b707634ea5ab0bb64516fd1802bc88c46e935ab1
[mesa.git] / src / glsl / ir.h
1 /* -*- c++ -*- */
2 /*
3 * Copyright © 2010 Intel Corporation
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
14 * Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
21 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
22 * DEALINGS IN THE SOFTWARE.
23 */
24
25 #pragma once
26 #ifndef IR_H
27 #define IR_H
28
29 #include <stdio.h>
30 #include <stdlib.h>
31
32 #include "ralloc.h"
33 #include "glsl_types.h"
34 #include "list.h"
35 #include "ir_visitor.h"
36 #include "ir_hierarchical_visitor.h"
37
38 /**
39 * \defgroup IR Intermediate representation nodes
40 *
41 * @{
42 */
43
44 /**
45 * Class tags
46 *
47 * Each concrete class derived from \c ir_instruction has a value in this
48 * enumerant. The value for the type is stored in \c ir_instruction::ir_type
49 * by the constructor. While using type tags is not very C++, it is extremely
50 * convenient. For example, during debugging you can simply inspect
51 * \c ir_instruction::ir_type to find out the actual type of the object.
52 *
53 * In addition, it is possible to use a switch-statement based on \c
54 * \c ir_instruction::ir_type to select different behavior for different object
55 * types. For functions that have only slight differences for several object
56 * types, this allows writing very straightforward, readable code.
57 */
58 enum ir_node_type {
59 /**
60 * Zero is unused so that the IR validator can detect cases where
61 * \c ir_instruction::ir_type has not been initialized.
62 */
63 ir_type_unset,
64 ir_type_variable,
65 ir_type_assignment,
66 ir_type_call,
67 ir_type_constant,
68 ir_type_dereference_array,
69 ir_type_dereference_record,
70 ir_type_dereference_variable,
71 ir_type_discard,
72 ir_type_expression,
73 ir_type_function,
74 ir_type_function_signature,
75 ir_type_if,
76 ir_type_loop,
77 ir_type_loop_jump,
78 ir_type_return,
79 ir_type_swizzle,
80 ir_type_texture,
81 ir_type_max /**< maximum ir_type enum number, for validation */
82 };
83
84 /**
85 * Base class of all IR instructions
86 */
87 class ir_instruction : public exec_node {
88 public:
89 enum ir_node_type ir_type;
90 const struct glsl_type *type;
91
92 /** ir_print_visitor helper for debugging. */
93 void print(void) const;
94
95 virtual void accept(ir_visitor *) = 0;
96 virtual ir_visitor_status accept(ir_hierarchical_visitor *) = 0;
97 virtual ir_instruction *clone(void *mem_ctx,
98 struct hash_table *ht) const = 0;
99
100 /**
101 * \name IR instruction downcast functions
102 *
103 * These functions either cast the object to a derived class or return
104 * \c NULL if the object's type does not match the specified derived class.
105 * Additional downcast functions will be added as needed.
106 */
107 /*@{*/
108 virtual class ir_variable * as_variable() { return NULL; }
109 virtual class ir_function * as_function() { return NULL; }
110 virtual class ir_dereference * as_dereference() { return NULL; }
111 virtual class ir_dereference_array * as_dereference_array() { return NULL; }
112 virtual class ir_dereference_variable *as_dereference_variable() { return NULL; }
113 virtual class ir_expression * as_expression() { return NULL; }
114 virtual class ir_rvalue * as_rvalue() { return NULL; }
115 virtual class ir_loop * as_loop() { return NULL; }
116 virtual class ir_assignment * as_assignment() { return NULL; }
117 virtual class ir_call * as_call() { return NULL; }
118 virtual class ir_return * as_return() { return NULL; }
119 virtual class ir_if * as_if() { return NULL; }
120 virtual class ir_swizzle * as_swizzle() { return NULL; }
121 virtual class ir_constant * as_constant() { return NULL; }
122 virtual class ir_discard * as_discard() { return NULL; }
123 /*@}*/
124
125 protected:
126 ir_instruction()
127 {
128 ir_type = ir_type_unset;
129 type = NULL;
130 }
131 };
132
133
134 class ir_rvalue : public ir_instruction {
135 public:
136 virtual ir_rvalue *clone(void *mem_ctx, struct hash_table *) const = 0;
137
138 virtual ir_constant *constant_expression_value() = 0;
139
140 virtual ir_rvalue * as_rvalue()
141 {
142 return this;
143 }
144
145 ir_rvalue *as_rvalue_to_saturate();
146
147 virtual bool is_lvalue() const
148 {
149 return false;
150 }
151
152 /**
153 * Get the variable that is ultimately referenced by an r-value
154 */
155 virtual ir_variable *variable_referenced() const
156 {
157 return NULL;
158 }
159
160
161 /**
162 * If an r-value is a reference to a whole variable, get that variable
163 *
164 * \return
165 * Pointer to a variable that is completely dereferenced by the r-value. If
166 * the r-value is not a dereference or the dereference does not access the
167 * entire variable (i.e., it's just one array element, struct field), \c NULL
168 * is returned.
169 */
170 virtual ir_variable *whole_variable_referenced()
171 {
172 return NULL;
173 }
174
175 /**
176 * Determine if an r-value has the value zero
177 *
178 * The base implementation of this function always returns \c false. The
179 * \c ir_constant class over-rides this function to return \c true \b only
180 * for vector and scalar types that have all elements set to the value
181 * zero (or \c false for booleans).
182 *
183 * \sa ir_constant::has_value, ir_rvalue::is_one, ir_rvalue::is_negative_one
184 */
185 virtual bool is_zero() const;
186
187 /**
188 * Determine if an r-value has the value one
189 *
190 * The base implementation of this function always returns \c false. The
191 * \c ir_constant class over-rides this function to return \c true \b only
192 * for vector and scalar types that have all elements set to the value
193 * one (or \c true for booleans).
