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 this->type = glsl_type::int_type;
442 break;
443
444 case ir_unop_bitcast_i642d:
445 case ir_unop_bitcast_u642d:
446 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE,
447 op0->type->vector_elements, 1);
448 break;
449
450 case ir_unop_bitcast_d2i64:
451 this->type = glsl_type::get_instance(GLSL_TYPE_INT64,
452 op0->type->vector_elements, 1);
453 break;
454 case ir_unop_bitcast_d2u64:
455 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64,
456 op0->type->vector_elements, 1);
457 break;
458
459 default:
460 assert(!"not reached: missing automatic type setup for ir_expression");
461 this->type = op0->type;
462 break;
463 }
464 }
465
ir_expression(int op,ir_rvalue * op0,ir_rvalue * op1)466 ir_expression::ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1)
467 : ir_rvalue(ir_type_expression)
468 {
469 this->operation = ir_expression_operation(op);
470 this->operands[0] = op0;
471 this->operands[1] = op1;
472 this->operands[2] = NULL;
473 this->operands[3] = NULL;
474
475 assert(op > ir_last_unop);
476 init_num_operands();
477 assert(num_operands == 2);
478 for (unsigned i = 0; i < num_operands; i++) {
479 assert(this->operands[i] != NULL);
480 }
481
482 switch (this->operation) {
483 case ir_binop_all_equal:
484 case ir_binop_any_nequal:
485 this->type = glsl_type::bool_type;
486 break;
487
488 case ir_binop_add:
489 case ir_binop_sub:
490 case ir_binop_min:
491 case ir_binop_max:
492 case ir_binop_pow:
493 case ir_binop_mul:
494 case ir_binop_div:
495 case ir_binop_mod:
496 case ir_binop_atan2:
497 if (op0->type->is_scalar()) {
498 this->type = op1->type;
499 } else if (op1->type->is_scalar()) {
500 this->type = op0->type;
501 } else {
502 if (this->operation == ir_binop_mul) {
503 this->type = glsl_type::get_mul_type(op0->type, op1->type);
504 } else {
505 assert(op0->type == op1->type);
506 this->type = op0->type;
507 }
508 }
509 break;
510
511 case ir_binop_logic_and:
512 case ir_binop_logic_xor:
513 case ir_binop_logic_or:
514 case ir_binop_bit_and:
515 case ir_binop_bit_xor:
516 case ir_binop_bit_or:
517 assert(!op0->type->is_matrix());
518 assert(!op1->type->is_matrix());
519 if (op0->type->is_scalar()) {
520 this->type = op1->type;
521 } else if (op1->type->is_scalar()) {
522 this->type = op0->type;
523 } else {
524 assert(op0->type->vector_elements == op1->type->vector_elements);
525 this->type = op0->type;
526 }
527 break;
528
529 case ir_binop_equal:
530 case ir_binop_nequal:
531 case ir_binop_gequal:
532 case ir_binop_less:
533 assert(op0->type == op1->type);
534 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL,
535 op0->type->vector_elements, 1);
536 break;
537
538 case ir_binop_dot:
539 this->type = op0->type->get_base_type();
540 break;
541
542 case ir_binop_imul_high:
543 case ir_binop_mul_32x16:
544 case ir_binop_carry:
545 case ir_binop_borrow:
546 case ir_binop_lshift:
547 case ir_binop_rshift:
548 case ir_binop_ldexp:
549 case ir_binop_interpolate_at_offset:
550 case ir_binop_interpolate_at_sample:
551 this->type = op0->type;
552 break;
553
554 case ir_binop_add_sat:
555 case ir_binop_sub_sat:
556 case ir_binop_avg:
557 case ir_binop_avg_round:
558 assert(op0->type == op1->type);
559 this->type = op0->type;
560 break;
561
562 case ir_binop_abs_sub: {
563 enum glsl_base_type base;
564
565 assert(op0->type == op1->type);
566
567 switch (op0->type->base_type) {
568 case GLSL_TYPE_UINT:
569 case GLSL_TYPE_INT:
570 base = GLSL_TYPE_UINT;
571 break;
572 case GLSL_TYPE_UINT8:
573 case GLSL_TYPE_INT8:
574 base = GLSL_TYPE_UINT8;
575 break;
576 case GLSL_TYPE_UINT16:
577 case GLSL_TYPE_INT16:
578 base = GLSL_TYPE_UINT16;
579 break;
580 case GLSL_TYPE_UINT64:
581 case GLSL_TYPE_INT64:
582 base = GLSL_TYPE_UINT64;
583 break;
584 default:
585 unreachable(!"Invalid base type.");
586 }
587
588 this->type = glsl_type::get_instance(base, op0->type->vector_elements, 1);
589 break;
590 }
591
592 case ir_binop_vector_extract:
593 this->type = op0->type->get_scalar_type();
594 break;
595
596 default:
597 assert(!"not reached: missing automatic type setup for ir_expression");
598 this->type = glsl_type::float_type;
599 }
600 }
601
ir_expression(int op,ir_rvalue * op0,ir_rvalue * op1,ir_rvalue * op2)602 ir_expression::ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1,
603 ir_rvalue *op2)
604 : ir_rvalue(ir_type_expression)
605 {
606 this->operation = ir_expression_operation(op);
607 this->operands[0] = op0;
608 this->operands[1] = op1;
609 this->operands[2] = op2;
610 this->operands[3] = NULL;
611
612 assert(op > ir_last_binop && op <= ir_last_triop);
613 init_num_operands();
614 assert(num_operands == 3);
615 for (unsigned i = 0; i < num_operands; i++) {
616 assert(this->operands[i] != NULL);
617 }
618
619 switch (this->operation) {
620 case ir_triop_fma:
621 case ir_triop_lrp:
622 case ir_triop_bitfield_extract:
623 case ir_triop_vector_insert:
624 this->type = op0->type;
625 break;
626
627 case ir_triop_csel:
628 this->type = op1->type;
629 break;
630
631 default:
632 assert(!"not reached: missing automatic type setup for ir_expression");
633 this->type = glsl_type::float_type;
634 }
635 }
636
637 /**
638 * This is only here for ir_reader to used for testing purposes. Please use
639 * the precomputed num_operands field if you need the number of operands.
640 */
641 unsigned
get_num_operands(ir_expression_operation op)642 ir_expression::get_num_operands(ir_expression_operation op)
643 {
644 assert(op <= ir_last_opcode);
645
646 if (op <= ir_last_unop)
647 return 1;
648
649 if (op <= ir_last_binop)
650 return 2;
651
652 if (op <= ir_last_triop)
653 return 3;
654
655 if (op <= ir_last_quadop)
656 return 4;
657
658 unreachable("Could not calculate number of operands");
659 }
660
661 #include "ir_expression_operation_strings.h"
662
663 const char*
depth_layout_string(ir_depth_layout layout)664 depth_layout_string(ir_depth_layout layout)
665 {
666 switch(layout) {
667 case ir_depth_layout_none: return "";
668 case ir_depth_layout_any: return "depth_any";
669 case ir_depth_layout_greater: return "depth_greater";
670 case ir_depth_layout_less: return "depth_less";
671 case ir_depth_layout_unchanged: return "depth_unchanged";
672
673 default:
674 assert(0);
675 return "";
676 }
677 }
678
679 ir_expression_operation
get_operator(const char * str)680 ir_expression::get_operator(const char *str)
681 {
682 for (int op = 0; op <= int(ir_last_opcode); op++) {
683 if (strcmp(str, ir_expression_operation_strings[op]) == 0)
684 return (ir_expression_operation) op;
685 }
686 return (ir_expression_operation) -1;
687 }
688
689 ir_variable *
variable_referenced() const690 ir_expression::variable_referenced() const
691 {
692 switch (operation) {
693 case ir_binop_vector_extract:
694 case ir_triop_vector_insert:
695 /* We get these for things like a[0] where a is a vector type. In these
696 * cases we want variable_referenced() to return the actual vector
697 * variable this is wrapping.
