glsl: add texture gather changes
[mesa.git] / src / glsl / ir.h
1 /* -*- c++ -*- */
2 /*
3 * Copyright © 2010 Intel Corporation
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
14 * Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
21 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
22 * DEALINGS IN THE SOFTWARE.
23 */
24
25 #pragma once
26 #ifndef IR_H
27 #define IR_H
28
29 #include <stdio.h>
30 #include <stdlib.h>
31
32 #include "ralloc.h"
33 #include "glsl_types.h"
34 #include "list.h"
35 #include "ir_visitor.h"
36 #include "ir_hierarchical_visitor.h"
37 #include "main/mtypes.h"
38
39 #ifdef __cplusplus
40
41 /**
42 * \defgroup IR Intermediate representation nodes
43 *
44 * @{
45 */
46
47 /**
48 * Class tags
49 *
50 * Each concrete class derived from \c ir_instruction has a value in this
51 * enumerant. The value for the type is stored in \c ir_instruction::ir_type
52 * by the constructor. While using type tags is not very C++, it is extremely
53 * convenient. For example, during debugging you can simply inspect
54 * \c ir_instruction::ir_type to find out the actual type of the object.
55 *
56 * In addition, it is possible to use a switch-statement based on \c
57 * \c ir_instruction::ir_type to select different behavior for different object
58 * types. For functions that have only slight differences for several object
59 * types, this allows writing very straightforward, readable code.
60 */
61 enum ir_node_type {
62 /**
63 * Zero is unused so that the IR validator can detect cases where
64 * \c ir_instruction::ir_type has not been initialized.
65 */
66 ir_type_unset,
67 ir_type_variable,
68 ir_type_assignment,
69 ir_type_call,
70 ir_type_constant,
71 ir_type_dereference_array,
72 ir_type_dereference_record,
73 ir_type_dereference_variable,
74 ir_type_discard,
75 ir_type_expression,
76 ir_type_function,
77 ir_type_function_signature,
78 ir_type_if,
79 ir_type_loop,
80 ir_type_loop_jump,
81 ir_type_return,
82 ir_type_swizzle,
83 ir_type_texture,
84 ir_type_emit_vertex,
85 ir_type_end_primitive,
86 ir_type_max /**< maximum ir_type enum number, for validation */
87 };
88
89 /**
90 * Base class of all IR instructions
91 */
92 class ir_instruction : public exec_node {
93 public:
94 enum ir_node_type ir_type;
95
96 /**
97 * GCC 4.7+ and clang warn when deleting an ir_instruction unless
98 * there's a virtual destructor present. Because we almost
99 * universally use ralloc for our memory management of
100 * ir_instructions, the destructor doesn't need to do any work.
101 */
102 virtual ~ir_instruction()
103 {
104 }
105
106 /** ir_print_visitor helper for debugging. */
107 void print(void) const;
108
109 virtual void accept(ir_visitor *) = 0;
110 virtual ir_visitor_status accept(ir_hierarchical_visitor *) = 0;
111 virtual ir_instruction *clone(void *mem_ctx,
112 struct hash_table *ht) const = 0;
113
114 /**
115 * \name IR instruction downcast functions
116 *
117 * These functions either cast the object to a derived class or return
118 * \c NULL if the object's type does not match the specified derived class.
119 * Additional downcast functions will be added as needed.
120 */
121 /*@{*/
122 virtual class ir_variable * as_variable() { return NULL; }
123 virtual class ir_function * as_function() { return NULL; }
124 virtual class ir_dereference * as_dereference() { return NULL; }
125 virtual class ir_dereference_array * as_dereference_array() { return NULL; }
126 virtual class ir_dereference_variable *as_dereference_variable() { return NULL; }
127 virtual class ir_dereference_record *as_dereference_record() { return NULL; }
128 virtual class ir_expression * as_expression() { return NULL; }
129 virtual class ir_rvalue * as_rvalue() { return NULL; }
130 virtual class ir_loop * as_loop() { return NULL; }
131 virtual class ir_assignment * as_assignment() { return NULL; }
132 virtual class ir_call * as_call() { return NULL; }
133 virtual class ir_return * as_return() { return NULL; }
134 virtual class ir_if * as_if() { return NULL; }
135 virtual class ir_swizzle * as_swizzle() { return NULL; }
136 virtual class ir_constant * as_constant() { return NULL; }
137 virtual class ir_discard * as_discard() { return NULL; }
138 virtual class ir_jump * as_jump() { return NULL; }
139 /*@}*/
140
141 protected:
142 ir_instruction()
143 {
144 ir_type = ir_type_unset;
145 }
146 };
147
148
149 /**
150 * The base class for all "values"/expression trees.