194 *
195 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_negative_one
196 */
197 virtual bool is_one() const;
198
199 /**
200 * Determine if an r-value has the value negative one
201 *
202 * The base implementation of this function always returns \c false. The
203 * \c ir_constant class over-rides this function to return \c true \b only
204 * for vector and scalar types that have all elements set to the value
205 * negative one. For boolean times, the result is always \c false.
206 *
207 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one
208 */
209 virtual bool is_negative_one() const;
210
211 protected:
212 ir_rvalue();
213 };
214
215
216 /**
217 * Variable storage classes
218 */
219 enum ir_variable_mode {
220 ir_var_auto = 0, /**< Function local variables and globals. */
221 ir_var_uniform, /**< Variable declared as a uniform. */
222 ir_var_in,
223 ir_var_out,
224 ir_var_inout,
225 ir_var_const_in, /**< "in" param that must be a constant expression */
226 ir_var_system_value, /**< Ex: front-face, instance-id, etc. */
227 ir_var_temporary /**< Temporary variable generated during compilation. */
228 };
229
230 enum ir_variable_interpolation {
231 ir_var_smooth = 0,
232 ir_var_flat,
233 ir_var_noperspective
234 };
235
236 /**
237 * \brief Layout qualifiers for gl_FragDepth.
238 *
239 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
240 * with a layout qualifier.
241 */
242 enum ir_depth_layout {
243 ir_depth_layout_none, /**< No depth layout is specified. */
244 ir_depth_layout_any,
245 ir_depth_layout_greater,
246 ir_depth_layout_less,
247 ir_depth_layout_unchanged
248 };
249
250 /**
251 * \brief Convert depth layout qualifier to string.
252 */
253 const char*
254 depth_layout_string(ir_depth_layout layout);
255
256 /**
257 * Description of built-in state associated with a uniform
258 *
259 * \sa ir_variable::state_slots
260 */
261 struct ir_state_slot {
262 int tokens[5];
263 int swizzle;
264 };
265
266 class ir_variable : public ir_instruction {
267 public:
268 ir_variable(const struct glsl_type *, const char *, ir_variable_mode);
269
270 virtual ir_variable *clone(void *mem_ctx, struct hash_table *ht) const;
271
272 virtual ir_variable *as_variable()
273 {
274 return this;
275 }
276
277 virtual void accept(ir_visitor *v)
278 {
279 v->visit(this);
280 }
281
282 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
283
284
285 /**
286 * Get the string value for the interpolation qualifier
287 *
288 * \return The string that would be used in a shader to specify \c
289 * mode will be returned.
290 *
291 * This function should only be used on a shader input or output variable.
292 */
293 const char *interpolation_string() const;
294
295 /**
296 * Delcared name of the variable
297 */
298 const char *name;
299
300 /**
301 * Highest element accessed with a constant expression array index
302 *
303 * Not used for non-array variables.
304 */
305 unsigned max_array_access;
306
307 /**
308 * Is the variable read-only?
309 *
310 * This is set for variables declared as \c const, shader inputs,
311 * and uniforms.
312 */
313 unsigned read_only:1;
314 unsigned centroid:1;
315 unsigned invariant:1;
316
317 /**
318 * Has this variable been used for reading or writing?
319 *
320 * Several GLSL semantic checks require knowledge of whether or not a
321 * variable has been used. For example, it is an error to redeclare a
322 * variable as invariant after it has been used.
323 */
324 unsigned used:1;
325
326 /**
327 * Storage class of the variable.
328 *
329 * \sa ir_variable_mode
330 */
331 unsigned mode:3;
332
333 /**
334 * Interpolation mode for shader inputs / outputs
335 *
336 * \sa ir_variable_interpolation
337 */
338 unsigned interpolation:2;
339
340 /**
341 * \name ARB_fragment_coord_conventions
342 * @{
343 */
344 unsigned origin_upper_left:1;
345 unsigned pixel_center_integer:1;
346 /*@}*/
347
348 /**
349 * \brief Layout qualifier for gl_FragDepth.
350 *
351 * This is not equal to \c ir_depth_layout_none if and only if this
352 * variable is \c gl_FragDepth and a layout qualifier is specified.
353 */
354 ir_depth_layout depth_layout;
355
356 /**
357 * Was the location explicitly set in the shader?
358 *
359 * If the location is explicitly set in the shader, it \b cannot be changed
360 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
361 * no effect).
362 */
363 unsigned explicit_location:1;
364
365 /**
366 * Storage location of the base of this variable
367 *
368 * The precise meaning of this field depends on the nature of the variable.
369 *
370 * - Vertex shader input: one of the values from \c gl_vert_attrib.
371 * - Vertex shader output: one of the values from \c gl_vert_result.
372 * - Fragment shader input: one of the values from \c gl_frag_attrib.
373 * - Fragment shader output: one of the values from \c gl_frag_result.
374 * - Uniforms: Per-stage uniform slot number.
375 * - Other: This field is not currently used.
376 *
377 * If the variable is a uniform, shader input, or shader output, and the
378 * slot has not been assigned, the value will be -1.
379 */
380 int location;
381
382 /**
383 * Built-in state that backs this uniform
384 *
385 * Once set at variable creation, \c state_slots must remain invariant.
386 * This is because, ideally, this array would be shared by all clones of
387 * this variable in the IR tree. In other words, we'd really like for it
388 * to be a fly-weight.
389 *
390 * If the variable is not a uniform, \c num_state_slots will be zero and
391 * \c state_slots will be \c NULL.
392 */
393 /*@{*/
394 unsigned num_state_slots; /**< Number of state slots used */
395 ir_state_slot *state_slots; /**< State descriptors. */
396 /*@}*/
397
398 /**
399 * Emit a warning if this variable is accessed.
400 */
401 const char *warn_extension;
402
403 /**
404 * Value assigned in the initializer of a variable declared "const"
405 */
406 ir_constant *constant_value;
407 };
408
409
410 /*@{*/
411 /**
412 * The representation of a function instance; may be the full definition or
413 * simply a prototype.
