1 /*
2 * Copyright © 2010 Intel Corporation
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
21 * DEALINGS IN THE SOFTWARE.
22 */
23 #include <string.h>
24 #include "main/core.h" /* for MAX2 */
25 #include "ir.h"
26 #include "compiler/glsl_types.h"
27 #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_saturate:
262 this->type = op0->type;
263 break;
264
265 case ir_unop_f2i:
266 case ir_unop_b2i:
267 case ir_unop_u2i:
268 case ir_unop_d2i:
269 case ir_unop_bitcast_f2i:
270 case ir_unop_bit_count:
271 case ir_unop_find_msb:
272 case ir_unop_find_lsb:
273 case ir_unop_subroutine_to_int:
274 case ir_unop_i642i:
275 case ir_unop_u642i:
276 this->type = glsl_type::get_instance(GLSL_TYPE_INT,
277 op0->type->vector_elements, 1);
278 break;
279
280 case ir_unop_b2f:
281 case ir_unop_i2f:
282 case ir_unop_u2f:
283 case ir_unop_d2f:
284 case ir_unop_bitcast_i2f:
285 case ir_unop_bitcast_u2f:
286 case ir_unop_i642f:
287 case ir_unop_u642f:
288 this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT,
289 op0->type->vector_elements, 1);
290 break;
291
292 case ir_unop_f2b:
293 case ir_unop_i2b:
294 case ir_unop_d2b:
295 case ir_unop_i642b:
296 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL,
297 op0->type->vector_elements, 1);
298 break;
299
300 case ir_unop_f2d:
301 case ir_unop_i2d:
302 case ir_unop_u2d:
303 case ir_unop_i642d:
304 case ir_unop_u642d:
305 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE,
306 op0->type->vector_elements, 1);
307 break;
308
309 case ir_unop_i2u:
310 case ir_unop_f2u:
311 case ir_unop_d2u:
312 case ir_unop_bitcast_f2u:
313 case ir_unop_i642u:
314 case ir_unop_u642u:
315 this->type = glsl_type::get_instance(GLSL_TYPE_UINT,
316 op0->type->vector_elements, 1);
317 break;
318
319 case ir_unop_i2i64:
320 case ir_unop_u2i64:
321 case ir_unop_b2i64:
322 case ir_unop_f2i64:
323 case ir_unop_d2i64:
324 case ir_unop_u642i64:
325 this->type = glsl_type::get_instance(GLSL_TYPE_INT64,
326 op0->type->vector_elements, 1);
327 break;
328
329 case ir_unop_i2u64:
330 case ir_unop_u2u64:
331 case ir_unop_f2u64:
332 case ir_unop_d2u64:
333 case ir_unop_i642u64:
334 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64,
335 op0->type->vector_elements, 1);
336 break;
337 case ir_unop_noise:
338 this->type = glsl_type::float_type;
339 break;
340
341 case ir_unop_unpack_double_2x32:
342 case ir_unop_unpack_uint_2x32:
343 this->type = glsl_type::uvec2_type;
344 break;
345
346 case ir_unop_unpack_int_2x32:
347 this->type = glsl_type::ivec2_type;
348 break;
349
350 case ir_unop_pack_snorm_2x16:
351 case ir_unop_pack_snorm_4x8:
352 case ir_unop_pack_unorm_2x16:
353 case ir_unop_pack_unorm_4x8:
354 case ir_unop_pack_half_2x16:
355 this->type = glsl_type::uint_type;
356 break;
357
358 case ir_unop_pack_double_2x32:
359 this->type = glsl_type::double_type;
360 break;
361
362 case ir_unop_pack_int_2x32:
363 this->type = glsl_type::int64_t_type;
364 break;
365
366 case ir_unop_pack_uint_2x32:
367 this->type = glsl_type::uint64_t_type;
368 break;
369
370 case ir_unop_unpack_snorm_2x16:
371 case ir_unop_unpack_unorm_2x16:
372 case ir_unop_unpack_half_2x16:
373 this->type = glsl_type::vec2_type;
374 break;
375
376 case ir_unop_unpack_snorm_4x8:
377 case ir_unop_unpack_unorm_4x8:
378 this->type = glsl_type::vec4_type;
379 break;
380
381 case ir_unop_unpack_sampler_2x32:
382 case ir_unop_unpack_image_2x32:
383 this->type = glsl_type::uvec2_type;
384 break;
385
386 case ir_unop_pack_sampler_2x32:
387 case ir_unop_pack_image_2x32:
388 this->type = op0->type;
389 break;
390
391 case ir_unop_frexp_sig:
392 this->type = op0->type;
393 break;
394 case ir_unop_frexp_exp:
395 this->type = glsl_type::get_instance(GLSL_TYPE_INT,
396 op0->type->vector_elements, 1);
397 break;
398
399 case ir_unop_get_buffer_size:
400 case ir_unop_ssbo_unsized_array_length:
401 this->type = glsl_type::int_type;
402 break;
403
404 case ir_unop_bitcast_i642d:
405 case ir_unop_bitcast_u642d:
406 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE,
407 op0->type->vector_elements, 1);
408 break;
409
410 case ir_unop_bitcast_d2i64:
411 this->type = glsl_type::get_instance(GLSL_TYPE_INT64,
412 op0->type->vector_elements, 1);
413 break;
414 case ir_unop_bitcast_d2u64:
415 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64,
416 op0->type->vector_elements, 1);
417 break;
418
419 default:
420 assert(!"not reached: missing automatic type setup for ir_expression");
421 this->type = op0->type;
422 break;
423 }
424 }
425
ir_expression(int op,ir_rvalue * op0,ir_rvalue * op1)426 ir_expression::ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1)
427 : ir_rvalue(ir_type_expression)
428 {
429 this->operation = ir_expression_operation(op);
430 this->operands[0] = op0;
431 this->operands[1] = op1;
432 this->operands[2] = NULL;
433 this->operands[3] = NULL;
434
435 assert(op > ir_last_unop);
436 init_num_operands();
437 assert(num_operands == 2);
438 for (unsigned i = 0; i < num_operands; i++) {
439 assert(this->operands[i] != NULL);
440 }
441
442 switch (this->operation) {
443 case ir_binop_all_equal:
444 case ir_binop_any_nequal:
445 this->type = glsl_type::bool_type;
446 break;
447
448 case ir_binop_add:
449 case ir_binop_sub:
450 case ir_binop_min:
451 case ir_binop_max:
452 case ir_binop_pow:
453 case ir_binop_mul:
454 case ir_binop_div:
455 case ir_binop_mod:
456 if (op0->type->is_scalar()) {
457 this->type = op1->type;
458 } else if (op1->type->is_scalar()) {
459 this->type = op0->type;
460 } else {
461 if (this->operation == ir_binop_mul) {
462 this->type = glsl_type::get_mul_type(op0->type, op1->type);
463 } else {
464 assert(op0->type == op1->type);
465 this->type = op0->type;
466 }
467 }
468 break;
469
470 case ir_binop_logic_and:
471 case ir_binop_logic_xor:
472 case ir_binop_logic_or:
473 case ir_binop_bit_and:
474 case ir_binop_bit_xor:
475 case ir_binop_bit_or:
476 assert(!op0->type->is_matrix());
477 assert(!op1->type->is_matrix());
478 if (op0->type->is_scalar()) {
479 this->type = op1->type;
480 } else if (op1->type->is_scalar()) {
481 this->type = op0->type;
482 } else {
483 assert(op0->type->vector_elements == op1->type->vector_elements);
484 this->type = op0->type;
485 }
486 break;
487
488 case ir_binop_equal:
489 case ir_binop_nequal:
490 case ir_binop_gequal:
491 case ir_binop_less:
492 assert(op0->type == op1->type);
493 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL,
494 op0->type->vector_elements, 1);
495 break;
496
497 case ir_binop_dot:
498 this->type = op0->type->get_base_type();
499 break;
500
501 case ir_binop_imul_high:
502 case ir_binop_carry:
503 case ir_binop_borrow:
504 case ir_binop_lshift:
505 case ir_binop_rshift:
506 case ir_binop_ldexp:
507 case ir_binop_interpolate_at_offset:
508 case ir_binop_interpolate_at_sample:
509 this->type = op0->type;
510 break;
511
512 case ir_binop_vector_extract:
513 this->type = op0->type->get_scalar_type();
514 break;
515
516 default:
517 assert(!"not reached: missing automatic type setup for ir_expression");
518 this->type = glsl_type::float_type;
519 }
520 }
521
ir_expression(int op,ir_rvalue * op0,ir_rvalue * op1,ir_rvalue * op2)522 ir_expression::ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1,
523 ir_rvalue *op2)
524 : ir_rvalue(ir_type_expression)
525 {
526 this->operation = ir_expression_operation(op);
527 this->operands[0] = op0;
528 this->operands[1] = op1;
529 this->operands[2] = op2;
530 this->operands[3] = NULL;
531
532 assert(op > ir_last_binop && op <= ir_last_triop);
533 init_num_operands();
534 assert(num_operands == 3);
535 for (unsigned i = 0; i < num_operands; i++) {
536 assert(this->operands[i] != NULL);
537 }
538
539 switch (this->operation) {
540 case ir_triop_fma:
541 case ir_triop_lrp:
542 case ir_triop_bitfield_extract:
543 case ir_triop_vector_insert:
544 this->type = op0->type;
545 break;
546
547 case ir_triop_csel:
548 this->type = op1->type;
549 break;
550
551 default:
552 assert(!"not reached: missing automatic type setup for ir_expression");
553 this->type = glsl_type::float_type;
554 }
555 }
556
557 /**
558 * This is only here for ir_reader to used for testing purposes. Please use
559 * the precomputed num_operands field if you need the number of operands.
