a927e90ca1f24418d432644093af316c8e2863ef
[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 #ifdef __cplusplus
40
41 /**
42 * \defgroup IR Intermediate representation nodes
43 *
44 * @{
45 */
46
47 /**
48 * Class tags
49 *
50 * Each concrete class derived from \c ir_instruction has a value in this
51 * enumerant. The value for the type is stored in \c ir_instruction::ir_type
52 * by the constructor. While using type tags is not very C++, it is extremely
53 * convenient. For example, during debugging you can simply inspect
54 * \c ir_instruction::ir_type to find out the actual type of the object.
55 *
56 * In addition, it is possible to use a switch-statement based on \c
57 * \c ir_instruction::ir_type to select different behavior for different object
58 * types. For functions that have only slight differences for several object
59 * types, this allows writing very straightforward, readable code.
60 */
61 enum ir_node_type {
62 /**
63 * Zero is unused so that the IR validator can detect cases where
64 * \c ir_instruction::ir_type has not been initialized.
65 */
66 ir_type_unset,
67 ir_type_variable,
68 ir_type_assignment,
69 ir_type_call,
70 ir_type_constant,
71 ir_type_dereference_array,
72 ir_type_dereference_record,
73 ir_type_dereference_variable,
74 ir_type_discard,
75 ir_type_expression,
76 ir_type_function,
77 ir_type_function_signature,
78 ir_type_if,
79 ir_type_loop,
80 ir_type_loop_jump,
81 ir_type_return,
82 ir_type_swizzle,
83 ir_type_texture,
84 ir_type_emit_vertex,
85 ir_type_end_primitive,
86 ir_type_max /**< maximum ir_type enum number, for validation */
87 };
88
89 /**
90 * Base class of all IR instructions
91 */
92 class ir_instruction : public exec_node {
93 public:
94 enum ir_node_type ir_type;
95
96 /**
97 * GCC 4.7+ and clang warn when deleting an ir_instruction unless
98 * there's a virtual destructor present. Because we almost
99 * universally use ralloc for our memory management of
100 * ir_instructions, the destructor doesn't need to do any work.
101 */
102 virtual ~ir_instruction()
103 {
104 }
105
106 /** ir_print_visitor helper for debugging. */
107 void print(void) const;
108
109 virtual void accept(ir_visitor *) = 0;
110 virtual ir_visitor_status accept(ir_hierarchical_visitor *) = 0;
111 virtual ir_instruction *clone(void *mem_ctx,
112 struct hash_table *ht) const = 0;
113
114 /**
115 * \name IR instruction downcast functions
116 *
117 * These functions either cast the object to a derived class or return
118 * \c NULL if the object's type does not match the specified derived class.
119 * Additional downcast functions will be added as needed.
120 */
121 /*@{*/
122 virtual class ir_variable * as_variable() { return NULL; }
123 virtual class ir_function * as_function() { return NULL; }
124 virtual class ir_dereference * as_dereference() { return NULL; }
125 virtual class ir_dereference_array * as_dereference_array() { return NULL; }
126 virtual class ir_dereference_variable *as_dereference_variable() { return NULL; }
127 virtual class ir_dereference_record *as_dereference_record() { return NULL; }
128 virtual class ir_expression * as_expression() { return NULL; }
129 virtual class ir_rvalue * as_rvalue() { return NULL; }
130 virtual class ir_loop * as_loop() { return NULL; }
131 virtual class ir_assignment * as_assignment() { return NULL; }
132 virtual class ir_call * as_call() { return NULL; }
133 virtual class ir_return * as_return() { return NULL; }
134 virtual class ir_if * as_if() { return NULL; }
135 virtual class ir_swizzle * as_swizzle() { return NULL; }
136 virtual class ir_constant * as_constant() { return NULL; }
137 virtual class ir_discard * as_discard() { return NULL; }
138 virtual class ir_jump * as_jump() { return NULL; }
139 /*@}*/
140
141 protected:
142 ir_instruction()
143 {
144 ir_type = ir_type_unset;
145 }
146 };
147
148
149 /**
150 * The base class for all "values"/expression trees.
151 */
152 class ir_rvalue : public ir_instruction {
153 public:
154 const struct glsl_type *type;
155
156 virtual ir_rvalue *clone(void *mem_ctx, struct hash_table *) const;
157
158 virtual void accept(ir_visitor *v)
159 {
160 v->visit(this);
161 }
162
163 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
164
165 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
166
167 virtual ir_rvalue * as_rvalue()
168 {
169 return this;
170 }
171
172 ir_rvalue *as_rvalue_to_saturate();
173
174 virtual bool is_lvalue() const
175 {
176 return false;
177 }
178
179 /**
180 * Get the variable that is ultimately referenced by an r-value
181 */
182 virtual ir_variable *variable_referenced() const
183 {
184 return NULL;
185 }
186
187
188 /**
189 * If an r-value is a reference to a whole variable, get that variable
190 *
191 * \return
192 * Pointer to a variable that is completely dereferenced by the r-value. If
193 * the r-value is not a dereference or the dereference does not access the
194 * entire variable (i.e., it's just one array element, struct field), \c NULL
195 * is returned.
196 */
197 virtual ir_variable *whole_variable_referenced()
198 {
199 return NULL;
200 }
201
202 /**
203 * Determine if an r-value has the value zero
204 *
205 * The base implementation of this function always returns \c false. The
206 * \c ir_constant class over-rides this function to return \c true \b only
207 * for vector and scalar types that have all elements set to the value
208 * zero (or \c false for booleans).
209 *
210 * \sa ir_constant::has_value, ir_rvalue::is_one, ir_rvalue::is_negative_one,
211 * ir_constant::is_basis
212 */
213 virtual bool is_zero() const;
214
215 /**
216 * Determine if an r-value has the value one
217 *
218 * The base implementation of this function always returns \c false. The
219 * \c ir_constant class over-rides this function to return \c true \b only
220 * for vector and scalar types that have all elements set to the value
221 * one (or \c true for booleans).
222 *
223 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_negative_one,
224 * ir_constant::is_basis
225 */
226 virtual bool is_one() const;
227
228 /**
229 * Determine if an r-value has the value negative one
230 *
231 * The base implementation of this function always returns \c false. The
232 * \c ir_constant class over-rides this function to return \c true \b only
233 * for vector and scalar types that have all elements set to the value
234 * negative one. For boolean types, the result is always \c false.
235 *
236 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one
237 * ir_constant::is_basis
238 */
239 virtual bool is_negative_one() const;
240
241 /**
242 * Determine if an r-value is a basis vector
243 *
244 * The base implementation of this function always returns \c false. The
245 * \c ir_constant class over-rides this function to return \c true \b only
246 * for vector and scalar types that have one element set to the value one,
247 * and the other elements set to the value zero. For boolean types, the
248 * result is always \c false.
249 *
250 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one,
251 * is_constant::is_negative_one
252 */
253 virtual bool is_basis() const;
254
255
256 /**
257 * Return a generic value of error_type.
258 *
259 * Allocation will be performed with 'mem_ctx' as ralloc owner.
260 */
261 static ir_rvalue *error_value(void *mem_ctx);
262
263 protected:
264 ir_rvalue();
265 };
266
267
268 /**
269 * Variable storage classes
270 */
271 enum ir_variable_mode {
272 ir_var_auto = 0, /**< Function local variables and globals. */
273 ir_var_uniform, /**< Variable declared as a uniform. */
274 ir_var_shader_in,
275 ir_var_shader_out,
276 ir_var_function_in,
277 ir_var_function_out,
278 ir_var_function_inout,
279 ir_var_const_in, /**< "in" param that must be a constant expression */
280 ir_var_system_value, /**< Ex: front-face, instance-id, etc. */
281 ir_var_temporary, /**< Temporary variable generated during compilation. */
282 ir_var_mode_count /**< Number of variable modes */
283 };
284
285 /**
286 * \brief Layout qualifiers for gl_FragDepth.