414 */
415 class ir_function_signature : public ir_instruction {
416 /* An ir_function_signature will be part of the list of signatures in
417 * an ir_function.
418 */
419 public:
420 ir_function_signature(const glsl_type *return_type);
421
422 virtual ir_function_signature *clone(void *mem_ctx,
423 struct hash_table *ht) const;
424 ir_function_signature *clone_prototype(void *mem_ctx,
425 struct hash_table *ht) const;
426
427 virtual void accept(ir_visitor *v)
428 {
429 v->visit(this);
430 }
431
432 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
433
434 /**
435 * Get the name of the function for which this is a signature
436 */
437 const char *function_name() const;
438
439 /**
440 * Get a handle to the function for which this is a signature
441 *
442 * There is no setter function, this function returns a \c const pointer,
443 * and \c ir_function_signature::_function is private for a reason. The
444 * only way to make a connection between a function and function signature
445 * is via \c ir_function::add_signature. This helps ensure that certain
446 * invariants (i.e., a function signature is in the list of signatures for
447 * its \c _function) are met.
448 *
449 * \sa ir_function::add_signature
450 */
451 inline const class ir_function *function() const
452 {
453 return this->_function;
454 }
455
456 /**
457 * Check whether the qualifiers match between this signature's parameters
458 * and the supplied parameter list. If not, returns the name of the first
459 * parameter with mismatched qualifiers (for use in error messages).
460 */
461 const char *qualifiers_match(exec_list *params);
462
463 /**
464 * Replace the current parameter list with the given one. This is useful
465 * if the current information came from a prototype, and either has invalid
466 * or missing parameter names.
467 */
468 void replace_parameters(exec_list *new_params);
469
470 /**
471 * Function return type.
472 *
473 * \note This discards the optional precision qualifier.
474 */
475 const struct glsl_type *return_type;
476
477 /**
478 * List of ir_variable of function parameters.
479 *
480 * This represents the storage. The paramaters passed in a particular
481 * call will be in ir_call::actual_paramaters.
482 */
483 struct exec_list parameters;
484
485 /** Whether or not this function has a body (which may be empty). */
486 unsigned is_defined:1;
487
488 /** Whether or not this function signature is a built-in. */
489 unsigned is_builtin:1;
490
491 /** Body of instructions in the function. */
492 struct exec_list body;
493
494 private:
495 /** Function of which this signature is one overload. */
496 class ir_function *_function;
497
498 friend class ir_function;
499 };
500
501
502 /**
503 * Header for tracking multiple overloaded functions with the same name.
504 * Contains a list of ir_function_signatures representing each of the
505 * actual functions.
506 */
507 class ir_function : public ir_instruction {
508 public:
509 ir_function(const char *name);
510
511 virtual ir_function *clone(void *mem_ctx, struct hash_table *ht) const;
512
513 virtual ir_function *as_function()
514 {
515 return this;
516 }
517
518 virtual void accept(ir_visitor *v)
519 {
520 v->visit(this);
521 }
522
523 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
524
525 void add_signature(ir_function_signature *sig)
526 {
527 sig->_function = this;
528 this->signatures.push_tail(sig);
529 }
530
531 /**
532 * Get an iterator for the set of function signatures
533 */
534 exec_list_iterator iterator()
535 {
536 return signatures.iterator();
537 }
538
539 /**
540 * Find a signature that matches a set of actual parameters, taking implicit
541 * conversions into account.
542 */
543 ir_function_signature *matching_signature(const exec_list *actual_param);
544
545 /**
546 * Find a signature that exactly matches a set of actual parameters without
547 * any implicit type conversions.
548 */
549 ir_function_signature *exact_matching_signature(const exec_list *actual_ps);
550
551 /**
552 * Name of the function.
553 */
554 const char *name;
555
556 /** Whether or not this function has a signature that isn't a built-in. */
557 bool has_user_signature();
558
559 /**
560 * List of ir_function_signature for each overloaded function with this name.
561 */
562 struct exec_list signatures;
563 };
564
565 inline const char *ir_function_signature::function_name() const
566 {
567 return this->_function->name;
568 }
569 /*@}*/
570
571
572 /**
573 * IR instruction representing high-level if-statements
574 */
575 class ir_if : public ir_instruction {
576 public:
577 ir_if(ir_rvalue *condition)
578 : condition(condition)
579 {
580 ir_type = ir_type_if;
581 }
582
583 virtual ir_if *clone(void *mem_ctx, struct hash_table *ht) const;
584
585 virtual ir_if *as_if()
586 {
587 return this;
588 }
589
590 virtual void accept(ir_visitor *v)
591 {
592 v->visit(this);
593 }
594
595 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
596
597 ir_rvalue *condition;
598 /** List of ir_instruction for the body of the then branch */
599 exec_list then_instructions;
600 /** List of ir_instruction for the body of the else branch */
601 exec_list else_instructions;
602 };
603
604
605 /**
606 * IR instruction representing a high-level loop structure.
607 */
608 class ir_loop : public ir_instruction {
609 public:
610 ir_loop();
611
612 virtual ir_loop *clone(void *mem_ctx, struct hash_table *ht) const;
613
614 virtual void accept(ir_visitor *v)
615 {
616 v->visit(this);
617 }
618
619 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
620
621 virtual ir_loop *as_loop()
622 {
623 return this;
624 }
625
626 /**
627 * Get an iterator for the instructions of the loop body
628 */
629 exec_list_iterator iterator()
630 {
631 return body_instructions.iterator();
632 }
633
634 /** List of ir_instruction that make up the body of the loop. */
635 exec_list body_instructions;
636
637 /**
638 * \name Loop counter and controls
639 *
640 * Represents a loop like a FORTRAN \c do-loop.
641 *
642 * \note
643 * If \c from and \c to are the same value, the loop will execute once.
