Merge remote 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 class ir_variable : public ir_instruction {
257 public:
258 ir_variable(const struct glsl_type *, const char *, ir_variable_mode);
259
260 virtual ir_variable *clone(void *mem_ctx, struct hash_table *ht) const;
261
262 virtual ir_variable *as_variable()
263 {
264 return this;
265 }
266
267 virtual void accept(ir_visitor *v)
268 {
269 v->visit(this);
270 }
271
272 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
273
274
275 /**
276 * Get the string value for the interpolation qualifier
277 *
278 * \return The string that would be used in a shader to specify \c
279 * mode will be returned.
280 *
281 * This function should only be used on a shader input or output variable.
282 */
283 const char *interpolation_string() const;
284
285 /**
286 * Calculate the number of slots required to hold this variable
287 *
288 * This is used to determine how many uniform or varying locations a variable
289 * occupies. The count is in units of floating point components.
290 */
291 unsigned component_slots() const;
292
293 /**
294 * Delcared name of the variable
295 */
296 const char *name;
297
298 /**
299 * Highest element accessed with a constant expression array index
300 *
301 * Not used for non-array variables.
302 */
303 unsigned max_array_access;
304
305 /**
306 * Is the variable read-only?
307 *
308 * This is set for variables declared as \c const, shader inputs,
309 * and uniforms.
310 */
311 unsigned read_only:1;
312 unsigned centroid:1;
313 unsigned invariant:1;
314
315 /**
316 * Has this variable been used for reading or writing?
317 *
318 * Several GLSL semantic checks require knowledge of whether or not a
319 * variable has been used. For example, it is an error to redeclare a
320 * variable as invariant after it has been used.
321 */
322 unsigned used:1;
323
324 /**
325 * Storage class of the variable.
326 *
327 * \sa ir_variable_mode
328 */
329 unsigned mode:3;
330
331 /**
332 * Interpolation mode for shader inputs / outputs
333 *
334 * \sa ir_variable_interpolation
335 */
336 unsigned interpolation:2;
337
338 /**
339 * Flag that the whole array is assignable
340 *
341 * In GLSL 1.20 and later whole arrays are assignable (and comparable for
342 * equality). This flag enables this behavior.
343 */
344 unsigned array_lvalue:1;
345
346 /**
347 * \name ARB_fragment_coord_conventions
348 * @{
349 */
350 unsigned origin_upper_left:1;
351 unsigned pixel_center_integer:1;
352 /*@}*/
353
354 /**
355 * \brief Layout qualifier for gl_FragDepth.
356 *
357 * This is not equal to \c ir_depth_layout_none if and only if this
358 * variable is \c gl_FragDepth and a layout qualifier is specified.
359 */
360 ir_depth_layout depth_layout;
361
362 /**
363 * Was the location explicitly set in the shader?
364 *
365 * If the location is explicitly set in the shader, it \b cannot be changed
366 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
367 * no effect).
368 */
369 unsigned explicit_location:1;
370
371 /**
372 * Storage location of the base of this variable
373 *
374 * The precise meaning of this field depends on the nature of the variable.
375 *
376 * - Vertex shader input: one of the values from \c gl_vert_attrib.
377 * - Vertex shader output: one of the values from \c gl_vert_result.
378 * - Fragment shader input: one of the values from \c gl_frag_attrib.
379 * - Fragment shader output: one of the values from \c gl_frag_result.
380 * - Uniforms: Per-stage uniform slot number.
381 * - Other: This field is not currently used.
382 *
383 * If the variable is a uniform, shader input, or shader output, and the
384 * slot has not been assigned, the value will be -1.
385 */
386 int location;
387
388 /**
389 * Emit a warning if this variable is accessed.
390 */
391 const char *warn_extension;
392
393 /**
394 * Value assigned in the initializer of a variable declared "const"
395 */
396 ir_constant *constant_value;
397 };
398
399
400 /*@{*/
401 /**
402 * The representation of a function instance; may be the full definition or
403 * simply a prototype.
404 */
405 class ir_function_signature : public ir_instruction {
406 /* An ir_function_signature will be part of the list of signatures in
407 * an ir_function.
408 */
409 public:
410 ir_function_signature(const glsl_type *return_type);
411
412 virtual ir_function_signature *clone(void *mem_ctx,
413 struct hash_table *ht) const;
414 ir_function_signature *clone_prototype(void *mem_ctx,
415 struct hash_table *ht) const;
416
417 virtual void accept(ir_visitor *v)
418 {
419 v->visit(this);
420 }
421
422 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
423
424 /**
425 * Get the name of the function for which this is a signature
426 */
427 const char *function_name() const;
428
429 /**
430 * Get a handle to the function for which this is a signature
431 *
432 * There is no setter function, this function returns a \c const pointer,
433 * and \c ir_function_signature::_function is private for a reason. The
434 * only way to make a connection between a function and function signature
435 * is via \c ir_function::add_signature. This helps ensure that certain
436 * invariants (i.e., a function signature is in the list of signatures for
437 * its \c _function) are met.