151 */
152 class ir_rvalue : public ir_instruction {
153 public:
154 const struct glsl_type *type;
155
156 virtual ir_rvalue *clone(void *mem_ctx, struct hash_table *) const;
157
158 virtual void accept(ir_visitor *v)
159 {
160 v->visit(this);
161 }
162
163 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
164
165 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
166
167 virtual ir_rvalue * as_rvalue()
168 {
169 return this;
170 }
171
172 ir_rvalue *as_rvalue_to_saturate();
173
174 virtual bool is_lvalue() const
175 {
176 return false;
177 }
178
179 /**
180 * Get the variable that is ultimately referenced by an r-value
181 */
182 virtual ir_variable *variable_referenced() const
183 {
184 return NULL;
185 }
186
187
188 /**
189 * If an r-value is a reference to a whole variable, get that variable
190 *
191 * \return
192 * Pointer to a variable that is completely dereferenced by the r-value. If
193 * the r-value is not a dereference or the dereference does not access the
194 * entire variable (i.e., it's just one array element, struct field), \c NULL
195 * is returned.
196 */
197 virtual ir_variable *whole_variable_referenced()
198 {
199 return NULL;
200 }
201
202 /**
203 * Determine if an r-value has the value zero
204 *
205 * The base implementation of this function always returns \c false. The
206 * \c ir_constant class over-rides this function to return \c true \b only
207 * for vector and scalar types that have all elements set to the value
208 * zero (or \c false for booleans).
209 *
210 * \sa ir_constant::has_value, ir_rvalue::is_one, ir_rvalue::is_negative_one,
211 * ir_constant::is_basis
212 */
213 virtual bool is_zero() const;
214
215 /**
216 * Determine if an r-value has the value one
217 *
218 * The base implementation of this function always returns \c false. The
219 * \c ir_constant class over-rides this function to return \c true \b only
220 * for vector and scalar types that have all elements set to the value
221 * one (or \c true for booleans).
222 *
223 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_negative_one,
224 * ir_constant::is_basis
225 */
226 virtual bool is_one() const;
227
228 /**
229 * Determine if an r-value has the value negative one
230 *
231 * The base implementation of this function always returns \c false. The
232 * \c ir_constant class over-rides this function to return \c true \b only
233 * for vector and scalar types that have all elements set to the value
234 * negative one. For boolean types, the result is always \c false.
235 *
236 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one
237 * ir_constant::is_basis
238 */
239 virtual bool is_negative_one() const;
240
241 /**
242 * Determine if an r-value is a basis vector
243 *
244 * The base implementation of this function always returns \c false. The
245 * \c ir_constant class over-rides this function to return \c true \b only
246 * for vector and scalar types that have one element set to the value one,
247 * and the other elements set to the value zero. For boolean types, the
248 * result is always \c false.
249 *
250 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one,
251 * is_constant::is_negative_one
252 */
253 virtual bool is_basis() const;
254
255
256 /**
257 * Return a generic value of error_type.
258 *
259 * Allocation will be performed with 'mem_ctx' as ralloc owner.
260 */
261 static ir_rvalue *error_value(void *mem_ctx);
262
263 protected:
264 ir_rvalue();
265 };
266
267
268 /**
269 * Variable storage classes
270 */
271 enum ir_variable_mode {
272 ir_var_auto = 0, /**< Function local variables and globals. */
273 ir_var_uniform, /**< Variable declared as a uniform. */
274 ir_var_shader_in,
275 ir_var_shader_out,
276 ir_var_function_in,
277 ir_var_function_out,
278 ir_var_function_inout,
279 ir_var_const_in, /**< "in" param that must be a constant expression */
280 ir_var_system_value, /**< Ex: front-face, instance-id, etc. */
281 ir_var_temporary, /**< Temporary variable generated during compilation. */
282 ir_var_mode_count /**< Number of variable modes */
283 };
284
285 /**
286 * \brief Layout qualifiers for gl_FragDepth.
