glsl: Modify array_sizing_visitor to handle unnamed interface blocks.
[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 * Set this->interface_type on a newly created variable.
396 */
397 void init_interface_type(const struct glsl_type *type)
398 {
399 assert(this->interface_type == NULL);
400 this->interface_type = type;
401 if (this->is_interface_instance()) {
402 this->max_ifc_array_access =
403 rzalloc_array(this, unsigned, type->length);
404 }
405 }
406
407 /**
408 * Change this->interface_type on a variable that previously had a
409 * different, but compatible, interface_type. This is used during linking
410 * to set the size of arrays in interface blocks.
411 */
412 void change_interface_type(const struct glsl_type *type)
413 {
414 if (this->max_ifc_array_access != NULL) {
415 /* max_ifc_array_access has already been allocated, so make sure the
416 * new interface has the same number of fields as the old one.
417 */
418 assert(this->interface_type->length == type->length);
419 }
420 this->interface_type = type;
421 }
422
423 const glsl_type *get_interface_type() const
424 {
425 return this->interface_type;
426 }
427
428 /**
429 * Declared type of the variable
430 */
431 const struct glsl_type *type;
432
433 /**
434 * Declared name of the variable
435 */
436 const char *name;
437
438 /**
439 * Highest element accessed with a constant expression array index
440 *
441 * Not used for non-array variables.
442 */
443 unsigned max_array_access;
444
445 /**
446 * For variables which satisfy the is_interface_instance() predicate, this
447 * points to an array of integers such that if the ith member of the
448 * interface block is an array, max_ifc_array_access[i] is the maximum
449 * array element of that member that has been accessed. If the ith member
450 * of the interface block is not an array, max_ifc_array_access[i] is
451 * unused.
452 *
453 * For variables whose type is not an interface block, this pointer is
454 * NULL.
455 */
456 unsigned *max_ifc_array_access;
457
458 /**
459 * Is the variable read-only?
460 *
461 * This is set for variables declared as \c const, shader inputs,
462 * and uniforms.
463 */
464 unsigned read_only:1;
465 unsigned centroid:1;
466 unsigned invariant:1;
467
468 /**
469 * Has this variable been used for reading or writing?
470 *
471 * Several GLSL semantic checks require knowledge of whether or not a
472 * variable has been used. For example, it is an error to redeclare a
473 * variable as invariant after it has been used.
474 *
475 * This is only maintained in the ast_to_hir.cpp path, not in
476 * Mesa's fixed function or ARB program paths.
477 */
478 unsigned used:1;
479
480 /**
481 * Has this variable been statically assigned?
482 *
483 * This answers whether the variable was assigned in any path of
484 * the shader during ast_to_hir. This doesn't answer whether it is
485 * still written after dead code removal, nor is it maintained in
486 * non-ast_to_hir.cpp (GLSL parsing) paths.
487 */
488 unsigned assigned:1;
489
490 /**
491 * Storage class of the variable.
492 *
493 * \sa ir_variable_mode
494 */
495 unsigned mode:4;
496
497 /**
498 * Interpolation mode for shader inputs / outputs
499 *
500 * \sa ir_variable_interpolation
501 */
502 unsigned interpolation:2;
503
504 /**
505 * \name ARB_fragment_coord_conventions
506 * @{
507 */
508 unsigned origin_upper_left:1;
509 unsigned pixel_center_integer:1;
510 /*@}*/
511
512 /**
513 * Was the location explicitly set in the shader?
514 *
515 * If the location is explicitly set in the shader, it \b cannot be changed
516 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
517 * no effect).
518 */
519 unsigned explicit_location:1;
520 unsigned explicit_index:1;
521
522 /**
523 * Was an initial binding explicitly set in the shader?
524 *
525 * If so, constant_value contains an integer ir_constant representing the
526 * initial binding point.
527 */
528 unsigned explicit_binding:1;
529
530 /**
531 * Does this variable have an initializer?
532 *
533 * This is used by the linker to cross-validiate initializers of global
534 * variables.
535 */
536 unsigned has_initializer:1;
537
538 /**
539 * Is this variable a generic output or input that has not yet been matched
540 * up to a variable in another stage of the pipeline?
541 *
542 * This is used by the linker as scratch storage while assigning locations
543 * to generic inputs and outputs.
544 */
545 unsigned is_unmatched_generic_inout:1;
546
547 /**
548 * If non-zero, then this variable may be packed along with other variables
549 * into a single varying slot, so this offset should be applied when
550 * accessing components. For example, an offset of 1 means that the x
551 * component of this variable is actually stored in component y of the
552 * location specified by \c location.
553 */
554 unsigned location_frac:2;
555
556 /**
557 * \brief Layout qualifier for gl_FragDepth.
558 *
559 * This is not equal to \c ir_depth_layout_none if and only if this
560 * variable is \c gl_FragDepth and a layout qualifier is specified.
561 */
562 ir_depth_layout depth_layout;
563
564 /**
565 * Storage location of the base of this variable
566 *
567 * The precise meaning of this field depends on the nature of the variable.
568 *
569 * - Vertex shader input: one of the values from \c gl_vert_attrib.
570 * - Vertex shader output: one of the values from \c gl_varying_slot.
571 * - Geometry shader input: one of the values from \c gl_varying_slot.
572 * - Geometry shader output: one of the values from \c gl_varying_slot.
573 * - Fragment shader input: one of the values from \c gl_varying_slot.
574 * - Fragment shader output: one of the values from \c gl_frag_result.
575 * - Uniforms: Per-stage uniform slot number for default uniform block.
576 * - Uniforms: Index within the uniform block definition for UBO members.
577 * - Other: This field is not currently used.
578 *
579 * If the variable is a uniform, shader input, or shader output, and the
580 * slot has not been assigned, the value will be -1.
581 */
582 int location;
583
584 /**
585 * output index for dual source blending.
586 */
587 int index;
588
589 /**
590 * Initial binding point for a sampler or UBO.
591 *
592 * For array types, this represents the binding point for the first element.
593 */
594 int binding;
595
596 /**
597 * Built-in state that backs this uniform
598 *
599 * Once set at variable creation, \c state_slots must remain invariant.
600 * This is because, ideally, this array would be shared by all clones of
601 * this variable in the IR tree. In other words, we'd really like for it
602 * to be a fly-weight.
603 *
604 * If the variable is not a uniform, \c num_state_slots will be zero and
605 * \c state_slots will be \c NULL.
606 */
607 /*@{*/
608 unsigned num_state_slots; /**< Number of state slots used */
609 ir_state_slot *state_slots; /**< State descriptors. */
610 /*@}*/
611
612 /**
613 * Emit a warning if this variable is accessed.