438 *
439 * \sa ir_function::add_signature
440 */
441 inline const class ir_function *function() const
442 {
443 return this->_function;
444 }
445
446 /**
447 * Check whether the qualifiers match between this signature's parameters
448 * and the supplied parameter list. If not, returns the name of the first
449 * parameter with mismatched qualifiers (for use in error messages).
450 */
451 const char *qualifiers_match(exec_list *params);
452
453 /**
454 * Replace the current parameter list with the given one. This is useful
455 * if the current information came from a prototype, and either has invalid
456 * or missing parameter names.
457 */
458 void replace_parameters(exec_list *new_params);
459
460 /**
461 * Function return type.
462 *
463 * \note This discards the optional precision qualifier.
464 */
465 const struct glsl_type *return_type;
466
467 /**
468 * List of ir_variable of function parameters.
469 *
470 * This represents the storage. The paramaters passed in a particular
471 * call will be in ir_call::actual_paramaters.
472 */
473 struct exec_list parameters;
474
475 /** Whether or not this function has a body (which may be empty). */
476 unsigned is_defined:1;
477
478 /** Whether or not this function signature is a built-in. */
479 unsigned is_builtin:1;
480
481 /** Body of instructions in the function. */
482 struct exec_list body;
483
484 private:
485 /** Function of which this signature is one overload. */
486 class ir_function *_function;
487
488 friend class ir_function;
489 };
490
491
492 /**
493 * Header for tracking multiple overloaded functions with the same name.
494 * Contains a list of ir_function_signatures representing each of the
495 * actual functions.
496 */
497 class ir_function : public ir_instruction {
498 public:
499 ir_function(const char *name);
500
501 virtual ir_function *clone(void *mem_ctx, struct hash_table *ht) const;
502
503 virtual ir_function *as_function()
504 {
505 return this;
506 }
507
508 virtual void accept(ir_visitor *v)
509 {
510 v->visit(this);
511 }
512
513 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
514
515 void add_signature(ir_function_signature *sig)
516 {
517 sig->_function = this;
518 this->signatures.push_tail(sig);
519 }
520
521 /**
522 * Get an iterator for the set of function signatures
523 */
524 exec_list_iterator iterator()
525 {
526 return signatures.iterator();
527 }
528
529 /**
530 * Find a signature that matches a set of actual parameters, taking implicit
531 * conversions into account.
532 */
533 ir_function_signature *matching_signature(const exec_list *actual_param);
534
535 /**
536 * Find a signature that exactly matches a set of actual parameters without
537 * any implicit type conversions.
538 */
539 ir_function_signature *exact_matching_signature(const exec_list *actual_ps);
540
541 /**
542 * Name of the function.
543 */
544 const char *name;
545
546 /** Whether or not this function has a signature that isn't a built-in. */
547 bool has_user_signature();
548
549 /**
550 * List of ir_function_signature for each overloaded function with this name.
551 */
552 struct exec_list signatures;
553 };
554
555 inline const char *ir_function_signature::function_name() const
556 {
557 return this->_function->name;
558 }
559 /*@}*/
560
561
562 /**
563 * IR instruction representing high-level if-statements
564 */
565 class ir_if : public ir_instruction {
566 public:
567 ir_if(ir_rvalue *condition)
568 : condition(condition)
569 {
570 ir_type = ir_type_if;
571 }
572
573 virtual ir_if *clone(void *mem_ctx, struct hash_table *ht) const;
574
575 virtual ir_if *as_if()
576 {
577 return this;
578 }
579
580 virtual void accept(ir_visitor *v)
581 {
582 v->visit(this);
583 }
584
585 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
586
587 ir_rvalue *condition;
588 /** List of ir_instruction for the body of the then branch */
589 exec_list then_instructions;
590 /** List of ir_instruction for the body of the else branch */
591 exec_list else_instructions;
592 };
593
594
595 /**
596 * IR instruction representing a high-level loop structure.
597 */
598 class ir_loop : public ir_instruction {
599 public:
600 ir_loop();
601
602 virtual ir_loop *clone(void *mem_ctx, struct hash_table *ht) const;
603
604 virtual void accept(ir_visitor *v)
605 {
606 v->visit(this);
607 }
608
609 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
610
611 virtual ir_loop *as_loop()
612 {
613 return this;
614 }
615
616 /**
617 * Get an iterator for the instructions of the loop body
618 */
619 exec_list_iterator iterator()
620 {
621 return body_instructions.iterator();
622 }
623
624 /** List of ir_instruction that make up the body of the loop. */
625 exec_list body_instructions;
626
627 /**
628 * \name Loop counter and controls
629 *
630 * Represents a loop like a FORTRAN \c do-loop.
