projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
soft-fp64/fadd: Use absolute value of expDiff
[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_f2fmp:
297 case ir_unop_b2f16:
298 this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT16,
299 op0->type->vector_elements, 1);
300 break;
301
302 case ir_unop_f2b:
303 case ir_unop_i2b:
304 case ir_unop_d2b:
305 case ir_unop_f162b:
306 case ir_unop_i642b:
307 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL,
308 op0->type->vector_elements, 1);
309 break;
310
311 case ir_unop_f2d:
312 case ir_unop_i2d:
313 case ir_unop_u2d:
314 case ir_unop_i642d:
315 case ir_unop_u642d:
316 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE,
317 op0->type->vector_elements, 1);
318 break;
319
320 case ir_unop_i2u:
321 case ir_unop_f2u:
322 case ir_unop_d2u:
323 case ir_unop_bitcast_f2u:
324 case ir_unop_i642u:
325 case ir_unop_u642u:
326 this->type = glsl_type::get_instance(GLSL_TYPE_UINT,
327 op0->type->vector_elements, 1);
328 break;
329
330 case ir_unop_i2i64:
331 case ir_unop_u2i64:
332 case ir_unop_b2i64:
333 case ir_unop_f2i64:
334 case ir_unop_d2i64:
335 case ir_unop_u642i64:
336 this->type = glsl_type::get_instance(GLSL_TYPE_INT64,
337 op0->type->vector_elements, 1);
338 break;
339
340 case ir_unop_i2u64:
341 case ir_unop_u2u64:
342 case ir_unop_f2u64:
343 case ir_unop_d2u64:
344 case ir_unop_i642u64:
345 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64,
346 op0->type->vector_elements, 1);
347 break;
348 case ir_unop_noise:
349 this->type = glsl_type::float_type;
350 break;
351
352 case ir_unop_unpack_double_2x32:
353 case ir_unop_unpack_uint_2x32:
354 this->type = glsl_type::uvec2_type;
355 break;
356
357 case ir_unop_unpack_int_2x32:
358 this->type = glsl_type::ivec2_type;
359 break;
360
361 case ir_unop_pack_snorm_2x16:
362 case ir_unop_pack_snorm_4x8:
363 case ir_unop_pack_unorm_2x16:
364 case ir_unop_pack_unorm_4x8:
365 case ir_unop_pack_half_2x16:
366 this->type = glsl_type::uint_type;
367 break;
368
369 case ir_unop_pack_double_2x32:
370 this->type = glsl_type::double_type;
371 break;
372
373 case ir_unop_pack_int_2x32:
374 this->type = glsl_type::int64_t_type;
375 break;
376
377 case ir_unop_pack_uint_2x32:
378 this->type = glsl_type::uint64_t_type;
379 break;
380
381 case ir_unop_unpack_snorm_2x16:
382 case ir_unop_unpack_unorm_2x16:
383 case ir_unop_unpack_half_2x16:
384 this->type = glsl_type::vec2_type;
385 break;
386
387 case ir_unop_unpack_snorm_4x8:
388 case ir_unop_unpack_unorm_4x8:
389 this->type = glsl_type::vec4_type;
390 break;
391
392 case ir_unop_unpack_sampler_2x32:
393 case ir_unop_unpack_image_2x32:
394 this->type = glsl_type::uvec2_type;
395 break;
396
397 case ir_unop_pack_sampler_2x32:
398 case ir_unop_pack_image_2x32:
399 this->type = op0->type;
400 break;
401
402 case ir_unop_frexp_sig:
403 this->type = op0->type;
404 break;
405 case ir_unop_frexp_exp:
406 this->type = glsl_type::get_instance(GLSL_TYPE_INT,
407 op0->type->vector_elements, 1);
408 break;
409
410 case ir_unop_get_buffer_size:
411 case ir_unop_ssbo_unsized_array_length:
412 this->type = glsl_type::int_type;
413 break;
414
415 case ir_unop_bitcast_i642d:
416 case ir_unop_bitcast_u642d:
417 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE,
418 op0->type->vector_elements, 1);
419 break;
420
421 case ir_unop_bitcast_d2i64:
422 this->type = glsl_type::get_instance(GLSL_TYPE_INT64,
423 op0->type->vector_elements, 1);
424 break;
425 case ir_unop_bitcast_d2u64:
426 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64,
427 op0->type->vector_elements, 1);
428 break;
429
430 default:
431 assert(!"not reached: missing automatic type setup for ir_expression");
432 this->type = op0->type;
433 break;
434 }
435 }
436
437 ir_expression::ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1)
438 : ir_rvalue(ir_type_expression)
439 {
440 this->operation = ir_expression_operation(op);
441 this->operands[0] = op0;
442 this->operands[1] = op1;
443 this->operands[2] = NULL;
444 this->operands[3] = NULL;
445
446 assert(op > ir_last_unop);
447 init_num_operands();
448 assert(num_operands == 2);
449 for (unsigned i = 0; i < num_operands; i++) {
450 assert(this->operands[i] != NULL);
451 }
452
453 switch (this->operation) {
454 case ir_binop_all_equal:
455 case ir_binop_any_nequal:
456 this->type = glsl_type::bool_type;
457 break;
458
459 case ir_binop_add:
460 case ir_binop_sub:
461 case ir_binop_min:
462 case ir_binop_max:
463 case ir_binop_pow:
464 case ir_binop_mul:
465 case ir_binop_div:
466 case ir_binop_mod:
467 case ir_binop_atan2:
468 if (op0->type->is_scalar()) {
469 this->type = op1->type;
470 } else if (op1->type->is_scalar()) {
471 this->type = op0->type;
472 } else {
473 if (this->operation == ir_binop_mul) {
474 this->type = glsl_type::get_mul_type(op0->type, op1->type);
475 } else {
476 assert(op0->type == op1->type);
477 this->type = op0->type;
478 }
479 }
480 break;
481
482 case ir_binop_logic_and:
483 case ir_binop_logic_xor:
484 case ir_binop_logic_or:
485 case ir_binop_bit_and:
486 case ir_binop_bit_xor:
487 case ir_binop_bit_or:
488 assert(!op0->type->is_matrix());
489 assert(!op1->type->is_matrix());
490 if (op0->type->is_scalar()) {
491 this->type = op1->type;
492 } else if (op1->type->is_scalar()) {
493 this->type = op0->type;
494 } else {
495 assert(op0->type->vector_elements == op1->type->vector_elements);
496 this->type = op0->type;
497 }
498 break;
499
500 case ir_binop_equal:
501 case ir_binop_nequal:
502 case ir_binop_gequal:
503 case ir_binop_less:
504 assert(op0->type == op1->type);
505 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL,
506 op0->type->vector_elements, 1);
507 break;
508
509 case ir_binop_dot:
510 this->type = op0->type->get_base_type();
511 break;
512
513 case ir_binop_imul_high:
514 case ir_binop_mul_32x16:
515 case ir_binop_carry:
516 case ir_binop_borrow:
517 case ir_binop_lshift:
518 case ir_binop_rshift:
519 case ir_binop_ldexp:
520 case ir_binop_interpolate_at_offset:
521 case ir_binop_interpolate_at_sample:
522 this->type = op0->type;
523 break;
524
525 case ir_binop_add_sat:
526 case ir_binop_sub_sat:
527 case ir_binop_avg:
528 case ir_binop_avg_round:
529 assert(op0->type == op1->type);
530 this->type = op0->type;
531 break;
532
533 case ir_binop_abs_sub: {
534 enum glsl_base_type base;
535
536 assert(op0->type == op1->type);
537
538 switch (op0->type->base_type) {
539 case GLSL_TYPE_UINT:
540 case GLSL_TYPE_INT:
541 base = GLSL_TYPE_UINT;
542 break;
543 case GLSL_TYPE_UINT8:
544 case GLSL_TYPE_INT8:
545 base = GLSL_TYPE_UINT8;
546 break;
547 case GLSL_TYPE_UINT16:
548 case GLSL_TYPE_INT16:
549 base = GLSL_TYPE_UINT16;
550 break;
551 case GLSL_TYPE_UINT64:
552 case GLSL_TYPE_INT64:
553 base = GLSL_TYPE_UINT64;
554 break;
555 default:
556 unreachable(!"Invalid base type.");
557 }
558
559 this->type = glsl_type::get_instance(base, op0->type->vector_elements, 1);
560 break;
561 }
562
563 case ir_binop_vector_extract:
564 this->type = op0->type->get_scalar_type();
565 break;
566
567 default:
568 assert(!"not reached: missing automatic type setup for ir_expression");
569 this->type = glsl_type::float_type;
570 }
571 }
572
573 ir_expression::ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1,
574 ir_rvalue *op2)
575 : ir_rvalue(ir_type_expression)
576 {
577 this->operation = ir_expression_operation(op);
578 this->operands[0] = op0;
579 this->operands[1] = op1;
580 this->operands[2] = op2;
581 this->operands[3] = NULL;
582
583 assert(op > ir_last_binop && op <= ir_last_triop);
584 init_num_operands();
585 assert(num_operands == 3);
586 for (unsigned i = 0; i < num_operands; i++) {
587 assert(this->operands[i] != NULL);
588 }
589
590 switch (this->operation) {
591 case ir_triop_fma:
592 case ir_triop_lrp:
593 case ir_triop_bitfield_extract:
594 case ir_triop_vector_insert:
595 this->type = op0->type;
596 break;
597
598 case ir_triop_csel:
599 this->type = op1->type;
600 break;
601
602 default:
603 assert(!"not reached: missing automatic type setup for ir_expression");
604 this->type = glsl_type::float_type;
605 }
606 }
607
608 /**
609 * This is only here for ir_reader to used for testing purposes. Please use
610 * the precomputed num_operands field if you need the number of operands.
