glsl: Add support for new fma built-in in ARB_gpu_shader5.
[mesa.git] / src / glsl / ir.h
1 /* -*- c++ -*- */
2 /*
3 * Copyright © 2010 Intel Corporation
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
14 * Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
21 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
22 * DEALINGS IN THE SOFTWARE.
23 */
24
25 #pragma once
26 #ifndef IR_H
27 #define IR_H
28
29 #include <stdio.h>
30 #include <stdlib.h>
31
32 #include "ralloc.h"
33 #include "glsl_types.h"
34 #include "list.h"
35 #include "ir_visitor.h"
36 #include "ir_hierarchical_visitor.h"
37 #include "main/mtypes.h"
38
39 #ifdef __cplusplus
40
41 /**
42 * \defgroup IR Intermediate representation nodes
43 *
44 * @{
45 */
46
47 /**
48 * Class tags
49 *
50 * Each concrete class derived from \c ir_instruction has a value in this
51 * enumerant. The value for the type is stored in \c ir_instruction::ir_type
52 * by the constructor. While using type tags is not very C++, it is extremely
53 * convenient. For example, during debugging you can simply inspect
54 * \c ir_instruction::ir_type to find out the actual type of the object.
55 *
56 * In addition, it is possible to use a switch-statement based on \c
57 * \c ir_instruction::ir_type to select different behavior for different object
58 * types. For functions that have only slight differences for several object
59 * types, this allows writing very straightforward, readable code.
60 */
61 enum ir_node_type {
62 /**
63 * Zero is unused so that the IR validator can detect cases where
64 * \c ir_instruction::ir_type has not been initialized.
65 */
66 ir_type_unset,
67 ir_type_variable,
68 ir_type_assignment,
69 ir_type_call,
70 ir_type_constant,
71 ir_type_dereference_array,
72 ir_type_dereference_record,
73 ir_type_dereference_variable,
74 ir_type_discard,
75 ir_type_expression,
76 ir_type_function,
77 ir_type_function_signature,
78 ir_type_if,
79 ir_type_loop,
80 ir_type_loop_jump,
81 ir_type_return,
82 ir_type_swizzle,
83 ir_type_texture,
84 ir_type_emit_vertex,
85 ir_type_end_primitive,
86 ir_type_max /**< maximum ir_type enum number, for validation */
87 };
88
89 /**
90 * Base class of all IR instructions
91 */
92 class ir_instruction : public exec_node {
93 public:
94 enum ir_node_type ir_type;
95
96 /**
97 * GCC 4.7+ and clang warn when deleting an ir_instruction unless
98 * there's a virtual destructor present. Because we almost
99 * universally use ralloc for our memory management of
100 * ir_instructions, the destructor doesn't need to do any work.
101 */
102 virtual ~ir_instruction()
103 {
104 }
105
106 /** ir_print_visitor helper for debugging. */
107 void print(void) const;
108
109 virtual void accept(ir_visitor *) = 0;
110 virtual ir_visitor_status accept(ir_hierarchical_visitor *) = 0;
111 virtual ir_instruction *clone(void *mem_ctx,
112 struct hash_table *ht) const = 0;
113
114 /**
115 * \name IR instruction downcast functions
116 *
117 * These functions either cast the object to a derived class or return
118 * \c NULL if the object's type does not match the specified derived class.
119 * Additional downcast functions will be added as needed.
120 */
121 /*@{*/
122 virtual class ir_variable * as_variable() { return NULL; }
123 virtual class ir_function * as_function() { return NULL; }
124 virtual class ir_dereference * as_dereference() { return NULL; }
125 virtual class ir_dereference_array * as_dereference_array() { return NULL; }
126 virtual class ir_dereference_variable *as_dereference_variable() { return NULL; }
127 virtual class ir_dereference_record *as_dereference_record() { return NULL; }
128 virtual class ir_expression * as_expression() { return NULL; }
129 virtual class ir_rvalue * as_rvalue() { return NULL; }
130 virtual class ir_loop * as_loop() { return NULL; }
131 virtual class ir_assignment * as_assignment() { return NULL; }
132 virtual class ir_call * as_call() { return NULL; }
133 virtual class ir_return * as_return() { return NULL; }
134 virtual class ir_if * as_if() { return NULL; }
135 virtual class ir_swizzle * as_swizzle() { return NULL; }
136 virtual class ir_constant * as_constant() { return NULL; }
137 virtual class ir_discard * as_discard() { return NULL; }
138 virtual class ir_jump * as_jump() { return NULL; }
139 /*@}*/
140
141 protected:
142 ir_instruction()
143 {
144 ir_type = ir_type_unset;
145 }
146 };
147
148
149 /**
150 * The base class for all "values"/expression trees.
151 */
152 class ir_rvalue : public ir_instruction {
153 public:
154 const struct glsl_type *type;
155
156 virtual ir_rvalue *clone(void *mem_ctx, struct hash_table *) const;
157
158 virtual void accept(ir_visitor *v)
159 {
160 v->visit(this);
161 }
162
163 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
164
165 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
166
167 virtual ir_rvalue * as_rvalue()
168 {
169 return this;
170 }
171
172 ir_rvalue *as_rvalue_to_saturate();
173
174 virtual bool is_lvalue() const
175 {
176 return false;
177 }
178
179 /**
180 * Get the variable that is ultimately referenced by an r-value
181 */
182 virtual ir_variable *variable_referenced() const
183 {
184 return NULL;
185 }
186
187
188 /**
189 * If an r-value is a reference to a whole variable, get that variable
190 *
191 * \return
192 * Pointer to a variable that is completely dereferenced by the r-value. If
193 * the r-value is not a dereference or the dereference does not access the
194 * entire variable (i.e., it's just one array element, struct field), \c NULL
195 * is returned.
196 */
197 virtual ir_variable *whole_variable_referenced()
198 {
199 return NULL;
200 }
201
202 /**
203 * Determine if an r-value has the value zero
204 *
205 * The base implementation of this function always returns \c false. The
206 * \c ir_constant class over-rides this function to return \c true \b only
207 * for vector and scalar types that have all elements set to the value
208 * zero (or \c false for booleans).
209 *
210 * \sa ir_constant::has_value, ir_rvalue::is_one, ir_rvalue::is_negative_one,
211 * ir_constant::is_basis
212 */
213 virtual bool is_zero() const;
214
215 /**
216 * Determine if an r-value has the value one
217 *
218 * The base implementation of this function always returns \c false. The
219 * \c ir_constant class over-rides this function to return \c true \b only
220 * for vector and scalar types that have all elements set to the value
221 * one (or \c true for booleans).
222 *
223 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_negative_one,
224 * ir_constant::is_basis
225 */
226 virtual bool is_one() const;
227
228 /**
229 * Determine if an r-value has the value negative one
230 *
231 * The base implementation of this function always returns \c false. The
232 * \c ir_constant class over-rides this function to return \c true \b only
233 * for vector and scalar types that have all elements set to the value
234 * negative one. For boolean types, the result is always \c false.
235 *
236 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one
237 * ir_constant::is_basis
238 */
239 virtual bool is_negative_one() const;
240
241 /**
242 * Determine if an r-value is a basis vector
243 *
244 * The base implementation of this function always returns \c false. The
245 * \c ir_constant class over-rides this function to return \c true \b only
246 * for vector and scalar types that have one element set to the value one,
247 * and the other elements set to the value zero. For boolean types, the
248 * result is always \c false.
249 *
250 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one,
251 * is_constant::is_negative_one
252 */
253 virtual bool is_basis() const;
254
255
256 /**
257 * Return a generic value of error_type.
258 *
259 * Allocation will be performed with 'mem_ctx' as ralloc owner.
