Merge remote-tracking branch 'origin/master' into pipe-video
[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()
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()
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_conservative_depth extension allows 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 * Calculate the number of slots required to hold this variable
297 *
298 * This is used to determine how many uniform or varying locations a variable
299 * occupies. The count is in units of floating point components.
300 */
301 unsigned component_slots() const;
302
303 /**
304 * Delcared name of the variable
305 */
306 const char *name;
307
308 /**
309 * Highest element accessed with a constant expression array index
310 *
311 * Not used for non-array variables.
312 */
313 unsigned max_array_access;
314
315 /**
316 * Is the variable read-only?
317 *
318 * This is set for variables declared as \c const, shader inputs,
319 * and uniforms.
320 */
321 unsigned read_only:1;
322 unsigned centroid:1;
323 unsigned invariant:1;
324
325 /**
326 * Has this variable been used for reading or writing?
327 *
328 * Several GLSL semantic checks require knowledge of whether or not a
329 * variable has been used. For example, it is an error to redeclare a
330 * variable as invariant after it has been used.
331 */
332 unsigned used:1;
333
334 /**
335 * Storage class of the variable.
336 *
337 * \sa ir_variable_mode
338 */
339 unsigned mode:3;
340
341 /**
342 * Interpolation mode for shader inputs / outputs
343 *
344 * \sa ir_variable_interpolation
345 */
346 unsigned interpolation:2;
347
348 /**
349 * Flag that the whole array is assignable
350 *
351 * In GLSL 1.20 and later whole arrays are assignable (and comparable for
352 * equality). This flag enables this behavior.
353 */
354 unsigned array_lvalue:1;
355
356 /**
357 * \name ARB_fragment_coord_conventions
358 * @{
359 */
360 unsigned origin_upper_left:1;
361 unsigned pixel_center_integer:1;
362 /*@}*/
363
364 /**
365 * \brief Layout qualifier for gl_FragDepth.
366 *
367 * This is not equal to \c ir_depth_layout_none if and only if this
368 * variable is \c gl_FragDepth and a layout qualifier is specified.
369 */
370 ir_depth_layout depth_layout;
371
372 /**
373 * Was the location explicitly set in the shader?
374 *
375 * If the location is explicitly set in the shader, it \b cannot be changed
376 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
377 * no effect).
378 */
379 unsigned explicit_location:1;
380
381 /**
382 * Storage location of the base of this variable
383 *
384 * The precise meaning of this field depends on the nature of the variable.
385 *
386 * - Vertex shader input: one of the values from \c gl_vert_attrib.
387 * - Vertex shader output: one of the values from \c gl_vert_result.
388 * - Fragment shader input: one of the values from \c gl_frag_attrib.
389 * - Fragment shader output: one of the values from \c gl_frag_result.
390 * - Uniforms: Per-stage uniform slot number.
391 * - Other: This field is not currently used.
392 *
393 * If the variable is a uniform, shader input, or shader output, and the
394 * slot has not been assigned, the value will be -1.
395 */
396 int location;
397
398 /**
399 * Built-in state that backs this uniform
400 *
401 * Once set at variable creation, \c state_slots must remain invariant.
402 * This is because, ideally, this array would be shared by all clones of
403 * this variable in the IR tree. In other words, we'd really like for it
404 * to be a fly-weight.
405 *
406 * If the variable is not a uniform, \c num_state_slots will be zero and
407 * \c state_slots will be \c NULL.
408 */
409 /*@{*/
410 unsigned num_state_slots; /**< Number of state slots used */
411 ir_state_slot *state_slots; /**< State descriptors. */
412 /*@}*/
413
414 /**
415 * Emit a warning if this variable is accessed.
416 */
417 const char *warn_extension;
418
419 /**
420 * Value assigned in the initializer of a variable declared "const"
421 */
422 ir_constant *constant_value;
423 };
424
425
426 /*@{*/
427 /**
428 * The representation of a function instance; may be the full definition or
429 * simply a prototype.
430 */
431 class ir_function_signature : public ir_instruction {
432 /* An ir_function_signature will be part of the list of signatures in
433 * an ir_function.
434 */
435 public:
436 ir_function_signature(const glsl_type *return_type);
437
438 virtual ir_function_signature *clone(void *mem_ctx,
439 struct hash_table *ht) const;
440 ir_function_signature *clone_prototype(void *mem_ctx,
441 struct hash_table *ht) const;
442
443 virtual void accept(ir_visitor *v)
444 {
445 v->visit(this);
446 }
447
448 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
449
450 /**
451 * Get the name of the function for which this is a signature
452 */
453 const char *function_name() const;
454
455 /**
456 * Get a handle to the function for which this is a signature
457 *
458 * There is no setter function, this function returns a \c const pointer,
459 * and \c ir_function_signature::_function is private for a reason. The
460 * only way to make a connection between a function and function signature
461 * is via \c ir_function::add_signature. This helps ensure that certain
462 * invariants (i.e., a function signature is in the list of signatures for
463 * its \c _function) are met.
