glsl: Add parameter to .equals() to ignore an IR type.
[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 /**
91 * Base class of all IR instructions
92 */
93 class ir_instruction : public exec_node {
94 public:
95 enum ir_node_type ir_type;
96
97 /**
98 * GCC 4.7+ and clang warn when deleting an ir_instruction unless
99 * there's a virtual destructor present. Because we almost
100 * universally use ralloc for our memory management of
101 * ir_instructions, the destructor doesn't need to do any work.
102 */
103 virtual ~ir_instruction()
104 {
105 }
106
107 /** ir_print_visitor helper for debugging. */
108 void print(void) const;
109
110 virtual void accept(ir_visitor *) = 0;
111 virtual ir_visitor_status accept(ir_hierarchical_visitor *) = 0;
112 virtual ir_instruction *clone(void *mem_ctx,
113 struct hash_table *ht) const = 0;
114
115 /**
116 * \name IR instruction downcast functions
117 *
118 * These functions either cast the object to a derived class or return
119 * \c NULL if the object's type does not match the specified derived class.
120 * Additional downcast functions will be added as needed.
121 */
122 /*@{*/
123 virtual class ir_variable * as_variable() { return NULL; }
124 virtual class ir_function * as_function() { return NULL; }
125 virtual class ir_dereference * as_dereference() { return NULL; }
126 virtual class ir_dereference_array * as_dereference_array() { return NULL; }
127 virtual class ir_dereference_variable *as_dereference_variable() { return NULL; }
128 virtual class ir_dereference_record *as_dereference_record() { return NULL; }
129 virtual class ir_expression * as_expression() { return NULL; }
130 virtual class ir_rvalue * as_rvalue() { return NULL; }
131 virtual class ir_loop * as_loop() { return NULL; }
132 virtual class ir_assignment * as_assignment() { return NULL; }
133 virtual class ir_call * as_call() { return NULL; }
134 virtual class ir_return * as_return() { return NULL; }
135 virtual class ir_if * as_if() { return NULL; }
136 virtual class ir_swizzle * as_swizzle() { return NULL; }
137 virtual class ir_texture * as_texture() { return NULL; }
138 virtual class ir_constant * as_constant() { return NULL; }
139 virtual class ir_discard * as_discard() { return NULL; }
140 virtual class ir_jump * as_jump() { return NULL; }
141 /*@}*/
142
143 /**
144 * IR equality method: Return true if the referenced instruction would
145 * return the same value as this one.
146 *
147 * This intended to be used for CSE and algebraic optimizations, on rvalues
148 * in particular. No support for other instruction types (assignments,
149 * jumps, calls, etc.) is planned.
150 */
151 virtual bool equals(ir_instruction *ir, enum ir_node_type ignore = ir_type_unset);
152
153 protected:
154 ir_instruction()
155 {
156 ir_type = ir_type_unset;
157 }
158 };
159
160
161 /**
162 * The base class for all "values"/expression trees.
163 */
164 class ir_rvalue : public ir_instruction {
165 public:
166 const struct glsl_type *type;
167
168 virtual ir_rvalue *clone(void *mem_ctx, struct hash_table *) const;
169
170 virtual void accept(ir_visitor *v)
171 {
172 v->visit(this);
173 }
174
175 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
176
177 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
178
179 virtual ir_rvalue * as_rvalue()
180 {
181 return this;
182 }
183
184 ir_rvalue *as_rvalue_to_saturate();
185
186 virtual bool is_lvalue() const
187 {
188 return false;
189 }
190
191 /**
192 * Get the variable that is ultimately referenced by an r-value
193 */
194 virtual ir_variable *variable_referenced() const
195 {
196 return NULL;
197 }
198
199
200 /**
201 * If an r-value is a reference to a whole variable, get that variable
202 *
203 * \return
204 * Pointer to a variable that is completely dereferenced by the r-value. If
205 * the r-value is not a dereference or the dereference does not access the
206 * entire variable (i.e., it's just one array element, struct field), \c NULL
207 * is returned.
208 */
209 virtual ir_variable *whole_variable_referenced()
210 {
211 return NULL;
212 }
213
214 /**
215 * Determine if an r-value has the value zero
216 *
217 * The base implementation of this function always returns \c false. The
218 * \c ir_constant class over-rides this function to return \c true \b only
219 * for vector and scalar types that have all elements set to the value
220 * zero (or \c false for booleans).
221 *
222 * \sa ir_constant::has_value, ir_rvalue::is_one, ir_rvalue::is_negative_one,
223 * ir_constant::is_basis
224 */
225 virtual bool is_zero() const;
226
227 /**
228 * Determine if an r-value has the value one
229 *
230 * The base implementation of this function always returns \c false. The
231 * \c ir_constant class over-rides this function to return \c true \b only
232 * for vector and scalar types that have all elements set to the value
233 * one (or \c true for booleans).
234 *
235 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_negative_one,
236 * ir_constant::is_basis
237 */
238 virtual bool is_one() const;
239
240 /**
241 * Determine if an r-value has the value negative one
242 *
243 * The base implementation of this function always returns \c false. The
244 * \c ir_constant class over-rides this function to return \c true \b only
245 * for vector and scalar types that have all elements set to the value
246 * negative one. For boolean types, the result is always \c false.
247 *
248 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one
249 * ir_constant::is_basis
250 */
251 virtual bool is_negative_one() const;
252
253 /**
254 * Determine if an r-value is a basis vector
255 *
256 * The base implementation of this function always returns \c false. The
257 * \c ir_constant class over-rides this function to return \c true \b only
258 * for vector and scalar types that have one element set to the value one,
259 * and the other elements set to the value zero. For boolean types, the
260 * result is always \c false.
261 *
262 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one,
263 * is_constant::is_negative_one
264 */
265 virtual bool is_basis() const;
266
267
268 /**
269 * Return a generic value of error_type.
270 *
271 * Allocation will be performed with 'mem_ctx' as ralloc owner.
272 */
273 static ir_rvalue *error_value(void *mem_ctx);
274
275 protected:
276 ir_rvalue();
277 };
278
279
280 /**
281 * Variable storage classes
282 */
283 enum ir_variable_mode {
284 ir_var_auto = 0, /**< Function local variables and globals. */
285 ir_var_uniform, /**< Variable declared as a uniform. */
286 ir_var_shader_in,
287 ir_var_shader_out,
288 ir_var_function_in,
289 ir_var_function_out,
290 ir_var_function_inout,
291 ir_var_const_in, /**< "in" param that must be a constant expression */
292 ir_var_system_value, /**< Ex: front-face, instance-id, etc. */
293 ir_var_temporary, /**< Temporary variable generated during compilation. */
294 ir_var_mode_count /**< Number of variable modes */
295 };
296
297 /**
298 * Enum keeping track of how a variable was declared. For error checking of
299 * the gl_PerVertex redeclaration rules.
300 */
301 enum ir_var_declaration_type {
302 /**
303 * Normal declaration (for most variables, this means an explicit
304 * declaration. Exception: temporaries are always implicitly declared, but
305 * they still use ir_var_declared_normally).
306 *
307 * Note: an ir_variable that represents a named interface block uses
308 * ir_var_declared_normally.
309 */
310 ir_var_declared_normally = 0,
311
312 /**
313 * Variable was explicitly declared (or re-declared) in an unnamed
314 * interface block.
315 */
316 ir_var_declared_in_block,
317
318 /**
319 * Variable is an implicitly declared built-in that has not been explicitly
320 * re-declared by the shader.
321 */
322 ir_var_declared_implicitly,
323 };
324
325 /**
326 * \brief Layout qualifiers for gl_FragDepth.
327 *
328 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
329 * with a layout qualifier.
330 */
331 enum ir_depth_layout {
332 ir_depth_layout_none, /**< No depth layout is specified. */
333 ir_depth_layout_any,
334 ir_depth_layout_greater,
335 ir_depth_layout_less,
336 ir_depth_layout_unchanged
337 };
338
339 /**
340 * \brief Convert depth layout qualifier to string.
341 */
342 const char*
343 depth_layout_string(ir_depth_layout layout);
344
345 /**
346 * Description of built-in state associated with a uniform
347 *
348 * \sa ir_variable::state_slots
349 */
350 struct ir_state_slot {
351 int tokens[5];
352 int swizzle;
353 };
354
355
356 /**
357 * Get the string value for an interpolation qualifier
358 *
359 * \return The string that would be used in a shader to specify \c
360 * mode will be returned.