260 */
261 static ir_rvalue *error_value(void *mem_ctx);
262
263 protected:
264 ir_rvalue();
265 };
266
267
268 /**
269 * Variable storage classes
270 */
271 enum ir_variable_mode {
272 ir_var_auto = 0, /**< Function local variables and globals. */
273 ir_var_uniform, /**< Variable declared as a uniform. */
274 ir_var_shader_in,
275 ir_var_shader_out,
276 ir_var_function_in,
277 ir_var_function_out,
278 ir_var_function_inout,
279 ir_var_const_in, /**< "in" param that must be a constant expression */
280 ir_var_system_value, /**< Ex: front-face, instance-id, etc. */
281 ir_var_temporary, /**< Temporary variable generated during compilation. */
282 ir_var_mode_count /**< Number of variable modes */
283 };
284
285 /**
286 * \brief Layout qualifiers for gl_FragDepth.
287 *
288 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
289 * with a layout qualifier.
290 */
291 enum ir_depth_layout {
292 ir_depth_layout_none, /**< No depth layout is specified. */
293 ir_depth_layout_any,
294 ir_depth_layout_greater,
295 ir_depth_layout_less,
296 ir_depth_layout_unchanged
297 };
298
299 /**
300 * \brief Convert depth layout qualifier to string.
301 */
302 const char*
303 depth_layout_string(ir_depth_layout layout);
304
305 /**
306 * Description of built-in state associated with a uniform
307 *
308 * \sa ir_variable::state_slots
309 */
310 struct ir_state_slot {
311 int tokens[5];
312 int swizzle;
313 };
314
315 class ir_variable : public ir_instruction {
316 public:
317 ir_variable(const struct glsl_type *, const char *, ir_variable_mode);
318
319 virtual ir_variable *clone(void *mem_ctx, struct hash_table *ht) const;
320
321 virtual ir_variable *as_variable()
322 {
323 return this;
324 }
325
326 virtual void accept(ir_visitor *v)
327 {
328 v->visit(this);
329 }
330
331 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
332
333
334 /**
335 * Get the string value for the interpolation qualifier
336 *
337 * \return The string that would be used in a shader to specify \c
338 * mode will be returned.
339 *
340 * This function is used to generate error messages of the form "shader
341 * uses %s interpolation qualifier", so in the case where there is no
342 * interpolation qualifier, it returns "no".
343 *
344 * This function should only be used on a shader input or output variable.
345 */
346 const char *interpolation_string() const;
347
348 /**
349 * Determine how this variable should be interpolated based on its
350 * interpolation qualifier (if present), whether it is gl_Color or
351 * gl_SecondaryColor, and whether flatshading is enabled in the current GL
352 * state.
353 *
354 * The return value will always be either INTERP_QUALIFIER_SMOOTH,
355 * INTERP_QUALIFIER_NOPERSPECTIVE, or INTERP_QUALIFIER_FLAT.
356 */
357 glsl_interp_qualifier determine_interpolation_mode(bool flat_shade);
358
359 /**
360 * Determine whether or not a variable is part of a uniform block.
361 */
362 inline bool is_in_uniform_block() const
363 {
364 return this->mode == ir_var_uniform && this->interface_type != NULL;
365 }
366
367 /**
368 * Determine whether or not a variable is the declaration of an interface
369 * block
370 *
371 * For the first declaration below, there will be an \c ir_variable named
372 * "instance" whose type and whose instance_type will be the same
373 * \cglsl_type. For the second declaration, there will be an \c ir_variable
374 * named "f" whose type is float and whose instance_type is B2.
375 *
376 * "instance" is an interface instance variable, but "f" is not.
377 *
378 * uniform B1 {
379 * float f;
380 * } instance;
381 *
382 * uniform B2 {
383 * float f;
384 * };
385 */
386 inline bool is_interface_instance() const
387 {
388 const glsl_type *const t = this->type;
389
390 return (t == this->interface_type)
391 || (t->is_array() && t->fields.array == this->interface_type);
392 }
393
394 /**
395 * Declared type of the variable
396 */
397 const struct glsl_type *type;
398
399 /**
400 * Declared name of the variable
401 */
402 const char *name;
403
404 /**
405 * Highest element accessed with a constant expression array index
406 *
407 * Not used for non-array variables.
408 */
409 unsigned max_array_access;
410
411 /**
412 * Is the variable read-only?
413 *
414 * This is set for variables declared as \c const, shader inputs,
415 * and uniforms.
416 */
417 unsigned read_only:1;
418 unsigned centroid:1;
419 unsigned invariant:1;
420
421 /**
422 * Has this variable been used for reading or writing?
423 *
424 * Several GLSL semantic checks require knowledge of whether or not a
425 * variable has been used. For example, it is an error to redeclare a
426 * variable as invariant after it has been used.
427 *
428 * This is only maintained in the ast_to_hir.cpp path, not in
429 * Mesa's fixed function or ARB program paths.
430 */
431 unsigned used:1;
432
433 /**
434 * Has this variable been statically assigned?
435 *
436 * This answers whether the variable was assigned in any path of
437 * the shader during ast_to_hir. This doesn't answer whether it is
438 * still written after dead code removal, nor is it maintained in
439 * non-ast_to_hir.cpp (GLSL parsing) paths.
440 */
441 unsigned assigned:1;
442
443 /**
444 * Storage class of the variable.
445 *
446 * \sa ir_variable_mode
447 */
448 unsigned mode:4;
449
450 /**
451 * Interpolation mode for shader inputs / outputs
452 *
453 * \sa ir_variable_interpolation
454 */
455 unsigned interpolation:2;
456
457 /**
458 * \name ARB_fragment_coord_conventions
459 * @{
460 */
461 unsigned origin_upper_left:1;
462 unsigned pixel_center_integer:1;
463 /*@}*/
464
465 /**
466 * Was the location explicitly set in the shader?
467 *
468 * If the location is explicitly set in the shader, it \b cannot be changed
469 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
470 * no effect).
471 */
472 unsigned explicit_location:1;
473 unsigned explicit_index:1;
474
475 /**
476 * Was an initial binding explicitly set in the shader?
477 *
478 * If so, constant_value contains an integer ir_constant representing the
479 * initial binding point.
480 */
481 unsigned explicit_binding:1;
482
483 /**
484 * Does this variable have an initializer?
485 *
486 * This is used by the linker to cross-validiate initializers of global
487 * variables.
488 */
489 unsigned has_initializer:1;
490
491 /**
492 * Is this variable a generic output or input that has not yet been matched
493 * up to a variable in another stage of the pipeline?
494 *
495 * This is used by the linker as scratch storage while assigning locations
496 * to generic inputs and outputs.
497 */
498 unsigned is_unmatched_generic_inout:1;
499
500 /**
501 * If non-zero, then this variable may be packed along with other variables
502 * into a single varying slot, so this offset should be applied when
503 * accessing components. For example, an offset of 1 means that the x
504 * component of this variable is actually stored in component y of the
505 * location specified by \c location.
506 */
507 unsigned location_frac:2;
508
509 /**
510 * \brief Layout qualifier for gl_FragDepth.
511 *
512 * This is not equal to \c ir_depth_layout_none if and only if this
513 * variable is \c gl_FragDepth and a layout qualifier is specified.
514 */
515 ir_depth_layout depth_layout;
516
517 /**
518 * Storage location of the base of this variable
519 *
520 * The precise meaning of this field depends on the nature of the variable.
521 *
522 * - Vertex shader input: one of the values from \c gl_vert_attrib.
523 * - Vertex shader output: one of the values from \c gl_varying_slot.
524 * - Geometry shader input: one of the values from \c gl_varying_slot.
