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
ir_rvalue(enum ir_node_type t)30 ir_rvalue::ir_rvalue(enum ir_node_type t)
31 : ir_instruction(t)
32 {
33 this->type = glsl_type::error_type;
34 }
35
is_zero() const36 bool ir_rvalue::is_zero() const
37 {
38 return false;
39 }
40
is_one() const41 bool ir_rvalue::is_one() const
42 {
43 return false;
44 }
45
is_negative_one() const46 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
update_rhs_swizzle(ir_swizzle_mask & m,unsigned from,unsigned to)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
set_lhs(ir_rvalue * lhs)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 *
whole_variable_written()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
ir_assignment(ir_dereference * lhs,ir_rvalue * rhs,ir_rvalue * condition,unsigned write_mask)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
ir_assignment(ir_rvalue * lhs,ir_rvalue * rhs,ir_rvalue * condition)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
ir_expression(int op,const struct glsl_type * type,ir_rvalue * op0,ir_rvalue * op1,ir_rvalue * op2,ir_rvalue * op3)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
ir_expression(int op,ir_rvalue * op0)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_i2imp:
303 this->type = glsl_type::get_instance(GLSL_TYPE_INT16,
304 op0->type->vector_elements, 1);
305 break;
306
307 case ir_unop_i2i:
308 if (op0->type->base_type == GLSL_TYPE_INT) {
309 this->type = glsl_type::get_instance(GLSL_TYPE_INT16,
310 op0->type->vector_elements, 1);
311 } else {
312 assert(op0->type->base_type == GLSL_TYPE_INT16);
313 this->type = glsl_type::get_instance(GLSL_TYPE_INT,
314 op0->type->vector_elements, 1);
315 }
316 break;
317
318 case ir_unop_u2u:
319 if (op0->type->base_type == GLSL_TYPE_UINT) {
320 this->type = glsl_type::get_instance(GLSL_TYPE_UINT16,
321 op0->type->vector_elements, 1);
322 } else {
323 assert(op0->type->base_type == GLSL_TYPE_UINT16);
324 this->type = glsl_type::get_instance(GLSL_TYPE_UINT,
325 op0->type->vector_elements, 1);
326 }
327 break;
328
329 case ir_unop_u2ump:
330 this->type = glsl_type::get_instance(GLSL_TYPE_UINT16,
331 op0->type->vector_elements, 1);
332 break;
333
334 case ir_unop_f2b:
335 case ir_unop_i2b:
336 case ir_unop_d2b:
337 case ir_unop_f162b:
338 case ir_unop_i642b:
339 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL,
340 op0->type->vector_elements, 1);
341 break;
342
343 case ir_unop_f2d:
344 case ir_unop_i2d:
345 case ir_unop_u2d:
346 case ir_unop_i642d:
347 case ir_unop_u642d:
348 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE,
349 op0->type->vector_elements, 1);
350 break;
351
352 case ir_unop_i2u:
353 case ir_unop_f2u:
354 case ir_unop_d2u:
355 case ir_unop_bitcast_f2u:
356 case ir_unop_i642u:
357 case ir_unop_u642u:
358 this->type = glsl_type::get_instance(GLSL_TYPE_UINT,
359 op0->type->vector_elements, 1);
360 break;
361
362 case ir_unop_i2i64:
363 case ir_unop_u2i64:
364 case ir_unop_b2i64:
365 case ir_unop_f2i64:
366 case ir_unop_d2i64:
367 case ir_unop_u642i64:
368 this->type = glsl_type::get_instance(GLSL_TYPE_INT64,
369 op0->type->vector_elements, 1);
370 break;
371
372 case ir_unop_i2u64:
373 case ir_unop_u2u64:
374 case ir_unop_f2u64:
375 case ir_unop_d2u64:
376 case ir_unop_i642u64:
377 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64,
378 op0->type->vector_elements, 1);
379 break;
380
381 case ir_unop_unpack_double_2x32:
382 case ir_unop_unpack_uint_2x32:
383 this->type = glsl_type::uvec2_type;
384 break;
385
386 case ir_unop_unpack_int_2x32:
387 this->type = glsl_type::ivec2_type;
388 break;
389
390 case ir_unop_pack_snorm_2x16:
391 case ir_unop_pack_snorm_4x8:
392 case ir_unop_pack_unorm_2x16:
393 case ir_unop_pack_unorm_4x8:
394 case ir_unop_pack_half_2x16:
395 this->type = glsl_type::uint_type;
396 break;
397
398 case ir_unop_pack_double_2x32:
399 this->type = glsl_type::double_type;
400 break;
401
402 case ir_unop_pack_int_2x32:
403 this->type = glsl_type::int64_t_type;
404 break;
405
406 case ir_unop_pack_uint_2x32:
407 this->type = glsl_type::uint64_t_type;
408 break;
409
410 case ir_unop_unpack_snorm_2x16:
411 case ir_unop_unpack_unorm_2x16:
412 case ir_unop_unpack_half_2x16:
413 this->type = glsl_type::vec2_type;
414 break;
415
416 case ir_unop_unpack_snorm_4x8:
417 case ir_unop_unpack_unorm_4x8:
418 this->type = glsl_type::vec4_type;
419 break;
420
421 case ir_unop_unpack_sampler_2x32:
422 case ir_unop_unpack_image_2x32:
423 this->type = glsl_type::uvec2_type;
424 break;
425
426 case ir_unop_pack_sampler_2x32:
427 case ir_unop_pack_image_2x32:
428 this->type = op0->type;
429 break;
430
431 case ir_unop_frexp_sig:
432 this->type = op0->type;
433 break;
434 case ir_unop_frexp_exp:
435 this->type = glsl_type::get_instance(GLSL_TYPE_INT,
436 op0->type->vector_elements, 1);
437 break;
438
439 case ir_unop_get_buffer_size:
440 case ir_unop_ssbo_unsized_array_length:
441 case ir_unop_implicitly_sized_array_length:
442 this->type = glsl_type::int_type;
443 break;
444
445 case ir_unop_bitcast_i642d:
446 case ir_unop_bitcast_u642d:
447 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE,
448 op0->type->vector_elements, 1);
449 break;
450
451 case ir_unop_bitcast_d2i64:
452 this->type = glsl_type::get_instance(GLSL_TYPE_INT64,
453 op0->type->vector_elements, 1);
454 break;
455 case ir_unop_bitcast_d2u64:
456 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64,
457 op0->type->vector_elements, 1);
458 break;
459
460 default:
461 assert(!"not reached: missing automatic type setup for ir_expression");
462 this->type = op0->type;
463 break;
464 }
465 }
466
ir_expression(int op,ir_rvalue * op0,ir_rvalue * op1)467 ir_expression::ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1)
468 : ir_rvalue(ir_type_expression)
469 {
470 this->operation = ir_expression_operation(op);
471 this->operands[0] = op0;
472 this->operands[1] = op1;
473 this->operands[2] = NULL;
474 this->operands[3] = NULL;
475
476 assert(op > ir_last_unop);
477 init_num_operands();
478 assert(num_operands == 2);
479 for (unsigned i = 0; i < num_operands; i++) {
480 assert(this->operands[i] != NULL);
481 }
482
483 switch (this->operation) {
484 case ir_binop_all_equal:
485 case ir_binop_any_nequal:
486 this->type = glsl_type::bool_type;
487 break;
488
489 case ir_binop_add:
490 case ir_binop_sub:
491 case ir_binop_min:
492 case ir_binop_max:
493 case ir_binop_pow:
494 case ir_binop_mul:
495 case ir_binop_div:
496 case ir_binop_mod:
497 case ir_binop_atan2:
498 if (op0->type->is_scalar()) {
499 this->type = op1->type;
500 } else if (op1->type->is_scalar()) {
501 this->type = op0->type;
502 } else {
503 if (this->operation == ir_binop_mul) {
504 this->type = glsl_type::get_mul_type(op0->type, op1->type);
505 } else {
506 assert(op0->type == op1->type);
507 this->type = op0->type;
508 }
509 }
510 break;
511
512 case ir_binop_logic_and:
513 case ir_binop_logic_xor:
514 case ir_binop_logic_or:
515 case ir_binop_bit_and:
516 case ir_binop_bit_xor:
517 case ir_binop_bit_or:
518 assert(!op0->type->is_matrix());
519 assert(!op1->type->is_matrix());
520 if (op0->type->is_scalar()) {
521 this->type = op1->type;
522 } else if (op1->type->is_scalar()) {
523 this->type = op0->type;
524 } else {
525 assert(op0->type->vector_elements == op1->type->vector_elements);
526 this->type = op0->type;
527 }
528 break;
529
530 case ir_binop_equal:
531 case ir_binop_nequal:
532 case ir_binop_gequal:
533 case ir_binop_less:
534 assert(op0->type == op1->type);
535 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL,
536 op0->type->vector_elements, 1);
537 break;
538
539 case ir_binop_dot:
540 this->type = op0->type->get_base_type();
541 break;
542
543 case ir_binop_imul_high:
544 case ir_binop_mul_32x16:
545 case ir_binop_carry:
546 case ir_binop_borrow:
547 case ir_binop_lshift:
548 case ir_binop_rshift:
549 case ir_binop_ldexp:
550 case ir_binop_interpolate_at_offset:
551 case ir_binop_interpolate_at_sample:
552 this->type = op0->type;
553 break;
554
555 case ir_binop_add_sat:
556 case ir_binop_sub_sat:
557 case ir_binop_avg:
558 case ir_binop_avg_round:
559 assert(op0->type == op1->type);
560 this->type = op0->type;
561 break;
562
563 case ir_binop_abs_sub: {
564 enum glsl_base_type base;
565
566 assert(op0->type == op1->type);
567
568 switch (op0->type->base_type) {
569 case GLSL_TYPE_UINT:
570 case GLSL_TYPE_INT:
571 base = GLSL_TYPE_UINT;
572 break;
573 case GLSL_TYPE_UINT8:
574 case GLSL_TYPE_INT8:
575 base = GLSL_TYPE_UINT8;
576 break;
577 case GLSL_TYPE_UINT16:
578 case GLSL_TYPE_INT16:
579 base = GLSL_TYPE_UINT16;
580 break;
581 case GLSL_TYPE_UINT64:
582 case GLSL_TYPE_INT64:
583 base = GLSL_TYPE_UINT64;
584 break;
585 default:
586 unreachable(!"Invalid base type.");
587 }
588
589 this->type = glsl_type::get_instance(base, op0->type->vector_elements, 1);
590 break;
591 }
592
593 case ir_binop_vector_extract:
594 this->type = op0->type->get_scalar_type();
595 break;
596
597 default:
598 assert(!"not reached: missing automatic type setup for ir_expression");
599 this->type = glsl_type::float_type;
600 }
601 }
602
ir_expression(int op,ir_rvalue * op0,ir_rvalue * op1,ir_rvalue * op2)603 ir_expression::ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1,
604 ir_rvalue *op2)
605 : ir_rvalue(ir_type_expression)
606 {
607 this->operation = ir_expression_operation(op);
608 this->operands[0] = op0;
609 this->operands[1] = op1;
610 this->operands[2] = op2;
611 this->operands[3] = NULL;
612
613 assert(op > ir_last_binop && op <= ir_last_triop);
614 init_num_operands();
615 assert(num_operands == 3);
616 for (unsigned i = 0; i < num_operands; i++) {
617 assert(this->operands[i] != NULL);
618 }
619
620 switch (this->operation) {
621 case ir_triop_fma:
622 case ir_triop_lrp:
623 case ir_triop_bitfield_extract:
624 case ir_triop_vector_insert:
625 this->type = op0->type;
626 break;
627
628 case ir_triop_csel:
629 this->type = op1->type;
630 break;
631
632 default:
633 assert(!"not reached: missing automatic type setup for ir_expression");
634 this->type = glsl_type::float_type;
635 }
636 }
637
638 /**
639 * This is only here for ir_reader to used for testing purposes. Please use
640 * the precomputed num_operands field if you need the number of operands.