614 */
615 const char *warn_extension;
616
617 /**
618 * Value assigned in the initializer of a variable declared "const"
619 */
620 ir_constant *constant_value;
621
622 /**
623 * Constant expression assigned in the initializer of the variable
624 *
625 * \warning
626 * This field and \c ::constant_value are distinct. Even if the two fields
627 * refer to constants with the same value, they must point to separate
628 * objects.
629 */
630 ir_constant *constant_initializer;
631
632 private:
633 /**
634 * For variables that are in an interface block or are an instance of an
635 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
636 *
637 * \sa ir_variable::location
638 */
639 const glsl_type *interface_type;
640 };
641
642 /**
643 * A function that returns whether a built-in function is available in the
644 * current shading language (based on version, ES or desktop, and extensions).
645 */
646 typedef bool (*builtin_available_predicate)(const _mesa_glsl_parse_state *);
647
648 /*@{*/
649 /**
650 * The representation of a function instance; may be the full definition or
651 * simply a prototype.
652 */
653 class ir_function_signature : public ir_instruction {
654 /* An ir_function_signature will be part of the list of signatures in
655 * an ir_function.
656 */
657 public:
658 ir_function_signature(const glsl_type *return_type,
659 builtin_available_predicate builtin_avail = NULL);
660
661 virtual ir_function_signature *clone(void *mem_ctx,
662 struct hash_table *ht) const;
663 ir_function_signature *clone_prototype(void *mem_ctx,
664 struct hash_table *ht) const;
665
666 virtual void accept(ir_visitor *v)
667 {
668 v->visit(this);
669 }
670
671 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
672
673 /**
674 * Attempt to evaluate this function as a constant expression,
675 * given a list of the actual parameters and the variable context.
676 * Returns NULL for non-built-ins.
677 */
678 ir_constant *constant_expression_value(exec_list *actual_parameters, struct hash_table *variable_context);
679
680 /**
681 * Get the name of the function for which this is a signature
682 */
683 const char *function_name() const;
684
685 /**
686 * Get a handle to the function for which this is a signature
687 *
688 * There is no setter function, this function returns a \c const pointer,
689 * and \c ir_function_signature::_function is private for a reason. The
690 * only way to make a connection between a function and function signature
691 * is via \c ir_function::add_signature. This helps ensure that certain
692 * invariants (i.e., a function signature is in the list of signatures for
693 * its \c _function) are met.
694 *
695 * \sa ir_function::add_signature
696 */
697 inline const class ir_function *function() const
698 {
699 return this->_function;
700 }
701
702 /**
703 * Check whether the qualifiers match between this signature's parameters
704 * and the supplied parameter list. If not, returns the name of the first
705 * parameter with mismatched qualifiers (for use in error messages).
706 */
707 const char *qualifiers_match(exec_list *params);
708
709 /**
710 * Replace the current parameter list with the given one. This is useful
711 * if the current information came from a prototype, and either has invalid
712 * or missing parameter names.
713 */
714 void replace_parameters(exec_list *new_params);
715
716 /**
717 * Function return type.
718 *
719 * \note This discards the optional precision qualifier.
720 */
721 const struct glsl_type *return_type;
722
723 /**
724 * List of ir_variable of function parameters.
725 *
726 * This represents the storage. The paramaters passed in a particular
727 * call will be in ir_call::actual_paramaters.
728 */
729 struct exec_list parameters;
730
731 /** Whether or not this function has a body (which may be empty). */
732 unsigned is_defined:1;
733
734 /** Whether or not this function signature is a built-in. */
735 bool is_builtin() const;
736
737 /** Whether or not a built-in is available for this shader. */
738 bool is_builtin_available(const _mesa_glsl_parse_state *state) const;
739
740 /** Body of instructions in the function. */
741 struct exec_list body;
742
743 private:
744 /**
745 * A function pointer to a predicate that answers whether a built-in
746 * function is available in the current shader. NULL if not a built-in.
747 */
748 builtin_available_predicate builtin_avail;
749
750 /** Function of which this signature is one overload. */
751 class ir_function *_function;
752
753 /** Function signature of which this one is a prototype clone */
754 const ir_function_signature *origin;
755
756 friend class ir_function;
757
758 /**
759 * Helper function to run a list of instructions for constant
760 * expression evaluation.
761 *
762 * The hash table represents the values of the visible variables.
763 * There are no scoping issues because the table is indexed on
764 * ir_variable pointers, not variable names.
765 *
766 * Returns false if the expression is not constant, true otherwise,
767 * and the value in *result if result is non-NULL.
768 */
769 bool constant_expression_evaluate_expression_list(const struct exec_list &body,
770 struct hash_table *variable_context,
771 ir_constant **result);
772 };
773
774
775 /**
776 * Header for tracking multiple overloaded functions with the same name.
777 * Contains a list of ir_function_signatures representing each of the
778 * actual functions.
779 */
780 class ir_function : public ir_instruction {
781 public:
782 ir_function(const char *name);
783
784 virtual ir_function *clone(void *mem_ctx, struct hash_table *ht) const;
785
786 virtual ir_function *as_function()
787 {
788 return this;
789 }
790
791 virtual void accept(ir_visitor *v)
792 {
793 v->visit(this);
794 }
795
796 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
797
798 void add_signature(ir_function_signature *sig)
799 {
800 sig->_function = this;
801 this->signatures.push_tail(sig);
802 }
803
804 /**
805 * Get an iterator for the set of function signatures
806 */
807 exec_list_iterator iterator()
808 {
809 return signatures.iterator();
810 }
811
812 /**
813 * Find a signature that matches a set of actual parameters, taking implicit
814 * conversions into account. Also flags whether the match was exact.
815 */
816 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
817 const exec_list *actual_param,
818 bool *match_is_exact);
819
820 /**
821 * Find a signature that matches a set of actual parameters, taking implicit
822 * conversions into account.
823 */
824 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
825 const exec_list *actual_param);
826
827 /**
828 * Find a signature that exactly matches a set of actual parameters without
829 * any implicit type conversions.
830 */
831 ir_function_signature *exact_matching_signature(_mesa_glsl_parse_state *state,
832 const exec_list *actual_ps);
833
834 /**
835 * Name of the function.
836 */
837 const char *name;
838
839 /** Whether or not this function has a signature that isn't a built-in. */
840 bool has_user_signature();
841
842 /**
843 * List of ir_function_signature for each overloaded function with this name.
