projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
glsl: Add b2f16 and f162b conversion operations
[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 "util/half_float.h"
26 #include "compiler/glsl_types.h"
27 #include "glsl_parser_extras.h"
28
29
30 ir_rvalue::ir_rvalue(enum ir_node_type t)
31 : ir_instruction(t)
32 {
33 this->type = glsl_type::error_type;
34 }
35
36 bool ir_rvalue::is_zero() const
37 {
38 return false;
39 }
40
41 bool ir_rvalue::is_one() const
42 {
43 return false;
44 }
45
46 bool ir_rvalue::is_negative_one() const
47 {
48 return false;
49 }
50
51 /**
52 * Modify the swizzle make to move one component to another
53 *
54 * \param m IR swizzle to be modified
55 * \param from Component in the RHS that is to be swizzled
56 * \param to Desired swizzle location of \c from
57 */
58 static void
59 update_rhs_swizzle(ir_swizzle_mask &m, unsigned from, unsigned to)
60 {
61 switch (to) {
62 case 0: m.x = from; break;
63 case 1: m.y = from; break;
64 case 2: m.z = from; break;
65 case 3: m.w = from; break;
66 default: assert(!"Should not get here.");
67 }
68 }
69
70 void
71 ir_assignment::set_lhs(ir_rvalue *lhs)
72 {
73 void *mem_ctx = this;
74 bool swizzled = false;
75
76 while (lhs != NULL) {
77 ir_swizzle *swiz = lhs->as_swizzle();
78
79 if (swiz == NULL)
80 break;
81
82 unsigned write_mask = 0;
83 ir_swizzle_mask rhs_swiz = { 0, 0, 0, 0, 0, 0 };
84
85 for (unsigned i = 0; i < swiz->mask.num_components; i++) {
86 unsigned c = 0;
87
88 switch (i) {
89 case 0: c = swiz->mask.x; break;
90 case 1: c = swiz->mask.y; break;
91 case 2: c = swiz->mask.z; break;
92 case 3: c = swiz->mask.w; break;
93 default: assert(!"Should not get here.");
94 }
95
96 write_mask |= (((this->write_mask >> i) & 1) << c);
97 update_rhs_swizzle(rhs_swiz, i, c);
98 rhs_swiz.num_components = swiz->val->type->vector_elements;
99 }
100
101 this->write_mask = write_mask;
102 lhs = swiz->val;
103
104 this->rhs = new(mem_ctx) ir_swizzle(this->rhs, rhs_swiz);
105 swizzled = true;
106 }
107
108 if (swizzled) {
109 /* Now, RHS channels line up with the LHS writemask. Collapse it
110 * to just the channels that will be written.
111 */
112 ir_swizzle_mask rhs_swiz = { 0, 0, 0, 0, 0, 0 };
113 int rhs_chan = 0;
114 for (int i = 0; i < 4; i++) {
115 if (write_mask & (1 << i))
116 update_rhs_swizzle(rhs_swiz, i, rhs_chan++);
117 }
118 rhs_swiz.num_components = rhs_chan;
119 this->rhs = new(mem_ctx) ir_swizzle(this->rhs, rhs_swiz);
120 }
121
122 assert((lhs == NULL) || lhs->as_dereference());
123
124 this->lhs = (ir_dereference *) lhs;
125 }
126
127 ir_variable *
128 ir_assignment::whole_variable_written()
129 {
130 ir_variable *v = this->lhs->whole_variable_referenced();
131
132 if (v == NULL)
133 return NULL;
134
135 if (v->type->is_scalar())
136 return v;
137
138 if (v->type->is_vector()) {
139 const unsigned mask = (1U << v->type->vector_elements) - 1;
140
141 if (mask != this->write_mask)
142 return NULL;
143 }
144
145 /* Either all the vector components are assigned or the variable is some
146 * composite type (and the whole thing is assigned.
147 */
148 return v;
149 }
150
151 ir_assignment::ir_assignment(ir_dereference *lhs, ir_rvalue *rhs,
152 ir_rvalue *condition, unsigned write_mask)
153 : ir_instruction(ir_type_assignment)
154 {
155 this->condition = condition;
156 this->rhs = rhs;
157 this->lhs = lhs;
158 this->write_mask = write_mask;
159
160 if (lhs->type->is_scalar() || lhs->type->is_vector()) {
161 int lhs_components = 0;
162 for (int i = 0; i < 4; i++) {
163 if (write_mask & (1 << i))
164 lhs_components++;
165 }
166
167 assert(lhs_components == this->rhs->type->vector_elements);
168 }
169 }
170
171 ir_assignment::ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs,
172 ir_rvalue *condition)
173 : ir_instruction(ir_type_assignment)
174 {
175 this->condition = condition;
176 this->rhs = rhs;
177
178 /* If the RHS is a vector type, assume that all components of the vector
179 * type are being written to the LHS. The write mask comes from the RHS
180 * because we can have a case where the LHS is a vec4 and the RHS is a
181 * vec3. In that case, the assignment is:
182 *
183 * (assign (...) (xyz) (var_ref lhs) (var_ref rhs))
184 */
185 if (rhs->type->is_vector())
186 this->write_mask = (1U << rhs->type->vector_elements) - 1;
187 else if (rhs->type->is_scalar())
188 this->write_mask = 1;
189 else
190 this->write_mask = 0;
191
192 this->set_lhs(lhs);
193 }
194
195 ir_expression::ir_expression(int op, const struct glsl_type *type,
196 ir_rvalue *op0, ir_rvalue *op1,
197 ir_rvalue *op2, ir_rvalue *op3)
198 : ir_rvalue(ir_type_expression)
199 {
200 this->type = type;
201 this->operation = ir_expression_operation(op);
202 this->operands[0] = op0;
203 this->operands[1] = op1;
204 this->operands[2] = op2;
205 this->operands[3] = op3;
206 init_num_operands();
207
208 #ifndef NDEBUG
209 for (unsigned i = num_operands; i < 4; i++) {
210 assert(this->operands[i] == NULL);
211 }
212
213 for (unsigned i = 0; i < num_operands; i++) {
214 assert(this->operands[i] != NULL);
215 }
216 #endif
217 }
218
219 ir_expression::ir_expression(int op, ir_rvalue *op0)
220 : ir_rvalue(ir_type_expression)
221 {
222 this->operation = ir_expression_operation(op);
223 this->operands[0] = op0;
224 this->operands[1] = NULL;
225 this->operands[2] = NULL;
226 this->operands[3] = NULL;
227
228 assert(op <= ir_last_unop);
229 init_num_operands();
230 assert(num_operands == 1);
231 assert(this->operands[0]);
232
233 switch (this->operation) {
234 case ir_unop_bit_not:
235 case ir_unop_logic_not:
236 case ir_unop_neg:
237 case ir_unop_abs:
238 case ir_unop_sign:
239 case ir_unop_rcp:
240 case ir_unop_rsq:
241 case ir_unop_sqrt:
242 case ir_unop_exp:
243 case ir_unop_log:
244 case ir_unop_exp2:
245 case ir_unop_log2:
246 case ir_unop_trunc:
247 case ir_unop_ceil:
248 case ir_unop_floor:
249 case ir_unop_fract:
250 case ir_unop_round_even:
251 case ir_unop_sin:
252 case ir_unop_cos:
253 case ir_unop_dFdx:
254 case ir_unop_dFdx_coarse:
255 case ir_unop_dFdx_fine:
256 case ir_unop_dFdy:
257 case ir_unop_dFdy_coarse:
258 case ir_unop_dFdy_fine:
259 case ir_unop_bitfield_reverse:
260 case ir_unop_interpolate_at_centroid:
261 case ir_unop_clz:
262 case ir_unop_saturate:
263 case ir_unop_atan:
264 this->type = op0->type;
265 break;
266
267 case ir_unop_f2i:
268 case ir_unop_b2i:
269 case ir_unop_u2i:
270 case ir_unop_d2i:
271 case ir_unop_bitcast_f2i:
272 case ir_unop_bit_count:
273 case ir_unop_find_msb:
274 case ir_unop_find_lsb:
275 case ir_unop_subroutine_to_int:
276 case ir_unop_i642i:
277 case ir_unop_u642i:
278 this->type = glsl_type::get_instance(GLSL_TYPE_INT,
279 op0->type->vector_elements, 1);
280 break;
281
282 case ir_unop_b2f:
283 case ir_unop_i2f:
284 case ir_unop_u2f:
285 case ir_unop_d2f:
286 case ir_unop_f162f:
287 case ir_unop_bitcast_i2f:
288 case ir_unop_bitcast_u2f:
289 case ir_unop_i642f:
290 case ir_unop_u642f:
291 this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT,
292 op0->type->vector_elements, 1);
293 break;
294
295 case ir_unop_f2f16:
296 case ir_unop_b2f16:
297 this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT16,
298 op0->type->vector_elements, 1);
299 break;
300
301 case ir_unop_f2b:
302 case ir_unop_i2b:
303 case ir_unop_d2b:
304 case ir_unop_f162b:
305 case ir_unop_i642b:
306 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL,
307 op0->type->vector_elements, 1);
308 break;
309
310 case ir_unop_f2d:
311 case ir_unop_i2d:
312 case ir_unop_u2d:
313 case ir_unop_i642d:
314 case ir_unop_u642d:
