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
ir_rvalue(enum ir_node_type t)28 ir_rvalue::ir_rvalue(enum ir_node_type t)
29 : ir_instruction(t)
30 {
31 this->type = glsl_type::error_type;
32 }
33
is_zero() const34 bool ir_rvalue::is_zero() const
35 {
36 return false;
37 }
38
is_one() const39 bool ir_rvalue::is_one() const
40 {
41 return false;
42 }
43
is_negative_one() const44 bool ir_rvalue::is_negative_one() const
45 {
46 return false;
47 }
48
49 /**
50 * Modify the swizzle make to move one component to another
51 *
52 * \param m IR swizzle to be modified
53 * \param from Component in the RHS that is to be swizzled
54 * \param to Desired swizzle location of \c from
55 */
56 static void
update_rhs_swizzle(ir_swizzle_mask & m,unsigned from,unsigned to)57 update_rhs_swizzle(ir_swizzle_mask &m, unsigned from, unsigned to)
58 {
59 switch (to) {
60 case 0: m.x = from; break;
61 case 1: m.y = from; break;
62 case 2: m.z = from; break;
63 case 3: m.w = from; break;
64 default: assert(!"Should not get here.");
65 }
66 }
67
68 void
set_lhs(ir_rvalue * lhs)69 ir_assignment::set_lhs(ir_rvalue *lhs)
70 {
71 void *mem_ctx = this;
72 bool swizzled = false;
73
74 while (lhs != NULL) {
75 ir_swizzle *swiz = lhs->as_swizzle();
76
77 if (swiz == NULL)
78 break;
79
80 unsigned write_mask = 0;
81 ir_swizzle_mask rhs_swiz = { 0, 0, 0, 0, 0, 0 };
82
83 for (unsigned i = 0; i < swiz->mask.num_components; i++) {
84 unsigned c = 0;
85
86 switch (i) {
87 case 0: c = swiz->mask.x; break;
88 case 1: c = swiz->mask.y; break;
89 case 2: c = swiz->mask.z; break;
90 case 3: c = swiz->mask.w; break;
91 default: assert(!"Should not get here.");
92 }
93
94 write_mask |= (((this->write_mask >> i) & 1) << c);
95 update_rhs_swizzle(rhs_swiz, i, c);
96 rhs_swiz.num_components = swiz->val->type->vector_elements;
97 }
98
99 this->write_mask = write_mask;
100 lhs = swiz->val;
101
102 this->rhs = new(mem_ctx) ir_swizzle(this->rhs, rhs_swiz);
103 swizzled = true;
104 }
105
106 if (swizzled) {
107 /* Now, RHS channels line up with the LHS writemask. Collapse it
108 * to just the channels that will be written.
109 */
110 ir_swizzle_mask rhs_swiz = { 0, 0, 0, 0, 0, 0 };
111 int rhs_chan = 0;
112 for (int i = 0; i < 4; i++) {
113 if (write_mask & (1 << i))
114 update_rhs_swizzle(rhs_swiz, i, rhs_chan++);
115 }
116 rhs_swiz.num_components = rhs_chan;
117 this->rhs = new(mem_ctx) ir_swizzle(this->rhs, rhs_swiz);
118 }
119
120 assert((lhs == NULL) || lhs->as_dereference());
121
122 this->lhs = (ir_dereference *) lhs;
123 }
124
125 ir_variable *
whole_variable_written()126 ir_assignment::whole_variable_written()
127 {
128 ir_variable *v = this->lhs->whole_variable_referenced();
129
130 if (v == NULL)
131 return NULL;
132
133 if (v->type->is_scalar())
134 return v;
135
136 if (v->type->is_vector()) {
137 const unsigned mask = (1U << v->type->vector_elements) - 1;
138
139 if (mask != this->write_mask)
140 return NULL;
141 }
142
143 /* Either all the vector components are assigned or the variable is some
144 * composite type (and the whole thing is assigned.
145 */
146 return v;
147 }
148
ir_assignment(ir_dereference * lhs,ir_rvalue * rhs,ir_rvalue * condition,unsigned write_mask)149 ir_assignment::ir_assignment(ir_dereference *lhs, ir_rvalue *rhs,
150 ir_rvalue *condition, unsigned write_mask)
151 : ir_instruction(ir_type_assignment)
152 {
153 this->condition = condition;
154 this->rhs = rhs;
155 this->lhs = lhs;
156 this->write_mask = write_mask;
157
158 if (lhs->type->is_scalar() || lhs->type->is_vector()) {
159 int lhs_components = 0;
160 for (int i = 0; i < 4; i++) {
161 if (write_mask & (1 << i))
162 lhs_components++;
163 }
164
165 assert(lhs_components == this->rhs->type->vector_elements);
166 }
167 }
168
ir_assignment(ir_rvalue * lhs,ir_rvalue * rhs,ir_rvalue * condition)169 ir_assignment::ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs,
170 ir_rvalue *condition)
171 : ir_instruction(ir_type_assignment)
172 {
173 this->condition = condition;
174 this->rhs = rhs;
175
176 /* If the RHS is a vector type, assume that all components of the vector
177 * type are being written to the LHS. The write mask comes from the RHS
178 * because we can have a case where the LHS is a vec4 and the RHS is a
179 * vec3. In that case, the assignment is:
180 *
181 * (assign (...) (xyz) (var_ref lhs) (var_ref rhs))
182 */
183 if (rhs->type->is_vector())
184 this->write_mask = (1U << rhs->type->vector_elements) - 1;
185 else if (rhs->type->is_scalar())
186 this->write_mask = 1;
187 else
188 this->write_mask = 0;
189
190 this->set_lhs(lhs);
191 }
192
ir_expression(int op,const struct glsl_type * type,ir_rvalue * op0,ir_rvalue * op1,ir_rvalue * op2,ir_rvalue * op3)193 ir_expression::ir_expression(int op, const struct glsl_type *type,
194 ir_rvalue *op0, ir_rvalue *op1,
195 ir_rvalue *op2, ir_rvalue *op3)
196 : ir_rvalue(ir_type_expression)
197 {
198 this->type = type;
199 this->operation = ir_expression_operation(op);
200 this->operands[0] = op0;
201 this->operands[1] = op1;
202 this->operands[2] = op2;
203 this->operands[3] = op3;
204 #ifndef NDEBUG
205 int num_operands = get_num_operands(this->operation);
206 for (int i = num_operands; i < 4; i++) {
207 assert(this->operands[i] == NULL);
208 }
209 #endif
210 }
211
ir_expression(int op,ir_rvalue * op0)212 ir_expression::ir_expression(int op, ir_rvalue *op0)
213 : ir_rvalue(ir_type_expression)
214 {
215 this->operation = ir_expression_operation(op);
216 this->operands[0] = op0;
217 this->operands[1] = NULL;
218 this->operands[2] = NULL;
219 this->operands[3] = NULL;
220
221 assert(op <= ir_last_unop);
222
223 switch (this->operation) {
224 case ir_unop_bit_not:
225 case ir_unop_logic_not:
226 case ir_unop_neg:
227 case ir_unop_abs:
228 case ir_unop_sign:
229 case ir_unop_rcp:
230 case ir_unop_rsq:
231 case ir_unop_sqrt:
232 case ir_unop_exp:
233 case ir_unop_log:
234 case ir_unop_exp2:
235 case ir_unop_log2:
236 case ir_unop_trunc:
237 case ir_unop_ceil:
238 case ir_unop_floor:
239 case ir_unop_fract:
240 case ir_unop_round_even:
241 case ir_unop_sin:
242 case ir_unop_cos:
243 case ir_unop_dFdx:
244 case ir_unop_dFdx_coarse:
245 case ir_unop_dFdx_fine:
246 case ir_unop_dFdy:
247 case ir_unop_dFdy_coarse:
248 case ir_unop_dFdy_fine:
249 case ir_unop_bitfield_reverse:
250 case ir_unop_interpolate_at_centroid:
251 case ir_unop_saturate:
252 this->type = op0->type;
253 break;
254
255 case ir_unop_f2i:
256 case ir_unop_b2i:
257 case ir_unop_u2i:
258 case ir_unop_d2i:
259 case ir_unop_bitcast_f2i:
260 case ir_unop_bit_count:
261 case ir_unop_find_msb:
262 case ir_unop_find_lsb:
263 case ir_unop_subroutine_to_int:
264 this->type = glsl_type::get_instance(GLSL_TYPE_INT,
265 op0->type->vector_elements, 1);
266 break;
267
268 case ir_unop_b2f:
269 case ir_unop_i2f:
270 case ir_unop_u2f:
271 case ir_unop_d2f:
272 case ir_unop_bitcast_i2f:
273 case ir_unop_bitcast_u2f:
274 this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT,