631 *
632 * \note
633 * If \c from and \c to are the same value, the loop will execute once.
634 */
635 /*@{*/
636 ir_rvalue *from; /** Value of the loop counter on the first
637 * iteration of the loop.
638 */
639 ir_rvalue *to; /** Value of the loop counter on the last
640 * iteration of the loop.
641 */
642 ir_rvalue *increment;
643 ir_variable *counter;
644
645 /**
646 * Comparison operation in the loop terminator.
647 *
648 * If any of the loop control fields are non-\c NULL, this field must be
649 * one of \c ir_binop_less, \c ir_binop_greater, \c ir_binop_lequal,
650 * \c ir_binop_gequal, \c ir_binop_equal, or \c ir_binop_nequal.
651 */
652 int cmp;
653 /*@}*/
654 };
655
656
657 class ir_assignment : public ir_instruction {
658 public:
659 ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs, ir_rvalue *condition);
660
661 /**
662 * Construct an assignment with an explicit write mask
663 *
664 * \note
665 * Since a write mask is supplied, the LHS must already be a bare
666 * \c ir_dereference. The cannot be any swizzles in the LHS.
667 */
668 ir_assignment(ir_dereference *lhs, ir_rvalue *rhs, ir_rvalue *condition,
669 unsigned write_mask);
670
671 virtual ir_assignment *clone(void *mem_ctx, struct hash_table *ht) const;
672
673 virtual ir_constant *constant_expression_value();
674
675 virtual void accept(ir_visitor *v)
676 {
677 v->visit(this);
678 }
679
680 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
681
682 virtual ir_assignment * as_assignment()
683 {
684 return this;
685 }
686
687 /**
688 * Get a whole variable written by an assignment
689 *
690 * If the LHS of the assignment writes a whole variable, the variable is
691 * returned. Otherwise \c NULL is returned. Examples of whole-variable
692 * assignment are:
693 *
694 * - Assigning to a scalar
695 * - Assigning to all components of a vector
696 * - Whole array (or matrix) assignment
697 * - Whole structure assignment
698 */
699 ir_variable *whole_variable_written();
700
701 /**
702 * Set the LHS of an assignment
703 */
704 void set_lhs(ir_rvalue *lhs);
705
706 /**
707 * Left-hand side of the assignment.
708 *
709 * This should be treated as read only. If you need to set the LHS of an
710 * assignment, use \c ir_assignment::set_lhs.
711 */
712 ir_dereference *lhs;
713
714 /**
715 * Value being assigned
716 */
717 ir_rvalue *rhs;
718
719 /**
720 * Optional condition for the assignment.
721 */
722 ir_rvalue *condition;
723
724
725 /**
726 * Component mask written
727 *
728 * For non-vector types in the LHS, this field will be zero. For vector
729 * types, a bit will be set for each component that is written. Note that
730 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
731 *
732 * A partially-set write mask means that each enabled channel gets
733 * the value from a consecutive channel of the rhs. For example,
734 * to write just .xyw of gl_FrontColor with color:
735 *
736 * (assign (constant bool (1)) (xyw)
737 * (var_ref gl_FragColor)
738 * (swiz xyw (var_ref color)))
739 */
740 unsigned write_mask:4;
741 };
742
743 /* Update ir_expression::num_operands() and operator_strs when
744 * updating this list.
745 */
746 enum ir_expression_operation {
747 ir_unop_bit_not,
748 ir_unop_logic_not,
749 ir_unop_neg,
750 ir_unop_abs,
751 ir_unop_sign,
752 ir_unop_rcp,
753 ir_unop_rsq,
754 ir_unop_sqrt,
755 ir_unop_exp, /**< Log base e on gentype */
756 ir_unop_log, /**< Natural log on gentype */
757 ir_unop_exp2,
758 ir_unop_log2,
759 ir_unop_f2i, /**< Float-to-integer conversion. */
760 ir_unop_i2f, /**< Integer-to-float conversion. */
761 ir_unop_f2b, /**< Float-to-boolean conversion */
762 ir_unop_b2f, /**< Boolean-to-float conversion */
763 ir_unop_i2b, /**< int-to-boolean conversion */
764 ir_unop_b2i, /**< Boolean-to-int conversion */
765 ir_unop_u2f, /**< Unsigned-to-float conversion. */
766 ir_unop_any,
767
768 /**
769 * \name Unary floating-point rounding operations.
770 */
771 /*@{*/
772 ir_unop_trunc,
773 ir_unop_ceil,
774 ir_unop_floor,
775 ir_unop_fract,
776 ir_unop_round_even,
777 /*@}*/
778
779 /**
780 * \name Trigonometric operations.
