glsl: Add ir_constant constructor for fp16
[mesa.git] / src / compiler / glsl / ir.cpp
1 /*
2 * Copyright © 2010 Intel Corporation
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
21 * DEALINGS IN THE SOFTWARE.
22 */
23 #include <string.h>
24 #include "ir.h"
25 #include "compiler/glsl_types.h"
26 #include "glsl_parser_extras.h"
27
28
29 ir_rvalue::ir_rvalue(enum ir_node_type t)
30 : ir_instruction(t)
31 {
32 this->type = glsl_type::error_type;
33 }
34
35 bool ir_rvalue::is_zero() const
36 {
37 return false;
38 }
39
40 bool ir_rvalue::is_one() const
41 {
42 return false;
43 }
44
45 bool ir_rvalue::is_negative_one() const
46 {
47 return false;
48 }
49
50 /**
51 * Modify the swizzle make to move one component to another
52 *
53 * \param m IR swizzle to be modified
54 * \param from Component in the RHS that is to be swizzled
55 * \param to Desired swizzle location of \c from
56 */
57 static void
58 update_rhs_swizzle(ir_swizzle_mask &m, unsigned from, unsigned to)
59 {
60 switch (to) {
61 case 0: m.x = from; break;
62 case 1: m.y = from; break;
63 case 2: m.z = from; break;
64 case 3: m.w = from; break;
65 default: assert(!"Should not get here.");
66 }
67 }
68
69 void
70 ir_assignment::set_lhs(ir_rvalue *lhs)
71 {
72 void *mem_ctx = this;
73 bool swizzled = false;
74
75 while (lhs != NULL) {
76 ir_swizzle *swiz = lhs->as_swizzle();
77
78 if (swiz == NULL)
79 break;
80
81 unsigned write_mask = 0;
82 ir_swizzle_mask rhs_swiz = { 0, 0, 0, 0, 0, 0 };
83
84 for (unsigned i = 0; i < swiz->mask.num_components; i++) {
85 unsigned c = 0;
86
87 switch (i) {
88 case 0: c = swiz->mask.x; break;
89 case 1: c = swiz->mask.y; break;
90 case 2: c = swiz->mask.z; break;
91 case 3: c = swiz->mask.w; break;
92 default: assert(!"Should not get here.");
93 }
94
95 write_mask |= (((this->write_mask >> i) & 1) << c);
96 update_rhs_swizzle(rhs_swiz, i, c);
97 rhs_swiz.num_components = swiz->val->type->vector_elements;
98 }
99
100 this->write_mask = write_mask;
101 lhs = swiz->val;
102
103 this->rhs = new(mem_ctx) ir_swizzle(this->rhs, rhs_swiz);
104 swizzled = true;
105 }
106
107 if (swizzled) {
108 /* Now, RHS channels line up with the LHS writemask. Collapse it
109 * to just the channels that will be written.
110 */
111 ir_swizzle_mask rhs_swiz = { 0, 0, 0, 0, 0, 0 };
112 int rhs_chan = 0;
113 for (int i = 0; i < 4; i++) {
114 if (write_mask & (1 << i))
115 update_rhs_swizzle(rhs_swiz, i, rhs_chan++);
116 }
117 rhs_swiz.num_components = rhs_chan;
118 this->rhs = new(mem_ctx) ir_swizzle(this->rhs, rhs_swiz);
119 }
120
121 assert((lhs == NULL) || lhs->as_dereference());
122
123 this->lhs = (ir_dereference *) lhs;
124 }
125
126 ir_variable *
127 ir_assignment::whole_variable_written()
128 {
129 ir_variable *v = this->lhs->whole_variable_referenced();
130
131 if (v == NULL)
132 return NULL;
133
134 if (v->type->is_scalar())
135 return v;
136
137 if (v->type->is_vector()) {
138 const unsigned mask = (1U << v->type->vector_elements) - 1;
139
140 if (mask != this->write_mask)
141 return NULL;
142 }
143
144 /* Either all the vector components are assigned or the variable is some
145 * composite type (and the whole thing is assigned.
146 */
147 return v;
148 }
149
150 ir_assignment::ir_assignment(ir_dereference *lhs, ir_rvalue *rhs,
151 ir_rvalue *condition, unsigned write_mask)
152 : ir_instruction(ir_type_assignment)
153 {
154 this->condition = condition;
155 this->rhs = rhs;
156 this->lhs = lhs;
157 this->write_mask = write_mask;
158
159 if (lhs->type->is_scalar() || lhs->type->is_vector()) {
160 int lhs_components = 0;
161 for (int i = 0; i < 4; i++) {
162 if (write_mask & (1 << i))
163 lhs_components++;
164 }
165
166 assert(lhs_components == this->rhs->type->vector_elements);
167 }
168 }
169
170 ir_assignment::ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs,
171 ir_rvalue *condition)
172 : ir_instruction(ir_type_assignment)
173 {
174 this->condition = condition;
175 this->rhs = rhs;
176
177 /* If the RHS is a vector type, assume that all components of the vector
178 * type are being written to the LHS. The write mask comes from the RHS
179 * because we can have a case where the LHS is a vec4 and the RHS is a
180 * vec3. In that case, the assignment is:
181 *
182 * (assign (...) (xyz) (var_ref lhs) (var_ref rhs))
183 */
184 if (rhs->type->is_vector())
185 this->write_mask = (1U << rhs->type->vector_elements) - 1;
186 else if (rhs->type->is_scalar())
187 this->write_mask = 1;
188 else
189 this->write_mask = 0;
190
191 this->set_lhs(lhs);
192 }
193
194 ir_expression::ir_expression(int op, const struct glsl_type *type,
195 ir_rvalue *op0, ir_rvalue *op1,
196 ir_rvalue *op2, ir_rvalue *op3)
197 : ir_rvalue(ir_type_expression)
198 {
199 this->type = type;
200 this->operation = ir_expression_operation(op);
201 this->operands[0] = op0;
202 this->operands[1] = op1;
203 this->operands[2] = op2;
204 this->operands[3] = op3;
205 init_num_operands();
206
207 #ifndef NDEBUG
208 for (unsigned i = num_operands; i < 4; i++) {
209 assert(this->operands[i] == NULL);
210 }
211
212 for (unsigned i = 0; i < num_operands; i++) {
213 assert(this->operands[i] != NULL);
214 }
215 #endif
216 }
217
218 ir_expression::ir_expression(int op, ir_rvalue *op0)
219 : ir_rvalue(ir_type_expression)
220 {
221 this->operation = ir_expression_operation(op);
222 this->operands[0] = op0;
223 this->operands[1] = NULL;
224 this->operands[2] = NULL;
225 this->operands[3] = NULL;
226
227 assert(op <= ir_last_unop);
228 init_num_operands();
229 assert(num_operands == 1);
230 assert(this->operands[0]);
231
232 switch (this->operation) {
233 case ir_unop_bit_not:
234 case ir_unop_logic_not:
235 case ir_unop_neg:
236 case ir_unop_abs:
237 case ir_unop_sign:
238 case ir_unop_rcp:
239 case ir_unop_rsq:
240 case ir_unop_sqrt:
241 case ir_unop_exp:
242 case ir_unop_log:
243 case ir_unop_exp2:
244 case ir_unop_log2:
245 case ir_unop_trunc:
246 case ir_unop_ceil:
247 case ir_unop_floor:
248 case ir_unop_fract:
249 case ir_unop_round_even:
250 case ir_unop_sin:
251 case ir_unop_cos:
252 case ir_unop_dFdx:
253 case ir_unop_dFdx_coarse:
254 case ir_unop_dFdx_fine:
255 case ir_unop_dFdy:
256 case ir_unop_dFdy_coarse:
257 case ir_unop_dFdy_fine:
258 case ir_unop_bitfield_reverse:
259 case ir_unop_interpolate_at_centroid:
260 case ir_unop_clz:
261 case ir_unop_saturate:
262 case ir_unop_atan:
263 this->type = op0->type;
264 break;
265
266 case ir_unop_f2i:
267 case ir_unop_b2i:
268 case ir_unop_u2i:
269 case ir_unop_d2i:
270 case ir_unop_bitcast_f2i:
271 case ir_unop_bit_count:
272 case ir_unop_find_msb:
273 case ir_unop_find_lsb:
274 case ir_unop_subroutine_to_int:
275 case ir_unop_i642i:
276 case ir_unop_u642i:
277 this->type = glsl_type::get_instance(GLSL_TYPE_INT,
278 op0->type->vector_elements, 1);
279 break;
280
281 case ir_unop_b2f:
282 case ir_unop_i2f:
283 case ir_unop_u2f:
284 case ir_unop_d2f:
285 case ir_unop_bitcast_i2f:
286 case ir_unop_bitcast_u2f:
287 case ir_unop_i642f:
288 case ir_unop_u642f:
289 this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT,
290 op0->type->vector_elements, 1);
291 break;
292
293 case ir_unop_f2b:
294 case ir_unop_i2b:
295 case ir_unop_d2b:
296 case ir_unop_i642b:
297 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL,
298 op0->type->vector_elements, 1);
299 break;
300
301 case ir_unop_f2d:
302 case ir_unop_i2d:
303 case ir_unop_u2d:
304 case ir_unop_i642d:
305 case ir_unop_u642d:
306 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE,
307 op0->type->vector_elements, 1);
308 break;