275 op0->type->vector_elements, 1);
276 break;
277
278 case ir_unop_f2b:
279 case ir_unop_i2b:
280 case ir_unop_d2b:
281 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL,
282 op0->type->vector_elements, 1);
283 break;
284
285 case ir_unop_f2d:
286 case ir_unop_i2d:
287 case ir_unop_u2d:
288 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE,
289 op0->type->vector_elements, 1);
290 break;
291
292 case ir_unop_i2u:
293 case ir_unop_f2u:
294 case ir_unop_d2u:
295 case ir_unop_bitcast_f2u:
296 this->type = glsl_type::get_instance(GLSL_TYPE_UINT,
297 op0->type->vector_elements, 1);
298 break;
299
300 case ir_unop_noise:
301 this->type = glsl_type::float_type;
302 break;
303
304 case ir_unop_unpack_double_2x32:
305 this->type = glsl_type::uvec2_type;
306 break;
307
308 case ir_unop_pack_snorm_2x16:
309 case ir_unop_pack_snorm_4x8:
310 case ir_unop_pack_unorm_2x16:
311 case ir_unop_pack_unorm_4x8:
312 case ir_unop_pack_half_2x16:
313 this->type = glsl_type::uint_type;
314 break;
315
316 case ir_unop_pack_double_2x32:
317 this->type = glsl_type::double_type;
318 break;
319
320 case ir_unop_unpack_snorm_2x16:
321 case ir_unop_unpack_unorm_2x16:
322 case ir_unop_unpack_half_2x16:
323 this->type = glsl_type::vec2_type;
324 break;
325
326 case ir_unop_unpack_snorm_4x8:
327 case ir_unop_unpack_unorm_4x8:
328 this->type = glsl_type::vec4_type;
329 break;
330
331 case ir_unop_frexp_sig:
332 this->type = op0->type;
333 break;
334 case ir_unop_frexp_exp:
335 this->type = glsl_type::get_instance(GLSL_TYPE_INT,
336 op0->type->vector_elements, 1);
337 break;
338
339 case ir_unop_get_buffer_size:
340 case ir_unop_ssbo_unsized_array_length:
341 this->type = glsl_type::int_type;
342 break;
343
344 case ir_unop_vote_any:
345 case ir_unop_vote_all:
346 case ir_unop_vote_eq:
347 this->type = glsl_type::bool_type;
348 break;
349
350 default:
351 assert(!"not reached: missing automatic type setup for ir_expression");
352 this->type = op0->type;
353 break;
354 }
355 }
356
ir_expression(int op,ir_rvalue * op0,ir_rvalue * op1)357 ir_expression::ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1)
358 : ir_rvalue(ir_type_expression)
359 {
360 this->operation = ir_expression_operation(op);
361 this->operands[0] = op0;
362 this->operands[1] = op1;
363 this->operands[2] = NULL;
364 this->operands[3] = NULL;
365
366 assert(op > ir_last_unop);
367
368 switch (this->operation) {
369 case ir_binop_all_equal:
370 case ir_binop_any_nequal:
371 this->type = glsl_type::bool_type;
372 break;
373
374 case ir_binop_add:
375 case ir_binop_sub:
376 case ir_binop_min:
377 case ir_binop_max:
378 case ir_binop_pow:
379 case ir_binop_mul:
380 case ir_binop_div:
381 case ir_binop_mod:
382 if (op0->type->is_scalar()) {
383 this->type = op1->type;
384 } else if (op1->type->is_scalar()) {
385 this->type = op0->type;
386 } else {
387 if (this->operation == ir_binop_mul) {
388 this->type = glsl_type::get_mul_type(op0->type, op1->type);
389 } else {
390 assert(op0->type == op1->type);
391 this->type = op0->type;
392 }
393 }
394 break;
395
396 case ir_binop_logic_and:
397 case ir_binop_logic_xor:
398 case ir_binop_logic_or:
399 case ir_binop_bit_and:
400 case ir_binop_bit_xor:
401 case ir_binop_bit_or:
402 assert(!op0->type->is_matrix());
403 assert(!op1->type->is_matrix());
404 if (op0->type->is_scalar()) {
405 this->type = op1->type;
406 } else if (op1->type->is_scalar()) {
407 this->type = op0->type;
408 } else {
409 assert(op0->type->vector_elements == op1->type->vector_elements);
410 this->type = op0->type;
411 }
412 break;
413
414 case ir_binop_equal:
415 case ir_binop_nequal:
416 case ir_binop_lequal:
417 case ir_binop_gequal:
418 case ir_binop_less:
419 case ir_binop_greater:
420 assert(op0->type == op1->type);
421 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL,
422 op0->type->vector_elements, 1);
423 break;
424
425 case ir_binop_dot:
426 this->type = op0->type->get_base_type();
427 break;
428
429 case ir_binop_imul_high:
430 case ir_binop_carry:
431 case ir_binop_borrow:
432 case ir_binop_lshift:
433 case ir_binop_rshift:
434 case ir_binop_ldexp:
435 case ir_binop_interpolate_at_offset:
436 case ir_binop_interpolate_at_sample:
437 this->type = op0->type;
438 break;
439
440 case ir_binop_vector_extract:
441 this->type = op0->type->get_scalar_type();
442 break;
443
444 default:
445 assert(!"not reached: missing automatic type setup for ir_expression");
446 this->type = glsl_type::float_type;
447 }
448 }
449
ir_expression(int op,ir_rvalue * op0,ir_rvalue * op1,ir_rvalue * op2)450 ir_expression::ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1,
451 ir_rvalue *op2)
452 : ir_rvalue(ir_type_expression)
453 {
454 this->operation = ir_expression_operation(op);
455 this->operands[0] = op0;
456 this->operands[1] = op1;
457 this->operands[2] = op2;
458 this->operands[3] = NULL;
459
460 assert(op > ir_last_binop && op <= ir_last_triop);
461
462 switch (this->operation) {
463 case ir_triop_fma:
464 case ir_triop_lrp:
465 case ir_triop_bitfield_extract:
466 case ir_triop_vector_insert:
467 this->type = op0->type;
468 break;
469
470 case ir_triop_csel:
471 this->type = op1->type;
472 break;
473
474 default:
475 assert(!"not reached: missing automatic type setup for ir_expression");
476 this->type = glsl_type::float_type;
477 }
478 }
479
480 unsigned int
get_num_operands(ir_expression_operation op)481 ir_expression::get_num_operands(ir_expression_operation op)
482 {
483 assert(op <= ir_last_opcode);
484
485 if (op <= ir_last_unop)
486 return 1;
487
488 if (op <= ir_last_binop)
489 return 2;
490
491 if (op <= ir_last_triop)
492 return 3;
493
494 if (op <= ir_last_quadop)
495 return 4;
496
497 assert(false);
498 return 0;
499 }
500
501 #include "ir_expression_operation_strings.h"
502
503 const char*
depth_layout_string(ir_depth_layout layout)504 depth_layout_string(ir_depth_layout layout)
505 {
506 switch(layout) {
507 case ir_depth_layout_none: return "";
508 case ir_depth_layout_any: return "depth_any";
509 case ir_depth_layout_greater: return "depth_greater";
510 case ir_depth_layout_less: return "depth_less";
511 case ir_depth_layout_unchanged: return "depth_unchanged";
512
513 default:
514 assert(0);
515 return "";
516 }
517 }
518
519 ir_expression_operation
get_operator(const char * str)520 ir_expression::get_operator(const char *str)
521 {
522 for (int op = 0; op <= int(ir_last_opcode); op++) {
523 if (strcmp(str, ir_expression_operation_strings[op]) == 0)
524 return (ir_expression_operation) op;
525 }
526 return (ir_expression_operation) -1;
527 }
528
529 ir_variable *
variable_referenced() const530 ir_expression::variable_referenced() const
531 {
532 switch (operation) {
533 case ir_binop_vector_extract:
534 case ir_triop_vector_insert:
535 /* We get these for things like a[0] where a is a vector type. In these
536 * cases we want variable_referenced() to return the actual vector
537 * variable this is wrapping.