781 */
782 /*@{*/
783 ir_unop_sin,
784 ir_unop_cos,
785 ir_unop_sin_reduced, /**< Reduced range sin. [-pi, pi] */
786 ir_unop_cos_reduced, /**< Reduced range cos. [-pi, pi] */
787 /*@}*/
788
789 /**
790 * \name Partial derivatives.
791 */
792 /*@{*/
793 ir_unop_dFdx,
794 ir_unop_dFdy,
795 /*@}*/
796
797 ir_unop_noise,
798
799 /**
800 * A sentinel marking the last of the unary operations.
801 */
802 ir_last_unop = ir_unop_noise,
803
804 ir_binop_add,
805 ir_binop_sub,
806 ir_binop_mul,
807 ir_binop_div,
808
809 /**
810 * Takes one of two combinations of arguments:
811 *
812 * - mod(vecN, vecN)
813 * - mod(vecN, float)
814 *
815 * Does not take integer types.
816 */
817 ir_binop_mod,
818
819 /**
820 * \name Binary comparison operators which return a boolean vector.
821 * The type of both operands must be equal.
822 */
823 /*@{*/
824 ir_binop_less,
825 ir_binop_greater,
826 ir_binop_lequal,
827 ir_binop_gequal,
828 ir_binop_equal,
829 ir_binop_nequal,
830 /**
831 * Returns single boolean for whether all components of operands[0]
832 * equal the components of operands[1].
833 */
834 ir_binop_all_equal,
835 /**
836 * Returns single boolean for whether any component of operands[0]
837 * is not equal to the corresponding component of operands[1].
838 */
839 ir_binop_any_nequal,
840 /*@}*/
841
842 /**
843 * \name Bit-wise binary operations.
844 */
845 /*@{*/
846 ir_binop_lshift,
847 ir_binop_rshift,
848 ir_binop_bit_and,
849 ir_binop_bit_xor,
850 ir_binop_bit_or,
851 /*@}*/
852
853 ir_binop_logic_and,
854 ir_binop_logic_xor,
855 ir_binop_logic_or,
856
857 ir_binop_dot,
858 ir_binop_min,
859 ir_binop_max,
860
861 ir_binop_pow,
862
863 /**
864 * A sentinel marking the last of the binary operations.
865 */
866 ir_last_binop = ir_binop_pow,
867
868 ir_quadop_vector,
869
870 /**
871 * A sentinel marking the last of all operations.
872 */
873 ir_last_opcode = ir_last_binop
874 };
875
876 class ir_expression : public ir_rvalue {
877 public:
878 /**
879 * Constructor for unary operation expressions
880 */
881 ir_expression(int op, const struct glsl_type *type, ir_rvalue *);
882 ir_expression(int op, ir_rvalue *);
883
884 /**
885 * Constructor for binary operation expressions
886 */
887 ir_expression(int op, const struct glsl_type *type,
888 ir_rvalue *, ir_rvalue *);
889 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1);
890
891 /**
892 * Constructor for quad operator expressions
893 */
894 ir_expression(int op, const struct glsl_type *type,
895 ir_rvalue *, ir_rvalue *, ir_rvalue *, ir_rvalue *);
896
897 virtual ir_expression *as_expression()
898 {
899 return this;
900 }
901
902 virtual ir_expression *clone(void *mem_ctx, struct hash_table *ht) const;
903
904 /**
905 * Attempt to constant-fold the expression
906 *
907 * If the expression cannot be constant folded, this method will return
908 * \c NULL.
909 */
910 virtual ir_constant *constant_expression_value();
911
912 /**
913 * Determine the number of operands used by an expression
914 */
915 static unsigned int get_num_operands(ir_expression_operation);
916
917 /**
918 * Determine the number of operands used by an expression
919 */
920 unsigned int get_num_operands() const
921 {
922 return (this->operation == ir_quadop_vector)
923 ? this->type->vector_elements : get_num_operands(operation);
924 }
925
926 /**
927 * Return a string representing this expression's operator.
928 */
929 const char *operator_string();
930
931 /**
932 * Return a string representing this expression's operator.
933 */
934 static const char *operator_string(ir_expression_operation);
935
936
937 /**
938 * Do a reverse-lookup to translate the given string into an operator.