309
310 case ir_unop_i2u:
311 case ir_unop_f2u:
312 case ir_unop_d2u:
313 case ir_unop_bitcast_f2u:
314 case ir_unop_i642u:
315 case ir_unop_u642u:
316 this->type = glsl_type::get_instance(GLSL_TYPE_UINT,
317 op0->type->vector_elements, 1);
318 break;
319
320 case ir_unop_i2i64:
321 case ir_unop_u2i64:
322 case ir_unop_b2i64:
323 case ir_unop_f2i64:
324 case ir_unop_d2i64:
325 case ir_unop_u642i64:
326 this->type = glsl_type::get_instance(GLSL_TYPE_INT64,
327 op0->type->vector_elements, 1);
328 break;
329
330 case ir_unop_i2u64:
331 case ir_unop_u2u64:
332 case ir_unop_f2u64:
333 case ir_unop_d2u64:
334 case ir_unop_i642u64:
335 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64,
336 op0->type->vector_elements, 1);
337 break;
338 case ir_unop_noise:
339 this->type = glsl_type::float_type;
340 break;
341
342 case ir_unop_unpack_double_2x32:
343 case ir_unop_unpack_uint_2x32:
344 this->type = glsl_type::uvec2_type;
345 break;
346
347 case ir_unop_unpack_int_2x32:
348 this->type = glsl_type::ivec2_type;
349 break;
350
351 case ir_unop_pack_snorm_2x16:
352 case ir_unop_pack_snorm_4x8:
353 case ir_unop_pack_unorm_2x16:
354 case ir_unop_pack_unorm_4x8:
355 case ir_unop_pack_half_2x16:
356 this->type = glsl_type::uint_type;
357 break;
358
359 case ir_unop_pack_double_2x32:
360 this->type = glsl_type::double_type;
361 break;
362
363 case ir_unop_pack_int_2x32:
364 this->type = glsl_type::int64_t_type;
365 break;
366
367 case ir_unop_pack_uint_2x32:
368 this->type = glsl_type::uint64_t_type;
369 break;
370
371 case ir_unop_unpack_snorm_2x16:
372 case ir_unop_unpack_unorm_2x16:
373 case ir_unop_unpack_half_2x16:
374 this->type = glsl_type::vec2_type;
375 break;
376
377 case ir_unop_unpack_snorm_4x8:
378 case ir_unop_unpack_unorm_4x8:
379 this->type = glsl_type::vec4_type;
380 break;
381
382 case ir_unop_unpack_sampler_2x32:
383 case ir_unop_unpack_image_2x32:
384 this->type = glsl_type::uvec2_type;
385 break;
386
387 case ir_unop_pack_sampler_2x32:
388 case ir_unop_pack_image_2x32:
389 this->type = op0->type;
390 break;
391
392 case ir_unop_frexp_sig:
393 this->type = op0->type;
394 break;
395 case ir_unop_frexp_exp:
396 this->type = glsl_type::get_instance(GLSL_TYPE_INT,
397 op0->type->vector_elements, 1);
398 break;
399
400 case ir_unop_get_buffer_size:
401 case ir_unop_ssbo_unsized_array_length:
402 this->type = glsl_type::int_type;
403 break;
404
405 case ir_unop_bitcast_i642d:
406 case ir_unop_bitcast_u642d:
407 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE,
408 op0->type->vector_elements, 1);
409 break;
410
411 case ir_unop_bitcast_d2i64:
412 this->type = glsl_type::get_instance(GLSL_TYPE_INT64,
413 op0->type->vector_elements, 1);
414 break;
415 case ir_unop_bitcast_d2u64:
416 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64,
417 op0->type->vector_elements, 1);
418 break;
419
420 default:
421 assert(!"not reached: missing automatic type setup for ir_expression");
422 this->type = op0->type;
423 break;
424 }
425 }
426
427 ir_expression::ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1)
428 : ir_rvalue(ir_type_expression)
429 {
430 this->operation = ir_expression_operation(op);
431 this->operands[0] = op0;
432 this->operands[1] = op1;
433 this->operands[2] = NULL;
434 this->operands[3] = NULL;
435
436 assert(op > ir_last_unop);
437 init_num_operands();
438 assert(num_operands == 2);
439 for (unsigned i = 0; i < num_operands; i++) {
440 assert(this->operands[i] != NULL);
441 }
442
443 switch (this->operation) {
444 case ir_binop_all_equal:
445 case ir_binop_any_nequal:
446 this->type = glsl_type::bool_type;
447 break;
448
449 case ir_binop_add:
450 case ir_binop_sub:
451 case ir_binop_min:
452 case ir_binop_max:
453 case ir_binop_pow:
454 case ir_binop_mul:
455 case ir_binop_div:
456 case ir_binop_mod:
457 case ir_binop_atan2:
458 if (op0->type->is_scalar()) {
459 this->type = op1->type;
460 } else if (op1->type->is_scalar()) {
461 this->type = op0->type;
462 } else {
463 if (this->operation == ir_binop_mul) {
464 this->type = glsl_type::get_mul_type(op0->type, op1->type);
465 } else {
466 assert(op0->type == op1->type);
467 this->type = op0->type;
468 }
469 }
470 break;
471
472 case ir_binop_logic_and:
473 case ir_binop_logic_xor:
474 case ir_binop_logic_or:
475 case ir_binop_bit_and:
476 case ir_binop_bit_xor:
477 case ir_binop_bit_or:
478 assert(!op0->type->is_matrix());
479 assert(!op1->type->is_matrix());
480 if (op0->type->is_scalar()) {
481 this->type = op1->type;
482 } else if (op1->type->is_scalar()) {
483 this->type = op0->type;
484 } else {
485 assert(op0->type->vector_elements == op1->type->vector_elements);
486 this->type = op0->type;
487 }
488 break;
489
490 case ir_binop_equal:
491 case ir_binop_nequal:
492 case ir_binop_gequal:
493 case ir_binop_less:
494 assert(op0->type == op1->type);
495 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL,
496 op0->type->vector_elements, 1);
497 break;
498
499 case ir_binop_dot:
500 this->type = op0->type->get_base_type();
501 break;
502
503 case ir_binop_imul_high:
504 case ir_binop_mul_32x16:
505 case ir_binop_carry:
506 case ir_binop_borrow:
507 case ir_binop_lshift:
508 case ir_binop_rshift:
509 case ir_binop_ldexp:
510 case ir_binop_interpolate_at_offset:
511 case ir_binop_interpolate_at_sample:
512 this->type = op0->type;
513 break;
514
515 case ir_binop_add_sat:
516 case ir_binop_sub_sat:
517 case ir_binop_avg:
518 case ir_binop_avg_round:
519 assert(op0->type == op1->type);
520 this->type = op0->type;
521 break;
522
523 case ir_binop_abs_sub: {
524 enum glsl_base_type base;
525
526 assert(op0->type == op1->type);
527
528 switch (op0->type->base_type) {
529 case GLSL_TYPE_UINT:
530 case GLSL_TYPE_INT:
531 base = GLSL_TYPE_UINT;
532 break;
533 case GLSL_TYPE_UINT8:
534 case GLSL_TYPE_INT8:
535 base = GLSL_TYPE_UINT8;
536 break;
537 case GLSL_TYPE_UINT16:
538 case GLSL_TYPE_INT16:
539 base = GLSL_TYPE_UINT16;
540 break;
541 case GLSL_TYPE_UINT64:
542 case GLSL_TYPE_INT64:
543 base = GLSL_TYPE_UINT64;
544 break;
545 default:
546 unreachable(!"Invalid base type.");
547 }
548
549 this->type = glsl_type::get_instance(base, op0->type->vector_elements, 1);
550 break;
551 }
552
553 case ir_binop_vector_extract:
554 this->type = op0->type->get_scalar_type();
555 break;
556
557 default:
558 assert(!"not reached: missing automatic type setup for ir_expression");
559 this->type = glsl_type::float_type;
560 }
561 }
562
563 ir_expression::ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1,
564 ir_rvalue *op2)
565 : ir_rvalue(ir_type_expression)
566 {
567 this->operation = ir_expression_operation(op);
568 this->operands[0] = op0;
569 this->operands[1] = op1;
570 this->operands[2] = op2;
571 this->operands[3] = NULL;
572
573 assert(op > ir_last_binop && op <= ir_last_triop);
574 init_num_operands();
575 assert(num_operands == 3);
576 for (unsigned i = 0; i < num_operands; i++) {
577 assert(this->operands[i] != NULL);
578 }
579
580 switch (this->operation) {
581 case ir_triop_fma:
582 case ir_triop_lrp:
583 case ir_triop_bitfield_extract:
584 case ir_triop_vector_insert:
585 this->type = op0->type;
586 break;
587
588 case ir_triop_csel:
589 this->type = op1->type;
590 break;
591
592 default:
593 assert(!"not reached: missing automatic type setup for ir_expression");
594 this->type = glsl_type::float_type;
595 }
596 }
597
598 /**
599 * This is only here for ir_reader to used for testing purposes. Please use
600 * the precomputed num_operands field if you need the number of operands.