641 */
642 unsigned
get_num_operands(ir_expression_operation op)643 ir_expression::get_num_operands(ir_expression_operation op)
644 {
645 assert(op <= ir_last_opcode);
646
647 if (op <= ir_last_unop)
648 return 1;
649
650 if (op <= ir_last_binop)
651 return 2;
652
653 if (op <= ir_last_triop)
654 return 3;
655
656 if (op <= ir_last_quadop)
657 return 4;
658
659 unreachable("Could not calculate number of operands");
660 }
661
662 #include "ir_expression_operation_strings.h"
663
664 const char*
depth_layout_string(ir_depth_layout layout)665 depth_layout_string(ir_depth_layout layout)
666 {
667 switch(layout) {
668 case ir_depth_layout_none: return "";
669 case ir_depth_layout_any: return "depth_any";
670 case ir_depth_layout_greater: return "depth_greater";
671 case ir_depth_layout_less: return "depth_less";
672 case ir_depth_layout_unchanged: return "depth_unchanged";
673
674 default:
675 assert(0);
676 return "";
677 }
678 }
679
680 ir_expression_operation
get_operator(const char * str)681 ir_expression::get_operator(const char *str)
682 {
683 for (int op = 0; op <= int(ir_last_opcode); op++) {
684 if (strcmp(str, ir_expression_operation_strings[op]) == 0)
685 return (ir_expression_operation) op;
686 }
687 return (ir_expression_operation) -1;
688 }
689
690 ir_variable *
variable_referenced() const691 ir_expression::variable_referenced() const
692 {
693 switch (operation) {
694 case ir_binop_vector_extract:
695 case ir_triop_vector_insert:
696 /* We get these for things like a[0] where a is a vector type. In these
697 * cases we want variable_referenced() to return the actual vector
698 * variable this is wrapping.
699 */
700 return operands[0]->variable_referenced();
701 default:
702 return ir_rvalue::variable_referenced();
703 }
704 }
705
ir_constant()706 ir_constant::ir_constant()
707 : ir_rvalue(ir_type_constant)
708 {
709 this->const_elements = NULL;
710 }
711
ir_constant(const struct glsl_type * type,const ir_constant_data * data)712 ir_constant::ir_constant(const struct glsl_type *type,
713 const ir_constant_data *data)
714 : ir_rvalue(ir_type_constant)
715 {
716 this->const_elements = NULL;
717
718 assert((type->base_type >= GLSL_TYPE_UINT)
719 && (type->base_type <= GLSL_TYPE_IMAGE));
720
721 this->type = type;
722 memcpy(& this->value, data, sizeof(this->value));
723 }
724
ir_constant(float16_t f16,unsigned vector_elements)725 ir_constant::ir_constant(float16_t f16, unsigned vector_elements)
726 : ir_rvalue(ir_type_constant)
727 {
728 this->const_elements = NULL;
729 assert(vector_elements <= 4);
730 this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT16, vector_elements, 1);
731 for (unsigned i = 0; i < vector_elements; i++) {
732 this->value.f16[i] = f16.bits;
733 }
734 for (unsigned i = vector_elements; i < 16; i++) {
735 this->value.f[i] = 0;
736 }
737 }
738
ir_constant(float f,unsigned vector_elements)739 ir_constant::ir_constant(float f, unsigned vector_elements)
740 : ir_rvalue(ir_type_constant)
741 {
742 this->const_elements = NULL;
743 assert(vector_elements <= 4);
744 this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT, vector_elements, 1);
745 for (unsigned i = 0; i < vector_elements; i++) {
746 this->value.f[i] = f;
747 }
748 for (unsigned i = vector_elements; i < 16; i++) {
749 this->value.f[i] = 0;
750 }
751 }
752
ir_constant(double d,unsigned vector_elements)753 ir_constant::ir_constant(double d, unsigned vector_elements)
754 : ir_rvalue(ir_type_constant)
755 {
756 this->const_elements = NULL;
757 assert(vector_elements <= 4);
758 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE, vector_elements, 1);
759 for (unsigned i = 0; i < vector_elements; i++) {
760 this->value.d[i] = d;
761 }
762 for (unsigned i = vector_elements; i < 16; i++) {
763 this->value.d[i] = 0.0;
764 }
765 }
766
ir_constant(int16_t i16,unsigned vector_elements)767 ir_constant::ir_constant(int16_t i16, unsigned vector_elements)
768 : ir_rvalue(ir_type_constant)
769 {
770 this->const_elements = NULL;
771 assert(vector_elements <= 4);
772 this->type = glsl_type::get_instance(GLSL_TYPE_INT16, vector_elements, 1);
773 for (unsigned i = 0; i < vector_elements; i++) {
774 this->value.i16[i] = i16;
775 }
776 for (unsigned i = vector_elements; i < 16; i++) {
777 this->value.i16[i] = 0;
778 }
779 }
780
ir_constant(uint16_t u16,unsigned vector_elements)781 ir_constant::ir_constant(uint16_t u16, unsigned vector_elements)
782 : ir_rvalue(ir_type_constant)
783 {
784 this->const_elements = NULL;
785 assert(vector_elements <= 4);
786 this->type = glsl_type::get_instance(GLSL_TYPE_UINT16, vector_elements, 1);
787 for (unsigned i = 0; i < vector_elements; i++) {
788 this->value.u16[i] = u16;
789 }
790 for (unsigned i = vector_elements; i < 16; i++) {
791 this->value.u16[i] = 0;
792 }
793 }
794
ir_constant(unsigned int u,unsigned vector_elements)795 ir_constant::ir_constant(unsigned int u, unsigned vector_elements)
796 : ir_rvalue(ir_type_constant)
797 {
798 this->const_elements = NULL;
799 assert(vector_elements <= 4);
800 this->type = glsl_type::get_instance(GLSL_TYPE_UINT, vector_elements, 1);
801 for (unsigned i = 0; i < vector_elements; i++) {
802 this->value.u[i] = u;
803 }
804 for (unsigned i = vector_elements; i < 16; i++) {
805 this->value.u[i] = 0;
806 }
807 }
808
ir_constant(int integer,unsigned vector_elements)809 ir_constant::ir_constant(int integer, unsigned vector_elements)
810 : ir_rvalue(ir_type_constant)
811 {
812 this->const_elements = NULL;
813 assert(vector_elements <= 4);
814 this->type = glsl_type::get_instance(GLSL_TYPE_INT, vector_elements, 1);
815 for (unsigned i = 0; i < vector_elements; i++) {
816 this->value.i[i] = integer;
817 }
818 for (unsigned i = vector_elements; i < 16; i++) {
819 this->value.i[i] = 0;
820 }
821 }
822
ir_constant(uint64_t u64,unsigned vector_elements)823 ir_constant::ir_constant(uint64_t u64, unsigned vector_elements)
824 : ir_rvalue(ir_type_constant)
825 {
826 this->const_elements = NULL;
827 assert(vector_elements <= 4);
828 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64, vector_elements, 1);
829 for (unsigned i = 0; i < vector_elements; i++) {
830 this->value.u64[i] = u64;
831 }
832 for (unsigned i = vector_elements; i < 16; i++) {
833 this->value.u64[i] = 0;
834 }
835 }
836
ir_constant(int64_t int64,unsigned vector_elements)837 ir_constant::ir_constant(int64_t int64, unsigned vector_elements)
838 : ir_rvalue(ir_type_constant)
839 {
840 this->const_elements = NULL;
841 assert(vector_elements <= 4);
842 this->type = glsl_type::get_instance(GLSL_TYPE_INT64, vector_elements, 1);
843 for (unsigned i = 0; i < vector_elements; i++) {
844 this->value.i64[i] = int64;
845 }
846 for (unsigned i = vector_elements; i < 16; i++) {
847 this->value.i64[i] = 0;
848 }
849 }
850
ir_constant(bool b,unsigned vector_elements)851 ir_constant::ir_constant(bool b, unsigned vector_elements)
852 : ir_rvalue(ir_type_constant)
853 {
854 this->const_elements = NULL;
855 assert(vector_elements <= 4);
856 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL, vector_elements, 1);
857 for (unsigned i = 0; i < vector_elements; i++) {
858 this->value.b[i] = b;
859 }
860 for (unsigned i = vector_elements; i < 16; i++) {
861 this->value.b[i] = false;
862 }
863 }
864
ir_constant(const ir_constant * c,unsigned i)865 ir_constant::ir_constant(const ir_constant *c, unsigned i)
866 : ir_rvalue(ir_type_constant)
867 {
868 this->const_elements = NULL;
869 this->type = c->type->get_base_type();
870
871 /* Section 5.11 (Out-of-Bounds Accesses) of the GLSL 4.60 spec says:
872 *
873 * In the subsections described above for array, vector, matrix and
874 * structure accesses, any out-of-bounds access produced undefined
875 * behavior....Out-of-bounds reads return undefined values, which
876 * include values from other variables of the active program or zero.