644 */
645 /*@{*/
646 ir_rvalue *from; /** Value of the loop counter on the first
647 * iteration of the loop.
648 */
649 ir_rvalue *to; /** Value of the loop counter on the last
650 * iteration of the loop.
651 */
652 ir_rvalue *increment;
653 ir_variable *counter;
654
655 /**
656 * Comparison operation in the loop terminator.
657 *
658 * If any of the loop control fields are non-\c NULL, this field must be
659 * one of \c ir_binop_less, \c ir_binop_greater, \c ir_binop_lequal,
660 * \c ir_binop_gequal, \c ir_binop_equal, or \c ir_binop_nequal.
661 */
662 int cmp;
663 /*@}*/
664 };
665
666
667 class ir_assignment : public ir_instruction {
668 public:
669 ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs, ir_rvalue *condition = NULL);
670
671 /**
672 * Construct an assignment with an explicit write mask
673 *
674 * \note
675 * Since a write mask is supplied, the LHS must already be a bare
676 * \c ir_dereference. The cannot be any swizzles in the LHS.
677 */
678 ir_assignment(ir_dereference *lhs, ir_rvalue *rhs, ir_rvalue *condition,
679 unsigned write_mask);
680
681 virtual ir_assignment *clone(void *mem_ctx, struct hash_table *ht) const;
682
683 virtual ir_constant *constant_expression_value();
684
685 virtual void accept(ir_visitor *v)
686 {
687 v->visit(this);
688 }
689
690 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
691
692 virtual ir_assignment * as_assignment()
693 {
694 return this;
695 }
696
697 /**
698 * Get a whole variable written by an assignment
699 *
700 * If the LHS of the assignment writes a whole variable, the variable is
701 * returned. Otherwise \c NULL is returned. Examples of whole-variable
702 * assignment are:
703 *
704 * - Assigning to a scalar
705 * - Assigning to all components of a vector
706 * - Whole array (or matrix) assignment
707 * - Whole structure assignment
708 */
709 ir_variable *whole_variable_written();
710
711 /**
712 * Set the LHS of an assignment
713 */
714 void set_lhs(ir_rvalue *lhs);
715
716 /**
717 * Left-hand side of the assignment.
718 *
719 * This should be treated as read only. If you need to set the LHS of an
720 * assignment, use \c ir_assignment::set_lhs.
721 */
722 ir_dereference *lhs;
723
724 /**
725 * Value being assigned
726 */
727 ir_rvalue *rhs;
728
729 /**
730 * Optional condition for the assignment.
731 */
732 ir_rvalue *condition;
733
734
735 /**
736 * Component mask written
737 *
738 * For non-vector types in the LHS, this field will be zero. For vector
739 * types, a bit will be set for each component that is written. Note that
740 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
741 *
742 * A partially-set write mask means that each enabled channel gets
743 * the value from a consecutive channel of the rhs. For example,
744 * to write just .xyw of gl_FrontColor with color:
745 *
746 * (assign (constant bool (1)) (xyw)
747 * (var_ref gl_FragColor)
748 * (swiz xyw (var_ref color)))
749 */
750 unsigned write_mask:4;
751 };
752
753 /* Update ir_expression::num_operands() and operator_strs when
754 * updating this list.
755 */
756 enum ir_expression_operation {
757 ir_unop_bit_not,
758 ir_unop_logic_not,
759 ir_unop_neg,
760 ir_unop_abs,
761 ir_unop_sign,
762 ir_unop_rcp,
763 ir_unop_rsq,
764 ir_unop_sqrt,
765 ir_unop_exp, /**< Log base e on gentype */
766 ir_unop_log, /**< Natural log on gentype */
767 ir_unop_exp2,
768 ir_unop_log2,
769 ir_unop_f2i, /**< Float-to-integer conversion. */
770 ir_unop_i2f, /**< Integer-to-float conversion. */
771 ir_unop_f2b, /**< Float-to-boolean conversion */
772 ir_unop_b2f, /**< Boolean-to-float conversion */
773 ir_unop_i2b, /**< int-to-boolean conversion */
774 ir_unop_b2i, /**< Boolean-to-int conversion */
775 ir_unop_u2f, /**< Unsigned-to-float conversion. */
776 ir_unop_i2u, /**< Integer-to-unsigned conversion. */
777 ir_unop_u2i, /**< Unsigned-to-integer conversion. */
778 ir_unop_any,
779
780 /**
781 * \name Unary floating-point rounding operations.
782 */
783 /*@{*/
784 ir_unop_trunc,
785 ir_unop_ceil,
786 ir_unop_floor,
787 ir_unop_fract,
788 ir_unop_round_even,
789 /*@}*/
790
791 /**
792 * \name Trigonometric operations.
793 */
794 /*@{*/
795 ir_unop_sin,
796 ir_unop_cos,
797 ir_unop_sin_reduced, /**< Reduced range sin. [-pi, pi] */
798 ir_unop_cos_reduced, /**< Reduced range cos. [-pi, pi] */
799 /*@}*/
800
801 /**
802 * \name Partial derivatives.
803 */
804 /*@{*/
805 ir_unop_dFdx,
806 ir_unop_dFdy,
807 /*@}*/
808
809 ir_unop_noise,
810
811 /**
812 * A sentinel marking the last of the unary operations.
813 */
814 ir_last_unop = ir_unop_noise,
815
816 ir_binop_add,
817 ir_binop_sub,
818 ir_binop_mul,
819 ir_binop_div,
820
821 /**
822 * Takes one of two combinations of arguments:
823 *
824 * - mod(vecN, vecN)
825 * - mod(vecN, float)
826 *
827 * Does not take integer types.
828 */
829 ir_binop_mod,
830
831 /**
832 * \name Binary comparison operators which return a boolean vector.
833 * The type of both operands must be equal.