315 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE,
316 op0->type->vector_elements, 1);
317 break;
318
319 case ir_unop_i2u:
320 case ir_unop_f2u:
321 case ir_unop_d2u:
322 case ir_unop_bitcast_f2u:
323 case ir_unop_i642u:
324 case ir_unop_u642u:
325 this->type = glsl_type::get_instance(GLSL_TYPE_UINT,
326 op0->type->vector_elements, 1);
327 break;
328
329 case ir_unop_i2i64:
330 case ir_unop_u2i64:
331 case ir_unop_b2i64:
332 case ir_unop_f2i64:
333 case ir_unop_d2i64:
334 case ir_unop_u642i64:
335 this->type = glsl_type::get_instance(GLSL_TYPE_INT64,
336 op0->type->vector_elements, 1);
337 break;
338
339 case ir_unop_i2u64:
340 case ir_unop_u2u64:
341 case ir_unop_f2u64:
342 case ir_unop_d2u64:
343 case ir_unop_i642u64:
344 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64,
345 op0->type->vector_elements, 1);
346 break;
347 case ir_unop_noise:
348 this->type = glsl_type::float_type;
349 break;
350
351 case ir_unop_unpack_double_2x32:
352 case ir_unop_unpack_uint_2x32:
353 this->type = glsl_type::uvec2_type;
354 break;
355
356 case ir_unop_unpack_int_2x32:
357 this->type = glsl_type::ivec2_type;
358 break;
359
360 case ir_unop_pack_snorm_2x16:
361 case ir_unop_pack_snorm_4x8:
362 case ir_unop_pack_unorm_2x16:
363 case ir_unop_pack_unorm_4x8:
364 case ir_unop_pack_half_2x16:
365 this->type = glsl_type::uint_type;
366 break;
367
368 case ir_unop_pack_double_2x32:
369 this->type = glsl_type::double_type;
370 break;
371
372 case ir_unop_pack_int_2x32:
373 this->type = glsl_type::int64_t_type;
374 break;
375
376 case ir_unop_pack_uint_2x32:
377 this->type = glsl_type::uint64_t_type;
378 break;
379
380 case ir_unop_unpack_snorm_2x16:
381 case ir_unop_unpack_unorm_2x16:
382 case ir_unop_unpack_half_2x16:
383 this->type = glsl_type::vec2_type;
384 break;
385
386 case ir_unop_unpack_snorm_4x8:
387 case ir_unop_unpack_unorm_4x8:
388 this->type = glsl_type::vec4_type;
389 break;
390
391 case ir_unop_unpack_sampler_2x32:
392 case ir_unop_unpack_image_2x32:
393 this->type = glsl_type::uvec2_type;
394 break;
395
396 case ir_unop_pack_sampler_2x32:
397 case ir_unop_pack_image_2x32:
398 this->type = op0->type;
399 break;
400
401 case ir_unop_frexp_sig:
402 this->type = op0->type;
403 break;
404 case ir_unop_frexp_exp:
405 this->type = glsl_type::get_instance(GLSL_TYPE_INT,
406 op0->type->vector_elements, 1);
407 break;
408
409 case ir_unop_get_buffer_size:
410 case ir_unop_ssbo_unsized_array_length:
411 this->type = glsl_type::int_type;
412 break;
413
414 case ir_unop_bitcast_i642d:
415 case ir_unop_bitcast_u642d:
416 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE,
417 op0->type->vector_elements, 1);
418 break;
419
420 case ir_unop_bitcast_d2i64:
421 this->type = glsl_type::get_instance(GLSL_TYPE_INT64,
422 op0->type->vector_elements, 1);
423 break;
424 case ir_unop_bitcast_d2u64:
425 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64,
426 op0->type->vector_elements, 1);
427 break;
428
429 default:
430 assert(!"not reached: missing automatic type setup for ir_expression");
431 this->type = op0->type;
432 break;
433 }
434 }
435
436 ir_expression::ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1)
437 : ir_rvalue(ir_type_expression)
438 {
439 this->operation = ir_expression_operation(op);
440 this->operands[0] = op0;
441 this->operands[1] = op1;
442 this->operands[2] = NULL;
443 this->operands[3] = NULL;
444
445 assert(op > ir_last_unop);
446 init_num_operands();
447 assert(num_operands == 2);
448 for (unsigned i = 0; i < num_operands; i++) {
449 assert(this->operands[i] != NULL);
450 }
451
452 switch (this->operation) {
453 case ir_binop_all_equal:
454 case ir_binop_any_nequal:
455 this->type = glsl_type::bool_type;
456 break;
457
458 case ir_binop_add:
459 case ir_binop_sub:
460 case ir_binop_min:
461 case ir_binop_max:
462 case ir_binop_pow:
463 case ir_binop_mul:
464 case ir_binop_div:
465 case ir_binop_mod:
466 case ir_binop_atan2:
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 if (this->operation == ir_binop_mul) {
473 this->type = glsl_type::get_mul_type(op0->type, op1->type);
474 } else {
475 assert(op0->type == op1->type);
476 this->type = op0->type;
477 }
478 }
479 break;
480
481 case ir_binop_logic_and:
482 case ir_binop_logic_xor:
483 case ir_binop_logic_or:
484 case ir_binop_bit_and:
485 case ir_binop_bit_xor:
486 case ir_binop_bit_or:
487 assert(!op0->type->is_matrix());
488 assert(!op1->type->is_matrix());
489 if (op0->type->is_scalar()) {
490 this->type = op1->type;
491 } else if (op1->type->is_scalar()) {
492 this->type = op0->type;
493 } else {
494 assert(op0->type->vector_elements == op1->type->vector_elements);
495 this->type = op0->type;
496 }
497 break;
498
499 case ir_binop_equal:
500 case ir_binop_nequal:
501 case ir_binop_gequal:
502 case ir_binop_less:
503 assert(op0->type == op1->type);
504 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL,
505 op0->type->vector_elements, 1);
506 break;
507
508 case ir_binop_dot:
509 this->type = op0->type->get_base_type();
510 break;
511
512 case ir_binop_imul_high:
513 case ir_binop_mul_32x16:
514 case ir_binop_carry:
515 case ir_binop_borrow:
516 case ir_binop_lshift:
517 case ir_binop_rshift:
518 case ir_binop_ldexp:
519 case ir_binop_interpolate_at_offset:
520 case ir_binop_interpolate_at_sample:
521 this->type = op0->type;
522 break;
523
524 case ir_binop_add_sat:
525 case ir_binop_sub_sat:
526 case ir_binop_avg:
527 case ir_binop_avg_round:
528 assert(op0->type == op1->type);
529 this->type = op0->type;
530 break;
531
532 case ir_binop_abs_sub: {
533 enum glsl_base_type base;
534
535 assert(op0->type == op1->type);
536
537 switch (op0->type->base_type) {
538 case GLSL_TYPE_UINT:
539 case GLSL_TYPE_INT:
540 base = GLSL_TYPE_UINT;
541 break;
542 case GLSL_TYPE_UINT8:
543 case GLSL_TYPE_INT8:
544 base = GLSL_TYPE_UINT8;
545 break;
546 case GLSL_TYPE_UINT16:
547 case GLSL_TYPE_INT16:
548 base = GLSL_TYPE_UINT16;
549 break;
550 case GLSL_TYPE_UINT64:
551 case GLSL_TYPE_INT64:
552 base = GLSL_TYPE_UINT64;
553 break;
554 default:
555 unreachable(!"Invalid base type.");
556 }
557
558 this->type = glsl_type::get_instance(base, op0->type->vector_elements, 1);
559 break;
560 }
561
562 case ir_binop_vector_extract:
563 this->type = op0->type->get_scalar_type();
564 break;
565
566 default:
567 assert(!"not reached: missing automatic type setup for ir_expression");
568 this->type = glsl_type::float_type;
569 }
570 }
571
572 ir_expression::ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1,
573 ir_rvalue *op2)
574 : ir_rvalue(ir_type_expression)
575 {
576 this->operation = ir_expression_operation(op);
577 this->operands[0] = op0;
578 this->operands[1] = op1;
579 this->operands[2] = op2;
580 this->operands[3] = NULL;
581
582 assert(op > ir_last_binop && op <= ir_last_triop);
583 init_num_operands();
584 assert(num_operands == 3);
585 for (unsigned i = 0; i < num_operands; i++) {
586 assert(this->operands[i] != NULL);
587 }
588
589 switch (this->operation) {
590 case ir_triop_fma:
591 case ir_triop_lrp:
592 case ir_triop_bitfield_extract:
593 case ir_triop_vector_insert:
594 this->type = op0->type;
595 break;
596
597 case ir_triop_csel:
598 this->type = op1->type;
599 break;
600
601 default:
602 assert(!"not reached: missing automatic type setup for ir_expression");
603 this->type = glsl_type::float_type;
604 }
605 }
606
607 /**
608 * This is only here for ir_reader to used for testing purposes. Please use
609 * the precomputed num_operands field if you need the number of operands.