877 *
878 * GL_KHR_robustness and GL_ARB_robustness encourage us to return zero.
879 */
880 if (i >= c->type->vector_elements) {
881 this->value = { { 0 } };
882 return;
883 }
884
885 switch (this->type->base_type) {
886 case GLSL_TYPE_UINT16: this->value.u16[0] = c->value.u16[i]; break;
887 case GLSL_TYPE_INT16: this->value.i16[0] = c->value.i16[i]; break;
888 case GLSL_TYPE_UINT: this->value.u[0] = c->value.u[i]; break;
889 case GLSL_TYPE_INT: this->value.i[0] = c->value.i[i]; break;
890 case GLSL_TYPE_FLOAT: this->value.f[0] = c->value.f[i]; break;
891 case GLSL_TYPE_FLOAT16: this->value.f16[0] = c->value.f16[i]; break;
892 case GLSL_TYPE_BOOL: this->value.b[0] = c->value.b[i]; break;
893 case GLSL_TYPE_DOUBLE: this->value.d[0] = c->value.d[i]; break;
894 default: assert(!"Should not get here."); break;
895 }
896 }
897
ir_constant(const struct glsl_type * type,exec_list * value_list)898 ir_constant::ir_constant(const struct glsl_type *type, exec_list *value_list)
899 : ir_rvalue(ir_type_constant)
900 {
901 this->const_elements = NULL;
902 this->type = type;
903
904 assert(type->is_scalar() || type->is_vector() || type->is_matrix()
905 || type->is_struct() || type->is_array());
906
907 /* If the constant is a record, the types of each of the entries in
908 * value_list must be a 1-for-1 match with the structure components. Each
909 * entry must also be a constant. Just move the nodes from the value_list
910 * to the list in the ir_constant.
911 */
912 if (type->is_array() || type->is_struct()) {
913 this->const_elements = ralloc_array(this, ir_constant *, type->length);
914 unsigned i = 0;
915 foreach_in_list(ir_constant, value, value_list) {
916 assert(value->as_constant() != NULL);
917
918 this->const_elements[i++] = value;
919 }
920 return;
921 }
922
923 for (unsigned i = 0; i < 16; i++) {
924 this->value.u[i] = 0;
925 }
926
927 ir_constant *value = (ir_constant *) (value_list->get_head_raw());
928
929 /* Constructors with exactly one scalar argument are special for vectors
930 * and matrices. For vectors, the scalar value is replicated to fill all
931 * the components. For matrices, the scalar fills the components of the
932 * diagonal while the rest is filled with 0.
933 */
934 if (value->type->is_scalar() && value->next->is_tail_sentinel()) {
935 if (type->is_matrix()) {
936 /* Matrix - fill diagonal (rest is already set to 0) */
937 for (unsigned i = 0; i < type->matrix_columns; i++) {
938 switch (type->base_type) {
939 case GLSL_TYPE_FLOAT:
940 this->value.f[i * type->vector_elements + i] =
941 value->value.f[0];
942 break;
943 case GLSL_TYPE_DOUBLE:
944 this->value.d[i * type->vector_elements + i] =
945 value->value.d[0];
946 break;
947 case GLSL_TYPE_FLOAT16:
948 this->value.f16[i * type->vector_elements + i] =
949 value->value.f16[0];
950 break;
951 default:
952 assert(!"unexpected matrix base type");
953 }
954 }
955 } else {
956 /* Vector or scalar - fill all components */
957 switch (type->base_type) {
958 case GLSL_TYPE_UINT16:
959 case GLSL_TYPE_INT16:
960 for (unsigned i = 0; i < type->components(); i++)
961 this->value.u16[i] = value->value.u16[0];
962 break;
963 case GLSL_TYPE_UINT:
964 case GLSL_TYPE_INT:
965 for (unsigned i = 0; i < type->components(); i++)
966 this->value.u[i] = value->value.u[0];
967 break;
968 case GLSL_TYPE_FLOAT:
969 for (unsigned i = 0; i < type->components(); i++)
970 this->value.f[i] = value->value.f[0];
971 break;
972 case GLSL_TYPE_FLOAT16:
973 for (unsigned i = 0; i < type->components(); i++)
974 this->value.f16[i] = value->value.f16[0];
975 break;
976 case GLSL_TYPE_DOUBLE:
977 for (unsigned i = 0; i < type->components(); i++)
978 this->value.d[i] = value->value.d[0];
979 break;
980 case GLSL_TYPE_UINT64:
981 case GLSL_TYPE_INT64:
982 for (unsigned i = 0; i < type->components(); i++)
983 this->value.u64[i] = value->value.u64[0];
984 break;
985 case GLSL_TYPE_BOOL:
986 for (unsigned i = 0; i < type->components(); i++)
987 this->value.b[i] = value->value.b[0];
988 break;
989 case GLSL_TYPE_SAMPLER:
990 case GLSL_TYPE_IMAGE:
991 this->value.u64[0] = value->value.u64[0];
992 break;
993 default:
994 assert(!"Should not get here.");
995 break;
996 }
997 }
998 return;
999 }
1000
1001 if (type->is_matrix() && value->type->is_matrix()) {
1002 assert(value->next->is_tail_sentinel());
1003
1004 /* From section 5.4.2 of the GLSL 1.20 spec:
1005 * "If a matrix is constructed from a matrix, then each component
1006 * (column i, row j) in the result that has a corresponding component
1007 * (column i, row j) in the argument will be initialized from there."
1008 */
1009 unsigned cols = MIN2(type->matrix_columns, value->type->matrix_columns);
1010 unsigned rows = MIN2(type->vector_elements, value->type->vector_elements);
1011 for (unsigned i = 0; i < cols; i++) {
1012 for (unsigned j = 0; j < rows; j++) {
1013 const unsigned src = i * value->type->vector_elements + j;
1014 const unsigned dst = i * type->vector_elements + j;
1015 this->value.f[dst] = value->value.f[src];
1016 }
1017 }
1018
1019 /* "All other components will be initialized to the identity matrix." */
1020 for (unsigned i = cols; i < type->matrix_columns; i++)
1021 this->value.f[i * type->vector_elements + i] = 1.0;
1022
1023 return;
1024 }
1025
1026 /* Use each component from each entry in the value_list to initialize one
1027 * component of the constant being constructed.
1028 */
1029 unsigned i = 0;
1030 for (;;) {
1031 assert(value->as_constant() != NULL);
1032 assert(!value->is_tail_sentinel());
1033
1034 for (unsigned j = 0; j < value->type->components(); j++) {
1035 switch (type->base_type) {
1036 case GLSL_TYPE_UINT16:
1037 this->value.u16[i] = value->get_uint16_component(j);
1038 break;
1039 case GLSL_TYPE_INT16:
1040 this->value.i16[i] = value->get_int16_component(j);
1041 break;
1042 case GLSL_TYPE_UINT:
1043 this->value.u[i] = value->get_uint_component(j);
1044 break;
1045 case GLSL_TYPE_INT:
1046 this->value.i[i] = value->get_int_component(j);
1047 break;
1048 case GLSL_TYPE_FLOAT:
1049 this->value.f[i] = value->get_float_component(j);
1050 break;
1051 case GLSL_TYPE_FLOAT16:
1052 this->value.f16[i] = value->get_float16_component(j);
1053 break;
1054 case GLSL_TYPE_BOOL:
1055 this->value.b[i] = value->get_bool_component(j);
1056 break;
1057 case GLSL_TYPE_DOUBLE:
1058 this->value.d[i] = value->get_double_component(j);
1059 break;
1060 case GLSL_TYPE_UINT64:
1061 this->value.u64[i] = value->get_uint64_component(j);
1062 break;
1063 case GLSL_TYPE_INT64:
1064 this->value.i64[i] = value->get_int64_component(j);
1065 break;
1066 default:
1067 /* FINISHME: What to do? Exceptions are not the answer.
1068 */
1069 break;
1070 }
1071
1072 i++;
1073 if (i >= type->components())
1074 break;
1075 }
1076
1077 if (i >= type->components())
1078 break; /* avoid downcasting a list sentinel */
1079 value = (ir_constant *) value->next;
1080 }
1081 }
1082
1083 ir_constant *
zero(void * mem_ctx,const glsl_type * type)1084 ir_constant::zero(void *mem_ctx, const glsl_type *type)
1085 {
1086 assert(type->is_scalar() || type->is_vector() || type->is_matrix()
1087 || type->is_struct() || type->is_array());
1088
1089 ir_constant *c = new(mem_ctx) ir_constant;
1090 c->type = type;
1091 memset(&c->value, 0, sizeof(c->value));
1092
1093 if (type->is_array()) {
1094 c->const_elements = ralloc_array(c, ir_constant *, type->length);
1095
1096 for (unsigned i = 0; i < type->length; i++)
1097 c->const_elements[i] = ir_constant::zero(c, type->fields.array);
1098 }
1099
1100 if (type->is_struct()) {
1101 c->const_elements = ralloc_array(c, ir_constant *, type->length);
1102
1103 for (unsigned i = 0; i < type->length; i++) {
1104 c->const_elements[i] =
1105 ir_constant::zero(mem_ctx, type->fields.structure[i].type);
1106 }
1107 }
1108
1109 return c;
1110 }
1111
1112 bool
get_bool_component(unsigned i) const1113 ir_constant::get_bool_component(unsigned i) const
1114 {
1115 switch (this->type->base_type) {
1116 case GLSL_TYPE_UINT16:return this->value.u16[i] != 0;
1117 case GLSL_TYPE_INT16: return this->value.i16[i] != 0;
1118 case GLSL_TYPE_UINT: return this->value.u[i] != 0;
1119 case GLSL_TYPE_INT: return this->value.i[i] != 0;
1120 case GLSL_TYPE_FLOAT: return ((int)this->value.f[i]) != 0;
1121 case GLSL_TYPE_FLOAT16: return ((int)_mesa_half_to_float(this->value.f16[i])) != 0;
1122 case GLSL_TYPE_BOOL: return this->value.b[i];
1123 case GLSL_TYPE_DOUBLE: return this->value.d[i] != 0.0;
1124 case GLSL_TYPE_SAMPLER:
1125 case GLSL_TYPE_IMAGE:
1126 case GLSL_TYPE_UINT64: return this->value.u64[i] != 0;
1127 case GLSL_TYPE_INT64: return this->value.i64[i] != 0;
1128 default: assert(!"Should not get here."); break;
1129 }
1130
1131 /* Must return something to make the compiler happy. This is clearly an
1132 * error case.