525 * - Geometry shader output: one of the values from \c gl_varying_slot.
526 * - Fragment shader input: one of the values from \c gl_varying_slot.
527 * - Fragment shader output: one of the values from \c gl_frag_result.
528 * - Uniforms: Per-stage uniform slot number for default uniform block.
529 * - Uniforms: Index within the uniform block definition for UBO members.
530 * - Other: This field is not currently used.
531 *
532 * If the variable is a uniform, shader input, or shader output, and the
533 * slot has not been assigned, the value will be -1.
534 */
535 int location;
536
537 /**
538 * output index for dual source blending.
539 */
540 int index;
541
542 /**
543 * Initial binding point for a sampler or UBO.
544 *
545 * For array types, this represents the binding point for the first element.
546 */
547 int binding;
548
549 /**
550 * Built-in state that backs this uniform
551 *
552 * Once set at variable creation, \c state_slots must remain invariant.
553 * This is because, ideally, this array would be shared by all clones of
554 * this variable in the IR tree. In other words, we'd really like for it
555 * to be a fly-weight.
556 *
557 * If the variable is not a uniform, \c num_state_slots will be zero and
558 * \c state_slots will be \c NULL.
559 */
560 /*@{*/
561 unsigned num_state_slots; /**< Number of state slots used */
562 ir_state_slot *state_slots; /**< State descriptors. */
563 /*@}*/
564
565 /**
566 * Emit a warning if this variable is accessed.
567 */
568 const char *warn_extension;
569
570 /**
571 * Value assigned in the initializer of a variable declared "const"
572 */
573 ir_constant *constant_value;
574
575 /**
576 * Constant expression assigned in the initializer of the variable
577 *
578 * \warning
579 * This field and \c ::constant_value are distinct. Even if the two fields
580 * refer to constants with the same value, they must point to separate
581 * objects.
582 */
583 ir_constant *constant_initializer;
584
585 /**
586 * For variables that are in an interface block or are an instance of an
587 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
588 *
589 * \sa ir_variable::location
590 */
591 const glsl_type *interface_type;
592 };
593
594
595 /*@{*/
596 /**
597 * The representation of a function instance; may be the full definition or
598 * simply a prototype.
599 */
600 class ir_function_signature : public ir_instruction {
601 /* An ir_function_signature will be part of the list of signatures in
602 * an ir_function.
603 */
604 public:
605 ir_function_signature(const glsl_type *return_type);
606
607 virtual ir_function_signature *clone(void *mem_ctx,
608 struct hash_table *ht) const;
609 ir_function_signature *clone_prototype(void *mem_ctx,
610 struct hash_table *ht) const;
611
612 virtual void accept(ir_visitor *v)
613 {
614 v->visit(this);
615 }
616
617 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
618
619 /**
620 * Attempt to evaluate this function as a constant expression,
621 * given a list of the actual parameters and the variable context.
622 * Returns NULL for non-built-ins.
623 */
624 ir_constant *constant_expression_value(exec_list *actual_parameters, struct hash_table *variable_context);
625
626 /**
627 * Get the name of the function for which this is a signature
628 */
629 const char *function_name() const;
630
631 /**
632 * Get a handle to the function for which this is a signature
633 *
634 * There is no setter function, this function returns a \c const pointer,
635 * and \c ir_function_signature::_function is private for a reason. The
636 * only way to make a connection between a function and function signature
637 * is via \c ir_function::add_signature. This helps ensure that certain
638 * invariants (i.e., a function signature is in the list of signatures for
639 * its \c _function) are met.
640 *
641 * \sa ir_function::add_signature
642 */
643 inline const class ir_function *function() const
644 {
645 return this->_function;
646 }
647
648 /**
649 * Check whether the qualifiers match between this signature's parameters
650 * and the supplied parameter list. If not, returns the name of the first
651 * parameter with mismatched qualifiers (for use in error messages).
652 */
653 const char *qualifiers_match(exec_list *params);
654
655 /**
656 * Replace the current parameter list with the given one. This is useful
657 * if the current information came from a prototype, and either has invalid
658 * or missing parameter names.
659 */
660 void replace_parameters(exec_list *new_params);
661
662 /**
663 * Function return type.
664 *
665 * \note This discards the optional precision qualifier.
666 */
667 const struct glsl_type *return_type;
668
669 /**
670 * List of ir_variable of function parameters.
671 *
672 * This represents the storage. The paramaters passed in a particular
673 * call will be in ir_call::actual_paramaters.
674 */
675 struct exec_list parameters;
676
677 /** Whether or not this function has a body (which may be empty). */
678 unsigned is_defined:1;
679
680 /** Whether or not this function signature is a built-in. */
681 unsigned is_builtin:1;
682
683 /** Body of instructions in the function. */
684 struct exec_list body;
685
686 private:
687 /** Function of which this signature is one overload. */
688 class ir_function *_function;
689
690 /** Function signature of which this one is a prototype clone */
691 const ir_function_signature *origin;
692
693 friend class ir_function;
694
695 /**
696 * Helper function to run a list of instructions for constant
697 * expression evaluation.
698 *
699 * The hash table represents the values of the visible variables.
700 * There are no scoping issues because the table is indexed on
701 * ir_variable pointers, not variable names.
702 *
703 * Returns false if the expression is not constant, true otherwise,
704 * and the value in *result if result is non-NULL.
705 */
706 bool constant_expression_evaluate_expression_list(const struct exec_list &body,
707 struct hash_table *variable_context,
708 ir_constant **result);
709 };
710
711
712 /**
713 * Header for tracking multiple overloaded functions with the same name.
714 * Contains a list of ir_function_signatures representing each of the
715 * actual functions.
716 */
717 class ir_function : public ir_instruction {
718 public:
719 ir_function(const char *name);
720
721 virtual ir_function *clone(void *mem_ctx, struct hash_table *ht) const;
722
723 virtual ir_function *as_function()
724 {
725 return this;
726 }
727
728 virtual void accept(ir_visitor *v)
729 {
730 v->visit(this);
731 }
732
733 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
734
735 void add_signature(ir_function_signature *sig)
736 {
737 sig->_function = this;
738 this->signatures.push_tail(sig);
739 }
740
741 /**
742 * Get an iterator for the set of function signatures
743 */
744 exec_list_iterator iterator()
745 {
746 return signatures.iterator();
747 }
748
749 /**
750 * Find a signature that matches a set of actual parameters, taking implicit
751 * conversions into account. Also flags whether the match was exact.
752 */
753 ir_function_signature *matching_signature(const exec_list *actual_param,
754 bool *match_is_exact);
755
756 /**
757 * Find a signature that matches a set of actual parameters, taking implicit
758 * conversions into account.
759 */
760 ir_function_signature *matching_signature(const exec_list *actual_param);
761
762 /**
763 * Find a signature that exactly matches a set of actual parameters without
764 * any implicit type conversions.
765 */
766 ir_function_signature *exact_matching_signature(const exec_list *actual_ps);
767
768 /**
769 * Name of the function.
770 */
771 const char *name;
772
773 /** Whether or not this function has a signature that isn't a built-in. */
774 bool has_user_signature();
775
776 /**
777 * List of ir_function_signature for each overloaded function with this name.