287 *
288 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
289 * with a layout qualifier.
290 */
291 enum ir_depth_layout {
292 ir_depth_layout_none, /**< No depth layout is specified. */
293 ir_depth_layout_any,
294 ir_depth_layout_greater,
295 ir_depth_layout_less,
296 ir_depth_layout_unchanged
297 };
298
299 /**
300 * \brief Convert depth layout qualifier to string.
301 */
302 const char*
303 depth_layout_string(ir_depth_layout layout);
304
305 /**
306 * Description of built-in state associated with a uniform
307 *
308 * \sa ir_variable::state_slots
309 */
310 struct ir_state_slot {
311 int tokens[5];
312 int swizzle;
313 };
314
315 class ir_variable : public ir_instruction {
316 public:
317 ir_variable(const struct glsl_type *, const char *, ir_variable_mode);
318
319 virtual ir_variable *clone(void *mem_ctx, struct hash_table *ht) const;
320
321 virtual ir_variable *as_variable()
322 {
323 return this;
324 }
325
326 virtual void accept(ir_visitor *v)
327 {
328 v->visit(this);
329 }
330
331 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
332
333
334 /**
335 * Get the string value for the interpolation qualifier
336 *
337 * \return The string that would be used in a shader to specify \c
338 * mode will be returned.
339 *
340 * This function is used to generate error messages of the form "shader
341 * uses %s interpolation qualifier", so in the case where there is no
342 * interpolation qualifier, it returns "no".
343 *
344 * This function should only be used on a shader input or output variable.
345 */
346 const char *interpolation_string() const;
347
348 /**
349 * Determine how this variable should be interpolated based on its
350 * interpolation qualifier (if present), whether it is gl_Color or
351 * gl_SecondaryColor, and whether flatshading is enabled in the current GL
352 * state.
353 *
354 * The return value will always be either INTERP_QUALIFIER_SMOOTH,
355 * INTERP_QUALIFIER_NOPERSPECTIVE, or INTERP_QUALIFIER_FLAT.
356 */
357 glsl_interp_qualifier determine_interpolation_mode(bool flat_shade);
358
359 /**
360 * Determine whether or not a variable is part of a uniform block.
361 */
362 inline bool is_in_uniform_block() const
363 {
364 return this->mode == ir_var_uniform && this->interface_type != NULL;
365 }
366
367 /**
368 * Determine whether or not a variable is the declaration of an interface
369 * block
370 *
371 * For the first declaration below, there will be an \c ir_variable named
372 * "instance" whose type and whose instance_type will be the same
373 * \cglsl_type. For the second declaration, there will be an \c ir_variable
374 * named "f" whose type is float and whose instance_type is B2.
375 *
376 * "instance" is an interface instance variable, but "f" is not.
377 *
378 * uniform B1 {
379 * float f;
380 * } instance;
381 *
382 * uniform B2 {
383 * float f;
384 * };
385 */
386 inline bool is_interface_instance() const
387 {
388 const glsl_type *const t = this->type;
389
390 return (t == this->interface_type)
391 || (t->is_array() && t->fields.array == this->interface_type);
392 }
393
394 /**
395 * Declared type of the variable
396 */
397 const struct glsl_type *type;
398
399 /**
400 * Declared name of the variable
401 */
402 const char *name;
403
404 /**
405 * Highest element accessed with a constant expression array index
406 *
407 * Not used for non-array variables.
408 */
409 unsigned max_array_access;
410
411 /**
412 * Is the variable read-only?
413 *
414 * This is set for variables declared as \c const, shader inputs,
415 * and uniforms.
416 */
417 unsigned read_only:1;
418 unsigned centroid:1;
419 unsigned invariant:1;
420
421 /**
422 * Has this variable been used for reading or writing?
423 *
424 * Several GLSL semantic checks require knowledge of whether or not a
425 * variable has been used. For example, it is an error to redeclare a
426 * variable as invariant after it has been used.
427 *
428 * This is only maintained in the ast_to_hir.cpp path, not in
429 * Mesa's fixed function or ARB program paths.
430 */
431 unsigned used:1;
432
433 /**
434 * Has this variable been statically assigned?
435 *
436 * This answers whether the variable was assigned in any path of
437 * the shader during ast_to_hir. This doesn't answer whether it is
438 * still written after dead code removal, nor is it maintained in
439 * non-ast_to_hir.cpp (GLSL parsing) paths.
440 */
441 unsigned assigned:1;
442
443 /**
444 * Storage class of the variable.
445 *
446 * \sa ir_variable_mode
447 */
448 unsigned mode:4;
449
450 /**
451 * Interpolation mode for shader inputs / outputs
452 *
453 * \sa ir_variable_interpolation
454 */
455 unsigned interpolation:2;
456
457 /**
458 * \name ARB_fragment_coord_conventions
459 * @{
460 */
461 unsigned origin_upper_left:1;
462 unsigned pixel_center_integer:1;
463 /*@}*/
464
465 /**
466 * Was the location explicitly set in the shader?
467 *
468 * If the location is explicitly set in the shader, it \b cannot be changed
469 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
470 * no effect).
471 */
472 unsigned explicit_location:1;
473 unsigned explicit_index:1;
474
475 /**
476 * Was an initial binding explicitly set in the shader?
477 *
478 * If so, constant_value contains an integer ir_constant representing the
479 * initial binding point.
480 */
481 unsigned explicit_binding:1;
482
483 /**
484 * Does this variable have an initializer?
485 *
486 * This is used by the linker to cross-validiate initializers of global
487 * variables.
488 */
489 unsigned has_initializer:1;
490
491 /**
492 * Is this variable a generic output or input that has not yet been matched
493 * up to a variable in another stage of the pipeline?
494 *
495 * This is used by the linker as scratch storage while assigning locations
496 * to generic inputs and outputs.
497 */
498 unsigned is_unmatched_generic_inout:1;
499
500 /**
501 * If non-zero, then this variable may be packed along with other variables
502 * into a single varying slot, so this offset should be applied when
503 * accessing components. For example, an offset of 1 means that the x
504 * component of this variable is actually stored in component y of the
505 * location specified by \c location.
506 */
507 unsigned location_frac:2;
508
509 /**
510 * \brief Layout qualifier for gl_FragDepth.
511 *
512 * This is not equal to \c ir_depth_layout_none if and only if this
513 * variable is \c gl_FragDepth and a layout qualifier is specified.
514 */
515 ir_depth_layout depth_layout;
516
517 /**
518 * Storage location of the base of this variable
519 *
520 * The precise meaning of this field depends on the nature of the variable.
521 *
522 * - Vertex shader input: one of the values from \c gl_vert_attrib.
523 * - Vertex shader output: one of the values from \c gl_varying_slot.
524 * - Geometry shader input: one of the values from \c gl_varying_slot.
525 * - Geometry shader output: one of the values from \c gl_varying_slot.
526 * - Fragment shader input: one of the values from \c gl_varying_slot.
527 * - Fragment shader output: one of the values from \c gl_frag_result.
528 * - Uniforms: Per-stage uniform slot number for default uniform block.
529 * - Uniforms: Index within the uniform block definition for UBO members.
530 * - Other: This field is not currently used.
531 *
532 * If the variable is a uniform, shader input, or shader output, and the
533 * slot has not been assigned, the value will be -1.