610 */
611 unsigned
612 ir_expression::get_num_operands(ir_expression_operation op)
613 {
614 assert(op <= ir_last_opcode);
615
616 if (op <= ir_last_unop)
617 return 1;
618
619 if (op <= ir_last_binop)
620 return 2;
621
622 if (op <= ir_last_triop)
623 return 3;
624
625 if (op <= ir_last_quadop)
626 return 4;
627
628 unreachable("Could not calculate number of operands");
629 }
630
631 #include "ir_expression_operation_strings.h"
632
633 const char*
634 depth_layout_string(ir_depth_layout layout)
635 {
636 switch(layout) {
637 case ir_depth_layout_none: return "";
638 case ir_depth_layout_any: return "depth_any";
639 case ir_depth_layout_greater: return "depth_greater";
640 case ir_depth_layout_less: return "depth_less";
641 case ir_depth_layout_unchanged: return "depth_unchanged";
642
643 default:
644 assert(0);
645 return "";
646 }
647 }
648
649 ir_expression_operation
650 ir_expression::get_operator(const char *str)
651 {
652 for (int op = 0; op <= int(ir_last_opcode); op++) {
653 if (strcmp(str, ir_expression_operation_strings[op]) == 0)
654 return (ir_expression_operation) op;
655 }
656 return (ir_expression_operation) -1;
657 }
658
659 ir_variable *
660 ir_expression::variable_referenced() const
661 {
662 switch (operation) {
663 case ir_binop_vector_extract:
664 case ir_triop_vector_insert:
665 /* We get these for things like a[0] where a is a vector type. In these
666 * cases we want variable_referenced() to return the actual vector
667 * variable this is wrapping.
668 */
669 return operands[0]->variable_referenced();
670 default:
671 return ir_rvalue::variable_referenced();
672 }
673 }
674
675 ir_constant::ir_constant()
676 : ir_rvalue(ir_type_constant)
677 {
678 this->const_elements = NULL;
679 }
680
681 ir_constant::ir_constant(const struct glsl_type *type,
682 const ir_constant_data *data)
683 : ir_rvalue(ir_type_constant)
684 {
685 this->const_elements = NULL;
686
687 assert((type->base_type >= GLSL_TYPE_UINT)
688 && (type->base_type <= GLSL_TYPE_IMAGE));
689
690 this->type = type;
691 memcpy(& this->value, data, sizeof(this->value));
692 }
693
694 ir_constant::ir_constant(float16_t f16, unsigned vector_elements)
695 : ir_rvalue(ir_type_constant)
696 {
697 assert(vector_elements <= 4);
698 this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT16, vector_elements, 1);
699 for (unsigned i = 0; i < vector_elements; i++) {
700 this->value.f16[i] = f16.bits;
701 }
702 for (unsigned i = vector_elements; i < 16; i++) {
703 this->value.f[i] = 0;
704 }
705 }
706
707 ir_constant::ir_constant(float f, unsigned vector_elements)
708 : ir_rvalue(ir_type_constant)
709 {
710 assert(vector_elements <= 4);
711 this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT, vector_elements, 1);
712 for (unsigned i = 0; i < vector_elements; i++) {
713 this->value.f[i] = f;
714 }
715 for (unsigned i = vector_elements; i < 16; i++) {
716 this->value.f[i] = 0;
717 }
718 }
719
720 ir_constant::ir_constant(double d, unsigned vector_elements)
721 : ir_rvalue(ir_type_constant)
722 {
723 assert(vector_elements <= 4);
724 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE, vector_elements, 1);
725 for (unsigned i = 0; i < vector_elements; i++) {
726 this->value.d[i] = d;
727 }
728 for (unsigned i = vector_elements; i < 16; i++) {
729 this->value.d[i] = 0.0;
730 }
731 }
732
733 ir_constant::ir_constant(unsigned int u, unsigned vector_elements)
734 : ir_rvalue(ir_type_constant)
735 {
736 assert(vector_elements <= 4);
737 this->type = glsl_type::get_instance(GLSL_TYPE_UINT, vector_elements, 1);
738 for (unsigned i = 0; i < vector_elements; i++) {
739 this->value.u[i] = u;
740 }
741 for (unsigned i = vector_elements; i < 16; i++) {
742 this->value.u[i] = 0;
743 }
744 }
745
746 ir_constant::ir_constant(int integer, unsigned vector_elements)
747 : ir_rvalue(ir_type_constant)
748 {
749 assert(vector_elements <= 4);
750 this->type = glsl_type::get_instance(GLSL_TYPE_INT, vector_elements, 1);
751 for (unsigned i = 0; i < vector_elements; i++) {
752 this->value.i[i] = integer;
753 }
754 for (unsigned i = vector_elements; i < 16; i++) {
755 this->value.i[i] = 0;
756 }
757 }
758
759 ir_constant::ir_constant(uint64_t u64, unsigned vector_elements)
760 : ir_rvalue(ir_type_constant)
761 {
762 assert(vector_elements <= 4);
763 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64, vector_elements, 1);
764 for (unsigned i = 0; i < vector_elements; i++) {
765 this->value.u64[i] = u64;
766 }
767 for (unsigned i = vector_elements; i < 16; i++) {
768 this->value.u64[i] = 0;
769 }
770 }
771
772 ir_constant::ir_constant(int64_t int64, unsigned vector_elements)
773 : ir_rvalue(ir_type_constant)
774 {
775 assert(vector_elements <= 4);
776 this->type = glsl_type::get_instance(GLSL_TYPE_INT64, vector_elements, 1);
777 for (unsigned i = 0; i < vector_elements; i++) {
778 this->value.i64[i] = int64;
779 }
780 for (unsigned i = vector_elements; i < 16; i++) {
781 this->value.i64[i] = 0;
782 }
783 }
784
785 ir_constant::ir_constant(bool b, unsigned vector_elements)
786 : ir_rvalue(ir_type_constant)
787 {
788 assert(vector_elements <= 4);
789 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL, vector_elements, 1);
790 for (unsigned i = 0; i < vector_elements; i++) {
791 this->value.b[i] = b;
792 }
793 for (unsigned i = vector_elements; i < 16; i++) {
794 this->value.b[i] = false;
795 }
796 }
797
798 ir_constant::ir_constant(const ir_constant *c, unsigned i)
799 : ir_rvalue(ir_type_constant)
800 {
801 this->const_elements = NULL;
802 this->type = c->type->get_base_type();
803
804 switch (this->type->base_type) {
805 case GLSL_TYPE_UINT: this->value.u[0] = c->value.u[i]; break;
806 case GLSL_TYPE_INT: this->value.i[0] = c->value.i[i]; break;
807 case GLSL_TYPE_FLOAT: this->value.f[0] = c->value.f[i]; break;
808 case GLSL_TYPE_FLOAT16: this->value.f16[0] = c->value.f16[i]; break;
809 case GLSL_TYPE_BOOL: this->value.b[0] = c->value.b[i]; break;
810 case GLSL_TYPE_DOUBLE: this->value.d[0] = c->value.d[i]; break;
811 default: assert(!"Should not get here."); break;
812 }
813 }
814
815 ir_constant::ir_constant(const struct glsl_type *type, exec_list *value_list)
816 : ir_rvalue(ir_type_constant)
817 {
818 this->const_elements = NULL;
819 this->type = type;
820
821 assert(type->is_scalar() || type->is_vector() || type->is_matrix()
822 || type->is_struct() || type->is_array());
823
824 /* If the constant is a record, the types of each of the entries in
825 * value_list must be a 1-for-1 match with the structure components. Each
826 * entry must also be a constant. Just move the nodes from the value_list
827 * to the list in the ir_constant.
828 */
829 if (type->is_array() || type->is_struct()) {
830 this->const_elements = ralloc_array(this, ir_constant *, type->length);
831 unsigned i = 0;
832 foreach_in_list(ir_constant, value, value_list) {
833 assert(value->as_constant() != NULL);
834
835 this->const_elements[i++] = value;
836 }
837 return;
838 }
839
840 for (unsigned i = 0; i < 16; i++) {
841 this->value.u[i] = 0;
842 }
843
844 ir_constant *value = (ir_constant *) (value_list->get_head_raw());
845
846 /* Constructors with exactly one scalar argument are special for vectors
847 * and matrices. For vectors, the scalar value is replicated to fill all
848 * the components. For matrices, the scalar fills the components of the
849 * diagonal while the rest is filled with 0.
850 */
851 if (value->type->is_scalar() && value->next->is_tail_sentinel()) {
852 if (type->is_matrix()) {
853 /* Matrix - fill diagonal (rest is already set to 0) */
854 for (unsigned i = 0; i < type->matrix_columns; i++) {
855 switch (type->base_type) {
856 case GLSL_TYPE_FLOAT:
857 this->value.f[i * type->vector_elements + i] =
858 value->value.f[0];
859 break;
860 case GLSL_TYPE_DOUBLE:
861 this->value.d[i * type->vector_elements + i] =
862 value->value.d[0];
863 break;
864 case GLSL_TYPE_FLOAT16:
865 this->value.f16[i * type->vector_elements + i] =
866 value->value.f16[0];
867 break;
868 default:
869 assert(!"unexpected matrix base type");
870 }
871 }
872 } else {
873 /* Vector or scalar - fill all components */
874 switch (type->base_type) {
875 case GLSL_TYPE_UINT:
876 case GLSL_TYPE_INT:
877 for (unsigned i = 0; i < type->components(); i++)
878 this->value.u[i] = value->value.u[0];
879 break;
880 case GLSL_TYPE_FLOAT:
881 for (unsigned i = 0; i < type->components(); i++)
882 this->value.f[i] = value->value.f[0];
883 break;
884 case GLSL_TYPE_FLOAT16:
885 for (unsigned i = 0; i < type->components(); i++)
886 this->value.f16[i] = value->value.f16[0];
887 break;
888 case GLSL_TYPE_DOUBLE:
889 for (unsigned i = 0; i < type->components(); i++)
890 this->value.d[i] = value->value.d[0];
891 break;
892 case GLSL_TYPE_UINT64:
893 case GLSL_TYPE_INT64:
894 for (unsigned i = 0; i < type->components(); i++)
895 this->value.u64[i] = value->value.u64[0];
896 break;
897 case GLSL_TYPE_BOOL:
898 for (unsigned i = 0; i < type->components(); i++)
899 this->value.b[i] = value->value.b[0];
900 break;
901 case GLSL_TYPE_SAMPLER:
902 case GLSL_TYPE_IMAGE:
903 this->value.u64[0] = value->value.u64[0];
904 break;
905 default:
906 assert(!"Should not get here.");
907 break;
908 }
909 }
910 return;
911 }
912
913 if (type->is_matrix() && value->type->is_matrix()) {
914 assert(value->next->is_tail_sentinel());
915
916 /* From section 5.4.2 of the GLSL 1.20 spec:
917 * "If a matrix is constructed from a matrix, then each component
918 * (column i, row j) in the result that has a corresponding component
919 * (column i, row j) in the argument will be initialized from there."