834 */
835 /*@{*/
836 ir_binop_less,
837 ir_binop_greater,
838 ir_binop_lequal,
839 ir_binop_gequal,
840 ir_binop_equal,
841 ir_binop_nequal,
842 /**
843 * Returns single boolean for whether all components of operands[0]
844 * equal the components of operands[1].
845 */
846 ir_binop_all_equal,
847 /**
848 * Returns single boolean for whether any component of operands[0]
849 * is not equal to the corresponding component of operands[1].
850 */
851 ir_binop_any_nequal,
852 /*@}*/
853
854 /**
855 * \name Bit-wise binary operations.
856 */
857 /*@{*/
858 ir_binop_lshift,
859 ir_binop_rshift,
860 ir_binop_bit_and,
861 ir_binop_bit_xor,
862 ir_binop_bit_or,
863 /*@}*/
864
865 ir_binop_logic_and,
866 ir_binop_logic_xor,
867 ir_binop_logic_or,
868
869 ir_binop_dot,
870 ir_binop_min,
871 ir_binop_max,
872
873 ir_binop_pow,
874
875 /**
876 * A sentinel marking the last of the binary operations.
877 */
878 ir_last_binop = ir_binop_pow,
879
880 ir_quadop_vector,
881
882 /**
883 * A sentinel marking the last of all operations.
884 */
885 ir_last_opcode = ir_last_binop
886 };
887
888 class ir_expression : public ir_rvalue {
889 public:
890 /**
891 * Constructor for unary operation expressions
892 */
893 ir_expression(int op, const struct glsl_type *type, ir_rvalue *);
894 ir_expression(int op, ir_rvalue *);
895
896 /**
897 * Constructor for binary operation expressions
898 */
899 ir_expression(int op, const struct glsl_type *type,
900 ir_rvalue *, ir_rvalue *);
901 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1);
902
903 /**
904 * Constructor for quad operator expressions
905 */
906 ir_expression(int op, const struct glsl_type *type,
907 ir_rvalue *, ir_rvalue *, ir_rvalue *, ir_rvalue *);
908
909 virtual ir_expression *as_expression()
910 {
911 return this;
912 }
913
914 virtual ir_expression *clone(void *mem_ctx, struct hash_table *ht) const;
915
916 /**
917 * Attempt to constant-fold the expression
918 *
919 * If the expression cannot be constant folded, this method will return
920 * \c NULL.
921 */
922 virtual ir_constant *constant_expression_value();
923
924 /**
925 * Determine the number of operands used by an expression
926 */
927 static unsigned int get_num_operands(ir_expression_operation);
928
929 /**
930 * Determine the number of operands used by an expression
931 */
932 unsigned int get_num_operands() const
933 {
934 return (this->operation == ir_quadop_vector)
935 ? this->type->vector_elements : get_num_operands(operation);
936 }
937
938 /**
939 * Return a string representing this expression's operator.
940 */
941 const char *operator_string();
942
943 /**
944 * Return a string representing this expression's operator.
945 */
946 static const char *operator_string(ir_expression_operation);
947
948
949 /**
950 * Do a reverse-lookup to translate the given string into an operator.
951 */
952 static ir_expression_operation get_operator(const char *);
953
954 virtual void accept(ir_visitor *v)
955 {
956 v->visit(this);
957 }
958
959 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
960
961 ir_expression_operation operation;
962 ir_rvalue *operands[4];
963 };
964
965
966 /**
967 * IR instruction representing a function call
968 */
969 class ir_call : public ir_rvalue {
970 public:
971 ir_call(ir_function_signature *callee, exec_list *actual_parameters)
972 : callee(callee)
973 {
974 ir_type = ir_type_call;
975 assert(callee->return_type != NULL);
976 type = callee->return_type;
977 actual_parameters->move_nodes_to(& this->actual_parameters);
978 this->use_builtin = callee->is_builtin;
979 }
980
981 virtual ir_call *clone(void *mem_ctx, struct hash_table *ht) const;
982
983 virtual ir_constant *constant_expression_value();
984
985 virtual ir_call *as_call()
986 {
987 return this;
988 }
989
990 virtual void accept(ir_visitor *v)
991 {
992 v->visit(this);
993 }
994
995 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
996
997 /**
998 * Get a generic ir_call object when an error occurs
999 *
1000 * Any allocation will be performed with 'ctx' as ralloc owner.
1001 */
1002 static ir_call *get_error_instruction(void *ctx);
1003
1004 /**
1005 * Get an iterator for the set of acutal parameters
1006 */
1007 exec_list_iterator iterator()
1008 {
1009 return actual_parameters.iterator();
1010 }
1011
1012 /**
1013 * Get the name of the function being called.
1014 */
1015 const char *callee_name() const
1016 {
1017 return callee->function_name();
1018 }
1019
1020 /**
1021 * Get the function signature bound to this function call
1022 */
1023 ir_function_signature *get_callee()
1024 {
1025 return callee;
1026 }
1027
1028 /**
1029 * Set the function call target
1030 */
1031 void set_callee(ir_function_signature *sig);
1032
1033 /**
1034 * Generates an inline version of the function before @ir,
1035 * returning the return value of the function.
1036 */
1037 ir_rvalue *generate_inline(ir_instruction *ir);
1038
1039 /* List of ir_rvalue of paramaters passed in this call. */
1040 exec_list actual_parameters;
1041
1042 /** Should this call only bind to a built-in function? */
1043 bool use_builtin;
1044
1045 private:
1046 ir_call()
1047 : callee(NULL)
1048 {
1049 this->ir_type = ir_type_call;
1050 }
1051
1052 ir_function_signature *callee;
1053 };
1054
1055
1056 /**
1057 * \name Jump-like IR instructions.
1058 *
1059 * These include \c break, \c continue, \c return, and \c discard.
