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