920 */
921 unsigned cols = MIN2(type->matrix_columns, value->type->matrix_columns);
922 unsigned rows = MIN2(type->vector_elements, value->type->vector_elements);
923 for (unsigned i = 0; i < cols; i++) {
924 for (unsigned j = 0; j < rows; j++) {
925 const unsigned src = i * value->type->vector_elements + j;
926 const unsigned dst = i * type->vector_elements + j;
927 this->value.f[dst] = value->value.f[src];
928 }
929 }
930
931 /* "All other components will be initialized to the identity matrix." */
932 for (unsigned i = cols; i < type->matrix_columns; i++)
933 this->value.f[i * type->vector_elements + i] = 1.0;
934
935 return;
936 }
937
938 /* Use each component from each entry in the value_list to initialize one
939 * component of the constant being constructed.
940 */
941 unsigned i = 0;
942 for (;;) {
943 assert(value->as_constant() != NULL);
944 assert(!value->is_tail_sentinel());
945
946 for (unsigned j = 0; j < value->type->components(); j++) {
947 switch (type->base_type) {
948 case GLSL_TYPE_UINT:
949 this->value.u[i] = value->get_uint_component(j);
950 break;
951 case GLSL_TYPE_INT:
952 this->value.i[i] = value->get_int_component(j);
953 break;
954 case GLSL_TYPE_FLOAT:
955 this->value.f[i] = value->get_float_component(j);
956 break;
957 case GLSL_TYPE_FLOAT16:
958 this->value.f16[i] = value->get_float16_component(j);
959 break;
960 case GLSL_TYPE_BOOL:
961 this->value.b[i] = value->get_bool_component(j);
962 break;
963 case GLSL_TYPE_DOUBLE:
964 this->value.d[i] = value->get_double_component(j);
965 break;
966 case GLSL_TYPE_UINT64:
967 this->value.u64[i] = value->get_uint64_component(j);
968 break;
969 case GLSL_TYPE_INT64:
970 this->value.i64[i] = value->get_int64_component(j);
971 break;
972 default:
973 /* FINISHME: What to do? Exceptions are not the answer.
974 */
975 break;
976 }
977
978 i++;
979 if (i >= type->components())
980 break;
981 }
982
983 if (i >= type->components())
984 break; /* avoid downcasting a list sentinel */
985 value = (ir_constant *) value->next;
986 }
987 }
988
989 ir_constant *
990 ir_constant::zero(void *mem_ctx, const glsl_type *type)
991 {
992 assert(type->is_scalar() || type->is_vector() || type->is_matrix()
993 || type->is_struct() || type->is_array());
994
995 ir_constant *c = new(mem_ctx) ir_constant;
996 c->type = type;
997 memset(&c->value, 0, sizeof(c->value));
998
999 if (type->is_array()) {
1000 c->const_elements = ralloc_array(c, ir_constant *, type->length);
1001
1002 for (unsigned i = 0; i < type->length; i++)
1003 c->const_elements[i] = ir_constant::zero(c, type->fields.array);
1004 }
1005
1006 if (type->is_struct()) {
1007 c->const_elements = ralloc_array(c, ir_constant *, type->length);
1008
1009 for (unsigned i = 0; i < type->length; i++) {
1010 c->const_elements[i] =
1011 ir_constant::zero(mem_ctx, type->fields.structure[i].type);
1012 }
1013 }
1014
1015 return c;
1016 }
1017
1018 bool
1019 ir_constant::get_bool_component(unsigned i) const
1020 {
1021 switch (this->type->base_type) {
1022 case GLSL_TYPE_UINT: return this->value.u[i] != 0;
1023 case GLSL_TYPE_INT: return this->value.i[i] != 0;
1024 case GLSL_TYPE_FLOAT: return ((int)this->value.f[i]) != 0;
1025 case GLSL_TYPE_FLOAT16: return ((int)_mesa_half_to_float(this->value.f16[i])) != 0;
1026 case GLSL_TYPE_BOOL: return this->value.b[i];
1027 case GLSL_TYPE_DOUBLE: return this->value.d[i] != 0.0;
1028 case GLSL_TYPE_SAMPLER:
1029 case GLSL_TYPE_IMAGE:
1030 case GLSL_TYPE_UINT64: return this->value.u64[i] != 0;
1031 case GLSL_TYPE_INT64: return this->value.i64[i] != 0;
1032 default: assert(!"Should not get here."); break;
1033 }
1034
1035 /* Must return something to make the compiler happy. This is clearly an
1036 * error case.
1037 */
1038 return false;
1039 }
1040
1041 float
1042 ir_constant::get_float_component(unsigned i) const
1043 {
1044 switch (this->type->base_type) {
1045 case GLSL_TYPE_UINT: return (float) this->value.u[i];
1046 case GLSL_TYPE_INT: return (float) this->value.i[i];
1047 case GLSL_TYPE_FLOAT: return this->value.f[i];
1048 case GLSL_TYPE_FLOAT16: return _mesa_half_to_float(this->value.f16[i]);
1049 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1.0f : 0.0f;
1050 case GLSL_TYPE_DOUBLE: return (float) this->value.d[i];
1051 case GLSL_TYPE_SAMPLER:
1052 case GLSL_TYPE_IMAGE:
1053 case GLSL_TYPE_UINT64: return (float) this->value.u64[i];
1054 case GLSL_TYPE_INT64: return (float) this->value.i64[i];
1055 default: assert(!"Should not get here."); break;
1056 }
1057
1058 /* Must return something to make the compiler happy. This is clearly an
1059 * error case.
1060 */
1061 return 0.0;
1062 }
1063
1064 uint16_t
1065 ir_constant::get_float16_component(unsigned i) const
1066 {
1067 if (this->type->base_type == GLSL_TYPE_FLOAT16)
1068 return this->value.f16[i];
1069 else
1070 return _mesa_float_to_half(get_float_component(i));
1071 }
1072
1073 double
1074 ir_constant::get_double_component(unsigned i) const
1075 {
1076 switch (this->type->base_type) {
1077 case GLSL_TYPE_UINT: return (double) this->value.u[i];
1078 case GLSL_TYPE_INT: return (double) this->value.i[i];
1079 case GLSL_TYPE_FLOAT: return (double) this->value.f[i];
1080 case GLSL_TYPE_FLOAT16: return (double) _mesa_half_to_float(this->value.f16[i]);
1081 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1.0 : 0.0;
1082 case GLSL_TYPE_DOUBLE: return this->value.d[i];
1083 case GLSL_TYPE_SAMPLER:
1084 case GLSL_TYPE_IMAGE:
1085 case GLSL_TYPE_UINT64: return (double) this->value.u64[i];
1086 case GLSL_TYPE_INT64: return (double) this->value.i64[i];
1087 default: assert(!"Should not get here."); break;
1088 }
1089
1090 /* Must return something to make the compiler happy. This is clearly an
1091 * error case.
1092 */
1093 return 0.0;
1094 }
1095
1096 int
1097 ir_constant::get_int_component(unsigned i) const
1098 {
1099 switch (this->type->base_type) {
1100 case GLSL_TYPE_UINT: return this->value.u[i];
1101 case GLSL_TYPE_INT: return this->value.i[i];
1102 case GLSL_TYPE_FLOAT: return (int) this->value.f[i];
1103 case GLSL_TYPE_FLOAT16: return (int) _mesa_half_to_float(this->value.f16[i]);
1104 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1105 case GLSL_TYPE_DOUBLE: return (int) this->value.d[i];
1106 case GLSL_TYPE_SAMPLER:
1107 case GLSL_TYPE_IMAGE:
1108 case GLSL_TYPE_UINT64: return (int) this->value.u64[i];
1109 case GLSL_TYPE_INT64: return (int) this->value.i64[i];
1110 default: assert(!"Should not get here."); break;
1111 }
1112
1113 /* Must return something to make the compiler happy. This is clearly an
1114 * error case.
1115 */
1116 return 0;
1117 }
1118
1119 unsigned
1120 ir_constant::get_uint_component(unsigned i) const
1121 {
1122 switch (this->type->base_type) {
1123 case GLSL_TYPE_UINT: return this->value.u[i];
1124 case GLSL_TYPE_INT: return this->value.i[i];
1125 case GLSL_TYPE_FLOAT: return (unsigned) this->value.f[i];
1126 case GLSL_TYPE_FLOAT16: return (unsigned) _mesa_half_to_float(this->value.f16[i]);
1127 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1128 case GLSL_TYPE_DOUBLE: return (unsigned) this->value.d[i];
1129 case GLSL_TYPE_SAMPLER:
1130 case GLSL_TYPE_IMAGE:
1131 case GLSL_TYPE_UINT64: return (unsigned) this->value.u64[i];
1132 case GLSL_TYPE_INT64: return (unsigned) this->value.i64[i];
1133 default: assert(!"Should not get here."); break;
1134 }
1135
1136 /* Must return something to make the compiler happy. This is clearly an
1137 * error case.