698 */
699 return operands[0]->variable_referenced();
700 default:
701 return ir_rvalue::variable_referenced();
702 }
703 }
704
ir_constant()705 ir_constant::ir_constant()
706 : ir_rvalue(ir_type_constant)
707 {
708 this->const_elements = NULL;
709 }
710
ir_constant(const struct glsl_type * type,const ir_constant_data * data)711 ir_constant::ir_constant(const struct glsl_type *type,
712 const ir_constant_data *data)
713 : ir_rvalue(ir_type_constant)
714 {
715 this->const_elements = NULL;
716
717 assert((type->base_type >= GLSL_TYPE_UINT)
718 && (type->base_type <= GLSL_TYPE_IMAGE));
719
720 this->type = type;
721 memcpy(& this->value, data, sizeof(this->value));
722 }
723
ir_constant(float16_t f16,unsigned vector_elements)724 ir_constant::ir_constant(float16_t f16, unsigned vector_elements)
725 : ir_rvalue(ir_type_constant)
726 {
727 this->const_elements = NULL;
728 assert(vector_elements <= 4);
729 this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT16, vector_elements, 1);
730 for (unsigned i = 0; i < vector_elements; i++) {
731 this->value.f16[i] = f16.bits;
732 }
733 for (unsigned i = vector_elements; i < 16; i++) {
734 this->value.f[i] = 0;
735 }
736 }
737
ir_constant(float f,unsigned vector_elements)738 ir_constant::ir_constant(float f, unsigned vector_elements)
739 : ir_rvalue(ir_type_constant)
740 {
741 this->const_elements = NULL;
742 assert(vector_elements <= 4);
743 this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT, vector_elements, 1);
744 for (unsigned i = 0; i < vector_elements; i++) {
745 this->value.f[i] = f;
746 }
747 for (unsigned i = vector_elements; i < 16; i++) {
748 this->value.f[i] = 0;
749 }
750 }
751
ir_constant(double d,unsigned vector_elements)752 ir_constant::ir_constant(double d, unsigned vector_elements)
753 : ir_rvalue(ir_type_constant)
754 {
755 this->const_elements = NULL;
756 assert(vector_elements <= 4);
757 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE, vector_elements, 1);
758 for (unsigned i = 0; i < vector_elements; i++) {
759 this->value.d[i] = d;
760 }
761 for (unsigned i = vector_elements; i < 16; i++) {
762 this->value.d[i] = 0.0;
763 }
764 }
765
ir_constant(int16_t i16,unsigned vector_elements)766 ir_constant::ir_constant(int16_t i16, unsigned vector_elements)
767 : ir_rvalue(ir_type_constant)
768 {
769 this->const_elements = NULL;
770 assert(vector_elements <= 4);
771 this->type = glsl_type::get_instance(GLSL_TYPE_INT16, vector_elements, 1);
772 for (unsigned i = 0; i < vector_elements; i++) {
773 this->value.i16[i] = i16;
774 }
775 for (unsigned i = vector_elements; i < 16; i++) {
776 this->value.i16[i] = 0;
777 }
778 }
779
ir_constant(uint16_t u16,unsigned vector_elements)780 ir_constant::ir_constant(uint16_t u16, unsigned vector_elements)
781 : ir_rvalue(ir_type_constant)
782 {
783 this->const_elements = NULL;
784 assert(vector_elements <= 4);
785 this->type = glsl_type::get_instance(GLSL_TYPE_UINT16, vector_elements, 1);
786 for (unsigned i = 0; i < vector_elements; i++) {
787 this->value.u16[i] = u16;
788 }
789 for (unsigned i = vector_elements; i < 16; i++) {
790 this->value.u16[i] = 0;
791 }
792 }
793
ir_constant(unsigned int u,unsigned vector_elements)794 ir_constant::ir_constant(unsigned int u, unsigned vector_elements)
795 : ir_rvalue(ir_type_constant)
796 {
797 this->const_elements = NULL;
798 assert(vector_elements <= 4);
799 this->type = glsl_type::get_instance(GLSL_TYPE_UINT, vector_elements, 1);
800 for (unsigned i = 0; i < vector_elements; i++) {
801 this->value.u[i] = u;
802 }
803 for (unsigned i = vector_elements; i < 16; i++) {
804 this->value.u[i] = 0;
805 }
806 }
807
ir_constant(int integer,unsigned vector_elements)808 ir_constant::ir_constant(int integer, unsigned vector_elements)
809 : ir_rvalue(ir_type_constant)
810 {
811 this->const_elements = NULL;
812 assert(vector_elements <= 4);
813 this->type = glsl_type::get_instance(GLSL_TYPE_INT, vector_elements, 1);
814 for (unsigned i = 0; i < vector_elements; i++) {
815 this->value.i[i] = integer;
816 }
817 for (unsigned i = vector_elements; i < 16; i++) {
818 this->value.i[i] = 0;
819 }
820 }
821
ir_constant(uint64_t u64,unsigned vector_elements)822 ir_constant::ir_constant(uint64_t u64, unsigned vector_elements)
823 : ir_rvalue(ir_type_constant)
824 {
825 this->const_elements = NULL;
826 assert(vector_elements <= 4);
827 this->type = glsl_type::get_instance(GLSL_TYPE_UINT64, vector_elements, 1);
828 for (unsigned i = 0; i < vector_elements; i++) {
829 this->value.u64[i] = u64;
830 }
831 for (unsigned i = vector_elements; i < 16; i++) {
832 this->value.u64[i] = 0;
833 }
834 }
835
ir_constant(int64_t int64,unsigned vector_elements)836 ir_constant::ir_constant(int64_t int64, unsigned vector_elements)
837 : ir_rvalue(ir_type_constant)
838 {
839 this->const_elements = NULL;
840 assert(vector_elements <= 4);
841 this->type = glsl_type::get_instance(GLSL_TYPE_INT64, vector_elements, 1);
842 for (unsigned i = 0; i < vector_elements; i++) {
843 this->value.i64[i] = int64;
844 }
845 for (unsigned i = vector_elements; i < 16; i++) {
846 this->value.i64[i] = 0;
847 }
848 }
849
ir_constant(bool b,unsigned vector_elements)850 ir_constant::ir_constant(bool b, unsigned vector_elements)
851 : ir_rvalue(ir_type_constant)
852 {
853 this->const_elements = NULL;
854 assert(vector_elements <= 4);
855 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL, vector_elements, 1);
856 for (unsigned i = 0; i < vector_elements; i++) {
857 this->value.b[i] = b;
858 }
859 for (unsigned i = vector_elements; i < 16; i++) {
860 this->value.b[i] = false;
861 }
862 }
863
ir_constant(const ir_constant * c,unsigned i)864 ir_constant::ir_constant(const ir_constant *c, unsigned i)
865 : ir_rvalue(ir_type_constant)
866 {
867 this->const_elements = NULL;
868 this->type = c->type->get_base_type();
869
870 /* Section 5.11 (Out-of-Bounds Accesses) of the GLSL 4.60 spec says:
871 *
872 * In the subsections described above for array, vector, matrix and
873 * structure accesses, any out-of-bounds access produced undefined
874 * behavior....Out-of-bounds reads return undefined values, which
875 * include values from other variables of the active program or zero.
876 *
877 * GL_KHR_robustness and GL_ARB_robustness encourage us to return zero.
878 */
879 if (i >= c->type->vector_elements) {
880 this->value = { { 0 } };
881 return;
882 }
883
884 switch (this->type->base_type) {
885 case GLSL_TYPE_UINT16: this->value.u16[0] = c->value.u16[i]; break;
886 case GLSL_TYPE_INT16: this->value.i16[0] = c->value.i16[i]; break;
887 case GLSL_TYPE_UINT: this->value.u[0] = c->value.u[i]; break;
888 case GLSL_TYPE_INT: this->value.i[0] = c->value.i[i]; break;
889 case GLSL_TYPE_FLOAT: this->value.f[0] = c->value.f[i]; break;
890 case GLSL_TYPE_FLOAT16: this->value.f16[0] = c->value.f16[i]; break;
891 case GLSL_TYPE_BOOL: this->value.b[0] = c->value.b[i]; break;
892 case GLSL_TYPE_DOUBLE: this->value.d[0] = c->value.d[i]; break;
893 default: assert(!"Should not get here."); break;
894 }
895 }
896
ir_constant(const struct glsl_type * type,exec_list * value_list)897 ir_constant::ir_constant(const struct glsl_type *type, exec_list *value_list)
898 : ir_rvalue(ir_type_constant)
899 {
900 this->const_elements = NULL;
901 this->type = type;
902
903 assert(type->is_scalar() || type->is_vector() || type->is_matrix()
904 || type->is_struct() || type->is_array());
905
906 /* If the constant is a record, the types of each of the entries in
907 * value_list must be a 1-for-1 match with the structure components. Each
908 * entry must also be a constant. Just move the nodes from the value_list
909 * to the list in the ir_constant.
910 */
911 if (type->is_array() || type->is_struct()) {
912 this->const_elements = ralloc_array(this, ir_constant *, type->length);
913 unsigned i = 0;
914 foreach_in_list(ir_constant, value, value_list) {
915 assert(value->as_constant() != NULL);
916
917 this->const_elements[i++] = value;
918 }
919 return;
920 }
921
922 for (unsigned i = 0; i < 16; i++) {
923 this->value.u[i] = 0;
924 }
925
926 ir_constant *value = (ir_constant *) (value_list->get_head_raw());
927
928 /* Constructors with exactly one scalar argument are special for vectors
929 * and matrices. For vectors, the scalar value is replicated to fill all
930 * the components. For matrices, the scalar fills the components of the
931 * diagonal while the rest is filled with 0.