361 *
362 * This function is used to generate error messages of the form "shader
363 * uses %s interpolation qualifier", so in the case where there is no
364 * interpolation qualifier, it returns "no".
365 *
366 * This function should only be used on a shader input or output variable.
367 */
368 const char *interpolation_string(unsigned interpolation);
369
370
371 class ir_variable : public ir_instruction {
372 public:
373 ir_variable(const struct glsl_type *, const char *, ir_variable_mode);
374
375 virtual ir_variable *clone(void *mem_ctx, struct hash_table *ht) const;
376
377 virtual ir_variable *as_variable()
378 {
379 return this;
380 }
381
382 virtual void accept(ir_visitor *v)
383 {
384 v->visit(this);
385 }
386
387 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
388
389
390 /**
391 * Determine how this variable should be interpolated based on its
392 * interpolation qualifier (if present), whether it is gl_Color or
393 * gl_SecondaryColor, and whether flatshading is enabled in the current GL
394 * state.
395 *
396 * The return value will always be either INTERP_QUALIFIER_SMOOTH,
397 * INTERP_QUALIFIER_NOPERSPECTIVE, or INTERP_QUALIFIER_FLAT.
398 */
399 glsl_interp_qualifier determine_interpolation_mode(bool flat_shade);
400
401 /**
402 * Determine whether or not a variable is part of a uniform block.
403 */
404 inline bool is_in_uniform_block() const
405 {
406 return this->data.mode == ir_var_uniform && this->interface_type != NULL;
407 }
408
409 /**
410 * Determine whether or not a variable is the declaration of an interface
411 * block
412 *
413 * For the first declaration below, there will be an \c ir_variable named
414 * "instance" whose type and whose instance_type will be the same
415 * \cglsl_type. For the second declaration, there will be an \c ir_variable
416 * named "f" whose type is float and whose instance_type is B2.
417 *
418 * "instance" is an interface instance variable, but "f" is not.
419 *
420 * uniform B1 {
421 * float f;
422 * } instance;
423 *
424 * uniform B2 {
425 * float f;
426 * };
427 */
428 inline bool is_interface_instance() const
429 {
430 const glsl_type *const t = this->type;
431
432 return (t == this->interface_type)
433 || (t->is_array() && t->fields.array == this->interface_type);
434 }
435
436 /**
437 * Set this->interface_type on a newly created variable.
438 */
439 void init_interface_type(const struct glsl_type *type)
440 {
441 assert(this->interface_type == NULL);
442 this->interface_type = type;
443 if (this->is_interface_instance()) {
444 this->max_ifc_array_access =
445 rzalloc_array(this, unsigned, type->length);
446 }
447 }
448
449 /**
450 * Change this->interface_type on a variable that previously had a
451 * different, but compatible, interface_type. This is used during linking
452 * to set the size of arrays in interface blocks.
453 */
454 void change_interface_type(const struct glsl_type *type)
455 {
456 if (this->max_ifc_array_access != NULL) {
457 /* max_ifc_array_access has already been allocated, so make sure the
458 * new interface has the same number of fields as the old one.
459 */
460 assert(this->interface_type->length == type->length);
461 }
462 this->interface_type = type;
463 }
464
465 /**
466 * Change this->interface_type on a variable that previously had a
467 * different, and incompatible, interface_type. This is used during
468 * compilation to handle redeclaration of the built-in gl_PerVertex
469 * interface block.
470 */
471 void reinit_interface_type(const struct glsl_type *type)
472 {
473 if (this->max_ifc_array_access != NULL) {
474 #ifndef _NDEBUG
475 /* Redeclaring gl_PerVertex is only allowed if none of the built-ins
476 * it defines have been accessed yet; so it's safe to throw away the
477 * old max_ifc_array_access pointer, since all of its values are
478 * zero.
479 */
480 for (unsigned i = 0; i < this->interface_type->length; i++)
481 assert(this->max_ifc_array_access[i] == 0);
482 #endif
483 ralloc_free(this->max_ifc_array_access);
484 this->max_ifc_array_access = NULL;
485 }
486 this->interface_type = NULL;
487 init_interface_type(type);
488 }
489
490 const glsl_type *get_interface_type() const
491 {
492 return this->interface_type;
493 }
494
495 /**
496 * Declared type of the variable
497 */
498 const struct glsl_type *type;
499
500 /**
501 * Declared name of the variable
502 */
503 const char *name;
504
505 /**
506 * For variables which satisfy the is_interface_instance() predicate, this
507 * points to an array of integers such that if the ith member of the
508 * interface block is an array, max_ifc_array_access[i] is the maximum
509 * array element of that member that has been accessed. If the ith member
510 * of the interface block is not an array, max_ifc_array_access[i] is
511 * unused.
512 *
513 * For variables whose type is not an interface block, this pointer is
514 * NULL.
515 */
516 unsigned *max_ifc_array_access;
517
518 struct ir_variable_data {
519
520 /**
521 * Is the variable read-only?
522 *
523 * This is set for variables declared as \c const, shader inputs,
524 * and uniforms.
525 */
526 unsigned read_only:1;
527 unsigned centroid:1;
528 unsigned sample:1;
529 unsigned invariant:1;
530
531 /**
532 * Has this variable been used for reading or writing?
533 *
534 * Several GLSL semantic checks require knowledge of whether or not a
535 * variable has been used. For example, it is an error to redeclare a
536 * variable as invariant after it has been used.
537 *
538 * This is only maintained in the ast_to_hir.cpp path, not in
539 * Mesa's fixed function or ARB program paths.
540 */
541 unsigned used:1;
542
543 /**
544 * Has this variable been statically assigned?
545 *
546 * This answers whether the variable was assigned in any path of
547 * the shader during ast_to_hir. This doesn't answer whether it is
548 * still written after dead code removal, nor is it maintained in
549 * non-ast_to_hir.cpp (GLSL parsing) paths.
550 */
551 unsigned assigned:1;
552
553 /**
554 * Enum indicating how the variable was declared. See
555 * ir_var_declaration_type.
556 *
557 * This is used to detect certain kinds of illegal variable redeclarations.
558 */
559 unsigned how_declared:2;
560
561 /**
562 * Storage class of the variable.
563 *
564 * \sa ir_variable_mode
565 */
566 unsigned mode:4;
567
568 /**
569 * Interpolation mode for shader inputs / outputs
570 *
571 * \sa ir_variable_interpolation
572 */
573 unsigned interpolation:2;
574
575 /**
576 * \name ARB_fragment_coord_conventions
577 * @{
578 */
579 unsigned origin_upper_left:1;
580 unsigned pixel_center_integer:1;
581 /*@}*/
582
583 /**
584 * Was the location explicitly set in the shader?
585 *
586 * If the location is explicitly set in the shader, it \b cannot be changed
587 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
588 * no effect).
589 */
590 unsigned explicit_location:1;
591 unsigned explicit_index:1;
592
593 /**
594 * Was an initial binding explicitly set in the shader?
595 *
596 * If so, constant_value contains an integer ir_constant representing the
597 * initial binding point.
598 */
599 unsigned explicit_binding:1;
600
601 /**
602 * Does this variable have an initializer?
603 *
604 * This is used by the linker to cross-validiate initializers of global
605 * variables.
606 */
607 unsigned has_initializer:1;
608
609 /**
610 * Is this variable a generic output or input that has not yet been matched
611 * up to a variable in another stage of the pipeline?
612 *
613 * This is used by the linker as scratch storage while assigning locations
614 * to generic inputs and outputs.
615 */
616 unsigned is_unmatched_generic_inout:1;
617
618 /**
619 * If non-zero, then this variable may be packed along with other variables
620 * into a single varying slot, so this offset should be applied when
621 * accessing components. For example, an offset of 1 means that the x
622 * component of this variable is actually stored in component y of the
623 * location specified by \c location.
624 */
625 unsigned location_frac:2;
626
627 /**
628 * Non-zero if this variable was created by lowering a named interface
629 * block which was not an array.
630 *
631 * Note that this variable and \c from_named_ifc_block_array will never
632 * both be non-zero.