1138 */
1139 return 0;
1140 }
1141
1142 int64_t
1143 ir_constant::get_int64_component(unsigned i) const
1144 {
1145 switch (this->type->base_type) {
1146 case GLSL_TYPE_UINT: return this->value.u[i];
1147 case GLSL_TYPE_INT: return this->value.i[i];
1148 case GLSL_TYPE_FLOAT: return (int64_t) this->value.f[i];
1149 case GLSL_TYPE_FLOAT16: return (int64_t) _mesa_half_to_float(this->value.f16[i]);
1150 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1151 case GLSL_TYPE_DOUBLE: return (int64_t) this->value.d[i];
1152 case GLSL_TYPE_SAMPLER:
1153 case GLSL_TYPE_IMAGE:
1154 case GLSL_TYPE_UINT64: return (int64_t) this->value.u64[i];
1155 case GLSL_TYPE_INT64: return this->value.i64[i];
1156 default: assert(!"Should not get here."); break;
1157 }
1158
1159 /* Must return something to make the compiler happy. This is clearly an
1160 * error case.
1161 */
1162 return 0;
1163 }
1164
1165 uint64_t
1166 ir_constant::get_uint64_component(unsigned i) const
1167 {
1168 switch (this->type->base_type) {
1169 case GLSL_TYPE_UINT: return this->value.u[i];
1170 case GLSL_TYPE_INT: return this->value.i[i];
1171 case GLSL_TYPE_FLOAT: return (uint64_t) this->value.f[i];
1172 case GLSL_TYPE_FLOAT16: return (uint64_t) _mesa_half_to_float(this->value.f16[i]);
1173 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1174 case GLSL_TYPE_DOUBLE: return (uint64_t) this->value.d[i];
1175 case GLSL_TYPE_SAMPLER:
1176 case GLSL_TYPE_IMAGE:
1177 case GLSL_TYPE_UINT64: return this->value.u64[i];
1178 case GLSL_TYPE_INT64: return (uint64_t) this->value.i64[i];
1179 default: assert(!"Should not get here."); break;
1180 }
1181
1182 /* Must return something to make the compiler happy. This is clearly an
1183 * error case.
1184 */
1185 return 0;
1186 }
1187
1188 ir_constant *
1189 ir_constant::get_array_element(unsigned i) const
1190 {
1191 assert(this->type->is_array());
1192
1193 /* From page 35 (page 41 of the PDF) of the GLSL 1.20 spec:
1194 *
1195 * "Behavior is undefined if a shader subscripts an array with an index
1196 * less than 0 or greater than or equal to the size the array was
1197 * declared with."
1198 *
1199 * Most out-of-bounds accesses are removed before things could get this far.
1200 * There are cases where non-constant array index values can get constant
1201 * folded.
1202 */
1203 if (int(i) < 0)
1204 i = 0;
1205 else if (i >= this->type->length)
1206 i = this->type->length - 1;
1207
1208 return const_elements[i];
1209 }
1210
1211 ir_constant *
1212 ir_constant::get_record_field(int idx)
1213 {
1214 assert(this->type->is_struct());
1215 assert(idx >= 0 && (unsigned) idx < this->type->length);
1216
1217 return const_elements[idx];
1218 }
1219
1220 void
1221 ir_constant::copy_offset(ir_constant *src, int offset)
1222 {
1223 switch (this->type->base_type) {
1224 case GLSL_TYPE_UINT:
1225 case GLSL_TYPE_INT:
1226 case GLSL_TYPE_FLOAT:
1227 case GLSL_TYPE_FLOAT16:
1228 case GLSL_TYPE_DOUBLE:
1229 case GLSL_TYPE_SAMPLER:
1230 case GLSL_TYPE_IMAGE:
1231 case GLSL_TYPE_UINT64:
1232 case GLSL_TYPE_INT64:
1233 case GLSL_TYPE_BOOL: {
1234 unsigned int size = src->type->components();
1235 assert (size <= this->type->components() - offset);
1236 for (unsigned int i=0; i<size; i++) {
1237 switch (this->type->base_type) {
1238 case GLSL_TYPE_UINT:
1239 value.u[i+offset] = src->get_uint_component(i);
1240 break;
1241 case GLSL_TYPE_INT:
1242 value.i[i+offset] = src->get_int_component(i);
1243 break;
1244 case GLSL_TYPE_FLOAT:
1245 value.f[i+offset] = src->get_float_component(i);
1246 break;
1247 case GLSL_TYPE_FLOAT16:
1248 value.f16[i+offset] = src->get_float16_component(i);
1249 break;
1250 case GLSL_TYPE_BOOL:
1251 value.b[i+offset] = src->get_bool_component(i);
1252 break;
1253 case GLSL_TYPE_DOUBLE:
1254 value.d[i+offset] = src->get_double_component(i);
1255 break;
1256 case GLSL_TYPE_SAMPLER:
1257 case GLSL_TYPE_IMAGE:
1258 case GLSL_TYPE_UINT64:
1259 value.u64[i+offset] = src->get_uint64_component(i);
1260 break;
1261 case GLSL_TYPE_INT64:
1262 value.i64[i+offset] = src->get_int64_component(i);
1263 break;
1264 default: // Shut up the compiler
1265 break;
1266 }
1267 }
1268 break;
1269 }
1270
1271 case GLSL_TYPE_STRUCT:
1272 case GLSL_TYPE_ARRAY: {
1273 assert (src->type == this->type);
1274 for (unsigned i = 0; i < this->type->length; i++) {
1275 this->const_elements[i] = src->const_elements[i]->clone(this, NULL);
1276 }
1277 break;
1278 }
1279
1280 default:
1281 assert(!"Should not get here.");
1282 break;
1283 }
1284 }
1285
1286 void
1287 ir_constant::copy_masked_offset(ir_constant *src, int offset, unsigned int mask)
1288 {
1289 assert (!type->is_array() && !type->is_struct());
1290
1291 if (!type->is_vector() && !type->is_matrix()) {
1292 offset = 0;
1293 mask = 1;
1294 }
1295
1296 int id = 0;
1297 for (int i=0; i<4; i++) {
1298 if (mask & (1 << i)) {
1299 switch (this->type->base_type) {
1300 case GLSL_TYPE_UINT:
1301 value.u[i+offset] = src->get_uint_component(id++);
1302 break;
1303 case GLSL_TYPE_INT:
1304 value.i[i+offset] = src->get_int_component(id++);
1305 break;
1306 case GLSL_TYPE_FLOAT:
1307 value.f[i+offset] = src->get_float_component(id++);
1308 break;
1309 case GLSL_TYPE_FLOAT16:
1310 value.f16[i+offset] = src->get_float16_component(id++);
1311 break;
1312 case GLSL_TYPE_BOOL:
1313 value.b[i+offset] = src->get_bool_component(id++);
1314 break;
1315 case GLSL_TYPE_DOUBLE:
1316 value.d[i+offset] = src->get_double_component(id++);
1317 break;
1318 case GLSL_TYPE_SAMPLER:
1319 case GLSL_TYPE_IMAGE:
1320 case GLSL_TYPE_UINT64:
1321 value.u64[i+offset] = src->get_uint64_component(id++);
1322 break;
1323 case GLSL_TYPE_INT64:
1324 value.i64[i+offset] = src->get_int64_component(id++);
1325 break;
1326 default:
1327 assert(!"Should not get here.");
1328 return;
1329 }
1330 }
1331 }
1332 }
1333
1334 bool
1335 ir_constant::has_value(const ir_constant *c) const
1336 {
1337 if (this->type != c->type)
1338 return false;
1339
1340 if (this->type->is_array() || this->type->is_struct()) {
1341 for (unsigned i = 0; i < this->type->length; i++) {
1342 if (!this->const_elements[i]->has_value(c->const_elements[i]))
1343 return false;
1344 }
1345 return true;
1346 }
1347
1348 for (unsigned i = 0; i < this->type->components(); i++) {
1349 switch (this->type->base_type) {
1350 case GLSL_TYPE_UINT:
1351 if (this->value.u[i] != c->value.u[i])
1352 return false;
1353 break;
1354 case GLSL_TYPE_INT:
1355 if (this->value.i[i] != c->value.i[i])
1356 return false;
1357 break;
1358 case GLSL_TYPE_FLOAT:
1359 if (this->value.f[i] != c->value.f[i])
1360 return false;
1361 break;
1362 case GLSL_TYPE_FLOAT16:
1363 /* Convert to float to make sure NaN and ±0.0 compares correctly */
1364 if (_mesa_half_to_float(this->value.f16[i]) !=
1365 _mesa_half_to_float(c->value.f16[i]))
1366 return false;
1367 break;
1368 case GLSL_TYPE_BOOL:
1369 if (this->value.b[i] != c->value.b[i])
1370 return false;
1371 break;
1372 case GLSL_TYPE_DOUBLE:
1373 if (this->value.d[i] != c->value.d[i])
1374 return false;
1375 break;
1376 case GLSL_TYPE_SAMPLER:
1377 case GLSL_TYPE_IMAGE:
1378 case GLSL_TYPE_UINT64:
1379 if (this->value.u64[i] != c->value.u64[i])
1380 return false;
1381 break;
1382 case GLSL_TYPE_INT64:
1383 if (this->value.i64[i] != c->value.i64[i])
1384 return false;
1385 break;
1386 default:
1387 assert(!"Should not get here.");
1388 return false;
1389 }
1390 }
1391
1392 return true;
1393 }
1394
1395 bool
1396 ir_constant::is_value(float f, int i) const
1397 {
1398 if (!this->type->is_scalar() && !this->type->is_vector())
1399 return false;
1400
1401 /* Only accept boolean values for 0/1. */
1402 if (int(bool(i)) != i && this->type->is_boolean())
1403 return false;
1404
1405 for (unsigned c = 0; c < this->type->vector_elements; c++) {
1406 switch (this->type->base_type) {
1407 case GLSL_TYPE_FLOAT:
1408 if (this->value.f[c] != f)
1409 return false;
1410 break;
1411 case GLSL_TYPE_FLOAT16:
1412 if (_mesa_half_to_float(this->value.f16[c]) != f)
1413 return false;
1414 break;
1415 case GLSL_TYPE_INT:
1416 if (this->value.i[c] != i)
1417 return false;
1418 break;
1419 case GLSL_TYPE_UINT:
1420 if (this->value.u[c] != unsigned(i))
1421 return false;
1422 break;
1423 case GLSL_TYPE_BOOL:
1424 if (this->value.b[c] != bool(i))
1425 return false;
1426 break;
1427 case GLSL_TYPE_DOUBLE:
1428 if (this->value.d[c] != double(f))
1429 return false;
1430 break;
1431 case GLSL_TYPE_SAMPLER:
1432 case GLSL_TYPE_IMAGE:
1433 case GLSL_TYPE_UINT64:
1434 if (this->value.u64[c] != uint64_t(i))
1435 return false;
1436 break;
1437 case GLSL_TYPE_INT64:
1438 if (this->value.i64[c] != i)
1439 return false;
1440 break;
1441 default:
1442 /* The only other base types are structures, arrays, and samplers.