611 */
612 unsigned
613 ir_expression::get_num_operands(ir_expression_operation op)
614 {
615 assert(op <= ir_last_opcode);
616
617 if (op <= ir_last_unop)
618 return 1;
619
620 if (op <= ir_last_binop)
621 return 2;
622
623 if (op <= ir_last_triop)
624 return 3;
625
626 if (op <= ir_last_quadop)
627 return 4;
628
629 unreachable("Could not calculate number of operands");
630 }
631
632 #include "ir_expression_operation_strings.h"
633
634 const char*
635 depth_layout_string(ir_depth_layout layout)
636 {
637 switch(layout) {
638 case ir_depth_layout_none: return "";
639 case ir_depth_layout_any: return "depth_any";
640 case ir_depth_layout_greater: return "depth_greater";
641 case ir_depth_layout_less: return "depth_less";
642 case ir_depth_layout_unchanged: return "depth_unchanged";
643
644 default:
645 assert(0);
646 return "";
647 }
648 }
649
650 ir_expression_operation
651 ir_expression::get_operator(const char *str)
652 {
653 for (int op = 0; op <= int(ir_last_opcode); op++) {
654 if (strcmp(str, ir_expression_operation_strings[op]) == 0)
655 return (ir_expression_operation) op;
656 }
657 return (ir_expression_operation) -1;
658 }
659
660 ir_variable *
661 ir_expression::variable_referenced() const
662 {
663 switch (operation) {
664 case ir_binop_vector_extract:
665 case ir_triop_vector_insert:
666 /* We get these for things like a[0] where a is a vector type. In these
667 * cases we want variable_referenced() to return the actual vector
668 * variable this is wrapping.
669 */
670 return operands[0]->variable_referenced();
671 default:
672 return ir_rvalue::variable_referenced();
673 }
674 }
675
676 ir_constant::ir_constant()
677 : ir_rvalue(ir_type_constant)
678 {
679 this->const_elements = NULL;
680 }
681
682 ir_constant::ir_constant(const struct glsl_type *type,
683 const ir_constant_data *data)
684 : ir_rvalue(ir_type_constant)
685 {
686 this->const_elements = NULL;
687
688 assert((type->base_type >= GLSL_TYPE_UINT)
689 && (type->base_type <= GLSL_TYPE_IMAGE));
690
691 this->type = type;
692 memcpy(& this->value, data, sizeof(this->value));
693 }
694
695 ir_constant::ir_constant(float16_t f16, unsigned vector_elements)
696 : ir_rvalue(ir_type_constant)
697 {
698 assert(vector_elements <= 4);
699 this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT16, vector_elements, 1);
700 for (unsigned i = 0; i < vector_elements; i++) {
701 this->value.f16[i] = f16.bits;
702 }
703 for (unsigned i = vector_elements; i < 16; i++) {
704 this->value.f[i] = 0;
705 }
706 }
707
708 ir_constant::ir_constant(float f, unsigned vector_elements)
709 : ir_rvalue(ir_type_constant)
710 {
711 assert(vector_elements <= 4);
712 this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT, vector_elements, 1);
713 for (unsigned i = 0; i < vector_elements; i++) {
714 this->value.f[i] = f;
715 }
716 for (unsigned i = vector_elements; i < 16; i++) {
717 this->value.f[i] = 0;
718 }
719 }
720
721 ir_constant::ir_constant(double d, unsigned vector_elements)
722 : ir_rvalue(ir_type_constant)
723 {
724 assert(vector_elements <= 4);
725 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE, vector_elements, 1);
726 for (unsigned i = 0; i < vector_elements; i++) {
727 this->value.d[i] = d;
728 }
729 for (unsigned i = vector_elements; i < 16; i++) {
730 this->value.d[i] = 0.0;
731 }
732 }
733
734 ir_constant::ir_constant(unsigned int u, unsigned vector_elements)
735 : ir_rvalue(ir_type_constant)
736 {
737 assert(vector_elements <= 4);
738 this->type = glsl_type::get_instance(GLSL_TYPE_UINT, vector_elements, 1);
739 for (unsigned i = 0; i < vector_elements; i++) {
740 this->value.u[i] = u;
741 }
742 for (unsigned i = vector_elements; i < 16; i++) {
743 this->value.u[i] = 0;
744 }
745 }
746
747 ir_constant::ir_constant(int integer, unsigned vector_elements)
748 : ir_rvalue(ir_type_constant)
749 {
750 assert(vector_elements <= 4);
751 this->type = glsl_type::get_instance(GLSL_TYPE_INT, vector_elements, 1);
752 for (unsigned i = 0; i < vector_elements; i++) {
753 this->value.i[i] = integer;
754 }
755 for (unsigned i = vector_elements; i < 16; i++) {
756 this->value.i[i] = 0;
757 }
758 }
759
760 ir_constant::ir_constant(uint64_t u64, unsigned vector_elements)
761 : ir_rvalue(ir_type_constant)
762 {
763 assert(vector_elements <= 4);
764 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64, vector_elements, 1);
765 for (unsigned i = 0; i < vector_elements; i++) {
766 this->value.u64[i] = u64;
767 }
768 for (unsigned i = vector_elements; i < 16; i++) {
769 this->value.u64[i] = 0;
770 }
771 }
772
773 ir_constant::ir_constant(int64_t int64, unsigned vector_elements)
774 : ir_rvalue(ir_type_constant)
775 {
776 assert(vector_elements <= 4);
777 this->type = glsl_type::get_instance(GLSL_TYPE_INT64, vector_elements, 1);
778 for (unsigned i = 0; i < vector_elements; i++) {
779 this->value.i64[i] = int64;
780 }
781 for (unsigned i = vector_elements; i < 16; i++) {
782 this->value.i64[i] = 0;
783 }
784 }
785
786 ir_constant::ir_constant(bool b, unsigned vector_elements)
787 : ir_rvalue(ir_type_constant)
788 {
789 assert(vector_elements <= 4);
790 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL, vector_elements, 1);
791 for (unsigned i = 0; i < vector_elements; i++) {
792 this->value.b[i] = b;
793 }
794 for (unsigned i = vector_elements; i < 16; i++) {
795 this->value.b[i] = false;
796 }
797 }
798
799 ir_constant::ir_constant(const ir_constant *c, unsigned i)
800 : ir_rvalue(ir_type_constant)
801 {
802 this->const_elements = NULL;
803 this->type = c->type->get_base_type();
804
805 switch (this->type->base_type) {
806 case GLSL_TYPE_UINT: this->value.u[0] = c->value.u[i]; break;
807 case GLSL_TYPE_INT: this->value.i[0] = c->value.i[i]; break;
808 case GLSL_TYPE_FLOAT: this->value.f[0] = c->value.f[i]; break;
809 case GLSL_TYPE_FLOAT16: this->value.f16[0] = c->value.f16[i]; break;
810 case GLSL_TYPE_BOOL: this->value.b[0] = c->value.b[i]; break;
811 case GLSL_TYPE_DOUBLE: this->value.d[0] = c->value.d[i]; break;
812 default: assert(!"Should not get here."); break;
813 }
814 }
815
816 ir_constant::ir_constant(const struct glsl_type *type, exec_list *value_list)
817 : ir_rvalue(ir_type_constant)
818 {
819 this->const_elements = NULL;
820 this->type = type;
821
822 assert(type->is_scalar() || type->is_vector() || type->is_matrix()
823 || type->is_struct() || type->is_array());
824
825 /* If the constant is a record, the types of each of the entries in
826 * value_list must be a 1-for-1 match with the structure components. Each
827 * entry must also be a constant. Just move the nodes from the value_list
828 * to the list in the ir_constant.