844 */
845 struct exec_list signatures;
846 };
847
848 inline const char *ir_function_signature::function_name() const
849 {
850 return this->_function->name;
851 }
852 /*@}*/
853
854
855 /**
856 * IR instruction representing high-level if-statements
857 */
858 class ir_if : public ir_instruction {
859 public:
860 ir_if(ir_rvalue *condition)
861 : condition(condition)
862 {
863 ir_type = ir_type_if;
864 }
865
866 virtual ir_if *clone(void *mem_ctx, struct hash_table *ht) const;
867
868 virtual ir_if *as_if()
869 {
870 return this;
871 }
872
873 virtual void accept(ir_visitor *v)
874 {
875 v->visit(this);
876 }
877
878 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
879
880 ir_rvalue *condition;
881 /** List of ir_instruction for the body of the then branch */
882 exec_list then_instructions;
883 /** List of ir_instruction for the body of the else branch */
884 exec_list else_instructions;
885 };
886
887
888 /**
889 * IR instruction representing a high-level loop structure.
890 */
891 class ir_loop : public ir_instruction {
892 public:
893 ir_loop();
894
895 virtual ir_loop *clone(void *mem_ctx, struct hash_table *ht) const;
896
897 virtual void accept(ir_visitor *v)
898 {
899 v->visit(this);
900 }
901
902 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
903
904 virtual ir_loop *as_loop()
905 {
906 return this;
907 }
908
909 /**
910 * Get an iterator for the instructions of the loop body
911 */
912 exec_list_iterator iterator()
913 {
914 return body_instructions.iterator();
915 }
916
917 /** List of ir_instruction that make up the body of the loop. */
918 exec_list body_instructions;
919
920 /**
921 * \name Loop counter and controls
922 *
923 * Represents a loop like a FORTRAN \c do-loop.
924 *
925 * \note
926 * If \c from and \c to are the same value, the loop will execute once.
927 */
928 /*@{*/
929 ir_rvalue *from; /** Value of the loop counter on the first
930 * iteration of the loop.
931 */
932 ir_rvalue *to; /** Value of the loop counter on the last
933 * iteration of the loop.
934 */
935 ir_rvalue *increment;
936 ir_variable *counter;
937
938 /**
939 * Comparison operation in the loop terminator.
940 *
941 * If any of the loop control fields are non-\c NULL, this field must be
942 * one of \c ir_binop_less, \c ir_binop_greater, \c ir_binop_lequal,
943 * \c ir_binop_gequal, \c ir_binop_equal, or \c ir_binop_nequal.
944 */
945 int cmp;
946 /*@}*/
947 };
948
949
950 class ir_assignment : public ir_instruction {
951 public:
952 ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs, ir_rvalue *condition = NULL);
953
954 /**
955 * Construct an assignment with an explicit write mask
956 *
957 * \note
958 * Since a write mask is supplied, the LHS must already be a bare
959 * \c ir_dereference. The cannot be any swizzles in the LHS.
960 */
961 ir_assignment(ir_dereference *lhs, ir_rvalue *rhs, ir_rvalue *condition,
962 unsigned write_mask);
963
964 virtual ir_assignment *clone(void *mem_ctx, struct hash_table *ht) const;
965
966 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
967
968 virtual void accept(ir_visitor *v)
969 {
970 v->visit(this);
971 }
972
973 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
974
975 virtual ir_assignment * as_assignment()
976 {
977 return this;
978 }
979
980 /**
981 * Get a whole variable written by an assignment
982 *
983 * If the LHS of the assignment writes a whole variable, the variable is
984 * returned. Otherwise \c NULL is returned. Examples of whole-variable
985 * assignment are:
986 *
987 * - Assigning to a scalar
988 * - Assigning to all components of a vector
989 * - Whole array (or matrix) assignment
990 * - Whole structure assignment
991 */
992 ir_variable *whole_variable_written();
993
994 /**
995 * Set the LHS of an assignment
996 */
997 void set_lhs(ir_rvalue *lhs);
998
999 /**
1000 * Left-hand side of the assignment.
1001 *
1002 * This should be treated as read only. If you need to set the LHS of an
1003 * assignment, use \c ir_assignment::set_lhs.
1004 */
1005 ir_dereference *lhs;
1006
1007 /**
1008 * Value being assigned
1009 */
1010 ir_rvalue *rhs;
1011
1012 /**
1013 * Optional condition for the assignment.
1014 */
1015 ir_rvalue *condition;
1016
1017
1018 /**
1019 * Component mask written
1020 *
1021 * For non-vector types in the LHS, this field will be zero. For vector
1022 * types, a bit will be set for each component that is written. Note that
1023 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1024 *
1025 * A partially-set write mask means that each enabled channel gets
1026 * the value from a consecutive channel of the rhs. For example,
1027 * to write just .xyw of gl_FrontColor with color:
1028 *
1029 * (assign (constant bool (1)) (xyw)
1030 * (var_ref gl_FragColor)
1031 * (swiz xyw (var_ref color)))
1032 */
1033 unsigned write_mask:4;
1034 };
1035
1036 /* Update ir_expression::get_num_operands() and operator_strs when
1037 * updating this list.
1038 */
1039 enum ir_expression_operation {
1040 ir_unop_bit_not,
1041 ir_unop_logic_not,
1042 ir_unop_neg,
1043 ir_unop_abs,
1044 ir_unop_sign,
1045 ir_unop_rcp,
1046 ir_unop_rsq,
1047 ir_unop_sqrt,
1048 ir_unop_exp, /**< Log base e on gentype */
1049 ir_unop_log, /**< Natural log on gentype */
1050 ir_unop_exp2,
1051 ir_unop_log2,
1052 ir_unop_f2i, /**< Float-to-integer conversion. */
1053 ir_unop_f2u, /**< Float-to-unsigned conversion. */
1054 ir_unop_i2f, /**< Integer-to-float conversion. */
1055 ir_unop_f2b, /**< Float-to-boolean conversion */
1056 ir_unop_b2f, /**< Boolean-to-float conversion */
1057 ir_unop_i2b, /**< int-to-boolean conversion */
1058 ir_unop_b2i, /**< Boolean-to-int conversion */
1059 ir_unop_u2f, /**< Unsigned-to-float conversion. */
1060 ir_unop_i2u, /**< Integer-to-unsigned conversion. */
1061 ir_unop_u2i, /**< Unsigned-to-integer conversion. */
1062 ir_unop_bitcast_i2f, /**< Bit-identical int-to-float "conversion" */
1063 ir_unop_bitcast_f2i, /**< Bit-identical float-to-int "conversion" */
1064 ir_unop_bitcast_u2f, /**< Bit-identical uint-to-float "conversion" */
1065 ir_unop_bitcast_f2u, /**< Bit-identical float-to-uint "conversion" */
1066 ir_unop_any,
1067
1068 /**
1069 * \name Unary floating-point rounding operations.
1070 */
1071 /*@{*/
1072 ir_unop_trunc,
1073 ir_unop_ceil,
1074 ir_unop_floor,
1075 ir_unop_fract,
1076 ir_unop_round_even,
1077 /*@}*/
1078
1079 /**
1080 * \name Trigonometric operations.
