projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
glsl: make use of glsl_type::is_double()
[mesa.git] / src / compiler / glsl / ir.cpp
1 /*
2 * Copyright © 2010 Intel Corporation
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
21 * DEALINGS IN THE SOFTWARE.
22 */
23 #include <string.h>
24 #include "main/core.h" /* for MAX2 */
25 #include "ir.h"
26 #include "compiler/glsl_types.h"
27
28 ir_rvalue::ir_rvalue(enum ir_node_type t)
29 : ir_instruction(t)
30 {
31 this->type = glsl_type::error_type;
32 }
33
34 bool ir_rvalue::is_zero() const
35 {
36 return false;
37 }
38
39 bool ir_rvalue::is_one() const
40 {
41 return false;
42 }
43
44 bool ir_rvalue::is_negative_one() const
45 {
46 return false;
47 }
48
49 /**
50 * Modify the swizzle make to move one component to another
51 *
52 * \param m IR swizzle to be modified
53 * \param from Component in the RHS that is to be swizzled
54 * \param to Desired swizzle location of \c from
55 */
56 static void
57 update_rhs_swizzle(ir_swizzle_mask &m, unsigned from, unsigned to)
58 {
59 switch (to) {
60 case 0: m.x = from; break;
61 case 1: m.y = from; break;
62 case 2: m.z = from; break;
63 case 3: m.w = from; break;
64 default: assert(!"Should not get here.");
65 }
66 }
67
68 void
69 ir_assignment::set_lhs(ir_rvalue *lhs)
70 {
71 void *mem_ctx = this;
72 bool swizzled = false;
73
74 while (lhs != NULL) {
75 ir_swizzle *swiz = lhs->as_swizzle();
76
77 if (swiz == NULL)
78 break;
79
80 unsigned write_mask = 0;
81 ir_swizzle_mask rhs_swiz = { 0, 0, 0, 0, 0, 0 };
82
83 for (unsigned i = 0; i < swiz->mask.num_components; i++) {
84 unsigned c = 0;
85
86 switch (i) {
87 case 0: c = swiz->mask.x; break;
88 case 1: c = swiz->mask.y; break;
89 case 2: c = swiz->mask.z; break;
90 case 3: c = swiz->mask.w; break;
91 default: assert(!"Should not get here.");
92 }
93
94 write_mask |= (((this->write_mask >> i) & 1) << c);
95 update_rhs_swizzle(rhs_swiz, i, c);
96 rhs_swiz.num_components = swiz->val->type->vector_elements;
97 }
98
99 this->write_mask = write_mask;
100 lhs = swiz->val;
101
102 this->rhs = new(mem_ctx) ir_swizzle(this->rhs, rhs_swiz);
103 swizzled = true;
104 }
105
106 if (swizzled) {
107 /* Now, RHS channels line up with the LHS writemask. Collapse it
108 * to just the channels that will be written.
109 */
110 ir_swizzle_mask rhs_swiz = { 0, 0, 0, 0, 0, 0 };
111 int rhs_chan = 0;
112 for (int i = 0; i < 4; i++) {
113 if (write_mask & (1 << i))
114 update_rhs_swizzle(rhs_swiz, i, rhs_chan++);
115 }
116 rhs_swiz.num_components = rhs_chan;
117 this->rhs = new(mem_ctx) ir_swizzle(this->rhs, rhs_swiz);
118 }
119
120 assert((lhs == NULL) || lhs->as_dereference());
121
122 this->lhs = (ir_dereference *) lhs;
123 }
124
125 ir_variable *
126 ir_assignment::whole_variable_written()
127 {
128 ir_variable *v = this->lhs->whole_variable_referenced();
129
130 if (v == NULL)
131 return NULL;
132
133 if (v->type->is_scalar())
134 return v;
135
136 if (v->type->is_vector()) {
137 const unsigned mask = (1U << v->type->vector_elements) - 1;
138
139 if (mask != this->write_mask)
140 return NULL;
141 }
142
143 /* Either all the vector components are assigned or the variable is some
144 * composite type (and the whole thing is assigned.
145 */
146 return v;
147 }
148
149 ir_assignment::ir_assignment(ir_dereference *lhs, ir_rvalue *rhs,
150 ir_rvalue *condition, unsigned write_mask)
151 : ir_instruction(ir_type_assignment)
152 {
153 this->condition = condition;
154 this->rhs = rhs;
155 this->lhs = lhs;
156 this->write_mask = write_mask;
157
158 if (lhs->type->is_scalar() || lhs->type->is_vector()) {
159 int lhs_components = 0;
160 for (int i = 0; i < 4; i++) {
161 if (write_mask & (1 << i))
162 lhs_components++;
163 }
164
165 assert(lhs_components == this->rhs->type->vector_elements);
166 }
167 }
168
169 ir_assignment::ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs,
170 ir_rvalue *condition)
171 : ir_instruction(ir_type_assignment)
172 {
173 this->condition = condition;
174 this->rhs = rhs;
175
176 /* If the RHS is a vector type, assume that all components of the vector
177 * type are being written to the LHS. The write mask comes from the RHS
178 * because we can have a case where the LHS is a vec4 and the RHS is a
179 * vec3. In that case, the assignment is:
180 *
181 * (assign (...) (xyz) (var_ref lhs) (var_ref rhs))
182 */
183 if (rhs->type->is_vector())
184 this->write_mask = (1U << rhs->type->vector_elements) - 1;
185 else if (rhs->type->is_scalar())
186 this->write_mask = 1;
187 else
188 this->write_mask = 0;
189
190 this->set_lhs(lhs);
191 }
192
193 ir_expression::ir_expression(int op, const struct glsl_type *type,
194 ir_rvalue *op0, ir_rvalue *op1,
195 ir_rvalue *op2, ir_rvalue *op3)
196 : ir_rvalue(ir_type_expression)
197 {
198 this->type = type;
199 this->operation = ir_expression_operation(op);
200 this->operands[0] = op0;
201 this->operands[1] = op1;
202 this->operands[2] = op2;
203 this->operands[3] = op3;
204 #ifndef NDEBUG
205 int num_operands = get_num_operands(this->operation);
206 for (int i = num_operands; i < 4; i++) {
207 assert(this->operands[i] == NULL);
208 }
209 #endif
210 }
211
212 ir_expression::ir_expression(int op, ir_rvalue *op0)
213 : ir_rvalue(ir_type_expression)
214 {
215 this->operation = ir_expression_operation(op);
216 this->operands[0] = op0;
217 this->operands[1] = NULL;
218 this->operands[2] = NULL;
219 this->operands[3] = NULL;
220
221 assert(op <= ir_last_unop);
222
223 switch (this->operation) {
224 case ir_unop_bit_not:
225 case ir_unop_logic_not:
226 case ir_unop_neg:
227 case ir_unop_abs:
228 case ir_unop_sign:
229 case ir_unop_rcp:
230 case ir_unop_rsq:
231 case ir_unop_sqrt:
232 case ir_unop_exp:
233 case ir_unop_log:
234 case ir_unop_exp2:
235 case ir_unop_log2:
236 case ir_unop_trunc:
237 case ir_unop_ceil:
238 case ir_unop_floor:
239 case ir_unop_fract:
240 case ir_unop_round_even:
241 case ir_unop_sin:
242 case ir_unop_cos:
243 case ir_unop_dFdx:
244 case ir_unop_dFdx_coarse:
245 case ir_unop_dFdx_fine:
246 case ir_unop_dFdy:
247 case ir_unop_dFdy_coarse:
248 case ir_unop_dFdy_fine:
249 case ir_unop_bitfield_reverse:
250 case ir_unop_interpolate_at_centroid:
251 case ir_unop_saturate:
252 this->type = op0->type;
253 break;
254
255 case ir_unop_f2i:
256 case ir_unop_b2i:
257 case ir_unop_u2i:
258 case ir_unop_d2i:
259 case ir_unop_bitcast_f2i:
260 case ir_unop_bit_count:
261 case ir_unop_find_msb:
262 case ir_unop_find_lsb:
263 case ir_unop_subroutine_to_int:
264 case ir_unop_i642i:
265 case ir_unop_u642i:
266 this->type = glsl_type::get_instance(GLSL_TYPE_INT,
267 op0->type->vector_elements, 1);
268 break;
269
270 case ir_unop_b2f:
271 case ir_unop_i2f:
272 case ir_unop_u2f:
273 case ir_unop_d2f:
274 case ir_unop_bitcast_i2f:
275 case ir_unop_bitcast_u2f:
276 case ir_unop_i642f:
277 case ir_unop_u642f:
278 this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT,
279 op0->type->vector_elements, 1);
280 break;
281
282 case ir_unop_f2b:
283 case ir_unop_i2b:
284 case ir_unop_d2b:
285 case ir_unop_i642b:
286 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL,
287 op0->type->vector_elements, 1);
288 break;
289
290 case ir_unop_f2d:
291 case ir_unop_i2d:
292 case ir_unop_u2d:
293 case ir_unop_i642d:
294 case ir_unop_u642d:
295 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE,
296 op0->type->vector_elements, 1);
297 break;
298
299 case ir_unop_i2u:
300 case ir_unop_f2u:
301 case ir_unop_d2u:
302 case ir_unop_bitcast_f2u:
303 case ir_unop_i642u:
304 case ir_unop_u642u:
305 this->type = glsl_type::get_instance(GLSL_TYPE_UINT,