829 */
830 if (type->is_array() || type->is_struct()) {
831 this->const_elements = ralloc_array(this, ir_constant *, type->length);
832 unsigned i = 0;
833 foreach_in_list(ir_constant, value, value_list) {
834 assert(value->as_constant() != NULL);
835
836 this->const_elements[i++] = value;
837 }
838 return;
839 }
840
841 for (unsigned i = 0; i < 16; i++) {
842 this->value.u[i] = 0;
843 }
844
845 ir_constant *value = (ir_constant *) (value_list->get_head_raw());
846
847 /* Constructors with exactly one scalar argument are special for vectors
848 * and matrices. For vectors, the scalar value is replicated to fill all
849 * the components. For matrices, the scalar fills the components of the
850 * diagonal while the rest is filled with 0.
851 */
852 if (value->type->is_scalar() && value->next->is_tail_sentinel()) {
853 if (type->is_matrix()) {
854 /* Matrix - fill diagonal (rest is already set to 0) */
855 for (unsigned i = 0; i < type->matrix_columns; i++) {
856 switch (type->base_type) {
857 case GLSL_TYPE_FLOAT:
858 this->value.f[i * type->vector_elements + i] =
859 value->value.f[0];
860 break;
861 case GLSL_TYPE_DOUBLE:
862 this->value.d[i * type->vector_elements + i] =
863 value->value.d[0];
864 break;
865 case GLSL_TYPE_FLOAT16:
866 this->value.f16[i * type->vector_elements + i] =
867 value->value.f16[0];
868 break;
869 default:
870 assert(!"unexpected matrix base type");
871 }
872 }
873 } else {
874 /* Vector or scalar - fill all components */
875 switch (type->base_type) {
876 case GLSL_TYPE_UINT:
877 case GLSL_TYPE_INT:
878 for (unsigned i = 0; i < type->components(); i++)
879 this->value.u[i] = value->value.u[0];
880 break;
881 case GLSL_TYPE_FLOAT:
882 for (unsigned i = 0; i < type->components(); i++)
883 this->value.f[i] = value->value.f[0];
884 break;
885 case GLSL_TYPE_FLOAT16:
886 for (unsigned i = 0; i < type->components(); i++)
887 this->value.f16[i] = value->value.f16[0];
888 break;
889 case GLSL_TYPE_DOUBLE:
890 for (unsigned i = 0; i < type->components(); i++)
891 this->value.d[i] = value->value.d[0];
892 break;
893 case GLSL_TYPE_UINT64:
894 case GLSL_TYPE_INT64:
895 for (unsigned i = 0; i < type->components(); i++)
896 this->value.u64[i] = value->value.u64[0];
897 break;
898 case GLSL_TYPE_BOOL:
899 for (unsigned i = 0; i < type->components(); i++)
900 this->value.b[i] = value->value.b[0];
901 break;
902 case GLSL_TYPE_SAMPLER:
903 case GLSL_TYPE_IMAGE:
904 this->value.u64[0] = value->value.u64[0];
905 break;
906 default:
907 assert(!"Should not get here.");
908 break;
909 }
910 }
911 return;
912 }
913
914 if (type->is_matrix() && value->type->is_matrix()) {
915 assert(value->next->is_tail_sentinel());
916
917 /* From section 5.4.2 of the GLSL 1.20 spec:
918 * "If a matrix is constructed from a matrix, then each component
919 * (column i, row j) in the result that has a corresponding component
920 * (column i, row j) in the argument will be initialized from there."
921 */
922 unsigned cols = MIN2(type->matrix_columns, value->type->matrix_columns);
923 unsigned rows = MIN2(type->vector_elements, value->type->vector_elements);
924 for (unsigned i = 0; i < cols; i++) {
925 for (unsigned j = 0; j < rows; j++) {
926 const unsigned src = i * value->type->vector_elements + j;
927 const unsigned dst = i * type->vector_elements + j;
928 this->value.f[dst] = value->value.f[src];
929 }
930 }
931
932 /* "All other components will be initialized to the identity matrix." */
933 for (unsigned i = cols; i < type->matrix_columns; i++)
934 this->value.f[i * type->vector_elements + i] = 1.0;
935
936 return;
937 }
938
939 /* Use each component from each entry in the value_list to initialize one
940 * component of the constant being constructed.
941 */
942 unsigned i = 0;
943 for (;;) {
944 assert(value->as_constant() != NULL);
945 assert(!value->is_tail_sentinel());
946
947 for (unsigned j = 0; j < value->type->components(); j++) {
948 switch (type->base_type) {
949 case GLSL_TYPE_UINT:
950 this->value.u[i] = value->get_uint_component(j);
951 break;
952 case GLSL_TYPE_INT:
953 this->value.i[i] = value->get_int_component(j);
954 break;
955 case GLSL_TYPE_FLOAT:
956 this->value.f[i] = value->get_float_component(j);
957 break;
958 case GLSL_TYPE_FLOAT16:
959 this->value.f16[i] = value->get_float16_component(j);
960 break;
961 case GLSL_TYPE_BOOL:
962 this->value.b[i] = value->get_bool_component(j);
963 break;
964 case GLSL_TYPE_DOUBLE:
965 this->value.d[i] = value->get_double_component(j);
966 break;
967 case GLSL_TYPE_UINT64:
968 this->value.u64[i] = value->get_uint64_component(j);
969 break;
970 case GLSL_TYPE_INT64:
971 this->value.i64[i] = value->get_int64_component(j);
972 break;
973 default:
974 /* FINISHME: What to do? Exceptions are not the answer.
975 */
976 break;
977 }
978
979 i++;
980 if (i >= type->components())
981 break;
982 }
983
984 if (i >= type->components())
985 break; /* avoid downcasting a list sentinel */
986 value = (ir_constant *) value->next;
987 }
988 }
989
990 ir_constant *
991 ir_constant::zero(void *mem_ctx, const glsl_type *type)
992 {
993 assert(type->is_scalar() || type->is_vector() || type->is_matrix()
994 || type->is_struct() || type->is_array());
995
996 ir_constant *c = new(mem_ctx) ir_constant;
997 c->type = type;
998 memset(&c->value, 0, sizeof(c->value));
999
1000 if (type->is_array()) {
1001 c->const_elements = ralloc_array(c, ir_constant *, type->length);
1002
1003 for (unsigned i = 0; i < type->length; i++)
1004 c->const_elements[i] = ir_constant::zero(c, type->fields.array);
1005 }
1006
1007 if (type->is_struct()) {
1008 c->const_elements = ralloc_array(c, ir_constant *, type->length);
1009
1010 for (unsigned i = 0; i < type->length; i++) {
1011 c->const_elements[i] =
1012 ir_constant::zero(mem_ctx, type->fields.structure[i].type);
1013 }
1014 }
1015
1016 return c;
1017 }
1018
1019 bool
1020 ir_constant::get_bool_component(unsigned i) const
1021 {
1022 switch (this->type->base_type) {
1023 case GLSL_TYPE_UINT: return this->value.u[i] != 0;
1024 case GLSL_TYPE_INT: return this->value.i[i] != 0;
1025 case GLSL_TYPE_FLOAT: return ((int)this->value.f[i]) != 0;
1026 case GLSL_TYPE_FLOAT16: return ((int)_mesa_half_to_float(this->value.f16[i])) != 0;
1027 case GLSL_TYPE_BOOL: return this->value.b[i];
1028 case GLSL_TYPE_DOUBLE: return this->value.d[i] != 0.0;
1029 case GLSL_TYPE_SAMPLER:
1030 case GLSL_TYPE_IMAGE:
1031 case GLSL_TYPE_UINT64: return this->value.u64[i] != 0;
1032 case GLSL_TYPE_INT64: return this->value.i64[i] != 0;
1033 default: assert(!"Should not get here."); break;
1034 }
1035
1036 /* Must return something to make the compiler happy. This is clearly an
1037 * error case.