1081 */
1082 /*@{*/
1083 ir_unop_sin,
1084 ir_unop_cos,
1085 ir_unop_sin_reduced, /**< Reduced range sin. [-pi, pi] */
1086 ir_unop_cos_reduced, /**< Reduced range cos. [-pi, pi] */
1087 /*@}*/
1088
1089 /**
1090 * \name Partial derivatives.
1091 */
1092 /*@{*/
1093 ir_unop_dFdx,
1094 ir_unop_dFdy,
1095 /*@}*/
1096
1097 /**
1098 * \name Floating point pack and unpack operations.
1099 */
1100 /*@{*/
1101 ir_unop_pack_snorm_2x16,
1102 ir_unop_pack_snorm_4x8,
1103 ir_unop_pack_unorm_2x16,
1104 ir_unop_pack_unorm_4x8,
1105 ir_unop_pack_half_2x16,
1106 ir_unop_unpack_snorm_2x16,
1107 ir_unop_unpack_snorm_4x8,
1108 ir_unop_unpack_unorm_2x16,
1109 ir_unop_unpack_unorm_4x8,
1110 ir_unop_unpack_half_2x16,
1111 /*@}*/
1112
1113 /**
1114 * \name Lowered floating point unpacking operations.
1115 *
1116 * \see lower_packing_builtins_visitor::split_unpack_half_2x16
1117 */
1118 /*@{*/
1119 ir_unop_unpack_half_2x16_split_x,
1120 ir_unop_unpack_half_2x16_split_y,
1121 /*@}*/
1122
1123 /**
1124 * \name Bit operations, part of ARB_gpu_shader5.
1125 */
1126 /*@{*/
1127 ir_unop_bitfield_reverse,
1128 ir_unop_bit_count,
1129 ir_unop_find_msb,
1130 ir_unop_find_lsb,
1131 /*@}*/
1132
1133 ir_unop_noise,
1134
1135 /**
1136 * A sentinel marking the last of the unary operations.
1137 */
1138 ir_last_unop = ir_unop_noise,
1139
1140 ir_binop_add,
1141 ir_binop_sub,
1142 ir_binop_mul, /**< Floating-point or low 32-bit integer multiply. */
1143 ir_binop_imul_high, /**< Calculates the high 32-bits of a 64-bit multiply. */
1144 ir_binop_div,
1145
1146 /**
1147 * Returns the carry resulting from the addition of the two arguments.
1148 */
1149 /*@{*/
1150 ir_binop_carry,
1151 /*@}*/
1152
1153 /**
1154 * Returns the borrow resulting from the subtraction of the second argument
1155 * from the first argument.
1156 */
1157 /*@{*/
1158 ir_binop_borrow,
1159 /*@}*/
1160
1161 /**
1162 * Takes one of two combinations of arguments:
1163 *
1164 * - mod(vecN, vecN)
1165 * - mod(vecN, float)
1166 *
1167 * Does not take integer types.
1168 */
1169 ir_binop_mod,
1170
1171 /**
1172 * \name Binary comparison operators which return a boolean vector.
1173 * The type of both operands must be equal.
1174 */
1175 /*@{*/
1176 ir_binop_less,
1177 ir_binop_greater,
1178 ir_binop_lequal,
1179 ir_binop_gequal,
1180 ir_binop_equal,
1181 ir_binop_nequal,
1182 /**
1183 * Returns single boolean for whether all components of operands[0]
1184 * equal the components of operands[1].
1185 */
1186 ir_binop_all_equal,
1187 /**
1188 * Returns single boolean for whether any component of operands[0]
1189 * is not equal to the corresponding component of operands[1].
1190 */
1191 ir_binop_any_nequal,
1192 /*@}*/
1193
1194 /**
1195 * \name Bit-wise binary operations.
1196 */
1197 /*@{*/
1198 ir_binop_lshift,
1199 ir_binop_rshift,
1200 ir_binop_bit_and,
1201 ir_binop_bit_xor,
1202 ir_binop_bit_or,
1203 /*@}*/
1204
1205 ir_binop_logic_and,
1206 ir_binop_logic_xor,
1207 ir_binop_logic_or,
1208
1209 ir_binop_dot,
1210 ir_binop_min,
1211 ir_binop_max,
1212
1213 ir_binop_pow,
1214
1215 /**
1216 * \name Lowered floating point packing operations.
1217 *
1218 * \see lower_packing_builtins_visitor::split_pack_half_2x16
1219 */
1220 /*@{*/
1221 ir_binop_pack_half_2x16_split,
1222 /*@}*/
1223
1224 /**
1225 * \name First half of a lowered bitfieldInsert() operation.
1226 *
1227 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1228 */
1229 /*@{*/
1230 ir_binop_bfm,
1231 /*@}*/
1232
1233 /**
1234 * Load a value the size of a given GLSL type from a uniform block.
1235 *
1236 * operand0 is the ir_constant uniform block index in the linked shader.
1237 * operand1 is a byte offset within the uniform block.
1238 */
1239 ir_binop_ubo_load,
1240
1241 /**
1242 * \name Multiplies a number by two to a power, part of ARB_gpu_shader5.
1243 */
1244 /*@{*/
1245 ir_binop_ldexp,
1246 /*@}*/
1247
1248 /**
1249 * Extract a scalar from a vector
1250 *
1251 * operand0 is the vector
1252 * operand1 is the index of the field to read from operand0
1253 */
1254 ir_binop_vector_extract,
1255
1256 /**
1257 * A sentinel marking the last of the binary operations.
1258 */
1259 ir_last_binop = ir_binop_vector_extract,
1260
1261 /**
1262 * \name Fused floating-point multiply-add, part of ARB_gpu_shader5.
1263 */
1264 /*@{*/
1265 ir_triop_fma,
1266 /*@}*/
1267
1268 ir_triop_lrp,
1269
1270 /**
1271 * \name Conditional Select
1272 *
1273 * A vector conditional select instruction (like ?:, but operating per-
1274 * component on vectors).
1275 *
1276 * \see lower_instructions_visitor::ldexp_to_arith
1277 */
1278 /*@{*/
1279 ir_triop_csel,
1280 /*@}*/
1281
1282 /**
1283 * \name Second half of a lowered bitfieldInsert() operation.
1284 *
1285 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1286 */
1287 /*@{*/
1288 ir_triop_bfi,
1289 /*@}*/
1290
1291 ir_triop_bitfield_extract,
1292
1293 /**
1294 * Generate a value with one field of a vector changed
1295 *
1296 * operand0 is the vector
1297 * operand1 is the value to write into the vector result
1298 * operand2 is the index in operand0 to be modified
1299 */
1300 ir_triop_vector_insert,
1301
1302 /**
1303 * A sentinel marking the last of the ternary operations.