1133 */
1134 return false;
1135 }
1136
1137 float
get_float_component(unsigned i) const1138 ir_constant::get_float_component(unsigned i) const
1139 {
1140 switch (this->type->base_type) {
1141 case GLSL_TYPE_UINT16:return (float) this->value.u16[i];
1142 case GLSL_TYPE_INT16: return (float) this->value.i16[i];
1143 case GLSL_TYPE_UINT: return (float) this->value.u[i];
1144 case GLSL_TYPE_INT: return (float) this->value.i[i];
1145 case GLSL_TYPE_FLOAT: return this->value.f[i];
1146 case GLSL_TYPE_FLOAT16: return _mesa_half_to_float(this->value.f16[i]);
1147 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1.0f : 0.0f;
1148 case GLSL_TYPE_DOUBLE: return (float) this->value.d[i];
1149 case GLSL_TYPE_SAMPLER:
1150 case GLSL_TYPE_IMAGE:
1151 case GLSL_TYPE_UINT64: return (float) this->value.u64[i];
1152 case GLSL_TYPE_INT64: return (float) this->value.i64[i];
1153 default: assert(!"Should not get here."); break;
1154 }
1155
1156 /* Must return something to make the compiler happy. This is clearly an
1157 * error case.
1158 */
1159 return 0.0;
1160 }
1161
1162 uint16_t
get_float16_component(unsigned i) const1163 ir_constant::get_float16_component(unsigned i) const
1164 {
1165 if (this->type->base_type == GLSL_TYPE_FLOAT16)
1166 return this->value.f16[i];
1167 else
1168 return _mesa_float_to_half(get_float_component(i));
1169 }
1170
1171 double
get_double_component(unsigned i) const1172 ir_constant::get_double_component(unsigned i) const
1173 {
1174 switch (this->type->base_type) {
1175 case GLSL_TYPE_UINT16:return (double) this->value.u16[i];
1176 case GLSL_TYPE_INT16: return (double) this->value.i16[i];
1177 case GLSL_TYPE_UINT: return (double) this->value.u[i];
1178 case GLSL_TYPE_INT: return (double) this->value.i[i];
1179 case GLSL_TYPE_FLOAT: return (double) this->value.f[i];
1180 case GLSL_TYPE_FLOAT16: return (double) _mesa_half_to_float(this->value.f16[i]);
1181 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1.0 : 0.0;
1182 case GLSL_TYPE_DOUBLE: return this->value.d[i];
1183 case GLSL_TYPE_SAMPLER:
1184 case GLSL_TYPE_IMAGE:
1185 case GLSL_TYPE_UINT64: return (double) this->value.u64[i];
1186 case GLSL_TYPE_INT64: return (double) this->value.i64[i];
1187 default: assert(!"Should not get here."); break;
1188 }
1189
1190 /* Must return something to make the compiler happy. This is clearly an
1191 * error case.
1192 */
1193 return 0.0;
1194 }
1195
1196 int16_t
get_int16_component(unsigned i) const1197 ir_constant::get_int16_component(unsigned i) const
1198 {
1199 switch (this->type->base_type) {
1200 case GLSL_TYPE_UINT16:return this->value.u16[i];
1201 case GLSL_TYPE_INT16: return this->value.i16[i];
1202 case GLSL_TYPE_UINT: return this->value.u[i];
1203 case GLSL_TYPE_INT: return this->value.i[i];
1204 case GLSL_TYPE_FLOAT: return (int16_t) this->value.f[i];
1205 case GLSL_TYPE_FLOAT16: return (int16_t) _mesa_half_to_float(this->value.f16[i]);
1206 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1207 case GLSL_TYPE_DOUBLE: return (int16_t) this->value.d[i];
1208 case GLSL_TYPE_SAMPLER:
1209 case GLSL_TYPE_IMAGE:
1210 case GLSL_TYPE_UINT64: return (int16_t) this->value.u64[i];
1211 case GLSL_TYPE_INT64: return (int16_t) this->value.i64[i];
1212 default: assert(!"Should not get here."); break;
1213 }
1214
1215 /* Must return something to make the compiler happy. This is clearly an
1216 * error case.
1217 */
1218 return 0;
1219 }
1220
1221 uint16_t
get_uint16_component(unsigned i) const1222 ir_constant::get_uint16_component(unsigned i) const
1223 {
1224 switch (this->type->base_type) {
1225 case GLSL_TYPE_UINT16:return this->value.u16[i];
1226 case GLSL_TYPE_INT16: return this->value.i16[i];
1227 case GLSL_TYPE_UINT: return this->value.u[i];
1228 case GLSL_TYPE_INT: return this->value.i[i];
1229 case GLSL_TYPE_FLOAT: return (uint16_t) this->value.f[i];
1230 case GLSL_TYPE_FLOAT16: return (uint16_t) _mesa_half_to_float(this->value.f16[i]);
1231 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1232 case GLSL_TYPE_DOUBLE: return (uint16_t) this->value.d[i];
1233 case GLSL_TYPE_SAMPLER:
1234 case GLSL_TYPE_IMAGE:
1235 case GLSL_TYPE_UINT64: return (uint16_t) this->value.u64[i];
1236 case GLSL_TYPE_INT64: return (uint16_t) this->value.i64[i];
1237 default: assert(!"Should not get here."); break;
1238 }
1239
1240 /* Must return something to make the compiler happy. This is clearly an
1241 * error case.
1242 */
1243 return 0;
1244 }
1245
1246 int
get_int_component(unsigned i) const1247 ir_constant::get_int_component(unsigned i) const
1248 {
1249 switch (this->type->base_type) {
1250 case GLSL_TYPE_UINT16:return this->value.u16[i];
1251 case GLSL_TYPE_INT16: return this->value.i16[i];
1252 case GLSL_TYPE_UINT: return this->value.u[i];
1253 case GLSL_TYPE_INT: return this->value.i[i];
1254 case GLSL_TYPE_FLOAT: return (int) this->value.f[i];
1255 case GLSL_TYPE_FLOAT16: return (int) _mesa_half_to_float(this->value.f16[i]);
1256 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1257 case GLSL_TYPE_DOUBLE: return (int) this->value.d[i];
1258 case GLSL_TYPE_SAMPLER:
1259 case GLSL_TYPE_IMAGE:
1260 case GLSL_TYPE_UINT64: return (int) this->value.u64[i];
1261 case GLSL_TYPE_INT64: return (int) this->value.i64[i];
1262 default: assert(!"Should not get here."); break;
1263 }
1264
1265 /* Must return something to make the compiler happy. This is clearly an
1266 * error case.
1267 */
1268 return 0;
1269 }
1270
1271 unsigned
get_uint_component(unsigned i) const1272 ir_constant::get_uint_component(unsigned i) const
1273 {
1274 switch (this->type->base_type) {
1275 case GLSL_TYPE_UINT16:return this->value.u16[i];
1276 case GLSL_TYPE_INT16: return this->value.i16[i];
1277 case GLSL_TYPE_UINT: return this->value.u[i];
1278 case GLSL_TYPE_INT: return this->value.i[i];
1279 case GLSL_TYPE_FLOAT: return (unsigned) this->value.f[i];
1280 case GLSL_TYPE_FLOAT16: return (unsigned) _mesa_half_to_float(this->value.f16[i]);
1281 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1282 case GLSL_TYPE_DOUBLE: return (unsigned) this->value.d[i];
1283 case GLSL_TYPE_SAMPLER:
1284 case GLSL_TYPE_IMAGE:
1285 case GLSL_TYPE_UINT64: return (unsigned) this->value.u64[i];
1286 case GLSL_TYPE_INT64: return (unsigned) this->value.i64[i];
1287 default: assert(!"Should not get here."); break;
1288 }
1289
1290 /* Must return something to make the compiler happy. This is clearly an
1291 * error case.
1292 */
1293 return 0;
1294 }
1295
1296 int64_t
get_int64_component(unsigned i) const1297 ir_constant::get_int64_component(unsigned i) const
1298 {
1299 switch (this->type->base_type) {
1300 case GLSL_TYPE_UINT16:return this->value.u16[i];
1301 case GLSL_TYPE_INT16: return this->value.i16[i];
1302 case GLSL_TYPE_UINT: return this->value.u[i];
1303 case GLSL_TYPE_INT: return this->value.i[i];
1304 case GLSL_TYPE_FLOAT: return (int64_t) this->value.f[i];
1305 case GLSL_TYPE_FLOAT16: return (int64_t) _mesa_half_to_float(this->value.f16[i]);
1306 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1307 case GLSL_TYPE_DOUBLE: return (int64_t) this->value.d[i];
1308 case GLSL_TYPE_SAMPLER:
1309 case GLSL_TYPE_IMAGE:
1310 case GLSL_TYPE_UINT64: return (int64_t) this->value.u64[i];
1311 case GLSL_TYPE_INT64: return this->value.i64[i];
1312 default: assert(!"Should not get here."); break;
1313 }
1314
1315 /* Must return something to make the compiler happy. This is clearly an
1316 * error case.