1038 */
1039 return false;
1040 }
1041
1042 float
1043 ir_constant::get_float_component(unsigned i) const
1044 {
1045 switch (this->type->base_type) {
1046 case GLSL_TYPE_UINT: return (float) this->value.u[i];
1047 case GLSL_TYPE_INT: return (float) this->value.i[i];
1048 case GLSL_TYPE_FLOAT: return this->value.f[i];
1049 case GLSL_TYPE_FLOAT16: return _mesa_half_to_float(this->value.f16[i]);
1050 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1.0f : 0.0f;
1051 case GLSL_TYPE_DOUBLE: return (float) this->value.d[i];
1052 case GLSL_TYPE_SAMPLER:
1053 case GLSL_TYPE_IMAGE:
1054 case GLSL_TYPE_UINT64: return (float) this->value.u64[i];
1055 case GLSL_TYPE_INT64: return (float) this->value.i64[i];
1056 default: assert(!"Should not get here."); break;
1057 }
1058
1059 /* Must return something to make the compiler happy. This is clearly an
1060 * error case.
1061 */
1062 return 0.0;
1063 }
1064
1065 uint16_t
1066 ir_constant::get_float16_component(unsigned i) const
1067 {
1068 if (this->type->base_type == GLSL_TYPE_FLOAT16)
1069 return this->value.f16[i];
1070 else
1071 return _mesa_float_to_half(get_float_component(i));
1072 }
1073
1074 double
1075 ir_constant::get_double_component(unsigned i) const
1076 {
1077 switch (this->type->base_type) {
1078 case GLSL_TYPE_UINT: return (double) this->value.u[i];
1079 case GLSL_TYPE_INT: return (double) this->value.i[i];
1080 case GLSL_TYPE_FLOAT: return (double) this->value.f[i];
1081 case GLSL_TYPE_FLOAT16: return (double) _mesa_half_to_float(this->value.f16[i]);
1082 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1.0 : 0.0;
1083 case GLSL_TYPE_DOUBLE: return this->value.d[i];
1084 case GLSL_TYPE_SAMPLER:
1085 case GLSL_TYPE_IMAGE:
1086 case GLSL_TYPE_UINT64: return (double) this->value.u64[i];
1087 case GLSL_TYPE_INT64: return (double) this->value.i64[i];
1088 default: assert(!"Should not get here."); break;
1089 }
1090
1091 /* Must return something to make the compiler happy. This is clearly an
1092 * error case.
1093 */
1094 return 0.0;
1095 }
1096
1097 int
1098 ir_constant::get_int_component(unsigned i) const
1099 {
1100 switch (this->type->base_type) {
1101 case GLSL_TYPE_UINT: return this->value.u[i];
1102 case GLSL_TYPE_INT: return this->value.i[i];
1103 case GLSL_TYPE_FLOAT: return (int) this->value.f[i];
1104 case GLSL_TYPE_FLOAT16: return (int) _mesa_half_to_float(this->value.f16[i]);
1105 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1106 case GLSL_TYPE_DOUBLE: return (int) this->value.d[i];
1107 case GLSL_TYPE_SAMPLER:
1108 case GLSL_TYPE_IMAGE:
1109 case GLSL_TYPE_UINT64: return (int) this->value.u64[i];
1110 case GLSL_TYPE_INT64: return (int) this->value.i64[i];
1111 default: assert(!"Should not get here."); break;
1112 }
1113
1114 /* Must return something to make the compiler happy. This is clearly an
1115 * error case.
1116 */
1117 return 0;
1118 }
1119
1120 unsigned
1121 ir_constant::get_uint_component(unsigned i) const
1122 {
1123 switch (this->type->base_type) {
1124 case GLSL_TYPE_UINT: return this->value.u[i];
1125 case GLSL_TYPE_INT: return this->value.i[i];
1126 case GLSL_TYPE_FLOAT: return (unsigned) this->value.f[i];
1127 case GLSL_TYPE_FLOAT16: return (unsigned) _mesa_half_to_float(this->value.f16[i]);
1128 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1129 case GLSL_TYPE_DOUBLE: return (unsigned) this->value.d[i];
1130 case GLSL_TYPE_SAMPLER:
1131 case GLSL_TYPE_IMAGE:
1132 case GLSL_TYPE_UINT64: return (unsigned) this->value.u64[i];
1133 case GLSL_TYPE_INT64: return (unsigned) this->value.i64[i];
1134 default: assert(!"Should not get here."); break;
1135 }
1136
1137 /* Must return something to make the compiler happy. This is clearly an
1138 * error case.
1139 */
1140 return 0;
1141 }
1142
1143 int64_t
1144 ir_constant::get_int64_component(unsigned i) const
1145 {
1146 switch (this->type->base_type) {
1147 case GLSL_TYPE_UINT: return this->value.u[i];
1148 case GLSL_TYPE_INT: return this->value.i[i];
1149 case GLSL_TYPE_FLOAT: return (int64_t) this->value.f[i];
1150 case GLSL_TYPE_FLOAT16: return (int64_t) _mesa_half_to_float(this->value.f16[i]);
1151 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1152 case GLSL_TYPE_DOUBLE: return (int64_t) this->value.d[i];
1153 case GLSL_TYPE_SAMPLER:
1154 case GLSL_TYPE_IMAGE:
1155 case GLSL_TYPE_UINT64: return (int64_t) this->value.u64[i];
1156 case GLSL_TYPE_INT64: return this->value.i64[i];
1157 default: assert(!"Should not get here."); break;
1158 }
1159
1160 /* Must return something to make the compiler happy. This is clearly an
1161 * error case.
1162 */
1163 return 0;
1164 }
1165
1166 uint64_t
1167 ir_constant::get_uint64_component(unsigned i) const
1168 {
1169 switch (this->type->base_type) {
1170 case GLSL_TYPE_UINT: return this->value.u[i];
1171 case GLSL_TYPE_INT: return this->value.i[i];
1172 case GLSL_TYPE_FLOAT: return (uint64_t) this->value.f[i];
1173 case GLSL_TYPE_FLOAT16: return (uint64_t) _mesa_half_to_float(this->value.f16[i]);
1174 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1175 case GLSL_TYPE_DOUBLE: return (uint64_t) this->value.d[i];
1176 case GLSL_TYPE_SAMPLER:
1177 case GLSL_TYPE_IMAGE:
1178 case GLSL_TYPE_UINT64: return this->value.u64[i];
1179 case GLSL_TYPE_INT64: return (uint64_t) this->value.i64[i];
1180 default: assert(!"Should not get here."); break;
1181 }
1182
1183 /* Must return something to make the compiler happy. This is clearly an
1184 * error case.
1185 */
1186 return 0;
1187 }
1188
1189 ir_constant *
1190 ir_constant::get_array_element(unsigned i) const
1191 {
1192 assert(this->type->is_array());
1193
1194 /* From page 35 (page 41 of the PDF) of the GLSL 1.20 spec:
1195 *
1196 * "Behavior is undefined if a shader subscripts an array with an index
1197 * less than 0 or greater than or equal to the size the array was
1198 * declared with."
1199 *
1200 * Most out-of-bounds accesses are removed before things could get this far.
1201 * There are cases where non-constant array index values can get constant
1202 * folded.