464 *
465 * \sa ir_function::add_signature
466 */
467 inline const class ir_function *function() const
468 {
469 return this->_function;
470 }
471
472 /**
473 * Check whether the qualifiers match between this signature's parameters
474 * and the supplied parameter list. If not, returns the name of the first
475 * parameter with mismatched qualifiers (for use in error messages).
476 */
477 const char *qualifiers_match(exec_list *params);
478
479 /**
480 * Replace the current parameter list with the given one. This is useful
481 * if the current information came from a prototype, and either has invalid
482 * or missing parameter names.
483 */
484 void replace_parameters(exec_list *new_params);
485
486 /**
487 * Function return type.
488 *
489 * \note This discards the optional precision qualifier.
490 */
491 const struct glsl_type *return_type;
492
493 /**
494 * List of ir_variable of function parameters.
495 *
496 * This represents the storage. The paramaters passed in a particular
497 * call will be in ir_call::actual_paramaters.
498 */
499 struct exec_list parameters;
500
501 /** Whether or not this function has a body (which may be empty). */
502 unsigned is_defined:1;
503
504 /** Whether or not this function signature is a built-in. */
505 unsigned is_builtin:1;
506
507 /** Body of instructions in the function. */
508 struct exec_list body;
509
510 private:
511 /** Function of which this signature is one overload. */
512 class ir_function *_function;
513
514 friend class ir_function;
515 };
516
517
518 /**
519 * Header for tracking multiple overloaded functions with the same name.
520 * Contains a list of ir_function_signatures representing each of the
521 * actual functions.
522 */
523 class ir_function : public ir_instruction {
524 public:
525 ir_function(const char *name);
526
527 virtual ir_function *clone(void *mem_ctx, struct hash_table *ht) const;
528
529 virtual ir_function *as_function()
530 {
531 return this;
532 }
533
534 virtual void accept(ir_visitor *v)
535 {
536 v->visit(this);
537 }
538
539 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
540
541 void add_signature(ir_function_signature *sig)
542 {
543 sig->_function = this;
544 this->signatures.push_tail(sig);
545 }
546
547 /**
548 * Get an iterator for the set of function signatures
549 */
550 exec_list_iterator iterator()
551 {
552 return signatures.iterator();
553 }
554
555 /**
556 * Find a signature that matches a set of actual parameters, taking implicit
557 * conversions into account.
558 */
559 ir_function_signature *matching_signature(const exec_list *actual_param);
560
561 /**
562 * Find a signature that exactly matches a set of actual parameters without
563 * any implicit type conversions.
564 */
565 ir_function_signature *exact_matching_signature(const exec_list *actual_ps);
566
567 /**
568 * Name of the function.
569 */
570 const char *name;
571
572 /** Whether or not this function has a signature that isn't a built-in. */
573 bool has_user_signature();
574
575 /**
576 * List of ir_function_signature for each overloaded function with this name.
577 */
578 struct exec_list signatures;
579 };
580
581 inline const char *ir_function_signature::function_name() const
582 {
583 return this->_function->name;
584 }
585 /*@}*/
586
587
588 /**
589 * IR instruction representing high-level if-statements
590 */
591 class ir_if : public ir_instruction {
592 public:
593 ir_if(ir_rvalue *condition)
594 : condition(condition)
595 {
596 ir_type = ir_type_if;
597 }
598
599 virtual ir_if *clone(void *mem_ctx, struct hash_table *ht) const;
600
601 virtual ir_if *as_if()
602 {
603 return this;
604 }
605
606 virtual void accept(ir_visitor *v)
607 {
608 v->visit(this);
609 }
610
611 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
612
613 ir_rvalue *condition;
614 /** List of ir_instruction for the body of the then branch */
615 exec_list then_instructions;
616 /** List of ir_instruction for the body of the else branch */
617 exec_list else_instructions;
618 };
619
620
621 /**
622 * IR instruction representing a high-level loop structure.
623 */
624 class ir_loop : public ir_instruction {
625 public:
626 ir_loop();
627
628 virtual ir_loop *clone(void *mem_ctx, struct hash_table *ht) const;
629
630 virtual void accept(ir_visitor *v)
631 {
632 v->visit(this);
633 }
634
635 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
636
637 virtual ir_loop *as_loop()
638 {
639 return this;
640 }
641
642 /**
643 * Get an iterator for the instructions of the loop body
644 */
645 exec_list_iterator iterator()
646 {
647 return body_instructions.iterator();
648 }
649
650 /** List of ir_instruction that make up the body of the loop. */
651 exec_list body_instructions;
652
653 /**
654 * \name Loop counter and controls
655 *
656 * Represents a loop like a FORTRAN \c do-loop.