1443 * Samplers cannot be constants, and the others should have been
1444 * filtered out above.
1445 */
1446 assert(!"Should not get here.");
1447 return false;
1448 }
1449 }
1450
1451 return true;
1452 }
1453
1454 bool
1455 ir_constant::is_zero() const
1456 {
1457 return is_value(0.0, 0);
1458 }
1459
1460 bool
1461 ir_constant::is_one() const
1462 {
1463 return is_value(1.0, 1);
1464 }
1465
1466 bool
1467 ir_constant::is_negative_one() const
1468 {
1469 return is_value(-1.0, -1);
1470 }
1471
1472 bool
1473 ir_constant::is_uint16_constant() const
1474 {
1475 if (!type->is_integer_32())
1476 return false;
1477
1478 return value.u[0] < (1 << 16);
1479 }
1480
1481 ir_loop::ir_loop()
1482 : ir_instruction(ir_type_loop)
1483 {
1484 }
1485
1486
1487 ir_dereference_variable::ir_dereference_variable(ir_variable *var)
1488 : ir_dereference(ir_type_dereference_variable)
1489 {
1490 assert(var != NULL);
1491
1492 this->var = var;
1493 this->type = var->type;
1494 }
1495
1496
1497 ir_dereference_array::ir_dereference_array(ir_rvalue *value,
1498 ir_rvalue *array_index)
1499 : ir_dereference(ir_type_dereference_array)
1500 {
1501 this->array_index = array_index;
1502 this->set_array(value);
1503 }
1504
1505
1506 ir_dereference_array::ir_dereference_array(ir_variable *var,
1507 ir_rvalue *array_index)
1508 : ir_dereference(ir_type_dereference_array)
1509 {
1510 void *ctx = ralloc_parent(var);
1511
1512 this->array_index = array_index;
1513 this->set_array(new(ctx) ir_dereference_variable(var));
1514 }
1515
1516
1517 void
1518 ir_dereference_array::set_array(ir_rvalue *value)
1519 {
1520 assert(value != NULL);
1521
1522 this->array = value;
1523
1524 const glsl_type *const vt = this->array->type;
1525
1526 if (vt->is_array()) {
1527 type = vt->fields.array;
1528 } else if (vt->is_matrix()) {
1529 type = vt->column_type();
1530 } else if (vt->is_vector()) {
1531 type = vt->get_base_type();
1532 }
1533 }
1534
1535
1536 ir_dereference_record::ir_dereference_record(ir_rvalue *value,
1537 const char *field)
1538 : ir_dereference(ir_type_dereference_record)
1539 {
1540 assert(value != NULL);
1541
1542 this->record = value;
1543 this->type = this->record->type->field_type(field);
1544 this->field_idx = this->record->type->field_index(field);
1545 }
1546
1547
1548 ir_dereference_record::ir_dereference_record(ir_variable *var,
1549 const char *field)
1550 : ir_dereference(ir_type_dereference_record)
1551 {
1552 void *ctx = ralloc_parent(var);
1553
1554 this->record = new(ctx) ir_dereference_variable(var);
1555 this->type = this->record->type->field_type(field);
1556 this->field_idx = this->record->type->field_index(field);
1557 }
1558
1559 bool
1560 ir_dereference::is_lvalue(const struct _mesa_glsl_parse_state *state) const
1561 {
1562 ir_variable *var = this->variable_referenced();
1563
1564 /* Every l-value derference chain eventually ends in a variable.
1565 */
1566 if ((var == NULL) || var->data.read_only)
1567 return false;
1568
1569 /* From section 4.1.7 of the ARB_bindless_texture spec:
1570 *
1571 * "Samplers can be used as l-values, so can be assigned into and used as
1572 * "out" and "inout" function parameters."
1573 *
1574 * From section 4.1.X of the ARB_bindless_texture spec:
1575 *
1576 * "Images can be used as l-values, so can be assigned into and used as
1577 * "out" and "inout" function parameters."
1578 */
1579 if ((!state || state->has_bindless()) &&
1580 (this->type->contains_sampler() || this->type->contains_image()))
1581 return true;
1582
1583 /* From section 4.1.7 of the GLSL 4.40 spec:
1584 *
1585 * "Opaque variables cannot be treated as l-values; hence cannot
1586 * be used as out or inout function parameters, nor can they be
1587 * assigned into."
1588 */
1589 if (this->type->contains_opaque())
1590 return false;
1591
1592 return true;
1593 }
1594
1595
1596 static const char * const tex_opcode_strs[] = { "tex", "txb", "txl", "txd", "txf", "txf_ms", "txs", "lod", "tg4", "query_levels", "texture_samples", "samples_identical" };
1597
1598 const char *ir_texture::opcode_string()
1599 {
1600 assert((unsigned int) op < ARRAY_SIZE(tex_opcode_strs));
1601 return tex_opcode_strs[op];
1602 }
1603
1604 ir_texture_opcode
1605 ir_texture::get_opcode(const char *str)
1606 {
1607 const int count = sizeof(tex_opcode_strs) / sizeof(tex_opcode_strs[0]);
1608 for (int op = 0; op < count; op++) {
1609 if (strcmp(str, tex_opcode_strs[op]) == 0)
1610 return (ir_texture_opcode) op;
1611 }
1612 return (ir_texture_opcode) -1;
1613 }
1614
1615
1616 void
1617 ir_texture::set_sampler(ir_dereference *sampler, const glsl_type *type)
1618 {
1619 assert(sampler != NULL);
1620 assert(type != NULL);
1621 this->sampler = sampler;
1622 this->type = type;
1623
1624 if (this->op == ir_txs || this->op == ir_query_levels ||
1625 this->op == ir_texture_samples) {
1626 assert(type->base_type == GLSL_TYPE_INT);
1627 } else if (this->op == ir_lod) {
1628 assert(type->vector_elements == 2);
1629 assert(type->is_float());
1630 } else if (this->op == ir_samples_identical) {
1631 assert(type == glsl_type::bool_type);
1632 assert(sampler->type->is_sampler());
1633 assert(sampler->type->sampler_dimensionality == GLSL_SAMPLER_DIM_MS);
1634 } else {
1635 assert(sampler->type->sampled_type == (int) type->base_type);
1636 if (sampler->type->sampler_shadow)
1637 assert(type->vector_elements == 4 || type->vector_elements == 1);
1638 else
1639 assert(type->vector_elements == 4);
1640 }
1641 }
1642
1643
1644 void
1645 ir_swizzle::init_mask(const unsigned *comp, unsigned count)
1646 {
1647 assert((count >= 1) && (count <= 4));
1648
1649 memset(&this->mask, 0, sizeof(this->mask));
1650 this->mask.num_components = count;
1651
1652 unsigned dup_mask = 0;
1653 switch (count) {
1654 case 4:
1655 assert(comp[3] <= 3);
1656 dup_mask |= (1U << comp[3])
1657 & ((1U << comp[0]) | (1U << comp[1]) | (1U << comp[2]));
1658 this->mask.w = comp[3];
1659
1660 case 3:
1661 assert(comp[2] <= 3);
1662 dup_mask |= (1U << comp[2])
1663 & ((1U << comp[0]) | (1U << comp[1]));
1664 this->mask.z = comp[2];
1665
1666 case 2:
1667 assert(comp[1] <= 3);
1668 dup_mask |= (1U << comp[1])
1669 & ((1U << comp[0]));
1670 this->mask.y = comp[1];
1671
1672 case 1:
1673 assert(comp[0] <= 3);
1674 this->mask.x = comp[0];
1675 }
1676
1677 this->mask.has_duplicates = dup_mask != 0;
1678
1679 /* Based on the number of elements in the swizzle and the base type
1680 * (i.e., float, int, unsigned, or bool) of the vector being swizzled,
1681 * generate the type of the resulting value.