1203 */
1204 if (int(i) < 0)
1205 i = 0;
1206 else if (i >= this->type->length)
1207 i = this->type->length - 1;
1208
1209 return const_elements[i];
1210 }
1211
1212 ir_constant *
1213 ir_constant::get_record_field(int idx)
1214 {
1215 assert(this->type->is_struct());
1216 assert(idx >= 0 && (unsigned) idx < this->type->length);
1217
1218 return const_elements[idx];
1219 }
1220
1221 void
1222 ir_constant::copy_offset(ir_constant *src, int offset)
1223 {
1224 switch (this->type->base_type) {
1225 case GLSL_TYPE_UINT:
1226 case GLSL_TYPE_INT:
1227 case GLSL_TYPE_FLOAT:
1228 case GLSL_TYPE_FLOAT16:
1229 case GLSL_TYPE_DOUBLE:
1230 case GLSL_TYPE_SAMPLER:
1231 case GLSL_TYPE_IMAGE:
1232 case GLSL_TYPE_UINT64:
1233 case GLSL_TYPE_INT64:
1234 case GLSL_TYPE_BOOL: {
1235 unsigned int size = src->type->components();
1236 assert (size <= this->type->components() - offset);
1237 for (unsigned int i=0; i<size; i++) {
1238 switch (this->type->base_type) {
1239 case GLSL_TYPE_UINT:
1240 value.u[i+offset] = src->get_uint_component(i);
1241 break;
1242 case GLSL_TYPE_INT:
1243 value.i[i+offset] = src->get_int_component(i);
1244 break;
1245 case GLSL_TYPE_FLOAT:
1246 value.f[i+offset] = src->get_float_component(i);
1247 break;
1248 case GLSL_TYPE_FLOAT16:
1249 value.f16[i+offset] = src->get_float16_component(i);
1250 break;
1251 case GLSL_TYPE_BOOL:
1252 value.b[i+offset] = src->get_bool_component(i);
1253 break;
1254 case GLSL_TYPE_DOUBLE:
1255 value.d[i+offset] = src->get_double_component(i);
1256 break;
1257 case GLSL_TYPE_SAMPLER:
1258 case GLSL_TYPE_IMAGE:
1259 case GLSL_TYPE_UINT64:
1260 value.u64[i+offset] = src->get_uint64_component(i);
1261 break;
1262 case GLSL_TYPE_INT64:
1263 value.i64[i+offset] = src->get_int64_component(i);
1264 break;
1265 default: // Shut up the compiler
1266 break;
1267 }
1268 }
1269 break;
1270 }
1271
1272 case GLSL_TYPE_STRUCT:
1273 case GLSL_TYPE_ARRAY: {
1274 assert (src->type == this->type);
1275 for (unsigned i = 0; i < this->type->length; i++) {
1276 this->const_elements[i] = src->const_elements[i]->clone(this, NULL);
1277 }
1278 break;
1279 }
1280
1281 default:
1282 assert(!"Should not get here.");
1283 break;
1284 }
1285 }
1286
1287 void
1288 ir_constant::copy_masked_offset(ir_constant *src, int offset, unsigned int mask)
1289 {
1290 assert (!type->is_array() && !type->is_struct());
1291
1292 if (!type->is_vector() && !type->is_matrix()) {
1293 offset = 0;
1294 mask = 1;
1295 }
1296
1297 int id = 0;
1298 for (int i=0; i<4; i++) {
1299 if (mask & (1 << i)) {
1300 switch (this->type->base_type) {
1301 case GLSL_TYPE_UINT:
1302 value.u[i+offset] = src->get_uint_component(id++);
1303 break;
1304 case GLSL_TYPE_INT:
1305 value.i[i+offset] = src->get_int_component(id++);
1306 break;
1307 case GLSL_TYPE_FLOAT:
1308 value.f[i+offset] = src->get_float_component(id++);
1309 break;
1310 case GLSL_TYPE_FLOAT16:
1311 value.f16[i+offset] = src->get_float16_component(id++);
1312 break;
1313 case GLSL_TYPE_BOOL:
1314 value.b[i+offset] = src->get_bool_component(id++);
1315 break;
1316 case GLSL_TYPE_DOUBLE:
1317 value.d[i+offset] = src->get_double_component(id++);
1318 break;
1319 case GLSL_TYPE_SAMPLER:
1320 case GLSL_TYPE_IMAGE:
1321 case GLSL_TYPE_UINT64:
1322 value.u64[i+offset] = src->get_uint64_component(id++);
1323 break;
1324 case GLSL_TYPE_INT64:
1325 value.i64[i+offset] = src->get_int64_component(id++);
1326 break;
1327 default:
1328 assert(!"Should not get here.");
1329 return;
1330 }
1331 }
1332 }
1333 }
1334
1335 bool
1336 ir_constant::has_value(const ir_constant *c) const
1337 {
1338 if (this->type != c->type)
1339 return false;
1340
1341 if (this->type->is_array() || this->type->is_struct()) {
1342 for (unsigned i = 0; i < this->type->length; i++) {
1343 if (!this->const_elements[i]->has_value(c->const_elements[i]))
1344 return false;
1345 }
1346 return true;
1347 }
1348
1349 for (unsigned i = 0; i < this->type->components(); i++) {
1350 switch (this->type->base_type) {
1351 case GLSL_TYPE_UINT:
1352 if (this->value.u[i] != c->value.u[i])
1353 return false;
1354 break;
1355 case GLSL_TYPE_INT:
1356 if (this->value.i[i] != c->value.i[i])
1357 return false;
1358 break;
1359 case GLSL_TYPE_FLOAT:
1360 if (this->value.f[i] != c->value.f[i])
1361 return false;
1362 break;
1363 case GLSL_TYPE_FLOAT16:
1364 /* Convert to float to make sure NaN and ±0.0 compares correctly */
1365 if (_mesa_half_to_float(this->value.f16[i]) !=
1366 _mesa_half_to_float(c->value.f16[i]))
1367 return false;
1368 break;
1369 case GLSL_TYPE_BOOL:
1370 if (this->value.b[i] != c->value.b[i])
1371 return false;
1372 break;
1373 case GLSL_TYPE_DOUBLE:
1374 if (this->value.d[i] != c->value.d[i])
1375 return false;
1376 break;
1377 case GLSL_TYPE_SAMPLER:
1378 case GLSL_TYPE_IMAGE:
1379 case GLSL_TYPE_UINT64:
1380 if (this->value.u64[i] != c->value.u64[i])
1381 return false;
1382 break;
1383 case GLSL_TYPE_INT64:
1384 if (this->value.i64[i] != c->value.i64[i])
1385 return false;
1386 break;
1387 default:
1388 assert(!"Should not get here.");
1389 return false;
1390 }
1391 }
1392
1393 return true;
1394 }
1395
1396 bool
1397 ir_constant::is_value(float f, int i) const
1398 {
1399 if (!this->type->is_scalar() && !this->type->is_vector())
1400 return false;
1401
1402 /* Only accept boolean values for 0/1. */
1403 if (int(bool(i)) != i && this->type->is_boolean())
1404 return false;
1405
1406 for (unsigned c = 0; c < this->type->vector_elements; c++) {
1407 switch (this->type->base_type) {
1408 case GLSL_TYPE_FLOAT:
1409 if (this->value.f[c] != f)
1410 return false;
1411 break;
1412 case GLSL_TYPE_FLOAT16:
1413 if (_mesa_half_to_float(this->value.f16[c]) != f)
1414 return false;
1415 break;
1416 case GLSL_TYPE_INT:
1417 if (this->value.i[c] != i)
1418 return false;
1419 break;
1420 case GLSL_TYPE_UINT:
1421 if (this->value.u[c] != unsigned(i))
1422 return false;
1423 break;
1424 case GLSL_TYPE_BOOL:
1425 if (this->value.b[c] != bool(i))
1426 return false;
1427 break;
1428 case GLSL_TYPE_DOUBLE:
1429 if (this->value.d[c] != double(f))
1430 return false;
1431 break;
1432 case GLSL_TYPE_SAMPLER:
1433 case GLSL_TYPE_IMAGE:
1434 case GLSL_TYPE_UINT64:
1435 if (this->value.u64[c] != uint64_t(i))
1436 return false;
1437 break;
1438 case GLSL_TYPE_INT64:
1439 if (this->value.i64[c] != i)
1440 return false;
1441 break;
1442 default:
1443 /* The only other base types are structures, arrays, and samplers.
1444 * Samplers cannot be constants, and the others should have been
1445 * filtered out above.