1317 */
1318 return 0;
1319 }
1320
1321 uint64_t
get_uint64_component(unsigned i) const1322 ir_constant::get_uint64_component(unsigned i) const
1323 {
1324 switch (this->type->base_type) {
1325 case GLSL_TYPE_UINT16:return this->value.u16[i];
1326 case GLSL_TYPE_INT16: return this->value.i16[i];
1327 case GLSL_TYPE_UINT: return this->value.u[i];
1328 case GLSL_TYPE_INT: return this->value.i[i];
1329 case GLSL_TYPE_FLOAT: return (uint64_t) this->value.f[i];
1330 case GLSL_TYPE_FLOAT16: return (uint64_t) _mesa_half_to_float(this->value.f16[i]);
1331 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1332 case GLSL_TYPE_DOUBLE: return (uint64_t) this->value.d[i];
1333 case GLSL_TYPE_SAMPLER:
1334 case GLSL_TYPE_IMAGE:
1335 case GLSL_TYPE_UINT64: return this->value.u64[i];
1336 case GLSL_TYPE_INT64: return (uint64_t) this->value.i64[i];
1337 default: assert(!"Should not get here."); break;
1338 }
1339
1340 /* Must return something to make the compiler happy. This is clearly an
1341 * error case.
1342 */
1343 return 0;
1344 }
1345
1346 ir_constant *
get_array_element(unsigned i) const1347 ir_constant::get_array_element(unsigned i) const
1348 {
1349 assert(this->type->is_array());
1350
1351 /* From page 35 (page 41 of the PDF) of the GLSL 1.20 spec:
1352 *
1353 * "Behavior is undefined if a shader subscripts an array with an index
1354 * less than 0 or greater than or equal to the size the array was
1355 * declared with."
1356 *
1357 * Most out-of-bounds accesses are removed before things could get this far.
1358 * There are cases where non-constant array index values can get constant
1359 * folded.
1360 */
1361 if (int(i) < 0)
1362 i = 0;
1363 else if (i >= this->type->length)
1364 i = this->type->length - 1;
1365
1366 return const_elements[i];
1367 }
1368
1369 ir_constant *
get_record_field(int idx)1370 ir_constant::get_record_field(int idx)
1371 {
1372 assert(this->type->is_struct());
1373 assert(idx >= 0 && (unsigned) idx < this->type->length);
1374
1375 return const_elements[idx];
1376 }
1377
1378 void
copy_offset(ir_constant * src,int offset)1379 ir_constant::copy_offset(ir_constant *src, int offset)
1380 {
1381 switch (this->type->base_type) {
1382 case GLSL_TYPE_UINT16:
1383 case GLSL_TYPE_INT16:
1384 case GLSL_TYPE_UINT:
1385 case GLSL_TYPE_INT:
1386 case GLSL_TYPE_FLOAT:
1387 case GLSL_TYPE_FLOAT16:
1388 case GLSL_TYPE_DOUBLE:
1389 case GLSL_TYPE_SAMPLER:
1390 case GLSL_TYPE_IMAGE:
1391 case GLSL_TYPE_UINT64:
1392 case GLSL_TYPE_INT64:
1393 case GLSL_TYPE_BOOL: {
1394 unsigned int size = src->type->components();
1395 assert (size <= this->type->components() - offset);
1396 for (unsigned int i=0; i<size; i++) {
1397 switch (this->type->base_type) {
1398 case GLSL_TYPE_UINT16:
1399 value.u16[i+offset] = src->get_uint16_component(i);
1400 break;
1401 case GLSL_TYPE_INT16:
1402 value.i16[i+offset] = src->get_int16_component(i);
1403 break;
1404 case GLSL_TYPE_UINT:
1405 value.u[i+offset] = src->get_uint_component(i);
1406 break;
1407 case GLSL_TYPE_INT:
1408 value.i[i+offset] = src->get_int_component(i);
1409 break;
1410 case GLSL_TYPE_FLOAT:
1411 value.f[i+offset] = src->get_float_component(i);
1412 break;
1413 case GLSL_TYPE_FLOAT16:
1414 value.f16[i+offset] = src->get_float16_component(i);
1415 break;
1416 case GLSL_TYPE_BOOL:
1417 value.b[i+offset] = src->get_bool_component(i);
1418 break;
1419 case GLSL_TYPE_DOUBLE:
1420 value.d[i+offset] = src->get_double_component(i);
1421 break;
1422 case GLSL_TYPE_SAMPLER:
1423 case GLSL_TYPE_IMAGE:
1424 case GLSL_TYPE_UINT64:
1425 value.u64[i+offset] = src->get_uint64_component(i);
1426 break;
1427 case GLSL_TYPE_INT64:
1428 value.i64[i+offset] = src->get_int64_component(i);
1429 break;
1430 default: // Shut up the compiler
1431 break;
1432 }
1433 }
1434 break;
1435 }
1436
1437 case GLSL_TYPE_STRUCT:
1438 case GLSL_TYPE_ARRAY: {
1439 assert (src->type == this->type);
1440 for (unsigned i = 0; i < this->type->length; i++) {
1441 this->const_elements[i] = src->const_elements[i]->clone(this, NULL);
1442 }
1443 break;
1444 }
1445
1446 default:
1447 assert(!"Should not get here.");
1448 break;
1449 }
1450 }
1451
1452 void
copy_masked_offset(ir_constant * src,int offset,unsigned int mask)1453 ir_constant::copy_masked_offset(ir_constant *src, int offset, unsigned int mask)
1454 {
1455 assert (!type->is_array() && !type->is_struct());
1456
1457 if (!type->is_vector() && !type->is_matrix()) {
1458 offset = 0;
1459 mask = 1;
1460 }
1461
1462 int id = 0;
1463 for (int i=0; i<4; i++) {
1464 if (mask & (1 << i)) {
1465 switch (this->type->base_type) {
1466 case GLSL_TYPE_UINT16:
1467 value.u16[i+offset] = src->get_uint16_component(id++);
1468 break;
1469 case GLSL_TYPE_INT16:
1470 value.i16[i+offset] = src->get_int16_component(id++);
1471 break;
1472 case GLSL_TYPE_UINT:
1473 value.u[i+offset] = src->get_uint_component(id++);
1474 break;
1475 case GLSL_TYPE_INT:
1476 value.i[i+offset] = src->get_int_component(id++);
1477 break;
1478 case GLSL_TYPE_FLOAT:
1479 value.f[i+offset] = src->get_float_component(id++);
1480 break;
1481 case GLSL_TYPE_FLOAT16:
1482 value.f16[i+offset] = src->get_float16_component(id++);
1483 break;
1484 case GLSL_TYPE_BOOL:
1485 value.b[i+offset] = src->get_bool_component(id++);
1486 break;
1487 case GLSL_TYPE_DOUBLE:
1488 value.d[i+offset] = src->get_double_component(id++);
1489 break;
1490 case GLSL_TYPE_SAMPLER:
1491 case GLSL_TYPE_IMAGE:
1492 case GLSL_TYPE_UINT64:
1493 value.u64[i+offset] = src->get_uint64_component(id++);
1494 break;
1495 case GLSL_TYPE_INT64:
1496 value.i64[i+offset] = src->get_int64_component(id++);
1497 break;
1498 default:
1499 assert(!"Should not get here.");
1500 return;
1501 }
1502 }
1503 }
1504 }
1505
1506 bool
has_value(const ir_constant * c) const1507 ir_constant::has_value(const ir_constant *c) const
1508 {
1509 if (this->type != c->type)
1510 return false;
1511
1512 if (this->type->is_array() || this->type->is_struct()) {
1513 for (unsigned i = 0; i < this->type->length; i++) {
1514 if (!this->const_elements[i]->has_value(c->const_elements[i]))
1515 return false;
1516 }
1517 return true;
1518 }
1519
1520 for (unsigned i = 0; i < this->type->components(); i++) {
1521 switch (this->type->base_type) {
1522 case GLSL_TYPE_UINT16:
1523 if (this->value.u16[i] != c->value.u16[i])
1524 return false;
1525 break;
1526 case GLSL_TYPE_INT16:
1527 if (this->value.i16[i] != c->value.i16[i])
1528 return false;
1529 break;
1530 case GLSL_TYPE_UINT:
1531 if (this->value.u[i] != c->value.u[i])
1532 return false;
1533 break;
1534 case GLSL_TYPE_INT:
1535 if (this->value.i[i] != c->value.i[i])
1536 return false;
1537 break;
1538 case GLSL_TYPE_FLOAT:
1539 if (this->value.f[i] != c->value.f[i])
1540 return false;
1541 break;
1542 case GLSL_TYPE_FLOAT16:
1543 /* Convert to float to make sure NaN and ±0.0 compares correctly */
1544 if (_mesa_half_to_float(this->value.f16[i]) !=
1545 _mesa_half_to_float(c->value.f16[i]))
1546 return false;
1547 break;
1548 case GLSL_TYPE_BOOL:
1549 if (this->value.b[i] != c->value.b[i])
1550 return false;
1551 break;
1552 case GLSL_TYPE_DOUBLE:
1553 if (this->value.d[i] != c->value.d[i])
1554 return false;
1555 break;
1556 case GLSL_TYPE_SAMPLER:
1557 case GLSL_TYPE_IMAGE:
1558 case GLSL_TYPE_UINT64:
1559 if (this->value.u64[i] != c->value.u64[i])
1560 return false;
1561 break;
1562 case GLSL_TYPE_INT64:
1563 if (this->value.i64[i] != c->value.i64[i])
1564 return false;
1565 break;
1566 default:
1567 assert(!"Should not get here.");
1568 return false;
1569 }
1570 }
1571
1572 return true;
1573 }
1574
1575 bool
is_value(float f,int i) const1576 ir_constant::is_value(float f, int i) const
1577 {
1578 if (!this->type->is_scalar() && !this->type->is_vector())
1579 return false;
1580
1581 /* Only accept boolean values for 0/1. */
1582 if (int(bool(i)) != i && this->type->is_boolean())
1583 return false;
1584
1585 for (unsigned c = 0; c < this->type->vector_elements; c++) {
1586 switch (this->type->base_type) {
1587 case GLSL_TYPE_FLOAT:
1588 if (this->value.f[c] != f)
1589 return false;
1590 break;
1591 case GLSL_TYPE_FLOAT16:
1592 if (_mesa_half_to_float(this->value.f16[c]) != f)
1593 return false;
1594 break;
1595 case GLSL_TYPE_INT16:
1596 if (this->value.i16[c] != int16_t(i))
1597 return false;
1598 break;
1599 case GLSL_TYPE_UINT16:
1600 if (this->value.u16[c] != uint16_t(i))
1601 return false;
1602 break;
1603 case GLSL_TYPE_INT:
1604 if (this->value.i[c] != i)
1605 return false;
1606 break;
1607 case GLSL_TYPE_UINT:
1608 if (this->value.u[c] != unsigned(i))
1609 return false;
1610 break;
1611 case GLSL_TYPE_BOOL:
1612 if (this->value.b[c] != bool(i))
1613 return false;
1614 break;
1615 case GLSL_TYPE_DOUBLE:
1616 if (this->value.d[c] != double(f))
1617 return false;
1618 break;
1619 case GLSL_TYPE_SAMPLER:
1620 case GLSL_TYPE_IMAGE:
1621 case GLSL_TYPE_UINT64:
1622 if (this->value.u64[c] != uint64_t(i))
1623 return false;
1624 break;
1625 case GLSL_TYPE_INT64:
1626 if (this->value.i64[c] != i)
1627 return false;
1628 break;
1629 default:
1630 /* The only other base types are structures, arrays, and samplers.