538 */
539 return operands[0]->variable_referenced();
540 default:
541 return ir_rvalue::variable_referenced();
542 }
543 }
544
ir_constant()545 ir_constant::ir_constant()
546 : ir_rvalue(ir_type_constant)
547 {
548 this->array_elements = NULL;
549 }
550
ir_constant(const struct glsl_type * type,const ir_constant_data * data)551 ir_constant::ir_constant(const struct glsl_type *type,
552 const ir_constant_data *data)
553 : ir_rvalue(ir_type_constant)
554 {
555 this->array_elements = NULL;
556
557 assert((type->base_type >= GLSL_TYPE_UINT)
558 && (type->base_type <= GLSL_TYPE_BOOL));
559
560 this->type = type;
561 memcpy(& this->value, data, sizeof(this->value));
562 }
563
ir_constant(float f,unsigned vector_elements)564 ir_constant::ir_constant(float f, unsigned vector_elements)
565 : ir_rvalue(ir_type_constant)
566 {
567 assert(vector_elements <= 4);
568 this->type = glsl_type::get_instance(GLSL_TYPE_FLOAT, vector_elements, 1);
569 for (unsigned i = 0; i < vector_elements; i++) {
570 this->value.f[i] = f;
571 }
572 for (unsigned i = vector_elements; i < 16; i++) {
573 this->value.f[i] = 0;
574 }
575 }
576
ir_constant(double d,unsigned vector_elements)577 ir_constant::ir_constant(double d, unsigned vector_elements)
578 : ir_rvalue(ir_type_constant)
579 {
580 assert(vector_elements <= 4);
581 this->type = glsl_type::get_instance(GLSL_TYPE_DOUBLE, vector_elements, 1);
582 for (unsigned i = 0; i < vector_elements; i++) {
583 this->value.d[i] = d;
584 }
585 for (unsigned i = vector_elements; i < 16; i++) {
586 this->value.d[i] = 0.0;
587 }
588 }
589
ir_constant(unsigned int u,unsigned vector_elements)590 ir_constant::ir_constant(unsigned int u, unsigned vector_elements)
591 : ir_rvalue(ir_type_constant)
592 {
593 assert(vector_elements <= 4);
594 this->type = glsl_type::get_instance(GLSL_TYPE_UINT, vector_elements, 1);
595 for (unsigned i = 0; i < vector_elements; i++) {
596 this->value.u[i] = u;
597 }
598 for (unsigned i = vector_elements; i < 16; i++) {
599 this->value.u[i] = 0;
600 }
601 }
602
ir_constant(int integer,unsigned vector_elements)603 ir_constant::ir_constant(int integer, unsigned vector_elements)
604 : ir_rvalue(ir_type_constant)
605 {
606 assert(vector_elements <= 4);
607 this->type = glsl_type::get_instance(GLSL_TYPE_INT, vector_elements, 1);
608 for (unsigned i = 0; i < vector_elements; i++) {
609 this->value.i[i] = integer;
610 }
611 for (unsigned i = vector_elements; i < 16; i++) {
612 this->value.i[i] = 0;
613 }
614 }
615
ir_constant(bool b,unsigned vector_elements)616 ir_constant::ir_constant(bool b, unsigned vector_elements)
617 : ir_rvalue(ir_type_constant)
618 {
619 assert(vector_elements <= 4);
620 this->type = glsl_type::get_instance(GLSL_TYPE_BOOL, vector_elements, 1);
621 for (unsigned i = 0; i < vector_elements; i++) {
622 this->value.b[i] = b;
623 }
624 for (unsigned i = vector_elements; i < 16; i++) {
625 this->value.b[i] = false;
626 }
627 }
628
ir_constant(const ir_constant * c,unsigned i)629 ir_constant::ir_constant(const ir_constant *c, unsigned i)
630 : ir_rvalue(ir_type_constant)
631 {
632 this->array_elements = NULL;
633 this->type = c->type->get_base_type();
634
635 switch (this->type->base_type) {
636 case GLSL_TYPE_UINT: this->value.u[0] = c->value.u[i]; break;
637 case GLSL_TYPE_INT: this->value.i[0] = c->value.i[i]; break;
638 case GLSL_TYPE_FLOAT: this->value.f[0] = c->value.f[i]; break;
639 case GLSL_TYPE_BOOL: this->value.b[0] = c->value.b[i]; break;
640 case GLSL_TYPE_DOUBLE: this->value.d[0] = c->value.d[i]; break;
641 default: assert(!"Should not get here."); break;
642 }
643 }
644
ir_constant(const struct glsl_type * type,exec_list * value_list)645 ir_constant::ir_constant(const struct glsl_type *type, exec_list *value_list)
646 : ir_rvalue(ir_type_constant)
647 {
648 this->array_elements = NULL;
649 this->type = type;
650
651 assert(type->is_scalar() || type->is_vector() || type->is_matrix()
652 || type->is_record() || type->is_array());
653
654 if (type->is_array()) {
655 this->array_elements = ralloc_array(this, ir_constant *, type->length);
656 unsigned i = 0;
657 foreach_in_list(ir_constant, value, value_list) {
658 assert(value->as_constant() != NULL);
659
660 this->array_elements[i++] = value;
661 }
662 return;
663 }
664
665 /* If the constant is a record, the types of each of the entries in
666 * value_list must be a 1-for-1 match with the structure components. Each
667 * entry must also be a constant. Just move the nodes from the value_list
668 * to the list in the ir_constant.
669 */
670 /* FINISHME: Should there be some type checking and / or assertions here? */
671 /* FINISHME: Should the new constant take ownership of the nodes from
672 * FINISHME: value_list, or should it make copies?
673 */
674 if (type->is_record()) {
675 value_list->move_nodes_to(& this->components);
676 return;
677 }
678
679 for (unsigned i = 0; i < 16; i++) {
680 this->value.u[i] = 0;
681 }
682
683 ir_constant *value = (ir_constant *) (value_list->get_head_raw());
684
685 /* Constructors with exactly one scalar argument are special for vectors
686 * and matrices. For vectors, the scalar value is replicated to fill all
687 * the components. For matrices, the scalar fills the components of the
688 * diagonal while the rest is filled with 0.
689 */
690 if (value->type->is_scalar() && value->next->is_tail_sentinel()) {
691 if (type->is_matrix()) {
692 /* Matrix - fill diagonal (rest is already set to 0) */
693 assert(type->base_type == GLSL_TYPE_FLOAT ||
694 type->base_type == GLSL_TYPE_DOUBLE);
695 for (unsigned i = 0; i < type->matrix_columns; i++) {
696 if (type->base_type == GLSL_TYPE_FLOAT)
697 this->value.f[i * type->vector_elements + i] =
698 value->value.f[0];
699 else
700 this->value.d[i * type->vector_elements + i] =
701 value->value.d[0];
702 }
703 } else {
704 /* Vector or scalar - fill all components */
705 switch (type->base_type) {
706 case GLSL_TYPE_UINT:
707 case GLSL_TYPE_INT:
708 for (unsigned i = 0; i < type->components(); i++)
709 this->value.u[i] = value->value.u[0];
710 break;
711 case GLSL_TYPE_FLOAT:
712 for (unsigned i = 0; i < type->components(); i++)
713 this->value.f[i] = value->value.f[0];
714 break;
715 case GLSL_TYPE_DOUBLE:
716 for (unsigned i = 0; i < type->components(); i++)
717 this->value.d[i] = value->value.d[0];
718 break;
719 case GLSL_TYPE_BOOL:
720 for (unsigned i = 0; i < type->components(); i++)
721 this->value.b[i] = value->value.b[0];
722 break;
723 default:
724 assert(!"Should not get here.");
725 break;
726 }
727 }
728 return;
729 }
730
731 if (type->is_matrix() && value->type->is_matrix()) {
732 assert(value->next->is_tail_sentinel());
733
734 /* From section 5.4.2 of the GLSL 1.20 spec:
735 * "If a matrix is constructed from a matrix, then each component
736 * (column i, row j) in the result that has a corresponding component
737 * (column i, row j) in the argument will be initialized from there."