1446 */
1447 assert(!"Should not get here.");
1448 return false;
1449 }
1450 }
1451
1452 return true;
1453 }
1454
1455 bool
1456 ir_constant::is_zero() const
1457 {
1458 return is_value(0.0, 0);
1459 }
1460
1461 bool
1462 ir_constant::is_one() const
1463 {
1464 return is_value(1.0, 1);
1465 }
1466
1467 bool
1468 ir_constant::is_negative_one() const
1469 {
1470 return is_value(-1.0, -1);
1471 }
1472
1473 bool
1474 ir_constant::is_uint16_constant() const
1475 {
1476 if (!type->is_integer_32())
1477 return false;
1478
1479 return value.u[0] < (1 << 16);
1480 }
1481
1482 ir_loop::ir_loop()
1483 : ir_instruction(ir_type_loop)
1484 {
1485 }
1486
1487
1488 ir_dereference_variable::ir_dereference_variable(ir_variable *var)
1489 : ir_dereference(ir_type_dereference_variable)
1490 {
1491 assert(var != NULL);
1492
1493 this->var = var;
1494 this->type = var->type;
1495 }
1496
1497
1498 ir_dereference_array::ir_dereference_array(ir_rvalue *value,
1499 ir_rvalue *array_index)
1500 : ir_dereference(ir_type_dereference_array)
1501 {
1502 this->array_index = array_index;
1503 this->set_array(value);
1504 }
1505
1506
1507 ir_dereference_array::ir_dereference_array(ir_variable *var,
1508 ir_rvalue *array_index)
1509 : ir_dereference(ir_type_dereference_array)
1510 {
1511 void *ctx = ralloc_parent(var);
1512
1513 this->array_index = array_index;
1514 this->set_array(new(ctx) ir_dereference_variable(var));
1515 }
1516
1517
1518 void
1519 ir_dereference_array::set_array(ir_rvalue *value)
1520 {
1521 assert(value != NULL);
1522
1523 this->array = value;
1524
1525 const glsl_type *const vt = this->array->type;
1526
1527 if (vt->is_array()) {
1528 type = vt->fields.array;
1529 } else if (vt->is_matrix()) {
1530 type = vt->column_type();
1531 } else if (vt->is_vector()) {
1532 type = vt->get_base_type();
1533 }
1534 }
1535
1536
1537 ir_dereference_record::ir_dereference_record(ir_rvalue *value,
1538 const char *field)
1539 : ir_dereference(ir_type_dereference_record)
1540 {
1541 assert(value != NULL);
1542
1543 this->record = value;
1544 this->type = this->record->type->field_type(field);
1545 this->field_idx = this->record->type->field_index(field);
1546 }
1547
1548
1549 ir_dereference_record::ir_dereference_record(ir_variable *var,
1550 const char *field)
1551 : ir_dereference(ir_type_dereference_record)
1552 {
1553 void *ctx = ralloc_parent(var);
1554
1555 this->record = new(ctx) ir_dereference_variable(var);
1556 this->type = this->record->type->field_type(field);
1557 this->field_idx = this->record->type->field_index(field);
1558 }
1559
1560 bool
1561 ir_dereference::is_lvalue(const struct _mesa_glsl_parse_state *state) const
1562 {
1563 ir_variable *var = this->variable_referenced();
1564
1565 /* Every l-value derference chain eventually ends in a variable.
1566 */
1567 if ((var == NULL) || var->data.read_only)
1568 return false;
1569
1570 /* From section 4.1.7 of the ARB_bindless_texture spec:
1571 *
1572 * "Samplers can be used as l-values, so can be assigned into and used as
1573 * "out" and "inout" function parameters."
1574 *
1575 * From section 4.1.X of the ARB_bindless_texture spec:
1576 *
1577 * "Images can be used as l-values, so can be assigned into and used as
1578 * "out" and "inout" function parameters."
1579 */
1580 if ((!state || state->has_bindless()) &&
1581 (this->type->contains_sampler() || this->type->contains_image()))
1582 return true;
1583
1584 /* From section 4.1.7 of the GLSL 4.40 spec:
1585 *
1586 * "Opaque variables cannot be treated as l-values; hence cannot
1587 * be used as out or inout function parameters, nor can they be
1588 * assigned into."
1589 */
1590 if (this->type->contains_opaque())
1591 return false;
1592
1593 return true;
1594 }
1595
1596
1597 static const char * const tex_opcode_strs[] = { "tex", "txb", "txl", "txd", "txf", "txf_ms", "txs", "lod", "tg4", "query_levels", "texture_samples", "samples_identical" };
1598
1599 const char *ir_texture::opcode_string()
1600 {
1601 assert((unsigned int) op < ARRAY_SIZE(tex_opcode_strs));
1602 return tex_opcode_strs[op];
1603 }
1604
1605 ir_texture_opcode
1606 ir_texture::get_opcode(const char *str)
1607 {
1608 const int count = sizeof(tex_opcode_strs) / sizeof(tex_opcode_strs[0]);
1609 for (int op = 0; op < count; op++) {
1610 if (strcmp(str, tex_opcode_strs[op]) == 0)
1611 return (ir_texture_opcode) op;
1612 }
1613 return (ir_texture_opcode) -1;
1614 }
1615
1616
1617 void
1618 ir_texture::set_sampler(ir_dereference *sampler, const glsl_type *type)
1619 {
1620 assert(sampler != NULL);
1621 assert(type != NULL);
1622 this->sampler = sampler;
1623 this->type = type;
1624
1625 if (this->op == ir_txs || this->op == ir_query_levels ||
1626 this->op == ir_texture_samples) {
1627 assert(type->base_type == GLSL_TYPE_INT);
1628 } else if (this->op == ir_lod) {
1629 assert(type->vector_elements == 2);
1630 assert(type->is_float());
1631 } else if (this->op == ir_samples_identical) {
1632 assert(type == glsl_type::bool_type);
1633 assert(sampler->type->is_sampler());
1634 assert(sampler->type->sampler_dimensionality == GLSL_SAMPLER_DIM_MS);
1635 } else {
1636 assert(sampler->type->sampled_type == (int) type->base_type);
1637 if (sampler->type->sampler_shadow)
1638 assert(type->vector_elements == 4 || type->vector_elements == 1);
1639 else
1640 assert(type->vector_elements == 4);
1641 }
1642 }
1643
1644
1645 void
1646 ir_swizzle::init_mask(const unsigned *comp, unsigned count)
1647 {
1648 assert((count >= 1) && (count <= 4));
1649
1650 memset(&this->mask, 0, sizeof(this->mask));
1651 this->mask.num_components = count;
1652
1653 unsigned dup_mask = 0;
1654 switch (count) {
1655 case 4:
1656 assert(comp[3] <= 3);
1657 dup_mask |= (1U << comp[3])
1658 & ((1U << comp[0]) | (1U << comp[1]) | (1U << comp[2]));
1659 this->mask.w = comp[3];
1660
1661 case 3:
1662 assert(comp[2] <= 3);
1663 dup_mask |= (1U << comp[2])
1664 & ((1U << comp[0]) | (1U << comp[1]));
1665 this->mask.z = comp[2];
1666
1667 case 2:
1668 assert(comp[1] <= 3);
1669 dup_mask |= (1U << comp[1])
1670 & ((1U << comp[0]));
1671 this->mask.y = comp[1];
1672
1673 case 1:
1674 assert(comp[0] <= 3);
1675 this->mask.x = comp[0];
1676 }
1677
1678 this->mask.has_duplicates = dup_mask != 0;
1679
1680 /* Based on the number of elements in the swizzle and the base type
1681 * (i.e., float, int, unsigned, or bool) of the vector being swizzled,
1682 * generate the type of the resulting value.
1683 */
1684 type = glsl_type::get_instance(val->type->base_type, mask.num_components, 1);
1685 }
1686
1687 ir_swizzle::ir_swizzle(ir_rvalue *val, unsigned x, unsigned y, unsigned z,
1688 unsigned w, unsigned count)
1689 : ir_rvalue(ir_type_swizzle), val(val)
1690 {
1691 const unsigned components[4] = { x, y, z, w };
1692 this->init_mask(components, count);
1693 }
1694
1695 ir_swizzle::ir_swizzle(ir_rvalue *val, const unsigned *comp,
1696 unsigned count)
1697 : ir_rvalue(ir_type_swizzle), val(val)
1698 {
1699 this->init_mask(comp, count);
1700 }
1701
1702 ir_swizzle::ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask)
1703 : ir_rvalue(ir_type_swizzle), val(val), mask(mask)
1704 {
1705 this->type = glsl_type::get_instance(val->type->base_type,
1706 mask.num_components, 1);
1707 }
1708
1709 #define X 1
1710 #define R 5
1711 #define S 9
1712 #define I 13
1713
1714 ir_swizzle *
1715 ir_swizzle::create(ir_rvalue *val, const char *str, unsigned vector_length)
1716 {
1717 void *ctx = ralloc_parent(val);
1718
1719 /* For each possible swizzle character, this table encodes the value in
1720 * \c idx_map that represents the 0th element of the vector. For invalid
1721 * swizzle characters (e.g., 'k'), a special value is used that will allow
1722 * detection of errors.