778 */
779 struct exec_list signatures;
780 };
781
782 inline const char *ir_function_signature::function_name() const
783 {
784 return this->_function->name;
785 }
786 /*@}*/
787
788
789 /**
790 * IR instruction representing high-level if-statements
791 */
792 class ir_if : public ir_instruction {
793 public:
794 ir_if(ir_rvalue *condition)
795 : condition(condition)
796 {
797 ir_type = ir_type_if;
798 }
799
800 virtual ir_if *clone(void *mem_ctx, struct hash_table *ht) const;
801
802 virtual ir_if *as_if()
803 {
804 return this;
805 }
806
807 virtual void accept(ir_visitor *v)
808 {
809 v->visit(this);
810 }
811
812 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
813
814 ir_rvalue *condition;
815 /** List of ir_instruction for the body of the then branch */
816 exec_list then_instructions;
817 /** List of ir_instruction for the body of the else branch */
818 exec_list else_instructions;
819 };
820
821
822 /**
823 * IR instruction representing a high-level loop structure.
824 */
825 class ir_loop : public ir_instruction {
826 public:
827 ir_loop();
828
829 virtual ir_loop *clone(void *mem_ctx, struct hash_table *ht) const;
830
831 virtual void accept(ir_visitor *v)
832 {
833 v->visit(this);
834 }
835
836 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
837
838 virtual ir_loop *as_loop()
839 {
840 return this;
841 }
842
843 /**
844 * Get an iterator for the instructions of the loop body
845 */
846 exec_list_iterator iterator()
847 {
848 return body_instructions.iterator();
849 }
850
851 /** List of ir_instruction that make up the body of the loop. */
852 exec_list body_instructions;
853
854 /**
855 * \name Loop counter and controls
856 *
857 * Represents a loop like a FORTRAN \c do-loop.
858 *
859 * \note
860 * If \c from and \c to are the same value, the loop will execute once.
861 */
862 /*@{*/
863 ir_rvalue *from; /** Value of the loop counter on the first
864 * iteration of the loop.
865 */
866 ir_rvalue *to; /** Value of the loop counter on the last
867 * iteration of the loop.
868 */
869 ir_rvalue *increment;
870 ir_variable *counter;
871
872 /**
873 * Comparison operation in the loop terminator.
874 *
875 * If any of the loop control fields are non-\c NULL, this field must be
876 * one of \c ir_binop_less, \c ir_binop_greater, \c ir_binop_lequal,
877 * \c ir_binop_gequal, \c ir_binop_equal, or \c ir_binop_nequal.
878 */
879 int cmp;
880 /*@}*/
881 };
882
883
884 class ir_assignment : public ir_instruction {
885 public:
886 ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs, ir_rvalue *condition = NULL);
887
888 /**
889 * Construct an assignment with an explicit write mask
890 *
891 * \note
892 * Since a write mask is supplied, the LHS must already be a bare
893 * \c ir_dereference. The cannot be any swizzles in the LHS.
894 */
895 ir_assignment(ir_dereference *lhs, ir_rvalue *rhs, ir_rvalue *condition,
896 unsigned write_mask);
897
898 virtual ir_assignment *clone(void *mem_ctx, struct hash_table *ht) const;
899
900 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
901
902 virtual void accept(ir_visitor *v)
903 {
904 v->visit(this);
905 }
906
907 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
908
909 virtual ir_assignment * as_assignment()
910 {
911 return this;
912 }
913
914 /**
915 * Get a whole variable written by an assignment
916 *
917 * If the LHS of the assignment writes a whole variable, the variable is
918 * returned. Otherwise \c NULL is returned. Examples of whole-variable
919 * assignment are:
920 *
921 * - Assigning to a scalar
922 * - Assigning to all components of a vector
923 * - Whole array (or matrix) assignment
924 * - Whole structure assignment
925 */
926 ir_variable *whole_variable_written();
927
928 /**
929 * Set the LHS of an assignment
930 */
931 void set_lhs(ir_rvalue *lhs);
932
933 /**
934 * Left-hand side of the assignment.
935 *
936 * This should be treated as read only. If you need to set the LHS of an
937 * assignment, use \c ir_assignment::set_lhs.
938 */
939 ir_dereference *lhs;
940
941 /**
942 * Value being assigned
943 */
944 ir_rvalue *rhs;
945
946 /**
947 * Optional condition for the assignment.
948 */
949 ir_rvalue *condition;
950
951
952 /**
953 * Component mask written
954 *
955 * For non-vector types in the LHS, this field will be zero. For vector
956 * types, a bit will be set for each component that is written. Note that
957 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
958 *
959 * A partially-set write mask means that each enabled channel gets
960 * the value from a consecutive channel of the rhs. For example,
961 * to write just .xyw of gl_FrontColor with color:
962 *
963 * (assign (constant bool (1)) (xyw)
964 * (var_ref gl_FragColor)
965 * (swiz xyw (var_ref color)))
966 */
967 unsigned write_mask:4;
968 };
969
970 /* Update ir_expression::get_num_operands() and operator_strs when
971 * updating this list.
972 */
973 enum ir_expression_operation {
974 ir_unop_bit_not,
975 ir_unop_logic_not,
976 ir_unop_neg,
977 ir_unop_abs,
978 ir_unop_sign,
979 ir_unop_rcp,
980 ir_unop_rsq,
981 ir_unop_sqrt,
982 ir_unop_exp, /**< Log base e on gentype */
983 ir_unop_log, /**< Natural log on gentype */
984 ir_unop_exp2,
985 ir_unop_log2,
986 ir_unop_f2i, /**< Float-to-integer conversion. */
987 ir_unop_f2u, /**< Float-to-unsigned conversion. */
988 ir_unop_i2f, /**< Integer-to-float conversion. */
989 ir_unop_f2b, /**< Float-to-boolean conversion */
990 ir_unop_b2f, /**< Boolean-to-float conversion */
991 ir_unop_i2b, /**< int-to-boolean conversion */
992 ir_unop_b2i, /**< Boolean-to-int conversion */
993 ir_unop_u2f, /**< Unsigned-to-float conversion. */
994 ir_unop_i2u, /**< Integer-to-unsigned conversion. */
995 ir_unop_u2i, /**< Unsigned-to-integer conversion. */
996 ir_unop_bitcast_i2f, /**< Bit-identical int-to-float "conversion" */
997 ir_unop_bitcast_f2i, /**< Bit-identical float-to-int "conversion" */
998 ir_unop_bitcast_u2f, /**< Bit-identical uint-to-float "conversion" */
999 ir_unop_bitcast_f2u, /**< Bit-identical float-to-uint "conversion" */
1000 ir_unop_any,
1001
1002 /**
1003 * \name Unary floating-point rounding operations.
1004 */
1005 /*@{*/
1006 ir_unop_trunc,
1007 ir_unop_ceil,
1008 ir_unop_floor,
1009 ir_unop_fract,
1010 ir_unop_round_even,
1011 /*@}*/
1012
1013 /**
1014 * \name Trigonometric operations.
1015 */
1016 /*@{*/
1017 ir_unop_sin,
1018 ir_unop_cos,
1019 ir_unop_sin_reduced, /**< Reduced range sin. [-pi, pi] */
1020 ir_unop_cos_reduced, /**< Reduced range cos. [-pi, pi] */
1021 /*@}*/
1022
1023 /**
1024 * \name Partial derivatives.
1025 */
1026 /*@{*/
1027 ir_unop_dFdx,
1028 ir_unop_dFdy,
1029 /*@}*/
1030
1031 /**
1032 * \name Floating point pack and unpack operations.
1033 */
1034 /*@{*/
1035 ir_unop_pack_snorm_2x16,
1036 ir_unop_pack_snorm_4x8,
1037 ir_unop_pack_unorm_2x16,
1038 ir_unop_pack_unorm_4x8,
1039 ir_unop_pack_half_2x16,
1040 ir_unop_unpack_snorm_2x16,
1041 ir_unop_unpack_snorm_4x8,
1042 ir_unop_unpack_unorm_2x16,
1043 ir_unop_unpack_unorm_4x8,
1044 ir_unop_unpack_half_2x16,
1045 /*@}*/
1046
1047 /**
1048 * \name Lowered floating point unpacking operations.