738 */
739 unsigned cols = MIN2(type->matrix_columns, value->type->matrix_columns);
740 unsigned rows = MIN2(type->vector_elements, value->type->vector_elements);
741 for (unsigned i = 0; i < cols; i++) {
742 for (unsigned j = 0; j < rows; j++) {
743 const unsigned src = i * value->type->vector_elements + j;
744 const unsigned dst = i * type->vector_elements + j;
745 this->value.f[dst] = value->value.f[src];
746 }
747 }
748
749 /* "All other components will be initialized to the identity matrix." */
750 for (unsigned i = cols; i < type->matrix_columns; i++)
751 this->value.f[i * type->vector_elements + i] = 1.0;
752
753 return;
754 }
755
756 /* Use each component from each entry in the value_list to initialize one
757 * component of the constant being constructed.
758 */
759 unsigned i = 0;
760 for (;;) {
761 assert(value->as_constant() != NULL);
762 assert(!value->is_tail_sentinel());
763
764 for (unsigned j = 0; j < value->type->components(); j++) {
765 switch (type->base_type) {
766 case GLSL_TYPE_UINT:
767 this->value.u[i] = value->get_uint_component(j);
768 break;
769 case GLSL_TYPE_INT:
770 this->value.i[i] = value->get_int_component(j);
771 break;
772 case GLSL_TYPE_FLOAT:
773 this->value.f[i] = value->get_float_component(j);
774 break;
775 case GLSL_TYPE_BOOL:
776 this->value.b[i] = value->get_bool_component(j);
777 break;
778 case GLSL_TYPE_DOUBLE:
779 this->value.d[i] = value->get_double_component(j);
780 break;
781 default:
782 /* FINISHME: What to do? Exceptions are not the answer.
783 */
784 break;
785 }
786
787 i++;
788 if (i >= type->components())
789 break;
790 }
791
792 if (i >= type->components())
793 break; /* avoid downcasting a list sentinel */
794 value = (ir_constant *) value->next;
795 }
796 }
797
798 ir_constant *
zero(void * mem_ctx,const glsl_type * type)799 ir_constant::zero(void *mem_ctx, const glsl_type *type)
800 {
801 assert(type->is_scalar() || type->is_vector() || type->is_matrix()
802 || type->is_record() || type->is_array());
803
804 ir_constant *c = new(mem_ctx) ir_constant;
805 c->type = type;
806 memset(&c->value, 0, sizeof(c->value));
807
808 if (type->is_array()) {
809 c->array_elements = ralloc_array(c, ir_constant *, type->length);
810
811 for (unsigned i = 0; i < type->length; i++)
812 c->array_elements[i] = ir_constant::zero(c, type->fields.array);
813 }
814
815 if (type->is_record()) {
816 for (unsigned i = 0; i < type->length; i++) {
817 ir_constant *comp = ir_constant::zero(mem_ctx, type->fields.structure[i].type);
818 c->components.push_tail(comp);
819 }
820 }
821
822 return c;
823 }
824
825 bool
get_bool_component(unsigned i) const826 ir_constant::get_bool_component(unsigned i) const
827 {
828 switch (this->type->base_type) {
829 case GLSL_TYPE_UINT: return this->value.u[i] != 0;
830 case GLSL_TYPE_INT: return this->value.i[i] != 0;
831 case GLSL_TYPE_FLOAT: return ((int)this->value.f[i]) != 0;
832 case GLSL_TYPE_BOOL: return this->value.b[i];
833 case GLSL_TYPE_DOUBLE: return this->value.d[i] != 0.0;
834 default: assert(!"Should not get here."); break;
835 }
836
837 /* Must return something to make the compiler happy. This is clearly an
838 * error case.
839 */
840 return false;
841 }
842
843 float
get_float_component(unsigned i) const844 ir_constant::get_float_component(unsigned i) const
845 {
846 switch (this->type->base_type) {
847 case GLSL_TYPE_UINT: return (float) this->value.u[i];
848 case GLSL_TYPE_INT: return (float) this->value.i[i];
849 case GLSL_TYPE_FLOAT: return this->value.f[i];
850 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1.0f : 0.0f;
851 case GLSL_TYPE_DOUBLE: return (float) this->value.d[i];
852 default: assert(!"Should not get here."); break;
853 }
854
855 /* Must return something to make the compiler happy. This is clearly an
856 * error case.
857 */
858 return 0.0;
859 }
860
861 double
get_double_component(unsigned i) const862 ir_constant::get_double_component(unsigned i) const
863 {
864 switch (this->type->base_type) {
865 case GLSL_TYPE_UINT: return (double) this->value.u[i];
866 case GLSL_TYPE_INT: return (double) this->value.i[i];
867 case GLSL_TYPE_FLOAT: return (double) this->value.f[i];
868 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1.0 : 0.0;
869 case GLSL_TYPE_DOUBLE: return this->value.d[i];
870 default: assert(!"Should not get here."); break;
871 }
872
873 /* Must return something to make the compiler happy. This is clearly an
874 * error case.
875 */
876 return 0.0;
877 }
878
879 int
get_int_component(unsigned i) const880 ir_constant::get_int_component(unsigned i) const
881 {
882 switch (this->type->base_type) {
883 case GLSL_TYPE_UINT: return this->value.u[i];
884 case GLSL_TYPE_INT: return this->value.i[i];
885 case GLSL_TYPE_FLOAT: return (int) this->value.f[i];
886 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
887 case GLSL_TYPE_DOUBLE: return (int) this->value.d[i];
888 default: assert(!"Should not get here."); break;
889 }
890
891 /* Must return something to make the compiler happy. This is clearly an
892 * error case.
893 */
894 return 0;
895 }
896
897 unsigned
get_uint_component(unsigned i) const898 ir_constant::get_uint_component(unsigned i) const
899 {
900 switch (this->type->base_type) {
901 case GLSL_TYPE_UINT: return this->value.u[i];
902 case GLSL_TYPE_INT: return this->value.i[i];
903 case GLSL_TYPE_FLOAT: return (unsigned) this->value.f[i];
904 case GLSL_TYPE_BOOL: return this->value.b[i] ? 1 : 0;
905 case GLSL_TYPE_DOUBLE: return (unsigned) this->value.d[i];
906 default: assert(!"Should not get here."); break;
907 }
908
909 /* Must return something to make the compiler happy. This is clearly an
910 * error case.
911 */
912 return 0;
913 }
914
915 ir_constant *
get_array_element(unsigned i) const916 ir_constant::get_array_element(unsigned i) const
917 {
918 assert(this->type->is_array());
919
920 /* From page 35 (page 41 of the PDF) of the GLSL 1.20 spec:
921 *
922 * "Behavior is undefined if a shader subscripts an array with an index
923 * less than 0 or greater than or equal to the size the array was
924 * declared with."
925 *
926 * Most out-of-bounds accesses are removed before things could get this far.
927 * There are cases where non-constant array index values can get constant
928 * folded.
929 */
930 if (int(i) < 0)
931 i = 0;
932 else if (i >= this->type->length)
933 i = this->type->length - 1;
934
935 return array_elements[i];
936 }
937
938 ir_constant *
get_record_field(const char * name)939 ir_constant::get_record_field(const char *name)
940 {
941 int idx = this->type->field_index(name);
942
943 if (idx < 0)
944 return NULL;
945
946 if (this->components.is_empty())
947 return NULL;
948
949 exec_node *node = this->components.get_head_raw();
950 for (int i = 0; i < idx; i++) {
951 node = node->next;
952
953 /* If the end of the list is encountered before the element matching the
954 * requested field is found, return NULL.