1304 */
1305 ir_last_triop = ir_triop_vector_insert,
1306
1307 ir_quadop_bitfield_insert,
1308
1309 ir_quadop_vector,
1310
1311 /**
1312 * A sentinel marking the last of the ternary operations.
1313 */
1314 ir_last_quadop = ir_quadop_vector,
1315
1316 /**
1317 * A sentinel marking the last of all operations.
1318 */
1319 ir_last_opcode = ir_quadop_vector
1320 };
1321
1322 class ir_expression : public ir_rvalue {
1323 public:
1324 ir_expression(int op, const struct glsl_type *type,
1325 ir_rvalue *op0, ir_rvalue *op1 = NULL,
1326 ir_rvalue *op2 = NULL, ir_rvalue *op3 = NULL);
1327
1328 /**
1329 * Constructor for unary operation expressions
1330 */
1331 ir_expression(int op, ir_rvalue *);
1332
1333 /**
1334 * Constructor for binary operation expressions
1335 */
1336 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1);
1337
1338 /**
1339 * Constructor for ternary operation expressions
1340 */
1341 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1, ir_rvalue *op2);
1342
1343 virtual ir_expression *as_expression()
1344 {
1345 return this;
1346 }
1347
1348 virtual ir_expression *clone(void *mem_ctx, struct hash_table *ht) const;
1349
1350 /**
1351 * Attempt to constant-fold the expression
1352 *
1353 * The "variable_context" hash table links ir_variable * to ir_constant *
1354 * that represent the variables' values. \c NULL represents an empty
1355 * context.
1356 *
1357 * If the expression cannot be constant folded, this method will return
1358 * \c NULL.
1359 */
1360 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1361
1362 /**
1363 * Determine the number of operands used by an expression
1364 */
1365 static unsigned int get_num_operands(ir_expression_operation);
1366
1367 /**
1368 * Determine the number of operands used by an expression
1369 */
1370 unsigned int get_num_operands() const
1371 {
1372 return (this->operation == ir_quadop_vector)
1373 ? this->type->vector_elements : get_num_operands(operation);
1374 }
1375
1376 /**
1377 * Return a string representing this expression's operator.
1378 */
1379 const char *operator_string();
1380
1381 /**
1382 * Return a string representing this expression's operator.
1383 */
1384 static const char *operator_string(ir_expression_operation);
1385
1386
1387 /**
1388 * Do a reverse-lookup to translate the given string into an operator.
1389 */
1390 static ir_expression_operation get_operator(const char *);
1391
1392 virtual void accept(ir_visitor *v)
1393 {
1394 v->visit(this);
1395 }
1396
1397 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1398
1399 ir_expression_operation operation;
1400 ir_rvalue *operands[4];
1401 };
1402
1403
1404 /**
1405 * HIR instruction representing a high-level function call, containing a list
1406 * of parameters and returning a value in the supplied temporary.
1407 */
1408 class ir_call : public ir_instruction {
1409 public:
1410 ir_call(ir_function_signature *callee,
1411 ir_dereference_variable *return_deref,
1412 exec_list *actual_parameters)
1413 : return_deref(return_deref), callee(callee)
1414 {
1415 ir_type = ir_type_call;
1416 assert(callee->return_type != NULL);
1417 actual_parameters->move_nodes_to(& this->actual_parameters);
1418 this->use_builtin = callee->is_builtin();
1419 }
1420
1421 virtual ir_call *clone(void *mem_ctx, struct hash_table *ht) const;
1422
1423 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1424
1425 virtual ir_call *as_call()
1426 {
1427 return this;
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 /**
1438 * Get an iterator for the set of acutal parameters
1439 */
1440 exec_list_iterator iterator()
1441 {
1442 return actual_parameters.iterator();
1443 }
1444
1445 /**
1446 * Get the name of the function being called.
1447 */
1448 const char *callee_name() const
1449 {
1450 return callee->function_name();
1451 }
1452
1453 /**
1454 * Generates an inline version of the function before @ir,
1455 * storing the return value in return_deref.
1456 */
1457 void generate_inline(ir_instruction *ir);
1458
1459 /**
1460 * Storage for the function's return value.
1461 * This must be NULL if the return type is void.
1462 */
1463 ir_dereference_variable *return_deref;
1464
1465 /**
1466 * The specific function signature being called.
1467 */
1468 ir_function_signature *callee;
1469
1470 /* List of ir_rvalue of paramaters passed in this call. */
1471 exec_list actual_parameters;
1472
1473 /** Should this call only bind to a built-in function? */
1474 bool use_builtin;
1475 };
1476
1477
1478 /**
1479 * \name Jump-like IR instructions.
1480 *
1481 * These include \c break, \c continue, \c return, and \c discard.
1482 */
1483 /*@{*/
1484 class ir_jump : public ir_instruction {
1485 protected:
1486 ir_jump()
1487 {
1488 ir_type = ir_type_unset;
1489 }
1490
1491 public:
1492 virtual ir_jump *as_jump()
1493 {
1494 return this;
1495 }
1496 };
1497
1498 class ir_return : public ir_jump {
1499 public:
1500 ir_return()
1501 : value(NULL)
1502 {
1503 this->ir_type = ir_type_return;
1504 }
1505
1506 ir_return(ir_rvalue *value)
1507 : value(value)
1508 {
1509 this->ir_type = ir_type_return;
1510 }
1511
1512 virtual ir_return *clone(void *mem_ctx, struct hash_table *) const;
1513
1514 virtual ir_return *as_return()
1515 {
1516 return this;
1517 }
1518
1519 ir_rvalue *get_value() const
1520 {
1521 return value;
1522 }
1523
1524 virtual void accept(ir_visitor *v)
1525 {
1526 v->visit(this);
1527 }
1528
1529 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1530
1531 ir_rvalue *value;
1532 };
1533
1534
1535 /**
1536 * Jump instructions used inside loops
1537 *
1538 * These include \c break and \c continue. The \c break within a loop is
1539 * different from the \c break within a switch-statement.
1540 *
1541 * \sa ir_switch_jump
1542 */
1543 class ir_loop_jump : public ir_jump {
1544 public:
1545 enum jump_mode {
1546 jump_break,
1547 jump_continue
1548 };
1549
1550 ir_loop_jump(jump_mode mode)
1551 {
1552 this->ir_type = ir_type_loop_jump;
1553 this->mode = mode;
1554 }
1555
1556 virtual ir_loop_jump *clone(void *mem_ctx, struct hash_table *) const;
1557
1558 virtual void accept(ir_visitor *v)
1559 {
1560 v->visit(this);
1561 }
1562
1563 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1564
1565 bool is_break() const
1566 {
1567 return mode == jump_break;
1568 }
1569
1570 bool is_continue() const
1571 {
1572 return mode == jump_continue;
1573 }
1574
1575 /** Mode selector for the jump instruction. */
1576 enum jump_mode mode;
1577 };
1578
1579 /**
1580 * IR instruction representing discard statements.