657 *
658 * \note
659 * If \c from and \c to are the same value, the loop will execute once.
660 */
661 /*@{*/
662 ir_rvalue *from; /** Value of the loop counter on the first
663 * iteration of the loop.
664 */
665 ir_rvalue *to; /** Value of the loop counter on the last
666 * iteration of the loop.
667 */
668 ir_rvalue *increment;
669 ir_variable *counter;
670
671 /**
672 * Comparison operation in the loop terminator.
673 *
674 * If any of the loop control fields are non-\c NULL, this field must be
675 * one of \c ir_binop_less, \c ir_binop_greater, \c ir_binop_lequal,
676 * \c ir_binop_gequal, \c ir_binop_equal, or \c ir_binop_nequal.
677 */
678 int cmp;
679 /*@}*/
680 };
681
682
683 class ir_assignment : public ir_instruction {
684 public:
685 ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs, ir_rvalue *condition);
686
687 /**
688 * Construct an assignment with an explicit write mask
689 *
690 * \note
691 * Since a write mask is supplied, the LHS must already be a bare
692 * \c ir_dereference. The cannot be any swizzles in the LHS.
693 */
694 ir_assignment(ir_dereference *lhs, ir_rvalue *rhs, ir_rvalue *condition,
695 unsigned write_mask);
696
697 virtual ir_assignment *clone(void *mem_ctx, struct hash_table *ht) const;
698
699 virtual ir_constant *constant_expression_value();
700
701 virtual void accept(ir_visitor *v)
702 {
703 v->visit(this);
704 }
705
706 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
707
708 virtual ir_assignment * as_assignment()
709 {
710 return this;
711 }
712
713 /**
714 * Get a whole variable written by an assignment
715 *
716 * If the LHS of the assignment writes a whole variable, the variable is
717 * returned. Otherwise \c NULL is returned. Examples of whole-variable
718 * assignment are:
719 *
720 * - Assigning to a scalar
721 * - Assigning to all components of a vector
722 * - Whole array (or matrix) assignment
723 * - Whole structure assignment
724 */
725 ir_variable *whole_variable_written();
726
727 /**
728 * Set the LHS of an assignment
729 */
730 void set_lhs(ir_rvalue *lhs);
731
732 /**
733 * Left-hand side of the assignment.
734 *
735 * This should be treated as read only. If you need to set the LHS of an
736 * assignment, use \c ir_assignment::set_lhs.
737 */
738 ir_dereference *lhs;
739
740 /**
741 * Value being assigned
742 */
743 ir_rvalue *rhs;
744
745 /**
746 * Optional condition for the assignment.
747 */
748 ir_rvalue *condition;
749
750
751 /**
752 * Component mask written
753 *
754 * For non-vector types in the LHS, this field will be zero. For vector
755 * types, a bit will be set for each component that is written. Note that
756 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
757 *
758 * A partially-set write mask means that each enabled channel gets
759 * the value from a consecutive channel of the rhs. For example,
760 * to write just .xyw of gl_FrontColor with color:
761 *
762 * (assign (constant bool (1)) (xyw)
763 * (var_ref gl_FragColor)
764 * (swiz xyw (var_ref color)))
765 */
766 unsigned write_mask:4;
767 };
768
769 /* Update ir_expression::num_operands() and operator_strs when
770 * updating this list.
771 */
772 enum ir_expression_operation {
773 ir_unop_bit_not,
774 ir_unop_logic_not,
775 ir_unop_neg,
776 ir_unop_abs,
777 ir_unop_sign,
778 ir_unop_rcp,
779 ir_unop_rsq,
780 ir_unop_sqrt,
781 ir_unop_exp, /**< Log base e on gentype */
782 ir_unop_log, /**< Natural log on gentype */
783 ir_unop_exp2,
784 ir_unop_log2,
785 ir_unop_f2i, /**< Float-to-integer conversion. */
786 ir_unop_i2f, /**< Integer-to-float conversion. */
787 ir_unop_f2b, /**< Float-to-boolean conversion */
788 ir_unop_b2f, /**< Boolean-to-float conversion */
789 ir_unop_i2b, /**< int-to-boolean conversion */
790 ir_unop_b2i, /**< Boolean-to-int conversion */
791 ir_unop_u2f, /**< Unsigned-to-float conversion. */
792 ir_unop_any,
793
794 /**
795 * \name Unary floating-point rounding operations.