306 op0->type->vector_elements, 1);
307 break;
308
309 case ir_unop_i2i64:
310 case ir_unop_u2i64:
311 case ir_unop_b2i64:
312 case ir_unop_f2i64:
313 case ir_unop_d2i64:
314 case ir_unop_u642i64:
315 this->type = glsl_type::get_instance(GLSL_TYPE_INT64,
316 op0->type->vector_elements, 1);
317 break;
318
319 case ir_unop_i2u64:
320 case ir_unop_u2u64:
321 case ir_unop_f2u64:
322 case ir_unop_d2u64:
323 case ir_unop_i642u64:
324 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64,
325 op0->type->vector_elements, 1);
326 break;
327 case ir_unop_noise:
328 this->type = glsl_type::float_type;
329 break;
330
331 case ir_unop_unpack_double_2x32:
332 case ir_unop_unpack_uint_2x32:
333 this->type = glsl_type::uvec2_type;
334 break;
335
336 case ir_unop_unpack_int_2x32:
337 this->type = glsl_type::ivec2_type;
338 break;
339
340 case ir_unop_pack_snorm_2x16:
341 case ir_unop_pack_snorm_4x8:
342 case ir_unop_pack_unorm_2x16:
343 case ir_unop_pack_unorm_4x8:
344 case ir_unop_pack_half_2x16:
345 this->type = glsl_type::uint_type;
346 break;
347
348 case ir_unop_pack_double_2x32:
349 this->type = glsl_type::double_type;
350 break;
351
352 case ir_unop_pack_int_2x32:
353 this->type = glsl_type::int64_t_type;
354 break;
355
356 case ir_unop_pack_uint_2x32:
357 this->type = glsl_type::uint64_t_type;
358 break;
359
360 case ir_unop_unpack_snorm_2x16:
361 case ir_unop_unpack_unorm_2x16:
362 case ir_unop_unpack_half_2x16:
363 this->type = glsl_type::vec2_type;
364 break;
365
366 case ir_unop_unpack_snorm_4x8:
367 case ir_unop_unpack_unorm_4x8:
368 this->type = glsl_type::vec4_type;
369 break;
370
371 case ir_unop_frexp_sig:
372 this->type = op0->type;
373 break;
374 case ir_unop_frexp_exp:
375 this->type = glsl_type::get_instance(GLSL_TYPE_INT,
376 op0->type->vector_elements, 1);
377 break;
378
379 case ir_unop_get_buffer_size:
380 case ir_unop_ssbo_unsized_array_length:
381 this->type = glsl_type::int_type;
382 break;
383
384 case ir_unop_ballot:
385 this->type = glsl_type::uint64_t_type;
386 break;
387
388 case ir_unop_read_first_invocation:
389 this->type = op0->type;
390 break;
391
392 case ir_unop_vote_any:
393 case ir_unop_vote_all:
394 case ir_unop_vote_eq:
395 this->type = glsl_type::bool_type;
396 break;
397
398 case ir_unop_bitcast_i642d:
399 case ir_unop_bitcast_u642d:
400 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE,
401 op0->type->vector_elements, 1);
402 break;
403
404 case ir_unop_bitcast_d2i64:
405 this->type = glsl_type::get_instance(GLSL_TYPE_INT64,
406 op0->type->vector_elements, 1);
407 break;
408 case ir_unop_bitcast_d2u64:
409 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64,
410 op0->type->vector_elements, 1);
411 break;
412
413 default:
414 assert(!"not reached: missing automatic type setup for ir_expression");
415 this->type = op0->type;
416 break;
417 }
418 }
419
420 ir_expression::ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1)
421 : ir_rvalue(ir_type_expression)
422 {
423 this->operation = ir_expression_operation(op);
424 this->operands[0] = op0;
425 this->operands[1] = op1;
426 this->operands[2] = NULL;
427 this->operands[3] = NULL;
428
429 assert(op > ir_last_unop);
430
431 switch (this->operation) {
432 case ir_binop_all_equal:
433 case ir_binop_any_nequal:
434 this->type = glsl_type::bool_type;
435 break;
436
437 case ir_binop_add:
438 case ir_binop_sub:
439 case ir_binop_min:
440 case ir_binop_max:
441 case ir_binop_pow:
442 case ir_binop_mul:
443 case ir_binop_div:
444 case ir_binop_mod:
445 if (op0->type->is_scalar()) {
446 this->type = op1->type;
447 } else if (op1->type->is_scalar()) {
448 this->type = op0->type;
449 } else {
450 if (this->operation == ir_binop_mul) {
451 this->type = glsl_type::get_mul_type(op0->type, op1->type);
452 } else {
453 assert(op0->type == op1->type);
454 this->type = op0->type;
455 }
456 }
457 break;
458
459 case ir_binop_logic_and:
460 case ir_binop_logic_xor:
461 case ir_binop_logic_or:
462 case ir_binop_bit_and:
463 case ir_binop_bit_xor:
464 case ir_binop_bit_or:
465 assert(!op0->type->is_matrix());
466 assert(!op1->type->is_matrix());
467 if (op0->type->is_scalar()) {
468 this->type = op1->type;
469 } else if (op1->type->is_scalar()) {
470 this->type = op0->type;
471 } else {
472 assert(op0->type->vector_elements == op1->type->vector_elements);
473 this->type = op0->type;
474 }
475 break;
476
477 case ir_binop_equal:
478 case ir_binop_nequal:
479 case ir_binop_lequal:
480 case ir_binop_gequal:
481 case ir_binop_less:
482 case ir_binop_greater:
483 assert(op0->type == op1->type);
484 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL,
485 op0->type->vector_elements, 1);
486 break;
487
488 case ir_binop_dot:
489 this->type = op0->type->get_base_type();
490 break;
491
492 case ir_binop_imul_high:
493 case ir_binop_carry:
494 case ir_binop_borrow:
495 case ir_binop_lshift:
496 case ir_binop_rshift:
497 case ir_binop_ldexp:
498 case ir_binop_interpolate_at_offset:
499 case ir_binop_interpolate_at_sample:
500 this->type = op0->type;
501 break;
502
503 case ir_binop_vector_extract:
504 this->type = op0->type->get_scalar_type();
505 break;
506
507 case ir_binop_read_invocation:
508 this->type = op0->type;
509 break;
510
511 default:
512 assert(!"not reached: missing automatic type setup for ir_expression");
513 this->type = glsl_type::float_type;
514 }
515 }
516
517 ir_expression::ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1,
518 ir_rvalue *op2)
519 : ir_rvalue(ir_type_expression)
520 {
521 this->operation = ir_expression_operation(op);
522 this->operands[0] = op0;
523 this->operands[1] = op1;
524 this->operands[2] = op2;
525 this->operands[3] = NULL;
526
527 assert(op > ir_last_binop && op <= ir_last_triop);
528
529 switch (this->operation) {
530 case ir_triop_fma:
531 case ir_triop_lrp:
532 case ir_triop_bitfield_extract:
533 case ir_triop_vector_insert:
534 this->type = op0->type;
535 break;
536
537 case ir_triop_csel:
538 this->type = op1->type;
539 break;
540
541 default:
542 assert(!"not reached: missing automatic type setup for ir_expression");
543 this->type = glsl_type::float_type;
544 }
545 }
546
547 unsigned int
548 ir_expression::get_num_operands(ir_expression_operation op)
549 {
550 assert(op <= ir_last_opcode);
551
552 if (op <= ir_last_unop)
553 return 1;
554
555 if (op <= ir_last_binop)
556 return 2;
557
558 if (op <= ir_last_triop)
559 return 3;
560
561 if (op <= ir_last_quadop)
562 return 4;
563
564 assert(false);
565 return 0;
566 }
567
568 #include "ir_expression_operation_strings.h"
569
570 const char*
571 depth_layout_string(ir_depth_layout layout)
572 {
573 switch(layout) {
574 case ir_depth_layout_none: return "";
575 case ir_depth_layout_any: return "depth_any";
576 case ir_depth_layout_greater: return "depth_greater";
577 case ir_depth_layout_less: return "depth_less";
578 case ir_depth_layout_unchanged: return "depth_unchanged";
579
580 default:
581 assert(0);
582 return "";
583 }
584 }
585
586 ir_expression_operation
587 ir_expression::get_operator(const char *str)
588 {
589 for (int op = 0; op <= int(ir_last_opcode); op++) {
590 if (strcmp(str, ir_expression_operation_strings[op]) == 0)
591 return (ir_expression_operation) op;
592 }
593 return (ir_expression_operation) -1;
594 }
595
596 ir_variable *
597 ir_expression::variable_referenced() const
598 {
599 switch (operation) {
600 case ir_binop_vector_extract:
601 case ir_triop_vector_insert:
602 /* We get these for things like a[0] where a is a vector type. In these
603 * cases we want variable_referenced() to return the actual vector
604 * variable this is wrapping.