1581 */
1582 class ir_discard : public ir_jump {
1583 public:
1584 ir_discard()
1585 {
1586 this->ir_type = ir_type_discard;
1587 this->condition = NULL;
1588 }
1589
1590 ir_discard(ir_rvalue *cond)
1591 {
1592 this->ir_type = ir_type_discard;
1593 this->condition = cond;
1594 }
1595
1596 virtual ir_discard *clone(void *mem_ctx, struct hash_table *ht) const;
1597
1598 virtual void accept(ir_visitor *v)
1599 {
1600 v->visit(this);
1601 }
1602
1603 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1604
1605 virtual ir_discard *as_discard()
1606 {
1607 return this;
1608 }
1609
1610 ir_rvalue *condition;
1611 };
1612 /*@}*/
1613
1614
1615 /**
1616 * Texture sampling opcodes used in ir_texture
1617 */
1618 enum ir_texture_opcode {
1619 ir_tex, /**< Regular texture look-up */
1620 ir_txb, /**< Texture look-up with LOD bias */
1621 ir_txl, /**< Texture look-up with explicit LOD */
1622 ir_txd, /**< Texture look-up with partial derivatvies */
1623 ir_txf, /**< Texel fetch with explicit LOD */
1624 ir_txf_ms, /**< Multisample texture fetch */
1625 ir_txs, /**< Texture size */
1626 ir_lod, /**< Texture lod query */
1627 ir_tg4, /**< Texture gather */
1628 ir_query_levels /**< Texture levels query */
1629 };
1630
1631
1632 /**
1633 * IR instruction to sample a texture
1634 *
1635 * The specific form of the IR instruction depends on the \c mode value
1636 * selected from \c ir_texture_opcodes. In the printed IR, these will
1637 * appear as:
1638 *
1639 * Texel offset (0 or an expression)
1640 * | Projection divisor
1641 * | | Shadow comparitor
1642 * | | |
1643 * v v v
1644 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1645 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1646 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1647 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1648 * (txf <type> <sampler> <coordinate> 0 <lod>)
1649 * (txf_ms
1650 * <type> <sampler> <coordinate> <sample_index>)
1651 * (txs <type> <sampler> <lod>)
1652 * (lod <type> <sampler> <coordinate>)
1653 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
1654 * (query_levels <type> <sampler>)
1655 */
1656 class ir_texture : public ir_rvalue {
1657 public:
1658 ir_texture(enum ir_texture_opcode op)
1659 : op(op), sampler(NULL), coordinate(NULL), projector(NULL),
1660 shadow_comparitor(NULL), offset(NULL)
1661 {
1662 this->ir_type = ir_type_texture;
1663 memset(&lod_info, 0, sizeof(lod_info));
1664 }
1665
1666 virtual ir_texture *clone(void *mem_ctx, struct hash_table *) const;
1667
1668 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1669
1670 virtual void accept(ir_visitor *v)
1671 {
1672 v->visit(this);
1673 }
1674
1675 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1676
1677 /**
1678 * Return a string representing the ir_texture_opcode.
1679 */
1680 const char *opcode_string();
1681
1682 /** Set the sampler and type. */
1683 void set_sampler(ir_dereference *sampler, const glsl_type *type);
1684
1685 /**
1686 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1687 */
1688 static ir_texture_opcode get_opcode(const char *);
1689
1690 enum ir_texture_opcode op;
1691
1692 /** Sampler to use for the texture access. */
1693 ir_dereference *sampler;
1694
1695 /** Texture coordinate to sample */
1696 ir_rvalue *coordinate;
1697
1698 /**
1699 * Value used for projective divide.
1700 *
1701 * If there is no projective divide (the common case), this will be
1702 * \c NULL. Optimization passes should check for this to point to a constant
1703 * of 1.0 and replace that with \c NULL.
1704 */
1705 ir_rvalue *projector;
1706
1707 /**
1708 * Coordinate used for comparison on shadow look-ups.
1709 *
1710 * If there is no shadow comparison, this will be \c NULL. For the
1711 * \c ir_txf opcode, this *must* be \c NULL.
1712 */
1713 ir_rvalue *shadow_comparitor;
1714
1715 /** Texel offset. */
1716 ir_rvalue *offset;
1717
1718 union {
1719 ir_rvalue *lod; /**< Floating point LOD */
1720 ir_rvalue *bias; /**< Floating point LOD bias */
1721 ir_rvalue *sample_index; /**< MSAA sample index */
1722 ir_rvalue *component; /**< Gather component selector */
1723 struct {
1724 ir_rvalue *dPdx; /**< Partial derivative of coordinate wrt X */
1725 ir_rvalue *dPdy; /**< Partial derivative of coordinate wrt Y */
1726 } grad;
1727 } lod_info;
1728 };
1729
1730
1731 struct ir_swizzle_mask {
1732 unsigned x:2;
1733 unsigned y:2;
1734 unsigned z:2;
1735 unsigned w:2;
1736
1737 /**
1738 * Number of components in the swizzle.
1739 */
1740 unsigned num_components:3;
1741
1742 /**
1743 * Does the swizzle contain duplicate components?
1744 *
1745 * L-value swizzles cannot contain duplicate components.
1746 */
1747 unsigned has_duplicates:1;
1748 };
1749
1750
1751 class ir_swizzle : public ir_rvalue {
1752 public:
1753 ir_swizzle(ir_rvalue *, unsigned x, unsigned y, unsigned z, unsigned w,
1754 unsigned count);
1755
1756 ir_swizzle(ir_rvalue *val, const unsigned *components, unsigned count);
1757
1758 ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask);
1759
1760 virtual ir_swizzle *clone(void *mem_ctx, struct hash_table *) const;
1761
1762 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1763
1764 virtual ir_swizzle *as_swizzle()
1765 {
1766 return this;
1767 }
1768
1769 /**
1770 * Construct an ir_swizzle from the textual representation. Can fail.
1771 */
1772 static ir_swizzle *create(ir_rvalue *, const char *, unsigned vector_length);
1773
1774 virtual void accept(ir_visitor *v)
1775 {
1776 v->visit(this);
1777 }
1778
1779 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1780
1781 bool is_lvalue() const
1782 {
1783 return val->is_lvalue() && !mask.has_duplicates;
1784 }
1785
1786 /**
1787 * Get the variable that is ultimately referenced by an r-value
1788 */
1789 virtual ir_variable *variable_referenced() const;
1790
1791 ir_rvalue *val;
1792 ir_swizzle_mask mask;
1793
1794 private:
1795 /**
1796 * Initialize the mask component of a swizzle
1797 *
1798 * This is used by the \c ir_swizzle constructors.