796 */
797 /*@{*/
798 ir_unop_trunc,
799 ir_unop_ceil,
800 ir_unop_floor,
801 ir_unop_fract,
802 ir_unop_round_even,
803 /*@}*/
804
805 /**
806 * \name Trigonometric operations.
807 */
808 /*@{*/
809 ir_unop_sin,
810 ir_unop_cos,
811 ir_unop_sin_reduced, /**< Reduced range sin. [-pi, pi] */
812 ir_unop_cos_reduced, /**< Reduced range cos. [-pi, pi] */
813 /*@}*/
814
815 /**
816 * \name Partial derivatives.
817 */
818 /*@{*/
819 ir_unop_dFdx,
820 ir_unop_dFdy,
821 /*@}*/
822
823 ir_unop_noise,
824
825 /**
826 * A sentinel marking the last of the unary operations.
827 */
828 ir_last_unop = ir_unop_noise,
829
830 ir_binop_add,
831 ir_binop_sub,
832 ir_binop_mul,
833 ir_binop_div,
834
835 /**
836 * Takes one of two combinations of arguments:
837 *
838 * - mod(vecN, vecN)
839 * - mod(vecN, float)
840 *
841 * Does not take integer types.
842 */
843 ir_binop_mod,
844
845 /**
846 * \name Binary comparison operators which return a boolean vector.
847 * The type of both operands must be equal.
848 */
849 /*@{*/
850 ir_binop_less,
851 ir_binop_greater,
852 ir_binop_lequal,
853 ir_binop_gequal,
854 ir_binop_equal,
855 ir_binop_nequal,
856 /**
857 * Returns single boolean for whether all components of operands[0]
858 * equal the components of operands[1].
859 */
860 ir_binop_all_equal,
861 /**
862 * Returns single boolean for whether any component of operands[0]
863 * is not equal to the corresponding component of operands[1].
864 */
865 ir_binop_any_nequal,
866 /*@}*/
867
868 /**
869 * \name Bit-wise binary operations.
870 */
871 /*@{*/
872 ir_binop_lshift,
873 ir_binop_rshift,
874 ir_binop_bit_and,
875 ir_binop_bit_xor,
876 ir_binop_bit_or,
877 /*@}*/
878
879 ir_binop_logic_and,
880 ir_binop_logic_xor,
881 ir_binop_logic_or,
882
883 ir_binop_dot,
884 ir_binop_min,
885 ir_binop_max,
886
887 ir_binop_pow,
888
889 /**
890 * A sentinel marking the last of the binary operations.
891 */
892 ir_last_binop = ir_binop_pow,
893
894 ir_quadop_vector,
895
896 /**
897 * A sentinel marking the last of all operations.
898 */
899 ir_last_opcode = ir_last_binop
900 };
901
902 class ir_expression : public ir_rvalue {
903 public:
904 /**
905 * Constructor for unary operation expressions
906 */
907 ir_expression(int op, const struct glsl_type *type, ir_rvalue *);
908 ir_expression(int op, ir_rvalue *);
909
910 /**
911 * Constructor for binary operation expressions
912 */
913 ir_expression(int op, const struct glsl_type *type,
914 ir_rvalue *, ir_rvalue *);
915 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1);
916
917 /**
918 * Constructor for quad operator expressions
919 */
920 ir_expression(int op, const struct glsl_type *type,
921 ir_rvalue *, ir_rvalue *, ir_rvalue *, ir_rvalue *);
922
923 virtual ir_expression *as_expression()
924 {
925 return this;
926 }
927
928 virtual ir_expression *clone(void *mem_ctx, struct hash_table *ht) const;
929
930 /**
931 * Attempt to constant-fold the expression
932 *
933 * If the expression cannot be constant folded, this method will return
934 * \c NULL.
935 */
936 virtual ir_constant *constant_expression_value();
937
938 /**
939 * Determine the number of operands used by an expression
940 */
941 static unsigned int get_num_operands(ir_expression_operation);
942
943 /**
944 * Determine the number of operands used by an expression
945 */
946 unsigned int get_num_operands() const
947 {
948 return (this->operation == ir_quadop_vector)
949 ? this->type->vector_elements : get_num_operands(operation);
950 }
951
952 /**
953 * Return a string representing this expression's operator.
954 */
955 const char *operator_string();
956
957 /**
958 * Return a string representing this expression's operator.
959 */
960 static const char *operator_string(ir_expression_operation);
961
962
963 /**
964 * Do a reverse-lookup to translate the given string into an operator.