955 */
956 if (node->is_tail_sentinel())
957 return NULL;
958 }
959
960 return (ir_constant *) node;
961 }
962
963 void
copy_offset(ir_constant * src,int offset)964 ir_constant::copy_offset(ir_constant *src, int offset)
965 {
966 switch (this->type->base_type) {
967 case GLSL_TYPE_UINT:
968 case GLSL_TYPE_INT:
969 case GLSL_TYPE_FLOAT:
970 case GLSL_TYPE_DOUBLE:
971 case GLSL_TYPE_BOOL: {
972 unsigned int size = src->type->components();
973 assert (size <= this->type->components() - offset);
974 for (unsigned int i=0; i<size; i++) {
975 switch (this->type->base_type) {
976 case GLSL_TYPE_UINT:
977 value.u[i+offset] = src->get_uint_component(i);
978 break;
979 case GLSL_TYPE_INT:
980 value.i[i+offset] = src->get_int_component(i);
981 break;
982 case GLSL_TYPE_FLOAT:
983 value.f[i+offset] = src->get_float_component(i);
984 break;
985 case GLSL_TYPE_BOOL:
986 value.b[i+offset] = src->get_bool_component(i);
987 break;
988 case GLSL_TYPE_DOUBLE:
989 value.d[i+offset] = src->get_double_component(i);
990 break;
991 default: // Shut up the compiler
992 break;
993 }
994 }
995 break;
996 }
997
998 case GLSL_TYPE_STRUCT: {
999 assert (src->type == this->type);
1000 this->components.make_empty();
1001 foreach_in_list(ir_constant, orig, &src->components) {
1002 this->components.push_tail(orig->clone(this, NULL));
1003 }
1004 break;
1005 }
1006
1007 case GLSL_TYPE_ARRAY: {
1008 assert (src->type == this->type);
1009 for (unsigned i = 0; i < this->type->length; i++) {
1010 this->array_elements[i] = src->array_elements[i]->clone(this, NULL);
1011 }
1012 break;
1013 }
1014
1015 default:
1016 assert(!"Should not get here.");
1017 break;
1018 }
1019 }
1020
1021 void
copy_masked_offset(ir_constant * src,int offset,unsigned int mask)1022 ir_constant::copy_masked_offset(ir_constant *src, int offset, unsigned int mask)
1023 {
1024 assert (!type->is_array() && !type->is_record());
1025
1026 if (!type->is_vector() && !type->is_matrix()) {
1027 offset = 0;
1028 mask = 1;
1029 }
1030
1031 int id = 0;
1032 for (int i=0; i<4; i++) {
1033 if (mask & (1 << i)) {
1034 switch (this->type->base_type) {
1035 case GLSL_TYPE_UINT:
1036 value.u[i+offset] = src->get_uint_component(id++);
1037 break;
1038 case GLSL_TYPE_INT:
1039 value.i[i+offset] = src->get_int_component(id++);
1040 break;
1041 case GLSL_TYPE_FLOAT:
1042 value.f[i+offset] = src->get_float_component(id++);
1043 break;
1044 case GLSL_TYPE_BOOL:
1045 value.b[i+offset] = src->get_bool_component(id++);
1046 break;
1047 case GLSL_TYPE_DOUBLE:
1048 value.d[i+offset] = src->get_double_component(id++);
1049 break;
1050 default:
1051 assert(!"Should not get here.");
1052 return;
1053 }
1054 }
1055 }
1056 }
1057
1058 bool
has_value(const ir_constant * c) const1059 ir_constant::has_value(const ir_constant *c) const
1060 {
1061 if (this->type != c->type)
1062 return false;
1063
1064 if (this->type->is_array()) {
1065 for (unsigned i = 0; i < this->type->length; i++) {
1066 if (!this->array_elements[i]->has_value(c->array_elements[i]))
1067 return false;
1068 }
1069 return true;
1070 }
1071
1072 if (this->type->base_type == GLSL_TYPE_STRUCT) {
1073 const exec_node *a_node = this->components.get_head_raw();
1074 const exec_node *b_node = c->components.get_head_raw();
1075
1076 while (!a_node->is_tail_sentinel()) {
1077 assert(!b_node->is_tail_sentinel());
1078
1079 const ir_constant *const a_field = (ir_constant *) a_node;
1080 const ir_constant *const b_field = (ir_constant *) b_node;
1081
1082 if (!a_field->has_value(b_field))
1083 return false;
1084
1085 a_node = a_node->next;
1086 b_node = b_node->next;
1087 }
1088
1089 return true;
1090 }
1091
1092 for (unsigned i = 0; i < this->type->components(); i++) {
1093 switch (this->type->base_type) {
1094 case GLSL_TYPE_UINT:
1095 if (this->value.u[i] != c->value.u[i])
1096 return false;
1097 break;
1098 case GLSL_TYPE_INT:
1099 if (this->value.i[i] != c->value.i[i])
1100 return false;
1101 break;
1102 case GLSL_TYPE_FLOAT:
1103 if (this->value.f[i] != c->value.f[i])
1104 return false;
1105 break;
1106 case GLSL_TYPE_BOOL:
1107 if (this->value.b[i] != c->value.b[i])
1108 return false;
1109 break;
1110 case GLSL_TYPE_DOUBLE:
1111 if (this->value.d[i] != c->value.d[i])
1112 return false;
1113 break;
1114 default:
1115 assert(!"Should not get here.");
1116 return false;
1117 }
1118 }
1119
1120 return true;
1121 }
1122
1123 bool
is_value(float f,int i) const1124 ir_constant::is_value(float f, int i) const
1125 {
1126 if (!this->type->is_scalar() && !this->type->is_vector())
1127 return false;
1128
1129 /* Only accept boolean values for 0/1. */
1130 if (int(bool(i)) != i && this->type->is_boolean())
1131 return false;
1132
1133 for (unsigned c = 0; c < this->type->vector_elements; c++) {
1134 switch (this->type->base_type) {
1135 case GLSL_TYPE_FLOAT:
1136 if (this->value.f[c] != f)
1137 return false;
1138 break;
1139 case GLSL_TYPE_INT:
1140 if (this->value.i[c] != i)
1141 return false;
1142 break;
1143 case GLSL_TYPE_UINT:
1144 if (this->value.u[c] != unsigned(i))
1145 return false;
1146 break;
1147 case GLSL_TYPE_BOOL:
1148 if (this->value.b[c] != bool(i))
1149 return false;
1150 break;
1151 case GLSL_TYPE_DOUBLE:
1152 if (this->value.d[c] != double(f))
1153 return false;
1154 break;
1155 default:
1156 /* The only other base types are structures, arrays, and samplers.
1157 * Samplers cannot be constants, and the others should have been
1158 * filtered out above.
1159 */
1160 assert(!"Should not get here.");
1161 return false;
1162 }
1163 }
1164
1165 return true;
1166 }
1167
1168 bool
is_zero() const1169 ir_constant::is_zero() const
1170 {
1171 return is_value(0.0, 0);
1172 }
1173
1174 bool
is_one() const1175 ir_constant::is_one() const
1176 {
1177 return is_value(1.0, 1);
1178 }
1179
1180 bool
is_negative_one() const1181 ir_constant::is_negative_one() const
1182 {
1183 return is_value(-1.0, -1);
1184 }
1185
1186 bool
is_uint16_constant() const1187 ir_constant::is_uint16_constant() const
1188 {
1189 if (!type->is_integer())
1190 return false;
1191
1192 return value.u[0] < (1 << 16);
1193 }
1194
ir_loop()1195 ir_loop::ir_loop()
1196 : ir_instruction(ir_type_loop)
1197 {
1198 }
1199
1200
ir_dereference_variable(ir_variable * var)1201 ir_dereference_variable::ir_dereference_variable(ir_variable *var)
1202 : ir_dereference(ir_type_dereference_variable)
1203 {
1204 assert(var != NULL);
1205
1206 this->var = var;
1207 this->type = var->type;
1208 }
1209
1210
ir_dereference_array(ir_rvalue * value,ir_rvalue * array_index)1211 ir_dereference_array::ir_dereference_array(ir_rvalue *value,
1212 ir_rvalue *array_index)
1213 : ir_dereference(ir_type_dereference_array)
1214 {
1215 this->array_index = array_index;
1216 this->set_array(value);
1217 }
1218
1219
ir_dereference_array(ir_variable * var,ir_rvalue * array_index)1220 ir_dereference_array::ir_dereference_array(ir_variable *var,
1221 ir_rvalue *array_index)
1222 : ir_dereference(ir_type_dereference_array)
1223 {
1224 void *ctx = ralloc_parent(var);
1225
1226 this->array_index = array_index;
1227 this->set_array(new(ctx) ir_dereference_variable(var));
1228 }
1229
1230
1231 void
set_array(ir_rvalue * value)1232 ir_dereference_array::set_array(ir_rvalue *value)
1233 {
1234 assert(value != NULL);
1235
1236 this->array = value;
1237
1238 const glsl_type *const vt = this->array->type;
1239
1240 if (vt->is_array()) {
1241 type = vt->fields.array;
1242 } else if (vt->is_matrix()) {
1243 type = vt->column_type();
1244 } else if (vt->is_vector()) {
1245 type = vt->get_base_type();
1246 }
1247 }
1248
1249
ir_dereference_record(ir_rvalue * value,const char * field)1250 ir_dereference_record::ir_dereference_record(ir_rvalue *value,
1251 const char *field)
1252 : ir_dereference(ir_type_dereference_record)
1253 {
1254 assert(value != NULL);
1255
1256 this->record = value;
1257 this->field = ralloc_strdup(this, field);
1258 this->type = this->record->type->field_type(field);
1259 }
1260
1261
ir_dereference_record(ir_variable * var,const char * field)1262 ir_dereference_record::ir_dereference_record(ir_variable *var,
1263 const char *field)
1264 : ir_dereference(ir_type_dereference_record)
1265 {
1266 void *ctx = ralloc_parent(var);
1267
1268 this->record = new(ctx) ir_dereference_variable(var);
1269 this->field = ralloc_strdup(this, field);
1270 this->type = this->record->type->field_type(field);
1271 }
1272
1273 bool
is_lvalue() const1274 ir_dereference::is_lvalue() const
1275 {
1276 ir_variable *var = this->variable_referenced();
1277
1278 /* Every l-value derference chain eventually ends in a variable.