534 */
535 int location;
536
537 /**
538 * output index for dual source blending.
539 */
540 int index;
541
542 /**
543 * Initial binding point for a sampler or UBO.
544 *
545 * For array types, this represents the binding point for the first element.
546 */
547 int binding;
548
549 /**
550 * Built-in state that backs this uniform
551 *
552 * Once set at variable creation, \c state_slots must remain invariant.
553 * This is because, ideally, this array would be shared by all clones of
554 * this variable in the IR tree. In other words, we'd really like for it
555 * to be a fly-weight.
556 *
557 * If the variable is not a uniform, \c num_state_slots will be zero and
558 * \c state_slots will be \c NULL.
559 */
560 /*@{*/
561 unsigned num_state_slots; /**< Number of state slots used */
562 ir_state_slot *state_slots; /**< State descriptors. */
563 /*@}*/
564
565 /**
566 * Emit a warning if this variable is accessed.
567 */
568 const char *warn_extension;
569
570 /**
571 * Value assigned in the initializer of a variable declared "const"
572 */
573 ir_constant *constant_value;
574
575 /**
576 * Constant expression assigned in the initializer of the variable
577 *
578 * \warning
579 * This field and \c ::constant_value are distinct. Even if the two fields
580 * refer to constants with the same value, they must point to separate
581 * objects.
582 */
583 ir_constant *constant_initializer;
584
585 /**
586 * For variables that are in an interface block or are an instance of an
587 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
588 *
589 * \sa ir_variable::location
590 */
591 const glsl_type *interface_type;
592 };
593
594 /**
595 * A function that returns whether a built-in function is available in the
596 * current shading language (based on version, ES or desktop, and extensions).
597 */
598 typedef bool (*builtin_available_predicate)(const _mesa_glsl_parse_state *);
599
600 /*@{*/
601 /**
602 * The representation of a function instance; may be the full definition or
603 * simply a prototype.
604 */
605 class ir_function_signature : public ir_instruction {
606 /* An ir_function_signature will be part of the list of signatures in
607 * an ir_function.
608 */
609 public:
610 ir_function_signature(const glsl_type *return_type,
611 builtin_available_predicate builtin_info = NULL);
612
613 virtual ir_function_signature *clone(void *mem_ctx,
614 struct hash_table *ht) const;
615 ir_function_signature *clone_prototype(void *mem_ctx,
616 struct hash_table *ht) const;
617
618 virtual void accept(ir_visitor *v)
619 {
620 v->visit(this);
621 }
622
623 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
624
625 /**
626 * Attempt to evaluate this function as a constant expression,
627 * given a list of the actual parameters and the variable context.
628 * Returns NULL for non-built-ins.
629 */
630 ir_constant *constant_expression_value(exec_list *actual_parameters, struct hash_table *variable_context);
631
632 /**
633 * Get the name of the function for which this is a signature
634 */
635 const char *function_name() const;
636
637 /**
638 * Get a handle to the function for which this is a signature
639 *
640 * There is no setter function, this function returns a \c const pointer,
641 * and \c ir_function_signature::_function is private for a reason. The
642 * only way to make a connection between a function and function signature
643 * is via \c ir_function::add_signature. This helps ensure that certain
644 * invariants (i.e., a function signature is in the list of signatures for
645 * its \c _function) are met.
646 *
647 * \sa ir_function::add_signature
648 */
649 inline const class ir_function *function() const
650 {
651 return this->_function;
652 }
653
654 /**
655 * Check whether the qualifiers match between this signature's parameters
656 * and the supplied parameter list. If not, returns the name of the first
657 * parameter with mismatched qualifiers (for use in error messages).
658 */
659 const char *qualifiers_match(exec_list *params);
660
661 /**
662 * Replace the current parameter list with the given one. This is useful
663 * if the current information came from a prototype, and either has invalid
664 * or missing parameter names.
665 */
666 void replace_parameters(exec_list *new_params);
667
668 /**
669 * Function return type.
670 *
671 * \note This discards the optional precision qualifier.
672 */
673 const struct glsl_type *return_type;
674
675 /**
676 * List of ir_variable of function parameters.
677 *
678 * This represents the storage. The paramaters passed in a particular
679 * call will be in ir_call::actual_paramaters.
680 */
681 struct exec_list parameters;
682
683 /** Whether or not this function has a body (which may be empty). */
684 unsigned is_defined:1;
685
686 /** Whether or not this function signature is a built-in. */
687 bool is_builtin() const;
688
689 /** Body of instructions in the function. */
690 struct exec_list body;
691
692 private:
693 /**
694 * A function pointer to a predicate that answers whether a built-in
695 * function is available in the current shader. NULL if not a built-in.
696 */
697 builtin_available_predicate builtin_info;
698
699 /** Function of which this signature is one overload. */
700 class ir_function *_function;
701
702 /** Function signature of which this one is a prototype clone */
703 const ir_function_signature *origin;
704
705 friend class ir_function;
706
707 /**
708 * Helper function to run a list of instructions for constant
709 * expression evaluation.
710 *
711 * The hash table represents the values of the visible variables.
712 * There are no scoping issues because the table is indexed on
713 * ir_variable pointers, not variable names.
714 *
715 * Returns false if the expression is not constant, true otherwise,
716 * and the value in *result if result is non-NULL.
717 */
718 bool constant_expression_evaluate_expression_list(const struct exec_list &body,
719 struct hash_table *variable_context,
720 ir_constant **result);
721 };
722
723
724 /**
725 * Header for tracking multiple overloaded functions with the same name.
726 * Contains a list of ir_function_signatures representing each of the
727 * actual functions.
728 */
729 class ir_function : public ir_instruction {
730 public:
731 ir_function(const char *name);
732
733 virtual ir_function *clone(void *mem_ctx, struct hash_table *ht) const;
734
735 virtual ir_function *as_function()
736 {
737 return this;
738 }
739
740 virtual void accept(ir_visitor *v)
741 {
742 v->visit(this);
743 }
744
745 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
746
747 void add_signature(ir_function_signature *sig)
748 {
749 sig->_function = this;
750 this->signatures.push_tail(sig);
751 }
752
753 /**
754 * Get an iterator for the set of function signatures
755 */
756 exec_list_iterator iterator()
757 {
758 return signatures.iterator();
759 }
760
761 /**
762 * Find a signature that matches a set of actual parameters, taking implicit
763 * conversions into account. Also flags whether the match was exact.
764 */
765 ir_function_signature *matching_signature(const exec_list *actual_param,
766 bool *match_is_exact);
767
768 /**
769 * Find a signature that matches a set of actual parameters, taking implicit
770 * conversions into account.
771 */
772 ir_function_signature *matching_signature(const exec_list *actual_param);
773
774 /**
775 * Find a signature that exactly matches a set of actual parameters without
776 * any implicit type conversions.
777 */
778 ir_function_signature *exact_matching_signature(const exec_list *actual_ps);
779
780 /**
781 * Name of the function.
782 */
783 const char *name;
784
785 /** Whether or not this function has a signature that isn't a built-in. */
786 bool has_user_signature();
787
788 /**
789 * List of ir_function_signature for each overloaded function with this name.