560 */
561 unsigned
get_num_operands(ir_expression_operation op)562 ir_expression::get_num_operands(ir_expression_operation op)
563 {
564 assert(op <= ir_last_opcode);
565
566 if (op <= ir_last_unop)
567 return 1;
568
569 if (op <= ir_last_binop)
570 return 2;
571
572 if (op <= ir_last_triop)
573 return 3;
574
575 if (op <= ir_last_quadop)
576 return 4;
577
578 unreachable("Could not calculate number of operands");
579 }
580
581 #include "ir_expression_operation_strings.h"
582
583 const char*
depth_layout_string(ir_depth_layout layout)584 depth_layout_string(ir_depth_layout layout)
585 {
586 switch(layout) {
587 case ir_depth_layout_none: return "";
588 case ir_depth_layout_any: return "depth_any";
589 case ir_depth_layout_greater: return "depth_greater";
590 case ir_depth_layout_less: return "depth_less";
591 case ir_depth_layout_unchanged: return "depth_unchanged";
592
593 default:
594 assert(0);
595 return "";
596 }
597 }
598
599 ir_expression_operation
get_operator(const char * str)600 ir_expression::get_operator(const char *str)
601 {
602 for (int op = 0; op <= int(ir_last_opcode); op++) {
603 if (strcmp(str, ir_expression_operation_strings[op]) == 0)
604 return (ir_expression_operation) op;
605 }
606 return (ir_expression_operation) -1;
607 }
608
609 ir_variable *
variable_referenced() const610 ir_expression::variable_referenced() const
611 {
612 switch (operation) {
613 case ir_binop_vector_extract:
614 case ir_triop_vector_insert:
615 /* We get these for things like a[0] where a is a vector type. In these
616 * cases we want variable_referenced() to return the actual vector
617 * variable this is wrapping.
618 */
619 return operands[0]->variable_referenced();
620 default:
621 return ir_rvalue::variable_referenced();
622 }
623 }
624
ir_constant()625 ir_constant::ir_constant()
626 : ir_rvalue(ir_type_constant)
627 {
628 this->const_elements = NULL;
629 }
630
ir_constant(const struct glsl_type * type,const ir_constant_data * data)631 ir_constant::ir_constant(const struct glsl_type *type,
632 const ir_constant_data *data)
633 : ir_rvalue(ir_type_constant)
634 {
635 this->const_elements = NULL;
636
637 assert((type->base_type >= GLSL_TYPE_UINT)
638 && (type->base_type <= GLSL_TYPE_IMAGE));
639
640 this->type = type;
641 memcpy(& this->value, data, sizeof(this->value));
642 }
643
ir_constant(float f,unsigned vector_elements)644 ir_constant::ir_constant(float f, unsigned vector_elements)
645 : ir_rvalue(ir_type_constant)
646 {
647 assert(vector_elements <= 4);
648 this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT, vector_elements, 1);
649 for (unsigned i = 0; i < vector_elements; i++) {
650 this->value.f[i] = f;
651 }
652 for (unsigned i = vector_elements; i < 16; i++) {
653 this->value.f[i] = 0;
654 }
655 }
656
ir_constant(double d,unsigned vector_elements)657 ir_constant::ir_constant(double d, unsigned vector_elements)
658 : ir_rvalue(ir_type_constant)
659 {
660 assert(vector_elements <= 4);
661 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE, vector_elements, 1);
662 for (unsigned i = 0; i < vector_elements; i++) {
663 this->value.d[i] = d;
664 }
665 for (unsigned i = vector_elements; i < 16; i++) {
666 this->value.d[i] = 0.0;
667 }
668 }
669
ir_constant(unsigned int u,unsigned vector_elements)670 ir_constant::ir_constant(unsigned int u, unsigned vector_elements)
671 : ir_rvalue(ir_type_constant)
672 {
673 assert(vector_elements <= 4);
674 this->type = glsl_type::get_instance(GLSL_TYPE_UINT, vector_elements, 1);
675 for (unsigned i = 0; i < vector_elements; i++) {
676 this->value.u[i] = u;
677 }
678 for (unsigned i = vector_elements; i < 16; i++) {
679 this->value.u[i] = 0;
680 }
681 }
682
ir_constant(int integer,unsigned vector_elements)683 ir_constant::ir_constant(int integer, unsigned vector_elements)
684 : ir_rvalue(ir_type_constant)
685 {
686 assert(vector_elements <= 4);
687 this->type = glsl_type::get_instance(GLSL_TYPE_INT, vector_elements, 1);
688 for (unsigned i = 0; i < vector_elements; i++) {
689 this->value.i[i] = integer;
690 }
691 for (unsigned i = vector_elements; i < 16; i++) {
692 this->value.i[i] = 0;
693 }
694 }
695
ir_constant(uint64_t u64,unsigned vector_elements)696 ir_constant::ir_constant(uint64_t u64, unsigned vector_elements)
697 : ir_rvalue(ir_type_constant)
698 {
699 assert(vector_elements <= 4);
700 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64, vector_elements, 1);
701 for (unsigned i = 0; i < vector_elements; i++) {
702 this->value.u64[i] = u64;
703 }
704 for (unsigned i = vector_elements; i < 16; i++) {
705 this->value.u64[i] = 0;
706 }
707 }
708
ir_constant(int64_t int64,unsigned vector_elements)709 ir_constant::ir_constant(int64_t int64, unsigned vector_elements)
710 : ir_rvalue(ir_type_constant)
711 {
712 assert(vector_elements <= 4);
713 this->type = glsl_type::get_instance(GLSL_TYPE_INT64, vector_elements, 1);
714 for (unsigned i = 0; i < vector_elements; i++) {
715 this->value.i64[i] = int64;
716 }
717 for (unsigned i = vector_elements; i < 16; i++) {
718 this->value.i64[i] = 0;
719 }
720 }
721
ir_constant(bool b,unsigned vector_elements)722 ir_constant::ir_constant(bool b, unsigned vector_elements)
723 : ir_rvalue(ir_type_constant)
724 {
725 assert(vector_elements <= 4);
726 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL, vector_elements, 1);
727 for (unsigned i = 0; i < vector_elements; i++) {
728 this->value.b[i] = b;
729 }
730 for (unsigned i = vector_elements; i < 16; i++) {
731 this->value.b[i] = false;
732 }
733 }
734
ir_constant(const ir_constant * c,unsigned i)735 ir_constant::ir_constant(const ir_constant *c, unsigned i)
736 : ir_rvalue(ir_type_constant)
737 {
738 this->const_elements = NULL;
739 this->type = c->type->get_base_type();
740
741 switch (this->type->base_type) {
742 case GLSL_TYPE_UINT: this->value.u[0] = c->value.u[i]; break;
743 case GLSL_TYPE_INT: this->value.i[0] = c->value.i[i]; break;
744 case GLSL_TYPE_FLOAT: this->value.f[0] = c->value.f[i]; break;
745 case GLSL_TYPE_BOOL: this->value.b[0] = c->value.b[i]; break;
746 case GLSL_TYPE_DOUBLE: this->value.d[0] = c->value.d[i]; break;
747 default: assert(!"Should not get here."); break;
748 }
749 }
750
ir_constant(const struct glsl_type * type,exec_list * value_list)751 ir_constant::ir_constant(const struct glsl_type *type, exec_list *value_list)
752 : ir_rvalue(ir_type_constant)
753 {
754 this->const_elements = NULL;
755 this->type = type;
756
757 assert(type->is_scalar() || type->is_vector() || type->is_matrix()
758 || type->is_record() || type->is_array());
759
760 /* If the constant is a record, the types of each of the entries in
761 * value_list must be a 1-for-1 match with the structure components. Each
762 * entry must also be a constant. Just move the nodes from the value_list
763 * to the list in the ir_constant.