1682 */
1683 type = glsl_type::get_instance(val->type->base_type, mask.num_components, 1);
1684 }
1685
1686 ir_swizzle::ir_swizzle(ir_rvalue *val, unsigned x, unsigned y, unsigned z,
1687 unsigned w, unsigned count)
1688 : ir_rvalue(ir_type_swizzle), val(val)
1689 {
1690 const unsigned components[4] = { x, y, z, w };
1691 this->init_mask(components, count);
1692 }
1693
1694 ir_swizzle::ir_swizzle(ir_rvalue *val, const unsigned *comp,
1695 unsigned count)
1696 : ir_rvalue(ir_type_swizzle), val(val)
1697 {
1698 this->init_mask(comp, count);
1699 }
1700
1701 ir_swizzle::ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask)
1702 : ir_rvalue(ir_type_swizzle), val(val), mask(mask)
1703 {
1704 this->type = glsl_type::get_instance(val->type->base_type,
1705 mask.num_components, 1);
1706 }
1707
1708 #define X 1
1709 #define R 5
1710 #define S 9
1711 #define I 13
1712
1713 ir_swizzle *
1714 ir_swizzle::create(ir_rvalue *val, const char *str, unsigned vector_length)
1715 {
1716 void *ctx = ralloc_parent(val);
1717
1718 /* For each possible swizzle character, this table encodes the value in
1719 * \c idx_map that represents the 0th element of the vector. For invalid
1720 * swizzle characters (e.g., 'k'), a special value is used that will allow
1721 * detection of errors.
1722 */
1723 static const unsigned char base_idx[26] = {
1724 /* a b c d e f g h i j k l m */
1725 R, R, I, I, I, I, R, I, I, I, I, I, I,
1726 /* n o p q r s t u v w x y z */
1727 I, I, S, S, R, S, S, I, I, X, X, X, X
1728 };
1729
1730 /* Each valid swizzle character has an entry in the previous table. This
1731 * table encodes the base index encoded in the previous table plus the actual
1732 * index of the swizzle character. When processing swizzles, the first
1733 * character in the string is indexed in the previous table. Each character
1734 * in the string is indexed in this table, and the value found there has the
1735 * value form the first table subtracted. The result must be on the range
1736 * [0,3].
1737 *
1738 * For example, the string "wzyx" will get X from the first table. Each of
1739 * the charcaters will get X+3, X+2, X+1, and X+0 from this table. After
1740 * subtraction, the swizzle values are { 3, 2, 1, 0 }.
1741 *
1742 * The string "wzrg" will get X from the first table. Each of the characters
1743 * will get X+3, X+2, R+0, and R+1 from this table. After subtraction, the
1744 * swizzle values are { 3, 2, 4, 5 }. Since 4 and 5 are outside the range
1745 * [0,3], the error is detected.
1746 */
1747 static const unsigned char idx_map[26] = {
1748 /* a b c d e f g h i j k l m */
1749 R+3, R+2, 0, 0, 0, 0, R+1, 0, 0, 0, 0, 0, 0,
1750 /* n o p q r s t u v w x y z */
1751 0, 0, S+2, S+3, R+0, S+0, S+1, 0, 0, X+3, X+0, X+1, X+2
1752 };
1753
1754 int swiz_idx[4] = { 0, 0, 0, 0 };
1755 unsigned i;
1756
1757
1758 /* Validate the first character in the swizzle string and look up the base
1759 * index value as described above.
1760 */
1761 if ((str[0] < 'a') || (str[0] > 'z'))
1762 return NULL;
1763
1764 const unsigned base = base_idx[str[0] - 'a'];
1765
1766
1767 for (i = 0; (i < 4) && (str[i] != '\0'); i++) {
1768 /* Validate the next character, and, as described above, convert it to a
1769 * swizzle index.
1770 */
1771 if ((str[i] < 'a') || (str[i] > 'z'))
1772 return NULL;
1773
1774 swiz_idx[i] = idx_map[str[i] - 'a'] - base;
1775 if ((swiz_idx[i] < 0) || (swiz_idx[i] >= (int) vector_length))
1776 return NULL;
1777 }
1778
1779 if (str[i] != '\0')
1780 return NULL;
1781
1782 return new(ctx) ir_swizzle(val, swiz_idx[0], swiz_idx[1], swiz_idx[2],
1783 swiz_idx[3], i);
1784 }
1785
1786 #undef X
1787 #undef R
1788 #undef S
1789 #undef I
1790
1791 ir_variable *
1792 ir_swizzle::variable_referenced() const
1793 {
1794 return this->val->variable_referenced();
1795 }
1796
1797
1798 bool ir_variable::temporaries_allocate_names = false;
1799
1800 const char ir_variable::tmp_name[] = "compiler_temp";
1801
1802 ir_variable::ir_variable(const struct glsl_type *type, const char *name,
1803 ir_variable_mode mode)
1804 : ir_instruction(ir_type_variable)
1805 {
1806 this->type = type;
1807
1808 if (mode == ir_var_temporary && !ir_variable::temporaries_allocate_names)
1809 name = NULL;
1810
1811 /* The ir_variable clone method may call this constructor with name set to
1812 * tmp_name.
1813 */
1814 assert(name != NULL
1815 || mode == ir_var_temporary
1816 || mode == ir_var_function_in
1817 || mode == ir_var_function_out
1818 || mode == ir_var_function_inout);
1819 assert(name != ir_variable::tmp_name
1820 || mode == ir_var_temporary);
1821 if (mode == ir_var_temporary
1822 && (name == NULL || name == ir_variable::tmp_name)) {
1823 this->name = ir_variable::tmp_name;
1824 } else if (name == NULL ||
1825 strlen(name) < ARRAY_SIZE(this->name_storage)) {
1826 strcpy(this->name_storage, name ? name : "");
1827 this->name = this->name_storage;
1828 } else {
1829 this->name = ralloc_strdup(this, name);
1830 }
1831
1832 this->u.max_ifc_array_access = NULL;
1833
1834 this->data.explicit_location = false;
1835 this->data.explicit_index = false;
1836 this->data.explicit_binding = false;
1837 this->data.explicit_component = false;
1838 this->data.has_initializer = false;
1839 this->data.is_unmatched_generic_inout = false;
1840 this->data.is_xfb_only = false;
1841 this->data.explicit_xfb_buffer = false;
1842 this->data.explicit_xfb_offset = false;
1843 this->data.explicit_xfb_stride = false;
1844 this->data.location = -1;
1845 this->data.location_frac = 0;
1846 this->data.matrix_layout = GLSL_MATRIX_LAYOUT_INHERITED;
1847 this->data.from_named_ifc_block = false;
1848 this->data.must_be_shader_input = false;
1849 this->data.index = 0;
1850 this->data.binding = 0;
1851 this->data.warn_extension_index = 0;
1852 this->constant_value = NULL;
1853 this->constant_initializer = NULL;
1854 this->data.depth_layout = ir_depth_layout_none;
1855 this->data.used = false;
1856 this->data.assigned = false;
1857 this->data.always_active_io = false;
1858 this->data.read_only = false;
1859 this->data.centroid = false;
1860 this->data.sample = false;
1861 this->data.patch = false;
1862 this->data.explicit_invariant = false;
1863 this->data.invariant = false;
1864 this->data.precise = false;
1865 this->data.how_declared = ir_var_declared_normally;
1866 this->data.mode = mode;
1867 this->data.interpolation = INTERP_MODE_NONE;
1868 this->data.max_array_access = -1;
1869 this->data.offset = 0;
1870 this->data.precision = GLSL_PRECISION_NONE;
1871 this->data.memory_read_only = false;
1872 this->data.memory_write_only = false;
1873 this->data.memory_coherent = false;
1874 this->data.memory_volatile = false;
1875 this->data.memory_restrict = false;
1876 this->data.from_ssbo_unsized_array = false;
1877 this->data.implicit_sized_array = false;
1878 this->data.fb_fetch_output = false;
1879 this->data.bindless = false;
1880 this->data.bound = false;
1881 this->data.image_format = PIPE_FORMAT_NONE;
1882 this->data._num_state_slots = 0;
1883 this->data.param_index = 0;
1884 this->data.stream = 0;
1885 this->data.xfb_buffer = -1;
1886 this->data.xfb_stride = -1;
1887
1888 this->interface_type = NULL;
1889
1890 if (type != NULL) {
1891 if (type->is_interface())
1892 this->init_interface_type(type);
1893 else if (type->without_array()->is_interface())
1894 this->init_interface_type(type->without_array());
1895 }
1896 }
1897
1898
1899 const char *
1900 interpolation_string(unsigned interpolation)
1901 {
1902 switch (interpolation) {
1903 case INTERP_MODE_NONE: return "no";
1904 case INTERP_MODE_SMOOTH: return "smooth";
1905 case INTERP_MODE_FLAT: return "flat";
1906 case INTERP_MODE_NOPERSPECTIVE: return "noperspective";
1907 }
1908
1909 assert(!"Should not get here.");
1910 return "";
1911 }
1912
1913 const char *const ir_variable::warn_extension_table[] = {
1914 "",
1915 "GL_ARB_shader_stencil_export",
1916 "GL_AMD_shader_stencil_export",
1917 };
1918
1919 void
1920 ir_variable::enable_extension_warning(const char *extension)
1921 {
1922 for (unsigned i = 0; i < ARRAY_SIZE(warn_extension_table); i++) {
1923 if (strcmp(warn_extension_table[i], extension) == 0) {
1924 this->data.warn_extension_index = i;
1925 return;
1926 }
1927 }
1928
1929 assert(!"Should not get here.");
1930 this->data.warn_extension_index = 0;
1931 }
1932
1933 const char *
1934 ir_variable::get_extension_warning() const
1935 {
1936 return this->data.warn_extension_index == 0
1937 ? NULL : warn_extension_table[this->data.warn_extension_index];
1938 }
1939
1940 ir_function_signature::ir_function_signature(const glsl_type *return_type,
1941 builtin_available_predicate b)
1942 : ir_instruction(ir_type_function_signature),
1943 return_type(return_type), is_defined(false),
1944 return_precision(GLSL_PRECISION_NONE),
1945 intrinsic_id(ir_intrinsic_invalid), builtin_avail(b), _function(NULL)
1946 {
1947 this->origin = NULL;
1948 }
1949
1950
1951 bool
1952 ir_function_signature::is_builtin() const
1953 {
1954 return builtin_avail != NULL;
1955 }
1956
1957
1958 bool
1959 ir_function_signature::is_builtin_available(const _mesa_glsl_parse_state *state) const
1960 {
1961 /* We can't call the predicate without a state pointer, so just say that
1962 * the signature is available. At compile time, we need the filtering,
1963 * but also receive a valid state pointer. At link time, we're resolving
1964 * imported built-in prototypes to their definitions, which will always
1965 * be an exact match. So we can skip the filtering.