790 */
791 struct exec_list signatures;
792 };
793
794 inline const char *ir_function_signature::function_name() const
795 {
796 return this->_function->name;
797 }
798 /*@}*/
799
800
801 /**
802 * IR instruction representing high-level if-statements
803 */
804 class ir_if : public ir_instruction {
805 public:
806 ir_if(ir_rvalue *condition)
807 : condition(condition)
808 {
809 ir_type = ir_type_if;
810 }
811
812 virtual ir_if *clone(void *mem_ctx, struct hash_table *ht) const;
813
814 virtual ir_if *as_if()
815 {
816 return this;
817 }
818
819 virtual void accept(ir_visitor *v)
820 {
821 v->visit(this);
822 }
823
824 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
825
826 ir_rvalue *condition;
827 /** List of ir_instruction for the body of the then branch */
828 exec_list then_instructions;
829 /** List of ir_instruction for the body of the else branch */
830 exec_list else_instructions;
831 };
832
833
834 /**
835 * IR instruction representing a high-level loop structure.
836 */
837 class ir_loop : public ir_instruction {
838 public:
839 ir_loop();
840
841 virtual ir_loop *clone(void *mem_ctx, struct hash_table *ht) const;
842
843 virtual void accept(ir_visitor *v)
844 {
845 v->visit(this);
846 }
847
848 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
849
850 virtual ir_loop *as_loop()
851 {
852 return this;
853 }
854
855 /**
856 * Get an iterator for the instructions of the loop body
857 */
858 exec_list_iterator iterator()
859 {
860 return body_instructions.iterator();
861 }
862
863 /** List of ir_instruction that make up the body of the loop. */
864 exec_list body_instructions;
865
866 /**
867 * \name Loop counter and controls
868 *
869 * Represents a loop like a FORTRAN \c do-loop.
870 *
871 * \note
872 * If \c from and \c to are the same value, the loop will execute once.
873 */
874 /*@{*/
875 ir_rvalue *from; /** Value of the loop counter on the first
876 * iteration of the loop.
877 */
878 ir_rvalue *to; /** Value of the loop counter on the last
879 * iteration of the loop.
880 */
881 ir_rvalue *increment;
882 ir_variable *counter;
883
884 /**
885 * Comparison operation in the loop terminator.
886 *
887 * If any of the loop control fields are non-\c NULL, this field must be
888 * one of \c ir_binop_less, \c ir_binop_greater, \c ir_binop_lequal,
889 * \c ir_binop_gequal, \c ir_binop_equal, or \c ir_binop_nequal.
890 */
891 int cmp;
892 /*@}*/
893 };
894
895
896 class ir_assignment : public ir_instruction {
897 public:
898 ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs, ir_rvalue *condition = NULL);
899
900 /**
901 * Construct an assignment with an explicit write mask
902 *
903 * \note
904 * Since a write mask is supplied, the LHS must already be a bare
905 * \c ir_dereference. The cannot be any swizzles in the LHS.
906 */
907 ir_assignment(ir_dereference *lhs, ir_rvalue *rhs, ir_rvalue *condition,
908 unsigned write_mask);
909
910 virtual ir_assignment *clone(void *mem_ctx, struct hash_table *ht) const;
911
912 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
913
914 virtual void accept(ir_visitor *v)
915 {
916 v->visit(this);
917 }
918
919 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
920
921 virtual ir_assignment * as_assignment()
922 {
923 return this;
924 }
925
926 /**
927 * Get a whole variable written by an assignment
928 *
929 * If the LHS of the assignment writes a whole variable, the variable is
930 * returned. Otherwise \c NULL is returned. Examples of whole-variable
931 * assignment are:
932 *
933 * - Assigning to a scalar
934 * - Assigning to all components of a vector
935 * - Whole array (or matrix) assignment
936 * - Whole structure assignment
937 */
938 ir_variable *whole_variable_written();
939
940 /**
941 * Set the LHS of an assignment
942 */
943 void set_lhs(ir_rvalue *lhs);
944
945 /**
946 * Left-hand side of the assignment.
947 *
948 * This should be treated as read only. If you need to set the LHS of an
949 * assignment, use \c ir_assignment::set_lhs.
950 */
951 ir_dereference *lhs;
952
953 /**
954 * Value being assigned
955 */
956 ir_rvalue *rhs;
957
958 /**
959 * Optional condition for the assignment.
960 */
961 ir_rvalue *condition;
962
963
964 /**
965 * Component mask written
966 *
967 * For non-vector types in the LHS, this field will be zero. For vector
968 * types, a bit will be set for each component that is written. Note that
969 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
970 *
971 * A partially-set write mask means that each enabled channel gets
972 * the value from a consecutive channel of the rhs. For example,
973 * to write just .xyw of gl_FrontColor with color:
974 *
975 * (assign (constant bool (1)) (xyw)
976 * (var_ref gl_FragColor)
977 * (swiz xyw (var_ref color)))
978 */
979 unsigned write_mask:4;
980 };
981
982 /* Update ir_expression::get_num_operands() and operator_strs when
983 * updating this list.
984 */
985 enum ir_expression_operation {
986 ir_unop_bit_not,
987 ir_unop_logic_not,
988 ir_unop_neg,
989 ir_unop_abs,
990 ir_unop_sign,
991 ir_unop_rcp,
992 ir_unop_rsq,
993 ir_unop_sqrt,
994 ir_unop_exp, /**< Log base e on gentype */
995 ir_unop_log, /**< Natural log on gentype */
996 ir_unop_exp2,
997 ir_unop_log2,
998 ir_unop_f2i, /**< Float-to-integer conversion. */
999 ir_unop_f2u, /**< Float-to-unsigned conversion. */
1000 ir_unop_i2f, /**< Integer-to-float conversion. */
1001 ir_unop_f2b, /**< Float-to-boolean conversion */
1002 ir_unop_b2f, /**< Boolean-to-float conversion */
1003 ir_unop_i2b, /**< int-to-boolean conversion */
1004 ir_unop_b2i, /**< Boolean-to-int conversion */
1005 ir_unop_u2f, /**< Unsigned-to-float conversion. */
1006 ir_unop_i2u, /**< Integer-to-unsigned conversion. */
1007 ir_unop_u2i, /**< Unsigned-to-integer conversion. */
1008 ir_unop_bitcast_i2f, /**< Bit-identical int-to-float "conversion" */
1009 ir_unop_bitcast_f2i, /**< Bit-identical float-to-int "conversion" */
1010 ir_unop_bitcast_u2f, /**< Bit-identical uint-to-float "conversion" */
1011 ir_unop_bitcast_f2u, /**< Bit-identical float-to-uint "conversion" */
1012 ir_unop_any,
1013
1014 /**
1015 * \name Unary floating-point rounding operations.
1016 */
1017 /*@{*/
1018 ir_unop_trunc,
1019 ir_unop_ceil,
1020 ir_unop_floor,
1021 ir_unop_fract,
1022 ir_unop_round_even,
1023 /*@}*/
1024
1025 /**
1026 * \name Trigonometric operations.
1027 */
1028 /*@{*/
1029 ir_unop_sin,
1030 ir_unop_cos,
1031 ir_unop_sin_reduced, /**< Reduced range sin. [-pi, pi] */
1032 ir_unop_cos_reduced, /**< Reduced range cos. [-pi, pi] */
1033 /*@}*/
1034
1035 /**
1036 * \name Partial derivatives.
1037 */
1038 /*@{*/
1039 ir_unop_dFdx,
1040 ir_unop_dFdy,
1041 /*@}*/
1042
1043 /**
1044 * \name Floating point pack and unpack operations.
1045 */
1046 /*@{*/
1047 ir_unop_pack_snorm_2x16,
1048 ir_unop_pack_snorm_4x8,
1049 ir_unop_pack_unorm_2x16,
1050 ir_unop_pack_unorm_4x8,
1051 ir_unop_pack_half_2x16,
1052 ir_unop_unpack_snorm_2x16,
1053 ir_unop_unpack_snorm_4x8,
1054 ir_unop_unpack_unorm_2x16,
1055 ir_unop_unpack_unorm_4x8,
1056 ir_unop_unpack_half_2x16,
1057 /*@}*/
1058
1059 /**
1060 * \name Lowered floating point unpacking operations.