601 */
602 unsigned
603 ir_expression::get_num_operands(ir_expression_operation op)
604 {
605 assert(op <= ir_last_opcode);
606
607 if (op <= ir_last_unop)
608 return 1;
609
610 if (op <= ir_last_binop)
611 return 2;
612
613 if (op <= ir_last_triop)
614 return 3;
615
616 if (op <= ir_last_quadop)
617 return 4;
618
619 unreachable("Could not calculate number of operands");
620 }
621
622 #include "ir_expression_operation_strings.h"
623
624 const char*
625 depth_layout_string(ir_depth_layout layout)
626 {
627 switch(layout) {
628 case ir_depth_layout_none: return "";
629 case ir_depth_layout_any: return "depth_any";
630 case ir_depth_layout_greater: return "depth_greater";
631 case ir_depth_layout_less: return "depth_less";
632 case ir_depth_layout_unchanged: return "depth_unchanged";
633
634 default:
635 assert(0);
636 return "";
637 }
638 }
639
640 ir_expression_operation
641 ir_expression::get_operator(const char *str)
642 {
643 for (int op = 0; op <= int(ir_last_opcode); op++) {
644 if (strcmp(str, ir_expression_operation_strings[op]) == 0)
645 return (ir_expression_operation) op;
646 }
647 return (ir_expression_operation) -1;
648 }
649
650 ir_variable *
651 ir_expression::variable_referenced() const
652 {
653 switch (operation) {
654 case ir_binop_vector_extract:
655 case ir_triop_vector_insert:
656 /* We get these for things like a[0] where a is a vector type. In these
657 * cases we want variable_referenced() to return the actual vector
658 * variable this is wrapping.
659 */
660 return operands[0]->variable_referenced();
661 default:
662 return ir_rvalue::variable_referenced();
663 }
664 }
665
666 ir_constant::ir_constant()
667 : ir_rvalue(ir_type_constant)
668 {
669 this->const_elements = NULL;
670 }
671
672 ir_constant::ir_constant(const struct glsl_type *type,
673 const ir_constant_data *data)
674 : ir_rvalue(ir_type_constant)
675 {
676 this->const_elements = NULL;
677
678 assert((type->base_type >= GLSL_TYPE_UINT)
679 && (type->base_type <= GLSL_TYPE_IMAGE));
680
681 this->type = type;
682 memcpy(& this->value, data, sizeof(this->value));
683 }
684
685 ir_constant::ir_constant(float16_t f16, unsigned vector_elements)
686 : ir_rvalue(ir_type_constant)
687 {
688 assert(vector_elements <= 4);
689 this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT16, vector_elements, 1);
690 for (unsigned i = 0; i < vector_elements; i++) {
691 this->value.f16[i] = f16.bits;
692 }
693 for (unsigned i = vector_elements; i < 16; i++) {
694 this->value.f[i] = 0;
695 }
696 }
697
698 ir_constant::ir_constant(float f, unsigned vector_elements)
699 : ir_rvalue(ir_type_constant)
700 {
701 assert(vector_elements <= 4);
702 this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT, vector_elements, 1);
703 for (unsigned i = 0; i < vector_elements; i++) {
704 this->value.f[i] = f;
705 }
706 for (unsigned i = vector_elements; i < 16; i++) {
707 this->value.f[i] = 0;
708 }
709 }
710
711 ir_constant::ir_constant(double d, unsigned vector_elements)
712 : ir_rvalue(ir_type_constant)
713 {
714 assert(vector_elements <= 4);
715 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE, vector_elements, 1);
716 for (unsigned i = 0; i < vector_elements; i++) {
717 this->value.d[i] = d;
718 }
719 for (unsigned i = vector_elements; i < 16; i++) {
720 this->value.d[i] = 0.0;
721 }
722 }
723
724 ir_constant::ir_constant(unsigned int u, unsigned vector_elements)
725 : ir_rvalue(ir_type_constant)
726 {
727 assert(vector_elements <= 4);
728 this->type = glsl_type::get_instance(GLSL_TYPE_UINT, vector_elements, 1);
729 for (unsigned i = 0; i < vector_elements; i++) {
730 this->value.u[i] = u;
731 }
732 for (unsigned i = vector_elements; i < 16; i++) {
733 this->value.u[i] = 0;
734 }
735 }
736
737 ir_constant::ir_constant(int integer, unsigned vector_elements)
738 : ir_rvalue(ir_type_constant)
739 {
740 assert(vector_elements <= 4);
741 this->type = glsl_type::get_instance(GLSL_TYPE_INT, vector_elements, 1);
742 for (unsigned i = 0; i < vector_elements; i++) {
743 this->value.i[i] = integer;
744 }
745 for (unsigned i = vector_elements; i < 16; i++) {
746 this->value.i[i] = 0;
747 }
748 }
749
750 ir_constant::ir_constant(uint64_t u64, unsigned vector_elements)
751 : ir_rvalue(ir_type_constant)
752 {
753 assert(vector_elements <= 4);
754 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64, vector_elements, 1);
755 for (unsigned i = 0; i < vector_elements; i++) {
756 this->value.u64[i] = u64;
757 }
758 for (unsigned i = vector_elements; i < 16; i++) {
759 this->value.u64[i] = 0;
760 }
761 }
762
763 ir_constant::ir_constant(int64_t int64, unsigned vector_elements)
764 : ir_rvalue(ir_type_constant)
765 {
766 assert(vector_elements <= 4);
767 this->type = glsl_type::get_instance(GLSL_TYPE_INT64, vector_elements, 1);
768 for (unsigned i = 0; i < vector_elements; i++) {
769 this->value.i64[i] = int64;
770 }
771 for (unsigned i = vector_elements; i < 16; i++) {
772 this->value.i64[i] = 0;
773 }
774 }
775
776 ir_constant::ir_constant(bool b, unsigned vector_elements)
777 : ir_rvalue(ir_type_constant)
778 {
779 assert(vector_elements <= 4);
780 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL, vector_elements, 1);
781 for (unsigned i = 0; i < vector_elements; i++) {
782 this->value.b[i] = b;
783 }
784 for (unsigned i = vector_elements; i < 16; i++) {
785 this->value.b[i] = false;
786 }
787 }
788
789 ir_constant::ir_constant(const ir_constant *c, unsigned i)
790 : ir_rvalue(ir_type_constant)
791 {
792 this->const_elements = NULL;
793 this->type = c->type->get_base_type();
794
795 switch (this->type->base_type) {
796 case GLSL_TYPE_UINT: this->value.u[0] = c->value.u[i]; break;
797 case GLSL_TYPE_INT: this->value.i[0] = c->value.i[i]; break;
798 case GLSL_TYPE_FLOAT: this->value.f[0] = c->value.f[i]; break;
799 case GLSL_TYPE_BOOL: this->value.b[0] = c->value.b[i]; break;
800 case GLSL_TYPE_DOUBLE: this->value.d[0] = c->value.d[i]; break;
801 default: assert(!"Should not get here."); break;
802 }
803 }
804
805 ir_constant::ir_constant(const struct glsl_type *type, exec_list *value_list)
806 : ir_rvalue(ir_type_constant)
807 {
808 this->const_elements = NULL;
809 this->type = type;
810
811 assert(type->is_scalar() || type->is_vector() || type->is_matrix()
812 || type->is_struct() || type->is_array());
813
814 /* If the constant is a record, the types of each of the entries in
815 * value_list must be a 1-for-1 match with the structure components. Each
816 * entry must also be a constant. Just move the nodes from the value_list
817 * to the list in the ir_constant.
818 */
819 if (type->is_array() || type->is_struct()) {
820 this->const_elements = ralloc_array(this, ir_constant *, type->length);
821 unsigned i = 0;
822 foreach_in_list(ir_constant, value, value_list) {
823 assert(value->as_constant() != NULL);
824
825 this->const_elements[i++] = value;
826 }
827 return;
828 }
829
830 for (unsigned i = 0; i < 16; i++) {
831 this->value.u[i] = 0;
832 }
833
834 ir_constant *value = (ir_constant *) (value_list->get_head_raw());
835
836 /* Constructors with exactly one scalar argument are special for vectors
837 * and matrices. For vectors, the scalar value is replicated to fill all
838 * the components. For matrices, the scalar fills the components of the
839 * diagonal while the rest is filled with 0.