605 */
606 return operands[0]->variable_referenced();
607 default:
608 return ir_rvalue::variable_referenced();
609 }
610 }
611
612 ir_constant::ir_constant()
613 : ir_rvalue(ir_type_constant)
614 {
615 this->array_elements = NULL;
616 }
617
618 ir_constant::ir_constant(const struct glsl_type *type,
619 const ir_constant_data *data)
620 : ir_rvalue(ir_type_constant)
621 {
622 this->array_elements = NULL;
623
624 assert((type->base_type >= GLSL_TYPE_UINT)
625 && (type->base_type <= GLSL_TYPE_BOOL));
626
627 this->type = type;
628 memcpy(& this->value, data, sizeof(this->value));
629 }
630
631 ir_constant::ir_constant(float f, unsigned vector_elements)
632 : ir_rvalue(ir_type_constant)
633 {
634 assert(vector_elements <= 4);
635 this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT, vector_elements, 1);
636 for (unsigned i = 0; i < vector_elements; i++) {
637 this->value.f[i] = f;
638 }
639 for (unsigned i = vector_elements; i < 16; i++) {
640 this->value.f[i] = 0;
641 }
642 }
643
644 ir_constant::ir_constant(double d, unsigned vector_elements)
645 : ir_rvalue(ir_type_constant)
646 {
647 assert(vector_elements <= 4);
648 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE, vector_elements, 1);
649 for (unsigned i = 0; i < vector_elements; i++) {
650 this->value.d[i] = d;
651 }
652 for (unsigned i = vector_elements; i < 16; i++) {
653 this->value.d[i] = 0.0;
654 }
655 }
656
657 ir_constant::ir_constant(unsigned int u, unsigned vector_elements)
658 : ir_rvalue(ir_type_constant)
659 {
660 assert(vector_elements <= 4);
661 this->type = glsl_type::get_instance(GLSL_TYPE_UINT, vector_elements, 1);
662 for (unsigned i = 0; i < vector_elements; i++) {
663 this->value.u[i] = u;
664 }
665 for (unsigned i = vector_elements; i < 16; i++) {
666 this->value.u[i] = 0;
667 }
668 }
669
670 ir_constant::ir_constant(int integer, unsigned vector_elements)
671 : ir_rvalue(ir_type_constant)
672 {
673 assert(vector_elements <= 4);
674 this->type = glsl_type::get_instance(GLSL_TYPE_INT, vector_elements, 1);
675 for (unsigned i = 0; i < vector_elements; i++) {
676 this->value.i[i] = integer;
677 }
678 for (unsigned i = vector_elements; i < 16; i++) {
679 this->value.i[i] = 0;
680 }
681 }
682
683 ir_constant::ir_constant(uint64_t u64, unsigned vector_elements)
684 : ir_rvalue(ir_type_constant)
685 {
686 assert(vector_elements <= 4);
687 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64, vector_elements, 1);
688 for (unsigned i = 0; i < vector_elements; i++) {
689 this->value.u64[i] = u64;
690 }
691 for (unsigned i = vector_elements; i < 16; i++) {
692 this->value.u64[i] = 0;
693 }
694 }
695
696 ir_constant::ir_constant(int64_t int64, unsigned vector_elements)
697 : ir_rvalue(ir_type_constant)
698 {
699 assert(vector_elements <= 4);
700 this->type = glsl_type::get_instance(GLSL_TYPE_INT64, vector_elements, 1);
701 for (unsigned i = 0; i < vector_elements; i++) {
702 this->value.i64[i] = int64;
703 }
704 for (unsigned i = vector_elements; i < 16; i++) {
705 this->value.i64[i] = 0;
706 }
707 }
708
709 ir_constant::ir_constant(bool b, unsigned vector_elements)
710 : ir_rvalue(ir_type_constant)
711 {
712 assert(vector_elements <= 4);
713 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL, vector_elements, 1);
714 for (unsigned i = 0; i < vector_elements; i++) {
715 this->value.b[i] = b;
716 }
717 for (unsigned i = vector_elements; i < 16; i++) {
718 this->value.b[i] = false;
719 }
720 }
721
722 ir_constant::ir_constant(const ir_constant *c, unsigned i)
723 : ir_rvalue(ir_type_constant)
724 {
725 this->array_elements = NULL;
726 this->type = c->type->get_base_type();
727
728 switch (this->type->base_type) {
729 case GLSL_TYPE_UINT: this->value.u[0] = c->value.u[i]; break;
730 case GLSL_TYPE_INT: this->value.i[0] = c->value.i[i]; break;
731 case GLSL_TYPE_FLOAT: this->value.f[0] = c->value.f[i]; break;
732 case GLSL_TYPE_BOOL: this->value.b[0] = c->value.b[i]; break;
733 case GLSL_TYPE_DOUBLE: this->value.d[0] = c->value.d[i]; break;
734 default: assert(!"Should not get here."); break;
735 }
736 }
737
738 ir_constant::ir_constant(const struct glsl_type *type, exec_list *value_list)
739 : ir_rvalue(ir_type_constant)
740 {
741 this->array_elements = NULL;
742 this->type = type;
743
744 assert(type->is_scalar() || type->is_vector() || type->is_matrix()
745 || type->is_record() || type->is_array());
746
747 if (type->is_array()) {
748 this->array_elements = ralloc_array(this, ir_constant *, type->length);
749 unsigned i = 0;
750 foreach_in_list(ir_constant, value, value_list) {
751 assert(value->as_constant() != NULL);
752
753 this->array_elements[i++] = value;
754 }
755 return;
756 }
757
758 /* If the constant is a record, the types of each of the entries in
759 * value_list must be a 1-for-1 match with the structure components. Each
760 * entry must also be a constant. Just move the nodes from the value_list
761 * to the list in the ir_constant.
762 */
763 /* FINISHME: Should there be some type checking and / or assertions here? */
764 /* FINISHME: Should the new constant take ownership of the nodes from
765 * FINISHME: value_list, or should it make copies?
766 */
767 if (type->is_record()) {
768 value_list->move_nodes_to(& this->components);
769 return;
770 }
771
772 for (unsigned i = 0; i < 16; i++) {
773 this->value.u[i] = 0;
774 }
775
776 ir_constant *value = (ir_constant *) (value_list->get_head_raw());
777
778 /* Constructors with exactly one scalar argument are special for vectors
779 * and matrices. For vectors, the scalar value is replicated to fill all
780 * the components. For matrices, the scalar fills the components of the
781 * diagonal while the rest is filled with 0.