1049 *
1050 * \see lower_packing_builtins_visitor::split_unpack_half_2x16
1051 */
1052 /*@{*/
1053 ir_unop_unpack_half_2x16_split_x,
1054 ir_unop_unpack_half_2x16_split_y,
1055 /*@}*/
1056
1057 /**
1058 * \name Bit operations, part of ARB_gpu_shader5.
1059 */
1060 /*@{*/
1061 ir_unop_bitfield_reverse,
1062 ir_unop_bit_count,
1063 ir_unop_find_msb,
1064 ir_unop_find_lsb,
1065 /*@}*/
1066
1067 ir_unop_noise,
1068
1069 /**
1070 * A sentinel marking the last of the unary operations.
1071 */
1072 ir_last_unop = ir_unop_noise,
1073
1074 ir_binop_add,
1075 ir_binop_sub,
1076 ir_binop_mul,
1077 ir_binop_div,
1078
1079 /**
1080 * Takes one of two combinations of arguments:
1081 *
1082 * - mod(vecN, vecN)
1083 * - mod(vecN, float)
1084 *
1085 * Does not take integer types.
1086 */
1087 ir_binop_mod,
1088
1089 /**
1090 * \name Binary comparison operators which return a boolean vector.
1091 * The type of both operands must be equal.
1092 */
1093 /*@{*/
1094 ir_binop_less,
1095 ir_binop_greater,
1096 ir_binop_lequal,
1097 ir_binop_gequal,
1098 ir_binop_equal,
1099 ir_binop_nequal,
1100 /**
1101 * Returns single boolean for whether all components of operands[0]
1102 * equal the components of operands[1].
1103 */
1104 ir_binop_all_equal,
1105 /**
1106 * Returns single boolean for whether any component of operands[0]
1107 * is not equal to the corresponding component of operands[1].
1108 */
1109 ir_binop_any_nequal,
1110 /*@}*/
1111
1112 /**
1113 * \name Bit-wise binary operations.
1114 */
1115 /*@{*/
1116 ir_binop_lshift,
1117 ir_binop_rshift,
1118 ir_binop_bit_and,
1119 ir_binop_bit_xor,
1120 ir_binop_bit_or,
1121 /*@}*/
1122
1123 ir_binop_logic_and,
1124 ir_binop_logic_xor,
1125 ir_binop_logic_or,
1126
1127 ir_binop_dot,
1128 ir_binop_min,
1129 ir_binop_max,
1130
1131 ir_binop_pow,
1132
1133 /**
1134 * \name Lowered floating point packing operations.
1135 *
1136 * \see lower_packing_builtins_visitor::split_pack_half_2x16
1137 */
1138 /*@{*/
1139 ir_binop_pack_half_2x16_split,
1140 /*@}*/
1141
1142 /**
1143 * \name First half of a lowered bitfieldInsert() operation.
1144 *
1145 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1146 */
1147 /*@{*/
1148 ir_binop_bfm,
1149 /*@}*/
1150
1151 /**
1152 * Load a value the size of a given GLSL type from a uniform block.
1153 *
1154 * operand0 is the ir_constant uniform block index in the linked shader.
1155 * operand1 is a byte offset within the uniform block.
1156 */
1157 ir_binop_ubo_load,
1158
1159 /**
1160 * Extract a scalar from a vector
1161 *
1162 * operand0 is the vector
1163 * operand1 is the index of the field to read from operand0
1164 */
1165 ir_binop_vector_extract,
1166
1167 /**
1168 * A sentinel marking the last of the binary operations.
1169 */
1170 ir_last_binop = ir_binop_vector_extract,
1171
1172 /**
1173 * \name Fused floating-point multiply-add, part of ARB_gpu_shader5.
1174 */
1175 /*@{*/
1176 ir_triop_fma,
1177 /*@}*/
1178
1179 ir_triop_lrp,
1180
1181 /**
1182 * \name Second half of a lowered bitfieldInsert() operation.
1183 *
1184 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1185 */
1186 /*@{*/
1187 ir_triop_bfi,
1188 /*@}*/
1189
1190 ir_triop_bitfield_extract,
1191
1192 /**
1193 * Generate a value with one field of a vector changed
1194 *
1195 * operand0 is the vector
1196 * operand1 is the value to write into the vector result
1197 * operand2 is the index in operand0 to be modified
1198 */
1199 ir_triop_vector_insert,
1200
1201 /**
1202 * A sentinel marking the last of the ternary operations.
1203 */
1204 ir_last_triop = ir_triop_vector_insert,
1205
1206 ir_quadop_bitfield_insert,
1207
1208 ir_quadop_vector,
1209
1210 /**
1211 * A sentinel marking the last of the ternary operations.
1212 */
1213 ir_last_quadop = ir_quadop_vector,
1214
1215 /**
1216 * A sentinel marking the last of all operations.
1217 */
1218 ir_last_opcode = ir_quadop_vector
1219 };
1220
1221 class ir_expression : public ir_rvalue {
1222 public:
1223 ir_expression(int op, const struct glsl_type *type,
1224 ir_rvalue *op0, ir_rvalue *op1 = NULL,
1225 ir_rvalue *op2 = NULL, ir_rvalue *op3 = NULL);
1226
1227 /**
1228 * Constructor for unary operation expressions
1229 */
1230 ir_expression(int op, ir_rvalue *);
1231
1232 /**
1233 * Constructor for binary operation expressions
1234 */
1235 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1);
1236
1237 virtual ir_expression *as_expression()
1238 {
1239 return this;
1240 }
1241
1242 virtual ir_expression *clone(void *mem_ctx, struct hash_table *ht) const;
1243
1244 /**
1245 * Attempt to constant-fold the expression
1246 *
1247 * The "variable_context" hash table links ir_variable * to ir_constant *
1248 * that represent the variables' values. \c NULL represents an empty
1249 * context.
1250 *
1251 * If the expression cannot be constant folded, this method will return
1252 * \c NULL.
1253 */
1254 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1255
1256 /**
1257 * Determine the number of operands used by an expression
1258 */
1259 static unsigned int get_num_operands(ir_expression_operation);
1260
1261 /**
1262 * Determine the number of operands used by an expression
1263 */
1264 unsigned int get_num_operands() const
1265 {
1266 return (this->operation == ir_quadop_vector)
1267 ? this->type->vector_elements : get_num_operands(operation);
1268 }
1269
1270 /**
1271 * Return a string representing this expression's operator.
1272 */
1273 const char *operator_string();
1274
1275 /**
1276 * Return a string representing this expression's operator.
1277 */
1278 static const char *operator_string(ir_expression_operation);
1279
1280
1281 /**
1282 * Do a reverse-lookup to translate the given string into an operator.
1283 */
1284 static ir_expression_operation get_operator(const char *);
1285
1286 virtual void accept(ir_visitor *v)
1287 {
1288 v->visit(this);
1289 }
1290
1291 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1292
1293 ir_expression_operation operation;
1294 ir_rvalue *operands[4];
1295 };
1296
1297
1298 /**
1299 * HIR instruction representing a high-level function call, containing a list
1300 * of parameters and returning a value in the supplied temporary.