1279 */
1280 if ((var == NULL) || var->data.read_only)
1281 return false;
1282
1283 /* From section 4.1.7 of the GLSL 4.40 spec:
1284 *
1285 * "Opaque variables cannot be treated as l-values; hence cannot
1286 * be used as out or inout function parameters, nor can they be
1287 * assigned into."
1288 */
1289 if (this->type->contains_opaque())
1290 return false;
1291
1292 return true;
1293 }
1294
1295
1296 static const char * const tex_opcode_strs[] = { "tex", "txb", "txl", "txd", "txf", "txf_ms", "txs", "lod", "tg4", "query_levels", "texture_samples", "samples_identical" };
1297
opcode_string()1298 const char *ir_texture::opcode_string()
1299 {
1300 assert((unsigned int) op < ARRAY_SIZE(tex_opcode_strs));
1301 return tex_opcode_strs[op];
1302 }
1303
1304 ir_texture_opcode
get_opcode(const char * str)1305 ir_texture::get_opcode(const char *str)
1306 {
1307 const int count = sizeof(tex_opcode_strs) / sizeof(tex_opcode_strs[0]);
1308 for (int op = 0; op < count; op++) {
1309 if (strcmp(str, tex_opcode_strs[op]) == 0)
1310 return (ir_texture_opcode) op;
1311 }
1312 return (ir_texture_opcode) -1;
1313 }
1314
1315
1316 void
set_sampler(ir_dereference * sampler,const glsl_type * type)1317 ir_texture::set_sampler(ir_dereference *sampler, const glsl_type *type)
1318 {
1319 assert(sampler != NULL);
1320 assert(type != NULL);
1321 this->sampler = sampler;
1322 this->type = type;
1323
1324 if (this->op == ir_txs || this->op == ir_query_levels ||
1325 this->op == ir_texture_samples) {
1326 assert(type->base_type == GLSL_TYPE_INT);
1327 } else if (this->op == ir_lod) {
1328 assert(type->vector_elements == 2);
1329 assert(type->base_type == GLSL_TYPE_FLOAT);
1330 } else if (this->op == ir_samples_identical) {
1331 assert(type == glsl_type::bool_type);
1332 assert(sampler->type->base_type == GLSL_TYPE_SAMPLER);
1333 assert(sampler->type->sampler_dimensionality == GLSL_SAMPLER_DIM_MS);
1334 } else {
1335 assert(sampler->type->sampled_type == (int) type->base_type);
1336 if (sampler->type->sampler_shadow)
1337 assert(type->vector_elements == 4 || type->vector_elements == 1);
1338 else
1339 assert(type->vector_elements == 4);
1340 }
1341 }
1342
1343
1344 void
init_mask(const unsigned * comp,unsigned count)1345 ir_swizzle::init_mask(const unsigned *comp, unsigned count)
1346 {
1347 assert((count >= 1) && (count <= 4));
1348
1349 memset(&this->mask, 0, sizeof(this->mask));
1350 this->mask.num_components = count;
1351
1352 unsigned dup_mask = 0;
1353 switch (count) {
1354 case 4:
1355 assert(comp[3] <= 3);
1356 dup_mask |= (1U << comp[3])
1357 & ((1U << comp[0]) | (1U << comp[1]) | (1U << comp[2]));
1358 this->mask.w = comp[3];
1359
1360 case 3:
1361 assert(comp[2] <= 3);
1362 dup_mask |= (1U << comp[2])
1363 & ((1U << comp[0]) | (1U << comp[1]));
1364 this->mask.z = comp[2];
1365
1366 case 2:
1367 assert(comp[1] <= 3);
1368 dup_mask |= (1U << comp[1])
1369 & ((1U << comp[0]));
1370 this->mask.y = comp[1];
1371
1372 case 1:
1373 assert(comp[0] <= 3);
1374 this->mask.x = comp[0];
1375 }
1376
1377 this->mask.has_duplicates = dup_mask != 0;
1378
1379 /* Based on the number of elements in the swizzle and the base type
1380 * (i.e., float, int, unsigned, or bool) of the vector being swizzled,
1381 * generate the type of the resulting value.
1382 */
1383 type = glsl_type::get_instance(val->type->base_type, mask.num_components, 1);
1384 }
1385
ir_swizzle(ir_rvalue * val,unsigned x,unsigned y,unsigned z,unsigned w,unsigned count)1386 ir_swizzle::ir_swizzle(ir_rvalue *val, unsigned x, unsigned y, unsigned z,
1387 unsigned w, unsigned count)
1388 : ir_rvalue(ir_type_swizzle), val(val)
1389 {
1390 const unsigned components[4] = { x, y, z, w };
1391 this->init_mask(components, count);
1392 }
1393
ir_swizzle(ir_rvalue * val,const unsigned * comp,unsigned count)1394 ir_swizzle::ir_swizzle(ir_rvalue *val, const unsigned *comp,
1395 unsigned count)
1396 : ir_rvalue(ir_type_swizzle), val(val)
1397 {
1398 this->init_mask(comp, count);
1399 }
1400
ir_swizzle(ir_rvalue * val,ir_swizzle_mask mask)1401 ir_swizzle::ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask)
1402 : ir_rvalue(ir_type_swizzle)
1403 {
1404 this->val = val;
1405 this->mask = mask;
1406 this->type = glsl_type::get_instance(val->type->base_type,
1407 mask.num_components, 1);
1408 }
1409
1410 #define X 1
1411 #define R 5
1412 #define S 9
1413 #define I 13
1414
1415 ir_swizzle *
create(ir_rvalue * val,const char * str,unsigned vector_length)1416 ir_swizzle::create(ir_rvalue *val, const char *str, unsigned vector_length)
1417 {
1418 void *ctx = ralloc_parent(val);
1419
1420 /* For each possible swizzle character, this table encodes the value in
1421 * \c idx_map that represents the 0th element of the vector. For invalid
1422 * swizzle characters (e.g., 'k'), a special value is used that will allow
1423 * detection of errors.
1424 */
1425 static const unsigned char base_idx[26] = {
1426 /* a b c d e f g h i j k l m */
1427 R, R, I, I, I, I, R, I, I, I, I, I, I,
1428 /* n o p q r s t u v w x y z */
1429 I, I, S, S, R, S, S, I, I, X, X, X, X
1430 };
1431
1432 /* Each valid swizzle character has an entry in the previous table. This
1433 * table encodes the base index encoded in the previous table plus the actual
1434 * index of the swizzle character. When processing swizzles, the first
1435 * character in the string is indexed in the previous table. Each character
1436 * in the string is indexed in this table, and the value found there has the
1437 * value form the first table subtracted. The result must be on the range
1438 * [0,3].