287 *
288 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
289 * with a layout qualifier.
290 */
291 enum ir_depth_layout {
292 ir_depth_layout_none, /**< No depth layout is specified. */
293 ir_depth_layout_any,
294 ir_depth_layout_greater,
295 ir_depth_layout_less,
296 ir_depth_layout_unchanged
297 };
298
299 /**
300 * \brief Convert depth layout qualifier to string.
301 */
302 const char*
303 depth_layout_string(ir_depth_layout layout);
304
305 /**
306 * Description of built-in state associated with a uniform
307 *
308 * \sa ir_variable::state_slots
309 */
310 struct ir_state_slot {
311 int tokens[5];
312 int swizzle;
313 };
314
315 class ir_variable : public ir_instruction {
316 public:
317 ir_variable(const struct glsl_type *, const char *, ir_variable_mode);
318
319 virtual ir_variable *clone(void *mem_ctx, struct hash_table *ht) const;
320
321 virtual ir_variable *as_variable()
322 {
323 return this;
324 }
325
326 virtual void accept(ir_visitor *v)
327 {
328 v->visit(this);
329 }
330
331 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
332
333
334 /**
335 * Get the string value for the interpolation qualifier
336 *
337 * \return The string that would be used in a shader to specify \c
338 * mode will be returned.
339 *
340 * This function is used to generate error messages of the form "shader
341 * uses %s interpolation qualifier", so in the case where there is no
342 * interpolation qualifier, it returns "no".
343 *
344 * This function should only be used on a shader input or output variable.
345 */
346 const char *interpolation_string() const;
347
348 /**
349 * Determine how this variable should be interpolated based on its
350 * interpolation qualifier (if present), whether it is gl_Color or
351 * gl_SecondaryColor, and whether flatshading is enabled in the current GL
352 * state.
353 *
354 * The return value will always be either INTERP_QUALIFIER_SMOOTH,
355 * INTERP_QUALIFIER_NOPERSPECTIVE, or INTERP_QUALIFIER_FLAT.
356 */
357 glsl_interp_qualifier determine_interpolation_mode(bool flat_shade);
358
359 /**
360 * Determine whether or not a variable is part of a uniform block.
361 */
362 inline bool is_in_uniform_block() const
363 {
364 return this->mode == ir_var_uniform && this->interface_type != NULL;
365 }
366
367 /**
368 * Determine whether or not a variable is the declaration of an interface
369 * block
370 *
371 * For the first declaration below, there will be an \c ir_variable named
372 * "instance" whose type and whose instance_type will be the same
373 * \cglsl_type. For the second declaration, there will be an \c ir_variable
374 * named "f" whose type is float and whose instance_type is B2.
375 *
376 * "instance" is an interface instance variable, but "f" is not.
377 *
378 * uniform B1 {
379 * float f;
380 * } instance;
381 *
382 * uniform B2 {
383 * float f;
384 * };
385 */
386 inline bool is_interface_instance() const
387 {
388 const glsl_type *const t = this->type;
389
390 return (t == this->interface_type)
391 || (t->is_array() && t->fields.array == this->interface_type);
392 }
393
394 /**
395 * Declared type of the variable
396 */
397 const struct glsl_type *type;
398
399 /**
400 * Declared name of the variable
401 */
402 const char *name;
403
404 /**
405 * Highest element accessed with a constant expression array index
406 *
407 * Not used for non-array variables.
408 */
409 unsigned max_array_access;
410
411 /**
412 * Is the variable read-only?
413 *
414 * This is set for variables declared as \c const, shader inputs,
415 * and uniforms.
416 */
417 unsigned read_only:1;
418 unsigned centroid:1;
419 unsigned invariant:1;
420
421 /**
422 * Has this variable been used for reading or writing?
423 *
424 * Several GLSL semantic checks require knowledge of whether or not a
425 * variable has been used. For example, it is an error to redeclare a
426 * variable as invariant after it has been used.
427 *
428 * This is only maintained in the ast_to_hir.cpp path, not in
429 * Mesa's fixed function or ARB program paths.