939 */
940 static ir_expression_operation get_operator(const char *);
941
942 virtual void accept(ir_visitor *v)
943 {
944 v->visit(this);
945 }
946
947 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
948
949 ir_expression_operation operation;
950 ir_rvalue *operands[4];
951 };
952
953
954 /**
955 * IR instruction representing a function call
956 */
957 class ir_call : public ir_rvalue {
958 public:
959 ir_call(ir_function_signature *callee, exec_list *actual_parameters)
960 : callee(callee)
961 {
962 ir_type = ir_type_call;
963 assert(callee->return_type != NULL);
964 type = callee->return_type;
965 actual_parameters->move_nodes_to(& this->actual_parameters);
966 }
967
968 virtual ir_call *clone(void *mem_ctx, struct hash_table *ht) const;
969
970 virtual ir_constant *constant_expression_value();
971
972 virtual ir_call *as_call()
973 {
974 return this;
975 }
976
977 virtual void accept(ir_visitor *v)
978 {
979 v->visit(this);
980 }
981
982 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
983
984 /**
985 * Get a generic ir_call object when an error occurs
986 *
987 * Any allocation will be performed with 'ctx' as ralloc owner.
988 */
989 static ir_call *get_error_instruction(void *ctx);
990
991 /**
992 * Get an iterator for the set of acutal parameters
993 */
994 exec_list_iterator iterator()
995 {
996 return actual_parameters.iterator();
997 }
998
999 /**
1000 * Get the name of the function being called.
1001 */
1002 const char *callee_name() const
1003 {
1004 return callee->function_name();
1005 }
1006
1007 /**
1008 * Get the function signature bound to this function call
1009 */
1010 ir_function_signature *get_callee()
1011 {
1012 return callee;
1013 }
1014
1015 /**
1016 * Set the function call target
1017 */
1018 void set_callee(ir_function_signature *sig);
1019
1020 /**
1021 * Generates an inline version of the function before @ir,
1022 * returning the return value of the function.
1023 */
1024 ir_rvalue *generate_inline(ir_instruction *ir);
1025
1026 /* List of ir_rvalue of paramaters passed in this call. */
1027 exec_list actual_parameters;
1028
1029 private:
1030 ir_call()
1031 : callee(NULL)
1032 {
1033 this->ir_type = ir_type_call;
1034 }
1035
1036 ir_function_signature *callee;
1037 };
1038
1039
1040 /**
1041 * \name Jump-like IR instructions.
1042 *
1043 * These include \c break, \c continue, \c return, and \c discard.
1044 */
1045 /*@{*/
1046 class ir_jump : public ir_instruction {
1047 protected:
1048 ir_jump()
1049 {
1050 ir_type = ir_type_unset;
1051 }
1052 };
1053
1054 class ir_return : public ir_jump {
1055 public:
1056 ir_return()
1057 : value(NULL)
1058 {
1059 this->ir_type = ir_type_return;
1060 }
1061
1062 ir_return(ir_rvalue *value)
1063 : value(value)
1064 {
1065 this->ir_type = ir_type_return;
1066 }
1067
1068 virtual ir_return *clone(void *mem_ctx, struct hash_table *) const;
1069
1070 virtual ir_return *as_return()
1071 {
1072 return this;
1073 }
1074
1075 ir_rvalue *get_value() const
1076 {
1077 return value;
1078 }
1079
1080 virtual void accept(ir_visitor *v)
1081 {
1082 v->visit(this);
1083 }
1084
1085 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1086
1087 ir_rvalue *value;
1088 };
1089
1090
1091 /**
1092 * Jump instructions used inside loops
1093 *
1094 * These include \c break and \c continue. The \c break within a loop is
1095 * different from the \c break within a switch-statement.
1096 *
1097 * \sa ir_switch_jump
1098 */
1099 class ir_loop_jump : public ir_jump {
1100 public:
1101 enum jump_mode {
1102 jump_break,
1103 jump_continue
1104 };
1105
1106 ir_loop_jump(jump_mode mode)
1107 {
1108 this->ir_type = ir_type_loop_jump;
1109 this->mode = mode;
1110 this->loop = loop;
1111 }
1112
1113 virtual ir_loop_jump *clone(void *mem_ctx, struct hash_table *) const;
1114
1115 virtual void accept(ir_visitor *v)
1116 {
1117 v->visit(this);
1118 }
1119
1120 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1121
1122 bool is_break() const
1123 {
1124 return mode == jump_break;
1125 }
1126
1127 bool is_continue() const
1128 {
1129 return mode == jump_continue;
1130 }
1131
1132 /** Mode selector for the jump instruction. */
1133 enum jump_mode mode;
1134 private:
1135 /** Loop containing this break instruction. */
1136 ir_loop *loop;
1137 };
1138
1139 /**
1140 * IR instruction representing discard statements.