1061 *
1062 * \see lower_packing_builtins_visitor::split_unpack_half_2x16
1063 */
1064 /*@{*/
1065 ir_unop_unpack_half_2x16_split_x,
1066 ir_unop_unpack_half_2x16_split_y,
1067 /*@}*/
1068
1069 /**
1070 * \name Bit operations, part of ARB_gpu_shader5.
1071 */
1072 /*@{*/
1073 ir_unop_bitfield_reverse,
1074 ir_unop_bit_count,
1075 ir_unop_find_msb,
1076 ir_unop_find_lsb,
1077 /*@}*/
1078
1079 ir_unop_noise,
1080
1081 /**
1082 * A sentinel marking the last of the unary operations.
1083 */
1084 ir_last_unop = ir_unop_noise,
1085
1086 ir_binop_add,
1087 ir_binop_sub,
1088 ir_binop_mul,
1089 ir_binop_div,
1090
1091 /**
1092 * Takes one of two combinations of arguments:
1093 *
1094 * - mod(vecN, vecN)
1095 * - mod(vecN, float)
1096 *
1097 * Does not take integer types.
1098 */
1099 ir_binop_mod,
1100
1101 /**
1102 * \name Binary comparison operators which return a boolean vector.
1103 * The type of both operands must be equal.
1104 */
1105 /*@{*/
1106 ir_binop_less,
1107 ir_binop_greater,
1108 ir_binop_lequal,
1109 ir_binop_gequal,
1110 ir_binop_equal,
1111 ir_binop_nequal,
1112 /**
1113 * Returns single boolean for whether all components of operands[0]
1114 * equal the components of operands[1].
1115 */
1116 ir_binop_all_equal,
1117 /**
1118 * Returns single boolean for whether any component of operands[0]
1119 * is not equal to the corresponding component of operands[1].
1120 */
1121 ir_binop_any_nequal,
1122 /*@}*/
1123
1124 /**
1125 * \name Bit-wise binary operations.
1126 */
1127 /*@{*/
1128 ir_binop_lshift,
1129 ir_binop_rshift,
1130 ir_binop_bit_and,
1131 ir_binop_bit_xor,
1132 ir_binop_bit_or,
1133 /*@}*/
1134
1135 ir_binop_logic_and,
1136 ir_binop_logic_xor,
1137 ir_binop_logic_or,
1138
1139 ir_binop_dot,
1140 ir_binop_min,
1141 ir_binop_max,
1142
1143 ir_binop_pow,
1144
1145 /**
1146 * \name Lowered floating point packing operations.
1147 *
1148 * \see lower_packing_builtins_visitor::split_pack_half_2x16
1149 */
1150 /*@{*/
1151 ir_binop_pack_half_2x16_split,
1152 /*@}*/
1153
1154 /**
1155 * \name First half of a lowered bitfieldInsert() operation.
1156 *
1157 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1158 */
1159 /*@{*/
1160 ir_binop_bfm,
1161 /*@}*/
1162
1163 /**
1164 * Load a value the size of a given GLSL type from a uniform block.
1165 *
1166 * operand0 is the ir_constant uniform block index in the linked shader.
1167 * operand1 is a byte offset within the uniform block.
1168 */
1169 ir_binop_ubo_load,
1170
1171 /**
1172 * Extract a scalar from a vector
1173 *
1174 * operand0 is the vector
1175 * operand1 is the index of the field to read from operand0
1176 */
1177 ir_binop_vector_extract,
1178
1179 /**
1180 * A sentinel marking the last of the binary operations.
1181 */
1182 ir_last_binop = ir_binop_vector_extract,
1183
1184 /**
1185 * \name Fused floating-point multiply-add, part of ARB_gpu_shader5.
1186 */
1187 /*@{*/
1188 ir_triop_fma,
1189 /*@}*/
1190
1191 ir_triop_lrp,
1192
1193 /**
1194 * \name Second half of a lowered bitfieldInsert() operation.
1195 *
1196 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1197 */
1198 /*@{*/
1199 ir_triop_bfi,
1200 /*@}*/
1201
1202 ir_triop_bitfield_extract,
1203
1204 /**
1205 * Generate a value with one field of a vector changed
1206 *
1207 * operand0 is the vector
1208 * operand1 is the value to write into the vector result
1209 * operand2 is the index in operand0 to be modified
1210 */
1211 ir_triop_vector_insert,
1212
1213 /**
1214 * A sentinel marking the last of the ternary operations.
1215 */
1216 ir_last_triop = ir_triop_vector_insert,
1217
1218 ir_quadop_bitfield_insert,
1219
1220 ir_quadop_vector,
1221
1222 /**
1223 * A sentinel marking the last of the ternary operations.
1224 */
1225 ir_last_quadop = ir_quadop_vector,
1226
1227 /**
1228 * A sentinel marking the last of all operations.
1229 */
1230 ir_last_opcode = ir_quadop_vector
1231 };
1232
1233 class ir_expression : public ir_rvalue {
1234 public:
1235 ir_expression(int op, const struct glsl_type *type,
1236 ir_rvalue *op0, ir_rvalue *op1 = NULL,
1237 ir_rvalue *op2 = NULL, ir_rvalue *op3 = NULL);
1238
1239 /**
1240 * Constructor for unary operation expressions
1241 */
1242 ir_expression(int op, ir_rvalue *);
1243
1244 /**
1245 * Constructor for binary operation expressions
1246 */
1247 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1);
1248
1249 virtual ir_expression *as_expression()
1250 {
1251 return this;
1252 }
1253
1254 virtual ir_expression *clone(void *mem_ctx, struct hash_table *ht) const;
1255
1256 /**
1257 * Attempt to constant-fold the expression
1258 *
1259 * The "variable_context" hash table links ir_variable * to ir_constant *
1260 * that represent the variables' values. \c NULL represents an empty
1261 * context.
1262 *
1263 * If the expression cannot be constant folded, this method will return
1264 * \c NULL.
1265 */
1266 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1267
1268 /**
1269 * Determine the number of operands used by an expression
1270 */
1271 static unsigned int get_num_operands(ir_expression_operation);
1272
1273 /**
1274 * Determine the number of operands used by an expression
1275 */
1276 unsigned int get_num_operands() const
1277 {
1278 return (this->operation == ir_quadop_vector)
1279 ? this->type->vector_elements : get_num_operands(operation);
1280 }
1281
1282 /**
1283 * Return a string representing this expression's operator.
1284 */
1285 const char *operator_string();
1286
1287 /**
1288 * Return a string representing this expression's operator.
1289 */
1290 static const char *operator_string(ir_expression_operation);
1291
1292
1293 /**
1294 * Do a reverse-lookup to translate the given string into an operator.
1295 */
1296 static ir_expression_operation get_operator(const char *);
1297
1298 virtual void accept(ir_visitor *v)
1299 {
1300 v->visit(this);
1301 }
1302
1303 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1304
1305 ir_expression_operation operation;
1306 ir_rvalue *operands[4];
1307 };
1308
1309
1310 /**
1311 * HIR instruction representing a high-level function call, containing a list
1312 * of parameters and returning a value in the supplied temporary.