430 */
431 unsigned used:1;
432
433 /**
434 * Has this variable been statically assigned?
435 *
436 * This answers whether the variable was assigned in any path of
437 * the shader during ast_to_hir. This doesn't answer whether it is
438 * still written after dead code removal, nor is it maintained in
439 * non-ast_to_hir.cpp (GLSL parsing) paths.
440 */
441 unsigned assigned:1;
442
443 /**
444 * Storage class of the variable.
445 *
446 * \sa ir_variable_mode
447 */
448 unsigned mode:4;
449
450 /**
451 * Interpolation mode for shader inputs / outputs
452 *
453 * \sa ir_variable_interpolation
454 */
455 unsigned interpolation:2;
456
457 /**
458 * \name ARB_fragment_coord_conventions
459 * @{
460 */
461 unsigned origin_upper_left:1;
462 unsigned pixel_center_integer:1;
463 /*@}*/
464
465 /**
466 * Was the location explicitly set in the shader?
467 *
468 * If the location is explicitly set in the shader, it \b cannot be changed
469 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
470 * no effect).
471 */
472 unsigned explicit_location:1;
473 unsigned explicit_index:1;
474
475 /**
476 * Was an initial binding explicitly set in the shader?
477 *
478 * If so, constant_value contains an integer ir_constant representing the
479 * initial binding point.
480 */
481 unsigned explicit_binding:1;
482
483 /**
484 * Does this variable have an initializer?
485 *
486 * This is used by the linker to cross-validiate initializers of global
487 * variables.
488 */
489 unsigned has_initializer:1;
490
491 /**
492 * Is this variable a generic output or input that has not yet been matched
493 * up to a variable in another stage of the pipeline?
494 *
495 * This is used by the linker as scratch storage while assigning locations
496 * to generic inputs and outputs.
497 */
498 unsigned is_unmatched_generic_inout:1;
499
500 /**
501 * If non-zero, then this variable may be packed along with other variables
502 * into a single varying slot, so this offset should be applied when
503 * accessing components. For example, an offset of 1 means that the x
504 * component of this variable is actually stored in component y of the
505 * location specified by \c location.
506 */
507 unsigned location_frac:2;
508
509 /**
510 * \brief Layout qualifier for gl_FragDepth.
511 *
512 * This is not equal to \c ir_depth_layout_none if and only if this
513 * variable is \c gl_FragDepth and a layout qualifier is specified.
514 */
515 ir_depth_layout depth_layout;
516
517 /**
518 * Storage location of the base of this variable
519 *
520 * The precise meaning of this field depends on the nature of the variable.
521 *
522 * - Vertex shader input: one of the values from \c gl_vert_attrib.
523 * - Vertex shader output: one of the values from \c gl_varying_slot.
524 * - Geometry shader input: one of the values from \c gl_varying_slot.
525 * - Geometry shader output: one of the values from \c gl_varying_slot.
526 * - Fragment shader input: one of the values from \c gl_varying_slot.
527 * - Fragment shader output: one of the values from \c gl_frag_result.
528 * - Uniforms: Per-stage uniform slot number for default uniform block.
529 * - Uniforms: Index within the uniform block definition for UBO members.
530 * - Other: This field is not currently used.
531 *
532 * If the variable is a uniform, shader input, or shader output, and the
533 * slot has not been assigned, the value will be -1.
534 */
535 int location;
536
537 /**
538 * output index for dual source blending.
539 */
540 int index;
541
542 /**
543 * Initial binding point for a sampler or UBO.
544 *
545 * For array types, this represents the binding point for the first element.
546 */
547 int binding;
548
549 /**
550 * Built-in state that backs this uniform
551 *
552 * Once set at variable creation, \c state_slots must remain invariant.
553 * This is because, ideally, this array would be shared by all clones of
554 * this variable in the IR tree. In other words, we'd really like for it
555 * to be a fly-weight.
556 *
557 * If the variable is not a uniform, \c num_state_slots will be zero and
558 * \c state_slots will be \c NULL.
559 */
560 /*@{*/
561 unsigned num_state_slots; /**< Number of state slots used */
562 ir_state_slot *state_slots; /**< State descriptors. */
563 /*@}*/
564
565 /**
566 * Emit a warning if this variable is accessed.