633 */
634 unsigned from_named_ifc_block_nonarray:1;
635
636 /**
637 * Non-zero if this variable was created by lowering a named interface
638 * block which was an array.
639 *
640 * Note that this variable and \c from_named_ifc_block_nonarray will never
641 * both be non-zero.
642 */
643 unsigned from_named_ifc_block_array:1;
644
645 /**
646 * \brief Layout qualifier for gl_FragDepth.
647 *
648 * This is not equal to \c ir_depth_layout_none if and only if this
649 * variable is \c gl_FragDepth and a layout qualifier is specified.
650 */
651 ir_depth_layout depth_layout;
652
653 /**
654 * Storage location of the base of this variable
655 *
656 * The precise meaning of this field depends on the nature of the variable.
657 *
658 * - Vertex shader input: one of the values from \c gl_vert_attrib.
659 * - Vertex shader output: one of the values from \c gl_varying_slot.
660 * - Geometry shader input: one of the values from \c gl_varying_slot.
661 * - Geometry shader output: one of the values from \c gl_varying_slot.
662 * - Fragment shader input: one of the values from \c gl_varying_slot.
663 * - Fragment shader output: one of the values from \c gl_frag_result.
664 * - Uniforms: Per-stage uniform slot number for default uniform block.
665 * - Uniforms: Index within the uniform block definition for UBO members.
666 * - Other: This field is not currently used.
667 *
668 * If the variable is a uniform, shader input, or shader output, and the
669 * slot has not been assigned, the value will be -1.
670 */
671 int location;
672
673 /**
674 * output index for dual source blending.
675 */
676 int index;
677
678 /**
679 * Initial binding point for a sampler or UBO.
680 *
681 * For array types, this represents the binding point for the first element.
682 */
683 int binding;
684
685 /**
686 * Location an atomic counter is stored at.
687 */
688 struct {
689 unsigned buffer_index;
690 unsigned offset;
691 } atomic;
692
693 /**
694 * Highest element accessed with a constant expression array index
695 *
696 * Not used for non-array variables.
697 */
698 unsigned max_array_access;
699
700 } data;
701
702 /**
703 * Built-in state that backs this uniform
704 *
705 * Once set at variable creation, \c state_slots must remain invariant.
706 * This is because, ideally, this array would be shared by all clones of
707 * this variable in the IR tree. In other words, we'd really like for it
708 * to be a fly-weight.
709 *
710 * If the variable is not a uniform, \c num_state_slots will be zero and
711 * \c state_slots will be \c NULL.
712 */
713 /*@{*/
714 unsigned num_state_slots; /**< Number of state slots used */
715 ir_state_slot *state_slots; /**< State descriptors. */
716 /*@}*/
717
718 /**
719 * Emit a warning if this variable is accessed.
720 */
721 const char *warn_extension;
722
723 /**
724 * Value assigned in the initializer of a variable declared "const"
725 */
726 ir_constant *constant_value;
727
728 /**
729 * Constant expression assigned in the initializer of the variable
730 *
731 * \warning
732 * This field and \c ::constant_value are distinct. Even if the two fields
733 * refer to constants with the same value, they must point to separate
734 * objects.
735 */
736 ir_constant *constant_initializer;
737
738 private:
739 /**
740 * For variables that are in an interface block or are an instance of an
741 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
742 *
743 * \sa ir_variable::location
744 */
745 const glsl_type *interface_type;
746 };
747
748 /**
749 * A function that returns whether a built-in function is available in the
750 * current shading language (based on version, ES or desktop, and extensions).
751 */
752 typedef bool (*builtin_available_predicate)(const _mesa_glsl_parse_state *);
753
754 /*@{*/
755 /**
756 * The representation of a function instance; may be the full definition or
757 * simply a prototype.
758 */
759 class ir_function_signature : public ir_instruction {
760 /* An ir_function_signature will be part of the list of signatures in
761 * an ir_function.
762 */
763 public:
764 ir_function_signature(const glsl_type *return_type,
765 builtin_available_predicate builtin_avail = NULL);
766
767 virtual ir_function_signature *clone(void *mem_ctx,
768 struct hash_table *ht) const;
769 ir_function_signature *clone_prototype(void *mem_ctx,
770 struct hash_table *ht) const;
771
772 virtual void accept(ir_visitor *v)
773 {
774 v->visit(this);
775 }
776
777 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
778
779 /**
780 * Attempt to evaluate this function as a constant expression,
781 * given a list of the actual parameters and the variable context.
782 * Returns NULL for non-built-ins.
783 */
784 ir_constant *constant_expression_value(exec_list *actual_parameters, struct hash_table *variable_context);
785
786 /**
787 * Get the name of the function for which this is a signature
788 */
789 const char *function_name() const;
790
791 /**
792 * Get a handle to the function for which this is a signature
793 *
794 * There is no setter function, this function returns a \c const pointer,
795 * and \c ir_function_signature::_function is private for a reason. The
796 * only way to make a connection between a function and function signature
797 * is via \c ir_function::add_signature. This helps ensure that certain
798 * invariants (i.e., a function signature is in the list of signatures for
799 * its \c _function) are met.
800 *
801 * \sa ir_function::add_signature
802 */
803 inline const class ir_function *function() const
804 {
805 return this->_function;
806 }
807
808 /**
809 * Check whether the qualifiers match between this signature's parameters
810 * and the supplied parameter list. If not, returns the name of the first
811 * parameter with mismatched qualifiers (for use in error messages).
812 */
813 const char *qualifiers_match(exec_list *params);
814
815 /**
816 * Replace the current parameter list with the given one. This is useful
817 * if the current information came from a prototype, and either has invalid
818 * or missing parameter names.
819 */
820 void replace_parameters(exec_list *new_params);
821
822 /**
823 * Function return type.
824 *
825 * \note This discards the optional precision qualifier.
826 */
827 const struct glsl_type *return_type;
828
829 /**
830 * List of ir_variable of function parameters.
831 *
832 * This represents the storage. The paramaters passed in a particular
833 * call will be in ir_call::actual_paramaters.
834 */
835 struct exec_list parameters;
836
837 /** Whether or not this function has a body (which may be empty). */
838 unsigned is_defined:1;
839
840 /** Whether or not this function signature is a built-in. */
841 bool is_builtin() const;
842
843 /**
844 * Whether or not this function is an intrinsic to be implemented
845 * by the driver.
846 */
847 bool is_intrinsic;
848
849 /** Whether or not a built-in is available for this shader. */
850 bool is_builtin_available(const _mesa_glsl_parse_state *state) const;
851
852 /** Body of instructions in the function. */
853 struct exec_list body;
854
855 private:
856 /**
857 * A function pointer to a predicate that answers whether a built-in
858 * function is available in the current shader. NULL if not a built-in.
859 */
860 builtin_available_predicate builtin_avail;
861
862 /** Function of which this signature is one overload. */
863 class ir_function *_function;
864
865 /** Function signature of which this one is a prototype clone */
866 const ir_function_signature *origin;
867
868 friend class ir_function;
869
870 /**
871 * Helper function to run a list of instructions for constant
872 * expression evaluation.
873 *
874 * The hash table represents the values of the visible variables.
875 * There are no scoping issues because the table is indexed on
876 * ir_variable pointers, not variable names.
877 *
878 * Returns false if the expression is not constant, true otherwise,
879 * and the value in *result if result is non-NULL.
880 */
881 bool constant_expression_evaluate_expression_list(const struct exec_list &body,
882 struct hash_table *variable_context,
883 ir_constant **result);
884 };
885
886
887 /**
888 * Header for tracking multiple overloaded functions with the same name.
889 * Contains a list of ir_function_signatures representing each of the
890 * actual functions.
891 */
892 class ir_function : public ir_instruction {
893 public:
894 ir_function(const char *name);
895
896 virtual ir_function *clone(void *mem_ctx, struct hash_table *ht) const;
897
898 virtual ir_function *as_function()
899 {
900 return this;
901 }
902
903 virtual void accept(ir_visitor *v)
904 {
905 v->visit(this);
906 }
907
908 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
909
910 void add_signature(ir_function_signature *sig)
911 {
912 sig->_function = this;
913 this->signatures.push_tail(sig);
914 }
915
916 /**
917 * Find a signature that matches a set of actual parameters, taking implicit
918 * conversions into account. Also flags whether the match was exact.