965 */
966 static ir_expression_operation get_operator(const char *);
967
968 virtual void accept(ir_visitor *v)
969 {
970 v->visit(this);
971 }
972
973 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
974
975 ir_expression_operation operation;
976 ir_rvalue *operands[4];
977 };
978
979
980 /**
981 * IR instruction representing a function call
982 */
983 class ir_call : public ir_rvalue {
984 public:
985 ir_call(ir_function_signature *callee, exec_list *actual_parameters)
986 : callee(callee)
987 {
988 ir_type = ir_type_call;
989 assert(callee->return_type != NULL);
990 type = callee->return_type;
991 actual_parameters->move_nodes_to(& this->actual_parameters);
992 }
993
994 virtual ir_call *clone(void *mem_ctx, struct hash_table *ht) const;
995
996 virtual ir_constant *constant_expression_value();
997
998 virtual ir_call *as_call()
999 {
1000 return this;
1001 }
1002
1003 virtual void accept(ir_visitor *v)
1004 {
1005 v->visit(this);
1006 }
1007
1008 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1009
1010 /**
1011 * Get a generic ir_call object when an error occurs
1012 *
1013 * Any allocation will be performed with 'ctx' as ralloc owner.
1014 */
1015 static ir_call *get_error_instruction(void *ctx);
1016
1017 /**
1018 * Get an iterator for the set of acutal parameters
1019 */
1020 exec_list_iterator iterator()
1021 {
1022 return actual_parameters.iterator();
1023 }
1024
1025 /**
1026 * Get the name of the function being called.
1027 */
1028 const char *callee_name() const
1029 {
1030 return callee->function_name();
1031 }
1032
1033 /**
1034 * Get the function signature bound to this function call
1035 */
1036 ir_function_signature *get_callee()
1037 {
1038 return callee;
1039 }
1040
1041 /**
1042 * Set the function call target
1043 */
1044 void set_callee(ir_function_signature *sig);
1045
1046 /**
1047 * Generates an inline version of the function before @ir,
1048 * returning the return value of the function.
1049 */
1050 ir_rvalue *generate_inline(ir_instruction *ir);
1051
1052 /* List of ir_rvalue of paramaters passed in this call. */
1053 exec_list actual_parameters;
1054
1055 private:
1056 ir_call()
1057 : callee(NULL)
1058 {
1059 this->ir_type = ir_type_call;
1060 }
1061
1062 ir_function_signature *callee;
1063 };
1064
1065
1066 /**
1067 * \name Jump-like IR instructions.
1068 *
1069 * These include \c break, \c continue, \c return, and \c discard.
1070 */
1071 /*@{*/
1072 class ir_jump : public ir_instruction {
1073 protected:
1074 ir_jump()
1075 {
1076 ir_type = ir_type_unset;
1077 }
1078 };
1079
1080 class ir_return : public ir_jump {
1081 public:
1082 ir_return()
1083 : value(NULL)
1084 {
1085 this->ir_type = ir_type_return;
1086 }
1087
1088 ir_return(ir_rvalue *value)
1089 : value(value)
1090 {
1091 this->ir_type = ir_type_return;
1092 }
1093
1094 virtual ir_return *clone(void *mem_ctx, struct hash_table *) const;
1095
1096 virtual ir_return *as_return()
1097 {
1098 return this;
1099 }
1100
1101 ir_rvalue *get_value() const
1102 {
1103 return value;
1104 }
1105
1106 virtual void accept(ir_visitor *v)
1107 {
1108 v->visit(this);
1109 }
1110
1111 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1112
1113 ir_rvalue *value;
1114 };
1115
1116
1117 /**
1118 * Jump instructions used inside loops
1119 *
1120 * These include \c break and \c continue. The \c break within a loop is
1121 * different from the \c break within a switch-statement.
1122 *
1123 * \sa ir_switch_jump
1124 */
1125 class ir_loop_jump : public ir_jump {
1126 public:
1127 enum jump_mode {
1128 jump_break,
1129 jump_continue
1130 };
1131
1132 ir_loop_jump(jump_mode mode)
1133 {
1134 this->ir_type = ir_type_loop_jump;
1135 this->mode = mode;
1136 this->loop = loop;
1137 }
1138
1139 virtual ir_loop_jump *clone(void *mem_ctx, struct hash_table *) const;
1140
1141 virtual void accept(ir_visitor *v)
1142 {
1143 v->visit(this);
1144 }
1145
1146 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1147
1148 bool is_break() const
1149 {
1150 return mode == jump_break;
1151 }
1152
1153 bool is_continue() const
1154 {
1155 return mode == jump_continue;
1156 }
1157
1158 /** Mode selector for the jump instruction. */
1159 enum jump_mode mode;
1160 private:
1161 /** Loop containing this break instruction. */
1162 ir_loop *loop;
1163 };
1164
1165 /**
1166 * IR instruction representing discard statements.