840 */
841 if (value->type->is_scalar() && value->next->is_tail_sentinel()) {
842 if (type->is_matrix()) {
843 /* Matrix - fill diagonal (rest is already set to 0) */
844 assert(type->is_float() || type->is_double());
845 for (unsigned i = 0; i < type->matrix_columns; i++) {
846 if (type->is_float())
847 this->value.f[i * type->vector_elements + i] =
848 value->value.f[0];
849 else
850 this->value.d[i * type->vector_elements + i] =
851 value->value.d[0];
852 }
853 } else {
854 /* Vector or scalar - fill all components */
855 switch (type->base_type) {
856 case GLSL_TYPE_UINT:
857 case GLSL_TYPE_INT:
858 for (unsigned i = 0; i < type->components(); i++)
859 this->value.u[i] = value->value.u[0];
860 break;
861 case GLSL_TYPE_FLOAT:
862 for (unsigned i = 0; i < type->components(); i++)
863 this->value.f[i] = value->value.f[0];
864 break;
865 case GLSL_TYPE_DOUBLE:
866 for (unsigned i = 0; i < type->components(); i++)
867 this->value.d[i] = value->value.d[0];
868 break;
869 case GLSL_TYPE_UINT64:
870 case GLSL_TYPE_INT64:
871 for (unsigned i = 0; i < type->components(); i++)
872 this->value.u64[i] = value->value.u64[0];
873 break;
874 case GLSL_TYPE_BOOL:
875 for (unsigned i = 0; i < type->components(); i++)
876 this->value.b[i] = value->value.b[0];
877 break;
878 case GLSL_TYPE_SAMPLER:
879 case GLSL_TYPE_IMAGE:
880 this->value.u64[0] = value->value.u64[0];
881 break;
882 default:
883 assert(!"Should not get here.");
884 break;
885 }
886 }
887 return;
888 }
889
890 if (type->is_matrix() && value->type->is_matrix()) {
891 assert(value->next->is_tail_sentinel());
892
893 /* From section 5.4.2 of the GLSL 1.20 spec:
894 * "If a matrix is constructed from a matrix, then each component
895 * (column i, row j) in the result that has a corresponding component
896 * (column i, row j) in the argument will be initialized from there."
897 */
898 unsigned cols = MIN2(type->matrix_columns, value->type->matrix_columns);
899 unsigned rows = MIN2(type->vector_elements, value->type->vector_elements);
900 for (unsigned i = 0; i < cols; i++) {
901 for (unsigned j = 0; j < rows; j++) {
902 const unsigned src = i * value->type->vector_elements + j;
903 const unsigned dst = i * type->vector_elements + j;
904 this->value.f[dst] = value->value.f[src];
905 }
906 }
907
908 /* "All other components will be initialized to the identity matrix." */
909 for (unsigned i = cols; i < type->matrix_columns; i++)
910 this->value.f[i * type->vector_elements + i] = 1.0;
911
912 return;
913 }
914
915 /* Use each component from each entry in the value_list to initialize one
916 * component of the constant being constructed.
917 */
918 unsigned i = 0;
919 for (;;) {
920 assert(value->as_constant() != NULL);
921 assert(!value->is_tail_sentinel());
922
923 for (unsigned j = 0; j < value->type->components(); j++) {
924 switch (type->base_type) {
925 case GLSL_TYPE_UINT:
926 this->value.u[i] = value->get_uint_component(j);
927 break;
928 case GLSL_TYPE_INT:
929 this->value.i[i] = value->get_int_component(j);
930 break;
931 case GLSL_TYPE_FLOAT:
932 this->value.f[i] = value->get_float_component(j);
933 break;
934 case GLSL_TYPE_BOOL:
935 this->value.b[i] = value->get_bool_component(j);
936 break;
937 case GLSL_TYPE_DOUBLE:
938 this->value.d[i] = value->get_double_component(j);
939 break;
940 case GLSL_TYPE_UINT64:
941 this->value.u64[i] = value->get_uint64_component(j);
942 break;
943 case GLSL_TYPE_INT64:
944 this->value.i64[i] = value->get_int64_component(j);
945 break;
946 default:
947 /* FINISHME: What to do? Exceptions are not the answer.
948 */
949 break;
950 }
951
952 i++;
953 if (i >= type->components())
954 break;
955 }
956
957 if (i >= type->components())
958 break; /* avoid downcasting a list sentinel */
959 value = (ir_constant *) value->next;
960 }
961 }
962
963 ir_constant *
964 ir_constant::zero(void *mem_ctx, const glsl_type *type)
965 {
966 assert(type->is_scalar() || type->is_vector() || type->is_matrix()
967 || type->is_struct() || type->is_array());
968
969 ir_constant *c = new(mem_ctx) ir_constant;
970 c->type = type;
971 memset(&c->value, 0, sizeof(c->value));
972
973 if (type->is_array()) {
974 c->const_elements = ralloc_array(c, ir_constant *, type->length);
975
976 for (unsigned i = 0; i < type->length; i++)
977 c->const_elements[i] = ir_constant::zero(c, type->fields.array);
978 }
979
980 if (type->is_struct()) {
981 c->const_elements = ralloc_array(c, ir_constant *, type->length);
982
983 for (unsigned i = 0; i < type->length; i++) {
984 c->const_elements[i] =
985 ir_constant::zero(mem_ctx, type->fields.structure[i].type);
986 }
987 }
988
989 return c;
990 }
991
992 bool
993 ir_constant::get_bool_component(unsigned i) const
994 {
995 switch (this->type->base_type) {
996 case GLSL_TYPE_UINT: return this->value.u[i] != 0;
997 case GLSL_TYPE_INT: return this->value.i[i] != 0;
998 case GLSL_TYPE_FLOAT: return ((int)this->value.f[i]) != 0;
999 case GLSL_TYPE_BOOL: return this->value.b[i];
1000 case GLSL_TYPE_DOUBLE: return this->value.d[i] != 0.0;
1001 case GLSL_TYPE_SAMPLER:
1002 case GLSL_TYPE_IMAGE:
1003 case GLSL_TYPE_UINT64: return this->value.u64[i] != 0;
1004 case GLSL_TYPE_INT64: return this->value.i64[i] != 0;
1005 default: assert(!"Should not get here."); break;
1006 }
1007
1008 /* Must return something to make the compiler happy. This is clearly an
1009 * error case.
1010 */
1011 return false;
1012 }
1013
1014 float
1015 ir_constant::get_float_component(unsigned i) const
1016 {
1017 switch (this->type->base_type) {
1018 case GLSL_TYPE_UINT: return (float) this->value.u[i];
1019 case GLSL_TYPE_INT: return (float) this->value.i[i];
1020 case GLSL_TYPE_FLOAT: return this->value.f[i];
1021 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1.0f : 0.0f;
1022 case GLSL_TYPE_DOUBLE: return (float) this->value.d[i];
1023 case GLSL_TYPE_SAMPLER:
1024 case GLSL_TYPE_IMAGE:
1025 case GLSL_TYPE_UINT64: return (float) this->value.u64[i];
1026 case GLSL_TYPE_INT64: return (float) this->value.i64[i];
1027 default: assert(!"Should not get here."); break;
1028 }
1029
1030 /* Must return something to make the compiler happy. This is clearly an
1031 * error case.
1032 */
1033 return 0.0;
1034 }
1035
1036 double
1037 ir_constant::get_double_component(unsigned i) const
1038 {
1039 switch (this->type->base_type) {
1040 case GLSL_TYPE_UINT: return (double) this->value.u[i];
1041 case GLSL_TYPE_INT: return (double) this->value.i[i];
1042 case GLSL_TYPE_FLOAT: return (double) this->value.f[i];
1043 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1.0 : 0.0;
1044 case GLSL_TYPE_DOUBLE: return this->value.d[i];
1045 case GLSL_TYPE_SAMPLER:
1046 case GLSL_TYPE_IMAGE:
1047 case GLSL_TYPE_UINT64: return (double) this->value.u64[i];
1048 case GLSL_TYPE_INT64: return (double) this->value.i64[i];
1049 default: assert(!"Should not get here."); break;
1050 }
1051
1052 /* Must return something to make the compiler happy. This is clearly an
1053 * error case.
1054 */
1055 return 0.0;
1056 }
1057
1058 int
1059 ir_constant::get_int_component(unsigned i) const
1060 {
1061 switch (this->type->base_type) {
1062 case GLSL_TYPE_UINT: return this->value.u[i];
1063 case GLSL_TYPE_INT: return this->value.i[i];
1064 case GLSL_TYPE_FLOAT: return (int) this->value.f[i];
1065 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1066 case GLSL_TYPE_DOUBLE: return (int) this->value.d[i];
1067 case GLSL_TYPE_SAMPLER:
1068 case GLSL_TYPE_IMAGE:
1069 case GLSL_TYPE_UINT64: return (int) this->value.u64[i];
1070 case GLSL_TYPE_INT64: return (int) this->value.i64[i];
1071 default: assert(!"Should not get here."); break;
1072 }
1073
1074 /* Must return something to make the compiler happy. This is clearly an
1075 * error case.
1076 */
1077 return 0;
1078 }
1079
1080 unsigned
1081 ir_constant::get_uint_component(unsigned i) const
1082 {
1083 switch (this->type->base_type) {
1084 case GLSL_TYPE_UINT: return this->value.u[i];
1085 case GLSL_TYPE_INT: return this->value.i[i];
1086 case GLSL_TYPE_FLOAT: return (unsigned) this->value.f[i];
1087 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1088 case GLSL_TYPE_DOUBLE: return (unsigned) this->value.d[i];
1089 case GLSL_TYPE_SAMPLER:
1090 case GLSL_TYPE_IMAGE:
1091 case GLSL_TYPE_UINT64: return (unsigned) this->value.u64[i];
1092 case GLSL_TYPE_INT64: return (unsigned) this->value.i64[i];
1093 default: assert(!"Should not get here."); break;
1094 }
1095
1096 /* Must return something to make the compiler happy. This is clearly an
1097 * error case.