1060 */
1061 /*@{*/
1062 class ir_jump : public ir_instruction {
1063 protected:
1064 ir_jump()
1065 {
1066 ir_type = ir_type_unset;
1067 }
1068 };
1069
1070 class ir_return : public ir_jump {
1071 public:
1072 ir_return()
1073 : value(NULL)
1074 {
1075 this->ir_type = ir_type_return;
1076 }
1077
1078 ir_return(ir_rvalue *value)
1079 : value(value)
1080 {
1081 this->ir_type = ir_type_return;
1082 }
1083
1084 virtual ir_return *clone(void *mem_ctx, struct hash_table *) const;
1085
1086 virtual ir_return *as_return()
1087 {
1088 return this;
1089 }
1090
1091 ir_rvalue *get_value() const
1092 {
1093 return value;
1094 }
1095
1096 virtual void accept(ir_visitor *v)
1097 {
1098 v->visit(this);
1099 }
1100
1101 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1102
1103 ir_rvalue *value;
1104 };
1105
1106
1107 /**
1108 * Jump instructions used inside loops
1109 *
1110 * These include \c break and \c continue. The \c break within a loop is
1111 * different from the \c break within a switch-statement.
1112 *
1113 * \sa ir_switch_jump
1114 */
1115 class ir_loop_jump : public ir_jump {
1116 public:
1117 enum jump_mode {
1118 jump_break,
1119 jump_continue
1120 };
1121
1122 ir_loop_jump(jump_mode mode)
1123 {
1124 this->ir_type = ir_type_loop_jump;
1125 this->mode = mode;
1126 this->loop = loop;
1127 }
1128
1129 virtual ir_loop_jump *clone(void *mem_ctx, struct hash_table *) const;
1130
1131 virtual void accept(ir_visitor *v)
1132 {
1133 v->visit(this);
1134 }
1135
1136 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1137
1138 bool is_break() const
1139 {
1140 return mode == jump_break;
1141 }
1142
1143 bool is_continue() const
1144 {
1145 return mode == jump_continue;
1146 }
1147
1148 /** Mode selector for the jump instruction. */
1149 enum jump_mode mode;
1150 private:
1151 /** Loop containing this break instruction. */
1152 ir_loop *loop;
1153 };
1154
1155 /**
1156 * IR instruction representing discard statements.
1157 */
1158 class ir_discard : public ir_jump {
1159 public:
1160 ir_discard()
1161 {
1162 this->ir_type = ir_type_discard;
1163 this->condition = NULL;
1164 }
1165
1166 ir_discard(ir_rvalue *cond)
1167 {
1168 this->ir_type = ir_type_discard;
1169 this->condition = cond;
1170 }
1171
1172 virtual ir_discard *clone(void *mem_ctx, struct hash_table *ht) const;
1173
1174 virtual void accept(ir_visitor *v)
1175 {
1176 v->visit(this);
1177 }
1178
1179 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1180
1181 virtual ir_discard *as_discard()
1182 {
1183 return this;
1184 }
1185
1186 ir_rvalue *condition;
1187 };
1188 /*@}*/
1189
1190
1191 /**
1192 * Texture sampling opcodes used in ir_texture
1193 */
1194 enum ir_texture_opcode {
1195 ir_tex, /**< Regular texture look-up */
1196 ir_txb, /**< Texture look-up with LOD bias */
1197 ir_txl, /**< Texture look-up with explicit LOD */
1198 ir_txd, /**< Texture look-up with partial derivatvies */
1199 ir_txf, /**< Texel fetch with explicit LOD */
1200 ir_txs /**< Texture size */
1201 };
1202
1203
1204 /**
1205 * IR instruction to sample a texture
1206 *
1207 * The specific form of the IR instruction depends on the \c mode value
1208 * selected from \c ir_texture_opcodes. In the printed IR, these will
1209 * appear as:
1210 *
1211 * Texel offset (0 or an expression)
1212 * | Projection divisor
1213 * | | Shadow comparitor
1214 * | | |
1215 * v v v
1216 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1217 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1218 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1219 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1220 * (txf <type> <sampler> <coordinate> 0 <lod>)
1221 * (txs <type> <sampler> <lod>)
1222 */
1223 class ir_texture : public ir_rvalue {
1224 public:
1225 ir_texture(enum ir_texture_opcode op)
1226 : op(op), projector(NULL), shadow_comparitor(NULL), offset(NULL)
1227 {
1228 this->ir_type = ir_type_texture;
1229 }
1230
1231 virtual ir_texture *clone(void *mem_ctx, struct hash_table *) const;
1232
1233 virtual ir_constant *constant_expression_value();
1234
1235 virtual void accept(ir_visitor *v)
1236 {
1237 v->visit(this);
1238 }
1239
1240 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1241
1242 /**
1243 * Return a string representing the ir_texture_opcode.
1244 */
1245 const char *opcode_string();
1246
1247 /** Set the sampler and type. */
1248 void set_sampler(ir_dereference *sampler, const glsl_type *type);
1249
1250 /**
1251 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1252 */
1253 static ir_texture_opcode get_opcode(const char *);
1254
1255 enum ir_texture_opcode op;
1256
1257 /** Sampler to use for the texture access. */
1258 ir_dereference *sampler;
1259
1260 /** Texture coordinate to sample */
1261 ir_rvalue *coordinate;
1262
1263 /**
1264 * Value used for projective divide.
1265 *
1266 * If there is no projective divide (the common case), this will be
1267 * \c NULL. Optimization passes should check for this to point to a constant
1268 * of 1.0 and replace that with \c NULL.
1269 */
1270 ir_rvalue *projector;
1271
1272 /**
1273 * Coordinate used for comparison on shadow look-ups.
1274 *
1275 * If there is no shadow comparison, this will be \c NULL. For the
1276 * \c ir_txf opcode, this *must* be \c NULL.