932 */
933 if (value->type->is_scalar() && value->next->is_tail_sentinel()) {
934 if (type->is_matrix()) {
935 /* Matrix - fill diagonal (rest is already set to 0) */
936 for (unsigned i = 0; i < type->matrix_columns; i++) {
937 switch (type->base_type) {
938 case GLSL_TYPE_FLOAT:
939 this->value.f[i * type->vector_elements + i] =
940 value->value.f[0];
941 break;
942 case GLSL_TYPE_DOUBLE:
943 this->value.d[i * type->vector_elements + i] =
944 value->value.d[0];
945 break;
946 case GLSL_TYPE_FLOAT16:
947 this->value.f16[i * type->vector_elements + i] =
948 value->value.f16[0];
949 break;
950 default:
951 assert(!"unexpected matrix base type");
952 }
953 }
954 } else {
955 /* Vector or scalar - fill all components */
956 switch (type->base_type) {
957 case GLSL_TYPE_UINT16:
958 case GLSL_TYPE_INT16:
959 for (unsigned i = 0; i < type->components(); i++)
960 this->value.u16[i] = value->value.u16[0];
961 break;
962 case GLSL_TYPE_UINT:
963 case GLSL_TYPE_INT:
964 for (unsigned i = 0; i < type->components(); i++)
965 this->value.u[i] = value->value.u[0];
966 break;
967 case GLSL_TYPE_FLOAT:
968 for (unsigned i = 0; i < type->components(); i++)
969 this->value.f[i] = value->value.f[0];
970 break;
971 case GLSL_TYPE_FLOAT16:
972 for (unsigned i = 0; i < type->components(); i++)
973 this->value.f16[i] = value->value.f16[0];
974 break;
975 case GLSL_TYPE_DOUBLE:
976 for (unsigned i = 0; i < type->components(); i++)
977 this->value.d[i] = value->value.d[0];
978 break;
979 case GLSL_TYPE_UINT64:
980 case GLSL_TYPE_INT64:
981 for (unsigned i = 0; i < type->components(); i++)
982 this->value.u64[i] = value->value.u64[0];
983 break;
984 case GLSL_TYPE_BOOL:
985 for (unsigned i = 0; i < type->components(); i++)
986 this->value.b[i] = value->value.b[0];
987 break;
988 case GLSL_TYPE_SAMPLER:
989 case GLSL_TYPE_IMAGE:
990 this->value.u64[0] = value->value.u64[0];
991 break;
992 default:
993 assert(!"Should not get here.");
994 break;
995 }
996 }
997 return;
998 }
999
1000 if (type->is_matrix() && value->type->is_matrix()) {
1001 assert(value->next->is_tail_sentinel());
1002
1003 /* From section 5.4.2 of the GLSL 1.20 spec:
1004 * "If a matrix is constructed from a matrix, then each component
1005 * (column i, row j) in the result that has a corresponding component
1006 * (column i, row j) in the argument will be initialized from there."
1007 */
1008 unsigned cols = MIN2(type->matrix_columns, value->type->matrix_columns);
1009 unsigned rows = MIN2(type->vector_elements, value->type->vector_elements);
1010 for (unsigned i = 0; i < cols; i++) {
1011 for (unsigned j = 0; j < rows; j++) {
1012 const unsigned src = i * value->type->vector_elements + j;
1013 const unsigned dst = i * type->vector_elements + j;
1014 this->value.f[dst] = value->value.f[src];
1015 }
1016 }
1017
1018 /* "All other components will be initialized to the identity matrix." */
1019 for (unsigned i = cols; i < type->matrix_columns; i++)
1020 this->value.f[i * type->vector_elements + i] = 1.0;
1021
1022 return;
1023 }
1024
1025 /* Use each component from each entry in the value_list to initialize one
1026 * component of the constant being constructed.
1027 */
1028 unsigned i = 0;
1029 for (;;) {
1030 assert(value->as_constant() != NULL);
1031 assert(!value->is_tail_sentinel());
1032
1033 for (unsigned j = 0; j < value->type->components(); j++) {
1034 switch (type->base_type) {
1035 case GLSL_TYPE_UINT16:
1036 this->value.u16[i] = value->get_uint16_component(j);
1037 break;
1038 case GLSL_TYPE_INT16:
1039 this->value.i16[i] = value->get_int16_component(j);
1040 break;
1041 case GLSL_TYPE_UINT:
1042 this->value.u[i] = value->get_uint_component(j);
1043 break;
1044 case GLSL_TYPE_INT:
1045 this->value.i[i] = value->get_int_component(j);
1046 break;
1047 case GLSL_TYPE_FLOAT:
1048 this->value.f[i] = value->get_float_component(j);
1049 break;
1050 case GLSL_TYPE_FLOAT16:
1051 this->value.f16[i] = value->get_float16_component(j);
1052 break;
1053 case GLSL_TYPE_BOOL:
1054 this->value.b[i] = value->get_bool_component(j);
1055 break;
1056 case GLSL_TYPE_DOUBLE:
1057 this->value.d[i] = value->get_double_component(j);
1058 break;
1059 case GLSL_TYPE_UINT64:
1060 this->value.u64[i] = value->get_uint64_component(j);
1061 break;
1062 case GLSL_TYPE_INT64:
1063 this->value.i64[i] = value->get_int64_component(j);
1064 break;
1065 default:
1066 /* FINISHME: What to do? Exceptions are not the answer.
1067 */
1068 break;
1069 }
1070
1071 i++;
1072 if (i >= type->components())
1073 break;
1074 }
1075
1076 if (i >= type->components())
1077 break; /* avoid downcasting a list sentinel */
1078 value = (ir_constant *) value->next;
1079 }
1080 }
1081
1082 ir_constant *
zero(void * mem_ctx,const glsl_type * type)1083 ir_constant::zero(void *mem_ctx, const glsl_type *type)
1084 {
1085 assert(type->is_scalar() || type->is_vector() || type->is_matrix()
1086 || type->is_struct() || type->is_array());
1087
1088 ir_constant *c = new(mem_ctx) ir_constant;
1089 c->type = type;
1090 memset(&c->value, 0, sizeof(c->value));
1091
1092 if (type->is_array()) {
1093 c->const_elements = ralloc_array(c, ir_constant *, type->length);
1094
1095 for (unsigned i = 0; i < type->length; i++)
1096 c->const_elements[i] = ir_constant::zero(c, type->fields.array);
1097 }
1098
1099 if (type->is_struct()) {
1100 c->const_elements = ralloc_array(c, ir_constant *, type->length);
1101
1102 for (unsigned i = 0; i < type->length; i++) {
1103 c->const_elements[i] =
1104 ir_constant::zero(mem_ctx, type->fields.structure[i].type);
1105 }
1106 }
1107
1108 return c;
1109 }
1110
1111 bool
get_bool_component(unsigned i) const1112 ir_constant::get_bool_component(unsigned i) const
1113 {
1114 switch (this->type->base_type) {
1115 case GLSL_TYPE_UINT16:return this->value.u16[i] != 0;
1116 case GLSL_TYPE_INT16: return this->value.i16[i] != 0;
1117 case GLSL_TYPE_UINT: return this->value.u[i] != 0;
1118 case GLSL_TYPE_INT: return this->value.i[i] != 0;
1119 case GLSL_TYPE_FLOAT: return ((int)this->value.f[i]) != 0;
1120 case GLSL_TYPE_FLOAT16: return ((int)_mesa_half_to_float(this->value.f16[i])) != 0;
1121 case GLSL_TYPE_BOOL: return this->value.b[i];
1122 case GLSL_TYPE_DOUBLE: return this->value.d[i] != 0.0;
1123 case GLSL_TYPE_SAMPLER:
1124 case GLSL_TYPE_IMAGE:
1125 case GLSL_TYPE_UINT64: return this->value.u64[i] != 0;
1126 case GLSL_TYPE_INT64: return this->value.i64[i] != 0;
1127 default: assert(!"Should not get here."); break;
1128 }
1129
1130 /* Must return something to make the compiler happy. This is clearly an
1131 * error case.
1132 */
1133 return false;
1134 }
1135
1136 float
get_float_component(unsigned i) const1137 ir_constant::get_float_component(unsigned i) const
1138 {
1139 switch (this->type->base_type) {
1140 case GLSL_TYPE_UINT16:return (float) this->value.u16[i];
1141 case GLSL_TYPE_INT16: return (float) this->value.i16[i];
1142 case GLSL_TYPE_UINT: return (float) this->value.u[i];
1143 case GLSL_TYPE_INT: return (float) this->value.i[i];
1144 case GLSL_TYPE_FLOAT: return this->value.f[i];
1145 case GLSL_TYPE_FLOAT16: return _mesa_half_to_float(this->value.f16[i]);
1146 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1.0f : 0.0f;
1147 case GLSL_TYPE_DOUBLE: return (float) this->value.d[i];
1148 case GLSL_TYPE_SAMPLER:
1149 case GLSL_TYPE_IMAGE:
1150 case GLSL_TYPE_UINT64: return (float) this->value.u64[i];
1151 case GLSL_TYPE_INT64: return (float) this->value.i64[i];
1152 default: assert(!"Should not get here."); break;
1153 }
1154
1155 /* Must return something to make the compiler happy. This is clearly an
1156 * error case.
1157 */
1158 return 0.0;
1159 }
1160
1161 uint16_t
get_float16_component(unsigned i) const1162 ir_constant::get_float16_component(unsigned i) const
1163 {
1164 if (this->type->base_type == GLSL_TYPE_FLOAT16)
1165 return this->value.f16[i];
1166 else
1167 return _mesa_float_to_half(get_float_component(i));
1168 }
1169
1170 double
get_double_component(unsigned i) const1171 ir_constant::get_double_component(unsigned i) const
1172 {
1173 switch (this->type->base_type) {
1174 case GLSL_TYPE_UINT16:return (double) this->value.u16[i];
1175 case GLSL_TYPE_INT16: return (double) this->value.i16[i];
1176 case GLSL_TYPE_UINT: return (double) this->value.u[i];
1177 case GLSL_TYPE_INT: return (double) this->value.i[i];
1178 case GLSL_TYPE_FLOAT: return (double) this->value.f[i];
1179 case GLSL_TYPE_FLOAT16: return (double) _mesa_half_to_float(this->value.f16[i]);
1180 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1.0 : 0.0;
1181 case GLSL_TYPE_DOUBLE: return this->value.d[i];
1182 case GLSL_TYPE_SAMPLER:
1183 case GLSL_TYPE_IMAGE:
1184 case GLSL_TYPE_UINT64: return (double) this->value.u64[i];
1185 case GLSL_TYPE_INT64: return (double) this->value.i64[i];
1186 default: assert(!"Should not get here."); break;
1187 }
1188
1189 /* Must return something to make the compiler happy. This is clearly an
1190 * error case.