919 */
920 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
921 const exec_list *actual_param,
922 bool *match_is_exact);
923
924 /**
925 * Find a signature that matches a set of actual parameters, taking implicit
926 * conversions into account.
927 */
928 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
929 const exec_list *actual_param);
930
931 /**
932 * Find a signature that exactly matches a set of actual parameters without
933 * any implicit type conversions.
934 */
935 ir_function_signature *exact_matching_signature(_mesa_glsl_parse_state *state,
936 const exec_list *actual_ps);
937
938 /**
939 * Name of the function.
940 */
941 const char *name;
942
943 /** Whether or not this function has a signature that isn't a built-in. */
944 bool has_user_signature();
945
946 /**
947 * List of ir_function_signature for each overloaded function with this name.
948 */
949 struct exec_list signatures;
950 };
951
952 inline const char *ir_function_signature::function_name() const
953 {
954 return this->_function->name;
955 }
956 /*@}*/
957
958
959 /**
960 * IR instruction representing high-level if-statements
961 */
962 class ir_if : public ir_instruction {
963 public:
964 ir_if(ir_rvalue *condition)
965 : condition(condition)
966 {
967 ir_type = ir_type_if;
968 }
969
970 virtual ir_if *clone(void *mem_ctx, struct hash_table *ht) const;
971
972 virtual ir_if *as_if()
973 {
974 return this;
975 }
976
977 virtual void accept(ir_visitor *v)
978 {
979 v->visit(this);
980 }
981
982 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
983
984 ir_rvalue *condition;
985 /** List of ir_instruction for the body of the then branch */
986 exec_list then_instructions;
987 /** List of ir_instruction for the body of the else branch */
988 exec_list else_instructions;
989 };
990
991
992 /**
993 * IR instruction representing a high-level loop structure.
994 */
995 class ir_loop : public ir_instruction {
996 public:
997 ir_loop();
998
999 virtual ir_loop *clone(void *mem_ctx, struct hash_table *ht) const;
1000
1001 virtual void accept(ir_visitor *v)
1002 {
1003 v->visit(this);
1004 }
1005
1006 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1007
1008 virtual ir_loop *as_loop()
1009 {
1010 return this;
1011 }
1012
1013 /** List of ir_instruction that make up the body of the loop. */
1014 exec_list body_instructions;
1015 };
1016
1017
1018 class ir_assignment : public ir_instruction {
1019 public:
1020 ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs, ir_rvalue *condition = NULL);
1021
1022 /**
1023 * Construct an assignment with an explicit write mask
1024 *
1025 * \note
1026 * Since a write mask is supplied, the LHS must already be a bare
1027 * \c ir_dereference. The cannot be any swizzles in the LHS.
1028 */
1029 ir_assignment(ir_dereference *lhs, ir_rvalue *rhs, ir_rvalue *condition,
1030 unsigned write_mask);
1031
1032 virtual ir_assignment *clone(void *mem_ctx, struct hash_table *ht) const;
1033
1034 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1035
1036 virtual void accept(ir_visitor *v)
1037 {
1038 v->visit(this);
1039 }
1040
1041 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1042
1043 virtual ir_assignment * as_assignment()
1044 {
1045 return this;
1046 }
1047
1048 /**
1049 * Get a whole variable written by an assignment
1050 *
1051 * If the LHS of the assignment writes a whole variable, the variable is
1052 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1053 * assignment are:
1054 *
1055 * - Assigning to a scalar
1056 * - Assigning to all components of a vector
1057 * - Whole array (or matrix) assignment
1058 * - Whole structure assignment
1059 */
1060 ir_variable *whole_variable_written();
1061
1062 /**
1063 * Set the LHS of an assignment
1064 */
1065 void set_lhs(ir_rvalue *lhs);
1066
1067 /**
1068 * Left-hand side of the assignment.
1069 *
1070 * This should be treated as read only. If you need to set the LHS of an
1071 * assignment, use \c ir_assignment::set_lhs.
1072 */
1073 ir_dereference *lhs;
1074
1075 /**
1076 * Value being assigned
1077 */
1078 ir_rvalue *rhs;
1079
1080 /**
1081 * Optional condition for the assignment.
1082 */
1083 ir_rvalue *condition;
1084
1085
1086 /**
1087 * Component mask written
1088 *
1089 * For non-vector types in the LHS, this field will be zero. For vector
1090 * types, a bit will be set for each component that is written. Note that
1091 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1092 *
1093 * A partially-set write mask means that each enabled channel gets
1094 * the value from a consecutive channel of the rhs. For example,
1095 * to write just .xyw of gl_FrontColor with color:
1096 *
1097 * (assign (constant bool (1)) (xyw)
1098 * (var_ref gl_FragColor)
1099 * (swiz xyw (var_ref color)))
1100 */
1101 unsigned write_mask:4;
1102 };
1103
1104 /* Update ir_expression::get_num_operands() and operator_strs when
1105 * updating this list.
1106 */
1107 enum ir_expression_operation {
1108 ir_unop_bit_not,
1109 ir_unop_logic_not,
1110 ir_unop_neg,
1111 ir_unop_abs,
1112 ir_unop_sign,
1113 ir_unop_rcp,
1114 ir_unop_rsq,
1115 ir_unop_sqrt,
1116 ir_unop_exp, /**< Log base e on gentype */
1117 ir_unop_log, /**< Natural log on gentype */
1118 ir_unop_exp2,
1119 ir_unop_log2,
1120 ir_unop_f2i, /**< Float-to-integer conversion. */
1121 ir_unop_f2u, /**< Float-to-unsigned conversion. */
1122 ir_unop_i2f, /**< Integer-to-float conversion. */
1123 ir_unop_f2b, /**< Float-to-boolean conversion */
1124 ir_unop_b2f, /**< Boolean-to-float conversion */
1125 ir_unop_i2b, /**< int-to-boolean conversion */
1126 ir_unop_b2i, /**< Boolean-to-int conversion */
1127 ir_unop_u2f, /**< Unsigned-to-float conversion. */
1128 ir_unop_i2u, /**< Integer-to-unsigned conversion. */
1129 ir_unop_u2i, /**< Unsigned-to-integer conversion. */
1130 ir_unop_bitcast_i2f, /**< Bit-identical int-to-float "conversion" */
1131 ir_unop_bitcast_f2i, /**< Bit-identical float-to-int "conversion" */
1132 ir_unop_bitcast_u2f, /**< Bit-identical uint-to-float "conversion" */
1133 ir_unop_bitcast_f2u, /**< Bit-identical float-to-uint "conversion" */
1134 ir_unop_any,
1135
1136 /**
1137 * \name Unary floating-point rounding operations.
1138 */
1139 /*@{*/
1140 ir_unop_trunc,
1141 ir_unop_ceil,
1142 ir_unop_floor,
1143 ir_unop_fract,
1144 ir_unop_round_even,
1145 /*@}*/
1146
1147 /**
1148 * \name Trigonometric operations.
1149 */
1150 /*@{*/
1151 ir_unop_sin,
1152 ir_unop_cos,
1153 ir_unop_sin_reduced, /**< Reduced range sin. [-pi, pi] */
1154 ir_unop_cos_reduced, /**< Reduced range cos. [-pi, pi] */
1155 /*@}*/
1156
1157 /**
1158 * \name Partial derivatives.
1159 */
1160 /*@{*/
1161 ir_unop_dFdx,
1162 ir_unop_dFdy,
1163 /*@}*/
1164
1165 /**
1166 * \name Floating point pack and unpack operations.
1167 */
1168 /*@{*/
1169 ir_unop_pack_snorm_2x16,
1170 ir_unop_pack_snorm_4x8,
1171 ir_unop_pack_unorm_2x16,
1172 ir_unop_pack_unorm_4x8,
1173 ir_unop_pack_half_2x16,
1174 ir_unop_unpack_snorm_2x16,
1175 ir_unop_unpack_snorm_4x8,
1176 ir_unop_unpack_unorm_2x16,
1177 ir_unop_unpack_unorm_4x8,
1178 ir_unop_unpack_half_2x16,
1179 /*@}*/
1180
1181 /**
1182 * \name Lowered floating point unpacking operations.
1183 *
1184 * \see lower_packing_builtins_visitor::split_unpack_half_2x16
1185 */
1186 /*@{*/
1187 ir_unop_unpack_half_2x16_split_x,
1188 ir_unop_unpack_half_2x16_split_y,
1189 /*@}*/
1190
1191 /**
1192 * \name Bit operations, part of ARB_gpu_shader5.