764 */
765 if (type->is_array() || type->is_record()) {
766 this->const_elements = ralloc_array(this, ir_constant *, type->length);
767 unsigned i = 0;
768 foreach_in_list(ir_constant, value, value_list) {
769 assert(value->as_constant() != NULL);
770
771 this->const_elements[i++] = value;
772 }
773 return;
774 }
775
776 for (unsigned i = 0; i < 16; i++) {
777 this->value.u[i] = 0;
778 }
779
780 ir_constant *value = (ir_constant *) (value_list->get_head_raw());
781
782 /* Constructors with exactly one scalar argument are special for vectors
783 * and matrices. For vectors, the scalar value is replicated to fill all
784 * the components. For matrices, the scalar fills the components of the
785 * diagonal while the rest is filled with 0.
786 */
787 if (value->type->is_scalar() && value->next->is_tail_sentinel()) {
788 if (type->is_matrix()) {
789 /* Matrix - fill diagonal (rest is already set to 0) */
790 assert(type->is_float() || type->is_double());
791 for (unsigned i = 0; i < type->matrix_columns; i++) {
792 if (type->is_float())
793 this->value.f[i * type->vector_elements + i] =
794 value->value.f[0];
795 else
796 this->value.d[i * type->vector_elements + i] =
797 value->value.d[0];
798 }
799 } else {
800 /* Vector or scalar - fill all components */
801 switch (type->base_type) {
802 case GLSL_TYPE_UINT:
803 case GLSL_TYPE_INT:
804 for (unsigned i = 0; i < type->components(); i++)
805 this->value.u[i] = value->value.u[0];
806 break;
807 case GLSL_TYPE_FLOAT:
808 for (unsigned i = 0; i < type->components(); i++)
809 this->value.f[i] = value->value.f[0];
810 break;
811 case GLSL_TYPE_DOUBLE:
812 for (unsigned i = 0; i < type->components(); i++)
813 this->value.d[i] = value->value.d[0];
814 break;
815 case GLSL_TYPE_UINT64:
816 case GLSL_TYPE_INT64:
817 for (unsigned i = 0; i < type->components(); i++)
818 this->value.u64[i] = value->value.u64[0];
819 break;
820 case GLSL_TYPE_BOOL:
821 for (unsigned i = 0; i < type->components(); i++)
822 this->value.b[i] = value->value.b[0];
823 break;
824 default:
825 assert(!"Should not get here.");
826 break;
827 }
828 }
829 return;
830 }
831
832 if (type->is_matrix() && value->type->is_matrix()) {
833 assert(value->next->is_tail_sentinel());
834
835 /* From section 5.4.2 of the GLSL 1.20 spec:
836 * "If a matrix is constructed from a matrix, then each component
837 * (column i, row j) in the result that has a corresponding component
838 * (column i, row j) in the argument will be initialized from there."
839 */
840 unsigned cols = MIN2(type->matrix_columns, value->type->matrix_columns);
841 unsigned rows = MIN2(type->vector_elements, value->type->vector_elements);
842 for (unsigned i = 0; i < cols; i++) {
843 for (unsigned j = 0; j < rows; j++) {
844 const unsigned src = i * value->type->vector_elements + j;
845 const unsigned dst = i * type->vector_elements + j;
846 this->value.f[dst] = value->value.f[src];
847 }
848 }
849
850 /* "All other components will be initialized to the identity matrix." */
851 for (unsigned i = cols; i < type->matrix_columns; i++)
852 this->value.f[i * type->vector_elements + i] = 1.0;
853
854 return;
855 }
856
857 /* Use each component from each entry in the value_list to initialize one
858 * component of the constant being constructed.
859 */
860 unsigned i = 0;
861 for (;;) {
862 assert(value->as_constant() != NULL);
863 assert(!value->is_tail_sentinel());
864
865 for (unsigned j = 0; j < value->type->components(); j++) {
866 switch (type->base_type) {
867 case GLSL_TYPE_UINT:
868 this->value.u[i] = value->get_uint_component(j);
869 break;
870 case GLSL_TYPE_INT:
871 this->value.i[i] = value->get_int_component(j);
872 break;
873 case GLSL_TYPE_FLOAT:
874 this->value.f[i] = value->get_float_component(j);
875 break;
876 case GLSL_TYPE_BOOL:
877 this->value.b[i] = value->get_bool_component(j);
878 break;
879 case GLSL_TYPE_DOUBLE:
880 this->value.d[i] = value->get_double_component(j);
881 break;
882 case GLSL_TYPE_UINT64:
883 this->value.u64[i] = value->get_uint64_component(j);
884 break;
885 case GLSL_TYPE_INT64:
886 this->value.i64[i] = value->get_int64_component(j);
887 break;
888 default:
889 /* FINISHME: What to do? Exceptions are not the answer.
890 */
891 break;
892 }
893
894 i++;
895 if (i >= type->components())
896 break;
897 }
898
899 if (i >= type->components())
900 break; /* avoid downcasting a list sentinel */
901 value = (ir_constant *) value->next;
902 }
903 }
904
905 ir_constant *
zero(void * mem_ctx,const glsl_type * type)906 ir_constant::zero(void *mem_ctx, const glsl_type *type)
907 {
908 assert(type->is_scalar() || type->is_vector() || type->is_matrix()
909 || type->is_record() || type->is_array());
910
911 ir_constant *c = new(mem_ctx) ir_constant;
912 c->type = type;
913 memset(&c->value, 0, sizeof(c->value));
914
915 if (type->is_array()) {
916 c->const_elements = ralloc_array(c, ir_constant *, type->length);
917
918 for (unsigned i = 0; i < type->length; i++)
919 c->const_elements[i] = ir_constant::zero(c, type->fields.array);
920 }
921
922 if (type->is_record()) {
923 c->const_elements = ralloc_array(c, ir_constant *, type->length);
924
925 for (unsigned i = 0; i < type->length; i++) {
926 c->const_elements[i] =
927 ir_constant::zero(mem_ctx, type->fields.structure[i].type);
928 }
929 }
930
931 return c;
932 }
933
934 bool
get_bool_component(unsigned i) const935 ir_constant::get_bool_component(unsigned i) const
936 {
937 switch (this->type->base_type) {
938 case GLSL_TYPE_UINT: return this->value.u[i] != 0;
939 case GLSL_TYPE_INT: return this->value.i[i] != 0;
940 case GLSL_TYPE_FLOAT: return ((int)this->value.f[i]) != 0;
941 case GLSL_TYPE_BOOL: return this->value.b[i];
942 case GLSL_TYPE_DOUBLE: return this->value.d[i] != 0.0;
943 case GLSL_TYPE_UINT64: return this->value.u64[i] != 0;
944 case GLSL_TYPE_INT64: return this->value.i64[i] != 0;
945 default: assert(!"Should not get here."); break;
946 }
947
948 /* Must return something to make the compiler happy. This is clearly an
949 * error case.