1277 */
1278 ir_rvalue *shadow_comparitor;
1279
1280 /** Texel offset. */
1281 ir_rvalue *offset;
1282
1283 union {
1284 ir_rvalue *lod; /**< Floating point LOD */
1285 ir_rvalue *bias; /**< Floating point LOD bias */
1286 struct {
1287 ir_rvalue *dPdx; /**< Partial derivative of coordinate wrt X */
1288 ir_rvalue *dPdy; /**< Partial derivative of coordinate wrt Y */
1289 } grad;
1290 } lod_info;
1291 };
1292
1293
1294 struct ir_swizzle_mask {
1295 unsigned x:2;
1296 unsigned y:2;
1297 unsigned z:2;
1298 unsigned w:2;
1299
1300 /**
1301 * Number of components in the swizzle.
1302 */
1303 unsigned num_components:3;
1304
1305 /**
1306 * Does the swizzle contain duplicate components?
1307 *
1308 * L-value swizzles cannot contain duplicate components.
1309 */
1310 unsigned has_duplicates:1;
1311 };
1312
1313
1314 class ir_swizzle : public ir_rvalue {
1315 public:
1316 ir_swizzle(ir_rvalue *, unsigned x, unsigned y, unsigned z, unsigned w,
1317 unsigned count);
1318
1319 ir_swizzle(ir_rvalue *val, const unsigned *components, unsigned count);
1320
1321 ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask);
1322
1323 virtual ir_swizzle *clone(void *mem_ctx, struct hash_table *) const;
1324
1325 virtual ir_constant *constant_expression_value();
1326
1327 virtual ir_swizzle *as_swizzle()
1328 {
1329 return this;
1330 }
1331
1332 /**
1333 * Construct an ir_swizzle from the textual representation. Can fail.
1334 */
1335 static ir_swizzle *create(ir_rvalue *, const char *, unsigned vector_length);
1336
1337 virtual void accept(ir_visitor *v)
1338 {
1339 v->visit(this);
1340 }
1341
1342 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1343
1344 bool is_lvalue() const
1345 {
1346 return val->is_lvalue() && !mask.has_duplicates;
1347 }
1348
1349 /**
1350 * Get the variable that is ultimately referenced by an r-value
1351 */
1352 virtual ir_variable *variable_referenced() const;
1353
1354 ir_rvalue *val;
1355 ir_swizzle_mask mask;
1356
1357 private:
1358 /**
1359 * Initialize the mask component of a swizzle
1360 *
1361 * This is used by the \c ir_swizzle constructors.
1362 */
1363 void init_mask(const unsigned *components, unsigned count);
1364 };
1365
1366
1367 class ir_dereference : public ir_rvalue {
1368 public:
1369 virtual ir_dereference *clone(void *mem_ctx, struct hash_table *) const = 0;
1370
1371 virtual ir_dereference *as_dereference()
1372 {
1373 return this;
1374 }
1375
1376 bool is_lvalue() const;
1377
1378 /**
1379 * Get the variable that is ultimately referenced by an r-value
1380 */
1381 virtual ir_variable *variable_referenced() const = 0;
1382 };
1383
1384
1385 class ir_dereference_variable : public ir_dereference {
1386 public:
1387 ir_dereference_variable(ir_variable *var);
1388
1389 virtual ir_dereference_variable *clone(void *mem_ctx,
1390 struct hash_table *) const;
1391
1392 virtual ir_constant *constant_expression_value();
1393
1394 virtual ir_dereference_variable *as_dereference_variable()
1395 {
1396 return this;
1397 }
1398
1399 /**
1400 * Get the variable that is ultimately referenced by an r-value
1401 */
1402 virtual ir_variable *variable_referenced() const
1403 {
1404 return this->var;
1405 }
1406
1407 virtual ir_variable *whole_variable_referenced()
1408 {
1409 /* ir_dereference_variable objects always dereference the entire
1410 * variable. However, if this dereference is dereferenced by anything
1411 * else, the complete deferefernce chain is not a whole-variable
1412 * dereference. This method should only be called on the top most
1413 * ir_rvalue in a dereference chain.
1414 */
1415 return this->var;
1416 }
1417
1418 virtual void accept(ir_visitor *v)
1419 {
1420 v->visit(this);
1421 }
1422
1423 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1424
1425 /**
1426 * Object being dereferenced.
1427 */
1428 ir_variable *var;
1429 };
1430
1431
1432 class ir_dereference_array : public ir_dereference {
1433 public:
1434 ir_dereference_array(ir_rvalue *value, ir_rvalue *array_index);
1435
1436 ir_dereference_array(ir_variable *var, ir_rvalue *array_index);
1437
1438 virtual ir_dereference_array *clone(void *mem_ctx,
1439 struct hash_table *) const;
1440
1441 virtual ir_constant *constant_expression_value();
1442
1443 virtual ir_dereference_array *as_dereference_array()
1444 {
1445 return this;
1446 }
1447
1448 /**
1449 * Get the variable that is ultimately referenced by an r-value
1450 */
1451 virtual ir_variable *variable_referenced() const
1452 {
1453 return this->array->variable_referenced();
1454 }
1455
1456 virtual void accept(ir_visitor *v)
1457 {
1458 v->visit(this);
1459 }
1460
1461 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1462
1463 ir_rvalue *array;
1464 ir_rvalue *array_index;
1465
1466 private:
1467 void set_array(ir_rvalue *value);
1468 };
1469
1470
1471 class ir_dereference_record : public ir_dereference {
1472 public:
1473 ir_dereference_record(ir_rvalue *value, const char *field);
1474
1475 ir_dereference_record(ir_variable *var, const char *field);
1476
1477 virtual ir_dereference_record *clone(void *mem_ctx,
1478 struct hash_table *) const;
1479
1480 virtual ir_constant *constant_expression_value();
1481
1482 /**
1483 * Get the variable that is ultimately referenced by an r-value
1484 */
1485 virtual ir_variable *variable_referenced() const
1486 {
1487 return this->record->variable_referenced();
1488 }
1489
1490 virtual void accept(ir_visitor *v)
1491 {
1492 v->visit(this);
1493 }
1494
1495 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1496
1497 ir_rvalue *record;
1498 const char *field;
1499 };
1500
1501
1502 /**
1503 * Data stored in an ir_constant
1504 */
1505 union ir_constant_data {
1506 unsigned u[16];
1507 int i[16];
1508 float f[16];
1509 bool b[16];
1510 };
1511
1512
1513 class ir_constant : public ir_rvalue {
1514 public:
1515 ir_constant(const struct glsl_type *type, const ir_constant_data *data);
1516 ir_constant(bool b);
1517 ir_constant(unsigned int u);
1518 ir_constant(int i);
1519 ir_constant(float f);
1520
1521 /**
1522 * Construct an ir_constant from a list of ir_constant values
1523 */
1524 ir_constant(const struct glsl_type *type, exec_list *values);
1525
1526 /**
1527 * Construct an ir_constant from a scalar component of another ir_constant
1528 *
1529 * The new \c ir_constant inherits the type of the component from the
1530 * source constant.