1193 */
1194 /*@{*/
1195 ir_unop_bitfield_reverse,
1196 ir_unop_bit_count,
1197 ir_unop_find_msb,
1198 ir_unop_find_lsb,
1199 /*@}*/
1200
1201 ir_unop_noise,
1202
1203 /**
1204 * A sentinel marking the last of the unary operations.
1205 */
1206 ir_last_unop = ir_unop_noise,
1207
1208 ir_binop_add,
1209 ir_binop_sub,
1210 ir_binop_mul, /**< Floating-point or low 32-bit integer multiply. */
1211 ir_binop_imul_high, /**< Calculates the high 32-bits of a 64-bit multiply. */
1212 ir_binop_div,
1213
1214 /**
1215 * Returns the carry resulting from the addition of the two arguments.
1216 */
1217 /*@{*/
1218 ir_binop_carry,
1219 /*@}*/
1220
1221 /**
1222 * Returns the borrow resulting from the subtraction of the second argument
1223 * from the first argument.
1224 */
1225 /*@{*/
1226 ir_binop_borrow,
1227 /*@}*/
1228
1229 /**
1230 * Takes one of two combinations of arguments:
1231 *
1232 * - mod(vecN, vecN)
1233 * - mod(vecN, float)
1234 *
1235 * Does not take integer types.
1236 */
1237 ir_binop_mod,
1238
1239 /**
1240 * \name Binary comparison operators which return a boolean vector.
1241 * The type of both operands must be equal.
1242 */
1243 /*@{*/
1244 ir_binop_less,
1245 ir_binop_greater,
1246 ir_binop_lequal,
1247 ir_binop_gequal,
1248 ir_binop_equal,
1249 ir_binop_nequal,
1250 /**
1251 * Returns single boolean for whether all components of operands[0]
1252 * equal the components of operands[1].
1253 */
1254 ir_binop_all_equal,
1255 /**
1256 * Returns single boolean for whether any component of operands[0]
1257 * is not equal to the corresponding component of operands[1].
1258 */
1259 ir_binop_any_nequal,
1260 /*@}*/
1261
1262 /**
1263 * \name Bit-wise binary operations.
1264 */
1265 /*@{*/
1266 ir_binop_lshift,
1267 ir_binop_rshift,
1268 ir_binop_bit_and,
1269 ir_binop_bit_xor,
1270 ir_binop_bit_or,
1271 /*@}*/
1272
1273 ir_binop_logic_and,
1274 ir_binop_logic_xor,
1275 ir_binop_logic_or,
1276
1277 ir_binop_dot,
1278 ir_binop_min,
1279 ir_binop_max,
1280
1281 ir_binop_pow,
1282
1283 /**
1284 * \name Lowered floating point packing operations.
1285 *
1286 * \see lower_packing_builtins_visitor::split_pack_half_2x16
1287 */
1288 /*@{*/
1289 ir_binop_pack_half_2x16_split,
1290 /*@}*/
1291
1292 /**
1293 * \name First half of a lowered bitfieldInsert() operation.
1294 *
1295 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1296 */
1297 /*@{*/
1298 ir_binop_bfm,
1299 /*@}*/
1300
1301 /**
1302 * Load a value the size of a given GLSL type from a uniform block.
1303 *
1304 * operand0 is the ir_constant uniform block index in the linked shader.
1305 * operand1 is a byte offset within the uniform block.
1306 */
1307 ir_binop_ubo_load,
1308
1309 /**
1310 * \name Multiplies a number by two to a power, part of ARB_gpu_shader5.
1311 */
1312 /*@{*/
1313 ir_binop_ldexp,
1314 /*@}*/
1315
1316 /**
1317 * Extract a scalar from a vector
1318 *
1319 * operand0 is the vector
1320 * operand1 is the index of the field to read from operand0
1321 */
1322 ir_binop_vector_extract,
1323
1324 /**
1325 * A sentinel marking the last of the binary operations.
1326 */
1327 ir_last_binop = ir_binop_vector_extract,
1328
1329 /**
1330 * \name Fused floating-point multiply-add, part of ARB_gpu_shader5.
1331 */
1332 /*@{*/
1333 ir_triop_fma,
1334 /*@}*/
1335
1336 ir_triop_lrp,
1337
1338 /**
1339 * \name Conditional Select
1340 *
1341 * A vector conditional select instruction (like ?:, but operating per-
1342 * component on vectors).
1343 *
1344 * \see lower_instructions_visitor::ldexp_to_arith
1345 */
1346 /*@{*/
1347 ir_triop_csel,
1348 /*@}*/
1349
1350 /**
1351 * \name Second half of a lowered bitfieldInsert() operation.
1352 *
1353 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1354 */
1355 /*@{*/
1356 ir_triop_bfi,
1357 /*@}*/
1358
1359 ir_triop_bitfield_extract,
1360
1361 /**
1362 * Generate a value with one field of a vector changed
1363 *
1364 * operand0 is the vector
1365 * operand1 is the value to write into the vector result
1366 * operand2 is the index in operand0 to be modified
1367 */
1368 ir_triop_vector_insert,
1369
1370 /**
1371 * A sentinel marking the last of the ternary operations.
1372 */
1373 ir_last_triop = ir_triop_vector_insert,
1374
1375 ir_quadop_bitfield_insert,
1376
1377 ir_quadop_vector,
1378
1379 /**
1380 * A sentinel marking the last of the ternary operations.
1381 */
1382 ir_last_quadop = ir_quadop_vector,
1383
1384 /**
1385 * A sentinel marking the last of all operations.
1386 */
1387 ir_last_opcode = ir_quadop_vector
1388 };
1389
1390 class ir_expression : public ir_rvalue {
1391 public:
1392 ir_expression(int op, const struct glsl_type *type,
1393 ir_rvalue *op0, ir_rvalue *op1 = NULL,
1394 ir_rvalue *op2 = NULL, ir_rvalue *op3 = NULL);
1395
1396 /**
1397 * Constructor for unary operation expressions
1398 */
1399 ir_expression(int op, ir_rvalue *);
1400
1401 /**
1402 * Constructor for binary operation expressions
1403 */
1404 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1);
1405
1406 /**
1407 * Constructor for ternary operation expressions
1408 */
1409 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1, ir_rvalue *op2);
1410
1411 virtual ir_expression *as_expression()
1412 {
1413 return this;
1414 }
1415
1416 virtual bool equals(ir_instruction *ir, enum ir_node_type ignore = ir_type_unset);
1417
1418 virtual ir_expression *clone(void *mem_ctx, struct hash_table *ht) const;
1419
1420 /**
1421 * Attempt to constant-fold the expression
1422 *
1423 * The "variable_context" hash table links ir_variable * to ir_constant *
1424 * that represent the variables' values. \c NULL represents an empty
1425 * context.
1426 *
1427 * If the expression cannot be constant folded, this method will return
1428 * \c NULL.
1429 */
1430 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1431
1432 /**
1433 * Determine the number of operands used by an expression
1434 */
1435 static unsigned int get_num_operands(ir_expression_operation);
1436
1437 /**
1438 * Determine the number of operands used by an expression
1439 */
1440 unsigned int get_num_operands() const
1441 {
1442 return (this->operation == ir_quadop_vector)
1443 ? this->type->vector_elements : get_num_operands(operation);
1444 }
1445
1446 /**
1447 * Return a string representing this expression's operator.
1448 */
1449 const char *operator_string();
1450
1451 /**
1452 * Return a string representing this expression's operator.
1453 */
1454 static const char *operator_string(ir_expression_operation);
1455
1456
1457 /**
1458 * Do a reverse-lookup to translate the given string into an operator.
1459 */
1460 static ir_expression_operation get_operator(const char *);
1461
1462 virtual void accept(ir_visitor *v)
1463 {
1464 v->visit(this);
1465 }
1466
1467 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1468
1469 ir_expression_operation operation;
1470 ir_rvalue *operands[4];
1471 };
1472
1473
1474 /**
1475 * HIR instruction representing a high-level function call, containing a list
1476 * of parameters and returning a value in the supplied temporary.