1439 *
1440 * For example, the string "wzyx" will get X from the first table. Each of
1441 * the charcaters will get X+3, X+2, X+1, and X+0 from this table. After
1442 * subtraction, the swizzle values are { 3, 2, 1, 0 }.
1443 *
1444 * The string "wzrg" will get X from the first table. Each of the characters
1445 * will get X+3, X+2, R+0, and R+1 from this table. After subtraction, the
1446 * swizzle values are { 3, 2, 4, 5 }. Since 4 and 5 are outside the range
1447 * [0,3], the error is detected.
1448 */
1449 static const unsigned char idx_map[26] = {
1450 /* a b c d e f g h i j k l m */
1451 R+3, R+2, 0, 0, 0, 0, R+1, 0, 0, 0, 0, 0, 0,
1452 /* n o p q r s t u v w x y z */
1453 0, 0, S+2, S+3, R+0, S+0, S+1, 0, 0, X+3, X+0, X+1, X+2
1454 };
1455
1456 int swiz_idx[4] = { 0, 0, 0, 0 };
1457 unsigned i;
1458
1459
1460 /* Validate the first character in the swizzle string and look up the base
1461 * index value as described above.
1462 */
1463 if ((str[0] < 'a') || (str[0] > 'z'))
1464 return NULL;
1465
1466 const unsigned base = base_idx[str[0] - 'a'];
1467
1468
1469 for (i = 0; (i < 4) && (str[i] != '\0'); i++) {
1470 /* Validate the next character, and, as described above, convert it to a
1471 * swizzle index.
1472 */
1473 if ((str[i] < 'a') || (str[i] > 'z'))
1474 return NULL;
1475
1476 swiz_idx[i] = idx_map[str[i] - 'a'] - base;
1477 if ((swiz_idx[i] < 0) || (swiz_idx[i] >= (int) vector_length))
1478 return NULL;
1479 }
1480
1481 if (str[i] != '\0')
1482 return NULL;
1483
1484 return new(ctx) ir_swizzle(val, swiz_idx[0], swiz_idx[1], swiz_idx[2],
1485 swiz_idx[3], i);
1486 }
1487
1488 #undef X
1489 #undef R
1490 #undef S
1491 #undef I
1492
1493 ir_variable *
variable_referenced() const1494 ir_swizzle::variable_referenced() const
1495 {
1496 return this->val->variable_referenced();
1497 }
1498
1499
1500 bool ir_variable::temporaries_allocate_names = false;
1501
1502 const char ir_variable::tmp_name[] = "compiler_temp";
1503
ir_variable(const struct glsl_type * type,const char * name,ir_variable_mode mode)1504 ir_variable::ir_variable(const struct glsl_type *type, const char *name,
1505 ir_variable_mode mode)
1506 : ir_instruction(ir_type_variable)
1507 {
1508 this->type = type;
1509
1510 if (mode == ir_var_temporary && !ir_variable::temporaries_allocate_names)
1511 name = NULL;
1512
1513 /* The ir_variable clone method may call this constructor with name set to
1514 * tmp_name.
1515 */
1516 assert(name != NULL
1517 || mode == ir_var_temporary
1518 || mode == ir_var_function_in
1519 || mode == ir_var_function_out
1520 || mode == ir_var_function_inout);
1521 assert(name != ir_variable::tmp_name
1522 || mode == ir_var_temporary);
1523 if (mode == ir_var_temporary
1524 && (name == NULL || name == ir_variable::tmp_name)) {
1525 this->name = ir_variable::tmp_name;
1526 } else if (name == NULL ||
1527 strlen(name) < ARRAY_SIZE(this->name_storage)) {
1528 strcpy(this->name_storage, name ? name : "");
1529 this->name = this->name_storage;
1530 } else {
1531 this->name = ralloc_strdup(this, name);
1532 }
1533
1534 this->u.max_ifc_array_access = NULL;
1535
1536 this->data.explicit_location = false;
1537 this->data.has_initializer = false;
1538 this->data.location = -1;
1539 this->data.location_frac = 0;
1540 this->data.binding = 0;
1541 this->data.warn_extension_index = 0;
1542 this->constant_value = NULL;
1543 this->constant_initializer = NULL;
1544 this->data.origin_upper_left = false;
1545 this->data.pixel_center_integer = false;
1546 this->data.depth_layout = ir_depth_layout_none;
1547 this->data.used = false;
1548 this->data.always_active_io = false;
1549 this->data.read_only = false;
1550 this->data.centroid = false;
1551 this->data.sample = false;
1552 this->data.patch = false;
1553 this->data.invariant = false;
1554 this->data.how_declared = ir_var_declared_normally;
1555 this->data.mode = mode;
1556 this->data.interpolation = INTERP_MODE_NONE;
1557 this->data.max_array_access = -1;
1558 this->data.offset = 0;
1559 this->data.precision = GLSL_PRECISION_NONE;
1560 this->data.image_read_only = false;
1561 this->data.image_write_only = false;
1562 this->data.image_coherent = false;
1563 this->data.image_volatile = false;
1564 this->data.image_restrict = false;
1565 this->data.from_ssbo_unsized_array = false;
1566 this->data.fb_fetch_output = false;
1567
1568 if (type != NULL) {
1569 if (type->base_type == GLSL_TYPE_SAMPLER)
1570 this->data.read_only = true;
1571
1572 if (type->is_interface())
1573 this->init_interface_type(type);
1574 else if (type->without_array()->is_interface())
1575 this->init_interface_type(type->without_array());
1576 }
1577 }
1578
1579
1580 const char *
interpolation_string(unsigned interpolation)1581 interpolation_string(unsigned interpolation)
1582 {
1583 switch (interpolation) {
1584 case INTERP_MODE_NONE: return "no";
1585 case INTERP_MODE_SMOOTH: return "smooth";
1586 case INTERP_MODE_FLAT: return "flat";
1587 case INTERP_MODE_NOPERSPECTIVE: return "noperspective";
1588 }
1589
1590 assert(!"Should not get here.");
1591 return "";
1592 }
1593
1594 const char *const ir_variable::warn_extension_table[] = {
1595 "",
1596 "GL_ARB_shader_stencil_export",
1597 "GL_AMD_shader_stencil_export",
1598 };
1599
1600 void
enable_extension_warning(const char * extension)1601 ir_variable::enable_extension_warning(const char *extension)
1602 {
1603 for (unsigned i = 0; i < ARRAY_SIZE(warn_extension_table); i++) {
1604 if (strcmp(warn_extension_table[i], extension) == 0) {
1605 this->data.warn_extension_index = i;
1606 return;
1607 }
1608 }
1609
1610 assert(!"Should not get here.");
1611 this->data.warn_extension_index = 0;
1612 }
1613
1614 const char *
get_extension_warning() const1615 ir_variable::get_extension_warning() const
1616 {
1617 return this->data.warn_extension_index == 0
1618 ? NULL : warn_extension_table[this->data.warn_extension_index];
1619 }
1620
ir_function_signature(const glsl_type * return_type,builtin_available_predicate b)1621 ir_function_signature::ir_function_signature(const glsl_type *return_type,
1622 builtin_available_predicate b)
1623 : ir_instruction(ir_type_function_signature),
1624 return_type(return_type), is_defined(false),
1625 intrinsic_id(ir_intrinsic_invalid), builtin_avail(b), _function(NULL)
1626 {
1627 this->origin = NULL;
1628 }
1629
1630
1631 bool
is_builtin() const1632 ir_function_signature::is_builtin() const
1633 {
1634 return builtin_avail != NULL;
1635 }
1636
1637
1638 bool
is_builtin_available(const _mesa_glsl_parse_state * state) const1639 ir_function_signature::is_builtin_available(const _mesa_glsl_parse_state *state) const
1640 {
1641 /* We can't call the predicate without a state pointer, so just say that
1642 * the signature is available. At compile time, we need the filtering,
1643 * but also receive a valid state pointer. At link time, we're resolving
1644 * imported built-in prototypes to their definitions, which will always
1645 * be an exact match. So we can skip the filtering.