1098 */
1099 return 0;
1100 }
1101
1102 int64_t
1103 ir_constant::get_int64_component(unsigned i) const
1104 {
1105 switch (this->type->base_type) {
1106 case GLSL_TYPE_UINT: return this->value.u[i];
1107 case GLSL_TYPE_INT: return this->value.i[i];
1108 case GLSL_TYPE_FLOAT: return (int64_t) this->value.f[i];
1109 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1110 case GLSL_TYPE_DOUBLE: return (int64_t) this->value.d[i];
1111 case GLSL_TYPE_SAMPLER:
1112 case GLSL_TYPE_IMAGE:
1113 case GLSL_TYPE_UINT64: return (int64_t) this->value.u64[i];
1114 case GLSL_TYPE_INT64: return this->value.i64[i];
1115 default: assert(!"Should not get here."); break;
1116 }
1117
1118 /* Must return something to make the compiler happy. This is clearly an
1119 * error case.
1120 */
1121 return 0;
1122 }
1123
1124 uint64_t
1125 ir_constant::get_uint64_component(unsigned i) const
1126 {
1127 switch (this->type->base_type) {
1128 case GLSL_TYPE_UINT: return this->value.u[i];
1129 case GLSL_TYPE_INT: return this->value.i[i];
1130 case GLSL_TYPE_FLOAT: return (uint64_t) this->value.f[i];
1131 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1132 case GLSL_TYPE_DOUBLE: return (uint64_t) this->value.d[i];
1133 case GLSL_TYPE_SAMPLER:
1134 case GLSL_TYPE_IMAGE:
1135 case GLSL_TYPE_UINT64: return this->value.u64[i];
1136 case GLSL_TYPE_INT64: return (uint64_t) this->value.i64[i];
1137 default: assert(!"Should not get here."); break;
1138 }
1139
1140 /* Must return something to make the compiler happy. This is clearly an
1141 * error case.
1142 */
1143 return 0;
1144 }
1145
1146 ir_constant *
1147 ir_constant::get_array_element(unsigned i) const
1148 {
1149 assert(this->type->is_array());
1150
1151 /* From page 35 (page 41 of the PDF) of the GLSL 1.20 spec:
1152 *
1153 * "Behavior is undefined if a shader subscripts an array with an index
1154 * less than 0 or greater than or equal to the size the array was
1155 * declared with."
1156 *
1157 * Most out-of-bounds accesses are removed before things could get this far.
1158 * There are cases where non-constant array index values can get constant
1159 * folded.
1160 */
1161 if (int(i) < 0)
1162 i = 0;
1163 else if (i >= this->type->length)
1164 i = this->type->length - 1;
1165
1166 return const_elements[i];
1167 }
1168
1169 ir_constant *
1170 ir_constant::get_record_field(int idx)
1171 {
1172 assert(this->type->is_struct());
1173 assert(idx >= 0 && (unsigned) idx < this->type->length);
1174
1175 return const_elements[idx];
1176 }
1177
1178 void
1179 ir_constant::copy_offset(ir_constant *src, int offset)
1180 {
1181 switch (this->type->base_type) {
1182 case GLSL_TYPE_UINT:
1183 case GLSL_TYPE_INT:
1184 case GLSL_TYPE_FLOAT:
1185 case GLSL_TYPE_DOUBLE:
1186 case GLSL_TYPE_SAMPLER:
1187 case GLSL_TYPE_IMAGE:
1188 case GLSL_TYPE_UINT64:
1189 case GLSL_TYPE_INT64:
1190 case GLSL_TYPE_BOOL: {
1191 unsigned int size = src->type->components();
1192 assert (size <= this->type->components() - offset);
1193 for (unsigned int i=0; i<size; i++) {
1194 switch (this->type->base_type) {
1195 case GLSL_TYPE_UINT:
1196 value.u[i+offset] = src->get_uint_component(i);
1197 break;
1198 case GLSL_TYPE_INT:
1199 value.i[i+offset] = src->get_int_component(i);
1200 break;
1201 case GLSL_TYPE_FLOAT:
1202 value.f[i+offset] = src->get_float_component(i);
1203 break;
1204 case GLSL_TYPE_BOOL:
1205 value.b[i+offset] = src->get_bool_component(i);
1206 break;
1207 case GLSL_TYPE_DOUBLE:
1208 value.d[i+offset] = src->get_double_component(i);
1209 break;
1210 case GLSL_TYPE_SAMPLER:
1211 case GLSL_TYPE_IMAGE:
1212 case GLSL_TYPE_UINT64:
1213 value.u64[i+offset] = src->get_uint64_component(i);
1214 break;
1215 case GLSL_TYPE_INT64:
1216 value.i64[i+offset] = src->get_int64_component(i);
1217 break;
1218 default: // Shut up the compiler
1219 break;
1220 }
1221 }
1222 break;
1223 }
1224
1225 case GLSL_TYPE_STRUCT:
1226 case GLSL_TYPE_ARRAY: {
1227 assert (src->type == this->type);
1228 for (unsigned i = 0; i < this->type->length; i++) {
1229 this->const_elements[i] = src->const_elements[i]->clone(this, NULL);
1230 }
1231 break;
1232 }
1233
1234 default:
1235 assert(!"Should not get here.");
1236 break;
1237 }
1238 }
1239
1240 void
1241 ir_constant::copy_masked_offset(ir_constant *src, int offset, unsigned int mask)
1242 {
1243 assert (!type->is_array() && !type->is_struct());
1244
1245 if (!type->is_vector() && !type->is_matrix()) {
1246 offset = 0;
1247 mask = 1;
1248 }
1249
1250 int id = 0;
1251 for (int i=0; i<4; i++) {
1252 if (mask & (1 << i)) {
1253 switch (this->type->base_type) {
1254 case GLSL_TYPE_UINT:
1255 value.u[i+offset] = src->get_uint_component(id++);
1256 break;
1257 case GLSL_TYPE_INT:
1258 value.i[i+offset] = src->get_int_component(id++);
1259 break;
1260 case GLSL_TYPE_FLOAT:
1261 value.f[i+offset] = src->get_float_component(id++);
1262 break;
1263 case GLSL_TYPE_BOOL:
1264 value.b[i+offset] = src->get_bool_component(id++);
1265 break;
1266 case GLSL_TYPE_DOUBLE:
1267 value.d[i+offset] = src->get_double_component(id++);
1268 break;
1269 case GLSL_TYPE_SAMPLER:
1270 case GLSL_TYPE_IMAGE:
1271 case GLSL_TYPE_UINT64:
1272 value.u64[i+offset] = src->get_uint64_component(id++);
1273 break;
1274 case GLSL_TYPE_INT64:
1275 value.i64[i+offset] = src->get_int64_component(id++);
1276 break;
1277 default:
1278 assert(!"Should not get here.");
1279 return;
1280 }
1281 }
1282 }
1283 }
1284
1285 bool
1286 ir_constant::has_value(const ir_constant *c) const
1287 {
1288 if (this->type != c->type)
1289 return false;
1290
1291 if (this->type->is_array() || this->type->is_struct()) {
1292 for (unsigned i = 0; i < this->type->length; i++) {
1293 if (!this->const_elements[i]->has_value(c->const_elements[i]))
1294 return false;
1295 }
1296 return true;
1297 }
1298
1299 for (unsigned i = 0; i < this->type->components(); i++) {
1300 switch (this->type->base_type) {
1301 case GLSL_TYPE_UINT:
1302 if (this->value.u[i] != c->value.u[i])
1303 return false;
1304 break;
1305 case GLSL_TYPE_INT:
1306 if (this->value.i[i] != c->value.i[i])
1307 return false;
1308 break;
1309 case GLSL_TYPE_FLOAT:
1310 if (this->value.f[i] != c->value.f[i])
1311 return false;
1312 break;
1313 case GLSL_TYPE_BOOL:
1314 if (this->value.b[i] != c->value.b[i])
1315 return false;
1316 break;
1317 case GLSL_TYPE_DOUBLE:
1318 if (this->value.d[i] != c->value.d[i])
1319 return false;
1320 break;
1321 case GLSL_TYPE_SAMPLER:
1322 case GLSL_TYPE_IMAGE:
1323 case GLSL_TYPE_UINT64:
1324 if (this->value.u64[i] != c->value.u64[i])
1325 return false;
1326 break;
1327 case GLSL_TYPE_INT64:
1328 if (this->value.i64[i] != c->value.i64[i])
1329 return false;
1330 break;
1331 default:
1332 assert(!"Should not get here.");
1333 return false;
1334 }
1335 }
1336
1337 return true;
1338 }
1339
1340 bool
1341 ir_constant::is_value(float f, int i) const
1342 {
1343 if (!this->type->is_scalar() && !this->type->is_vector())
1344 return false;
1345
1346 /* Only accept boolean values for 0/1. */
1347 if (int(bool(i)) != i && this->type->is_boolean())
1348 return false;
1349
1350 for (unsigned c = 0; c < this->type->vector_elements; c++) {
1351 switch (this->type->base_type) {
1352 case GLSL_TYPE_FLOAT:
1353 if (this->value.f[c] != f)
1354 return false;
1355 break;
1356 case GLSL_TYPE_INT:
1357 if (this->value.i[c] != i)
1358 return false;
1359 break;
1360 case GLSL_TYPE_UINT:
1361 if (this->value.u[c] != unsigned(i))
1362 return false;
1363 break;
1364 case GLSL_TYPE_BOOL:
1365 if (this->value.b[c] != bool(i))
1366 return false;
1367 break;
1368 case GLSL_TYPE_DOUBLE:
1369 if (this->value.d[c] != double(f))
1370 return false;
1371 break;
1372 case GLSL_TYPE_SAMPLER:
1373 case GLSL_TYPE_IMAGE:
1374 case GLSL_TYPE_UINT64:
1375 if (this->value.u64[c] != uint64_t(i))
1376 return false;
1377 break;
1378 case GLSL_TYPE_INT64:
1379 if (this->value.i64[c] != i)
1380 return false;
1381 break;
1382 default:
1383 /* The only other base types are structures, arrays, and samplers.
