reparent_ir(exec_list *list, void *mem_ctx)
1973 {
1974 foreach_in_list(ir_instruction, node, list) {
1975 visit_tree(node, steal_memory, mem_ctx);
1976 }
1977 }
1978
1979
1980 static ir_rvalue *
1981 try_min_one(ir_rvalue *ir)
1982 {
1983 ir_expression *expr = ir->as_expression();
1984
1985 if (!expr || expr->operation != ir_binop_min)
1986 return NULL;
1987
1988 if (expr->operands[0]->is_one())
1989 return expr->operands[1];
1990
1991 if (expr->operands[1]->is_one())
1992 return expr->operands[0];
1993
1994 return NULL;
1995 }
1996
1997 static ir_rvalue *
1998 try_max_zero(ir_rvalue *ir)
1999 {
2000 ir_expression *expr = ir->as_expression();
2001
2002 if (!expr || expr->operation != ir_binop_max)
2003 return NULL;
2004
2005 if (expr->operands[0]->is_zero())
2006 return expr->operands[1];
2007
2008 if (expr->operands[1]->is_zero())
2009 return expr->operands[0];
2010
2011 return NULL;
2012 }
2013
2014 ir_rvalue *
2015 ir_rvalue::as_rvalue_to_saturate()
2016 {
2017 ir_expression *expr = this->as_expression();
2018
2019 if (!expr)
2020 return NULL;
2021
2022 ir_rvalue *max_zero = try_max_zero(expr);
2023 if (max_zero) {
2024 return try_min_one(max_zero);
2025 } else {
2026 ir_rvalue *min_one = try_min_one(expr);
2027 if (min_one) {
2028 return try_max_zero(min_one);
2029 }
2030 }
2031
2032 return NULL;
2033 }
2034
2035
2036 unsigned
2037 vertices_per_prim(GLenum prim)
2038 {
2039 switch (prim) {
2040 case GL_POINTS:
2041 return 1;
2042 case GL_LINES:
2043 return 2;
2044 case GL_TRIANGLES:
2045 return 3;
2046 case GL_LINES_ADJACENCY:
2047 return 4;
2048 case GL_TRIANGLES_ADJACENCY:
2049 return 6;
2050 default:
2051 assert(!"Bad primitive");
2052 return 3;
2053 }
2054 }
2055
2056 /**
2057 * Generate a string describing the mode of a variable
2058 */
2059 const char *
2060 mode_string(const ir_variable *var)
2061 {
2062 switch (var->data.mode) {
2063 case ir_var_auto:
2064 return (var->data.read_only) ? "global constant" : "global variable";
2065
2066 case ir_var_uniform:
2067 return "uniform";
2068
2069 case ir_var_shader_storage:
2070 return "buffer";
2071
2072 case ir_var_shader_in:
2073 return "shader input";
2074
2075 case ir_var_shader_out:
2076 return "shader output";
2077
2078 case ir_var_function_in:
2079 case ir_var_const_in:
2080 return "function input";
2081
2082 case ir_var_function_out:
2083 return "function output";
2084
2085 case ir_var_function_inout:
2086 return "function inout";
2087
2088 case ir_var_system_value:
2089 return "shader input";
2090
2091 case ir_var_temporary:
2092 return "compiler temporary";
2093
2094 case ir_var_mode_count:
2095 break;
2096 }
2097
2098 assert(!"Should not get here.");
2099 return "invalid variable";
2100 }