782 */
783 if (value->type->is_scalar() && value->next->is_tail_sentinel()) {
784 if (type->is_matrix()) {
785 /* Matrix - fill diagonal (rest is already set to 0) */
786 assert(type->base_type == GLSL_TYPE_FLOAT ||
787 type->is_double());
788 for (unsigned i = 0; i < type->matrix_columns; i++) {
789 if (type->base_type == GLSL_TYPE_FLOAT)
790 this->value.f[i * type->vector_elements + i] =
791 value->value.f[0];
792 else
793 this->value.d[i * type->vector_elements + i] =
794 value->value.d[0];
795 }
796 } else {
797 /* Vector or scalar - fill all components */
798 switch (type->base_type) {
799 case GLSL_TYPE_UINT:
800 case GLSL_TYPE_INT:
801 for (unsigned i = 0; i < type->components(); i++)
802 this->value.u[i] = value->value.u[0];
803 break;
804 case GLSL_TYPE_FLOAT:
805 for (unsigned i = 0; i < type->components(); i++)
806 this->value.f[i] = value->value.f[0];
807 break;
808 case GLSL_TYPE_DOUBLE:
809 for (unsigned i = 0; i < type->components(); i++)
810 this->value.d[i] = value->value.d[0];
811 break;
812 case GLSL_TYPE_UINT64:
813 case GLSL_TYPE_INT64:
814 for (unsigned i = 0; i < type->components(); i++)
815 this->value.u64[i] = value->value.u64[0];
816 break;
817 case GLSL_TYPE_BOOL:
818 for (unsigned i = 0; i < type->components(); i++)
819 this->value.b[i] = value->value.b[0];
820 break;
821 default:
822 assert(!"Should not get here.");
823 break;
824 }
825 }
826 return;
827 }
828
829 if (type->is_matrix() && value->type->is_matrix()) {
830 assert(value->next->is_tail_sentinel());
831
832 /* From section 5.4.2 of the GLSL 1.20 spec:
833 * "If a matrix is constructed from a matrix, then each component
834 * (column i, row j) in the result that has a corresponding component
835 * (column i, row j) in the argument will be initialized from there."
836 */
837 unsigned cols = MIN2(type->matrix_columns, value->type->matrix_columns);
838 unsigned rows = MIN2(type->vector_elements, value->type->vector_elements);
839 for (unsigned i = 0; i < cols; i++) {
840 for (unsigned j = 0; j < rows; j++) {
841 const unsigned src = i * value->type->vector_elements + j;
842 const unsigned dst = i * type->vector_elements + j;
843 this->value.f[dst] = value->value.f[src];
844 }
845 }
846
847 /* "All other components will be initialized to the identity matrix." */
848 for (unsigned i = cols; i < type->matrix_columns; i++)
849 this->value.f[i * type->vector_elements + i] = 1.0;
850
851 return;
852 }
853
854 /* Use each component from each entry in the value_list to initialize one
855 * component of the constant being constructed.
856 */
857 unsigned i = 0;
858 for (;;) {
859 assert(value->as_constant() != NULL);
860 assert(!value->is_tail_sentinel());
861
862 for (unsigned j = 0; j < value->type->components(); j++) {
863 switch (type->base_type) {
864 case GLSL_TYPE_UINT:
865 this->value.u[i] = value->get_uint_component(j);
866 break;
867 case GLSL_TYPE_INT:
868 this->value.i[i] = value->get_int_component(j);
869 break;
870 case GLSL_TYPE_FLOAT:
871 this->value.f[i] = value->get_float_component(j);
872 break;
873 case GLSL_TYPE_BOOL:
874 this->value.b[i] = value->get_bool_component(j);
875 break;
876 case GLSL_TYPE_DOUBLE:
877 this->value.d[i] = value->get_double_component(j);
878 break;
879 case GLSL_TYPE_UINT64:
880 this->value.u64[i] = value->get_uint64_component(j);
881 break;
882 case GLSL_TYPE_INT64:
883 this->value.i64[i] = value->get_int64_component(j);
884 break;
885 default:
886 /* FINISHME: What to do? Exceptions are not the answer.
887 */
888 break;
889 }
890
891 i++;
892 if (i >= type->components())
893 break;
894 }
895
896 if (i >= type->components())
897 break; /* avoid downcasting a list sentinel */
898 value = (ir_constant *) value->next;
899 }
900 }
901
902 ir_constant *
903 ir_constant::zero(void *mem_ctx, const glsl_type *type)
904 {
905 assert(type->is_scalar() || type->is_vector() || type->is_matrix()
906 || type->is_record() || type->is_array());
907
908 ir_constant *c = new(mem_ctx) ir_constant;
909 c->type = type;
910 memset(&c->value, 0, sizeof(c->value));
911
912 if (type->is_array()) {
913 c->array_elements = ralloc_array(c, ir_constant *, type->length);
914
915 for (unsigned i = 0; i < type->length; i++)
916 c->array_elements[i] = ir_constant::zero(c, type->fields.array);
917 }
918
919 if (type->is_record()) {
920 for (unsigned i = 0; i < type->length; i++) {
921 ir_constant *comp = ir_constant::zero(mem_ctx, type->fields.structure[i].type);
922 c->components.push_tail(comp);
923 }
924 }
925
926 return c;
927 }
928
929 bool
930 ir_constant::get_bool_component(unsigned i) const
931 {
932 switch (this->type->base_type) {
933 case GLSL_TYPE_UINT: return this->value.u[i] != 0;
934 case GLSL_TYPE_INT: return this->value.i[i] != 0;
935 case GLSL_TYPE_FLOAT: return ((int)this->value.f[i]) != 0;
936 case GLSL_TYPE_BOOL: return this->value.b[i];
937 case GLSL_TYPE_DOUBLE: return this->value.d[i] != 0.0;
938 case GLSL_TYPE_UINT64: return this->value.u64[i] != 0;
939 case GLSL_TYPE_INT64: return this->value.i64[i] != 0;
940 default: assert(!"Should not get here."); break;
941 }
942
943 /* Must return something to make the compiler happy. This is clearly an
944 * error case.
945 */
946 return false;
947 }
948
949 float
950 ir_constant::get_float_component(unsigned i) const
951 {
952 switch (this->type->base_type) {
953 case GLSL_TYPE_UINT: return (float) this->value.u[i];
954 case GLSL_TYPE_INT: return (float) this->value.i[i];
955 case GLSL_TYPE_FLOAT: return this->value.f[i];
956 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1.0f : 0.0f;
957 case GLSL_TYPE_DOUBLE: return (float) this->value.d[i];
958 case GLSL_TYPE_UINT64: return (float) this->value.u64[i];
959 case GLSL_TYPE_INT64: return (float) this->value.i64[i];
960 default: assert(!"Should not get here."); break;
961 }
962
963 /* Must return something to make the compiler happy. This is clearly an
964 * error case.
965 */
966 return 0.0;
967 }
968
969 double
970 ir_constant::get_double_component(unsigned i) const
971 {
972 switch (this->type->base_type) {
973 case GLSL_TYPE_UINT: return (double) this->value.u[i];
974 case GLSL_TYPE_INT: return (double) this->value.i[i];
975 case GLSL_TYPE_FLOAT: return (double) this->value.f[i];
976 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1.0 : 0.0;
977 case GLSL_TYPE_DOUBLE: return this->value.d[i];
978 case GLSL_TYPE_UINT64: return (double) this->value.u64[i];
979 case GLSL_TYPE_INT64: return (double) this->value.i64[i];
980 default: assert(!"Should not get here."); break;
981 }
982
983 /* Must return something to make the compiler happy. This is clearly an
984 * error case.
985 */
986 return 0.0;
987 }
988
989 int
990 ir_constant::get_int_component(unsigned i) const
991 {
992 switch (this->type->base_type) {
993 case GLSL_TYPE_UINT: return this->value.u[i];
994 case GLSL_TYPE_INT: return this->value.i[i];
995 case GLSL_TYPE_FLOAT: return (int) this->value.f[i];
996 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
997 case GLSL_TYPE_DOUBLE: return (int) this->value.d[i];
998 case GLSL_TYPE_UINT64: return (int) this->value.u64[i];
999 case GLSL_TYPE_INT64: return (int) this->value.i64[i];
1000 default: assert(!"Should not get here."); break;
1001 }
1002
1003 /* Must return something to make the compiler happy. This is clearly an
1004 * error case.