1167 */
1168 class ir_discard : public ir_jump {
1169 public:
1170 ir_discard()
1171 {
1172 this->ir_type = ir_type_discard;
1173 this->condition = NULL;
1174 }
1175
1176 ir_discard(ir_rvalue *cond)
1177 {
1178 this->ir_type = ir_type_discard;
1179 this->condition = cond;
1180 }
1181
1182 virtual ir_discard *clone(void *mem_ctx, struct hash_table *ht) const;
1183
1184 virtual void accept(ir_visitor *v)
1185 {
1186 v->visit(this);
1187 }
1188
1189 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1190
1191 virtual ir_discard *as_discard()
1192 {
1193 return this;
1194 }
1195
1196 ir_rvalue *condition;
1197 };
1198 /*@}*/
1199
1200
1201 /**
1202 * Texture sampling opcodes used in ir_texture
1203 */
1204 enum ir_texture_opcode {
1205 ir_tex, /**< Regular texture look-up */
1206 ir_txb, /**< Texture look-up with LOD bias */
1207 ir_txl, /**< Texture look-up with explicit LOD */
1208 ir_txd, /**< Texture look-up with partial derivatvies */
1209 ir_txf /**< Texel fetch with explicit LOD */
1210 };
1211
1212
1213 /**
1214 * IR instruction to sample a texture
1215 *
1216 * The specific form of the IR instruction depends on the \c mode value
1217 * selected from \c ir_texture_opcodes. In the printed IR, these will
1218 * appear as:
1219 *
1220 * Texel offset (0 or an expression)
1221 * | Projection divisor
1222 * | | Shadow comparitor
1223 * | | |
1224 * v v v
1225 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1226 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1227 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1228 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1229 * (txf <type> <sampler> <coordinate> 0 <lod>)
1230 */
1231 class ir_texture : public ir_rvalue {
1232 public:
1233 ir_texture(enum ir_texture_opcode op)
1234 : op(op), projector(NULL), shadow_comparitor(NULL), offset(NULL)
1235 {
1236 this->ir_type = ir_type_texture;
1237 }
1238
1239 virtual ir_texture *clone(void *mem_ctx, struct hash_table *) const;
1240
1241 virtual ir_constant *constant_expression_value();
1242
1243 virtual void accept(ir_visitor *v)
1244 {
1245 v->visit(this);
1246 }
1247
1248 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1249
1250 /**
1251 * Return a string representing the ir_texture_opcode.
1252 */
1253 const char *opcode_string();
1254
1255 /** Set the sampler and type. */
1256 void set_sampler(ir_dereference *sampler, const glsl_type *type);
1257
1258 /**
1259 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1260 */
1261 static ir_texture_opcode get_opcode(const char *);
1262
1263 enum ir_texture_opcode op;
1264
1265 /** Sampler to use for the texture access. */
1266 ir_dereference *sampler;
1267
1268 /** Texture coordinate to sample */
1269 ir_rvalue *coordinate;
1270
1271 /**
1272 * Value used for projective divide.
1273 *
1274 * If there is no projective divide (the common case), this will be
1275 * \c NULL. Optimization passes should check for this to point to a constant
1276 * of 1.0 and replace that with \c NULL.
1277 */
1278 ir_rvalue *projector;
1279
1280 /**
1281 * Coordinate used for comparison on shadow look-ups.
1282 *
1283 * If there is no shadow comparison, this will be \c NULL. For the
1284 * \c ir_txf opcode, this *must* be \c NULL.
1285 */
1286 ir_rvalue *shadow_comparitor;
1287
1288 /** Texel offset. */
1289 ir_rvalue *offset;
1290
1291 union {
1292 ir_rvalue *lod; /**< Floating point LOD */
1293 ir_rvalue *bias; /**< Floating point LOD bias */
1294 struct {
1295 ir_rvalue *dPdx; /**< Partial derivative of coordinate wrt X */
1296 ir_rvalue *dPdy; /**< Partial derivative of coordinate wrt Y */
1297 } grad;
1298 } lod_info;
1299 };
1300
1301
1302 struct ir_swizzle_mask {
1303 unsigned x:2;
1304 unsigned y:2;
1305 unsigned z:2;
1306 unsigned w:2;
1307
1308 /**
1309 * Number of components in the swizzle.
1310 */
1311 unsigned num_components:3;
1312
1313 /**
1314 * Does the swizzle contain duplicate components?
1315 *
1316 * L-value swizzles cannot contain duplicate components.
1317 */
1318 unsigned has_duplicates:1;
1319 };
1320
1321
1322 class ir_swizzle : public ir_rvalue {
1323 public:
1324 ir_swizzle(ir_rvalue *, unsigned x, unsigned y, unsigned z, unsigned w,
1325 unsigned count);
1326
1327 ir_swizzle(ir_rvalue *val, const unsigned *components, unsigned count);
1328
1329 ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask);
1330
1331 virtual ir_swizzle *clone(void *mem_ctx, struct hash_table *) const;
1332
1333 virtual ir_constant *constant_expression_value();
1334
1335 virtual ir_swizzle *as_swizzle()
1336 {
1337 return this;
1338 }
1339
1340 /**
1341 * Construct an ir_swizzle from the textual representation. Can fail.