1531 *
1532 * \note
1533 * In the case of a matrix constant, the new constant is a scalar, \b not
1534 * a vector.
1535 */
1536 ir_constant(const ir_constant *c, unsigned i);
1537
1538 /**
1539 * Return a new ir_constant of the specified type containing all zeros.
1540 */
1541 static ir_constant *zero(void *mem_ctx, const glsl_type *type);
1542
1543 virtual ir_constant *clone(void *mem_ctx, struct hash_table *) const;
1544
1545 virtual ir_constant *constant_expression_value();
1546
1547 virtual ir_constant *as_constant()
1548 {
1549 return this;
1550 }
1551
1552 virtual void accept(ir_visitor *v)
1553 {
1554 v->visit(this);
1555 }
1556
1557 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1558
1559 /**
1560 * Get a particular component of a constant as a specific type
1561 *
1562 * This is useful, for example, to get a value from an integer constant
1563 * as a float or bool. This appears frequently when constructors are
1564 * called with all constant parameters.
1565 */
1566 /*@{*/
1567 bool get_bool_component(unsigned i) const;
1568 float get_float_component(unsigned i) const;
1569 int get_int_component(unsigned i) const;
1570 unsigned get_uint_component(unsigned i) const;
1571 /*@}*/
1572
1573 ir_constant *get_array_element(unsigned i) const;
1574
1575 ir_constant *get_record_field(const char *name);
1576
1577 /**
1578 * Determine whether a constant has the same value as another constant
1579 *
1580 * \sa ir_constant::is_zero, ir_constant::is_one,
1581 * ir_constant::is_negative_one
1582 */
1583 bool has_value(const ir_constant *) const;
1584
1585 virtual bool is_zero() const;
1586 virtual bool is_one() const;
1587 virtual bool is_negative_one() const;
1588
1589 /**
1590 * Value of the constant.
1591 *
1592 * The field used to back the values supplied by the constant is determined
1593 * by the type associated with the \c ir_instruction. Constants may be
1594 * scalars, vectors, or matrices.
1595 */
1596 union ir_constant_data value;
1597
1598 /* Array elements */
1599 ir_constant **array_elements;
1600
1601 /* Structure fields */
1602 exec_list components;
1603
1604 private:
1605 /**
1606 * Parameterless constructor only used by the clone method
1607 */
1608 ir_constant(void);
1609 };
1610
1611 /*@}*/
1612
1613 /**
1614 * Apply a visitor to each IR node in a list
1615 */
1616 void
1617 visit_exec_list(exec_list *list, ir_visitor *visitor);
1618
1619 /**
1620 * Validate invariants on each IR node in a list
1621 */
1622 void validate_ir_tree(exec_list *instructions);
1623
1624 struct _mesa_glsl_parse_state;
1625 struct gl_shader_program;
1626
1627 /**
1628 * Detect whether an unlinked shader contains static recursion
1629 *
1630 * If the list of instructions is determined to contain static recursion,
1631 * \c _mesa_glsl_error will be called to emit error messages for each function
1632 * that is in the recursion cycle.
1633 */
1634 void
1635 detect_recursion_unlinked(struct _mesa_glsl_parse_state *state,
1636 exec_list *instructions);
1637
1638 /**
1639 * Detect whether a linked shader contains static recursion
1640 *
1641 * If the list of instructions is determined to contain static recursion,
1642 * \c link_error_printf will be called to emit error messages for each function
1643 * that is in the recursion cycle. In addition,
1644 * \c gl_shader_program::LinkStatus will be set to false.
1645 */
1646 void
1647 detect_recursion_linked(struct gl_shader_program *prog,
1648 exec_list *instructions);
1649
1650 /**
1651 * Make a clone of each IR instruction in a list
1652 *
1653 * \param in List of IR instructions that are to be cloned
1654 * \param out List to hold the cloned instructions
1655 */
1656 void
1657 clone_ir_list(void *mem_ctx, exec_list *out, const exec_list *in);
1658
1659 extern void
1660 _mesa_glsl_initialize_variables(exec_list *instructions,
1661 struct _mesa_glsl_parse_state *state);
1662
1663 extern void
1664 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state *state);
1665
1666 extern void
1667 _mesa_glsl_release_functions(void);
1668
1669 extern void
1670 reparent_ir(exec_list *list, void *mem_ctx);
1671
1672 struct glsl_symbol_table;
1673
1674 extern void
1675 import_prototypes(const exec_list *source, exec_list *dest,
1676 struct glsl_symbol_table *symbols, void *mem_ctx);
1677
1678 extern bool
1679 ir_has_call(ir_instruction *ir);
1680
1681 extern void
1682 do_set_program_inouts(exec_list *instructions, struct gl_program *prog);
1683
1684 extern char *
1685 prototype_string(const glsl_type *return_type, const char *name,
1686 exec_list *parameters);
1687
1688 #endif /* IR_H */