1191 */
1192 return 0.0;
1193 }
1194
1195 int16_t
get_int16_component(unsigned i) const1196 ir_constant::get_int16_component(unsigned i) const
1197 {
1198 switch (this->type->base_type) {
1199 case GLSL_TYPE_UINT16:return this->value.u16[i];
1200 case GLSL_TYPE_INT16: return this->value.i16[i];
1201 case GLSL_TYPE_UINT: return this->value.u[i];
1202 case GLSL_TYPE_INT: return this->value.i[i];
1203 case GLSL_TYPE_FLOAT: return (int16_t) this->value.f[i];
1204 case GLSL_TYPE_FLOAT16: return (int16_t) _mesa_half_to_float(this->value.f16[i]);
1205 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1206 case GLSL_TYPE_DOUBLE: return (int16_t) this->value.d[i];
1207 case GLSL_TYPE_SAMPLER:
1208 case GLSL_TYPE_IMAGE:
1209 case GLSL_TYPE_UINT64: return (int16_t) this->value.u64[i];
1210 case GLSL_TYPE_INT64: return (int16_t) this->value.i64[i];
1211 default: assert(!"Should not get here."); break;
1212 }
1213
1214 /* Must return something to make the compiler happy. This is clearly an
1215 * error case.
1216 */
1217 return 0;
1218 }
1219
1220 uint16_t
get_uint16_component(unsigned i) const1221 ir_constant::get_uint16_component(unsigned i) const
1222 {
1223 switch (this->type->base_type) {
1224 case GLSL_TYPE_UINT16:return this->value.u16[i];
1225 case GLSL_TYPE_INT16: return this->value.i16[i];
1226 case GLSL_TYPE_UINT: return this->value.u[i];
1227 case GLSL_TYPE_INT: return this->value.i[i];
1228 case GLSL_TYPE_FLOAT: return (uint16_t) this->value.f[i];
1229 case GLSL_TYPE_FLOAT16: return (uint16_t) _mesa_half_to_float(this->value.f16[i]);
1230 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1231 case GLSL_TYPE_DOUBLE: return (uint16_t) this->value.d[i];
1232 case GLSL_TYPE_SAMPLER:
1233 case GLSL_TYPE_IMAGE:
1234 case GLSL_TYPE_UINT64: return (uint16_t) this->value.u64[i];
1235 case GLSL_TYPE_INT64: return (uint16_t) this->value.i64[i];
1236 default: assert(!"Should not get here."); break;
1237 }
1238
1239 /* Must return something to make the compiler happy. This is clearly an
1240 * error case.
1241 */
1242 return 0;
1243 }
1244
1245 int
get_int_component(unsigned i) const1246 ir_constant::get_int_component(unsigned i) const
1247 {
1248 switch (this->type->base_type) {
1249 case GLSL_TYPE_UINT16:return this->value.u16[i];
1250 case GLSL_TYPE_INT16: return this->value.i16[i];
1251 case GLSL_TYPE_UINT: return this->value.u[i];
1252 case GLSL_TYPE_INT: return this->value.i[i];
1253 case GLSL_TYPE_FLOAT: return (int) this->value.f[i];
1254 case GLSL_TYPE_FLOAT16: return (int) _mesa_half_to_float(this->value.f16[i]);
1255 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1256 case GLSL_TYPE_DOUBLE: return (int) this->value.d[i];
1257 case GLSL_TYPE_SAMPLER:
1258 case GLSL_TYPE_IMAGE:
1259 case GLSL_TYPE_UINT64: return (int) this->value.u64[i];
1260 case GLSL_TYPE_INT64: return (int) this->value.i64[i];
1261 default: assert(!"Should not get here."); break;
1262 }
1263
1264 /* Must return something to make the compiler happy. This is clearly an
1265 * error case.
1266 */
1267 return 0;
1268 }
1269
1270 unsigned
get_uint_component(unsigned i) const1271 ir_constant::get_uint_component(unsigned i) const
1272 {
1273 switch (this->type->base_type) {
1274 case GLSL_TYPE_UINT16:return this->value.u16[i];
1275 case GLSL_TYPE_INT16: return this->value.i16[i];
1276 case GLSL_TYPE_UINT: return this->value.u[i];
1277 case GLSL_TYPE_INT: return this->value.i[i];
1278 case GLSL_TYPE_FLOAT: return (unsigned) this->value.f[i];
1279 case GLSL_TYPE_FLOAT16: return (unsigned) _mesa_half_to_float(this->value.f16[i]);
1280 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1281 case GLSL_TYPE_DOUBLE: return (unsigned) this->value.d[i];
1282 case GLSL_TYPE_SAMPLER:
1283 case GLSL_TYPE_IMAGE:
1284 case GLSL_TYPE_UINT64: return (unsigned) this->value.u64[i];
1285 case GLSL_TYPE_INT64: return (unsigned) this->value.i64[i];
1286 default: assert(!"Should not get here."); break;
1287 }
1288
1289 /* Must return something to make the compiler happy. This is clearly an
1290 * error case.
1291 */
1292 return 0;
1293 }
1294
1295 int64_t
get_int64_component(unsigned i) const1296 ir_constant::get_int64_component(unsigned i) const
1297 {
1298 switch (this->type->base_type) {
1299 case GLSL_TYPE_UINT16:return this->value.u16[i];
1300 case GLSL_TYPE_INT16: return this->value.i16[i];
1301 case GLSL_TYPE_UINT: return this->value.u[i];
1302 case GLSL_TYPE_INT: return this->value.i[i];
1303 case GLSL_TYPE_FLOAT: return (int64_t) this->value.f[i];
1304 case GLSL_TYPE_FLOAT16: return (int64_t) _mesa_half_to_float(this->value.f16[i]);
1305 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1306 case GLSL_TYPE_DOUBLE: return (int64_t) this->value.d[i];
1307 case GLSL_TYPE_SAMPLER:
1308 case GLSL_TYPE_IMAGE:
1309 case GLSL_TYPE_UINT64: return (int64_t) this->value.u64[i];
1310 case GLSL_TYPE_INT64: return this->value.i64[i];
1311 default: assert(!"Should not get here."); break;
1312 }
1313
1314 /* Must return something to make the compiler happy. This is clearly an
1315 * error case.
1316 */
1317 return 0;
1318 }
1319
1320 uint64_t
get_uint64_component(unsigned i) const1321 ir_constant::get_uint64_component(unsigned i) const
1322 {
1323 switch (this->type->base_type) {
1324 case GLSL_TYPE_UINT16:return this->value.u16[i];
1325 case GLSL_TYPE_INT16: return this->value.i16[i];
1326 case GLSL_TYPE_UINT: return this->value.u[i];
1327 case GLSL_TYPE_INT: return this->value.i[i];
1328 case GLSL_TYPE_FLOAT: return (uint64_t) this->value.f[i];
1329 case GLSL_TYPE_FLOAT16: return (uint64_t) _mesa_half_to_float(this->value.f16[i]);
1330 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
1331 case GLSL_TYPE_DOUBLE: return (uint64_t) this->value.d[i];
1332 case GLSL_TYPE_SAMPLER:
1333 case GLSL_TYPE_IMAGE:
1334 case GLSL_TYPE_UINT64: return this->value.u64[i];
1335 case GLSL_TYPE_INT64: return (uint64_t) this->value.i64[i];
1336 default: assert(!"Should not get here."); break;
1337 }
1338
1339 /* Must return something to make the compiler happy. This is clearly an
1340 * error case.
1341 */
1342 return 0;
1343 }
1344
1345 ir_constant *
get_array_element(unsigned i) const1346 ir_constant::get_array_element(unsigned i) const
1347 {
1348 assert(this->type->is_array());
1349
1350 /* From page 35 (page 41 of the PDF) of the GLSL 1.20 spec:
1351 *
1352 * "Behavior is undefined if a shader subscripts an array with an index
1353 * less than 0 or greater than or equal to the size the array was
1354 * declared with."
1355 *
1356 * Most out-of-bounds accesses are removed before things could get this far.
1357 * There are cases where non-constant array index values can get constant
1358 * folded.