1966 */
1967 if (state == NULL)
1968 return true;
1969
1970 assert(builtin_avail != NULL);
1971 return builtin_avail(state);
1972 }
1973
1974
1975 static bool
1976 modes_match(unsigned a, unsigned b)
1977 {
1978 if (a == b)
1979 return true;
1980
1981 /* Accept "in" vs. "const in" */
1982 if ((a == ir_var_const_in && b == ir_var_function_in) ||
1983 (b == ir_var_const_in && a == ir_var_function_in))
1984 return true;
1985
1986 return false;
1987 }
1988
1989
1990 const char *
1991 ir_function_signature::qualifiers_match(exec_list *params)
1992 {
1993 /* check that the qualifiers match. */
1994 foreach_two_lists(a_node, &this->parameters, b_node, params) {
1995 ir_variable *a = (ir_variable *) a_node;
1996 ir_variable *b = (ir_variable *) b_node;
1997
1998 if (a->data.read_only != b->data.read_only ||
1999 !modes_match(a->data.mode, b->data.mode) ||
2000 a->data.interpolation != b->data.interpolation ||
2001 a->data.centroid != b->data.centroid ||
2002 a->data.sample != b->data.sample ||
2003 a->data.patch != b->data.patch ||
2004 a->data.memory_read_only != b->data.memory_read_only ||
2005 a->data.memory_write_only != b->data.memory_write_only ||
2006 a->data.memory_coherent != b->data.memory_coherent ||
2007 a->data.memory_volatile != b->data.memory_volatile ||
2008 a->data.memory_restrict != b->data.memory_restrict) {
2009
2010 /* parameter a's qualifiers don't match */
2011 return a->name;
2012 }
2013 }
2014 return NULL;
2015 }
2016
2017
2018 void
2019 ir_function_signature::replace_parameters(exec_list *new_params)
2020 {
2021 /* Destroy all of the previous parameter information. If the previous
2022 * parameter information comes from the function prototype, it may either
2023 * specify incorrect parameter names or not have names at all.
2024 */
2025 new_params->move_nodes_to(&parameters);
2026 }
2027
2028
2029 ir_function::ir_function(const char *name)
2030 : ir_instruction(ir_type_function)
2031 {
2032 this->subroutine_index = -1;
2033 this->name = ralloc_strdup(this, name);
2034 }
2035
2036
2037 bool
2038 ir_function::has_user_signature()
2039 {
2040 foreach_in_list(ir_function_signature, sig, &this->signatures) {
2041 if (!sig->is_builtin())
2042 return true;
2043 }
2044 return false;
2045 }
2046
2047
2048 ir_rvalue *
2049 ir_rvalue::error_value(void *mem_ctx)
2050 {
2051 ir_rvalue *v = new(mem_ctx) ir_rvalue(ir_type_unset);
2052
2053 v->type = glsl_type::error_type;
2054 return v;
2055 }
2056
2057
2058 void
2059 visit_exec_list(exec_list *list, ir_visitor *visitor)
2060 {
2061 foreach_in_list_safe(ir_instruction, node, list) {
2062 node->accept(visitor);
2063 }
2064 }
2065
2066
2067 static void
2068 steal_memory(ir_instruction *ir, void *new_ctx)
2069 {
2070 ir_variable *var = ir->as_variable();
2071 ir_function *fn = ir->as_function();
2072 ir_constant *constant = ir->as_constant();
2073 if (var != NULL && var->constant_value != NULL)
2074 steal_memory(var->constant_value, ir);
2075
2076 if (var != NULL && var->constant_initializer != NULL)
2077 steal_memory(var->constant_initializer, ir);
2078
2079 if (fn != NULL && fn->subroutine_types)
2080 ralloc_steal(new_ctx, fn->subroutine_types);
2081
2082 /* The components of aggregate constants are not visited by the normal
2083 * visitor, so steal their values by hand.
2084 */
2085 if (constant != NULL &&
2086 (constant->type->is_array() || constant->type->is_struct())) {
2087 for (unsigned int i = 0; i < constant->type->length; i++) {
2088 steal_memory(constant->const_elements[i], ir);
2089 }
2090 }
2091
2092 ralloc_steal(new_ctx, ir);
2093 }
2094
2095
2096 void
2097 reparent_ir(exec_list *list, void *mem_ctx)
2098 {
2099 foreach_in_list(ir_instruction, node, list) {
2100 visit_tree(node, steal_memory, mem_ctx);
2101 }
2102 }
2103
2104
2105 static ir_rvalue *
2106 try_min_one(ir_rvalue *ir)
2107 {
2108 ir_expression *expr = ir->as_expression();
2109
2110 if (!expr || expr->operation != ir_binop_min)
2111 return NULL;
2112
2113 if (expr->operands[0]->is_one())
2114 return expr->operands[1];
2115
2116 if (expr->operands[1]->is_one())
2117 return expr->operands[0];
2118
2119 return NULL;
2120 }
2121
2122 static ir_rvalue *
2123 try_max_zero(ir_rvalue *ir)
2124 {
2125 ir_expression *expr = ir->as_expression();
2126
2127 if (!expr || expr->operation != ir_binop_max)
2128 return NULL;
2129
2130 if (expr->operands[0]->is_zero())
2131 return expr->operands[1];
2132
2133 if (expr->operands[1]->is_zero())
2134 return expr->operands[0];
2135
2136 return NULL;
2137 }
2138
2139 ir_rvalue *
2140 ir_rvalue::as_rvalue_to_saturate()
2141 {
2142 ir_expression *expr = this->as_expression();
2143
2144 if (!expr)
2145 return NULL;
2146
2147 ir_rvalue *max_zero = try_max_zero(expr);
2148 if (max_zero) {
2149 return try_min_one(max_zero);
2150 } else {
2151 ir_rvalue *min_one = try_min_one(expr);
2152 if (min_one) {
2153 return try_max_zero(min_one);
2154 }
2155 }
2156
2157 return NULL;
2158 }
2159
2160
2161 unsigned
2162 vertices_per_prim(GLenum prim)
2163 {
2164 switch (prim) {
2165 case GL_POINTS:
2166 return 1;
2167 case GL_LINES:
2168 return 2;
2169 case GL_TRIANGLES:
2170 return 3;
2171 case GL_LINES_ADJACENCY:
2172 return 4;
2173 case GL_TRIANGLES_ADJACENCY:
2174 return 6;
2175 default:
2176 assert(!"Bad primitive");
2177 return 3;
2178 }
2179 }
2180
2181 /**
2182 * Generate a string describing the mode of a variable
2183 */
2184 const char *
2185 mode_string(const ir_variable *var)
2186 {
2187 switch (var->data.mode) {
2188 case ir_var_auto:
2189 return (var->data.read_only) ? "global constant" : "global variable";
2190
2191 case ir_var_uniform:
2192 return "uniform";
2193
2194 case ir_var_shader_storage:
2195 return "buffer";
2196
2197 case ir_var_shader_in:
2198 return "shader input";
2199
2200 case ir_var_shader_out:
2201 return "shader output";
2202
2203 case ir_var_function_in:
2204 case ir_var_const_in:
2205 return "function input";
2206
2207 case ir_var_function_out:
2208 return "function output";
2209
2210 case ir_var_function_inout:
2211 return "function inout";
2212
2213 case ir_var_system_value:
2214 return "shader input";
2215
2216 case ir_var_temporary:
2217 return "compiler temporary";
2218
2219 case ir_var_mode_count:
2220 break;
2221 }
2222
2223 assert(!"Should not get here.");
2224 return "invalid variable";
2225 }