1477 */
1478 class ir_call : public ir_instruction {
1479 public:
1480 ir_call(ir_function_signature *callee,
1481 ir_dereference_variable *return_deref,
1482 exec_list *actual_parameters)
1483 : return_deref(return_deref), callee(callee)
1484 {
1485 ir_type = ir_type_call;
1486 assert(callee->return_type != NULL);
1487 actual_parameters->move_nodes_to(& this->actual_parameters);
1488 this->use_builtin = callee->is_builtin();
1489 }
1490
1491 virtual ir_call *clone(void *mem_ctx, struct hash_table *ht) const;
1492
1493 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1494
1495 virtual ir_call *as_call()
1496 {
1497 return this;
1498 }
1499
1500 virtual void accept(ir_visitor *v)
1501 {
1502 v->visit(this);
1503 }
1504
1505 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1506
1507 /**
1508 * Get the name of the function being called.
1509 */
1510 const char *callee_name() const
1511 {
1512 return callee->function_name();
1513 }
1514
1515 /**
1516 * Generates an inline version of the function before @ir,
1517 * storing the return value in return_deref.
1518 */
1519 void generate_inline(ir_instruction *ir);
1520
1521 /**
1522 * Storage for the function's return value.
1523 * This must be NULL if the return type is void.
1524 */
1525 ir_dereference_variable *return_deref;
1526
1527 /**
1528 * The specific function signature being called.
1529 */
1530 ir_function_signature *callee;
1531
1532 /* List of ir_rvalue of paramaters passed in this call. */
1533 exec_list actual_parameters;
1534
1535 /** Should this call only bind to a built-in function? */
1536 bool use_builtin;
1537 };
1538
1539
1540 /**
1541 * \name Jump-like IR instructions.
1542 *
1543 * These include \c break, \c continue, \c return, and \c discard.
1544 */
1545 /*@{*/
1546 class ir_jump : public ir_instruction {
1547 protected:
1548 ir_jump()
1549 {
1550 ir_type = ir_type_unset;
1551 }
1552
1553 public:
1554 virtual ir_jump *as_jump()
1555 {
1556 return this;
1557 }
1558 };
1559
1560 class ir_return : public ir_jump {
1561 public:
1562 ir_return()
1563 : value(NULL)
1564 {
1565 this->ir_type = ir_type_return;
1566 }
1567
1568 ir_return(ir_rvalue *value)
1569 : value(value)
1570 {
1571 this->ir_type = ir_type_return;
1572 }
1573
1574 virtual ir_return *clone(void *mem_ctx, struct hash_table *) const;
1575
1576 virtual ir_return *as_return()
1577 {
1578 return this;
1579 }
1580
1581 ir_rvalue *get_value() const
1582 {
1583 return value;
1584 }
1585
1586 virtual void accept(ir_visitor *v)
1587 {
1588 v->visit(this);
1589 }
1590
1591 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1592
1593 ir_rvalue *value;
1594 };
1595
1596
1597 /**
1598 * Jump instructions used inside loops
1599 *
1600 * These include \c break and \c continue. The \c break within a loop is
1601 * different from the \c break within a switch-statement.
1602 *
1603 * \sa ir_switch_jump
1604 */
1605 class ir_loop_jump : public ir_jump {
1606 public:
1607 enum jump_mode {
1608 jump_break,
1609 jump_continue
1610 };
1611
1612 ir_loop_jump(jump_mode mode)
1613 {
1614 this->ir_type = ir_type_loop_jump;
1615 this->mode = mode;
1616 }
1617
1618 virtual ir_loop_jump *clone(void *mem_ctx, struct hash_table *) const;
1619
1620 virtual void accept(ir_visitor *v)
1621 {
1622 v->visit(this);
1623 }
1624
1625 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1626
1627 bool is_break() const
1628 {
1629 return mode == jump_break;
1630 }
1631
1632 bool is_continue() const
1633 {
1634 return mode == jump_continue;
1635 }
1636
1637 /** Mode selector for the jump instruction. */
1638 enum jump_mode mode;
1639 };
1640
1641 /**
1642 * IR instruction representing discard statements.
1643 */
1644 class ir_discard : public ir_jump {
1645 public:
1646 ir_discard()
1647 {
1648 this->ir_type = ir_type_discard;
1649 this->condition = NULL;
1650 }
1651
1652 ir_discard(ir_rvalue *cond)
1653 {
1654 this->ir_type = ir_type_discard;
1655 this->condition = cond;
1656 }
1657
1658 virtual ir_discard *clone(void *mem_ctx, struct hash_table *ht) const;
1659
1660 virtual void accept(ir_visitor *v)
1661 {
1662 v->visit(this);
1663 }
1664
1665 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1666
1667 virtual ir_discard *as_discard()
1668 {
1669 return this;
1670 }
1671
1672 ir_rvalue *condition;
1673 };
1674 /*@}*/
1675
1676
1677 /**
1678 * Texture sampling opcodes used in ir_texture
1679 */
1680 enum ir_texture_opcode {
1681 ir_tex, /**< Regular texture look-up */
1682 ir_txb, /**< Texture look-up with LOD bias */
1683 ir_txl, /**< Texture look-up with explicit LOD */
1684 ir_txd, /**< Texture look-up with partial derivatvies */
1685 ir_txf, /**< Texel fetch with explicit LOD */
1686 ir_txf_ms, /**< Multisample texture fetch */
1687 ir_txs, /**< Texture size */
1688 ir_lod, /**< Texture lod query */
1689 ir_tg4, /**< Texture gather */
1690 ir_query_levels /**< Texture levels query */
1691 };
1692
1693
1694 /**
1695 * IR instruction to sample a texture
1696 *
1697 * The specific form of the IR instruction depends on the \c mode value
1698 * selected from \c ir_texture_opcodes. In the printed IR, these will
1699 * appear as:
1700 *
1701 * Texel offset (0 or an expression)
1702 * | Projection divisor
1703 * | | Shadow comparitor
1704 * | | |
1705 * v v v
1706 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1707 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1708 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1709 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1710 * (txf <type> <sampler> <coordinate> 0 <lod>)
1711 * (txf_ms
1712 * <type> <sampler> <coordinate> <sample_index>)
1713 * (txs <type> <sampler> <lod>)
1714 * (lod <type> <sampler> <coordinate>)
1715 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
1716 * (query_levels <type> <sampler>)
1717 */
1718 class ir_texture : public ir_rvalue {
1719 public:
1720 ir_texture(enum ir_texture_opcode op)
1721 : op(op), sampler(NULL), coordinate(NULL), projector(NULL),
1722 shadow_comparitor(NULL), offset(NULL)
1723 {
1724 this->ir_type = ir_type_texture;
1725 memset(&lod_info, 0, sizeof(lod_info));
1726 }
1727
1728 virtual ir_texture *clone(void *mem_ctx, struct hash_table *) const;
1729
1730 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1731
1732 virtual void accept(ir_visitor *v)
1733 {
1734 v->visit(this);
1735 }
1736
1737 virtual ir_texture *as_texture()
1738 {
1739 return this;
1740 }
1741
1742 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1743
1744 virtual bool equals(ir_instruction *ir, enum ir_node_type ignore = ir_type_unset);
1745
1746 /**
1747 * Return a string representing the ir_texture_opcode.
1748 */
1749 const char *opcode_string();
1750
1751 /** Set the sampler and type. */
1752 void set_sampler(ir_dereference *sampler, const glsl_type *type);
1753
1754 /**
1755 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1756 */
1757 static ir_texture_opcode get_opcode(const char *);
1758
1759 enum ir_texture_opcode op;
1760
1761 /** Sampler to use for the texture access. */
1762 ir_dereference *sampler;
1763
1764 /** Texture coordinate to sample */
1765 ir_rvalue *coordinate;
1766
1767 /**
1768 * Value used for projective divide.
1769 *
1770 * If there is no projective divide (the common case), this will be
1771 * \c NULL. Optimization passes should check for this to point to a constant
1772 * of 1.0 and replace that with \c NULL.
1773 */
1774 ir_rvalue *projector;
1775
1776 /**
1777 * Coordinate used for comparison on shadow look-ups.
1778 *
1779 * If there is no shadow comparison, this will be \c NULL. For the
1780 * \c ir_txf opcode, this *must* be \c NULL.