1141 */
1142 class ir_discard : public ir_jump {
1143 public:
1144 ir_discard()
1145 {
1146 this->ir_type = ir_type_discard;
1147 this->condition = NULL;
1148 }
1149
1150 ir_discard(ir_rvalue *cond)
1151 {
1152 this->ir_type = ir_type_discard;
1153 this->condition = cond;
1154 }
1155
1156 virtual ir_discard *clone(void *mem_ctx, struct hash_table *ht) const;
1157
1158 virtual void accept(ir_visitor *v)
1159 {
1160 v->visit(this);
1161 }
1162
1163 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1164
1165 virtual ir_discard *as_discard()
1166 {
1167 return this;
1168 }
1169
1170 ir_rvalue *condition;
1171 };
1172 /*@}*/
1173
1174
1175 /**
1176 * Texture sampling opcodes used in ir_texture
1177 */
1178 enum ir_texture_opcode {
1179 ir_tex, /**< Regular texture look-up */
1180 ir_txb, /**< Texture look-up with LOD bias */
1181 ir_txl, /**< Texture look-up with explicit LOD */
1182 ir_txd, /**< Texture look-up with partial derivatvies */
1183 ir_txf /**< Texel fetch with explicit LOD */
1184 };
1185
1186
1187 /**
1188 * IR instruction to sample a texture
1189 *
1190 * The specific form of the IR instruction depends on the \c mode value
1191 * selected from \c ir_texture_opcodes. In the printed IR, these will
1192 * appear as:
1193 *
1194 * Texel offset (0 or an expression)
1195 * | Projection divisor
1196 * | | Shadow comparitor
1197 * | | |
1198 * v v v
1199 * (tex <sampler> <coordinate> 0 1 ( ))
1200 * (txb <sampler> <coordinate> 0 1 ( ) <bias>)
1201 * (txl <sampler> <coordinate> 0 1 ( ) <lod>)
1202 * (txd <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1203 * (txf <sampler> <coordinate> 0 <lod>)
1204 */
1205 class ir_texture : public ir_rvalue {
1206 public:
1207 ir_texture(enum ir_texture_opcode op)
1208 : op(op), projector(NULL), shadow_comparitor(NULL), offset(NULL)
1209 {
1210 this->ir_type = ir_type_texture;
1211 }
1212
1213 virtual ir_texture *clone(void *mem_ctx, struct hash_table *) const;
1214
1215 virtual ir_constant *constant_expression_value();
1216
1217 virtual void accept(ir_visitor *v)
1218 {
1219 v->visit(this);
1220 }
1221
1222 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1223
1224 /**
1225 * Return a string representing the ir_texture_opcode.
1226 */
1227 const char *opcode_string();
1228
1229 /** Set the sampler and infer the type. */
1230 void set_sampler(ir_dereference *sampler);
1231
1232 /**
1233 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1234 */
1235 static ir_texture_opcode get_opcode(const char *);
1236
1237 enum ir_texture_opcode op;
1238
1239 /** Sampler to use for the texture access. */
1240 ir_dereference *sampler;
1241
1242 /** Texture coordinate to sample */
1243 ir_rvalue *coordinate;
1244
1245 /**
1246 * Value used for projective divide.
1247 *
1248 * If there is no projective divide (the common case), this will be
1249 * \c NULL. Optimization passes should check for this to point to a constant
1250 * of 1.0 and replace that with \c NULL.
1251 */
1252 ir_rvalue *projector;
1253
1254 /**
1255 * Coordinate used for comparison on shadow look-ups.
1256 *
1257 * If there is no shadow comparison, this will be \c NULL. For the
1258 * \c ir_txf opcode, this *must* be \c NULL.
1259 */
1260 ir_rvalue *shadow_comparitor;
1261
1262 /** Texel offset. */
1263 ir_rvalue *offset;
1264
1265 union {
1266 ir_rvalue *lod; /**< Floating point LOD */
1267 ir_rvalue *bias; /**< Floating point LOD bias */
1268 struct {
1269 ir_rvalue *dPdx; /**< Partial derivative of coordinate wrt X */
1270 ir_rvalue *dPdy; /**< Partial derivative of coordinate wrt Y */
1271 } grad;
1272 } lod_info;
1273 };
1274
1275
1276 struct ir_swizzle_mask {
1277 unsigned x:2;
1278 unsigned y:2;
1279 unsigned z:2;
1280 unsigned w:2;
1281
1282 /**
1283 * Number of components in the swizzle.
1284 */
1285 unsigned num_components:3;
1286
1287 /**
1288 * Does the swizzle contain duplicate components?
1289 *
1290 * L-value swizzles cannot contain duplicate components.
1291 */
1292 unsigned has_duplicates:1;
1293 };
1294
1295
1296 class ir_swizzle : public ir_rvalue {
1297 public:
1298 ir_swizzle(ir_rvalue *, unsigned x, unsigned y, unsigned z, unsigned w,
1299 unsigned count);
1300
1301 ir_swizzle(ir_rvalue *val, const unsigned *components, unsigned count);
1302
1303 ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask);
1304
1305 virtual ir_swizzle *clone(void *mem_ctx, struct hash_table *) const;
1306
1307 virtual ir_constant *constant_expression_value();
1308
1309 virtual ir_swizzle *as_swizzle()
1310 {
1311 return this;
1312 }
1313
1314 /**
1315 * Construct an ir_swizzle from the textual representation. Can fail.