1342 */
1343 static ir_swizzle *create(ir_rvalue *, const char *, unsigned vector_length);
1344
1345 virtual void accept(ir_visitor *v)
1346 {
1347 v->visit(this);
1348 }
1349
1350 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1351
1352 bool is_lvalue()
1353 {
1354 return val->is_lvalue() && !mask.has_duplicates;
1355 }
1356
1357 /**
1358 * Get the variable that is ultimately referenced by an r-value
1359 */
1360 virtual ir_variable *variable_referenced();
1361
1362 ir_rvalue *val;
1363 ir_swizzle_mask mask;
1364
1365 private:
1366 /**
1367 * Initialize the mask component of a swizzle
1368 *
1369 * This is used by the \c ir_swizzle constructors.
1370 */
1371 void init_mask(const unsigned *components, unsigned count);
1372 };
1373
1374
1375 class ir_dereference : public ir_rvalue {
1376 public:
1377 virtual ir_dereference *clone(void *mem_ctx, struct hash_table *) const = 0;
1378
1379 virtual ir_dereference *as_dereference()
1380 {
1381 return this;
1382 }
1383
1384 bool is_lvalue();
1385
1386 /**
1387 * Get the variable that is ultimately referenced by an r-value
1388 */
1389 virtual ir_variable *variable_referenced() = 0;
1390 };
1391
1392
1393 class ir_dereference_variable : public ir_dereference {
1394 public:
1395 ir_dereference_variable(ir_variable *var);
1396
1397 virtual ir_dereference_variable *clone(void *mem_ctx,
1398 struct hash_table *) const;
1399
1400 virtual ir_constant *constant_expression_value();
1401
1402 virtual ir_dereference_variable *as_dereference_variable()
1403 {
1404 return this;
1405 }
1406
1407 /**
1408 * Get the variable that is ultimately referenced by an r-value
1409 */
1410 virtual ir_variable *variable_referenced()
1411 {
1412 return this->var;
1413 }
1414
1415 virtual ir_variable *whole_variable_referenced()
1416 {
1417 /* ir_dereference_variable objects always dereference the entire
1418 * variable. However, if this dereference is dereferenced by anything
1419 * else, the complete deferefernce chain is not a whole-variable
1420 * dereference. This method should only be called on the top most
1421 * ir_rvalue in a dereference chain.
1422 */
1423 return this->var;
1424 }
1425
1426 virtual void accept(ir_visitor *v)
1427 {
1428 v->visit(this);
1429 }
1430
1431 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1432
1433 /**
1434 * Object being dereferenced.
1435 */
1436 ir_variable *var;
1437 };
1438
1439
1440 class ir_dereference_array : public ir_dereference {
1441 public:
1442 ir_dereference_array(ir_rvalue *value, ir_rvalue *array_index);
1443
1444 ir_dereference_array(ir_variable *var, ir_rvalue *array_index);
1445
1446 virtual ir_dereference_array *clone(void *mem_ctx,
1447 struct hash_table *) const;
1448
1449 virtual ir_constant *constant_expression_value();
1450
1451 virtual ir_dereference_array *as_dereference_array()
1452 {
1453 return this;
1454 }
1455
1456 /**
1457 * Get the variable that is ultimately referenced by an r-value
1458 */
1459 virtual ir_variable *variable_referenced()
1460 {
1461 return this->array->variable_referenced();
1462 }
1463
1464 virtual void accept(ir_visitor *v)
1465 {
1466 v->visit(this);
1467 }
1468
1469 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1470
1471 ir_rvalue *array;
1472 ir_rvalue *array_index;
1473
1474 private:
1475 void set_array(ir_rvalue *value);
1476 };
1477
1478
1479 class ir_dereference_record : public ir_dereference {
1480 public:
1481 ir_dereference_record(ir_rvalue *value, const char *field);
1482
1483 ir_dereference_record(ir_variable *var, const char *field);
1484
1485 virtual ir_dereference_record *clone(void *mem_ctx,
1486 struct hash_table *) const;
1487
1488 virtual ir_constant *constant_expression_value();
1489
1490 /**
1491 * Get the variable that is ultimately referenced by an r-value
1492 */
1493 virtual ir_variable *variable_referenced()
1494 {
1495 return this->record->variable_referenced();
1496 }
1497
1498 virtual void accept(ir_visitor *v)
1499 {
1500 v->visit(this);
1501 }
1502
1503 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1504
1505 ir_rvalue *record;
1506 const char *field;
1507 };
1508
1509
1510 /**
1511 * Data stored in an ir_constant
1512 */
1513 union ir_constant_data {
1514 unsigned u[16];
1515 int i[16];
1516 float f[16];
1517 bool b[16];
1518 };
1519
1520
1521 class ir_constant : public ir_rvalue {
1522 public:
1523 ir_constant(const struct glsl_type *type, const ir_constant_data *data);
1524 ir_constant(bool b);
1525 ir_constant(unsigned int u);
1526 ir_constant(int i);
1527 ir_constant(float f);
1528
1529 /**
1530 * Construct an ir_constant from a list of ir_constant values
1531 */
1532 ir_constant(const struct glsl_type *type, exec_list *values);
1533
1534 /**
1535 * Construct an ir_constant from a scalar component of another ir_constant
1536 *
1537 * The new \c ir_constant inherits the type of the component from the
1538 * source constant.