1781 */
1782 ir_rvalue *shadow_comparitor;
1783
1784 /** Texel offset. */
1785 ir_rvalue *offset;
1786
1787 union {
1788 ir_rvalue *lod; /**< Floating point LOD */
1789 ir_rvalue *bias; /**< Floating point LOD bias */
1790 ir_rvalue *sample_index; /**< MSAA sample index */
1791 ir_rvalue *component; /**< Gather component selector */
1792 struct {
1793 ir_rvalue *dPdx; /**< Partial derivative of coordinate wrt X */
1794 ir_rvalue *dPdy; /**< Partial derivative of coordinate wrt Y */
1795 } grad;
1796 } lod_info;
1797 };
1798
1799
1800 struct ir_swizzle_mask {
1801 unsigned x:2;
1802 unsigned y:2;
1803 unsigned z:2;
1804 unsigned w:2;
1805
1806 /**
1807 * Number of components in the swizzle.
1808 */
1809 unsigned num_components:3;
1810
1811 /**
1812 * Does the swizzle contain duplicate components?
1813 *
1814 * L-value swizzles cannot contain duplicate components.
1815 */
1816 unsigned has_duplicates:1;
1817 };
1818
1819
1820 class ir_swizzle : public ir_rvalue {
1821 public:
1822 ir_swizzle(ir_rvalue *, unsigned x, unsigned y, unsigned z, unsigned w,
1823 unsigned count);
1824
1825 ir_swizzle(ir_rvalue *val, const unsigned *components, unsigned count);
1826
1827 ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask);
1828
1829 virtual ir_swizzle *clone(void *mem_ctx, struct hash_table *) const;
1830
1831 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1832
1833 virtual ir_swizzle *as_swizzle()
1834 {
1835 return this;
1836 }
1837
1838 /**
1839 * Construct an ir_swizzle from the textual representation. Can fail.
1840 */
1841 static ir_swizzle *create(ir_rvalue *, const char *, unsigned vector_length);
1842
1843 virtual void accept(ir_visitor *v)
1844 {
1845 v->visit(this);
1846 }
1847
1848 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1849
1850 virtual bool equals(ir_instruction *ir, enum ir_node_type ignore = ir_type_unset);
1851
1852 bool is_lvalue() const
1853 {
1854 return val->is_lvalue() && !mask.has_duplicates;
1855 }
1856
1857 /**
1858 * Get the variable that is ultimately referenced by an r-value
1859 */
1860 virtual ir_variable *variable_referenced() const;
1861
1862 ir_rvalue *val;
1863 ir_swizzle_mask mask;
1864
1865 private:
1866 /**
1867 * Initialize the mask component of a swizzle
1868 *
1869 * This is used by the \c ir_swizzle constructors.
1870 */
1871 void init_mask(const unsigned *components, unsigned count);
1872 };
1873
1874
1875 class ir_dereference : public ir_rvalue {
1876 public:
1877 virtual ir_dereference *clone(void *mem_ctx, struct hash_table *) const = 0;
1878
1879 virtual ir_dereference *as_dereference()
1880 {
1881 return this;
1882 }
1883
1884 bool is_lvalue() const;
1885
1886 /**
1887 * Get the variable that is ultimately referenced by an r-value
1888 */
1889 virtual ir_variable *variable_referenced() const = 0;
1890
1891 /**
1892 * Get the constant that is ultimately referenced by an r-value,
1893 * in a constant expression evaluation context.
1894 *
1895 * The offset is used when the reference is to a specific column of
1896 * a matrix.
1897 */
1898 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const = 0;
1899 };
1900
1901
1902 class ir_dereference_variable : public ir_dereference {
1903 public:
1904 ir_dereference_variable(ir_variable *var);
1905
1906 virtual ir_dereference_variable *clone(void *mem_ctx,
1907 struct hash_table *) const;
1908
1909 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1910
1911 virtual ir_dereference_variable *as_dereference_variable()
1912 {
1913 return this;
1914 }
1915
1916 virtual bool equals(ir_instruction *ir, enum ir_node_type ignore = ir_type_unset);
1917
1918 /**
1919 * Get the variable that is ultimately referenced by an r-value
1920 */
1921 virtual ir_variable *variable_referenced() const
1922 {
1923 return this->var;
1924 }
1925
1926 /**
1927 * Get the constant that is ultimately referenced by an r-value,
1928 * in a constant expression evaluation context.
1929 *
1930 * The offset is used when the reference is to a specific column of
1931 * a matrix.
1932 */
1933 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
1934
1935 virtual ir_variable *whole_variable_referenced()
1936 {
1937 /* ir_dereference_variable objects always dereference the entire
1938 * variable. However, if this dereference is dereferenced by anything
1939 * else, the complete deferefernce chain is not a whole-variable
1940 * dereference. This method should only be called on the top most
1941 * ir_rvalue in a dereference chain.
1942 */
1943 return this->var;
1944 }
1945
1946 virtual void accept(ir_visitor *v)
1947 {
1948 v->visit(this);
1949 }
1950
1951 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1952
1953 /**
1954 * Object being dereferenced.
1955 */
1956 ir_variable *var;
1957 };
1958
1959
1960 class ir_dereference_array : public ir_dereference {
1961 public:
1962 ir_dereference_array(ir_rvalue *value, ir_rvalue *array_index);
1963
1964 ir_dereference_array(ir_variable *var, ir_rvalue *array_index);
1965
1966 virtual ir_dereference_array *clone(void *mem_ctx,
1967 struct hash_table *) const;
1968
1969 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1970
1971 virtual ir_dereference_array *as_dereference_array()
1972 {
1973 return this;
1974 }
1975
1976 virtual bool equals(ir_instruction *ir, enum ir_node_type ignore = ir_type_unset);
1977
1978 /**
1979 * Get the variable that is ultimately referenced by an r-value
1980 */
1981 virtual ir_variable *variable_referenced() const
1982 {
1983 return this->array->variable_referenced();
1984 }
1985
1986 /**
1987 * Get the constant that is ultimately referenced by an r-value,
1988 * in a constant expression evaluation context.
1989 *
1990 * The offset is used when the reference is to a specific column of
1991 * a matrix.
1992 */
1993 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
1994
1995 virtual void accept(ir_visitor *v)
1996 {
1997 v->visit(this);
1998 }
1999
2000 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2001
2002 ir_rvalue *array;
2003 ir_rvalue *array_index;
2004
2005 private:
2006 void set_array(ir_rvalue *value);
2007 };
2008
2009
2010 class ir_dereference_record : public ir_dereference {
2011 public:
2012 ir_dereference_record(ir_rvalue *value, const char *field);
2013
2014 ir_dereference_record(ir_variable *var, const char *field);
2015
2016 virtual ir_dereference_record *clone(void *mem_ctx,
2017 struct hash_table *) const;
2018
2019 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
2020
2021 virtual ir_dereference_record *as_dereference_record()
2022 {
2023 return this;
2024 }
2025
2026 /**
2027 * Get the variable that is ultimately referenced by an r-value
2028 */
2029 virtual ir_variable *variable_referenced() const
2030 {
2031 return this->record->variable_referenced();
2032 }
2033
2034 /**
2035 * Get the constant that is ultimately referenced by an r-value,
2036 * in a constant expression evaluation context.
2037 *
2038 * The offset is used when the reference is to a specific column of
2039 * a matrix.
2040 */
2041 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
2042
2043 virtual void accept(ir_visitor *v)
2044 {
2045 v->visit(this);
2046 }
2047
2048 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2049
2050 ir_rvalue *record;
2051 const char *field;
2052 };
2053
2054
2055 /**
2056 * Data stored in an ir_constant
2057 */
2058 union ir_constant_data {
2059 unsigned u[16];
2060 int i[16];
2061 float f[16];
2062 bool b[16];
2063 };
2064
2065
2066 class ir_constant : public ir_rvalue {
2067 public:
2068 ir_constant(const struct glsl_type *type, const ir_constant_data *data);
2069 ir_constant(bool b, unsigned vector_elements=1);
2070 ir_constant(unsigned int u, unsigned vector_elements=1);
2071 ir_constant(int i, unsigned vector_elements=1);
2072 ir_constant(float f, unsigned vector_elements=1);
2073
2074 /**
2075 * Construct an ir_constant from a list of ir_constant values
2076 */
2077 ir_constant(const struct glsl_type *type, exec_list *values);
2078
2079 /**
2080 * Construct an ir_constant from a scalar component of another ir_constant
2081 *
2082 * The new \c ir_constant inherits the type of the component from the
2083 * source constant.