1359 */
1360 if (int(i) < 0)
1361 i = 0;
1362 else if (i >= this->type->length)
1363 i = this->type->length - 1;
1364
1365 return const_elements[i];
1366 }
1367
1368 ir_constant *
get_record_field(int idx)1369 ir_constant::get_record_field(int idx)
1370 {
1371 assert(this->type->is_struct());
1372 assert(idx >= 0 && (unsigned) idx < this->type->length);
1373
1374 return const_elements[idx];
1375 }
1376
1377 void
copy_offset(ir_constant * src,int offset)1378 ir_constant::copy_offset(ir_constant *src, int offset)
1379 {
1380 switch (this->type->base_type) {
1381 case GLSL_TYPE_UINT16:
1382 case GLSL_TYPE_INT16:
1383 case GLSL_TYPE_UINT:
1384 case GLSL_TYPE_INT:
1385 case GLSL_TYPE_FLOAT:
1386 case GLSL_TYPE_FLOAT16:
1387 case GLSL_TYPE_DOUBLE:
1388 case GLSL_TYPE_SAMPLER:
1389 case GLSL_TYPE_IMAGE:
1390 case GLSL_TYPE_UINT64:
1391 case GLSL_TYPE_INT64:
1392 case GLSL_TYPE_BOOL: {
1393 unsigned int size = src->type->components();
1394 assert (size <= this->type->components() - offset);
1395 for (unsigned int i=0; i<size; i++) {
1396 switch (this->type->base_type) {
1397 case GLSL_TYPE_UINT16:
1398 value.u16[i+offset] = src->get_uint16_component(i);
1399 break;
1400 case GLSL_TYPE_INT16:
1401 value.i16[i+offset] = src->get_int16_component(i);
1402 break;
1403 case GLSL_TYPE_UINT:
1404 value.u[i+offset] = src->get_uint_component(i);
1405 break;
1406 case GLSL_TYPE_INT:
1407 value.i[i+offset] = src->get_int_component(i);
1408 break;
1409 case GLSL_TYPE_FLOAT:
1410 value.f[i+offset] = src->get_float_component(i);
1411 break;
1412 case GLSL_TYPE_FLOAT16:
1413 value.f16[i+offset] = src->get_float16_component(i);
1414 break;
1415 case GLSL_TYPE_BOOL:
1416 value.b[i+offset] = src->get_bool_component(i);
1417 break;
1418 case GLSL_TYPE_DOUBLE:
1419 value.d[i+offset] = src->get_double_component(i);
1420 break;
1421 case GLSL_TYPE_SAMPLER:
1422 case GLSL_TYPE_IMAGE:
1423 case GLSL_TYPE_UINT64:
1424 value.u64[i+offset] = src->get_uint64_component(i);
1425 break;
1426 case GLSL_TYPE_INT64:
1427 value.i64[i+offset] = src->get_int64_component(i);
1428 break;
1429 default: // Shut up the compiler
1430 break;
1431 }
1432 }
1433 break;
1434 }
1435
1436 case GLSL_TYPE_STRUCT:
1437 case GLSL_TYPE_ARRAY: {
1438 assert (src->type == this->type);
1439 for (unsigned i = 0; i < this->type->length; i++) {
1440 this->const_elements[i] = src->const_elements[i]->clone(this, NULL);
1441 }
1442 break;
1443 }
1444
1445 default:
1446 assert(!"Should not get here.");
1447 break;
1448 }
1449 }
1450
1451 void
copy_masked_offset(ir_constant * src,int offset,unsigned int mask)1452 ir_constant::copy_masked_offset(ir_constant *src, int offset, unsigned int mask)
1453 {
1454 assert (!type->is_array() && !type->is_struct());
1455
1456 if (!type->is_vector() && !type->is_matrix()) {
1457 offset = 0;
1458 mask = 1;
1459 }
1460
1461 int id = 0;
1462 for (int i=0; i<4; i++) {
1463 if (mask & (1 << i)) {
1464 switch (this->type->base_type) {
1465 case GLSL_TYPE_UINT16:
1466 value.u16[i+offset] = src->get_uint16_component(id++);
1467 break;
1468 case GLSL_TYPE_INT16:
1469 value.i16[i+offset] = src->get_int16_component(id++);
1470 break;
1471 case GLSL_TYPE_UINT:
1472 value.u[i+offset] = src->get_uint_component(id++);
1473 break;
1474 case GLSL_TYPE_INT:
1475 value.i[i+offset] = src->get_int_component(id++);
1476 break;
1477 case GLSL_TYPE_FLOAT:
1478 value.f[i+offset] = src->get_float_component(id++);
1479 break;
1480 case GLSL_TYPE_FLOAT16:
1481 value.f16[i+offset] = src->get_float16_component(id++);
1482 break;
1483 case GLSL_TYPE_BOOL:
1484 value.b[i+offset] = src->get_bool_component(id++);
1485 break;
1486 case GLSL_TYPE_DOUBLE:
1487 value.d[i+offset] = src->get_double_component(id++);
1488 break;
1489 case GLSL_TYPE_SAMPLER:
1490 case GLSL_TYPE_IMAGE:
1491 case GLSL_TYPE_UINT64:
1492 value.u64[i+offset] = src->get_uint64_component(id++);
1493 break;
1494 case GLSL_TYPE_INT64:
1495 value.i64[i+offset] = src->get_int64_component(id++);
1496 break;
1497 default:
1498 assert(!"Should not get here.");
1499 return;
1500 }
1501 }
1502 }
1503 }
1504
1505 bool
has_value(const ir_constant * c) const1506 ir_constant::has_value(const ir_constant *c) const
1507 {
1508 if (this->type != c->type)
1509 return false;
1510
1511 if (this->type->is_array() || this->type->is_struct()) {
1512 for (unsigned i = 0; i < this->type->length; i++) {
1513 if (!this->const_elements[i]->has_value(c->const_elements[i]))
1514 return false;
1515 }
1516 return true;
1517 }
1518
1519 for (unsigned i = 0; i < this->type->components(); i++) {
1520 switch (this->type->base_type) {
1521 case GLSL_TYPE_UINT16:
1522 if (this->value.u16[i] != c->value.u16[i])
1523 return false;
1524 break;
1525 case GLSL_TYPE_INT16:
1526 if (this->value.i16[i] != c->value.i16[i])
1527 return false;
1528 break;
1529 case GLSL_TYPE_UINT:
1530 if (this->value.u[i] != c->value.u[i])
1531 return false;
1532 break;
1533 case GLSL_TYPE_INT:
1534 if (this->value.i[i] != c->value.i[i])
1535 return false;
1536 break;
1537 case GLSL_TYPE_FLOAT:
1538 if (this->value.f[i] != c->value.f[i])
1539 return false;
1540 break;
1541 case GLSL_TYPE_FLOAT16:
1542 /* Convert to float to make sure NaN and ±0.0 compares correctly */
1543 if (_mesa_half_to_float(this->value.f16[i]) !=
1544 _mesa_half_to_float(c->value.f16[i]))
1545 return false;
1546 break;
1547 case GLSL_TYPE_BOOL:
1548 if (this->value.b[i] != c->value.b[i])
1549 return false;
1550 break;
1551 case GLSL_TYPE_DOUBLE:
1552 if (this->value.d[i] != c->value.d[i])
1553 return false;
1554 break;
1555 case GLSL_TYPE_SAMPLER:
1556 case GLSL_TYPE_IMAGE:
1557 case GLSL_TYPE_UINT64:
1558 if (this->value.u64[i] != c->value.u64[i])
1559 return false;
1560 break;
1561 case GLSL_TYPE_INT64:
1562 if (this->value.i64[i] != c->value.i64[i])
1563 return false;
1564 break;
1565 default:
1566 assert(!"Should not get here.");
1567 return false;
1568 }
1569 }
1570
1571 return true;
1572 }
1573
1574 bool
is_value(float f,int i) const1575 ir_constant::is_value(float f, int i) const
1576 {
1577 if (!this->type->is_scalar() && !this->type->is_vector())
1578 return false;
1579
1580 /* Only accept boolean values for 0/1. */
1581 if (int(bool(i)) != i && this->type->is_boolean())
1582 return false;
1583
1584 for (unsigned c = 0; c < this->type->vector_elements; c++) {
1585 switch (this->type->base_type) {
1586 case GLSL_TYPE_FLOAT:
1587 if (this->value.f[c] != f)
1588 return false;
1589 break;
1590 case GLSL_TYPE_FLOAT16:
1591 if (_mesa_half_to_float(this->value.f16[c]) != f)
1592 return false;
1593 break;
1594 case GLSL_TYPE_INT16:
1595 if (this->value.i16[c] != int16_t(i))
1596 return false;
1597 break;
1598 case GLSL_TYPE_UINT16:
1599 if (this->value.u16[c] != uint16_t(i))
1600 return false;
1601 break;
1602 case GLSL_TYPE_INT:
1603 if (this->value.i[c] != i)
1604 return false;
1605 break;
1606 case GLSL_TYPE_UINT:
1607 if (this->value.u[c] != unsigned(i))
1608 return false;
1609 break;
1610 case GLSL_TYPE_BOOL:
1611 if (this->value.b[c] != bool(i))
1612 return false;
1613 break;
1614 case GLSL_TYPE_DOUBLE:
1615 if (this->value.d[c] != double(f))
1616 return false;
1617 break;
1618 case GLSL_TYPE_SAMPLER:
1619 case GLSL_TYPE_IMAGE:
1620 case GLSL_TYPE_UINT64:
1621 if (this->value.u64[c] != uint64_t(i))
1622 return false;
1623 break;
1624 case GLSL_TYPE_INT64:
1625 if (this->value.i64[c] != i)
1626 return false;
1627 break;
1628 default:
1629 /* The only other base types are structures, arrays, and samplers.
1630 * Samplers cannot be constants, and the others should have been
1631 * filtered out above.