1313 */
1314 class ir_call : public ir_instruction {
1315 public:
1316 ir_call(ir_function_signature *callee,
1317 ir_dereference_variable *return_deref,
1318 exec_list *actual_parameters)
1319 : return_deref(return_deref), callee(callee)
1320 {
1321 ir_type = ir_type_call;
1322 assert(callee->return_type != NULL);
1323 actual_parameters->move_nodes_to(& this->actual_parameters);
1324 this->use_builtin = callee->is_builtin();
1325 }
1326
1327 virtual ir_call *clone(void *mem_ctx, struct hash_table *ht) const;
1328
1329 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1330
1331 virtual ir_call *as_call()
1332 {
1333 return this;
1334 }
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 /**
1344 * Get an iterator for the set of acutal parameters
1345 */
1346 exec_list_iterator iterator()
1347 {
1348 return actual_parameters.iterator();
1349 }
1350
1351 /**
1352 * Get the name of the function being called.
1353 */
1354 const char *callee_name() const
1355 {
1356 return callee->function_name();
1357 }
1358
1359 /**
1360 * Generates an inline version of the function before @ir,
1361 * storing the return value in return_deref.
1362 */
1363 void generate_inline(ir_instruction *ir);
1364
1365 /**
1366 * Storage for the function's return value.
1367 * This must be NULL if the return type is void.
1368 */
1369 ir_dereference_variable *return_deref;
1370
1371 /**
1372 * The specific function signature being called.
1373 */
1374 ir_function_signature *callee;
1375
1376 /* List of ir_rvalue of paramaters passed in this call. */
1377 exec_list actual_parameters;
1378
1379 /** Should this call only bind to a built-in function? */
1380 bool use_builtin;
1381 };
1382
1383
1384 /**
1385 * \name Jump-like IR instructions.
1386 *
1387 * These include \c break, \c continue, \c return, and \c discard.
1388 */
1389 /*@{*/
1390 class ir_jump : public ir_instruction {
1391 protected:
1392 ir_jump()
1393 {
1394 ir_type = ir_type_unset;
1395 }
1396
1397 public:
1398 virtual ir_jump *as_jump()
1399 {
1400 return this;
1401 }
1402 };
1403
1404 class ir_return : public ir_jump {
1405 public:
1406 ir_return()
1407 : value(NULL)
1408 {
1409 this->ir_type = ir_type_return;
1410 }
1411
1412 ir_return(ir_rvalue *value)
1413 : value(value)
1414 {
1415 this->ir_type = ir_type_return;
1416 }
1417
1418 virtual ir_return *clone(void *mem_ctx, struct hash_table *) const;
1419
1420 virtual ir_return *as_return()
1421 {
1422 return this;
1423 }
1424
1425 ir_rvalue *get_value() const
1426 {
1427 return value;
1428 }
1429
1430 virtual void accept(ir_visitor *v)
1431 {
1432 v->visit(this);
1433 }
1434
1435 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1436
1437 ir_rvalue *value;
1438 };
1439
1440
1441 /**
1442 * Jump instructions used inside loops
1443 *
1444 * These include \c break and \c continue. The \c break within a loop is
1445 * different from the \c break within a switch-statement.
1446 *
1447 * \sa ir_switch_jump
1448 */
1449 class ir_loop_jump : public ir_jump {
1450 public:
1451 enum jump_mode {
1452 jump_break,
1453 jump_continue
1454 };
1455
1456 ir_loop_jump(jump_mode mode)
1457 {
1458 this->ir_type = ir_type_loop_jump;
1459 this->mode = mode;
1460 }
1461
1462 virtual ir_loop_jump *clone(void *mem_ctx, struct hash_table *) const;
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 bool is_break() const
1472 {
1473 return mode == jump_break;
1474 }
1475
1476 bool is_continue() const
1477 {
1478 return mode == jump_continue;
1479 }
1480
1481 /** Mode selector for the jump instruction. */
1482 enum jump_mode mode;
1483 };
1484
1485 /**
1486 * IR instruction representing discard statements.
1487 */
1488 class ir_discard : public ir_jump {
1489 public:
1490 ir_discard()
1491 {
1492 this->ir_type = ir_type_discard;
1493 this->condition = NULL;
1494 }
1495
1496 ir_discard(ir_rvalue *cond)
1497 {
1498 this->ir_type = ir_type_discard;
1499 this->condition = cond;
1500 }
1501
1502 virtual ir_discard *clone(void *mem_ctx, struct hash_table *ht) const;
1503
1504 virtual void accept(ir_visitor *v)
1505 {
1506 v->visit(this);
1507 }
1508
1509 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1510
1511 virtual ir_discard *as_discard()
1512 {
1513 return this;
1514 }
1515
1516 ir_rvalue *condition;
1517 };
1518 /*@}*/
1519
1520
1521 /**
1522 * Texture sampling opcodes used in ir_texture
1523 */
1524 enum ir_texture_opcode {
1525 ir_tex, /**< Regular texture look-up */
1526 ir_txb, /**< Texture look-up with LOD bias */
1527 ir_txl, /**< Texture look-up with explicit LOD */
1528 ir_txd, /**< Texture look-up with partial derivatvies */
1529 ir_txf, /**< Texel fetch with explicit LOD */
1530 ir_txf_ms, /**< Multisample texture fetch */
1531 ir_txs, /**< Texture size */
1532 ir_lod /**< Texture lod query */
1533 };
1534
1535
1536 /**
1537 * IR instruction to sample a texture
1538 *
1539 * The specific form of the IR instruction depends on the \c mode value
1540 * selected from \c ir_texture_opcodes. In the printed IR, these will
1541 * appear as:
1542 *
1543 * Texel offset (0 or an expression)
1544 * | Projection divisor
1545 * | | Shadow comparitor
1546 * | | |
1547 * v v v
1548 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1549 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1550 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1551 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1552 * (txf <type> <sampler> <coordinate> 0 <lod>)
1553 * (txf_ms
1554 * <type> <sampler> <coordinate> <sample_index>)
1555 * (txs <type> <sampler> <lod>)
1556 * (lod <type> <sampler> <coordinate>)
1557 */
1558 class ir_texture : public ir_rvalue {
1559 public:
1560 ir_texture(enum ir_texture_opcode op)
1561 : op(op), sampler(NULL), coordinate(NULL), projector(NULL),
1562 shadow_comparitor(NULL), offset(NULL)
1563 {
1564 this->ir_type = ir_type_texture;
1565 }
1566
1567 virtual ir_texture *clone(void *mem_ctx, struct hash_table *) const;
1568
1569 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1570
1571 virtual void accept(ir_visitor *v)
1572 {
1573 v->visit(this);
1574 }
1575
1576 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1577
1578 /**
1579 * Return a string representing the ir_texture_opcode.
1580 */
1581 const char *opcode_string();
1582
1583 /** Set the sampler and type. */
1584 void set_sampler(ir_dereference *sampler, const glsl_type *type);
1585
1586 /**
1587 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1588 */
1589 static ir_texture_opcode get_opcode(const char *);
1590
1591 enum ir_texture_opcode op;
1592
1593 /** Sampler to use for the texture access. */
1594 ir_dereference *sampler;
1595
1596 /** Texture coordinate to sample */
1597 ir_rvalue *coordinate;
1598
1599 /**
1600 * Value used for projective divide.
1601 *
1602 * If there is no projective divide (the common case), this will be
1603 * \c NULL. Optimization passes should check for this to point to a constant
1604 * of 1.0 and replace that with \c NULL.
1605 */
1606 ir_rvalue *projector;
1607
1608 /**
1609 * Coordinate used for comparison on shadow look-ups.
1610 *
1611 * If there is no shadow comparison, this will be \c NULL. For the
1612 * \c ir_txf opcode, this *must* be \c NULL.