1384 * Samplers cannot be constants, and the others should have been
1385 * filtered out above.
1386 */
1387 assert(!"Should not get here.");
1388 return false;
1389 }
1390 }
1391
1392 return true;
1393 }
1394
1395 bool
1396 ir_constant::is_zero() const
1397 {
1398 return is_value(0.0, 0);
1399 }
1400
1401 bool
1402 ir_constant::is_one() const
1403 {
1404 return is_value(1.0, 1);
1405 }
1406
1407 bool
1408 ir_constant::is_negative_one() const
1409 {
1410 return is_value(-1.0, -1);
1411 }
1412
1413 bool
1414 ir_constant::is_uint16_constant() const
1415 {
1416 if (!type->is_integer_32())
1417 return false;
1418
1419 return value.u[0] < (1 << 16);
1420 }
1421
1422 ir_loop::ir_loop()
1423 : ir_instruction(ir_type_loop)
1424 {
1425 }
1426
1427
1428 ir_dereference_variable::ir_dereference_variable(ir_variable *var)
1429 : ir_dereference(ir_type_dereference_variable)
1430 {
1431 assert(var != NULL);
1432
1433 this->var = var;
1434 this->type = var->type;
1435 }
1436
1437
1438 ir_dereference_array::ir_dereference_array(ir_rvalue *value,
1439 ir_rvalue *array_index)
1440 : ir_dereference(ir_type_dereference_array)
1441 {
1442 this->array_index = array_index;
1443 this->set_array(value);
1444 }
1445
1446
1447 ir_dereference_array::ir_dereference_array(ir_variable *var,
1448 ir_rvalue *array_index)
1449 : ir_dereference(ir_type_dereference_array)
1450 {
1451 void *ctx = ralloc_parent(var);
1452
1453 this->array_index = array_index;
1454 this->set_array(new(ctx) ir_dereference_variable(var));
1455 }
1456
1457
1458 void
1459 ir_dereference_array::set_array(ir_rvalue *value)
1460 {
1461 assert(value != NULL);
1462
1463 this->array = value;
1464
1465 const glsl_type *const vt = this->array->type;
1466
1467 if (vt->is_array()) {
1468 type = vt->fields.array;
1469 } else if (vt->is_matrix()) {
1470 type = vt->column_type();
1471 } else if (vt->is_vector()) {
1472 type = vt->get_base_type();
1473 }
1474 }
1475
1476
1477 ir_dereference_record::ir_dereference_record(ir_rvalue *value,
1478 const char *field)
1479 : ir_dereference(ir_type_dereference_record)
1480 {
1481 assert(value != NULL);
1482
1483 this->record = value;
1484 this->type = this->record->type->field_type(field);
1485 this->field_idx = this->record->type->field_index(field);
1486 }
1487
1488
1489 ir_dereference_record::ir_dereference_record(ir_variable *var,
1490 const char *field)
1491 : ir_dereference(ir_type_dereference_record)
1492 {
1493 void *ctx = ralloc_parent(var);
1494
1495 this->record = new(ctx) ir_dereference_variable(var);
1496 this->type = this->record->type->field_type(field);
1497 this->field_idx = this->record->type->field_index(field);
1498 }
1499
1500 bool
1501 ir_dereference::is_lvalue(const struct _mesa_glsl_parse_state *state) const
1502 {
1503 ir_variable *var = this->variable_referenced();
1504
1505 /* Every l-value derference chain eventually ends in a variable.
1506 */
1507 if ((var == NULL) || var->data.read_only)
1508 return false;
1509
1510 /* From section 4.1.7 of the ARB_bindless_texture spec:
1511 *
1512 * "Samplers can be used as l-values, so can be assigned into and used as
1513 * "out" and "inout" function parameters."
1514 *
1515 * From section 4.1.X of the ARB_bindless_texture spec:
1516 *
1517 * "Images can be used as l-values, so can be assigned into and used as
1518 * "out" and "inout" function parameters."
1519 */
1520 if ((!state || state->has_bindless()) &&
1521 (this->type->contains_sampler() || this->type->contains_image()))
1522 return true;
1523
1524 /* From section 4.1.7 of the GLSL 4.40 spec:
1525 *
1526 * "Opaque variables cannot be treated as l-values; hence cannot
1527 * be used as out or inout function parameters, nor can they be
1528 * assigned into."
1529 */
1530 if (this->type->contains_opaque())
1531 return false;
1532
1533 return true;
1534 }
1535
1536
1537 static const char * const tex_opcode_strs[] = { "tex", "txb", "txl", "txd", "txf", "txf_ms", "txs", "lod", "tg4", "query_levels", "texture_samples", "samples_identical" };
1538
1539 const char *ir_texture::opcode_string()
1540 {
1541 assert((unsigned int) op < ARRAY_SIZE(tex_opcode_strs));
1542 return tex_opcode_strs[op];
1543 }
1544
1545 ir_texture_opcode
1546 ir_texture::get_opcode(const char *str)
1547 {
1548 const int count = sizeof(tex_opcode_strs) / sizeof(tex_opcode_strs[0]);
1549 for (int op = 0; op < count; op++) {
1550 if (strcmp(str, tex_opcode_strs[op]) == 0)
1551 return (ir_texture_opcode) op;
1552 }
1553 return (ir_texture_opcode) -1;
1554 }
1555
1556
1557 void
1558 ir_texture::set_sampler(ir_dereference *sampler, const glsl_type *type)
1559 {
1560 assert(sampler != NULL);
1561 assert(type != NULL);
1562 this->sampler = sampler;
1563 this->type = type;
1564
1565 if (this->op == ir_txs || this->op == ir_query_levels ||
1566 this->op == ir_texture_samples) {
1567 assert(type->base_type == GLSL_TYPE_INT);
1568 } else if (this->op == ir_lod) {
1569 assert(type->vector_elements == 2);
1570 assert(type->is_float());
1571 } else if (this->op == ir_samples_identical) {
1572 assert(type == glsl_type::bool_type);
1573 assert(sampler->type->is_sampler());
1574 assert(sampler->type->sampler_dimensionality == GLSL_SAMPLER_DIM_MS);
1575 } else {
1576 assert(sampler->type->sampled_type == (int) type->base_type);
1577 if (sampler->type->sampler_shadow)
1578 assert(type->vector_elements == 4 || type->vector_elements == 1);
1579 else
1580 assert(type->vector_elements == 4);
1581 }
1582 }
1583
1584
1585 void
1586 ir_swizzle::init_mask(const unsigned *comp, unsigned count)
1587 {
1588 assert((count >= 1) && (count <= 4));
1589
1590 memset(&this->mask, 0, sizeof(this->mask));
1591 this->mask.num_components = count;
1592
1593 unsigned dup_mask = 0;
1594 switch (count) {
1595 case 4:
1596 assert(comp[3] <= 3);
1597 dup_mask |= (1U << comp[3])
1598 & ((1U << comp[0]) | (1U << comp[1]) | (1U << comp[2]));
1599 this->mask.w = comp[3];
1600
1601 case 3:
1602 assert(comp[2] <= 3);
1603 dup_mask |= (1U << comp[2])
1604 & ((1U << comp[0]) | (1U << comp[1]));
1605 this->mask.z = comp[2];
1606
1607 case 2:
1608 assert(comp[1] <= 3);
1609 dup_mask |= (1U << comp[1])
1610 & ((1U << comp[0]));
1611 this->mask.y = comp[1];
1612
1613 case 1:
1614 assert(comp[0] <= 3);
1615 this->mask.x = comp[0];
1616 }
1617
1618 this->mask.has_duplicates = dup_mask != 0;
1619
1620 /* Based on the number of elements in the swizzle and the base type
1621 * (i.e., float, int, unsigned, or bool) of the vector being swizzled,
1622 * generate the type of the resulting value.