1005 */
1006 return 0;
1007 }
1008
1009 unsigned
1010 ir_constant::get_uint_component(unsigned i) const
1011 {
1012 switch (this->type->base_type) {
1013 case GLSL_TYPE_UINT: return this->value.u[i];
1014 case GLSL_TYPE_INT: return this->value.i[i];
1015 case GLSL_TYPE_FLOAT: return (unsigned) this->value.f[i];
1016 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1017 case GLSL_TYPE_DOUBLE: return (unsigned) this->value.d[i];
1018 case GLSL_TYPE_UINT64: return (unsigned) this->value.u64[i];
1019 case GLSL_TYPE_INT64: return (unsigned) this->value.i64[i];
1020 default: assert(!"Should not get here."); break;
1021 }
1022
1023 /* Must return something to make the compiler happy. This is clearly an
1024 * error case.
1025 */
1026 return 0;
1027 }
1028
1029 int64_t
1030 ir_constant::get_int64_component(unsigned i) const
1031 {
1032 switch (this->type->base_type) {
1033 case GLSL_TYPE_UINT: return this->value.u[i];
1034 case GLSL_TYPE_INT: return this->value.i[i];
1035 case GLSL_TYPE_FLOAT: return (int64_t) this->value.f[i];
1036 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1037 case GLSL_TYPE_DOUBLE: return (int64_t) this->value.d[i];
1038 case GLSL_TYPE_UINT64: return (int64_t) this->value.u64[i];
1039 case GLSL_TYPE_INT64: return 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 uint64_t
1050 ir_constant::get_uint64_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 (uint64_t) this->value.f[i];
1056 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1057 case GLSL_TYPE_DOUBLE: return (uint64_t) this->value.d[i];
1058 case GLSL_TYPE_UINT64: return this->value.u64[i];
1059 case GLSL_TYPE_INT64: return (uint64_t) this->value.i64[i];
1060 default: assert(!"Should not get here."); break;
1061 }
1062
1063 /* Must return something to make the compiler happy. This is clearly an
1064 * error case.
1065 */
1066 return 0;
1067 }
1068
1069 ir_constant *
1070 ir_constant::get_array_element(unsigned i) const
1071 {
1072 assert(this->type->is_array());
1073
1074 /* From page 35 (page 41 of the PDF) of the GLSL 1.20 spec:
1075 *
1076 * "Behavior is undefined if a shader subscripts an array with an index
1077 * less than 0 or greater than or equal to the size the array was
1078 * declared with."
1079 *
1080 * Most out-of-bounds accesses are removed before things could get this far.
1081 * There are cases where non-constant array index values can get constant
1082 * folded.
1083 */
1084 if (int(i) < 0)
1085 i = 0;
1086 else if (i >= this->type->length)
1087 i = this->type->length - 1;
1088
1089 return array_elements[i];
1090 }
1091
1092 ir_constant *
1093 ir_constant::get_record_field(const char *name)
1094 {
1095 int idx = this->type->field_index(name);
1096
1097 if (idx < 0)
1098 return NULL;
1099
1100 if (this->components.is_empty())
1101 return NULL;
1102
1103 exec_node *node = this->components.get_head_raw();
1104 for (int i = 0; i < idx; i++) {
1105 node = node->next;
1106
1107 /* If the end of the list is encountered before the element matching the
1108 * requested field is found, return NULL.
1109 */
1110 if (node->is_tail_sentinel())
1111 return NULL;
1112 }
1113
1114 return (ir_constant *) node;
1115 }
1116
1117 void
1118 ir_constant::copy_offset(ir_constant *src, int offset)
1119 {
1120 switch (this->type->base_type) {
1121 case GLSL_TYPE_UINT:
1122 case GLSL_TYPE_INT:
1123 case GLSL_TYPE_FLOAT:
1124 case GLSL_TYPE_DOUBLE:
1125 case GLSL_TYPE_UINT64:
1126 case GLSL_TYPE_INT64:
1127 case GLSL_TYPE_BOOL: {
1128 unsigned int size = src->type->components();
1129 assert (size <= this->type->components() - offset);
1130 for (unsigned int i=0; i<size; i++) {
1131 switch (this->type->base_type) {
1132 case GLSL_TYPE_UINT:
1133 value.u[i+offset] = src->get_uint_component(i);
1134 break;
1135 case GLSL_TYPE_INT:
1136 value.i[i+offset] = src->get_int_component(i);
1137 break;
1138 case GLSL_TYPE_FLOAT:
1139 value.f[i+offset] = src->get_float_component(i);
1140 break;
1141 case GLSL_TYPE_BOOL:
1142 value.b[i+offset] = src->get_bool_component(i);
1143 break;
1144 case GLSL_TYPE_DOUBLE:
1145 value.d[i+offset] = src->get_double_component(i);
1146 break;
1147 case GLSL_TYPE_UINT64:
1148 value.u64[i+offset] = src->get_uint64_component(i);
1149 break;
1150 case GLSL_TYPE_INT64:
1151 value.i64[i+offset] = src->get_int64_component(i);
1152 break;
1153 default: // Shut up the compiler
1154 break;
1155 }
1156 }
1157 break;
1158 }
1159
1160 case GLSL_TYPE_STRUCT: {
1161 assert (src->type == this->type);
1162 this->components.make_empty();
1163 foreach_in_list(ir_constant, orig, &src->components) {
1164 this->components.push_tail(orig->clone(this, NULL));
1165 }
1166 break;
1167 }
1168
1169 case GLSL_TYPE_ARRAY: {
1170 assert (src->type == this->type);
1171 for (unsigned i = 0; i < this->type->length; i++) {
1172 this->array_elements[i] = src->array_elements[i]->clone(this, NULL);
1173 }
1174 break;
1175 }
1176
1177 default:
1178 assert(!"Should not get here.");
1179 break;
1180 }
1181 }
1182
1183 void
1184 ir_constant::copy_masked_offset(ir_constant *src, int offset, unsigned int mask)
1185 {
1186 assert (!type->is_array() && !type->is_record());
1187
1188 if (!type->is_vector() && !type->is_matrix()) {
1189 offset = 0;
1190 mask = 1;
1191 }
1192
1193 int id = 0;
1194 for (int i=0; i<4; i++) {
1195 if (mask & (1 << i)) {
1196 switch (this->type->base_type) {
1197 case GLSL_TYPE_UINT:
1198 value.u[i+offset] = src->get_uint_component(id++);
1199 break;
1200 case GLSL_TYPE_INT:
1201 value.i[i+offset] = src->get_int_component(id++);
1202 break;
1203 case GLSL_TYPE_FLOAT:
1204 value.f[i+offset] = src->get_float_component(id++);
1205 break;
1206 case GLSL_TYPE_BOOL:
1207 value.b[i+offset] = src->get_bool_component(id++);
1208 break;
1209 case GLSL_TYPE_DOUBLE:
1210 value.d[i+offset] = src->get_double_component(id++);
1211 break;
1212 case GLSL_TYPE_UINT64:
1213 value.u64[i+offset] = src->get_uint64_component(id++);
1214 break;
1215 case GLSL_TYPE_INT64:
1216 value.i64[i+offset] = src->get_int64_component(id++);
1217 break;
1218 default:
1219 assert(!"Should not get here.");
1220 return;
1221 }
1222 }
1223 }
1224 }
1225
1226 bool
1227 ir_constant::has_value(const ir_constant *c) const
1228 {
1229 if (this->type != c->type)
1230 return false;
1231
1232 if (this->type->is_array()) {
1233 for (unsigned i = 0; i < this->type->length; i++) {
1234 if (!this->array_elements[i]->has_value(c->array_elements[i]))
1235 return false;
1236 }
1237 return true;
1238 }
1239
1240 if (this->type->is_record()) {
1241 const exec_node *a_node = this->components.get_head_raw();
1242 const exec_node *b_node = c->components.get_head_raw();
1243
1244 while (!a_node->is_tail_sentinel()) {
1245 assert(!b_node->is_tail_sentinel());
1246
1247 const ir_constant *const a_field = (ir_constant *) a_node;
1248 const ir_constant *const b_field = (ir_constant *) b_node;
1249
1250 if (!a_field->has_value(b_field))
1251 return false;
1252
1253 a_node = a_node->next;
1254 b_node = b_node->next;
1255 }
1256
1257 return true;
1258 }
1259
1260 for (unsigned i = 0; i < this->type->components(); i++) {
1261 switch (this->type->base_type) {
1262 case GLSL_TYPE_UINT:
1263 if (this->value.u[i] != c->value.u[i])
1264 return false;
1265 break;
1266 case GLSL_TYPE_INT:
1267 if (this->value.i[i] != c->value.i[i])
1268 return false;
1269 break;
1270 case GLSL_TYPE_FLOAT:
1271 if (this->value.f[i] != c->value.f[i])
1272 return false;
1273 break;
1274 case GLSL_TYPE_BOOL:
1275 if (this->value.b[i] != c->value.b[i])
1276 return false;
1277 break;
1278 case GLSL_TYPE_DOUBLE:
1279 if (this->value.d[i] != c->value.d[i])
1280 return false;
1281 break;
1282 case GLSL_TYPE_UINT64:
1283 if (this->value.u64[i] != c->value.u64[i])
1284 return false;
1285 break;
1286 case GLSL_TYPE_INT64:
1287 if (this->value.i64[i] != c->value.i64[i])
1288 return false;
1289 break;
1290 default:
1291 assert(!"Should not get here.");
1292 return false;
1293 }
1294 }
1295
1296 return true;
1297 }
1298
1299 bool
1300 ir_constant::is_value(float f, int i) const
1301 {
1302 if (!this->type->is_scalar() && !this->type->is_vector())
1303 return false;
1304
1305 /* Only accept boolean values for 0/1. */
1306 if (int(bool(i)) != i && this->type->is_boolean())
1307 return false;
1308
1309 for (unsigned c = 0; c < this->type->vector_elements; c++) {
1310 switch (this->type->base_type) {
1311 case GLSL_TYPE_FLOAT:
1312 if (this->value.f[c] != f)
1313 return false;
1314 break;
1315 case GLSL_TYPE_INT:
1316 if (this->value.i[c] != i)
1317 return false;
1318 break;
1319 case GLSL_TYPE_UINT:
1320 if (this->value.u[c] != unsigned(i))
1321 return false;
1322 break;
1323 case GLSL_TYPE_BOOL:
1324 if (this->value.b[c] != bool(i))
1325 return false;
1326 break;
1327 case GLSL_TYPE_DOUBLE:
1328 if (this->value.d[c] != double(f))
1329 return false;
1330 break;
1331 case GLSL_TYPE_UINT64:
1332 if (this->value.u64[c] != uint64_t(i))
1333 return false;
1334 break;
1335 case GLSL_TYPE_INT64:
1336 if (this->value.i64[c] != i)
1337 return false;
1338 break;
1339 default:
1340 /* The only other base types are structures, arrays, and samplers.