950 */
951 return false;
952 }
953
954 float
get_float_component(unsigned i) const955 ir_constant::get_float_component(unsigned i) const
956 {
957 switch (this->type->base_type) {
958 case GLSL_TYPE_UINT: return (float) this->value.u[i];
959 case GLSL_TYPE_INT: return (float) this->value.i[i];
960 case GLSL_TYPE_FLOAT: return this->value.f[i];
961 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1.0f : 0.0f;
962 case GLSL_TYPE_DOUBLE: return (float) this->value.d[i];
963 case GLSL_TYPE_UINT64: return (float) this->value.u64[i];
964 case GLSL_TYPE_INT64: return (float) this->value.i64[i];
965 default: assert(!"Should not get here."); break;
966 }
967
968 /* Must return something to make the compiler happy. This is clearly an
969 * error case.
970 */
971 return 0.0;
972 }
973
974 double
get_double_component(unsigned i) const975 ir_constant::get_double_component(unsigned i) const
976 {
977 switch (this->type->base_type) {
978 case GLSL_TYPE_UINT: return (double) this->value.u[i];
979 case GLSL_TYPE_INT: return (double) this->value.i[i];
980 case GLSL_TYPE_FLOAT: return (double) this->value.f[i];
981 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1.0 : 0.0;
982 case GLSL_TYPE_DOUBLE: return this->value.d[i];
983 case GLSL_TYPE_UINT64: return (double) this->value.u64[i];
984 case GLSL_TYPE_INT64: return (double) this->value.i64[i];
985 default: assert(!"Should not get here."); break;
986 }
987
988 /* Must return something to make the compiler happy. This is clearly an
989 * error case.
990 */
991 return 0.0;
992 }
993
994 int
get_int_component(unsigned i) const995 ir_constant::get_int_component(unsigned i) const
996 {
997 switch (this->type->base_type) {
998 case GLSL_TYPE_UINT: return this->value.u[i];
999 case GLSL_TYPE_INT: return this->value.i[i];
1000 case GLSL_TYPE_FLOAT: return (int) this->value.f[i];
1001 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1002 case GLSL_TYPE_DOUBLE: return (int) this->value.d[i];
1003 case GLSL_TYPE_UINT64: return (int) this->value.u64[i];
1004 case GLSL_TYPE_INT64: return (int) this->value.i64[i];
1005 default: assert(!"Should not get here."); break;
1006 }
1007
1008 /* Must return something to make the compiler happy. This is clearly an
1009 * error case.
1010 */
1011 return 0;
1012 }
1013
1014 unsigned
get_uint_component(unsigned i) const1015 ir_constant::get_uint_component(unsigned i) const
1016 {
1017 switch (this->type->base_type) {
1018 case GLSL_TYPE_UINT: return this->value.u[i];
1019 case GLSL_TYPE_INT: return this->value.i[i];
1020 case GLSL_TYPE_FLOAT: return (unsigned) this->value.f[i];
1021 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1022 case GLSL_TYPE_DOUBLE: return (unsigned) this->value.d[i];
1023 case GLSL_TYPE_UINT64: return (unsigned) this->value.u64[i];
1024 case GLSL_TYPE_INT64: return (unsigned) this->value.i64[i];
1025 default: assert(!"Should not get here."); break;
1026 }
1027
1028 /* Must return something to make the compiler happy. This is clearly an
1029 * error case.
1030 */
1031 return 0;
1032 }
1033
1034 int64_t
get_int64_component(unsigned i) const1035 ir_constant::get_int64_component(unsigned i) const
1036 {
1037 switch (this->type->base_type) {
1038 case GLSL_TYPE_UINT: return this->value.u[i];
1039 case GLSL_TYPE_INT: return this->value.i[i];
1040 case GLSL_TYPE_FLOAT: return (int64_t) this->value.f[i];
1041 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1042 case GLSL_TYPE_DOUBLE: return (int64_t) this->value.d[i];
1043 case GLSL_TYPE_UINT64: return (int64_t) this->value.u64[i];
1044 case GLSL_TYPE_INT64: return this->value.i64[i];
1045 default: assert(!"Should not get here."); break;
1046 }
1047
1048 /* Must return something to make the compiler happy. This is clearly an
1049 * error case.
1050 */
1051 return 0;
1052 }
1053
1054 uint64_t
get_uint64_component(unsigned i) const1055 ir_constant::get_uint64_component(unsigned i) const
1056 {
1057 switch (this->type->base_type) {
1058 case GLSL_TYPE_UINT: return this->value.u[i];
1059 case GLSL_TYPE_INT: return this->value.i[i];
1060 case GLSL_TYPE_FLOAT: return (uint64_t) this->value.f[i];
1061 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1062 case GLSL_TYPE_DOUBLE: return (uint64_t) this->value.d[i];
1063 case GLSL_TYPE_UINT64: return this->value.u64[i];
1064 case GLSL_TYPE_INT64: return (uint64_t) this->value.i64[i];
1065 default: assert(!"Should not get here."); break;
1066 }
1067
1068 /* Must return something to make the compiler happy. This is clearly an
1069 * error case.
1070 */
1071 return 0;
1072 }
1073
1074 ir_constant *
get_array_element(unsigned i) const1075 ir_constant::get_array_element(unsigned i) const
1076 {
1077 assert(this->type->is_array());
1078
1079 /* From page 35 (page 41 of the PDF) of the GLSL 1.20 spec:
1080 *
1081 * "Behavior is undefined if a shader subscripts an array with an index
1082 * less than 0 or greater than or equal to the size the array was
1083 * declared with."
1084 *
1085 * Most out-of-bounds accesses are removed before things could get this far.
1086 * There are cases where non-constant array index values can get constant
1087 * folded.