1623 */
1624 type = glsl_type::get_instance(val->type->base_type, mask.num_components, 1);
1625 }
1626
1627 ir_swizzle::ir_swizzle(ir_rvalue *val, unsigned x, unsigned y, unsigned z,
1628 unsigned w, unsigned count)
1629 : ir_rvalue(ir_type_swizzle), val(val)
1630 {
1631 const unsigned components[4] = { x, y, z, w };
1632 this->init_mask(components, count);
1633 }
1634
1635 ir_swizzle::ir_swizzle(ir_rvalue *val, const unsigned *comp,
1636 unsigned count)
1637 : ir_rvalue(ir_type_swizzle), val(val)
1638 {
1639 this->init_mask(comp, count);
1640 }
1641
1642 ir_swizzle::ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask)
1643 : ir_rvalue(ir_type_swizzle), val(val), mask(mask)
1644 {
1645 this->type = glsl_type::get_instance(val->type->base_type,
1646 mask.num_components, 1);
1647 }
1648
1649 #define X 1
1650 #define R 5
1651 #define S 9
1652 #define I 13
1653
1654 ir_swizzle *
1655 ir_swizzle::create(ir_rvalue *val, const char *str, unsigned vector_length)
1656 {
1657 void *ctx = ralloc_parent(val);
1658
1659 /* For each possible swizzle character, this table encodes the value in
1660 * \c idx_map that represents the 0th element of the vector. For invalid
1661 * swizzle characters (e.g., 'k'), a special value is used that will allow
1662 * detection of errors.
1663 */
1664 static const unsigned char base_idx[26] = {
1665 /* a b c d e f g h i j k l m */
1666 R, R, I, I, I, I, R, I, I, I, I, I, I,
1667 /* n o p q r s t u v w x y z */
1668 I, I, S, S, R, S, S, I, I, X, X, X, X
1669 };
1670
1671 /* Each valid swizzle character has an entry in the previous table. This
1672 * table encodes the base index encoded in the previous table plus the actual
1673 * index of the swizzle character. When processing swizzles, the first
1674 * character in the string is indexed in the previous table. Each character
1675 * in the string is indexed in this table, and the value found there has the
1676 * value form the first table subtracted. The result must be on the range
1677 * [0,3].
1678 *
1679 * For example, the string "wzyx" will get X from the first table. Each of
1680 * the charcaters will get X+3, X+2, X+1, and X+0 from this table. After
1681 * subtraction, the swizzle values are { 3, 2, 1, 0 }.
1682 *
1683 * The string "wzrg" will get X from the first table. Each of the characters
1684 * will get X+3, X+2, R+0, and R+1 from this table. After subtraction, the
1685 * swizzle values are { 3, 2, 4, 5 }. Since 4 and 5 are outside the range
1686 * [0,3], the error is detected.
1687 */
1688 static const unsigned char idx_map[26] = {
1689 /* a b c d e f g h i j k l m */
1690 R+3, R+2, 0, 0, 0, 0, R+1, 0, 0, 0, 0, 0, 0,
1691 /* n o p q r s t u v w x y z */
1692 0, 0, S+2, S+3, R+0, S+0, S+1, 0, 0, X+3, X+0, X+1, X+2
1693 };
1694
1695 int swiz_idx[4] = { 0, 0, 0, 0 };
1696 unsigned i;
1697
1698
1699 /* Validate the first character in the swizzle string and look up the base
1700 * index value as described above.
1701 */
1702 if ((str[0] < 'a') || (str[0] > 'z'))
1703 return NULL;
1704
1705 const unsigned base = base_idx[str[0] - 'a'];
1706
1707
1708 for (i = 0; (i < 4) && (str[i] != '\0'); i++) {
1709 /* Validate the next character, and, as described above, convert it to a
1710 * swizzle index.
1711 */
1712 if ((str[i] < 'a') || (str[i] > 'z'))
1713 return NULL;
1714
1715 swiz_idx[i] = idx_map[str[i] - 'a'] - base;
1716 if ((swiz_idx[i] < 0) || (swiz_idx[i] >= (int) vector_length))
1717 return NULL;
1718 }
1719
1720 if (str[i] != '\0')
1721 return NULL;
1722
1723 return new(ctx) ir_swizzle(val, swiz_idx[0], swiz_idx[1], swiz_idx[2],
1724 swiz_idx[3], i);
1725 }
1726
1727 #undef X
1728 #undef R
1729 #undef S
1730 #undef I
1731
1732 ir_variable *
1733 ir_swizzle::variable_referenced() const
1734 {
1735 return this->val->variable_referenced();
1736 }
1737
1738
1739 bool ir_variable::temporaries_allocate_names = false;
1740
1741 const char ir_variable::tmp_name[] = "compiler_temp";
1742
1743 ir_variable::ir_variable(const struct glsl_type *type, const char *name,
1744 ir_variable_mode mode)
1745 : ir_instruction(ir_type_variable)
1746 {
1747 this->type = type;
1748
1749 if (mode == ir_var_temporary && !ir_variable::temporaries_allocate_names)
1750 name = NULL;
1751
1752 /* The ir_variable clone method may call this constructor with name set to
1753 * tmp_name.
1754 */
1755 assert(name != NULL
1756 || mode == ir_var_temporary
1757 || mode == ir_var_function_in
1758 || mode == ir_var_function_out
1759 || mode == ir_var_function_inout);
1760 assert(name != ir_variable::tmp_name
1761 || mode == ir_var_temporary);
1762 if (mode == ir_var_temporary
1763 && (name == NULL || name == ir_variable::tmp_name)) {
1764 this->name = ir_variable::tmp_name;
1765 } else if (name == NULL ||
1766 strlen(name) < ARRAY_SIZE(this->name_storage)) {
1767 strcpy(this->name_storage, name ? name : "");
1768 this->name = this->name_storage;
1769 } else {
1770 this->name = ralloc_strdup(this, name);
1771 }
1772
1773 this->u.max_ifc_array_access = NULL;
1774
1775 this->data.explicit_location = false;
1776 this->data.explicit_index = false;
1777 this->data.explicit_binding = false;
1778 this->data.explicit_component = false;
1779 this->data.has_initializer = false;
1780 this->data.is_unmatched_generic_inout = false;
1781 this->data.is_xfb_only = false;
1782 this->data.explicit_xfb_buffer = false;
1783 this->data.explicit_xfb_offset = false;
1784 this->data.explicit_xfb_stride = false;
1785 this->data.location = -1;
1786 this->data.location_frac = 0;
1787 this->data.matrix_layout = GLSL_MATRIX_LAYOUT_INHERITED;
1788 this->data.from_named_ifc_block = false;
1789 this->data.must_be_shader_input = false;
1790 this->data.index = 0;
1791 this->data.binding = 0;
1792 this->data.warn_extension_index = 0;
1793 this->constant_value = NULL;
1794 this->constant_initializer = NULL;
1795 this->data.depth_layout = ir_depth_layout_none;
1796 this->data.used = false;
1797 this->data.assigned = false;
1798 this->data.always_active_io = false;
1799 this->data.read_only = false;
1800 this->data.centroid = false;
1801 this->data.sample = false;
1802 this->data.patch = false;
1803 this->data.explicit_invariant = false;
1804 this->data.invariant = false;
1805 this->data.precise = false;
1806 this->data.how_declared = ir_var_declared_normally;
1807 this->data.mode = mode;
1808 this->data.interpolation = INTERP_MODE_NONE;
1809 this->data.max_array_access = -1;
1810 this->data.offset = 0;
1811 this->data.precision = GLSL_PRECISION_NONE;
1812 this->data.memory_read_only = false;
1813 this->data.memory_write_only = false;
1814 this->data.memory_coherent = false;
1815 this->data.memory_volatile = false;
1816 this->data.memory_restrict = false;
1817 this->data.from_ssbo_unsized_array = false;
1818 this->data.implicit_sized_array = false;
1819 this->data.fb_fetch_output = false;
1820 this->data.bindless = false;
1821 this->data.bound = false;
1822 this->data.image_format = PIPE_FORMAT_NONE;
1823 this->data._num_state_slots = 0;
1824 this->data.param_index = 0;
1825 this->data.stream = 0;
1826 this->data.xfb_buffer = -1;
1827 this->data.xfb_stride = -1;
1828
1829 this->interface_type = NULL;
1830
1831 if (type != NULL) {
1832 if (type->is_interface())
1833 this->init_interface_type(type);
1834 else if (type->without_array()->is_interface())
1835 this->init_interface_type(type->without_array());
1836 }
1837 }
1838
1839
1840 const char *
1841 interpolation_string(unsigned interpolation)
1842 {
1843 switch (interpolation) {
1844 case INTERP_MODE_NONE: return "no";
1845 case INTERP_MODE_SMOOTH: return "smooth";
1846 case INTERP_MODE_FLAT: return "flat";
1847 case INTERP_MODE_NOPERSPECTIVE: return "noperspective";
1848 }
1849
1850 assert(!"Should not get here.");
1851 return "";
1852 }
1853
1854 const char *const ir_variable::warn_extension_table[] = {
1855 "",
1856 "GL_ARB_shader_stencil_export",
1857 "GL_AMD_shader_stencil_export",
1858 };
1859
1860 void
1861 ir_variable::enable_extension_warning(const char *extension)
1862 {
1863 for (unsigned i = 0; i < ARRAY_SIZE(warn_extension_table); i++) {
1864 if (strcmp(warn_extension_table[i], extension) == 0) {
1865 this->data.warn_extension_index = i;
1866 return;
1867 }
1868 }
1869
1870 assert(!"Should not get here.");
1871 this->data.warn_extension_index = 0;
1872 }
1873
1874 const char *
1875 ir_variable::get_extension_warning() const
1876 {
1877 return this->data.warn_extension_index == 0
1878 ? NULL : warn_extension_table[this->data.warn_extension_index];
1879 }
1880
1881 ir_function_signature::ir_function_signature(const glsl_type *return_type,
1882 builtin_available_predicate b)
1883 : ir_instruction(ir_type_function_signature),
1884 return_type(return_type), is_defined(false),
1885 return_precision(GLSL_PRECISION_NONE),
1886 intrinsic_id(ir_intrinsic_invalid), builtin_avail(b), _function(NULL)
1887 {
1888 this->origin = NULL;
1889 }
1890
1891
1892 bool
1893 ir_function_signature::is_builtin() const
1894 {
1895 return builtin_avail != NULL;
1896 }
1897
1898
1899 bool
1900 ir_function_signature::is_builtin_available(const _mesa_glsl_parse_state *state) const
1901 {
1902 /* We can't call the predicate without a state pointer, so just say that
1903 * the signature is available. At compile time, we need the filtering,
1904 * but also receive a valid state pointer. At link time, we're resolving
1905 * imported built-in prototypes to their definitions, which will always
1906 * be an exact match. So we can skip the filtering.