1799 */
1800 void init_mask(const unsigned *components, unsigned count);
1801 };
1802
1803
1804 class ir_dereference : public ir_rvalue {
1805 public:
1806 virtual ir_dereference *clone(void *mem_ctx, struct hash_table *) const = 0;
1807
1808 virtual ir_dereference *as_dereference()
1809 {
1810 return this;
1811 }
1812
1813 bool is_lvalue() const;
1814
1815 /**
1816 * Get the variable that is ultimately referenced by an r-value
1817 */
1818 virtual ir_variable *variable_referenced() const = 0;
1819
1820 /**
1821 * Get the constant that is ultimately referenced by an r-value,
1822 * in a constant expression evaluation context.
1823 *
1824 * The offset is used when the reference is to a specific column of
1825 * a matrix.
1826 */
1827 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const = 0;
1828 };
1829
1830
1831 class ir_dereference_variable : public ir_dereference {
1832 public:
1833 ir_dereference_variable(ir_variable *var);
1834
1835 virtual ir_dereference_variable *clone(void *mem_ctx,
1836 struct hash_table *) const;
1837
1838 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1839
1840 virtual ir_dereference_variable *as_dereference_variable()
1841 {
1842 return this;
1843 }
1844
1845 /**
1846 * Get the variable that is ultimately referenced by an r-value
1847 */
1848 virtual ir_variable *variable_referenced() const
1849 {
1850 return this->var;
1851 }
1852
1853 /**
1854 * Get the constant that is ultimately referenced by an r-value,
1855 * in a constant expression evaluation context.
1856 *
1857 * The offset is used when the reference is to a specific column of
1858 * a matrix.
1859 */
1860 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
1861
1862 virtual ir_variable *whole_variable_referenced()
1863 {
1864 /* ir_dereference_variable objects always dereference the entire
1865 * variable. However, if this dereference is dereferenced by anything
1866 * else, the complete deferefernce chain is not a whole-variable
1867 * dereference. This method should only be called on the top most
1868 * ir_rvalue in a dereference chain.
1869 */
1870 return this->var;
1871 }
1872
1873 virtual void accept(ir_visitor *v)
1874 {
1875 v->visit(this);
1876 }
1877
1878 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1879
1880 /**
1881 * Object being dereferenced.
1882 */
1883 ir_variable *var;
1884 };
1885
1886
1887 class ir_dereference_array : public ir_dereference {
1888 public:
1889 ir_dereference_array(ir_rvalue *value, ir_rvalue *array_index);
1890
1891 ir_dereference_array(ir_variable *var, ir_rvalue *array_index);
1892
1893 virtual ir_dereference_array *clone(void *mem_ctx,
1894 struct hash_table *) const;
1895
1896 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1897
1898 virtual ir_dereference_array *as_dereference_array()
1899 {
1900 return this;
1901 }
1902
1903 /**
1904 * Get the variable that is ultimately referenced by an r-value
1905 */
1906 virtual ir_variable *variable_referenced() const
1907 {
1908 return this->array->variable_referenced();
1909 }
1910
1911 /**
1912 * Get the constant that is ultimately referenced by an r-value,
1913 * in a constant expression evaluation context.
1914 *
1915 * The offset is used when the reference is to a specific column of
1916 * a matrix.
1917 */
1918 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
1919
1920 virtual void accept(ir_visitor *v)
1921 {
1922 v->visit(this);
1923 }
1924
1925 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1926
1927 ir_rvalue *array;
1928 ir_rvalue *array_index;
1929
1930 private:
1931 void set_array(ir_rvalue *value);
1932 };
1933
1934
1935 class ir_dereference_record : public ir_dereference {
1936 public:
1937 ir_dereference_record(ir_rvalue *value, const char *field);
1938
1939 ir_dereference_record(ir_variable *var, const char *field);
1940
1941 virtual ir_dereference_record *clone(void *mem_ctx,
1942 struct hash_table *) const;
1943
1944 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1945
1946 virtual ir_dereference_record *as_dereference_record()
1947 {
1948 return this;
1949 }
1950
1951 /**
1952 * Get the variable that is ultimately referenced by an r-value
1953 */
1954 virtual ir_variable *variable_referenced() const
1955 {
1956 return this->record->variable_referenced();
1957 }
1958
1959 /**
1960 * Get the constant that is ultimately referenced by an r-value,
1961 * in a constant expression evaluation context.
1962 *
1963 * The offset is used when the reference is to a specific column of
1964 * a matrix.
1965 */
1966 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
1967
1968 virtual void accept(ir_visitor *v)
1969 {
1970 v->visit(this);
1971 }
1972
1973 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1974
1975 ir_rvalue *record;
1976 const char *field;
1977 };
1978
1979
1980 /**
1981 * Data stored in an ir_constant
1982 */
1983 union ir_constant_data {
1984 unsigned u[16];
1985 int i[16];
1986 float f[16];
1987 bool b[16];
1988 };
1989
1990
1991 class ir_constant : public ir_rvalue {
1992 public:
1993 ir_constant(const struct glsl_type *type, const ir_constant_data *data);
1994 ir_constant(bool b, unsigned vector_elements=1);
1995 ir_constant(unsigned int u, unsigned vector_elements=1);
1996 ir_constant(int i, unsigned vector_elements=1);
1997 ir_constant(float f, unsigned vector_elements=1);
1998
1999 /**
2000 * Construct an ir_constant from a list of ir_constant values
2001 */
2002 ir_constant(const struct glsl_type *type, exec_list *values);
2003
2004 /**
2005 * Construct an ir_constant from a scalar component of another ir_constant
2006 *
2007 * The new \c ir_constant inherits the type of the component from the
2008 * source constant.
2009 *
2010 * \note
2011 * In the case of a matrix constant, the new constant is a scalar, \b not
2012 * a vector.
2013 */
2014 ir_constant(const ir_constant *c, unsigned i);
2015
2016 /**
2017 * Return a new ir_constant of the specified type containing all zeros.
2018 */
2019 static ir_constant *zero(void *mem_ctx, const glsl_type *type);
2020
2021 virtual ir_constant *clone(void *mem_ctx, struct hash_table *) const;
2022
2023 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
2024
2025 virtual ir_constant *as_constant()
2026 {
2027 return this;
2028 }
2029
2030 virtual void accept(ir_visitor *v)
2031 {
2032 v->visit(this);
2033 }
2034
2035 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2036
2037 /**
2038 * Get a particular component of a constant as a specific type
2039 *
2040 * This is useful, for example, to get a value from an integer constant
2041 * as a float or bool. This appears frequently when constructors are
2042 * called with all constant parameters.