1316 */
1317 static ir_swizzle *create(ir_rvalue *, const char *, unsigned vector_length);
1318
1319 virtual void accept(ir_visitor *v)
1320 {
1321 v->visit(this);
1322 }
1323
1324 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1325
1326 bool is_lvalue()
1327 {
1328 return val->is_lvalue() && !mask.has_duplicates;
1329 }
1330
1331 /**
1332 * Get the variable that is ultimately referenced by an r-value
1333 */
1334 virtual ir_variable *variable_referenced();
1335
1336 ir_rvalue *val;
1337 ir_swizzle_mask mask;
1338
1339 private:
1340 /**
1341 * Initialize the mask component of a swizzle
1342 *
1343 * This is used by the \c ir_swizzle constructors.
1344 */
1345 void init_mask(const unsigned *components, unsigned count);
1346 };
1347
1348
1349 class ir_dereference : public ir_rvalue {
1350 public:
1351 virtual ir_dereference *clone(void *mem_ctx, struct hash_table *) const = 0;
1352
1353 virtual ir_dereference *as_dereference()
1354 {
1355 return this;
1356 }
1357
1358 bool is_lvalue();
1359
1360 /**
1361 * Get the variable that is ultimately referenced by an r-value
1362 */
1363 virtual ir_variable *variable_referenced() = 0;
1364 };
1365
1366
1367 class ir_dereference_variable : public ir_dereference {
1368 public:
1369 ir_dereference_variable(ir_variable *var);
1370
1371 virtual ir_dereference_variable *clone(void *mem_ctx,
1372 struct hash_table *) const;
1373
1374 virtual ir_constant *constant_expression_value();
1375
1376 virtual ir_dereference_variable *as_dereference_variable()
1377 {
1378 return this;
1379 }
1380
1381 /**
1382 * Get the variable that is ultimately referenced by an r-value
1383 */
1384 virtual ir_variable *variable_referenced()
1385 {
1386 return this->var;
1387 }
1388
1389 virtual ir_variable *whole_variable_referenced()
1390 {
1391 /* ir_dereference_variable objects always dereference the entire
1392 * variable. However, if this dereference is dereferenced by anything
1393 * else, the complete deferefernce chain is not a whole-variable
1394 * dereference. This method should only be called on the top most
1395 * ir_rvalue in a dereference chain.
1396 */
1397 return this->var;
1398 }
1399
1400 virtual void accept(ir_visitor *v)
1401 {
1402 v->visit(this);
1403 }
1404
1405 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1406
1407 /**
1408 * Object being dereferenced.
1409 */
1410 ir_variable *var;
1411 };
1412
1413
1414 class ir_dereference_array : public ir_dereference {
1415 public:
1416 ir_dereference_array(ir_rvalue *value, ir_rvalue *array_index);
1417
1418 ir_dereference_array(ir_variable *var, ir_rvalue *array_index);
1419
1420 virtual ir_dereference_array *clone(void *mem_ctx,
1421 struct hash_table *) const;
1422
1423 virtual ir_constant *constant_expression_value();
1424
1425 virtual ir_dereference_array *as_dereference_array()
1426 {
1427 return this;
1428 }
1429
1430 /**
1431 * Get the variable that is ultimately referenced by an r-value
1432 */
1433 virtual ir_variable *variable_referenced()
1434 {
1435 return this->array->variable_referenced();
1436 }
1437
1438 virtual void accept(ir_visitor *v)
1439 {
1440 v->visit(this);
1441 }
1442
1443 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1444
1445 ir_rvalue *array;
1446 ir_rvalue *array_index;
1447
1448 private:
1449 void set_array(ir_rvalue *value);
1450 };
1451
1452
1453 class ir_dereference_record : public ir_dereference {
1454 public:
1455 ir_dereference_record(ir_rvalue *value, const char *field);
1456
1457 ir_dereference_record(ir_variable *var, const char *field);
1458
1459 virtual ir_dereference_record *clone(void *mem_ctx,
1460 struct hash_table *) const;
1461
1462 virtual ir_constant *constant_expression_value();
1463
1464 /**
1465 * Get the variable that is ultimately referenced by an r-value
1466 */
1467 virtual ir_variable *variable_referenced()
1468 {
1469 return this->record->variable_referenced();
1470 }
1471
1472 virtual void accept(ir_visitor *v)
1473 {
1474 v->visit(this);
1475 }
1476
1477 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1478
1479 ir_rvalue *record;
1480 const char *field;
1481 };
1482
1483
1484 /**
1485 * Data stored in an ir_constant
1486 */
1487 union ir_constant_data {
1488 unsigned u[16];
1489 int i[16];
1490 float f[16];
1491 bool b[16];
1492 };
1493
1494
1495 class ir_constant : public ir_rvalue {
1496 public:
1497 ir_constant(const struct glsl_type *type, const ir_constant_data *data);
1498 ir_constant(bool b);
1499 ir_constant(unsigned int u);
1500 ir_constant(int i);
1501 ir_constant(float f);
1502
1503 /**
1504 * Construct an ir_constant from a list of ir_constant values
1505 */
1506 ir_constant(const struct glsl_type *type, exec_list *values);
1507
1508 /**
1509 * Construct an ir_constant from a scalar component of another ir_constant
1510 *
1511 * The new \c ir_constant inherits the type of the component from the
1512 * source constant.