1907 */
1908 if (state == NULL)
1909 return true;
1910
1911 assert(builtin_avail != NULL);
1912 return builtin_avail(state);
1913 }
1914
1915
1916 static bool
1917 modes_match(unsigned a, unsigned b)
1918 {
1919 if (a == b)
1920 return true;
1921
1922 /* Accept "in" vs. "const in" */
1923 if ((a == ir_var_const_in && b == ir_var_function_in) ||
1924 (b == ir_var_const_in && a == ir_var_function_in))
1925 return true;
1926
1927 return false;
1928 }
1929
1930
1931 const char *
1932 ir_function_signature::qualifiers_match(exec_list *params)
1933 {
1934 /* check that the qualifiers match. */
1935 foreach_two_lists(a_node, &this->parameters, b_node, params) {
1936 ir_variable *a = (ir_variable *) a_node;
1937 ir_variable *b = (ir_variable *) b_node;
1938
1939 if (a->data.read_only != b->data.read_only ||
1940 !modes_match(a->data.mode, b->data.mode) ||
1941 a->data.interpolation != b->data.interpolation ||
1942 a->data.centroid != b->data.centroid ||
1943 a->data.sample != b->data.sample ||
1944 a->data.patch != b->data.patch ||
1945 a->data.memory_read_only != b->data.memory_read_only ||
1946 a->data.memory_write_only != b->data.memory_write_only ||
1947 a->data.memory_coherent != b->data.memory_coherent ||
1948 a->data.memory_volatile != b->data.memory_volatile ||
1949 a->data.memory_restrict != b->data.memory_restrict) {
1950
1951 /* parameter a's qualifiers don't match */
1952 return a->name;
1953 }
1954 }
1955 return NULL;
1956 }
1957
1958
1959 void
1960 ir_function_signature::replace_parameters(exec_list *new_params)
1961 {
1962 /* Destroy all of the previous parameter information. If the previous
1963 * parameter information comes from the function prototype, it may either
1964 * specify incorrect parameter names or not have names at all.
1965 */
1966 new_params->move_nodes_to(&parameters);
1967 }
1968
1969
1970 ir_function::ir_function(const char *name)
1971 : ir_instruction(ir_type_function)
1972 {
1973 this->subroutine_index = -1;
1974 this->name = ralloc_strdup(this, name);
1975 }
1976
1977
1978 bool
1979 ir_function::has_user_signature()
1980 {
1981 foreach_in_list(ir_function_signature, sig, &this->signatures) {
1982 if (!sig->is_builtin())
1983 return true;
1984 }
1985 return false;
1986 }
1987
1988
1989 ir_rvalue *
1990 ir_rvalue::error_value(void *mem_ctx)
1991 {
1992 ir_rvalue *v = new(mem_ctx) ir_rvalue(ir_type_unset);
1993
1994 v->type = glsl_type::error_type;
1995 return v;
1996 }
1997
1998
1999 void
2000 visit_exec_list(exec_list *list, ir_visitor *visitor)
2001 {
2002 foreach_in_list_safe(ir_instruction, node, list) {
2003 node->accept(visitor);
2004 }
2005 }
2006
2007
2008 static void
2009 steal_memory(ir_instruction *ir, void *new_ctx)
2010 {
2011 ir_variable *var = ir->as_variable();
2012 ir_function *fn = ir->as_function();
2013 ir_constant *constant = ir->as_constant();
2014 if (var != NULL && var->constant_value != NULL)
2015 steal_memory(var->constant_value, ir);
2016
2017 if (var != NULL && var->constant_initializer != NULL)
2018 steal_memory(var->constant_initializer, ir);
2019
2020 if (fn != NULL && fn->subroutine_types)
2021 ralloc_steal(new_ctx, fn->subroutine_types);
2022
2023 /* The components of aggregate constants are not visited by the normal
2024 * visitor, so steal their values by hand.
2025 */
2026 if (constant != NULL &&
2027 (constant->type->is_array() || constant->type->is_struct())) {
2028 for (unsigned int i = 0; i < constant->type->length; i++) {
2029 steal_memory(constant->const_elements[i], ir);
2030 }
2031 }
2032
2033 ralloc_steal(new_ctx, ir);
2034 }
2035
2036
2037 void
2038 reparent_ir(exec_list *list, void *mem_ctx)
2039 {
2040 foreach_in_list(ir_instruction, node, list) {
2041 visit_tree(node, steal_memory, mem_ctx);
2042 }
2043 }
2044
2045
2046 static ir_rvalue *
2047 try_min_one(ir_rvalue *ir)
2048 {
2049 ir_expression *expr = ir->as_expression();
2050
2051 if (!expr || expr->operation != ir_binop_min)
2052 return NULL;
2053
2054 if (expr->operands[0]->is_one())
2055 return expr->operands[1];
2056
2057 if (expr->operands[1]->is_one())
2058 return expr->operands[0];
2059
2060 return NULL;
2061 }
2062
2063 static ir_rvalue *
2064 try_max_zero(ir_rvalue *ir)
2065 {
2066 ir_expression *expr = ir->as_expression();
2067
2068 if (!expr || expr->operation != ir_binop_max)
2069 return NULL;
2070
2071 if (expr->operands[0]->is_zero())
2072 return expr->operands[1];
2073
2074 if (expr->operands[1]->is_zero())
2075 return expr->operands[0];
2076
2077 return NULL;
2078 }
2079
2080 ir_rvalue *
2081 ir_rvalue::as_rvalue_to_saturate()
2082 {
2083 ir_expression *expr = this->as_expression();
2084
2085 if (!expr)
2086 return NULL;
2087
2088 ir_rvalue *max_zero = try_max_zero(expr);
2089 if (max_zero) {
2090 return try_min_one(max_zero);
2091 } else {
2092 ir_rvalue *min_one = try_min_one(expr);
2093 if (min_one) {
2094 return try_max_zero(min_one);
2095 }
2096 }
2097
2098 return NULL;
2099 }
2100
2101
2102 unsigned
2103 vertices_per_prim(GLenum prim)
2104 {
2105 switch (prim) {
2106 case GL_POINTS:
2107 return 1;
2108 case GL_LINES:
2109 return 2;
2110 case GL_TRIANGLES:
2111 return 3;
2112 case GL_LINES_ADJACENCY:
2113 return 4;
2114 case GL_TRIANGLES_ADJACENCY:
2115 return 6;
2116 default:
2117 assert(!"Bad primitive");
2118 return 3;
2119 }
2120 }
2121
2122 /**
2123 * Generate a string describing the mode of a variable
2124 */
2125 const char *
2126 mode_string(const ir_variable *var)
2127 {
2128 switch (var->data.mode) {
2129 case ir_var_auto:
2130 return (var->data.read_only) ? "global constant" : "global variable";
2131
2132 case ir_var_uniform:
2133 return "uniform";
2134
2135 case ir_var_shader_storage:
2136 return "buffer";
2137
2138 case ir_var_shader_in:
2139 return "shader input";
2140
2141 case ir_var_shader_out:
2142 return "shader output";
2143
2144 case ir_var_function_in:
2145 case ir_var_const_in:
2146 return "function input";
2147
2148 case ir_var_function_out:
2149 return "function output";
2150
2151 case ir_var_function_inout:
2152 return "function inout";
2153
2154 case ir_var_system_value:
2155 return "shader input";
2156
2157 case ir_var_temporary:
2158 return "compiler temporary";
2159
2160 case ir_var_mode_count:
2161 break;
2162 }
2163
2164 assert(!"Should not get here.");
2165 return "invalid variable";
2166 }