2084 *
2085 * \note
2086 * In the case of a matrix constant, the new constant is a scalar, \b not
2087 * a vector.
2088 */
2089 ir_constant(const ir_constant *c, unsigned i);
2090
2091 /**
2092 * Return a new ir_constant of the specified type containing all zeros.
2093 */
2094 static ir_constant *zero(void *mem_ctx, const glsl_type *type);
2095
2096 virtual ir_constant *clone(void *mem_ctx, struct hash_table *) const;
2097
2098 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
2099
2100 virtual ir_constant *as_constant()
2101 {
2102 return this;
2103 }
2104
2105 virtual void accept(ir_visitor *v)
2106 {
2107 v->visit(this);
2108 }
2109
2110 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2111
2112 virtual bool equals(ir_instruction *ir, enum ir_node_type ignore = ir_type_unset);
2113
2114 /**
2115 * Get a particular component of a constant as a specific type
2116 *
2117 * This is useful, for example, to get a value from an integer constant
2118 * as a float or bool. This appears frequently when constructors are
2119 * called with all constant parameters.
2120 */
2121 /*@{*/
2122 bool get_bool_component(unsigned i) const;
2123 float get_float_component(unsigned i) const;
2124 int get_int_component(unsigned i) const;
2125 unsigned get_uint_component(unsigned i) const;
2126 /*@}*/
2127
2128 ir_constant *get_array_element(unsigned i) const;
2129
2130 ir_constant *get_record_field(const char *name);
2131
2132 /**
2133 * Copy the values on another constant at a given offset.
2134 *
2135 * The offset is ignored for array or struct copies, it's only for
2136 * scalars or vectors into vectors or matrices.
2137 *
2138 * With identical types on both sides and zero offset it's clone()
2139 * without creating a new object.
2140 */
2141
2142 void copy_offset(ir_constant *src, int offset);
2143
2144 /**
2145 * Copy the values on another constant at a given offset and
2146 * following an assign-like mask.
2147 *
2148 * The mask is ignored for scalars.
2149 *
2150 * Note that this function only handles what assign can handle,
2151 * i.e. at most a vector as source and a column of a matrix as
2152 * destination.
2153 */
2154
2155 void copy_masked_offset(ir_constant *src, int offset, unsigned int mask);
2156
2157 /**
2158 * Determine whether a constant has the same value as another constant
2159 *
2160 * \sa ir_constant::is_zero, ir_constant::is_one,
2161 * ir_constant::is_negative_one, ir_constant::is_basis
2162 */
2163 bool has_value(const ir_constant *) const;
2164
2165 /**
2166 * Return true if this ir_constant represents the given value.
2167 *
2168 * For vectors, this checks that each component is the given value.
2169 */
2170 virtual bool is_value(float f, int i) const;
2171 virtual bool is_zero() const;
2172 virtual bool is_one() const;
2173 virtual bool is_negative_one() const;
2174 virtual bool is_basis() const;
2175
2176 /**
2177 * Value of the constant.
2178 *
2179 * The field used to back the values supplied by the constant is determined
2180 * by the type associated with the \c ir_instruction. Constants may be
2181 * scalars, vectors, or matrices.
2182 */
2183 union ir_constant_data value;
2184
2185 /* Array elements */
2186 ir_constant **array_elements;
2187
2188 /* Structure fields */
2189 exec_list components;
2190
2191 private:
2192 /**
2193 * Parameterless constructor only used by the clone method
2194 */
2195 ir_constant(void);
2196 };
2197
2198 /*@}*/
2199
2200 /**
2201 * IR instruction to emit a vertex in a geometry shader.
2202 */
2203 class ir_emit_vertex : public ir_instruction {
2204 public:
2205 ir_emit_vertex()
2206 {
2207 ir_type = ir_type_emit_vertex;
2208 }
2209
2210 virtual void accept(ir_visitor *v)
2211 {
2212 v->visit(this);
2213 }
2214
2215 virtual ir_emit_vertex *clone(void *mem_ctx, struct hash_table *) const
2216 {
2217 return new(mem_ctx) ir_emit_vertex();
2218 }
2219
2220 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2221 };
2222
2223 /**
2224 * IR instruction to complete the current primitive and start a new one in a
2225 * geometry shader.
2226 */
2227 class ir_end_primitive : public ir_instruction {
2228 public:
2229 ir_end_primitive()
2230 {
2231 ir_type = ir_type_end_primitive;
2232 }
2233
2234 virtual void accept(ir_visitor *v)
2235 {
2236 v->visit(this);
2237 }
2238
2239 virtual ir_end_primitive *clone(void *mem_ctx, struct hash_table *) const
2240 {
2241 return new(mem_ctx) ir_end_primitive();
2242 }
2243
2244 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2245 };
2246
2247 /**
2248 * Apply a visitor to each IR node in a list
2249 */
2250 void
2251 visit_exec_list(exec_list *list, ir_visitor *visitor);
2252
2253 /**
2254 * Validate invariants on each IR node in a list
2255 */
2256 void validate_ir_tree(exec_list *instructions);
2257
2258 struct _mesa_glsl_parse_state;
2259 struct gl_shader_program;
2260
2261 /**
2262 * Detect whether an unlinked shader contains static recursion
2263 *
2264 * If the list of instructions is determined to contain static recursion,
2265 * \c _mesa_glsl_error will be called to emit error messages for each function
2266 * that is in the recursion cycle.
2267 */
2268 void
2269 detect_recursion_unlinked(struct _mesa_glsl_parse_state *state,
2270 exec_list *instructions);
2271
2272 /**
2273 * Detect whether a linked shader contains static recursion
2274 *
2275 * If the list of instructions is determined to contain static recursion,
2276 * \c link_error_printf will be called to emit error messages for each function
2277 * that is in the recursion cycle. In addition,
2278 * \c gl_shader_program::LinkStatus will be set to false.
2279 */
2280 void
2281 detect_recursion_linked(struct gl_shader_program *prog,
2282 exec_list *instructions);
2283
2284 /**
2285 * Make a clone of each IR instruction in a list
2286 *
2287 * \param in List of IR instructions that are to be cloned
2288 * \param out List to hold the cloned instructions
2289 */
2290 void
2291 clone_ir_list(void *mem_ctx, exec_list *out, const exec_list *in);
2292
2293 extern void
2294 _mesa_glsl_initialize_variables(exec_list *instructions,
2295 struct _mesa_glsl_parse_state *state);
2296
2297 extern void
2298 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state *state);
2299
2300 extern void
2301 _mesa_glsl_initialize_builtin_functions();
2302
2303 extern ir_function_signature *
2304 _mesa_glsl_find_builtin_function(_mesa_glsl_parse_state *state,
2305 const char *name, exec_list *actual_parameters);
2306
2307 extern gl_shader *
2308 _mesa_glsl_get_builtin_function_shader(void);
2309
2310 extern void
2311 _mesa_glsl_release_functions(void);
2312
2313 extern void
2314 _mesa_glsl_release_builtin_functions(void);
2315
2316 extern void
2317 reparent_ir(exec_list *list, void *mem_ctx);
2318
2319 struct glsl_symbol_table;
2320
2321 extern void
2322 import_prototypes(const exec_list *source, exec_list *dest,
2323 struct glsl_symbol_table *symbols, void *mem_ctx);
2324
2325 extern bool
2326 ir_has_call(ir_instruction *ir);
2327
2328 extern void
2329 do_set_program_inouts(exec_list *instructions, struct gl_program *prog,
2330 gl_shader_stage shader_stage);
2331
2332 extern char *
2333 prototype_string(const glsl_type *return_type, const char *name,
2334 exec_list *parameters);
2335
2336 const char *
2337 mode_string(const ir_variable *var);
2338
2339 extern "C" {
2340 #endif /* __cplusplus */
2341
2342 extern void _mesa_print_ir(struct exec_list *instructions,
2343 struct _mesa_glsl_parse_state *state);
2344
2345 #ifdef __cplusplus
2346 } /* extern "C" */
2347 #endif
2348
2349 unsigned
2350 vertices_per_prim(GLenum prim);
2351
2352 #endif /* IR_H */