1301 */
1302 class ir_call : public ir_instruction {
1303 public:
1304 ir_call(ir_function_signature *callee,
1305 ir_dereference_variable *return_deref,
1306 exec_list *actual_parameters)
1307 : return_deref(return_deref), callee(callee)
1308 {
1309 ir_type = ir_type_call;
1310 assert(callee->return_type != NULL);
1311 actual_parameters->move_nodes_to(& this->actual_parameters);
1312 this->use_builtin = callee->is_builtin;
1313 }
1314
1315 virtual ir_call *clone(void *mem_ctx, struct hash_table *ht) const;
1316
1317 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1318
1319 virtual ir_call *as_call()
1320 {
1321 return this;
1322 }
1323
1324 virtual void accept(ir_visitor *v)
1325 {
1326 v->visit(this);
1327 }
1328
1329 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1330
1331 /**
1332 * Get an iterator for the set of acutal parameters
1333 */
1334 exec_list_iterator iterator()
1335 {
1336 return actual_parameters.iterator();
1337 }
1338
1339 /**
1340 * Get the name of the function being called.
1341 */
1342 const char *callee_name() const
1343 {
1344 return callee->function_name();
1345 }
1346
1347 /**
1348 * Generates an inline version of the function before @ir,
1349 * storing the return value in return_deref.
1350 */
1351 void generate_inline(ir_instruction *ir);
1352
1353 /**
1354 * Storage for the function's return value.
1355 * This must be NULL if the return type is void.
1356 */
1357 ir_dereference_variable *return_deref;
1358
1359 /**
1360 * The specific function signature being called.
1361 */
1362 ir_function_signature *callee;
1363
1364 /* List of ir_rvalue of paramaters passed in this call. */
1365 exec_list actual_parameters;
1366
1367 /** Should this call only bind to a built-in function? */
1368 bool use_builtin;
1369 };
1370
1371
1372 /**
1373 * \name Jump-like IR instructions.
1374 *
1375 * These include \c break, \c continue, \c return, and \c discard.
1376 */
1377 /*@{*/
1378 class ir_jump : public ir_instruction {
1379 protected:
1380 ir_jump()
1381 {
1382 ir_type = ir_type_unset;
1383 }
1384
1385 public:
1386 virtual ir_jump *as_jump()
1387 {
1388 return this;
1389 }
1390 };
1391
1392 class ir_return : public ir_jump {
1393 public:
1394 ir_return()
1395 : value(NULL)
1396 {
1397 this->ir_type = ir_type_return;
1398 }
1399
1400 ir_return(ir_rvalue *value)
1401 : value(value)
1402 {
1403 this->ir_type = ir_type_return;
1404 }
1405
1406 virtual ir_return *clone(void *mem_ctx, struct hash_table *) const;
1407
1408 virtual ir_return *as_return()
1409 {
1410 return this;
1411 }
1412
1413 ir_rvalue *get_value() const
1414 {
1415 return value;
1416 }
1417
1418 virtual void accept(ir_visitor *v)
1419 {
1420 v->visit(this);
1421 }
1422
1423 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1424
1425 ir_rvalue *value;
1426 };
1427
1428
1429 /**
1430 * Jump instructions used inside loops
1431 *
1432 * These include \c break and \c continue. The \c break within a loop is
1433 * different from the \c break within a switch-statement.
1434 *
1435 * \sa ir_switch_jump
1436 */
1437 class ir_loop_jump : public ir_jump {
1438 public:
1439 enum jump_mode {
1440 jump_break,
1441 jump_continue
1442 };
1443
1444 ir_loop_jump(jump_mode mode)
1445 {
1446 this->ir_type = ir_type_loop_jump;
1447 this->mode = mode;
1448 }
1449
1450 virtual ir_loop_jump *clone(void *mem_ctx, struct hash_table *) const;
1451
1452 virtual void accept(ir_visitor *v)
1453 {
1454 v->visit(this);
1455 }
1456
1457 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1458
1459 bool is_break() const
1460 {
1461 return mode == jump_break;
1462 }
1463
1464 bool is_continue() const
1465 {
1466 return mode == jump_continue;
1467 }
1468
1469 /** Mode selector for the jump instruction. */
1470 enum jump_mode mode;
1471 };
1472
1473 /**
1474 * IR instruction representing discard statements.
1475 */
1476 class ir_discard : public ir_jump {
1477 public:
1478 ir_discard()
1479 {
1480 this->ir_type = ir_type_discard;
1481 this->condition = NULL;
1482 }
1483
1484 ir_discard(ir_rvalue *cond)
1485 {
1486 this->ir_type = ir_type_discard;
1487 this->condition = cond;
1488 }
1489
1490 virtual ir_discard *clone(void *mem_ctx, struct hash_table *ht) const;
1491
1492 virtual void accept(ir_visitor *v)
1493 {
1494 v->visit(this);
1495 }
1496
1497 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1498
1499 virtual ir_discard *as_discard()
1500 {
1501 return this;
1502 }
1503
1504 ir_rvalue *condition;
1505 };
1506 /*@}*/
1507
1508
1509 /**
1510 * Texture sampling opcodes used in ir_texture
1511 */
1512 enum ir_texture_opcode {
1513 ir_tex, /**< Regular texture look-up */
1514 ir_txb, /**< Texture look-up with LOD bias */
1515 ir_txl, /**< Texture look-up with explicit LOD */
1516 ir_txd, /**< Texture look-up with partial derivatvies */
1517 ir_txf, /**< Texel fetch with explicit LOD */
1518 ir_txf_ms, /**< Multisample texture fetch */
1519 ir_txs, /**< Texture size */
1520 ir_lod /**< Texture lod query */
1521 };
1522
1523
1524 /**
1525 * IR instruction to sample a texture
1526 *
1527 * The specific form of the IR instruction depends on the \c mode value
1528 * selected from \c ir_texture_opcodes. In the printed IR, these will
1529 * appear as:
1530 *
1531 * Texel offset (0 or an expression)
1532 * | Projection divisor
1533 * | | Shadow comparitor
1534 * | | |
1535 * v v v
1536 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1537 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1538 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1539 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1540 * (txf <type> <sampler> <coordinate> 0 <lod>)
1541 * (txf_ms
1542 * <type> <sampler> <coordinate> <sample_index>)
1543 * (txs <type> <sampler> <lod>)
1544 * (lod <type> <sampler> <coordinate>)
1545 */
1546 class ir_texture : public ir_rvalue {
1547 public:
1548 ir_texture(enum ir_texture_opcode op)
1549 : op(op), sampler(NULL), coordinate(NULL), projector(NULL),
1550 shadow_comparitor(NULL), offset(NULL)
1551 {
1552 this->ir_type = ir_type_texture;
1553 }
1554
1555 virtual ir_texture *clone(void *mem_ctx, struct hash_table *) const;
1556
1557 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1558
1559 virtual void accept(ir_visitor *v)
1560 {
1561 v->visit(this);
1562 }
1563
1564 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1565
1566 /**
1567 * Return a string representing the ir_texture_opcode.
1568 */
1569 const char *opcode_string();
1570
1571 /** Set the sampler and type. */
1572 void set_sampler(ir_dereference *sampler, const glsl_type *type);
1573
1574 /**
1575 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1576 */
1577 static ir_texture_opcode get_opcode(const char *);
1578
1579 enum ir_texture_opcode op;
1580
1581 /** Sampler to use for the texture access. */
1582 ir_dereference *sampler;
1583
1584 /** Texture coordinate to sample */
1585 ir_rvalue *coordinate;
1586
1587 /**
1588 * Value used for projective divide.
1589 *
1590 * If there is no projective divide (the common case), this will be
1591 * \c NULL. Optimization passes should check for this to point to a constant
1592 * of 1.0 and replace that with \c NULL.
1593 */
1594 ir_rvalue *projector;
1595
1596 /**
1597 * Coordinate used for comparison on shadow look-ups.
1598 *
1599 * If there is no shadow comparison, this will be \c NULL. For the
1600 * \c ir_txf opcode, this *must* be \c NULL.