1341 * Samplers cannot be constants, and the others should have been
1342 * filtered out above.
1343 */
1344 assert(!"Should not get here.");
1345 return false;
1346 }
1347 }
1348
1349 return true;
1350 }
1351
1352 bool
1353 ir_constant::is_zero() const
1354 {
1355 return is_value(0.0, 0);
1356 }
1357
1358 bool
1359 ir_constant::is_one() const
1360 {
1361 return is_value(1.0, 1);
1362 }
1363
1364 bool
1365 ir_constant::is_negative_one() const
1366 {
1367 return is_value(-1.0, -1);
1368 }
1369
1370 bool
1371 ir_constant::is_uint16_constant() const
1372 {
1373 if (!type->is_integer())
1374 return false;
1375
1376 return value.u[0] < (1 << 16);
1377 }
1378
1379 ir_loop::ir_loop()
1380 : ir_instruction(ir_type_loop)
1381 {
1382 }
1383
1384
1385 ir_dereference_variable::ir_dereference_variable(ir_variable *var)
1386 : ir_dereference(ir_type_dereference_variable)
1387 {
1388 assert(var != NULL);
1389
1390 this->var = var;
1391 this->type = var->type;
1392 }
1393
1394
1395 ir_dereference_array::ir_dereference_array(ir_rvalue *value,
1396 ir_rvalue *array_index)
1397 : ir_dereference(ir_type_dereference_array)
1398 {
1399 this->array_index = array_index;
1400 this->set_array(value);
1401 }
1402
1403
1404 ir_dereference_array::ir_dereference_array(ir_variable *var,
1405 ir_rvalue *array_index)
1406 : ir_dereference(ir_type_dereference_array)
1407 {
1408 void *ctx = ralloc_parent(var);
1409
1410 this->array_index = array_index;
1411 this->set_array(new(ctx) ir_dereference_variable(var));
1412 }
1413
1414
1415 void
1416 ir_dereference_array::set_array(ir_rvalue *value)
1417 {
1418 assert(value != NULL);
1419
1420 this->array = value;
1421
1422 const glsl_type *const vt = this->array->type;
1423
1424 if (vt->is_array()) {
1425 type = vt->fields.array;
1426 } else if (vt->is_matrix()) {
1427 type = vt->column_type();
1428 } else if (vt->is_vector()) {
1429 type = vt->get_base_type();
1430 }
1431 }
1432
1433
1434 ir_dereference_record::ir_dereference_record(ir_rvalue *value,
1435 const char *field)
1436 : ir_dereference(ir_type_dereference_record)
1437 {
1438 assert(value != NULL);
1439
1440 this->record = value;
1441 this->field = ralloc_strdup(this, field);
1442 this->type = this->record->type->field_type(field);
1443 }
1444
1445
1446 ir_dereference_record::ir_dereference_record(ir_variable *var,
1447 const char *field)
1448 : ir_dereference(ir_type_dereference_record)
1449 {
1450 void *ctx = ralloc_parent(var);
1451
1452 this->record = new(ctx) ir_dereference_variable(var);
1453 this->field = ralloc_strdup(this, field);
1454 this->type = this->record->type->field_type(field);
1455 }
1456
1457 bool
1458 ir_dereference::is_lvalue() const
1459 {
1460 ir_variable *var = this->variable_referenced();
1461
1462 /* Every l-value derference chain eventually ends in a variable.
1463 */
1464 if ((var == NULL) || var->data.read_only)
1465 return false;
1466
1467 /* From section 4.1.7 of the GLSL 4.40 spec:
1468 *
1469 * "Opaque variables cannot be treated as l-values; hence cannot
1470 * be used as out or inout function parameters, nor can they be
1471 * assigned into."
1472 */
1473 if (this->type->contains_opaque())
1474 return false;
1475
1476 return true;
1477 }
1478
1479
1480 static const char * const tex_opcode_strs[] = { "tex", "txb", "txl", "txd", "txf", "txf_ms", "txs", "lod", "tg4", "query_levels", "texture_samples", "samples_identical" };
1481
1482 const char *ir_texture::opcode_string()
1483 {
1484 assert((unsigned int) op < ARRAY_SIZE(tex_opcode_strs));
1485 return tex_opcode_strs[op];
1486 }
1487
1488 ir_texture_opcode
1489 ir_texture::get_opcode(const char *str)
1490 {
1491 const int count = sizeof(tex_opcode_strs) / sizeof(tex_opcode_strs[0]);
1492 for (int op = 0; op < count; op++) {
1493 if (strcmp(str, tex_opcode_strs[op]) == 0)
1494 return (ir_texture_opcode) op;
1495 }
1496 return (ir_texture_opcode) -1;
1497 }
1498
1499
1500 void
1501 ir_texture::set_sampler(ir_dereference *sampler, const glsl_type *type)
1502 {
1503 assert(sampler != NULL);
1504 assert(type != NULL);
1505 this->sampler = sampler;
1506 this->type = type;
1507
1508 if (this->op == ir_txs || this->op == ir_query_levels ||
1509 this->op == ir_texture_samples) {
1510 assert(type->base_type == GLSL_TYPE_INT);
1511 } else if (this->op == ir_lod) {
1512 assert(type->vector_elements == 2);
1513 assert(type->base_type == GLSL_TYPE_FLOAT);
1514 } else if (this->op == ir_samples_identical) {
1515 assert(type == glsl_type::bool_type);
1516 assert(sampler->type->is_sampler());
1517 assert(sampler->type->sampler_dimensionality == GLSL_SAMPLER_DIM_MS);
1518 } else {
1519 assert(sampler->type->sampled_type == (int) type->base_type);
1520 if (sampler->type->sampler_shadow)
1521 assert(type->vector_elements == 4 || type->vector_elements == 1);
1522 else
1523 assert(type->vector_elements == 4);
1524 }
1525 }
1526
1527
1528 void
1529 ir_swizzle::init_mask(const unsigned *comp, unsigned count)
1530 {
1531 assert((count >= 1) && (count <= 4));
1532
1533 memset(&this->mask, 0, sizeof(this->mask));
1534 this->mask.num_components = count;
1535
1536 unsigned dup_mask = 0;
1537 switch (count) {
1538 case 4:
1539 assert(comp[3] <= 3);
1540 dup_mask |= (1U << comp[3])
1541 & ((1U << comp[0]) | (1U << comp[1]) | (1U << comp[2]));
1542 this->mask.w = comp[3];
1543
1544 case 3:
1545 assert(comp[2] <= 3);
1546 dup_mask |= (1U << comp[2])
1547 & ((1U << comp[0]) | (1U << comp[1]));
1548 this->mask.z = comp[2];
1549
1550 case 2:
1551 assert(comp[1] <= 3);
1552 dup_mask |= (1U << comp[1])
1553 & ((1U << comp[0]));
1554 this->mask.y = comp[1];
1555
1556 case 1:
1557 assert(comp[0] <= 3);
1558 this->mask.x = comp[0];
1559 }
1560
1561 this->mask.has_duplicates = dup_mask != 0;
1562
1563 /* Based on the number of elements in the swizzle and the base type
1564 * (i.e., float, int, unsigned, or bool) of the vector being swizzled,
1565 * generate the type of the resulting value.