1631 * Samplers cannot be constants, and the others should have been
1632 * filtered out above.
1633 */
1634 assert(!"Should not get here.");
1635 return false;
1636 }
1637 }
1638
1639 return true;
1640 }
1641
1642 bool
is_zero() const1643 ir_constant::is_zero() const
1644 {
1645 return is_value(0.0, 0);
1646 }
1647
1648 bool
is_one() const1649 ir_constant::is_one() const
1650 {
1651 return is_value(1.0, 1);
1652 }
1653
1654 bool
is_negative_one() const1655 ir_constant::is_negative_one() const
1656 {
1657 return is_value(-1.0, -1);
1658 }
1659
1660 bool
is_uint16_constant() const1661 ir_constant::is_uint16_constant() const
1662 {
1663 if (!type->is_integer_32())
1664 return false;
1665
1666 return value.u[0] < (1 << 16);
1667 }
1668
ir_loop()1669 ir_loop::ir_loop()
1670 : ir_instruction(ir_type_loop)
1671 {
1672 }
1673
1674
ir_dereference_variable(ir_variable * var)1675 ir_dereference_variable::ir_dereference_variable(ir_variable *var)
1676 : ir_dereference(ir_type_dereference_variable)
1677 {
1678 assert(var != NULL);
1679
1680 this->var = var;
1681 this->type = var->type;
1682 }
1683
1684
ir_dereference_array(ir_rvalue * value,ir_rvalue * array_index)1685 ir_dereference_array::ir_dereference_array(ir_rvalue *value,
1686 ir_rvalue *array_index)
1687 : ir_dereference(ir_type_dereference_array)
1688 {
1689 this->array_index = array_index;
1690 this->set_array(value);
1691 }
1692
1693
ir_dereference_array(ir_variable * var,ir_rvalue * array_index)1694 ir_dereference_array::ir_dereference_array(ir_variable *var,
1695 ir_rvalue *array_index)
1696 : ir_dereference(ir_type_dereference_array)
1697 {
1698 void *ctx = ralloc_parent(var);
1699
1700 this->array_index = array_index;
1701 this->set_array(new(ctx) ir_dereference_variable(var));
1702 }
1703
1704
1705 void
set_array(ir_rvalue * value)1706 ir_dereference_array::set_array(ir_rvalue *value)
1707 {
1708 assert(value != NULL);
1709
1710 this->array = value;
1711
1712 const glsl_type *const vt = this->array->type;
1713
1714 if (vt->is_array()) {
1715 type = vt->fields.array;
1716 } else if (vt->is_matrix()) {
1717 type = vt->column_type();
1718 } else if (vt->is_vector()) {
1719 type = vt->get_base_type();
1720 }
1721 }
1722
1723
ir_dereference_record(ir_rvalue * value,const char * field)1724 ir_dereference_record::ir_dereference_record(ir_rvalue *value,
1725 const char *field)
1726 : ir_dereference(ir_type_dereference_record)
1727 {
1728 assert(value != NULL);
1729
1730 this->record = value;
1731 this->type = this->record->type->field_type(field);
1732 this->field_idx = this->record->type->field_index(field);
1733 }
1734
1735
ir_dereference_record(ir_variable * var,const char * field)1736 ir_dereference_record::ir_dereference_record(ir_variable *var,
1737 const char *field)
1738 : ir_dereference(ir_type_dereference_record)
1739 {
1740 void *ctx = ralloc_parent(var);
1741
1742 this->record = new(ctx) ir_dereference_variable(var);
1743 this->type = this->record->type->field_type(field);
1744 this->field_idx = this->record->type->field_index(field);
1745 }
1746
1747 bool
is_lvalue(const struct _mesa_glsl_parse_state * state) const1748 ir_dereference::is_lvalue(const struct _mesa_glsl_parse_state *state) const
1749 {
1750 ir_variable *var = this->variable_referenced();
1751
1752 /* Every l-value dereference chain eventually ends in a variable.
1753 */
1754 if ((var == NULL) || var->data.read_only)
1755 return false;
1756
1757 /* From section 4.1.7 of the ARB_bindless_texture spec:
1758 *
1759 * "Samplers can be used as l-values, so can be assigned into and used as
1760 * "out" and "inout" function parameters."
1761 *
1762 * From section 4.1.X of the ARB_bindless_texture spec:
1763 *
1764 * "Images can be used as l-values, so can be assigned into and used as
1765 * "out" and "inout" function parameters."
1766 */
1767 if ((!state || state->has_bindless()) &&
1768 (this->type->contains_sampler() || this->type->contains_image()))
1769 return true;
1770
1771 /* From section 4.1.7 of the GLSL 4.40 spec:
1772 *
1773 * "Opaque variables cannot be treated as l-values; hence cannot
1774 * be used as out or inout function parameters, nor can they be
1775 * assigned into."
1776 */
1777 if (this->type->contains_opaque())
1778 return false;
1779
1780 return true;
1781 }
1782
1783
1784 static const char * const tex_opcode_strs[] = { "tex", "txb", "txl", "txd", "txf", "txf_ms", "txs", "lod", "tg4", "query_levels", "texture_samples", "samples_identical" };
1785
opcode_string()1786 const char *ir_texture::opcode_string()
1787 {
1788 assert((unsigned int) op < ARRAY_SIZE(tex_opcode_strs));
1789 return tex_opcode_strs[op];
1790 }
1791
1792 ir_texture_opcode
get_opcode(const char * str)1793 ir_texture::get_opcode(const char *str)
1794 {
1795 const int count = sizeof(tex_opcode_strs) / sizeof(tex_opcode_strs[0]);
1796 for (int op = 0; op < count; op++) {
1797 if (strcmp(str, tex_opcode_strs[op]) == 0)
1798 return (ir_texture_opcode) op;
1799 }
1800 return (ir_texture_opcode) -1;
1801 }
1802
1803
1804 void
set_sampler(ir_dereference * sampler,const glsl_type * type)1805 ir_texture::set_sampler(ir_dereference *sampler, const glsl_type *type)
1806 {
1807 assert(sampler != NULL);
1808 assert(type != NULL);
1809 this->sampler = sampler;
1810 this->type = type;
1811
1812 if (this->op == ir_txs || this->op == ir_query_levels ||
1813 this->op == ir_texture_samples) {
1814 assert(type->base_type == GLSL_TYPE_INT);
1815 } else if (this->op == ir_lod) {
1816 assert(type->vector_elements == 2);
1817 assert(type->is_float());
1818 } else if (this->op == ir_samples_identical) {
1819 assert(type == glsl_type::bool_type);
1820 assert(sampler->type->is_sampler());
1821 assert(sampler->type->sampler_dimensionality == GLSL_SAMPLER_DIM_MS);
1822 } else {
1823 assert(sampler->type->sampled_type == (int) type->base_type);
1824 if (sampler->type->sampler_shadow)
1825 assert(type->vector_elements == 4 || type->vector_elements == 1);
1826 else
1827 assert(type->vector_elements == 4);
1828 }
1829 }
1830
1831
1832 void
init_mask(const unsigned * comp,unsigned count)1833 ir_swizzle::init_mask(const unsigned *comp, unsigned count)
1834 {
1835 assert((count >= 1) && (count <= 4));
1836
1837 memset(&this->mask, 0, sizeof(this->mask));
1838 this->mask.num_components = count;
1839
1840 unsigned dup_mask = 0;
1841 switch (count) {
1842 case 4:
1843 assert(comp[3] <= 3);
1844 dup_mask |= (1U << comp[3])
1845 & ((1U << comp[0]) | (1U << comp[1]) | (1U << comp[2]));
1846 this->mask.w = comp[3];
1847
1848 case 3:
1849 assert(comp[2] <= 3);
1850 dup_mask |= (1U << comp[2])
1851 & ((1U << comp[0]) | (1U << comp[1]));
1852 this->mask.z = comp[2];
1853
1854 case 2:
1855 assert(comp[1] <= 3);
1856 dup_mask |= (1U << comp[1])
1857 & ((1U << comp[0]));
1858 this->mask.y = comp[1];
1859
1860 case 1:
1861 assert(comp[0] <= 3);
1862 this->mask.x = comp[0];
1863 }
1864
1865 this->mask.has_duplicates = dup_mask != 0;
1866
1867 /* Based on the number of elements in the swizzle and the base type
1868 * (i.e., float, int, unsigned, or bool) of the vector being swizzled,
1869 * generate the type of the resulting value.