2043 */
2044 /*@{*/
2045 bool get_bool_component(unsigned i) const;
2046 float get_float_component(unsigned i) const;
2047 int get_int_component(unsigned i) const;
2048 unsigned get_uint_component(unsigned i) const;
2049 /*@}*/
2050
2051 ir_constant *get_array_element(unsigned i) const;
2052
2053 ir_constant *get_record_field(const char *name);
2054
2055 /**
2056 * Copy the values on another constant at a given offset.
2057 *
2058 * The offset is ignored for array or struct copies, it's only for
2059 * scalars or vectors into vectors or matrices.
2060 *
2061 * With identical types on both sides and zero offset it's clone()
2062 * without creating a new object.
2063 */
2064
2065 void copy_offset(ir_constant *src, int offset);
2066
2067 /**
2068 * Copy the values on another constant at a given offset and
2069 * following an assign-like mask.
2070 *
2071 * The mask is ignored for scalars.
2072 *
2073 * Note that this function only handles what assign can handle,
2074 * i.e. at most a vector as source and a column of a matrix as
2075 * destination.
2076 */
2077
2078 void copy_masked_offset(ir_constant *src, int offset, unsigned int mask);
2079
2080 /**
2081 * Determine whether a constant has the same value as another constant
2082 *
2083 * \sa ir_constant::is_zero, ir_constant::is_one,
2084 * ir_constant::is_negative_one, ir_constant::is_basis
2085 */
2086 bool has_value(const ir_constant *) const;
2087
2088 virtual bool is_zero() const;
2089 virtual bool is_one() const;
2090 virtual bool is_negative_one() const;
2091 virtual bool is_basis() const;
2092
2093 /**
2094 * Value of the constant.
2095 *
2096 * The field used to back the values supplied by the constant is determined
2097 * by the type associated with the \c ir_instruction. Constants may be
2098 * scalars, vectors, or matrices.
2099 */
2100 union ir_constant_data value;
2101
2102 /* Array elements */
2103 ir_constant **array_elements;
2104
2105 /* Structure fields */
2106 exec_list components;
2107
2108 private:
2109 /**
2110 * Parameterless constructor only used by the clone method
2111 */
2112 ir_constant(void);
2113 };
2114
2115 /*@}*/
2116
2117 /**
2118 * IR instruction to emit a vertex in a geometry shader.
2119 */
2120 class ir_emit_vertex : public ir_instruction {
2121 public:
2122 ir_emit_vertex()
2123 {
2124 ir_type = ir_type_emit_vertex;
2125 }
2126
2127 virtual void accept(ir_visitor *v)
2128 {
2129 v->visit(this);
2130 }
2131
2132 virtual ir_emit_vertex *clone(void *mem_ctx, struct hash_table *) const
2133 {
2134 return new(mem_ctx) ir_emit_vertex();
2135 }
2136
2137 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2138 };
2139
2140 /**
2141 * IR instruction to complete the current primitive and start a new one in a
2142 * geometry shader.
2143 */
2144 class ir_end_primitive : public ir_instruction {
2145 public:
2146 ir_end_primitive()
2147 {
2148 ir_type = ir_type_end_primitive;
2149 }
2150
2151 virtual void accept(ir_visitor *v)
2152 {
2153 v->visit(this);
2154 }
2155
2156 virtual ir_end_primitive *clone(void *mem_ctx, struct hash_table *) const
2157 {
2158 return new(mem_ctx) ir_end_primitive();
2159 }
2160
2161 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2162 };
2163
2164 /**
2165 * Apply a visitor to each IR node in a list
2166 */
2167 void
2168 visit_exec_list(exec_list *list, ir_visitor *visitor);
2169
2170 /**
2171 * Validate invariants on each IR node in a list
2172 */
2173 void validate_ir_tree(exec_list *instructions);
2174
2175 struct _mesa_glsl_parse_state;
2176 struct gl_shader_program;
2177
2178 /**
2179 * Detect whether an unlinked shader contains static recursion
2180 *
2181 * If the list of instructions is determined to contain static recursion,
2182 * \c _mesa_glsl_error will be called to emit error messages for each function
2183 * that is in the recursion cycle.
2184 */
2185 void
2186 detect_recursion_unlinked(struct _mesa_glsl_parse_state *state,
2187 exec_list *instructions);
2188
2189 /**
2190 * Detect whether a linked shader contains static recursion
2191 *
2192 * If the list of instructions is determined to contain static recursion,
2193 * \c link_error_printf will be called to emit error messages for each function
2194 * that is in the recursion cycle. In addition,
2195 * \c gl_shader_program::LinkStatus will be set to false.
2196 */
2197 void
2198 detect_recursion_linked(struct gl_shader_program *prog,
2199 exec_list *instructions);
2200
2201 /**
2202 * Make a clone of each IR instruction in a list
2203 *
2204 * \param in List of IR instructions that are to be cloned
2205 * \param out List to hold the cloned instructions
2206 */
2207 void
2208 clone_ir_list(void *mem_ctx, exec_list *out, const exec_list *in);
2209
2210 extern void
2211 _mesa_glsl_initialize_variables(exec_list *instructions,
2212 struct _mesa_glsl_parse_state *state);
2213
2214 extern void
2215 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state *state);
2216
2217 extern void
2218 _mesa_glsl_initialize_builtin_functions();
2219
2220 extern ir_function_signature *
2221 _mesa_glsl_find_builtin_function(_mesa_glsl_parse_state *state,
2222 const char *name, exec_list *actual_parameters);
2223
2224 extern void
2225 _mesa_glsl_release_functions(void);
2226
2227 extern void
2228 _mesa_glsl_release_builtin_functions(void);
2229
2230 extern void
2231 reparent_ir(exec_list *list, void *mem_ctx);
2232
2233 struct glsl_symbol_table;
2234
2235 extern void
2236 import_prototypes(const exec_list *source, exec_list *dest,
2237 struct glsl_symbol_table *symbols, void *mem_ctx);
2238
2239 extern bool
2240 ir_has_call(ir_instruction *ir);
2241
2242 extern void
2243 do_set_program_inouts(exec_list *instructions, struct gl_program *prog,
2244 GLenum shader_type);
2245
2246 extern char *
2247 prototype_string(const glsl_type *return_type, const char *name,
2248 exec_list *parameters);
2249
2250 extern "C" {
2251 #endif /* __cplusplus */
2252
2253 extern void _mesa_print_ir(struct exec_list *instructions,
2254 struct _mesa_glsl_parse_state *state);
2255
2256 #ifdef __cplusplus
2257 } /* extern "C" */
2258 #endif
2259
2260 unsigned
2261 vertices_per_prim(GLenum prim);
2262
2263 #endif /* IR_H */