1723 */
1724 static const unsigned char base_idx[26] = {
1725 /* a b c d e f g h i j k l m */
1726 R, R, I, I, I, I, R, I, I, I, I, I, I,
1727 /* n o p q r s t u v w x y z */
1728 I, I, S, S, R, S, S, I, I, X, X, X, X
1729 };
1730
1731 /* Each valid swizzle character has an entry in the previous table. This
1732 * table encodes the base index encoded in the previous table plus the actual
1733 * index of the swizzle character. When processing swizzles, the first
1734 * character in the string is indexed in the previous table. Each character
1735 * in the string is indexed in this table, and the value found there has the
1736 * value form the first table subtracted. The result must be on the range
1737 * [0,3].
1738 *
1739 * For example, the string "wzyx" will get X from the first table. Each of
1740 * the charcaters will get X+3, X+2, X+1, and X+0 from this table. After
1741 * subtraction, the swizzle values are { 3, 2, 1, 0 }.
1742 *
1743 * The string "wzrg" will get X from the first table. Each of the characters
1744 * will get X+3, X+2, R+0, and R+1 from this table. After subtraction, the
1745 * swizzle values are { 3, 2, 4, 5 }. Since 4 and 5 are outside the range
1746 * [0,3], the error is detected.
1747 */
1748 static const unsigned char idx_map[26] = {
1749 /* a b c d e f g h i j k l m */
1750 R+3, R+2, 0, 0, 0, 0, R+1, 0, 0, 0, 0, 0, 0,
1751 /* n o p q r s t u v w x y z */
1752 0, 0, S+2, S+3, R+0, S+0, S+1, 0, 0, X+3, X+0, X+1, X+2
1753 };
1754
1755 int swiz_idx[4] = { 0, 0, 0, 0 };
1756 unsigned i;
1757
1758
1759 /* Validate the first character in the swizzle string and look up the base
1760 * index value as described above.
1761 */
1762 if ((str[0] < 'a') || (str[0] > 'z'))
1763 return NULL;
1764
1765 const unsigned base = base_idx[str[0] - 'a'];
1766
1767
1768 for (i = 0; (i < 4) && (str[i] != '\0'); i++) {
1769 /* Validate the next character, and, as described above, convert it to a
1770 * swizzle index.
1771 */
1772 if ((str[i] < 'a') || (str[i] > 'z'))
1773 return NULL;
1774
1775 swiz_idx[i] = idx_map[str[i] - 'a'] - base;
1776 if ((swiz_idx[i] < 0) || (swiz_idx[i] >= (int) vector_length))
1777 return NULL;
1778 }
1779
1780 if (str[i] != '\0')
1781 return NULL;
1782
1783 return new(ctx) ir_swizzle(val, swiz_idx[0], swiz_idx[1], swiz_idx[2],
1784 swiz_idx[3], i);
1785 }
1786
1787 #undef X
1788 #undef R
1789 #undef S
1790 #undef I
1791
1792 ir_variable *
1793 ir_swizzle::variable_referenced() const
1794 {
1795 return this->val->variable_referenced();
1796 }
1797
1798
1799 bool ir_variable::temporaries_allocate_names = false;
1800
1801 const char ir_variable::tmp_name[] = "compiler_temp";
1802
1803 ir_variable::ir_variable(const struct glsl_type *type, const char *name,
1804 ir_variable_mode mode)
1805 : ir_instruction(ir_type_variable)
1806 {
1807 this->type = type;
1808
1809 if (mode == ir_var_temporary && !ir_variable::temporaries_allocate_names)
1810 name = NULL;
1811
1812 /* The ir_variable clone method may call this constructor with name set to
1813 * tmp_name.
1814 */
1815 assert(name != NULL
1816 || mode == ir_var_temporary
1817 || mode == ir_var_function_in
1818 || mode == ir_var_function_out
1819 || mode == ir_var_function_inout);
1820 assert(name != ir_variable::tmp_name
1821 || mode == ir_var_temporary);
1822 if (mode == ir_var_temporary
1823 && (name == NULL || name == ir_variable::tmp_name)) {
1824 this->name = ir_variable::tmp_name;
1825 } else if (name == NULL ||
1826 strlen(name) < ARRAY_SIZE(this->name_storage)) {
1827 strcpy(this->name_storage, name ? name : "");
1828 this->name = this->name_storage;
1829 } else {
1830 this->name = ralloc_strdup(this, name);
1831 }
1832
1833 this->u.max_ifc_array_access = NULL;
1834
1835 this->data.explicit_location = false;
1836 this->data.explicit_index = false;
1837 this->data.explicit_binding = false;
1838 this->data.explicit_component = false;
1839 this->data.has_initializer = false;
1840 this->data.is_unmatched_generic_inout = false;
1841 this->data.is_xfb_only = false;
1842 this->data.explicit_xfb_buffer = false;
1843 this->data.explicit_xfb_offset = false;
1844 this->data.explicit_xfb_stride = false;
1845 this->data.location = -1;
1846 this->data.location_frac = 0;
1847 this->data.matrix_layout = GLSL_MATRIX_LAYOUT_INHERITED;
1848 this->data.from_named_ifc_block = false;
1849 this->data.must_be_shader_input = false;
1850 this->data.index = 0;
1851 this->data.binding = 0;
1852 this->data.warn_extension_index = 0;
1853 this->constant_value = NULL;
1854 this->constant_initializer = NULL;
1855 this->data.depth_layout = ir_depth_layout_none;
1856 this->data.used = false;
1857 this->data.assigned = false;
1858 this->data.always_active_io = false;
1859 this->data.read_only = false;
1860 this->data.centroid = false;
1861 this->data.sample = false;
1862 this->data.patch = false;
1863 this->data.explicit_invariant = false;
1864 this->data.invariant = false;
1865 this->data.precise = false;
1866 this->data.how_declared = ir_var_declared_normally;
1867 this->data.mode = mode;
1868 this->data.interpolation = INTERP_MODE_NONE;
1869 this->data.max_array_access = -1;
1870 this->data.offset = 0;
1871 this->data.precision = GLSL_PRECISION_NONE;
1872 this->data.memory_read_only = false;
1873 this->data.memory_write_only = false;
1874 this->data.memory_coherent = false;
1875 this->data.memory_volatile = false;
1876 this->data.memory_restrict = false;
1877 this->data.from_ssbo_unsized_array = false;
1878 this->data.implicit_sized_array = false;
1879 this->data.fb_fetch_output = false;
1880 this->data.bindless = false;
1881 this->data.bound = false;
1882 this->data.image_format = PIPE_FORMAT_NONE;
1883 this->data._num_state_slots = 0;
1884 this->data.param_index = 0;
1885 this->data.stream = 0;
1886 this->data.xfb_buffer = -1;
1887 this->data.xfb_stride = -1;
1888
1889 this->interface_type = NULL;
1890
1891 if (type != NULL) {
1892 if (type->is_interface())
1893 this->init_interface_type(type);
1894 else if (type->without_array()->is_interface())
1895 this->init_interface_type(type->without_array());
1896 }
1897 }
1898
1899
1900 const char *
1901 interpolation_string(unsigned interpolation)
1902 {
1903 switch (interpolation) {
1904 case INTERP_MODE_NONE: return "no";
1905 case INTERP_MODE_SMOOTH: return "smooth";
1906 case INTERP_MODE_FLAT: return "flat";
1907 case INTERP_MODE_NOPERSPECTIVE: return "noperspective";
1908 }
1909
1910 assert(!"Should not get here.");
1911 return "";
1912 }
1913
1914 const char *const ir_variable::warn_extension_table[] = {
1915 "",
1916 "GL_ARB_shader_stencil_export",
1917 "GL_AMD_shader_stencil_export",
1918 };
1919
1920 void
1921 ir_variable::enable_extension_warning(const char *extension)
1922 {
1923 for (unsigned i = 0; i < ARRAY_SIZE(warn_extension_table); i++) {
1924 if (strcmp(warn_extension_table[i], extension) == 0) {
1925 this->data.warn_extension_index = i;
1926 return;
1927 }
1928 }
1929
1930 assert(!"Should not get here.");
1931 this->data.warn_extension_index = 0;
1932 }
1933
1934 const char *
1935 ir_variable::get_extension_warning() const
1936 {
1937 return this->data.warn_extension_index == 0
1938 ? NULL : warn_extension_table[this->data.warn_extension_index];
1939 }
1940
1941 ir_function_signature::ir_function_signature(const glsl_type *return_type,
1942 builtin_available_predicate b)
1943 : ir_instruction(ir_type_function_signature),
1944 return_type(return_type), is_defined(false),
1945 return_precision(GLSL_PRECISION_NONE),
1946 intrinsic_id(ir_intrinsic_invalid), builtin_avail(b), _function(NULL)
1947 {
1948 this->origin = NULL;
1949 }
1950
1951
1952 bool
1953 ir_function_signature::is_builtin() const
1954 {
1955 return builtin_avail != NULL;
1956 }
1957
1958
1959 bool
1960 ir_function_signature::is_builtin_available(const _mesa_glsl_parse_state *state) const
1961 {
1962 /* We can't call the predicate without a state pointer, so just say that
1963 * the signature is available. At compile time, we need the filtering,
1964 * but also receive a valid state pointer. At link time, we're resolving
1965 * imported built-in prototypes to their definitions, which will always
1966 * be an exact match. So we can skip the filtering.