1870 */
1871 type = glsl_type::get_instance(val->type->base_type, mask.num_components, 1);
1872 }
1873
ir_swizzle(ir_rvalue * val,unsigned x,unsigned y,unsigned z,unsigned w,unsigned count)1874 ir_swizzle::ir_swizzle(ir_rvalue *val, unsigned x, unsigned y, unsigned z,
1875 unsigned w, unsigned count)
1876 : ir_rvalue(ir_type_swizzle), val(val)
1877 {
1878 const unsigned components[4] = { x, y, z, w };
1879 this->init_mask(components, count);
1880 }
1881
ir_swizzle(ir_rvalue * val,const unsigned * comp,unsigned count)1882 ir_swizzle::ir_swizzle(ir_rvalue *val, const unsigned *comp,
1883 unsigned count)
1884 : ir_rvalue(ir_type_swizzle), val(val)
1885 {
1886 this->init_mask(comp, count);
1887 }
1888
ir_swizzle(ir_rvalue * val,ir_swizzle_mask mask)1889 ir_swizzle::ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask)
1890 : ir_rvalue(ir_type_swizzle), val(val), mask(mask)
1891 {
1892 this->type = glsl_type::get_instance(val->type->base_type,
1893 mask.num_components, 1);
1894 }
1895
1896 #define X 1
1897 #define R 5
1898 #define S 9
1899 #define I 13
1900
1901 ir_swizzle *
create(ir_rvalue * val,const char * str,unsigned vector_length)1902 ir_swizzle::create(ir_rvalue *val, const char *str, unsigned vector_length)
1903 {
1904 void *ctx = ralloc_parent(val);
1905
1906 /* For each possible swizzle character, this table encodes the value in
1907 * \c idx_map that represents the 0th element of the vector. For invalid
1908 * swizzle characters (e.g., 'k'), a special value is used that will allow
1909 * detection of errors.
1910 */
1911 static const unsigned char base_idx[26] = {
1912 /* a b c d e f g h i j k l m */
1913 R, R, I, I, I, I, R, I, I, I, I, I, I,
1914 /* n o p q r s t u v w x y z */
1915 I, I, S, S, R, S, S, I, I, X, X, X, X
1916 };
1917
1918 /* Each valid swizzle character has an entry in the previous table. This
1919 * table encodes the base index encoded in the previous table plus the actual
1920 * index of the swizzle character. When processing swizzles, the first
1921 * character in the string is indexed in the previous table. Each character
1922 * in the string is indexed in this table, and the value found there has the
1923 * value form the first table subtracted. The result must be on the range
1924 * [0,3].
1925 *
1926 * For example, the string "wzyx" will get X from the first table. Each of
1927 * the charcaters will get X+3, X+2, X+1, and X+0 from this table. After
1928 * subtraction, the swizzle values are { 3, 2, 1, 0 }.
1929 *
1930 * The string "wzrg" will get X from the first table. Each of the characters
1931 * will get X+3, X+2, R+0, and R+1 from this table. After subtraction, the
1932 * swizzle values are { 3, 2, 4, 5 }. Since 4 and 5 are outside the range
1933 * [0,3], the error is detected.
1934 */
1935 static const unsigned char idx_map[26] = {
1936 /* a b c d e f g h i j k l m */
1937 R+3, R+2, 0, 0, 0, 0, R+1, 0, 0, 0, 0, 0, 0,
1938 /* n o p q r s t u v w x y z */
1939 0, 0, S+2, S+3, R+0, S+0, S+1, 0, 0, X+3, X+0, X+1, X+2
1940 };
1941
1942 int swiz_idx[4] = { 0, 0, 0, 0 };
1943 unsigned i;
1944
1945
1946 /* Validate the first character in the swizzle string and look up the base
1947 * index value as described above.
1948 */
1949 if ((str[0] < 'a') || (str[0] > 'z'))
1950 return NULL;
1951
1952 const unsigned base = base_idx[str[0] - 'a'];
1953
1954
1955 for (i = 0; (i < 4) && (str[i] != '\0'); i++) {
1956 /* Validate the next character, and, as described above, convert it to a
1957 * swizzle index.
1958 */
1959 if ((str[i] < 'a') || (str[i] > 'z'))
1960 return NULL;
1961
1962 swiz_idx[i] = idx_map[str[i] - 'a'] - base;
1963 if ((swiz_idx[i] < 0) || (swiz_idx[i] >= (int) vector_length))
1964 return NULL;
1965 }
1966
1967 if (str[i] != '\0')
1968 return NULL;
1969
1970 return new(ctx) ir_swizzle(val, swiz_idx[0], swiz_idx[1], swiz_idx[2],
1971 swiz_idx[3], i);
1972 }
1973
1974 #undef X
1975 #undef R
1976 #undef S
1977 #undef I
1978
1979 ir_variable *
variable_referenced() const1980 ir_swizzle::variable_referenced() const
1981 {
1982 return this->val->variable_referenced();
1983 }
1984
1985
1986 bool ir_variable::temporaries_allocate_names = false;
1987
1988 const char ir_variable::tmp_name[] = "compiler_temp";
1989
ir_variable(const struct glsl_type * type,const char * name,ir_variable_mode mode)1990 ir_variable::ir_variable(const struct glsl_type *type, const char *name,
1991 ir_variable_mode mode)
1992 : ir_instruction(ir_type_variable)
1993 {
1994 this->type = type;
1995
1996 if (mode == ir_var_temporary && !ir_variable::temporaries_allocate_names)
1997 name = NULL;
1998
1999 /* The ir_variable clone method may call this constructor with name set to
2000 * tmp_name.
2001 */
2002 assert(name != NULL
2003 || mode == ir_var_temporary
2004 || mode == ir_var_function_in
2005 || mode == ir_var_function_out
2006 || mode == ir_var_function_inout);
2007 assert(name != ir_variable::tmp_name
2008 || mode == ir_var_temporary);
2009 if (mode == ir_var_temporary
2010 && (name == NULL || name == ir_variable::tmp_name)) {
2011 this->name = ir_variable::tmp_name;
2012 } else if (name == NULL ||
2013 strlen(name) < ARRAY_SIZE(this->name_storage)) {
2014 strcpy(this->name_storage, name ? name : "");
2015 this->name = this->name_storage;
2016 } else {
2017 this->name = ralloc_strdup(this, name);
2018 }
2019
2020 this->u.max_ifc_array_access = NULL;
2021
2022 this->data.explicit_location = false;
2023 this->data.explicit_index = false;
2024 this->data.explicit_binding = false;
2025 this->data.explicit_component = false;
2026 this->data.has_initializer = false;
2027 this->data.is_implicit_initializer = false;
2028 this->data.is_unmatched_generic_inout = false;
2029 this->data.is_xfb = false;
2030 this->data.is_xfb_only = false;
2031 this->data.explicit_xfb_buffer = false;
2032 this->data.explicit_xfb_offset = false;
2033 this->data.explicit_xfb_stride = false;
2034 this->data.location = -1;
2035 this->data.location_frac = 0;
2036 this->data.matrix_layout = GLSL_MATRIX_LAYOUT_INHERITED;
2037 this->data.from_named_ifc_block = false;
2038 this->data.must_be_shader_input = false;
2039 this->data.index = 0;
2040 this->data.binding = 0;
2041 this->data.warn_extension_index = 0;
2042 this->constant_value = NULL;
2043 this->constant_initializer = NULL;
2044 this->data.depth_layout = ir_depth_layout_none;
2045 this->data.used = false;
2046 this->data.assigned = false;
2047 this->data.always_active_io = false;
2048 this->data.read_only = false;
2049 this->data.centroid = false;
2050 this->data.sample = false;
2051 this->data.patch = false;
2052 this->data.explicit_invariant = false;
2053 this->data.invariant = false;
2054 this->data.precise = false;
2055 this->data.how_declared = ir_var_declared_normally;
2056 this->data.mode = mode;
2057 this->data.interpolation = INTERP_MODE_NONE;
2058 this->data.max_array_access = -1;
2059 this->data.offset = 0;
2060 this->data.precision = GLSL_PRECISION_NONE;
2061 this->data.memory_read_only = false;
2062 this->data.memory_write_only = false;
2063 this->data.memory_coherent = false;
2064 this->data.memory_volatile = false;
2065 this->data.memory_restrict = false;
2066 this->data.from_ssbo_unsized_array = false;
2067 this->data.implicit_sized_array = false;
2068 this->data.fb_fetch_output = false;
2069 this->data.bindless = false;
2070 this->data.bound = false;
2071 this->data.image_format = PIPE_FORMAT_NONE;
2072 this->data._num_state_slots = 0;
2073 this->data.param_index = 0;
2074 this->data.stream = 0;
2075 this->data.xfb_buffer = -1;
2076 this->data.xfb_stride = -1;
2077 this->data.implicit_conversion_prohibited = false;
2078
2079 this->interface_type = NULL;
2080
2081 if (type != NULL) {
2082 if (type->is_interface())
2083 this->init_interface_type(type);
2084 else if (type->without_array()->is_interface())
2085 this->init_interface_type(type->without_array());
2086 }
2087 }
2088
2089
2090 const char *
interpolation_string(unsigned interpolation)2091 interpolation_string(unsigned interpolation)
2092 {
2093 switch (interpolation) {
2094 case INTERP_MODE_NONE: return "no";
2095 case INTERP_MODE_SMOOTH: return "smooth";
2096 case INTERP_MODE_FLAT: return "flat";
2097 case INTERP_MODE_NOPERSPECTIVE: return "noperspective";
2098 }
2099
2100 assert(!"Should not get here.");
2101 return "";
2102 }
2103
2104 const char *const ir_variable::warn_extension_table[] = {
2105 "",
2106 "GL_ARB_shader_stencil_export",
2107 "GL_AMD_shader_stencil_export",
2108 };
2109
2110 void
enable_extension_warning(const char * extension)2111 ir_variable::enable_extension_warning(const char *extension)
2112 {
2113 for (unsigned i = 0; i < ARRAY_SIZE(warn_extension_table); i++) {
2114 if (strcmp(warn_extension_table[i], extension) == 0) {
2115 this->data.warn_extension_index = i;
2116 return;
2117 }
2118 }
2119
2120 assert(!"Should not get here.");
2121 this->data.warn_extension_index = 0;
2122 }
2123
2124 const char *
get_extension_warning() const2125 ir_variable::get_extension_warning() const
2126 {
2127 return this->data.warn_extension_index == 0
2128 ? NULL : warn_extension_table[this->data.warn_extension_index];
2129 }
2130
ir_function_signature(const glsl_type * return_type,builtin_available_predicate b)2131 ir_function_signature::ir_function_signature(const glsl_type *return_type,
2132 builtin_available_predicate b)
2133 : ir_instruction(ir_type_function_signature),
2134 return_type(return_type), is_defined(false),
2135 return_precision(GLSL_PRECISION_NONE),
2136 intrinsic_id(ir_intrinsic_invalid), builtin_avail(b), _function(NULL)
2137 {
2138 this->origin = NULL;
2139 }
2140
2141
2142 bool
is_builtin() const2143 ir_function_signature::is_builtin() const
2144 {
2145 return builtin_avail != NULL;
2146 }
2147
2148
2149 bool
is_builtin_available(const _mesa_glsl_parse_state * state) const2150 ir_function_signature::is_builtin_available(const _mesa_glsl_parse_state *state) const
2151 {
2152 /* We can't call the predicate without a state pointer, so just say that
2153 * the signature is available. At compile time, we need the filtering,
2154 * but also receive a valid state pointer. At link time, we're resolving
2155 * imported built-in prototypes to their definitions, which will always
2156 * be an exact match. So we can skip the filtering.