1566 */
1567 type = glsl_type::get_instance(val->type->base_type, mask.num_components, 1);
1568 }
1569
1570 ir_swizzle::ir_swizzle(ir_rvalue *val, unsigned x, unsigned y, unsigned z,
1571 unsigned w, unsigned count)
1572 : ir_rvalue(ir_type_swizzle), val(val)
1573 {
1574 const unsigned components[4] = { x, y, z, w };
1575 this->init_mask(components, count);
1576 }
1577
1578 ir_swizzle::ir_swizzle(ir_rvalue *val, const unsigned *comp,
1579 unsigned count)
1580 : ir_rvalue(ir_type_swizzle), val(val)
1581 {
1582 this->init_mask(comp, count);
1583 }
1584
1585 ir_swizzle::ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask)
1586 : ir_rvalue(ir_type_swizzle)
1587 {
1588 this->val = val;
1589 this->mask = mask;
1590 this->type = glsl_type::get_instance(val->type->base_type,
1591 mask.num_components, 1);
1592 }
1593
1594 #define X 1
1595 #define R 5
1596 #define S 9
1597 #define I 13
1598
1599 ir_swizzle *
1600 ir_swizzle::create(ir_rvalue *val, const char *str, unsigned vector_length)
1601 {
1602 void *ctx = ralloc_parent(val);
1603
1604 /* For each possible swizzle character, this table encodes the value in
1605 * \c idx_map that represents the 0th element of the vector. For invalid
1606 * swizzle characters (e.g., 'k'), a special value is used that will allow
1607 * detection of errors.
1608 */
1609 static const unsigned char base_idx[26] = {
1610 /* a b c d e f g h i j k l m */
1611 R, R, I, I, I, I, R, I, I, I, I, I, I,
1612 /* n o p q r s t u v w x y z */
1613 I, I, S, S, R, S, S, I, I, X, X, X, X
1614 };
1615
1616 /* Each valid swizzle character has an entry in the previous table. This
1617 * table encodes the base index encoded in the previous table plus the actual
1618 * index of the swizzle character. When processing swizzles, the first
1619 * character in the string is indexed in the previous table. Each character
1620 * in the string is indexed in this table, and the value found there has the
1621 * value form the first table subtracted. The result must be on the range
1622 * [0,3].
1623 *
1624 * For example, the string "wzyx" will get X from the first table. Each of
1625 * the charcaters will get X+3, X+2, X+1, and X+0 from this table. After
1626 * subtraction, the swizzle values are { 3, 2, 1, 0 }.
1627 *
1628 * The string "wzrg" will get X from the first table. Each of the characters
1629 * will get X+3, X+2, R+0, and R+1 from this table. After subtraction, the
1630 * swizzle values are { 3, 2, 4, 5 }. Since 4 and 5 are outside the range
1631 * [0,3], the error is detected.
1632 */
1633 static const unsigned char idx_map[26] = {
1634 /* a b c d e f g h i j k l m */
1635 R+3, R+2, 0, 0, 0, 0, R+1, 0, 0, 0, 0, 0, 0,
1636 /* n o p q r s t u v w x y z */
1637 0, 0, S+2, S+3, R+0, S+0, S+1, 0, 0, X+3, X+0, X+1, X+2
1638 };
1639
1640 int swiz_idx[4] = { 0, 0, 0, 0 };
1641 unsigned i;
1642
1643
1644 /* Validate the first character in the swizzle string and look up the base
1645 * index value as described above.
1646 */
1647 if ((str[0] < 'a') || (str[0] > 'z'))
1648 return NULL;
1649
1650 const unsigned base = base_idx[str[0] - 'a'];
1651
1652
1653 for (i = 0; (i < 4) && (str[i] != '\0'); i++) {
1654 /* Validate the next character, and, as described above, convert it to a
1655 * swizzle index.
1656 */
1657 if ((str[i] < 'a') || (str[i] > 'z'))
1658 return NULL;
1659
1660 swiz_idx[i] = idx_map[str[i] - 'a'] - base;
1661 if ((swiz_idx[i] < 0) || (swiz_idx[i] >= (int) vector_length))
1662 return NULL;
1663 }
1664
1665 if (str[i] != '\0')
1666 return NULL;
1667
1668 return new(ctx) ir_swizzle(val, swiz_idx[0], swiz_idx[1], swiz_idx[2],
1669 swiz_idx[3], i);
1670 }
1671
1672 #undef X
1673 #undef R
1674 #undef S
1675 #undef I
1676
1677 ir_variable *
1678 ir_swizzle::variable_referenced() const
1679 {
1680 return this->val->variable_referenced();
1681 }
1682
1683
1684 bool ir_variable::temporaries_allocate_names = false;
1685
1686 const char ir_variable::tmp_name[] = "compiler_temp";
1687
1688 ir_variable::ir_variable(const struct glsl_type *type, const char *name,
1689 ir_variable_mode mode)
1690 : ir_instruction(ir_type_variable)
1691 {
1692 this->type = type;
1693
1694 if (mode == ir_var_temporary && !ir_variable::temporaries_allocate_names)
1695 name = NULL;
1696
1697 /* The ir_variable clone method may call this constructor with name set to
1698 * tmp_name.
1699 */
1700 assert(name != NULL
1701 || mode == ir_var_temporary
1702 || mode == ir_var_function_in
1703 || mode == ir_var_function_out
1704 || mode == ir_var_function_inout);
1705 assert(name != ir_variable::tmp_name
1706 || mode == ir_var_temporary);
1707 if (mode == ir_var_temporary
1708 && (name == NULL || name == ir_variable::tmp_name)) {
1709 this->name = ir_variable::tmp_name;
1710 } else if (name == NULL ||
1711 strlen(name) < ARRAY_SIZE(this->name_storage)) {
1712 strcpy(this->name_storage, name ? name : "");
1713 this->name = this->name_storage;
1714 } else {
1715 this->name = ralloc_strdup(this, name);
1716 }
1717
1718 this->u.max_ifc_array_access = NULL;
1719
1720 this->data.explicit_location = false;
1721 this->data.has_initializer = false;
1722 this->data.location = -1;
1723 this->data.location_frac = 0;
1724 this->data.binding = 0;
1725 this->data.warn_extension_index = 0;
1726 this->constant_value = NULL;
1727 this->constant_initializer = NULL;
1728 this->data.origin_upper_left = false;
1729 this->data.pixel_center_integer = false;
1730 this->data.depth_layout = ir_depth_layout_none;
1731 this->data.used = false;
1732 this->data.always_active_io = false;
1733 this->data.read_only = false;
1734 this->data.centroid = false;
1735 this->data.sample = false;
1736 this->data.patch = false;
1737 this->data.invariant = false;
1738 this->data.how_declared = ir_var_declared_normally;
1739 this->data.mode = mode;
1740 this->data.interpolation = INTERP_MODE_NONE;
1741 this->data.max_array_access = -1;
1742 this->data.offset = 0;
1743 this->data.precision = GLSL_PRECISION_NONE;
1744 this->data.image_read_only = false;
1745 this->data.image_write_only = false;
1746 this->data.image_coherent = false;
1747 this->data.image_volatile = false;
1748 this->data.image_restrict = false;
1749 this->data.from_ssbo_unsized_array = false;
1750 this->data.fb_fetch_output = false;
1751
1752 if (type != NULL) {
1753 if (type->is_sampler())
1754 this->data.read_only = true;
1755
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 intrinsic_id(ir_intrinsic_invalid), builtin_avail(b), _function(NULL)
1810 {
1811 this->origin = NULL;
1812 }
1813
1814
1815 bool
1816 ir_function_signature::is_builtin() const
1817 {
1818 return builtin_avail != NULL;
1819 }
1820
1821
1822 bool
1823 ir_function_signature::is_builtin_available(const _mesa_glsl_parse_state *state) const
1824 {
1825 /* We can't call the predicate without a state pointer, so just say that
1826 * the signature is available. At compile time, we need the filtering,
1827 * but also receive a valid state pointer. At link time, we're resolving
1828 * imported built-in prototypes to their definitions, which will always
1829 * be an exact match. So we can skip the filtering.