1088 */
1089 if (int(i) < 0)
1090 i = 0;
1091 else if (i >= this->type->length)
1092 i = this->type->length - 1;
1093
1094 return const_elements[i];
1095 }
1096
1097 ir_constant *
get_record_field(int idx)1098 ir_constant::get_record_field(int idx)
1099 {
1100 assert(this->type->is_record());
1101 assert(idx >= 0 && (unsigned) idx < this->type->length);
1102
1103 return const_elements[idx];
1104 }
1105
1106 void
copy_offset(ir_constant * src,int offset)1107 ir_constant::copy_offset(ir_constant *src, int offset)
1108 {
1109 switch (this->type->base_type) {
1110 case GLSL_TYPE_UINT:
1111 case GLSL_TYPE_INT:
1112 case GLSL_TYPE_FLOAT:
1113 case GLSL_TYPE_DOUBLE:
1114 case GLSL_TYPE_UINT64:
1115 case GLSL_TYPE_INT64:
1116 case GLSL_TYPE_BOOL: {
1117 unsigned int size = src->type->components();
1118 assert (size <= this->type->components() - offset);
1119 for (unsigned int i=0; i<size; i++) {
1120 switch (this->type->base_type) {
1121 case GLSL_TYPE_UINT:
1122 value.u[i+offset] = src->get_uint_component(i);
1123 break;
1124 case GLSL_TYPE_INT:
1125 value.i[i+offset] = src->get_int_component(i);
1126 break;
1127 case GLSL_TYPE_FLOAT:
1128 value.f[i+offset] = src->get_float_component(i);
1129 break;
1130 case GLSL_TYPE_BOOL:
1131 value.b[i+offset] = src->get_bool_component(i);
1132 break;
1133 case GLSL_TYPE_DOUBLE:
1134 value.d[i+offset] = src->get_double_component(i);
1135 break;
1136 case GLSL_TYPE_UINT64:
1137 value.u64[i+offset] = src->get_uint64_component(i);
1138 break;
1139 case GLSL_TYPE_INT64:
1140 value.i64[i+offset] = src->get_int64_component(i);
1141 break;
1142 default: // Shut up the compiler
1143 break;
1144 }
1145 }
1146 break;
1147 }
1148
1149 case GLSL_TYPE_STRUCT:
1150 case GLSL_TYPE_ARRAY: {
1151 assert (src->type == this->type);
1152 for (unsigned i = 0; i < this->type->length; i++) {
1153 this->const_elements[i] = src->const_elements[i]->clone(this, NULL);
1154 }
1155 break;
1156 }
1157
1158 default:
1159 assert(!"Should not get here.");
1160 break;
1161 }
1162 }
1163
1164 void
copy_masked_offset(ir_constant * src,int offset,unsigned int mask)1165 ir_constant::copy_masked_offset(ir_constant *src, int offset, unsigned int mask)
1166 {
1167 assert (!type->is_array() && !type->is_record());
1168
1169 if (!type->is_vector() && !type->is_matrix()) {
1170 offset = 0;
1171 mask = 1;
1172 }
1173
1174 int id = 0;
1175 for (int i=0; i<4; i++) {
1176 if (mask & (1 << i)) {
1177 switch (this->type->base_type) {
1178 case GLSL_TYPE_UINT:
1179 value.u[i+offset] = src->get_uint_component(id++);
1180 break;
1181 case GLSL_TYPE_INT:
1182 value.i[i+offset] = src->get_int_component(id++);
1183 break;
1184 case GLSL_TYPE_FLOAT:
1185 value.f[i+offset] = src->get_float_component(id++);
1186 break;
1187 case GLSL_TYPE_BOOL:
1188 value.b[i+offset] = src->get_bool_component(id++);
1189 break;
1190 case GLSL_TYPE_DOUBLE:
1191 value.d[i+offset] = src->get_double_component(id++);
1192 break;
1193 case GLSL_TYPE_UINT64:
1194 value.u64[i+offset] = src->get_uint64_component(id++);
1195 break;
1196 case GLSL_TYPE_INT64:
1197 value.i64[i+offset] = src->get_int64_component(id++);
1198 break;
1199 default:
1200 assert(!"Should not get here.");
1201 return;
1202 }
1203 }
1204 }
1205 }
1206
1207 bool
has_value(const ir_constant * c) const1208 ir_constant::has_value(const ir_constant *c) const
1209 {
1210 if (this->type != c->type)
1211 return false;
1212
1213 if (this->type->is_array() || this->type->is_record()) {
1214 for (unsigned i = 0; i < this->type->length; i++) {
1215 if (!this->const_elements[i]->has_value(c->const_elements[i]))
1216 return false;
1217 }
1218 return true;
1219 }
1220
1221 for (unsigned i = 0; i < this->type->components(); i++) {
1222 switch (this->type->base_type) {
1223 case GLSL_TYPE_UINT:
1224 if (this->value.u[i] != c->value.u[i])
1225 return false;
1226 break;
1227 case GLSL_TYPE_INT:
1228 if (this->value.i[i] != c->value.i[i])
1229 return false;
1230 break;
1231 case GLSL_TYPE_FLOAT:
1232 if (this->value.f[i] != c->value.f[i])
1233 return false;
1234 break;
1235 case GLSL_TYPE_BOOL:
1236 if (this->value.b[i] != c->value.b[i])
1237 return false;
1238 break;
1239 case GLSL_TYPE_DOUBLE:
1240 if (this->value.d[i] != c->value.d[i])
1241 return false;
1242 break;
1243 case GLSL_TYPE_UINT64:
1244 if (this->value.u64[i] != c->value.u64[i])
1245 return false;
1246 break;
1247 case GLSL_TYPE_INT64:
1248 if (this->value.i64[i] != c->value.i64[i])
1249 return false;
1250 break;
1251 default:
1252 assert(!"Should not get here.");
1253 return false;
1254 }
1255 }
1256
1257 return true;
1258 }
1259
1260 bool
is_value(float f,int i) const1261 ir_constant::is_value(float f, int i) const
1262 {
1263 if (!this->type->is_scalar() && !this->type->is_vector())
1264 return false;
1265
1266 /* Only accept boolean values for 0/1. */
1267 if (int(bool(i)) != i && this->type->is_boolean())
1268 return false;
1269
1270 for (unsigned c = 0; c < this->type->vector_elements; c++) {
1271 switch (this->type->base_type) {
1272 case GLSL_TYPE_FLOAT:
1273 if (this->value.f[c] != f)
1274 return false;
1275 break;
1276 case GLSL_TYPE_INT:
1277 if (this->value.i[c] != i)
1278 return false;
1279 break;
1280 case GLSL_TYPE_UINT:
1281 if (this->value.u[c] != unsigned(i))
1282 return false;
1283 break;
1284 case GLSL_TYPE_BOOL:
1285 if (this->value.b[c] != bool(i))
1286 return false;
1287 break;
1288 case GLSL_TYPE_DOUBLE:
1289 if (this->value.d[c] != double(f))
1290 return false;
1291 break;
1292 case GLSL_TYPE_UINT64:
1293 if (this->value.u64[c] != uint64_t(i))
1294 return false;
1295 break;
1296 case GLSL_TYPE_INT64:
1297 if (this->value.i64[c] != i)
1298 return false;
1299 break;
1300 default:
1301 /* The only other base types are structures, arrays, and samplers.
1302 * Samplers cannot be constants, and the others should have been
1303 * filtered out above.
1304 */
1305 assert(!"Should not get here.");
1306 return false;
1307 }
1308 }
1309
1310 return true;
1311 }
1312
1313 bool
is_zero() const1314 ir_constant::is_zero() const
1315 {
1316 return is_value(0.0, 0);
1317 }
1318
1319 bool
is_one() const1320 ir_constant::is_one() const
1321 {
1322 return is_value(1.0, 1);
1323 }
1324
1325 bool
is_negative_one() const1326 ir_constant::is_negative_one() const
1327 {
1328 return is_value(-1.0, -1);
1329 }
1330
1331 bool
is_uint16_constant() const1332 ir_constant::is_uint16_constant() const
1333 {
1334 if (!type->is_integer())
1335 return false;
1336
1337 return value.u[0] < (1 << 16);
1338 }
1339
ir_loop()1340 ir_loop::ir_loop()
1341 : ir_instruction(ir_type_loop)
1342 {
1343 }
1344
1345
ir_dereference_variable(ir_variable * var)1346 ir_dereference_variable::ir_dereference_variable(ir_variable *var)
1347 : ir_dereference(ir_type_dereference_variable)
1348 {
1349 assert(var != NULL);
1350
1351 this->var = var;
1352 this->type = var->type;
1353 }
1354
1355
ir_dereference_array(ir_rvalue * value,ir_rvalue * array_index)1356 ir_dereference_array::ir_dereference_array(ir_rvalue *value,
1357 ir_rvalue *array_index)
1358 : ir_dereference(ir_type_dereference_array)
1359 {
1360 this->array_index = array_index;
1361 this->set_array(value);
1362 }
1363
1364
ir_dereference_array(ir_variable * var,ir_rvalue * array_index)1365 ir_dereference_array::ir_dereference_array(ir_variable *var,
1366 ir_rvalue *array_index)
1367 : ir_dereference(ir_type_dereference_array)
1368 {
1369 void *ctx = ralloc_parent(var);
1370
1371 this->array_index = array_index;
1372 this->set_array(new(ctx) ir_dereference_variable(var));
1373 }
1374
1375
1376 void
set_array(ir_rvalue * value)1377 ir_dereference_array::set_array(ir_rvalue *value)
1378 {
1379 assert(value != NULL);
1380
1381 this->array = value;
1382
1383 const glsl_type *const vt = this->array->type;
1384
1385 if (vt->is_array()) {
1386 type = vt->fields.array;
1387 } else if (vt->is_matrix()) {
1388 type = vt->column_type();
1389 } else if (vt->is_vector()) {
1390 type = vt->get_base_type();
1391 }
1392 }
1393
1394
ir_dereference_record(ir_rvalue * value,const char * field)1395 ir_dereference_record::ir_dereference_record(ir_rvalue *value,
1396 const char *field)
1397 : ir_dereference(ir_type_dereference_record)
1398 {
1399 assert(value != NULL);
1400
1401 this->record = value;
1402 this->type = this->record->type->field_type(field);
1403 this->field_idx = this->record->type->field_index(field);
1404 }
1405
1406
ir_dereference_record(ir_variable * var,const char * field)1407 ir_dereference_record::ir_dereference_record(ir_variable *var,
1408 const char *field)
1409 : ir_dereference(ir_type_dereference_record)
1410 {
1411 void *ctx = ralloc_parent(var);
1412
1413 this->record = new(ctx) ir_dereference_variable(var);
1414 this->type = this->record->type->field_type(field);
1415 this->field_idx = this->record->type->field_index(field);
1416 }
1417
1418 bool
is_lvalue(const struct _mesa_glsl_parse_state * state) const1419 ir_dereference::is_lvalue(const struct _mesa_glsl_parse_state *state) const
1420 {
1421 ir_variable *var = this->variable_referenced();
1422
1423 /* Every l-value derference chain eventually ends in a variable.