1513 *
1514 * \note
1515 * In the case of a matrix constant, the new constant is a scalar, \b not
1516 * a vector.
1517 */
1518 ir_constant(const ir_constant *c, unsigned i);
1519
1520 /**
1521 * Return a new ir_constant of the specified type containing all zeros.
1522 */
1523 static ir_constant *zero(void *mem_ctx, const glsl_type *type);
1524
1525 virtual ir_constant *clone(void *mem_ctx, struct hash_table *) const;
1526
1527 virtual ir_constant *constant_expression_value();
1528
1529 virtual ir_constant *as_constant()
1530 {
1531 return this;
1532 }
1533
1534 virtual void accept(ir_visitor *v)
1535 {
1536 v->visit(this);
1537 }
1538
1539 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1540
1541 /**
1542 * Get a particular component of a constant as a specific type
1543 *
1544 * This is useful, for example, to get a value from an integer constant
1545 * as a float or bool. This appears frequently when constructors are
1546 * called with all constant parameters.
1547 */
1548 /*@{*/
1549 bool get_bool_component(unsigned i) const;
1550 float get_float_component(unsigned i) const;
1551 int get_int_component(unsigned i) const;
1552 unsigned get_uint_component(unsigned i) const;
1553 /*@}*/
1554
1555 ir_constant *get_array_element(unsigned i) const;
1556
1557 ir_constant *get_record_field(const char *name);
1558
1559 /**
1560 * Determine whether a constant has the same value as another constant
1561 *
1562 * \sa ir_constant::is_zero, ir_constant::is_one,
1563 * ir_constant::is_negative_one
1564 */
1565 bool has_value(const ir_constant *) const;
1566
1567 virtual bool is_zero() const;
1568 virtual bool is_one() const;
1569 virtual bool is_negative_one() const;
1570
1571 /**
1572 * Value of the constant.
1573 *
1574 * The field used to back the values supplied by the constant is determined
1575 * by the type associated with the \c ir_instruction. Constants may be
1576 * scalars, vectors, or matrices.
1577 */
1578 union ir_constant_data value;
1579
1580 /* Array elements */
1581 ir_constant **array_elements;
1582
1583 /* Structure fields */
1584 exec_list components;
1585
1586 private:
1587 /**
1588 * Parameterless constructor only used by the clone method
1589 */
1590 ir_constant(void);
1591 };
1592
1593 /*@}*/
1594
1595 /**
1596 * Apply a visitor to each IR node in a list
1597 */
1598 void
1599 visit_exec_list(exec_list *list, ir_visitor *visitor);
1600
1601 /**
1602 * Validate invariants on each IR node in a list
1603 */
1604 void validate_ir_tree(exec_list *instructions);
1605
1606 /**
1607 * Make a clone of each IR instruction in a list
1608 *
1609 * \param in List of IR instructions that are to be cloned
1610 * \param out List to hold the cloned instructions
1611 */
1612 void
1613 clone_ir_list(void *mem_ctx, exec_list *out, const exec_list *in);
1614
1615 extern void
1616 _mesa_glsl_initialize_variables(exec_list *instructions,
1617 struct _mesa_glsl_parse_state *state);
1618
1619 extern void
1620 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state *state);
1621
1622 extern void
1623 _mesa_glsl_release_functions(void);
1624
1625 extern void
1626 reparent_ir(exec_list *list, void *mem_ctx);
1627
1628 struct glsl_symbol_table;
1629
1630 extern void
1631 import_prototypes(const exec_list *source, exec_list *dest,
1632 struct glsl_symbol_table *symbols, void *mem_ctx);
1633
1634 extern bool
1635 ir_has_call(ir_instruction *ir);
1636
1637 extern void
1638 do_set_program_inouts(exec_list *instructions, struct gl_program *prog);
1639
1640 #endif /* IR_H */