2157 */
2158 if (state == NULL)
2159 return true;
2160
2161 assert(builtin_avail != NULL);
2162 return builtin_avail(state);
2163 }
2164
2165
2166 static bool
modes_match(unsigned a,unsigned b)2167 modes_match(unsigned a, unsigned b)
2168 {
2169 if (a == b)
2170 return true;
2171
2172 /* Accept "in" vs. "const in" */
2173 if ((a == ir_var_const_in && b == ir_var_function_in) ||
2174 (b == ir_var_const_in && a == ir_var_function_in))
2175 return true;
2176
2177 return false;
2178 }
2179
2180
2181 const char *
qualifiers_match(exec_list * params)2182 ir_function_signature::qualifiers_match(exec_list *params)
2183 {
2184 /* check that the qualifiers match. */
2185 foreach_two_lists(a_node, &this->parameters, b_node, params) {
2186 ir_variable *a = (ir_variable *) a_node;
2187 ir_variable *b = (ir_variable *) b_node;
2188
2189 if (a->data.read_only != b->data.read_only ||
2190 !modes_match(a->data.mode, b->data.mode) ||
2191 a->data.interpolation != b->data.interpolation ||
2192 a->data.centroid != b->data.centroid ||
2193 a->data.sample != b->data.sample ||
2194 a->data.patch != b->data.patch ||
2195 a->data.memory_read_only != b->data.memory_read_only ||
2196 a->data.memory_write_only != b->data.memory_write_only ||
2197 a->data.memory_coherent != b->data.memory_coherent ||
2198 a->data.memory_volatile != b->data.memory_volatile ||
2199 a->data.memory_restrict != b->data.memory_restrict) {
2200
2201 /* parameter a's qualifiers don't match */
2202 return a->name;
2203 }
2204 }
2205 return NULL;
2206 }
2207
2208
2209 void
replace_parameters(exec_list * new_params)2210 ir_function_signature::replace_parameters(exec_list *new_params)
2211 {
2212 /* Destroy all of the previous parameter information. If the previous
2213 * parameter information comes from the function prototype, it may either
2214 * specify incorrect parameter names or not have names at all.
2215 */
2216 new_params->move_nodes_to(¶meters);
2217 }
2218
2219
ir_function(const char * name)2220 ir_function::ir_function(const char *name)
2221 : ir_instruction(ir_type_function)
2222 {
2223 this->subroutine_index = -1;
2224 this->name = ralloc_strdup(this, name);
2225 }
2226
2227
2228 bool
has_user_signature()2229 ir_function::has_user_signature()
2230 {
2231 foreach_in_list(ir_function_signature, sig, &this->signatures) {
2232 if (!sig->is_builtin())
2233 return true;
2234 }
2235 return false;
2236 }
2237
2238
2239 ir_rvalue *
error_value(void * mem_ctx)2240 ir_rvalue::error_value(void *mem_ctx)
2241 {
2242 ir_rvalue *v = new(mem_ctx) ir_rvalue(ir_type_unset);
2243
2244 v->type = glsl_type::error_type;
2245 return v;
2246 }
2247
2248
2249 void
visit_exec_list(exec_list * list,ir_visitor * visitor)2250 visit_exec_list(exec_list *list, ir_visitor *visitor)
2251 {
2252 foreach_in_list_safe(ir_instruction, node, list) {
2253 node->accept(visitor);
2254 }
2255 }
2256
2257
2258 static void
steal_memory(ir_instruction * ir,void * new_ctx)2259 steal_memory(ir_instruction *ir, void *new_ctx)
2260 {
2261 ir_variable *var = ir->as_variable();
2262 ir_function *fn = ir->as_function();
2263 ir_constant *constant = ir->as_constant();
2264 if (var != NULL && var->constant_value != NULL)
2265 steal_memory(var->constant_value, ir);
2266
2267 if (var != NULL && var->constant_initializer != NULL)
2268 steal_memory(var->constant_initializer, ir);
2269
2270 if (fn != NULL && fn->subroutine_types)
2271 ralloc_steal(new_ctx, fn->subroutine_types);
2272
2273 /* The components of aggregate constants are not visited by the normal
2274 * visitor, so steal their values by hand.
2275 */
2276 if (constant != NULL &&
2277 (constant->type->is_array() || constant->type->is_struct())) {
2278 for (unsigned int i = 0; i < constant->type->length; i++) {
2279 steal_memory(constant->const_elements[i], ir);
2280 }
2281 }
2282
2283 ralloc_steal(new_ctx, ir);
2284 }
2285
2286
2287 void
reparent_ir(exec_list * list,void * mem_ctx)2288 reparent_ir(exec_list *list, void *mem_ctx)
2289 {
2290 foreach_in_list(ir_instruction, node, list) {
2291 visit_tree(node, steal_memory, mem_ctx);
2292 }
2293 }
2294
2295
2296 static ir_rvalue *
try_min_one(ir_rvalue * ir)2297 try_min_one(ir_rvalue *ir)
2298 {
2299 ir_expression *expr = ir->as_expression();
2300
2301 if (!expr || expr->operation != ir_binop_min)
2302 return NULL;
2303
2304 if (expr->operands[0]->is_one())
2305 return expr->operands[1];
2306
2307 if (expr->operands[1]->is_one())
2308 return expr->operands[0];
2309
2310 return NULL;
2311 }
2312
2313 static ir_rvalue *
try_max_zero(ir_rvalue * ir)2314 try_max_zero(ir_rvalue *ir)
2315 {
2316 ir_expression *expr = ir->as_expression();
2317
2318 if (!expr || expr->operation != ir_binop_max)
2319 return NULL;
2320
2321 if (expr->operands[0]->is_zero())
2322 return expr->operands[1];
2323
2324 if (expr->operands[1]->is_zero())
2325 return expr->operands[0];
2326
2327 return NULL;
2328 }
2329
2330 ir_rvalue *
as_rvalue_to_saturate()2331 ir_rvalue::as_rvalue_to_saturate()
2332 {
2333 ir_expression *expr = this->as_expression();
2334
2335 if (!expr)
2336 return NULL;
2337
2338 ir_rvalue *max_zero = try_max_zero(expr);
2339 if (max_zero) {
2340 return try_min_one(max_zero);
2341 } else {
2342 ir_rvalue *min_one = try_min_one(expr);
2343 if (min_one) {
2344 return try_max_zero(min_one);
2345 }
2346 }
2347
2348 return NULL;
2349 }
2350
2351
2352 unsigned
vertices_per_prim(GLenum prim)2353 vertices_per_prim(GLenum prim)
2354 {
2355 switch (prim) {
2356 case GL_POINTS:
2357 return 1;
2358 case GL_LINES:
2359 return 2;
2360 case GL_TRIANGLES:
2361 return 3;
2362 case GL_LINES_ADJACENCY:
2363 return 4;
2364 case GL_TRIANGLES_ADJACENCY:
2365 return 6;
2366 default:
2367 assert(!"Bad primitive");
2368 return 3;
2369 }
2370 }
2371
2372 /**
2373 * Generate a string describing the mode of a variable
2374 */
2375 const char *
mode_string(const ir_variable * var)2376 mode_string(const ir_variable *var)
2377 {
2378 switch (var->data.mode) {
2379 case ir_var_auto:
2380 return (var->data.read_only) ? "global constant" : "global variable";
2381
2382 case ir_var_uniform:
2383 return "uniform";
2384
2385 case ir_var_shader_storage:
2386 return "buffer";
2387
2388 case ir_var_shader_in:
2389 return "shader input";
2390
2391 case ir_var_shader_out:
2392 return "shader output";
2393
2394 case ir_var_function_in:
2395 case ir_var_const_in:
2396 return "function input";
2397
2398 case ir_var_function_out:
2399 return "function output";
2400
2401 case ir_var_function_inout:
2402 return "function inout";
2403
2404 case ir_var_system_value:
2405 return "shader input";
2406
2407 case ir_var_temporary:
2408 return "compiler temporary";
2409
2410 case ir_var_mode_count:
2411 break;
2412 }
2413
2414 assert(!"Should not get here.");
2415 return "invalid variable";
2416 }
2417