1424 */
1425 if ((var == NULL) || var->data.read_only)
1426 return false;
1427
1428 /* From section 4.1.7 of the ARB_bindless_texture spec:
1429 *
1430 * "Samplers can be used as l-values, so can be assigned into and used as
1431 * "out" and "inout" function parameters."
1432 *
1433 * From section 4.1.X of the ARB_bindless_texture spec:
1434 *
1435 * "Images can be used as l-values, so can be assigned into and used as
1436 * "out" and "inout" function parameters."
1437 */
1438 if ((!state || state->has_bindless()) &&
1439 (this->type->contains_sampler() || this->type->contains_image()))
1440 return true;
1441
1442 /* From section 4.1.7 of the GLSL 4.40 spec:
1443 *
1444 * "Opaque variables cannot be treated as l-values; hence cannot
1445 * be used as out or inout function parameters, nor can they be
1446 * assigned into."
1447 */
1448 if (this->type->contains_opaque())
1449 return false;
1450
1451 return true;
1452 }
1453
1454
1455 static const char * const tex_opcode_strs[] = { "tex", "txb", "txl", "txd", "txf", "txf_ms", "txs", "lod", "tg4", "query_levels", "texture_samples", "samples_identical" };
1456
opcode_string()1457 const char *ir_texture::opcode_string()
1458 {
1459 assert((unsigned int) op < ARRAY_SIZE(tex_opcode_strs));
1460 return tex_opcode_strs[op];
1461 }
1462
1463 ir_texture_opcode
get_opcode(const char * str)1464 ir_texture::get_opcode(const char *str)
1465 {
1466 const int count = sizeof(tex_opcode_strs) / sizeof(tex_opcode_strs[0]);
1467 for (int op = 0; op < count; op++) {
1468 if (strcmp(str, tex_opcode_strs[op]) == 0)
1469 return (ir_texture_opcode) op;
1470 }
1471 return (ir_texture_opcode) -1;
1472 }
1473
1474
1475 void
set_sampler(ir_dereference * sampler,const glsl_type * type)1476 ir_texture::set_sampler(ir_dereference *sampler, const glsl_type *type)
1477 {
1478 assert(sampler != NULL);
1479 assert(type != NULL);
1480 this->sampler = sampler;
1481 this->type = type;
1482
1483 if (this->op == ir_txs || this->op == ir_query_levels ||
1484 this->op == ir_texture_samples) {
1485 assert(type->base_type == GLSL_TYPE_INT);
1486 } else if (this->op == ir_lod) {
1487 assert(type->vector_elements == 2);
1488 assert(type->is_float());
1489 } else if (this->op == ir_samples_identical) {
1490 assert(type == glsl_type::bool_type);
1491 assert(sampler->type->is_sampler());
1492 assert(sampler->type->sampler_dimensionality == GLSL_SAMPLER_DIM_MS);
1493 } else {
1494 assert(sampler->type->sampled_type == (int) type->base_type);
1495 if (sampler->type->sampler_shadow)
1496 assert(type->vector_elements == 4 || type->vector_elements == 1);
1497 else
1498 assert(type->vector_elements == 4);
1499 }
1500 }
1501
1502
1503 void
init_mask(const unsigned * comp,unsigned count)1504 ir_swizzle::init_mask(const unsigned *comp, unsigned count)
1505 {
1506 assert((count >= 1) && (count <= 4));
1507
1508 memset(&this->mask, 0, sizeof(this->mask));
1509 this->mask.num_components = count;
1510
1511 unsigned dup_mask = 0;
1512 switch (count) {
1513 case 4:
1514 assert(comp[3] <= 3);
1515 dup_mask |= (1U << comp[3])
1516 & ((1U << comp[0]) | (1U << comp[1]) | (1U << comp[2]));
1517 this->mask.w = comp[3];
1518
1519 case 3:
1520 assert(comp[2] <= 3);
1521 dup_mask |= (1U << comp[2])
1522 & ((1U << comp[0]) | (1U << comp[1]));
1523 this->mask.z = comp[2];
1524
1525 case 2:
1526 assert(comp[1] <= 3);
1527 dup_mask |= (1U << comp[1])
1528 & ((1U << comp[0]));
1529 this->mask.y = comp[1];
1530
1531 case 1:
1532 assert(comp[0] <= 3);
1533 this->mask.x = comp[0];
1534 }
1535
1536 this->mask.has_duplicates = dup_mask != 0;
1537
1538 /* Based on the number of elements in the swizzle and the base type
1539 * (i.e., float, int, unsigned, or bool) of the vector being swizzled,
1540 * generate the type of the resulting value.
1541 */
1542 type = glsl_type::get_instance(val->type->base_type, mask.num_components, 1);
1543 }
1544
ir_swizzle(ir_rvalue * val,unsigned x,unsigned y,unsigned z,unsigned w,unsigned count)1545 ir_swizzle::ir_swizzle(ir_rvalue *val, unsigned x, unsigned y, unsigned z,
1546 unsigned w, unsigned count)
1547 : ir_rvalue(ir_type_swizzle), val(val)
1548 {
1549 const unsigned components[4] = { x, y, z, w };
1550 this->init_mask(components, count);
1551 }
1552
ir_swizzle(ir_rvalue * val,const unsigned * comp,unsigned count)1553 ir_swizzle::ir_swizzle(ir_rvalue *val, const unsigned *comp,
1554 unsigned count)
1555 : ir_rvalue(ir_type_swizzle), val(val)
1556 {
1557 this->init_mask(comp, count);
1558 }
1559
ir_swizzle(ir_rvalue * val,ir_swizzle_mask mask)1560 ir_swizzle::ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask)
1561 : ir_rvalue(ir_type_swizzle), val(val), mask(mask)
1562 {
1563 this->type = glsl_type::get_instance(val->type->base_type,
1564 mask.num_components, 1);
1565 }
1566
1567 #define X 1
1568 #define R 5
1569 #define S 9
1570 #define I 13
1571
1572 ir_swizzle *
create(ir_rvalue * val,const char * str,unsigned vector_length)1573 ir_swizzle::create(ir_rvalue *val, const char *str, unsigned vector_length)
1574 {
1575 void *ctx = ralloc_parent(val);
1576
1577 /* For each possible swizzle character, this table encodes the value in
1578 * \c idx_map that represents the 0th element of the vector. For invalid
1579 * swizzle characters (e.g., 'k'), a special value is used that will allow
1580 * detection of errors.
1581 */
1582 static const unsigned char base_idx[26] = {
1583 /* a b c d e f g h i j k l m */
1584 R, R, I, I, I, I, R, I, I, I, I, I, I,
1585 /* n o p q r s t u v w x y z */
1586 I, I, S, S, R, S, S, I, I, X, X, X, X
1587 };
1588
1589 /* Each valid swizzle character has an entry in the previous table. This
1590 * table encodes the base index encoded in the previous table plus the actual
1591 * index of the swizzle character. When processing swizzles, the first
1592 * character in the string is indexed in the previous table. Each character
1593 * in the string is indexed in this table, and the value found there has the
1594 * value form the first table subtracted. The result must be on the range
1595 * [0,3].