1967 */
1968 if (state == NULL)
1969 return true;
1970
1971 assert(builtin_avail != NULL);
1972 return builtin_avail(state);
1973 }
1974
1975
1976 static bool
1977 modes_match(unsigned a, unsigned b)
1978 {
1979 if (a == b)
1980 return true;
1981
1982 /* Accept "in" vs. "const in" */
1983 if ((a == ir_var_const_in && b == ir_var_function_in) ||
1984 (b == ir_var_const_in && a == ir_var_function_in))
1985 return true;
1986
1987 return false;
1988 }
1989
1990
1991 const char *
1992 ir_function_signature::qualifiers_match(exec_list *params)
1993 {
1994 /* check that the qualifiers match. */
1995 foreach_two_lists(a_node, &this->parameters, b_node, params) {
1996 ir_variable *a = (ir_variable *) a_node;
1997 ir_variable *b = (ir_variable *) b_node;
1998
1999 if (a->data.read_only != b->data.read_only ||
2000 !modes_match(a->data.mode, b->data.mode) ||
2001 a->data.interpolation != b->data.interpolation ||
2002 a->data.centroid != b->data.centroid ||
2003 a->data.sample != b->data.sample ||
2004 a->data.patch != b->data.patch ||
2005 a->data.memory_read_only != b->data.memory_read_only ||
2006 a->data.memory_write_only != b->data.memory_write_only ||
2007 a->data.memory_coherent != b->data.memory_coherent ||
2008 a->data.memory_volatile != b->data.memory_volatile ||
2009 a->data.memory_restrict != b->data.memory_restrict) {
2010
2011 /* parameter a's qualifiers don't match */
2012 return a->name;
2013 }
2014 }
2015 return NULL;
2016 }
2017
2018
2019 void
2020 ir_function_signature::replace_parameters(exec_list *new_params)
2021 {
2022 /* Destroy all of the previous parameter information. If the previous
2023 * parameter information comes from the function prototype, it may either
2024 * specify incorrect parameter names or not have names at all.
2025 */
2026 new_params->move_nodes_to(&parameters);
2027 }
2028
2029
2030 ir_function::ir_function(const char *name)
2031 : ir_instruction(ir_type_function)
2032 {
2033 this->subroutine_index = -1;
2034 this->name = ralloc_strdup(this, name);
2035 }
2036
2037
2038 bool
2039 ir_function::has_user_signature()
2040 {
2041 foreach_in_list(ir_function_signature, sig, &this->signatures) {
2042 if (!sig->is_builtin())
2043 return true;
2044 }
2045 return false;
2046 }
2047
2048
2049 ir_rvalue *
2050 ir_rvalue::error_value(void *mem_ctx)
2051 {
2052 ir_rvalue *v = new(mem_ctx) ir_rvalue(ir_type_unset);
2053
2054 v->type = glsl_type::error_type;
2055 return v;
2056 }
2057
2058
2059 void
2060 visit_exec_list(exec_list *list, ir_visitor *visitor)
2061 {
2062 foreach_in_list_safe(ir_instruction, node, list) {
2063 node->accept(visitor);
2064 }
2065 }
2066
2067
2068 static void
2069 steal_memory(ir_instruction *ir, void *new_ctx)
2070 {
2071 ir_variable *var = ir->as_variable();
2072 ir_function *fn = ir->as_function();
2073 ir_constant *constant = ir->as_constant();
2074 if (var != NULL && var->constant_value != NULL)
2075 steal_memory(var->constant_value, ir);
2076
2077 if (var != NULL && var->constant_initializer != NULL)
2078 steal_memory(var->constant_initializer, ir);
2079
2080 if (fn != NULL && fn->subroutine_types)
2081 ralloc_steal(new_ctx, fn->subroutine_types);
2082
2083 /* The components of aggregate constants are not visited by the normal
2084 * visitor, so steal their values by hand.
2085 */
2086 if (constant != NULL &&
2087 (constant->type->is_array() || constant->type->is_struct())) {
2088 for (unsigned int i = 0; i < constant->type->length; i++) {
2089 steal_memory(constant->const_elements[i], ir);
2090 }
2091 }
2092
2093 ralloc_steal(new_ctx, ir);
2094 }
2095
2096
2097 void
2098 reparent_ir(exec_list *list, void *mem_ctx)
2099 {
2100 foreach_in_list(ir_instruction, node, list) {
2101 visit_tree(node, steal_memory, mem_ctx);
2102 }
2103 }
2104
2105
2106 static ir_rvalue *
2107 try_min_one(ir_rvalue *ir)
2108 {
2109 ir_expression *expr = ir->as_expression();
2110
2111 if (!expr || expr->operation != ir_binop_min)
2112 return NULL;
2113
2114 if (expr->operands[0]->is_one())
2115 return expr->operands[1];
2116
2117 if (expr->operands[1]->is_one())
2118 return expr->operands[0];
2119
2120 return NULL;
2121 }
2122
2123 static ir_rvalue *
2124 try_max_zero(ir_rvalue *ir)
2125 {
2126 ir_expression *expr = ir->as_expression();
2127
2128 if (!expr || expr->operation != ir_binop_max)
2129 return NULL;
2130
2131 if (expr->operands[0]->is_zero())
2132 return expr->operands[1];
2133
2134 if (expr->operands[1]->is_zero())
2135 return expr->operands[0];
2136
2137 return NULL;
2138 }
2139
2140 ir_rvalue *
2141 ir_rvalue::as_rvalue_to_saturate()
2142 {
2143 ir_expression *expr = this->as_expression();
2144
2145 if (!expr)
2146 return NULL;
2147
2148 ir_rvalue *max_zero = try_max_zero(expr);
2149 if (max_zero) {
2150 return try_min_one(max_zero);
2151 } else {
2152 ir_rvalue *min_one = try_min_one(expr);
2153 if (min_one) {
2154 return try_max_zero(min_one);
2155 }
2156 }
2157
2158 return NULL;
2159 }
2160
2161
2162 unsigned
2163 vertices_per_prim(GLenum prim)
2164 {
2165 switch (prim) {
2166 case GL_POINTS:
2167 return 1;
2168 case GL_LINES:
2169 return 2;
2170 case GL_TRIANGLES:
2171 return 3;
2172 case GL_LINES_ADJACENCY:
2173 return 4;
2174 case GL_TRIANGLES_ADJACENCY:
2175 return 6;
2176 default:
2177 assert(!"Bad primitive");
2178 return 3;
2179 }
2180 }
2181
2182 /**
2183 * Generate a string describing the mode of a variable
2184 */
2185 const char *
2186 mode_string(const ir_variable *var)
2187 {
2188 switch (var->data.mode) {
2189 case ir_var_auto:
2190 return (var->data.read_only) ? "global constant" : "global variable";
2191
2192 case ir_var_uniform:
2193 return "uniform";
2194
2195 case ir_var_shader_storage:
2196 return "buffer";
2197
2198 case ir_var_shader_in:
2199 return "shader input";
2200
2201 case ir_var_shader_out:
2202 return "shader output";
2203
2204 case ir_var_function_in:
2205 case ir_var_const_in:
2206 return "function input";
2207
2208 case ir_var_function_out:
2209 return "function output";
2210
2211 case ir_var_function_inout:
2212 return "function inout";
2213
2214 case ir_var_system_value:
2215 return "shader input";
2216
2217 case ir_var_temporary:
2218 return "compiler temporary";
2219
2220 case ir_var_mode_count:
2221 break;
2222 }
2223
2224 assert(!"Should not get here.");
2225 return "invalid variable";
2226 }