1830 */
1831 if (state == NULL)
1832 return true;
1833
1834 assert(builtin_avail != NULL);
1835 return builtin_avail(state);
1836 }
1837
1838
1839 static bool
1840 modes_match(unsigned a, unsigned b)
1841 {
1842 if (a == b)
1843 return true;
1844
1845 /* Accept "in" vs. "const in" */
1846 if ((a == ir_var_const_in && b == ir_var_function_in) ||
1847 (b == ir_var_const_in && a == ir_var_function_in))
1848 return true;
1849
1850 return false;
1851 }
1852
1853
1854 const char *
1855 ir_function_signature::qualifiers_match(exec_list *params)
1856 {
1857 /* check that the qualifiers match. */
1858 foreach_two_lists(a_node, &this->parameters, b_node, params) {
1859 ir_variable *a = (ir_variable *) a_node;
1860 ir_variable *b = (ir_variable *) b_node;
1861
1862 if (a->data.read_only != b->data.read_only ||
1863 !modes_match(a->data.mode, b->data.mode) ||
1864 a->data.interpolation != b->data.interpolation ||
1865 a->data.centroid != b->data.centroid ||
1866 a->data.sample != b->data.sample ||
1867 a->data.patch != b->data.patch ||
1868 a->data.image_read_only != b->data.image_read_only ||
1869 a->data.image_write_only != b->data.image_write_only ||
1870 a->data.image_coherent != b->data.image_coherent ||
1871 a->data.image_volatile != b->data.image_volatile ||
1872 a->data.image_restrict != b->data.image_restrict) {
1873
1874 /* parameter a's qualifiers don't match */
1875 return a->name;
1876 }
1877 }
1878 return NULL;
1879 }
1880
1881
1882 void
1883 ir_function_signature::replace_parameters(exec_list *new_params)
1884 {
1885 /* Destroy all of the previous parameter information. If the previous
1886 * parameter information comes from the function prototype, it may either
1887 * specify incorrect parameter names or not have names at all.
1888 */
1889 new_params->move_nodes_to(&parameters);
1890 }
1891
1892
1893 ir_function::ir_function(const char *name)
1894 : ir_instruction(ir_type_function)
1895 {
1896 this->subroutine_index = -1;
1897 this->name = ralloc_strdup(this, name);
1898 }
1899
1900
1901 bool
1902 ir_function::has_user_signature()
1903 {
1904 foreach_in_list(ir_function_signature, sig, &this->signatures) {
1905 if (!sig->is_builtin())
1906 return true;
1907 }
1908 return false;
1909 }
1910
1911
1912 ir_rvalue *
1913 ir_rvalue::error_value(void *mem_ctx)
1914 {
1915 ir_rvalue *v = new(mem_ctx) ir_rvalue(ir_type_unset);
1916
1917 v->type = glsl_type::error_type;
1918 return v;
1919 }
1920
1921
1922 void
1923 visit_exec_list(exec_list *list, ir_visitor *visitor)
1924 {
1925 foreach_in_list_safe(ir_instruction, node, list) {
1926 node->accept(visitor);
1927 }
1928 }
1929
1930
1931 static void
1932 steal_memory(ir_instruction *ir, void *new_ctx)
1933 {
1934 ir_variable *var = ir->as_variable();
1935 ir_function *fn = ir->as_function();
1936 ir_constant *constant = ir->as_constant();
1937 if (var != NULL && var->constant_value != NULL)
1938 steal_memory(var->constant_value, ir);
1939
1940 if (var != NULL && var->constant_initializer != NULL)
1941 steal_memory(var->constant_initializer, ir);
1942
1943 if (fn != NULL && fn->subroutine_types)
1944 ralloc_steal(new_ctx, fn->subroutine_types);
1945
1946 /* The components of aggregate constants are not visited by the normal
1947 * visitor, so steal their values by hand.
1948 */
1949 if (constant != NULL) {
1950 if (constant->type->is_record()) {
1951 foreach_in_list(ir_constant, field, &constant->components) {
1952 steal_memory(field, ir);
1953 }
1954 } else if (constant->type->is_array()) {
1955 for (unsigned int i = 0; i < constant->type->length; i++) {
1956 steal_memory(constant->array_elements[i], ir);
1957 }
1958 }
1959 }
1960
1961 ralloc_steal(new_ctx, ir);
1962 }
1963
1964
1965 void
1966 reparent_ir(exec_list *list, void *mem_ctx)
1967 {
1968 foreach_in_list(ir_instruction, node, list) {
1969 visit_tree(node, steal_memory, mem_ctx);
1970 }
1971 }
1972
1973
1974 static ir_rvalue *
1975 try_min_one(ir_rvalue *ir)
1976 {
1977 ir_expression *expr = ir->as_expression();
1978
1979 if (!expr || expr->operation != ir_binop_min)
1980 return NULL;
1981
1982 if (expr->operands[0]->is_one())
1983 return expr->operands[1];
1984
1985 if (expr->operands[1]->is_one())
1986 return expr->operands[0];
1987
1988 return NULL;
1989 }
1990
1991 static ir_rvalue *
1992 try_max_zero(ir_rvalue *ir)
1993 {
1994 ir_expression *expr = ir->as_expression();
1995
1996 if (!expr || expr->operation != ir_binop_max)
1997 return NULL;
1998
1999 if (expr->operands[0]->is_zero())
2000 return expr->operands[1];
2001
2002 if (expr->operands[1]->is_zero())
2003 return expr->operands[0];
2004
2005 return NULL;
2006 }
2007
2008 ir_rvalue *
2009 ir_rvalue::as_rvalue_to_saturate()
2010 {
2011 ir_expression *expr = this->as_expression();
2012
2013 if (!expr)
2014 return NULL;
2015
2016 ir_rvalue *max_zero = try_max_zero(expr);
2017 if (max_zero) {
2018 return try_min_one(max_zero);
2019 } else {
2020 ir_rvalue *min_one = try_min_one(expr);
2021 if (min_one) {
2022 return try_max_zero(min_one);
2023 }
2024 }
2025
2026 return NULL;
2027 }
2028
2029
2030 unsigned
2031 vertices_per_prim(GLenum prim)
2032 {
2033 switch (prim) {
2034 case GL_POINTS:
2035 return 1;
2036 case GL_LINES:
2037 return 2;
2038 case GL_TRIANGLES:
2039 return 3;
2040 case GL_LINES_ADJACENCY:
2041 return 4;
2042 case GL_TRIANGLES_ADJACENCY:
2043 return 6;
2044 default:
2045 assert(!"Bad primitive");
2046 return 3;
2047 }
2048 }
2049
2050 /**
2051 * Generate a string describing the mode of a variable
2052 */
2053 const char *
2054 mode_string(const ir_variable *var)
2055 {
2056 switch (var->data.mode) {
2057 case ir_var_auto:
2058 return (var->data.read_only) ? "global constant" : "global variable";
2059
2060 case ir_var_uniform:
2061 return "uniform";
2062
2063 case ir_var_shader_storage:
2064 return "buffer";
2065
2066 case ir_var_shader_in:
2067 return "shader input";
2068
2069 case ir_var_shader_out:
2070 return "shader output";
2071
2072 case ir_var_function_in:
2073 case ir_var_const_in:
2074 return "function input";
2075
2076 case ir_var_function_out:
2077 return "function output";
2078
2079 case ir_var_function_inout:
2080 return "function inout";
2081
2082 case ir_var_system_value:
2083 return "shader input";
2084
2085 case ir_var_temporary:
2086 return "compiler temporary";
2087
2088 case ir_var_mode_count:
2089 break;
2090 }
2091
2092 assert(!"Should not get here.");
2093 return "invalid variable";
2094 }