1632 */
1633 assert(!"Should not get here.");
1634 return false;
1635 }
1636 }
1637
1638 return true;
1639 }
1640
1641 bool
is_zero() const1642 ir_constant::is_zero() const
1643 {
1644 return is_value(0.0, 0);
1645 }
1646
1647 bool
is_one() const1648 ir_constant::is_one() const
1649 {
1650 return is_value(1.0, 1);
1651 }
1652
1653 bool
is_negative_one() const1654 ir_constant::is_negative_one() const
1655 {
1656 return is_value(-1.0, -1);
1657 }
1658
1659 bool
is_uint16_constant() const1660 ir_constant::is_uint16_constant() const
1661 {
1662 if (!type->is_integer_32())
1663 return false;
1664
1665 return value.u[0] < (1 << 16);
1666 }
1667
ir_loop()1668 ir_loop::ir_loop()
1669 : ir_instruction(ir_type_loop)
1670 {
1671 }
1672
1673
ir_dereference_variable(ir_variable * var)1674 ir_dereference_variable::ir_dereference_variable(ir_variable *var)
1675 : ir_dereference(ir_type_dereference_variable)
1676 {
1677 assert(var != NULL);
1678
1679 this->var = var;
1680 this->type = var->type;
1681 }
1682
1683
ir_dereference_array(ir_rvalue * value,ir_rvalue * array_index)1684 ir_dereference_array::ir_dereference_array(ir_rvalue *value,
1685 ir_rvalue *array_index)
1686 : ir_dereference(ir_type_dereference_array)
1687 {
1688 this->array_index = array_index;
1689 this->set_array(value);
1690 }
1691
1692
ir_dereference_array(ir_variable * var,ir_rvalue * array_index)1693 ir_dereference_array::ir_dereference_array(ir_variable *var,
1694 ir_rvalue *array_index)
1695 : ir_dereference(ir_type_dereference_array)
1696 {
1697 void *ctx = ralloc_parent(var);
1698
1699 this->array_index = array_index;
1700 this->set_array(new(ctx) ir_dereference_variable(var));
1701 }
1702
1703
1704 void
set_array(ir_rvalue * value)1705 ir_dereference_array::set_array(ir_rvalue *value)
1706 {
1707 assert(value != NULL);
1708
1709 this->array = value;
1710
1711 const glsl_type *const vt = this->array->type;
1712
1713 if (vt->is_array()) {
1714 type = vt->fields.array;
1715 } else if (vt->is_matrix()) {
1716 type = vt->column_type();
1717 } else if (vt->is_vector()) {
1718 type = vt->get_base_type();
1719 }
1720 }
1721
1722
ir_dereference_record(ir_rvalue * value,const char * field)1723 ir_dereference_record::ir_dereference_record(ir_rvalue *value,
1724 const char *field)
1725 : ir_dereference(ir_type_dereference_record)
1726 {
1727 assert(value != NULL);
1728
1729 this->record = value;
1730 this->type = this->record->type->field_type(field);
1731 this->field_idx = this->record->type->field_index(field);
1732 }
1733
1734
ir_dereference_record(ir_variable * var,const char * field)1735 ir_dereference_record::ir_dereference_record(ir_variable *var,
1736 const char *field)
1737 : ir_dereference(ir_type_dereference_record)
1738 {
1739 void *ctx = ralloc_parent(var);
1740
1741 this->record = new(ctx) ir_dereference_variable(var);
1742 this->type = this->record->type->field_type(field);
1743 this->field_idx = this->record->type->field_index(field);
1744 }
1745
1746 bool
is_lvalue(const struct _mesa_glsl_parse_state * state) const1747 ir_dereference::is_lvalue(const struct _mesa_glsl_parse_state *state) const
1748 {
1749 ir_variable *var = this->variable_referenced();
1750
1751 /* Every l-value derference chain eventually ends in a variable.
1752 */
1753 if ((var == NULL) || var->data.read_only)
1754 return false;
1755
1756 /* From section 4.1.7 of the ARB_bindless_texture spec:
1757 *
1758 * "Samplers can be used as l-values, so can be assigned into and used as
1759 * "out" and "inout" function parameters."
1760 *
1761 * From section 4.1.X of the ARB_bindless_texture spec:
1762 *
1763 * "Images can be used as l-values, so can be assigned into and used as
1764 * "out" and "inout" function parameters."
1765 */
1766 if ((!state || state->has_bindless()) &&
1767 (this->type->contains_sampler() || this->type->contains_image()))
1768 return true;
1769
1770 /* From section 4.1.7 of the GLSL 4.40 spec:
1771 *
1772 * "Opaque variables cannot be treated as l-values; hence cannot
1773 * be used as out or inout function parameters, nor can they be
1774 * assigned into."
1775 */
1776 if (this->type->contains_opaque())
1777 return false;
1778
1779 return true;
1780 }
1781
1782
1783 static const char * const tex_opcode_strs[] = { "tex", "txb", "txl", "txd", "txf", "txf_ms", "txs", "lod", "tg4", "query_levels", "texture_samples", "samples_identical" };
1784
opcode_string()1785 const char *ir_texture::opcode_string()
1786 {
1787 assert((unsigned int) op < ARRAY_SIZE(tex_opcode_strs));
1788 return tex_opcode_strs[op];
1789 }
1790
1791 ir_texture_opcode
get_opcode(const char * str)1792 ir_texture::get_opcode(const char *str)
1793 {
1794 const int count = sizeof(tex_opcode_strs) / sizeof(tex_opcode_strs[0]);
1795 for (int op = 0; op < count; op++) {
1796 if (strcmp(str, tex_opcode_strs[op]) == 0)
1797 return (ir_texture_opcode) op;
1798 }
1799 return (ir_texture_opcode) -1;
1800 }
1801
1802
1803 void
set_sampler(ir_dereference * sampler,const glsl_type * type)1804 ir_texture::set_sampler(ir_dereference *sampler, const glsl_type *type)
1805 {
1806 assert(sampler != NULL);
1807 assert(type != NULL);
1808 this->sampler = sampler;
1809 this->type = type;
1810
1811 if (this->op == ir_txs || this->op == ir_query_levels ||
1812 this->op == ir_texture_samples) {
1813 assert(type->base_type == GLSL_TYPE_INT);
1814 } else if (this->op == ir_lod) {
1815 assert(type->vector_elements == 2);
1816 assert(type->is_float());
1817 } else if (this->op == ir_samples_identical) {
1818 assert(type == glsl_type::bool_type);
1819 assert(sampler->type->is_sampler());
1820 assert(sampler->type->sampler_dimensionality == GLSL_SAMPLER_DIM_MS);
1821 } else {
1822 assert(sampler->type->sampled_type == (int) type->base_type);
1823 if (sampler->type->sampler_shadow)
1824 assert(type->vector_elements == 4 || type->vector_elements == 1);
1825 else
1826 assert(type->vector_elements == 4);
1827 }
1828 }
1829
1830
1831 void
init_mask(const unsigned * comp,unsigned count)1832 ir_swizzle::init_mask(const unsigned *comp, unsigned count)
1833 {
1834 assert((count >= 1) && (count <= 4));
1835
1836 memset(&this->mask, 0, sizeof(this->mask));
1837 this->mask.num_components = count;
1838
1839 unsigned dup_mask = 0;
1840 switch (count) {
1841 case 4:
1842 assert(comp[3] <= 3);
1843 dup_mask |= (1U << comp[3])
1844 & ((1U << comp[0]) | (1U << comp[1]) | (1U << comp[2]));
1845 this->mask.w = comp[3];
1846
1847 case 3:
1848 assert(comp[2] <= 3);
1849 dup_mask |= (1U << comp[2])
1850 & ((1U << comp[0]) | (1U << comp[1]));
1851 this->mask.z = comp[2];
1852
1853 case 2:
1854 assert(comp[1] <= 3);
1855 dup_mask |= (1U << comp[1])
1856 & ((1U << comp[0]));
1857 this->mask.y = comp[1];
1858
1859 case 1:
1860 assert(comp[0] <= 3);
1861 this->mask.x = comp[0];
1862 }
1863
1864 this->mask.has_duplicates = dup_mask != 0;
1865
1866 /* Based on the number of elements in the swizzle and the base type
1867 * (i.e., float, int, unsigned, or bool) of the vector being swizzled,
1868 * generate the type of the resulting value.
1869 */
1870 type = glsl_type::get_instance(val->type->base_type, mask.num_components, 1);
1871 }
1872
ir_swizzle(ir_rvalue * val,unsigned x,unsigned y,unsigned z,unsigned w,unsigned count)1873 ir_swizzle::ir_swizzle(ir_rvalue *val, unsigned x, unsigned y, unsigned z,
1874 unsigned w, unsigned count)
1875 : ir_rvalue(ir_type_swizzle), val(val)
1876 {
1877 const unsigned components[4] = { x, y, z, w };
1878 this->init_mask(components, count);
1879 }
1880
ir_swizzle(ir_rvalue * val,const unsigned * comp,unsigned count)1881 ir_swizzle::ir_swizzle(ir_rvalue *val, const unsigned *comp,
1882 unsigned count)
1883 : ir_rvalue(ir_type_swizzle), val(val)
1884 {
1885 this->init_mask(comp, count);
1886 }
1887
ir_swizzle(ir_rvalue * val,ir_swizzle_mask mask)1888 ir_swizzle::ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask)
1889 : ir_rvalue(ir_type_swizzle), val(val), mask(mask)
1890 {
1891 this->type = glsl_type::get_instance(val->type->base_type,
1892 mask.num_components, 1);
1893 }
1894
1895 #define X 1
1896 #define R 5
1897 #define S 9
1898 #define I 13
1899
1900 ir_swizzle *
create(ir_rvalue * val,const char * str,unsigned vector_length)1901 ir_swizzle::create(ir_rvalue *val, const char *str, unsigned vector_length)
1902 {
1903 void *ctx = ralloc_parent(val);
1904
1905 /* For each possible swizzle character, this table encodes the value in
1906 * \c idx_map that represents the 0th element of the vector. For invalid
1907 * swizzle characters (e.g., 'k'), a special value is used that will allow
1908 * detection of errors.