1613 */
1614 ir_rvalue *shadow_comparitor;
1615
1616 /** Texel offset. */
1617 ir_rvalue *offset;
1618
1619 union {
1620 ir_rvalue *lod; /**< Floating point LOD */
1621 ir_rvalue *bias; /**< Floating point LOD bias */
1622 ir_rvalue *sample_index; /**< MSAA sample index */
1623 struct {
1624 ir_rvalue *dPdx; /**< Partial derivative of coordinate wrt X */
1625 ir_rvalue *dPdy; /**< Partial derivative of coordinate wrt Y */
1626 } grad;
1627 } lod_info;
1628 };
1629
1630
1631 struct ir_swizzle_mask {
1632 unsigned x:2;
1633 unsigned y:2;
1634 unsigned z:2;
1635 unsigned w:2;
1636
1637 /**
1638 * Number of components in the swizzle.
1639 */
1640 unsigned num_components:3;
1641
1642 /**
1643 * Does the swizzle contain duplicate components?
1644 *
1645 * L-value swizzles cannot contain duplicate components.
1646 */
1647 unsigned has_duplicates:1;
1648 };
1649
1650
1651 class ir_swizzle : public ir_rvalue {
1652 public:
1653 ir_swizzle(ir_rvalue *, unsigned x, unsigned y, unsigned z, unsigned w,
1654 unsigned count);
1655
1656 ir_swizzle(ir_rvalue *val, const unsigned *components, unsigned count);
1657
1658 ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask);
1659
1660 virtual ir_swizzle *clone(void *mem_ctx, struct hash_table *) const;
1661
1662 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1663
1664 virtual ir_swizzle *as_swizzle()
1665 {
1666 return this;
1667 }
1668
1669 /**
1670 * Construct an ir_swizzle from the textual representation. Can fail.
1671 */
1672 static ir_swizzle *create(ir_rvalue *, const char *, unsigned vector_length);
1673
1674 virtual void accept(ir_visitor *v)
1675 {
1676 v->visit(this);
1677 }
1678
1679 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1680
1681 bool is_lvalue() const
1682 {
1683 return val->is_lvalue() && !mask.has_duplicates;
1684 }
1685
1686 /**
1687 * Get the variable that is ultimately referenced by an r-value
1688 */
1689 virtual ir_variable *variable_referenced() const;
1690
1691 ir_rvalue *val;
1692 ir_swizzle_mask mask;
1693
1694 private:
1695 /**
1696 * Initialize the mask component of a swizzle
1697 *
1698 * This is used by the \c ir_swizzle constructors.
1699 */
1700 void init_mask(const unsigned *components, unsigned count);
1701 };
1702
1703
1704 class ir_dereference : public ir_rvalue {
1705 public:
1706 virtual ir_dereference *clone(void *mem_ctx, struct hash_table *) const = 0;
1707
1708 virtual ir_dereference *as_dereference()
1709 {
1710 return this;
1711 }
1712
1713 bool is_lvalue() const;
1714
1715 /**
1716 * Get the variable that is ultimately referenced by an r-value
1717 */
1718 virtual ir_variable *variable_referenced() const = 0;
1719
1720 /**
1721 * Get the constant that is ultimately referenced by an r-value,
1722 * in a constant expression evaluation context.
1723 *
1724 * The offset is used when the reference is to a specific column of
1725 * a matrix.
1726 */
1727 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const = 0;
1728 };
1729
1730
1731 class ir_dereference_variable : public ir_dereference {
1732 public:
1733 ir_dereference_variable(ir_variable *var);
1734
1735 virtual ir_dereference_variable *clone(void *mem_ctx,
1736 struct hash_table *) const;
1737
1738 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1739
1740 virtual ir_dereference_variable *as_dereference_variable()
1741 {
1742 return this;
1743 }
1744
1745 /**
1746 * Get the variable that is ultimately referenced by an r-value
1747 */
1748 virtual ir_variable *variable_referenced() const
1749 {
1750 return this->var;
1751 }
1752
1753 /**
1754 * Get the constant that is ultimately referenced by an r-value,
1755 * in a constant expression evaluation context.
1756 *
1757 * The offset is used when the reference is to a specific column of
1758 * a matrix.
1759 */
1760 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
1761
1762 virtual ir_variable *whole_variable_referenced()
1763 {
1764 /* ir_dereference_variable objects always dereference the entire
1765 * variable. However, if this dereference is dereferenced by anything
1766 * else, the complete deferefernce chain is not a whole-variable
1767 * dereference. This method should only be called on the top most
1768 * ir_rvalue in a dereference chain.
1769 */
1770 return this->var;
1771 }
1772
1773 virtual void accept(ir_visitor *v)
1774 {
1775 v->visit(this);
1776 }
1777
1778 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1779
1780 /**
1781 * Object being dereferenced.
1782 */
1783 ir_variable *var;
1784 };
1785
1786
1787 class ir_dereference_array : public ir_dereference {
1788 public:
1789 ir_dereference_array(ir_rvalue *value, ir_rvalue *array_index);
1790
1791 ir_dereference_array(ir_variable *var, ir_rvalue *array_index);
1792
1793 virtual ir_dereference_array *clone(void *mem_ctx,
1794 struct hash_table *) const;
1795
1796 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1797
1798 virtual ir_dereference_array *as_dereference_array()
1799 {
1800 return this;
1801 }
1802
1803 /**
1804 * Get the variable that is ultimately referenced by an r-value
1805 */
1806 virtual ir_variable *variable_referenced() const
1807 {
1808 return this->array->variable_referenced();
1809 }
1810
1811 /**
1812 * Get the constant that is ultimately referenced by an r-value,
1813 * in a constant expression evaluation context.
1814 *
1815 * The offset is used when the reference is to a specific column of
1816 * a matrix.
1817 */
1818 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
1819
1820 virtual void accept(ir_visitor *v)
1821 {
1822 v->visit(this);
1823 }
1824
1825 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1826
1827 ir_rvalue *array;
1828 ir_rvalue *array_index;
1829
1830 private:
1831 void set_array(ir_rvalue *value);
1832 };
1833
1834
1835 class ir_dereference_record : public ir_dereference {
1836 public:
1837 ir_dereference_record(ir_rvalue *value, const char *field);
1838
1839 ir_dereference_record(ir_variable *var, const char *field);
1840
1841 virtual ir_dereference_record *clone(void *mem_ctx,
1842 struct hash_table *) const;
1843
1844 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1845
1846 virtual ir_dereference_record *as_dereference_record()
1847 {
1848 return this;
1849 }
1850
1851 /**
1852 * Get the variable that is ultimately referenced by an r-value
1853 */
1854 virtual ir_variable *variable_referenced() const
1855 {
1856 return this->record->variable_referenced();
1857 }
1858
1859 /**
1860 * Get the constant that is ultimately referenced by an r-value,
1861 * in a constant expression evaluation context.
1862 *
1863 * The offset is used when the reference is to a specific column of
1864 * a matrix.
1865 */
1866 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
1867
1868 virtual void accept(ir_visitor *v)
1869 {
1870 v->visit(this);
1871 }
1872
1873 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1874
1875 ir_rvalue *record;
1876 const char *field;
1877 };
1878
1879
1880 /**
1881 * Data stored in an ir_constant
1882 */
1883 union ir_constant_data {
1884 unsigned u[16];
1885 int i[16];
1886 float f[16];
1887 bool b[16];
1888 };
1889
1890
1891 class ir_constant : public ir_rvalue {
1892 public:
1893 ir_constant(const struct glsl_type *type, const ir_constant_data *data);
1894 ir_constant(bool b);
1895 ir_constant(unsigned int u);
1896 ir_constant(int i);
1897 ir_constant(float f);
1898
1899 /**
1900 * Construct an ir_constant from a list of ir_constant values
1901 */
1902 ir_constant(const struct glsl_type *type, exec_list *values);
1903
1904 /**
1905 * Construct an ir_constant from a scalar component of another ir_constant
1906 *
1907 * The new \c ir_constant inherits the type of the component from the
1908 * source constant.