1601 */
1602 ir_rvalue *shadow_comparitor;
1603
1604 /** Texel offset. */
1605 ir_rvalue *offset;
1606
1607 union {
1608 ir_rvalue *lod; /**< Floating point LOD */
1609 ir_rvalue *bias; /**< Floating point LOD bias */
1610 ir_rvalue *sample_index; /**< MSAA sample index */
1611 struct {
1612 ir_rvalue *dPdx; /**< Partial derivative of coordinate wrt X */
1613 ir_rvalue *dPdy; /**< Partial derivative of coordinate wrt Y */
1614 } grad;
1615 } lod_info;
1616 };
1617
1618
1619 struct ir_swizzle_mask {
1620 unsigned x:2;
1621 unsigned y:2;
1622 unsigned z:2;
1623 unsigned w:2;
1624
1625 /**
1626 * Number of components in the swizzle.
1627 */
1628 unsigned num_components:3;
1629
1630 /**
1631 * Does the swizzle contain duplicate components?
1632 *
1633 * L-value swizzles cannot contain duplicate components.
1634 */
1635 unsigned has_duplicates:1;
1636 };
1637
1638
1639 class ir_swizzle : public ir_rvalue {
1640 public:
1641 ir_swizzle(ir_rvalue *, unsigned x, unsigned y, unsigned z, unsigned w,
1642 unsigned count);
1643
1644 ir_swizzle(ir_rvalue *val, const unsigned *components, unsigned count);
1645
1646 ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask);
1647
1648 virtual ir_swizzle *clone(void *mem_ctx, struct hash_table *) const;
1649
1650 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1651
1652 virtual ir_swizzle *as_swizzle()
1653 {
1654 return this;
1655 }
1656
1657 /**
1658 * Construct an ir_swizzle from the textual representation. Can fail.
1659 */
1660 static ir_swizzle *create(ir_rvalue *, const char *, unsigned vector_length);
1661
1662 virtual void accept(ir_visitor *v)
1663 {
1664 v->visit(this);
1665 }
1666
1667 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1668
1669 bool is_lvalue() const
1670 {
1671 return val->is_lvalue() && !mask.has_duplicates;
1672 }
1673
1674 /**
1675 * Get the variable that is ultimately referenced by an r-value
1676 */
1677 virtual ir_variable *variable_referenced() const;
1678
1679 ir_rvalue *val;
1680 ir_swizzle_mask mask;
1681
1682 private:
1683 /**
1684 * Initialize the mask component of a swizzle
1685 *
1686 * This is used by the \c ir_swizzle constructors.
1687 */
1688 void init_mask(const unsigned *components, unsigned count);
1689 };
1690
1691
1692 class ir_dereference : public ir_rvalue {
1693 public:
1694 virtual ir_dereference *clone(void *mem_ctx, struct hash_table *) const = 0;
1695
1696 virtual ir_dereference *as_dereference()
1697 {
1698 return this;
1699 }
1700
1701 bool is_lvalue() const;
1702
1703 /**
1704 * Get the variable that is ultimately referenced by an r-value
1705 */
1706 virtual ir_variable *variable_referenced() const = 0;
1707
1708 /**
1709 * Get the constant that is ultimately referenced by an r-value,
1710 * in a constant expression evaluation context.
1711 *
1712 * The offset is used when the reference is to a specific column of
1713 * a matrix.
1714 */
1715 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const = 0;
1716 };
1717
1718
1719 class ir_dereference_variable : public ir_dereference {
1720 public:
1721 ir_dereference_variable(ir_variable *var);
1722
1723 virtual ir_dereference_variable *clone(void *mem_ctx,
1724 struct hash_table *) const;
1725
1726 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1727
1728 virtual ir_dereference_variable *as_dereference_variable()
1729 {
1730 return this;
1731 }
1732
1733 /**
1734 * Get the variable that is ultimately referenced by an r-value
1735 */
1736 virtual ir_variable *variable_referenced() const
1737 {
1738 return this->var;
1739 }
1740
1741 /**
1742 * Get the constant that is ultimately referenced by an r-value,
1743 * in a constant expression evaluation context.
1744 *
1745 * The offset is used when the reference is to a specific column of
1746 * a matrix.
1747 */
1748 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
1749
1750 virtual ir_variable *whole_variable_referenced()
1751 {
1752 /* ir_dereference_variable objects always dereference the entire
1753 * variable. However, if this dereference is dereferenced by anything
1754 * else, the complete deferefernce chain is not a whole-variable
1755 * dereference. This method should only be called on the top most
1756 * ir_rvalue in a dereference chain.
1757 */
1758 return this->var;
1759 }
1760
1761 virtual void accept(ir_visitor *v)
1762 {
1763 v->visit(this);
1764 }
1765
1766 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1767
1768 /**
1769 * Object being dereferenced.
1770 */
1771 ir_variable *var;
1772 };
1773
1774
1775 class ir_dereference_array : public ir_dereference {
1776 public:
1777 ir_dereference_array(ir_rvalue *value, ir_rvalue *array_index);
1778
1779 ir_dereference_array(ir_variable *var, ir_rvalue *array_index);
1780
1781 virtual ir_dereference_array *clone(void *mem_ctx,
1782 struct hash_table *) const;
1783
1784 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1785
1786 virtual ir_dereference_array *as_dereference_array()
1787 {
1788 return this;
1789 }
1790
1791 /**
1792 * Get the variable that is ultimately referenced by an r-value
1793 */
1794 virtual ir_variable *variable_referenced() const
1795 {
1796 return this->array->variable_referenced();
1797 }
1798
1799 /**
1800 * Get the constant that is ultimately referenced by an r-value,
1801 * in a constant expression evaluation context.
1802 *
1803 * The offset is used when the reference is to a specific column of
1804 * a matrix.
1805 */
1806 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
1807
1808 virtual void accept(ir_visitor *v)
1809 {
1810 v->visit(this);
1811 }
1812
1813 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1814
1815 ir_rvalue *array;
1816 ir_rvalue *array_index;
1817
1818 private:
1819 void set_array(ir_rvalue *value);
1820 };
1821
1822
1823 class ir_dereference_record : public ir_dereference {
1824 public:
1825 ir_dereference_record(ir_rvalue *value, const char *field);
1826
1827 ir_dereference_record(ir_variable *var, const char *field);
1828
1829 virtual ir_dereference_record *clone(void *mem_ctx,
1830 struct hash_table *) const;
1831
1832 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1833
1834 virtual ir_dereference_record *as_dereference_record()
1835 {
1836 return this;
1837 }
1838
1839 /**
1840 * Get the variable that is ultimately referenced by an r-value
1841 */
1842 virtual ir_variable *variable_referenced() const
1843 {
1844 return this->record->variable_referenced();
1845 }
1846
1847 /**
1848 * Get the constant that is ultimately referenced by an r-value,
1849 * in a constant expression evaluation context.
1850 *
1851 * The offset is used when the reference is to a specific column of
1852 * a matrix.
1853 */
1854 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
1855
1856 virtual void accept(ir_visitor *v)
1857 {
1858 v->visit(this);
1859 }
1860
1861 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1862
1863 ir_rvalue *record;
1864 const char *field;
1865 };
1866
1867
1868 /**
1869 * Data stored in an ir_constant
1870 */
1871 union ir_constant_data {
1872 unsigned u[16];
1873 int i[16];
1874 float f[16];
1875 bool b[16];
1876 };
1877
1878
1879 class ir_constant : public ir_rvalue {
1880 public:
1881 ir_constant(const struct glsl_type *type, const ir_constant_data *data);
1882 ir_constant(bool b);
1883 ir_constant(unsigned int u);
1884 ir_constant(int i);
1885 ir_constant(float f);
1886
1887 /**
1888 * Construct an ir_constant from a list of ir_constant values
1889 */
1890 ir_constant(const struct glsl_type *type, exec_list *values);
1891
1892 /**
1893 * Construct an ir_constant from a scalar component of another ir_constant
1894 *
1895 * The new \c ir_constant inherits the type of the component from the
1896 * source constant.