1596 *
1597 * For example, the string "wzyx" will get X from the first table. Each of
1598 * the charcaters will get X+3, X+2, X+1, and X+0 from this table. After
1599 * subtraction, the swizzle values are { 3, 2, 1, 0 }.
1600 *
1601 * The string "wzrg" will get X from the first table. Each of the characters
1602 * will get X+3, X+2, R+0, and R+1 from this table. After subtraction, the
1603 * swizzle values are { 3, 2, 4, 5 }. Since 4 and 5 are outside the range
1604 * [0,3], the error is detected.
1605 */
1606 static const unsigned char idx_map[26] = {
1607 /* a b c d e f g h i j k l m */
1608 R+3, R+2, 0, 0, 0, 0, R+1, 0, 0, 0, 0, 0, 0,
1609 /* n o p q r s t u v w x y z */
1610 0, 0, S+2, S+3, R+0, S+0, S+1, 0, 0, X+3, X+0, X+1, X+2
1611 };
1612
1613 int swiz_idx[4] = { 0, 0, 0, 0 };
1614 unsigned i;
1615
1616
1617 /* Validate the first character in the swizzle string and look up the base
1618 * index value as described above.
1619 */
1620 if ((str[0] < 'a') || (str[0] > 'z'))
1621 return NULL;
1622
1623 const unsigned base = base_idx[str[0] - 'a'];
1624
1625
1626 for (i = 0; (i < 4) && (str[i] != '\0'); i++) {
1627 /* Validate the next character, and, as described above, convert it to a
1628 * swizzle index.
1629 */
1630 if ((str[i] < 'a') || (str[i] > 'z'))
1631 return NULL;
1632
1633 swiz_idx[i] = idx_map[str[i] - 'a'] - base;
1634 if ((swiz_idx[i] < 0) || (swiz_idx[i] >= (int) vector_length))
1635 return NULL;
1636 }
1637
1638 if (str[i] != '\0')
1639 return NULL;
1640
1641 return new(ctx) ir_swizzle(val, swiz_idx[0], swiz_idx[1], swiz_idx[2],
1642 swiz_idx[3], i);
1643 }
1644
1645 #undef X
1646 #undef R
1647 #undef S
1648 #undef I
1649
1650 ir_variable *
variable_referenced() const1651 ir_swizzle::variable_referenced() const
1652 {
1653 return this->val->variable_referenced();
1654 }
1655
1656
1657 bool ir_variable::temporaries_allocate_names = false;
1658
1659 const char ir_variable::tmp_name[] = "compiler_temp";
1660
ir_variable(const struct glsl_type * type,const char * name,ir_variable_mode mode)1661 ir_variable::ir_variable(const struct glsl_type *type, const char *name,
1662 ir_variable_mode mode)
1663 : ir_instruction(ir_type_variable)
1664 {
1665 this->type = type;
1666
1667 if (mode == ir_var_temporary && !ir_variable::temporaries_allocate_names)
1668 name = NULL;
1669
1670 /* The ir_variable clone method may call this constructor with name set to
1671 * tmp_name.
1672 */
1673 assert(name != NULL
1674 || mode == ir_var_temporary
1675 || mode == ir_var_function_in
1676 || mode == ir_var_function_out
1677 || mode == ir_var_function_inout);
1678 assert(name != ir_variable::tmp_name
1679 || mode == ir_var_temporary);
1680 if (mode == ir_var_temporary
1681 && (name == NULL || name == ir_variable::tmp_name)) {
1682 this->name = ir_variable::tmp_name;
1683 } else if (name == NULL ||
1684 strlen(name) < ARRAY_SIZE(this->name_storage)) {
1685 strcpy(this->name_storage, name ? name : "");
1686 this->name = this->name_storage;
1687 } else {
1688 this->name = ralloc_strdup(this, name);
1689 }
1690
1691 this->u.max_ifc_array_access = NULL;
1692
1693 this->data.explicit_location = false;
1694 this->data.has_initializer = false;
1695 this->data.location = -1;
1696 this->data.location_frac = 0;
1697 this->data.binding = 0;
1698 this->data.warn_extension_index = 0;
1699 this->constant_value = NULL;
1700 this->constant_initializer = NULL;
1701 this->data.origin_upper_left = false;
1702 this->data.pixel_center_integer = false;
1703 this->data.depth_layout = ir_depth_layout_none;
1704 this->data.used = false;
1705 this->data.always_active_io = false;
1706 this->data.read_only = false;
1707 this->data.centroid = false;
1708 this->data.sample = false;
1709 this->data.patch = false;
1710 this->data.invariant = false;
1711 this->data.how_declared = ir_var_declared_normally;
1712 this->data.mode = mode;
1713 this->data.interpolation = INTERP_MODE_NONE;
1714 this->data.max_array_access = -1;
1715 this->data.offset = 0;
1716 this->data.precision = GLSL_PRECISION_NONE;
1717 this->data.memory_read_only = false;
1718 this->data.memory_write_only = false;
1719 this->data.memory_coherent = false;
1720 this->data.memory_volatile = false;
1721 this->data.memory_restrict = false;
1722 this->data.from_ssbo_unsized_array = false;
1723 this->data.fb_fetch_output = false;
1724 this->data.bindless = false;
1725 this->data.bound = false;
1726
1727 if (type != NULL) {
1728 if (type->is_interface())
1729 this->init_interface_type(type);
1730 else if (type->without_array()->is_interface())
1731 this->init_interface_type(type->without_array());
1732 }
1733 }
1734
1735
1736 const char *
interpolation_string(unsigned interpolation)1737 interpolation_string(unsigned interpolation)
1738 {
1739 switch (interpolation) {
1740 case INTERP_MODE_NONE: return "no";
1741 case INTERP_MODE_SMOOTH: return "smooth";
1742 case INTERP_MODE_FLAT: return "flat";
1743 case INTERP_MODE_NOPERSPECTIVE: return "noperspective";
1744 }
1745
1746 assert(!"Should not get here.");
1747 return "";
1748 }
1749
1750 const char *const ir_variable::warn_extension_table[] = {
1751 "",
1752 "GL_ARB_shader_stencil_export",
1753 "GL_AMD_shader_stencil_export",
1754 };
1755
1756 void
enable_extension_warning(const char * extension)1757 ir_variable::enable_extension_warning(const char *extension)
1758 {
1759 for (unsigned i = 0; i < ARRAY_SIZE(warn_extension_table); i++) {
1760 if (strcmp(warn_extension_table[i], extension) == 0) {
1761 this->data.warn_extension_index = i;
1762 return;
1763 }
1764 }
1765
1766 assert(!"Should not get here.");
1767 this->data.warn_extension_index = 0;
1768 }
1769
1770 const char *
get_extension_warning() const1771 ir_variable::get_extension_warning() const
1772 {
1773 return this->data.warn_extension_index == 0
1774 ? NULL : warn_extension_table[this->data.warn_extension_index];
1775 }
1776
ir_function_signature(const glsl_type * return_type,builtin_available_predicate b)1777 ir_function_signature::ir_function_signature(const glsl_type *return_type,
1778 builtin_available_predicate b)
1779 : ir_instruction(ir_type_function_signature),
1780 return_type(return_type), is_defined(false),
1781 intrinsic_id(ir_intrinsic_invalid), builtin_avail(b), _function(NULL)
1782 {
1783 this->origin = NULL;
1784 }
1785
1786
1787 bool
is_builtin() const1788 ir_function_signature::is_builtin() const
1789 {
1790 return builtin_avail != NULL;
1791 }
1792
1793
1794 bool
is_builtin_available(const _mesa_glsl_parse_state * state) const1795 ir_function_signature::is_builtin_available(const _mesa_glsl_parse_state *state) const
1796 {
1797 /* We can't call the predicate without a state pointer, so just say that
1798 * the signature is available. At compile time, we need the filtering,
1799 * but also receive a valid state pointer. At link time, we're resolving
1800 * imported built-in prototypes to their definitions, which will always
1801 * be an exact match. So we can skip the filtering.