1646 */
1647 if (state == NULL)
1648 return true;
1649
1650 assert(builtin_avail != NULL);
1651 return builtin_avail(state);
1652 }
1653
1654
1655 static bool
modes_match(unsigned a,unsigned b)1656 modes_match(unsigned a, unsigned b)
1657 {
1658 if (a == b)
1659 return true;
1660
1661 /* Accept "in" vs. "const in" */
1662 if ((a == ir_var_const_in && b == ir_var_function_in) ||
1663 (b == ir_var_const_in && a == ir_var_function_in))
1664 return true;
1665
1666 return false;
1667 }
1668
1669
1670 const char *
qualifiers_match(exec_list * params)1671 ir_function_signature::qualifiers_match(exec_list *params)
1672 {
1673 /* check that the qualifiers match. */
1674 foreach_two_lists(a_node, &this->parameters, b_node, params) {
1675 ir_variable *a = (ir_variable *) a_node;
1676 ir_variable *b = (ir_variable *) b_node;
1677
1678 if (a->data.read_only != b->data.read_only ||
1679 !modes_match(a->data.mode, b->data.mode) ||
1680 a->data.interpolation != b->data.interpolation ||
1681 a->data.centroid != b->data.centroid ||
1682 a->data.sample != b->data.sample ||
1683 a->data.patch != b->data.patch ||
1684 a->data.image_read_only != b->data.image_read_only ||
1685 a->data.image_write_only != b->data.image_write_only ||
1686 a->data.image_coherent != b->data.image_coherent ||
1687 a->data.image_volatile != b->data.image_volatile ||
1688 a->data.image_restrict != b->data.image_restrict) {
1689
1690 /* parameter a's qualifiers don't match */
1691 return a->name;
1692 }
1693 }
1694 return NULL;
1695 }
1696
1697
1698 void
replace_parameters(exec_list * new_params)1699 ir_function_signature::replace_parameters(exec_list *new_params)
1700 {
1701 /* Destroy all of the previous parameter information. If the previous
1702 * parameter information comes from the function prototype, it may either
1703 * specify incorrect parameter names or not have names at all.
1704 */
1705 new_params->move_nodes_to(¶meters);
1706 }
1707
1708
ir_function(const char * name)1709 ir_function::ir_function(const char *name)
1710 : ir_instruction(ir_type_function)
1711 {
1712 this->subroutine_index = -1;
1713 this->name = ralloc_strdup(this, name);
1714 }
1715
1716
1717 bool
has_user_signature()1718 ir_function::has_user_signature()
1719 {
1720 foreach_in_list(ir_function_signature, sig, &this->signatures) {
1721 if (!sig->is_builtin())
1722 return true;
1723 }
1724 return false;
1725 }
1726
1727
1728 ir_rvalue *
error_value(void * mem_ctx)1729 ir_rvalue::error_value(void *mem_ctx)
1730 {
1731 ir_rvalue *v = new(mem_ctx) ir_rvalue(ir_type_unset);
1732
1733 v->type = glsl_type::error_type;
1734 return v;
1735 }
1736
1737
1738 void
visit_exec_list(exec_list * list,ir_visitor * visitor)1739 visit_exec_list(exec_list *list, ir_visitor *visitor)
1740 {
1741 foreach_in_list_safe(ir_instruction, node, list) {
1742 node->accept(visitor);
1743 }
1744 }
1745
1746
1747 static void
steal_memory(ir_instruction * ir,void * new_ctx)1748 steal_memory(ir_instruction *ir, void *new_ctx)
1749 {
1750 ir_variable *var = ir->as_variable();
1751 ir_function *fn = ir->as_function();
1752 ir_constant *constant = ir->as_constant();
1753 if (var != NULL && var->constant_value != NULL)
1754 steal_memory(var->constant_value, ir);
1755
1756 if (var != NULL && var->constant_initializer != NULL)
1757 steal_memory(var->constant_initializer, ir);
1758
1759 if (fn != NULL && fn->subroutine_types)
1760 ralloc_steal(new_ctx, fn->subroutine_types);
1761
1762 /* The components of aggregate constants are not visited by the normal
1763 * visitor, so steal their values by hand.
1764 */
1765 if (constant != NULL) {
1766 if (constant->type->is_record()) {
1767 foreach_in_list(ir_constant, field, &constant->components) {
1768 steal_memory(field, ir);
1769 }
1770 } else if (constant->type->is_array()) {
1771 for (unsigned int i = 0; i < constant->type->length; i++) {
1772 steal_memory(constant->array_elements[i], ir);
1773 }
1774 }
1775 }
1776
1777 ralloc_steal(new_ctx, ir);
1778 }
1779
1780
1781 void
reparent_ir(exec_list * list,void * mem_ctx)1782 reparent_ir(exec_list *list, void *mem_ctx)
1783 {
1784 foreach_in_list(ir_instruction, node, list) {
1785 visit_tree(node, steal_memory, mem_ctx);
1786 }
1787 }
1788
1789
1790 static ir_rvalue *
try_min_one(ir_rvalue * ir)1791 try_min_one(ir_rvalue *ir)
1792 {
1793 ir_expression *expr = ir->as_expression();
1794
1795 if (!expr || expr->operation != ir_binop_min)
1796 return NULL;
1797
1798 if (expr->operands[0]->is_one())
1799 return expr->operands[1];
1800
1801 if (expr->operands[1]->is_one())
1802 return expr->operands[0];
1803
1804 return NULL;
1805 }
1806
1807 static ir_rvalue *
try_max_zero(ir_rvalue * ir)1808 try_max_zero(ir_rvalue *ir)
1809 {
1810 ir_expression *expr = ir->as_expression();
1811
1812 if (!expr || expr->operation != ir_binop_max)
1813 return NULL;
1814
1815 if (expr->operands[0]->is_zero())
1816 return expr->operands[1];
1817
1818 if (expr->operands[1]->is_zero())
1819 return expr->operands[0];
1820
1821 return NULL;
1822 }
1823
1824 ir_rvalue *
as_rvalue_to_saturate()1825 ir_rvalue::as_rvalue_to_saturate()
1826 {
1827 ir_expression *expr = this->as_expression();
1828
1829 if (!expr)
1830 return NULL;
1831
1832 ir_rvalue *max_zero = try_max_zero(expr);
1833 if (max_zero) {
1834 return try_min_one(max_zero);
1835 } else {
1836 ir_rvalue *min_one = try_min_one(expr);
1837 if (min_one) {
1838 return try_max_zero(min_one);
1839 }
1840 }
1841
1842 return NULL;
1843 }
1844
1845
1846 unsigned
vertices_per_prim(GLenum prim)1847 vertices_per_prim(GLenum prim)
1848 {
1849 switch (prim) {
1850 case GL_POINTS:
1851 return 1;
1852 case GL_LINES:
1853 return 2;
1854 case GL_TRIANGLES:
1855 return 3;
1856 case GL_LINES_ADJACENCY:
1857 return 4;
1858 case GL_TRIANGLES_ADJACENCY:
1859 return 6;
1860 default:
1861 assert(!"Bad primitive");
1862 return 3;
1863 }
1864 }
1865
1866 /**
1867 * Generate a string describing the mode of a variable
1868 */
1869 const char *
mode_string(const ir_variable * var)1870 mode_string(const ir_variable *var)
1871 {
1872 switch (var->data.mode) {
1873 case ir_var_auto:
1874 return (var->data.read_only) ? "global constant" : "global variable";
1875
1876 case ir_var_uniform:
1877 return "uniform";
1878
1879 case ir_var_shader_storage:
1880 return "buffer";
1881
1882 case ir_var_shader_in:
1883 return "shader input";
1884
1885 case ir_var_shader_out:
1886 return "shader output";
1887
1888 case ir_var_function_in:
1889 case ir_var_const_in:
1890 return "function input";
1891
1892 case ir_var_function_out:
1893 return "function output";
1894
1895 case ir_var_function_inout:
1896 return "function inout";
1897
1898 case ir_var_system_value:
1899 return "shader input";
1900
1901 case ir_var_temporary:
1902 return "compiler temporary";
1903
1904 case ir_var_mode_count:
1905 break;
1906 }
1907
1908 assert(!"Should not get here.");
1909 return "invalid variable";
1910 }
1911