1909 */
1910 static const unsigned char base_idx[26] = {
1911 /* a b c d e f g h i j k l m */
1912 R, R, I, I, I, I, R, I, I, I, I, I, I,
1913 /* n o p q r s t u v w x y z */
1914 I, I, S, S, R, S, S, I, I, X, X, X, X
1915 };
1916
1917 /* Each valid swizzle character has an entry in the previous table. This
1918 * table encodes the base index encoded in the previous table plus the actual
1919 * index of the swizzle character. When processing swizzles, the first
1920 * character in the string is indexed in the previous table. Each character
1921 * in the string is indexed in this table, and the value found there has the
1922 * value form the first table subtracted. The result must be on the range
1923 * [0,3].
1924 *
1925 * For example, the string "wzyx" will get X from the first table. Each of
1926 * the charcaters will get X+3, X+2, X+1, and X+0 from this table. After
1927 * subtraction, the swizzle values are { 3, 2, 1, 0 }.
1928 *
1929 * The string "wzrg" will get X from the first table. Each of the characters
1930 * will get X+3, X+2, R+0, and R+1 from this table. After subtraction, the
1931 * swizzle values are { 3, 2, 4, 5 }. Since 4 and 5 are outside the range
1932 * [0,3], the error is detected.
1933 */
1934 static const unsigned char idx_map[26] = {
1935 /* a b c d e f g h i j k l m */
1936 R+3, R+2, 0, 0, 0, 0, R+1, 0, 0, 0, 0, 0, 0,
1937 /* n o p q r s t u v w x y z */
1938 0, 0, S+2, S+3, R+0, S+0, S+1, 0, 0, X+3, X+0, X+1, X+2
1939 };
1940
1941 int swiz_idx[4] = { 0, 0, 0, 0 };
1942 unsigned i;
1943
1944
1945 /* Validate the first character in the swizzle string and look up the base
1946 * index value as described above.
1947 */
1948 if ((str[0] < 'a') || (str[0] > 'z'))
1949 return NULL;
1950
1951 const unsigned base = base_idx[str[0] - 'a'];
1952
1953
1954 for (i = 0; (i < 4) && (str[i] != '\0'); i++) {
1955 /* Validate the next character, and, as described above, convert it to a
1956 * swizzle index.
1957 */
1958 if ((str[i] < 'a') || (str[i] > 'z'))
1959 return NULL;
1960
1961 swiz_idx[i] = idx_map[str[i] - 'a'] - base;
1962 if ((swiz_idx[i] < 0) || (swiz_idx[i] >= (int) vector_length))
1963 return NULL;
1964 }
1965
1966 if (str[i] != '\0')
1967 return NULL;
1968
1969 return new(ctx) ir_swizzle(val, swiz_idx[0], swiz_idx[1], swiz_idx[2],
1970 swiz_idx[3], i);
1971 }
1972
1973 #undef X
1974 #undef R
1975 #undef S
1976 #undef I
1977
1978 ir_variable *
variable_referenced() const1979 ir_swizzle::variable_referenced() const
1980 {
1981 return this->val->variable_referenced();
1982 }
1983
1984
1985 bool ir_variable::temporaries_allocate_names = false;
1986
1987 const char ir_variable::tmp_name[] = "compiler_temp";
1988
ir_variable(const struct glsl_type * type,const char * name,ir_variable_mode mode)1989 ir_variable::ir_variable(const struct glsl_type *type, const char *name,
1990 ir_variable_mode mode)
1991 : ir_instruction(ir_type_variable)
1992 {
1993 this->type = type;
1994
1995 if (mode == ir_var_temporary && !ir_variable::temporaries_allocate_names)
1996 name = NULL;
1997
1998 /* The ir_variable clone method may call this constructor with name set to
1999 * tmp_name.
2000 */
2001 assert(name != NULL
2002 || mode == ir_var_temporary
2003 || mode == ir_var_function_in
2004 || mode == ir_var_function_out
2005 || mode == ir_var_function_inout);
2006 assert(name != ir_variable::tmp_name
2007 || mode == ir_var_temporary);
2008 if (mode == ir_var_temporary
2009 && (name == NULL || name == ir_variable::tmp_name)) {
2010 this->name = ir_variable::tmp_name;
2011 } else if (name == NULL ||
2012 strlen(name) < ARRAY_SIZE(this->name_storage)) {
2013 strcpy(this->name_storage, name ? name : "");
2014 this->name = this->name_storage;
2015 } else {
2016 this->name = ralloc_strdup(this, name);
2017 }
2018
2019 this->u.max_ifc_array_access = NULL;
2020
2021 this->data.explicit_location = false;
2022 this->data.explicit_index = false;
2023 this->data.explicit_binding = false;
2024 this->data.explicit_component = false;
2025 this->data.has_initializer = false;
2026 this->data.is_implicit_initializer = false;
2027 this->data.is_unmatched_generic_inout = false;
2028 this->data.is_xfb_only = false;
2029 this->data.explicit_xfb_buffer = false;
2030 this->data.explicit_xfb_offset = false;
2031 this->data.explicit_xfb_stride = false;
2032 this->data.location = -1;
2033 this->data.location_frac = 0;
2034 this->data.matrix_layout = GLSL_MATRIX_LAYOUT_INHERITED;
2035 this->data.from_named_ifc_block = false;
2036 this->data.must_be_shader_input = false;
2037 this->data.index = 0;
2038 this->data.binding = 0;
2039 this->data.warn_extension_index = 0;
2040 this->constant_value = NULL;
2041 this->constant_initializer = NULL;
2042 this->data.depth_layout = ir_depth_layout_none;
2043 this->data.used = false;
2044 this->data.assigned = false;
2045 this->data.always_active_io = false;
2046 this->data.read_only = false;
2047 this->data.centroid = false;
2048 this->data.sample = false;
2049 this->data.patch = false;
2050 this->data.explicit_invariant = false;
2051 this->data.invariant = false;
2052 this->data.precise = false;
2053 this->data.how_declared = ir_var_declared_normally;
2054 this->data.mode = mode;
2055 this->data.interpolation = INTERP_MODE_NONE;
2056 this->data.max_array_access = -1;
2057 this->data.offset = 0;
2058 this->data.precision = GLSL_PRECISION_NONE;
2059 this->data.memory_read_only = false;
2060 this->data.memory_write_only = false;
2061 this->data.memory_coherent = false;
2062 this->data.memory_volatile = false;
2063 this->data.memory_restrict = false;
2064 this->data.from_ssbo_unsized_array = false;
2065 this->data.implicit_sized_array = false;
2066 this->data.fb_fetch_output = false;
2067 this->data.bindless = false;
2068 this->data.bound = false;
2069 this->data.image_format = PIPE_FORMAT_NONE;
2070 this->data._num_state_slots = 0;
2071 this->data.param_index = 0;
2072 this->data.stream = 0;
2073 this->data.xfb_buffer = -1;
2074 this->data.xfb_stride = -1;
2075
2076 this->interface_type = NULL;
2077
2078 if (type != NULL) {
2079 if (type->is_interface())
2080 this->init_interface_type(type);
2081 else if (type->without_array()->is_interface())
2082 this->init_interface_type(type->without_array());
2083 }
2084 }
2085
2086
2087 const char *
interpolation_string(unsigned interpolation)2088 interpolation_string(unsigned interpolation)
2089 {
2090 switch (interpolation) {
2091 case INTERP_MODE_NONE: return "no";
2092 case INTERP_MODE_SMOOTH: return "smooth";
2093 case INTERP_MODE_FLAT: return "flat";
2094 case INTERP_MODE_NOPERSPECTIVE: return "noperspective";
2095 }
2096
2097 assert(!"Should not get here.");
2098 return "";
2099 }
2100
2101 const char *const ir_variable::warn_extension_table[] = {
2102 "",
2103 "GL_ARB_shader_stencil_export",
2104 "GL_AMD_shader_stencil_export",
2105 };
2106
2107 void
enable_extension_warning(const char * extension)2108 ir_variable::enable_extension_warning(const char *extension)
2109 {
2110 for (unsigned i = 0; i < ARRAY_SIZE(warn_extension_table); i++) {
2111 if (strcmp(warn_extension_table[i], extension) == 0) {
2112 this->data.warn_extension_index = i;
2113 return;
2114 }
2115 }
2116
2117 assert(!"Should not get here.");
2118 this->data.warn_extension_index = 0;
2119 }
2120
2121 const char *
get_extension_warning() const2122 ir_variable::get_extension_warning() const
2123 {
2124 return this->data.warn_extension_index == 0
2125 ? NULL : warn_extension_table[this->data.warn_extension_index];
2126 }
2127
ir_function_signature(const glsl_type * return_type,builtin_available_predicate b)2128 ir_function_signature::ir_function_signature(const glsl_type *return_type,
2129 builtin_available_predicate b)
2130 : ir_instruction(ir_type_function_signature),
2131 return_type(return_type), is_defined(false),
2132 return_precision(GLSL_PRECISION_NONE),
2133 intrinsic_id(ir_intrinsic_invalid), builtin_avail(b), _function(NULL)
2134 {
2135 this->origin = NULL;
2136 }
2137
2138
2139 bool
is_builtin() const2140 ir_function_signature::is_builtin() const
2141 {
2142 return builtin_avail != NULL;
2143 }
2144
2145
2146 bool
is_builtin_available(const _mesa_glsl_parse_state * state) const2147 ir_function_signature::is_builtin_available(const _mesa_glsl_parse_state *state) const
2148 {
2149 /* We can't call the predicate without a state pointer, so just say that
2150 * the signature is available. At compile time, we need the filtering,
2151 * but also receive a valid state pointer. At link time, we're resolving
2152 * imported built-in prototypes to their definitions, which will always
2153 * be an exact match. So we can skip the filtering.