1802 */
1803 if (state == NULL)
1804 return true;
1805
1806 assert(builtin_avail != NULL);
1807 return builtin_avail(state);
1808 }
1809
1810
1811 static bool
modes_match(unsigned a,unsigned b)1812 modes_match(unsigned a, unsigned b)
1813 {
1814 if (a == b)
1815 return true;
1816
1817 /* Accept "in" vs. "const in" */
1818 if ((a == ir_var_const_in && b == ir_var_function_in) ||
1819 (b == ir_var_const_in && a == ir_var_function_in))
1820 return true;
1821
1822 return false;
1823 }
1824
1825
1826 const char *
qualifiers_match(exec_list * params)1827 ir_function_signature::qualifiers_match(exec_list *params)
1828 {
1829 /* check that the qualifiers match. */
1830 foreach_two_lists(a_node, &this->parameters, b_node, params) {
1831 ir_variable *a = (ir_variable *) a_node;
1832 ir_variable *b = (ir_variable *) b_node;
1833
1834 if (a->data.read_only != b->data.read_only ||
1835 !modes_match(a->data.mode, b->data.mode) ||
1836 a->data.interpolation != b->data.interpolation ||
1837 a->data.centroid != b->data.centroid ||
1838 a->data.sample != b->data.sample ||
1839 a->data.patch != b->data.patch ||
1840 a->data.memory_read_only != b->data.memory_read_only ||
1841 a->data.memory_write_only != b->data.memory_write_only ||
1842 a->data.memory_coherent != b->data.memory_coherent ||
1843 a->data.memory_volatile != b->data.memory_volatile ||
1844 a->data.memory_restrict != b->data.memory_restrict) {
1845
1846 /* parameter a's qualifiers don't match */
1847 return a->name;
1848 }
1849 }
1850 return NULL;
1851 }
1852
1853
1854 void
replace_parameters(exec_list * new_params)1855 ir_function_signature::replace_parameters(exec_list *new_params)
1856 {
1857 /* Destroy all of the previous parameter information. If the previous
1858 * parameter information comes from the function prototype, it may either
1859 * specify incorrect parameter names or not have names at all.
1860 */
1861 new_params->move_nodes_to(¶meters);
1862 }
1863
1864
ir_function(const char * name)1865 ir_function::ir_function(const char *name)
1866 : ir_instruction(ir_type_function)
1867 {
1868 this->subroutine_index = -1;
1869 this->name = ralloc_strdup(this, name);
1870 }
1871
1872
1873 bool
has_user_signature()1874 ir_function::has_user_signature()
1875 {
1876 foreach_in_list(ir_function_signature, sig, &this->signatures) {
1877 if (!sig->is_builtin())
1878 return true;
1879 }
1880 return false;
1881 }
1882
1883
1884 ir_rvalue *
error_value(void * mem_ctx)1885 ir_rvalue::error_value(void *mem_ctx)
1886 {
1887 ir_rvalue *v = new(mem_ctx) ir_rvalue(ir_type_unset);
1888
1889 v->type = glsl_type::error_type;
1890 return v;
1891 }
1892
1893
1894 void
visit_exec_list(exec_list * list,ir_visitor * visitor)1895 visit_exec_list(exec_list *list, ir_visitor *visitor)
1896 {
1897 foreach_in_list_safe(ir_instruction, node, list) {
1898 node->accept(visitor);
1899 }
1900 }
1901
1902
1903 static void
steal_memory(ir_instruction * ir,void * new_ctx)1904 steal_memory(ir_instruction *ir, void *new_ctx)
1905 {
1906 ir_variable *var = ir->as_variable();
1907 ir_function *fn = ir->as_function();
1908 ir_constant *constant = ir->as_constant();
1909 if (var != NULL && var->constant_value != NULL)
1910 steal_memory(var->constant_value, ir);
1911
1912 if (var != NULL && var->constant_initializer != NULL)
1913 steal_memory(var->constant_initializer, ir);
1914
1915 if (fn != NULL && fn->subroutine_types)
1916 ralloc_steal(new_ctx, fn->subroutine_types);
1917
1918 /* The components of aggregate constants are not visited by the normal
1919 * visitor, so steal their values by hand.
1920 */
1921 if (constant != NULL &&
1922 (constant->type->is_array() || constant->type->is_record())) {
1923 for (unsigned int i = 0; i < constant->type->length; i++) {
1924 steal_memory(constant->const_elements[i], ir);
1925 }
1926 }
1927
1928 ralloc_steal(new_ctx, ir);
1929 }
1930
1931
1932 void
reparent_ir(exec_list * list,void * mem_ctx)1933 reparent_ir(exec_list *list, void *mem_ctx)
1934 {
1935 foreach_in_list(ir_instruction, node, list) {
1936 visit_tree(node, steal_memory, mem_ctx);
1937 }
1938 }
1939
1940
1941 static ir_rvalue *
try_min_one(ir_rvalue * ir)1942 try_min_one(ir_rvalue *ir)
1943 {
1944 ir_expression *expr = ir->as_expression();
1945
1946 if (!expr || expr->operation != ir_binop_min)
1947 return NULL;
1948
1949 if (expr->operands[0]->is_one())
1950 return expr->operands[1];
1951
1952 if (expr->operands[1]->is_one())
1953 return expr->operands[0];
1954
1955 return NULL;
1956 }
1957
1958 static ir_rvalue *
try_max_zero(ir_rvalue * ir)1959 try_max_zero(ir_rvalue *ir)
1960 {
1961 ir_expression *expr = ir->as_expression();
1962
1963 if (!expr || expr->operation != ir_binop_max)
1964 return NULL;
1965
1966 if (expr->operands[0]->is_zero())
1967 return expr->operands[1];
1968
1969 if (expr->operands[1]->is_zero())
1970 return expr->operands[0];
1971
1972 return NULL;
1973 }
1974
1975 ir_rvalue *
as_rvalue_to_saturate()1976 ir_rvalue::as_rvalue_to_saturate()
1977 {
1978 ir_expression *expr = this->as_expression();
1979
1980 if (!expr)
1981 return NULL;
1982
1983 ir_rvalue *max_zero = try_max_zero(expr);
1984 if (max_zero) {
1985 return try_min_one(max_zero);
1986 } else {
1987 ir_rvalue *min_one = try_min_one(expr);
1988 if (min_one) {
1989 return try_max_zero(min_one);
1990 }
1991 }
1992
1993 return NULL;
1994 }
1995
1996
1997 unsigned
vertices_per_prim(GLenum prim)1998 vertices_per_prim(GLenum prim)
1999 {
2000 switch (prim) {
2001 case GL_POINTS:
2002 return 1;
2003 case GL_LINES:
2004 return 2;
2005 case GL_TRIANGLES:
2006 return 3;
2007 case GL_LINES_ADJACENCY:
2008 return 4;
2009 case GL_TRIANGLES_ADJACENCY:
2010 return 6;
2011 default:
2012 assert(!"Bad primitive");
2013 return 3;
2014 }
2015 }
2016
2017 /**
2018 * Generate a string describing the mode of a variable
2019 */
2020 const char *
mode_string(const ir_variable * var)2021 mode_string(const ir_variable *var)
2022 {
2023 switch (var->data.mode) {
2024 case ir_var_auto:
2025 return (var->data.read_only) ? "global constant" : "global variable";
2026
2027 case ir_var_uniform:
2028 return "uniform";
2029
2030 case ir_var_shader_storage:
2031 return "buffer";
2032
2033 case ir_var_shader_in:
2034 return "shader input";
2035
2036 case ir_var_shader_out:
2037 return "shader output";
2038
2039 case ir_var_function_in:
2040 case ir_var_const_in:
2041 return "function input";
2042
2043 case ir_var_function_out:
2044 return "function output";
2045
2046 case ir_var_function_inout:
2047 return "function inout";
2048
2049 case ir_var_system_value:
2050 return "shader input";
2051
2052 case ir_var_temporary:
2053 return "compiler temporary";
2054
2055 case ir_var_mode_count:
2056 break;
2057 }
2058
2059 assert(!"Should not get here.");
2060 return "invalid variable";
2061 }
2062