1909 *
1910 * \note
1911 * In the case of a matrix constant, the new constant is a scalar, \b not
1912 * a vector.
1913 */
1914 ir_constant(const ir_constant *c, unsigned i);
1915
1916 /**
1917 * Return a new ir_constant of the specified type containing all zeros.
1918 */
1919 static ir_constant *zero(void *mem_ctx, const glsl_type *type);
1920
1921 virtual ir_constant *clone(void *mem_ctx, struct hash_table *) const;
1922
1923 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1924
1925 virtual ir_constant *as_constant()
1926 {
1927 return this;
1928 }
1929
1930 virtual void accept(ir_visitor *v)
1931 {
1932 v->visit(this);
1933 }
1934
1935 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1936
1937 /**
1938 * Get a particular component of a constant as a specific type
1939 *
1940 * This is useful, for example, to get a value from an integer constant
1941 * as a float or bool. This appears frequently when constructors are
1942 * called with all constant parameters.
1943 */
1944 /*@{*/
1945 bool get_bool_component(unsigned i) const;
1946 float get_float_component(unsigned i) const;
1947 int get_int_component(unsigned i) const;
1948 unsigned get_uint_component(unsigned i) const;
1949 /*@}*/
1950
1951 ir_constant *get_array_element(unsigned i) const;
1952
1953 ir_constant *get_record_field(const char *name);
1954
1955 /**
1956 * Copy the values on another constant at a given offset.
1957 *
1958 * The offset is ignored for array or struct copies, it's only for
1959 * scalars or vectors into vectors or matrices.
1960 *
1961 * With identical types on both sides and zero offset it's clone()
1962 * without creating a new object.
1963 */
1964
1965 void copy_offset(ir_constant *src, int offset);
1966
1967 /**
1968 * Copy the values on another constant at a given offset and
1969 * following an assign-like mask.
1970 *
1971 * The mask is ignored for scalars.
1972 *
1973 * Note that this function only handles what assign can handle,
1974 * i.e. at most a vector as source and a column of a matrix as
1975 * destination.
1976 */
1977
1978 void copy_masked_offset(ir_constant *src, int offset, unsigned int mask);
1979
1980 /**
1981 * Determine whether a constant has the same value as another constant
1982 *
1983 * \sa ir_constant::is_zero, ir_constant::is_one,
1984 * ir_constant::is_negative_one, ir_constant::is_basis
1985 */
1986 bool has_value(const ir_constant *) const;
1987
1988 virtual bool is_zero() const;
1989 virtual bool is_one() const;
1990 virtual bool is_negative_one() const;
1991 virtual bool is_basis() const;
1992
1993 /**
1994 * Value of the constant.
1995 *
1996 * The field used to back the values supplied by the constant is determined
1997 * by the type associated with the \c ir_instruction. Constants may be
1998 * scalars, vectors, or matrices.
1999 */
2000 union ir_constant_data value;
2001
2002 /* Array elements */
2003 ir_constant **array_elements;
2004
2005 /* Structure fields */
2006 exec_list components;
2007
2008 private:
2009 /**
2010 * Parameterless constructor only used by the clone method
2011 */
2012 ir_constant(void);
2013 };
2014
2015 /*@}*/
2016
2017 /**
2018 * IR instruction to emit a vertex in a geometry shader.
2019 */
2020 class ir_emit_vertex : public ir_instruction {
2021 public:
2022 ir_emit_vertex()
2023 {
2024 ir_type = ir_type_emit_vertex;
2025 }
2026
2027 virtual void accept(ir_visitor *v)
2028 {
2029 v->visit(this);
2030 }
2031
2032 virtual ir_emit_vertex *clone(void *mem_ctx, struct hash_table *) const
2033 {
2034 return new(mem_ctx) ir_emit_vertex();
2035 }
2036
2037 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2038 };
2039
2040 /**
2041 * IR instruction to complete the current primitive and start a new one in a
2042 * geometry shader.
2043 */
2044 class ir_end_primitive : public ir_instruction {
2045 public:
2046 ir_end_primitive()
2047 {
2048 ir_type = ir_type_end_primitive;
2049 }
2050
2051 virtual void accept(ir_visitor *v)
2052 {
2053 v->visit(this);
2054 }
2055
2056 virtual ir_end_primitive *clone(void *mem_ctx, struct hash_table *) const
2057 {
2058 return new(mem_ctx) ir_end_primitive();
2059 }
2060
2061 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2062 };
2063
2064 /**
2065 * Apply a visitor to each IR node in a list
2066 */
2067 void
2068 visit_exec_list(exec_list *list, ir_visitor *visitor);
2069
2070 /**
2071 * Validate invariants on each IR node in a list
2072 */
2073 void validate_ir_tree(exec_list *instructions);
2074
2075 struct _mesa_glsl_parse_state;
2076 struct gl_shader_program;
2077
2078 /**
2079 * Detect whether an unlinked shader contains static recursion
2080 *
2081 * If the list of instructions is determined to contain static recursion,
2082 * \c _mesa_glsl_error will be called to emit error messages for each function
2083 * that is in the recursion cycle.
2084 */
2085 void
2086 detect_recursion_unlinked(struct _mesa_glsl_parse_state *state,
2087 exec_list *instructions);
2088
2089 /**
2090 * Detect whether a linked shader contains static recursion
2091 *
2092 * If the list of instructions is determined to contain static recursion,
2093 * \c link_error_printf will be called to emit error messages for each function
2094 * that is in the recursion cycle. In addition,
2095 * \c gl_shader_program::LinkStatus will be set to false.
2096 */
2097 void
2098 detect_recursion_linked(struct gl_shader_program *prog,
2099 exec_list *instructions);
2100
2101 /**
2102 * Make a clone of each IR instruction in a list
2103 *
2104 * \param in List of IR instructions that are to be cloned
2105 * \param out List to hold the cloned instructions
2106 */
2107 void
2108 clone_ir_list(void *mem_ctx, exec_list *out, const exec_list *in);
2109
2110 extern void
2111 _mesa_glsl_initialize_variables(exec_list *instructions,
2112 struct _mesa_glsl_parse_state *state);
2113
2114 extern void
2115 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state *state);
2116
2117 extern void
2118 _mesa_glsl_release_functions(void);
2119
2120 extern void
2121 reparent_ir(exec_list *list, void *mem_ctx);
2122
2123 struct glsl_symbol_table;
2124
2125 extern void
2126 import_prototypes(const exec_list *source, exec_list *dest,
2127 struct glsl_symbol_table *symbols, void *mem_ctx);
2128
2129 extern bool
2130 ir_has_call(ir_instruction *ir);
2131
2132 extern void
2133 do_set_program_inouts(exec_list *instructions, struct gl_program *prog,
2134 GLenum shader_type);
2135
2136 extern char *
2137 prototype_string(const glsl_type *return_type, const char *name,
2138 exec_list *parameters);
2139
2140 extern "C" {
2141 #endif /* __cplusplus */
2142
2143 extern void _mesa_print_ir(struct exec_list *instructions,
2144 struct _mesa_glsl_parse_state *state);
2145
2146 #ifdef __cplusplus
2147 } /* extern "C" */
2148 #endif
2149
2150 unsigned
2151 vertices_per_prim(GLenum prim);
2152
2153 #endif /* IR_H */