1897 *
1898 * \note
1899 * In the case of a matrix constant, the new constant is a scalar, \b not
1900 * a vector.
1901 */
1902 ir_constant(const ir_constant *c, unsigned i);
1903
1904 /**
1905 * Return a new ir_constant of the specified type containing all zeros.
1906 */
1907 static ir_constant *zero(void *mem_ctx, const glsl_type *type);
1908
1909 virtual ir_constant *clone(void *mem_ctx, struct hash_table *) const;
1910
1911 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1912
1913 virtual ir_constant *as_constant()
1914 {
1915 return this;
1916 }
1917
1918 virtual void accept(ir_visitor *v)
1919 {
1920 v->visit(this);
1921 }
1922
1923 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1924
1925 /**
1926 * Get a particular component of a constant as a specific type
1927 *
1928 * This is useful, for example, to get a value from an integer constant
1929 * as a float or bool. This appears frequently when constructors are
1930 * called with all constant parameters.
1931 */
1932 /*@{*/
1933 bool get_bool_component(unsigned i) const;
1934 float get_float_component(unsigned i) const;
1935 int get_int_component(unsigned i) const;
1936 unsigned get_uint_component(unsigned i) const;
1937 /*@}*/
1938
1939 ir_constant *get_array_element(unsigned i) const;
1940
1941 ir_constant *get_record_field(const char *name);
1942
1943 /**
1944 * Copy the values on another constant at a given offset.
1945 *
1946 * The offset is ignored for array or struct copies, it's only for
1947 * scalars or vectors into vectors or matrices.
1948 *
1949 * With identical types on both sides and zero offset it's clone()
1950 * without creating a new object.
1951 */
1952
1953 void copy_offset(ir_constant *src, int offset);
1954
1955 /**
1956 * Copy the values on another constant at a given offset and
1957 * following an assign-like mask.
1958 *
1959 * The mask is ignored for scalars.
1960 *
1961 * Note that this function only handles what assign can handle,
1962 * i.e. at most a vector as source and a column of a matrix as
1963 * destination.
1964 */
1965
1966 void copy_masked_offset(ir_constant *src, int offset, unsigned int mask);
1967
1968 /**
1969 * Determine whether a constant has the same value as another constant
1970 *
1971 * \sa ir_constant::is_zero, ir_constant::is_one,
1972 * ir_constant::is_negative_one, ir_constant::is_basis
1973 */
1974 bool has_value(const ir_constant *) const;
1975
1976 virtual bool is_zero() const;
1977 virtual bool is_one() const;
1978 virtual bool is_negative_one() const;
1979 virtual bool is_basis() const;
1980
1981 /**
1982 * Value of the constant.
1983 *
1984 * The field used to back the values supplied by the constant is determined
1985 * by the type associated with the \c ir_instruction. Constants may be
1986 * scalars, vectors, or matrices.
1987 */
1988 union ir_constant_data value;
1989
1990 /* Array elements */
1991 ir_constant **array_elements;
1992
1993 /* Structure fields */
1994 exec_list components;
1995
1996 private:
1997 /**
1998 * Parameterless constructor only used by the clone method
1999 */
2000 ir_constant(void);
2001 };
2002
2003 /*@}*/
2004
2005 /**
2006 * IR instruction to emit a vertex in a geometry shader.
2007 */
2008 class ir_emit_vertex : public ir_instruction {
2009 public:
2010 ir_emit_vertex()
2011 {
2012 ir_type = ir_type_emit_vertex;
2013 }
2014
2015 virtual void accept(ir_visitor *v)
2016 {
2017 v->visit(this);
2018 }
2019
2020 virtual ir_emit_vertex *clone(void *mem_ctx, struct hash_table *) const
2021 {
2022 return new(mem_ctx) ir_emit_vertex();
2023 }
2024
2025 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2026 };
2027
2028 /**
2029 * IR instruction to complete the current primitive and start a new one in a
2030 * geometry shader.
2031 */
2032 class ir_end_primitive : public ir_instruction {
2033 public:
2034 ir_end_primitive()
2035 {
2036 ir_type = ir_type_end_primitive;
2037 }
2038
2039 virtual void accept(ir_visitor *v)
2040 {
2041 v->visit(this);
2042 }
2043
2044 virtual ir_end_primitive *clone(void *mem_ctx, struct hash_table *) const
2045 {
2046 return new(mem_ctx) ir_end_primitive();
2047 }
2048
2049 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2050 };
2051
2052 /**
2053 * Apply a visitor to each IR node in a list
2054 */
2055 void
2056 visit_exec_list(exec_list *list, ir_visitor *visitor);
2057
2058 /**
2059 * Validate invariants on each IR node in a list
2060 */
2061 void validate_ir_tree(exec_list *instructions);
2062
2063 struct _mesa_glsl_parse_state;
2064 struct gl_shader_program;
2065
2066 /**
2067 * Detect whether an unlinked shader contains static recursion
2068 *
2069 * If the list of instructions is determined to contain static recursion,
2070 * \c _mesa_glsl_error will be called to emit error messages for each function
2071 * that is in the recursion cycle.
2072 */
2073 void
2074 detect_recursion_unlinked(struct _mesa_glsl_parse_state *state,
2075 exec_list *instructions);
2076
2077 /**
2078 * Detect whether a linked shader contains static recursion
2079 *
2080 * If the list of instructions is determined to contain static recursion,
2081 * \c link_error_printf will be called to emit error messages for each function
2082 * that is in the recursion cycle. In addition,
2083 * \c gl_shader_program::LinkStatus will be set to false.
2084 */
2085 void
2086 detect_recursion_linked(struct gl_shader_program *prog,
2087 exec_list *instructions);
2088
2089 /**
2090 * Make a clone of each IR instruction in a list
2091 *
2092 * \param in List of IR instructions that are to be cloned
2093 * \param out List to hold the cloned instructions
2094 */
2095 void
2096 clone_ir_list(void *mem_ctx, exec_list *out, const exec_list *in);
2097
2098 extern void
2099 _mesa_glsl_initialize_variables(exec_list *instructions,
2100 struct _mesa_glsl_parse_state *state);
2101
2102 extern void
2103 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state *state);
2104
2105 extern void
2106 _mesa_glsl_release_functions(void);
2107
2108 extern void
2109 reparent_ir(exec_list *list, void *mem_ctx);
2110
2111 struct glsl_symbol_table;
2112
2113 extern void
2114 import_prototypes(const exec_list *source, exec_list *dest,
2115 struct glsl_symbol_table *symbols, void *mem_ctx);
2116
2117 extern bool
2118 ir_has_call(ir_instruction *ir);
2119
2120 extern void
2121 do_set_program_inouts(exec_list *instructions, struct gl_program *prog,
2122 GLenum shader_type);
2123
2124 extern char *
2125 prototype_string(const glsl_type *return_type, const char *name,
2126 exec_list *parameters);
2127
2128 extern "C" {
2129 #endif /* __cplusplus */
2130
2131 extern void _mesa_print_ir(struct exec_list *instructions,
2132 struct _mesa_glsl_parse_state *state);
2133
2134 #ifdef __cplusplus
2135 } /* extern "C" */
2136 #endif
2137
2138 unsigned
2139 vertices_per_prim(GLenum prim);
2140
2141 #endif /* IR_H */