2154 */
2155 if (state == NULL)
2156 return true;
2157
2158 assert(builtin_avail != NULL);
2159 return builtin_avail(state);
2160 }
2161
2162
2163 static bool
modes_match(unsigned a,unsigned b)2164 modes_match(unsigned a, unsigned b)
2165 {
2166 if (a == b)
2167 return true;
2168
2169 /* Accept "in" vs. "const in" */
2170 if ((a == ir_var_const_in && b == ir_var_function_in) ||
2171 (b == ir_var_const_in && a == ir_var_function_in))
2172 return true;
2173
2174 return false;
2175 }
2176
2177
2178 const char *
qualifiers_match(exec_list * params)2179 ir_function_signature::qualifiers_match(exec_list *params)
2180 {
2181 /* check that the qualifiers match. */
2182 foreach_two_lists(a_node, &this->parameters, b_node, params) {
2183 ir_variable *a = (ir_variable *) a_node;
2184 ir_variable *b = (ir_variable *) b_node;
2185
2186 if (a->data.read_only != b->data.read_only ||
2187 !modes_match(a->data.mode, b->data.mode) ||
2188 a->data.interpolation != b->data.interpolation ||
2189 a->data.centroid != b->data.centroid ||
2190 a->data.sample != b->data.sample ||
2191 a->data.patch != b->data.patch ||
2192 a->data.memory_read_only != b->data.memory_read_only ||
2193 a->data.memory_write_only != b->data.memory_write_only ||
2194 a->data.memory_coherent != b->data.memory_coherent ||
2195 a->data.memory_volatile != b->data.memory_volatile ||
2196 a->data.memory_restrict != b->data.memory_restrict) {
2197
2198 /* parameter a's qualifiers don't match */
2199 return a->name;
2200 }
2201 }
2202 return NULL;
2203 }
2204
2205
2206 void
replace_parameters(exec_list * new_params)2207 ir_function_signature::replace_parameters(exec_list *new_params)
2208 {
2209 /* Destroy all of the previous parameter information. If the previous
2210 * parameter information comes from the function prototype, it may either
2211 * specify incorrect parameter names or not have names at all.
2212 */
2213 new_params->move_nodes_to(¶meters);
2214 }
2215
2216
ir_function(const char * name)2217 ir_function::ir_function(const char *name)
2218 : ir_instruction(ir_type_function)
2219 {
2220 this->subroutine_index = -1;
2221 this->name = ralloc_strdup(this, name);
2222 }
2223
2224
2225 bool
has_user_signature()2226 ir_function::has_user_signature()
2227 {
2228 foreach_in_list(ir_function_signature, sig, &this->signatures) {
2229 if (!sig->is_builtin())
2230 return true;
2231 }
2232 return false;
2233 }
2234
2235
2236 ir_rvalue *
error_value(void * mem_ctx)2237 ir_rvalue::error_value(void *mem_ctx)
2238 {
2239 ir_rvalue *v = new(mem_ctx) ir_rvalue(ir_type_unset);
2240
2241 v->type = glsl_type::error_type;
2242 return v;
2243 }
2244
2245
2246 void
visit_exec_list(exec_list * list,ir_visitor * visitor)2247 visit_exec_list(exec_list *list, ir_visitor *visitor)
2248 {
2249 foreach_in_list_safe(ir_instruction, node, list) {
2250 node->accept(visitor);
2251 }
2252 }
2253
2254
2255 static void
steal_memory(ir_instruction * ir,void * new_ctx)2256 steal_memory(ir_instruction *ir, void *new_ctx)
2257 {
2258 ir_variable *var = ir->as_variable();
2259 ir_function *fn = ir->as_function();
2260 ir_constant *constant = ir->as_constant();
2261 if (var != NULL && var->constant_value != NULL)
2262 steal_memory(var->constant_value, ir);
2263
2264 if (var != NULL && var->constant_initializer != NULL)
2265 steal_memory(var->constant_initializer, ir);
2266
2267 if (fn != NULL && fn->subroutine_types)
2268 ralloc_steal(new_ctx, fn->subroutine_types);
2269
2270 /* The components of aggregate constants are not visited by the normal
2271 * visitor, so steal their values by hand.
2272 */
2273 if (constant != NULL &&
2274 (constant->type->is_array() || constant->type->is_struct())) {
2275 for (unsigned int i = 0; i < constant->type->length; i++) {
2276 steal_memory(constant->const_elements[i], ir);
2277 }
2278 }
2279
2280 ralloc_steal(new_ctx, ir);
2281 }
2282
2283
2284 void
reparent_ir(exec_list * list,void * mem_ctx)2285 reparent_ir(exec_list *list, void *mem_ctx)
2286 {
2287 foreach_in_list(ir_instruction, node, list) {
2288 visit_tree(node, steal_memory, mem_ctx);
2289 }
2290 }
2291
2292
2293 static ir_rvalue *
try_min_one(ir_rvalue * ir)2294 try_min_one(ir_rvalue *ir)
2295 {
2296 ir_expression *expr = ir->as_expression();
2297
2298 if (!expr || expr->operation != ir_binop_min)
2299 return NULL;
2300
2301 if (expr->operands[0]->is_one())
2302 return expr->operands[1];
2303
2304 if (expr->operands[1]->is_one())
2305 return expr->operands[0];
2306
2307 return NULL;
2308 }
2309
2310 static ir_rvalue *
try_max_zero(ir_rvalue * ir)2311 try_max_zero(ir_rvalue *ir)
2312 {
2313 ir_expression *expr = ir->as_expression();
2314
2315 if (!expr || expr->operation != ir_binop_max)
2316 return NULL;
2317
2318 if (expr->operands[0]->is_zero())
2319 return expr->operands[1];
2320
2321 if (expr->operands[1]->is_zero())
2322 return expr->operands[0];
2323
2324 return NULL;
2325 }
2326
2327 ir_rvalue *
as_rvalue_to_saturate()2328 ir_rvalue::as_rvalue_to_saturate()
2329 {
2330 ir_expression *expr = this->as_expression();
2331
2332 if (!expr)
2333 return NULL;
2334
2335 ir_rvalue *max_zero = try_max_zero(expr);
2336 if (max_zero) {
2337 return try_min_one(max_zero);
2338 } else {
2339 ir_rvalue *min_one = try_min_one(expr);
2340 if (min_one) {
2341 return try_max_zero(min_one);
2342 }
2343 }
2344
2345 return NULL;
2346 }
2347
2348
2349 unsigned
vertices_per_prim(GLenum prim)2350 vertices_per_prim(GLenum prim)
2351 {
2352 switch (prim) {
2353 case GL_POINTS:
2354 return 1;
2355 case GL_LINES:
2356 return 2;
2357 case GL_TRIANGLES:
2358 return 3;
2359 case GL_LINES_ADJACENCY:
2360 return 4;
2361 case GL_TRIANGLES_ADJACENCY:
2362 return 6;
2363 default:
2364 assert(!"Bad primitive");
2365 return 3;
2366 }
2367 }
2368
2369 /**
2370 * Generate a string describing the mode of a variable
2371 */
2372 const char *
mode_string(const ir_variable * var)2373 mode_string(const ir_variable *var)
2374 {
2375 switch (var->data.mode) {
2376 case ir_var_auto:
2377 return (var->data.read_only) ? "global constant" : "global variable";
2378
2379 case ir_var_uniform:
2380 return "uniform";
2381
2382 case ir_var_shader_storage:
2383 return "buffer";
2384
2385 case ir_var_shader_in:
2386 return "shader input";
2387
2388 case ir_var_shader_out:
2389 return "shader output";
2390
2391 case ir_var_function_in:
2392 case ir_var_const_in:
2393 return "function input";
2394
2395 case ir_var_function_out:
2396 return "function output";
2397
2398 case ir_var_function_inout:
2399 return "function inout";
2400
2401 case ir_var_system_value:
2402 return "shader input";
2403
2404 case ir_var_temporary:
2405 return "compiler temporary";
2406
2407 case ir_var_mode_count:
2408 break;
2409 }
2410
2411 assert(!"Should not get here.");
2412 return "invalid variable";
2413 }
2414