1539 *
1540 * \note
1541 * In the case of a matrix constant, the new constant is a scalar, \b not
1542 * a vector.
1543 */
1544 ir_constant(const ir_constant *c, unsigned i);
1545
1546 /**
1547 * Return a new ir_constant of the specified type containing all zeros.
1548 */
1549 static ir_constant *zero(void *mem_ctx, const glsl_type *type);
1550
1551 virtual ir_constant *clone(void *mem_ctx, struct hash_table *) const;
1552
1553 virtual ir_constant *constant_expression_value();
1554
1555 virtual ir_constant *as_constant()
1556 {
1557 return this;
1558 }
1559
1560 virtual void accept(ir_visitor *v)
1561 {
1562 v->visit(this);
1563 }
1564
1565 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1566
1567 /**
1568 * Get a particular component of a constant as a specific type
1569 *
1570 * This is useful, for example, to get a value from an integer constant
1571 * as a float or bool. This appears frequently when constructors are
1572 * called with all constant parameters.
1573 */
1574 /*@{*/
1575 bool get_bool_component(unsigned i) const;
1576 float get_float_component(unsigned i) const;
1577 int get_int_component(unsigned i) const;
1578 unsigned get_uint_component(unsigned i) const;
1579 /*@}*/
1580
1581 ir_constant *get_array_element(unsigned i) const;
1582
1583 ir_constant *get_record_field(const char *name);
1584
1585 /**
1586 * Determine whether a constant has the same value as another constant
1587 *
1588 * \sa ir_constant::is_zero, ir_constant::is_one,
1589 * ir_constant::is_negative_one
1590 */
1591 bool has_value(const ir_constant *) const;
1592
1593 virtual bool is_zero() const;
1594 virtual bool is_one() const;
1595 virtual bool is_negative_one() const;
1596
1597 /**
1598 * Value of the constant.
1599 *
1600 * The field used to back the values supplied by the constant is determined
1601 * by the type associated with the \c ir_instruction. Constants may be
1602 * scalars, vectors, or matrices.
1603 */
1604 union ir_constant_data value;
1605
1606 /* Array elements */
1607 ir_constant **array_elements;
1608
1609 /* Structure fields */
1610 exec_list components;
1611
1612 private:
1613 /**
1614 * Parameterless constructor only used by the clone method
1615 */
1616 ir_constant(void);
1617 };
1618
1619 /*@}*/
1620
1621 /**
1622 * Apply a visitor to each IR node in a list
1623 */
1624 void
1625 visit_exec_list(exec_list *list, ir_visitor *visitor);
1626
1627 /**
1628 * Validate invariants on each IR node in a list
1629 */
1630 void validate_ir_tree(exec_list *instructions);
1631
1632 /**
1633 * Make a clone of each IR instruction in a list
1634 *
1635 * \param in List of IR instructions that are to be cloned
1636 * \param out List to hold the cloned instructions
1637 */
1638 void
1639 clone_ir_list(void *mem_ctx, exec_list *out, const exec_list *in);
1640
1641 extern void
1642 _mesa_glsl_initialize_variables(exec_list *instructions,
1643 struct _mesa_glsl_parse_state *state);
1644
1645 extern void
1646 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state *state);
1647
1648 extern void
1649 _mesa_glsl_release_functions(void);
1650
1651 extern void
1652 reparent_ir(exec_list *list, void *mem_ctx);
1653
1654 struct glsl_symbol_table;
1655
1656 extern void
1657 import_prototypes(const exec_list *source, exec_list *dest,
1658 struct glsl_symbol_table *symbols, void *mem_ctx);
1659
1660 extern bool
1661 ir_has_call(ir_instruction *ir);
1662
1663 extern void
1664 do_set_program_inouts(exec_list *instructions, struct gl_program *prog);
1665
1666 #endif /* IR_H */