567 */
568 const char *warn_extension;
569
570 /**
571 * Value assigned in the initializer of a variable declared "const"
572 */
573 ir_constant *constant_value;
574
575 /**
576 * Constant expression assigned in the initializer of the variable
577 *
578 * \warning
579 * This field and \c ::constant_value are distinct. Even if the two fields
580 * refer to constants with the same value, they must point to separate
581 * objects.
582 */
583 ir_constant *constant_initializer;
584
585 /**
586 * For variables that are in an interface block or are an instance of an
587 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
588 *
589 * \sa ir_variable::location
590 */
591 const glsl_type *interface_type;
592 };
593
594 /**
595 * A function that returns whether a built-in function is available in the
596 * current shading language (based on version, ES or desktop, and extensions).
597 */
598 typedef bool (*builtin_available_predicate)(const _mesa_glsl_parse_state *);
599
600 /*@{*/
601 /**
602 * The representation of a function instance; may be the full definition or
603 * simply a prototype.
604 */
605 class ir_function_signature : public ir_instruction {
606 /* An ir_function_signature will be part of the list of signatures in
607 * an ir_function.
608 */
609 public:
610 ir_function_signature(const glsl_type *return_type,
611 builtin_available_predicate builtin_avail = NULL);
612
613 virtual ir_function_signature *clone(void *mem_ctx,
614 struct hash_table *ht) const;
615 ir_function_signature *clone_prototype(void *mem_ctx,
616 struct hash_table *ht) const;
617
618 virtual void accept(ir_visitor *v)
619 {
620 v->visit(this);
621 }
622
623 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
624
625 /**
626 * Attempt to evaluate this function as a constant expression,
627 * given a list of the actual parameters and the variable context.
628 * Returns NULL for non-built-ins.
629 */
630 ir_constant *constant_expression_value(exec_list *actual_parameters, struct hash_table *variable_context);
631
632 /**
633 * Get the name of the function for which this is a signature
634 */
635 const char *function_name() const;
636
637 /**
638 * Get a handle to the function for which this is a signature
639 *
640 * There is no setter function, this function returns a \c const pointer,
641 * and \c ir_function_signature::_function is private for a reason. The
642 * only way to make a connection between a function and function signature
643 * is via \c ir_function::add_signature. This helps ensure that certain
644 * invariants (i.e., a function signature is in the list of signatures for
645 * its \c _function) are met.
646 *
647 * \sa ir_function::add_signature
648 */
649 inline const class ir_function *function() const
650 {
651 return this->_function;
652 }
653
654 /**
655 * Check whether the qualifiers match between this signature's parameters
656 * and the supplied parameter list. If not, returns the name of the first
657 * parameter with mismatched qualifiers (for use in error messages).
658 */
659 const char *qualifiers_match(exec_list *params);
660
661 /**
662 * Replace the current parameter list with the given one. This is useful
663 * if the current information came from a prototype, and either has invalid
664 * or missing parameter names.
665 */
666 void replace_parameters(exec_list *new_params);
667
668 /**
669 * Function return type.
670 *
671 * \note This discards the optional precision qualifier.
672 */
673 const struct glsl_type *return_type;
674
675 /**
676 * List of ir_variable of function parameters.
677 *
678 * This represents the storage. The paramaters passed in a particular
679 * call will be in ir_call::actual_paramaters.
680 */
681 struct exec_list parameters;
682
683 /** Whether or not this function has a body (which may be empty). */
684 unsigned is_defined:1;
685
686 /** Whether or not this function signature is a built-in. */
687 bool is_builtin() const;
688
689 /** Whether or not a built-in is available for this shader. */
690 bool is_builtin_available(const _mesa_glsl_parse_state *state) const;
691
692 /** Body of instructions in the function. */
693 struct exec_list body;
694
695 private:
696 /**
697 * A function pointer to a predicate that answers whether a built-in
698 * function is available in the current shader. NULL if not a built-in.
699 */
700 builtin_available_predicate builtin_avail;
701
702 /** Function of which this signature is one overload. */
703 class ir_function *_function;
704
705 /** Function signature of which this one is a prototype clone */
706 const ir_function_signature *origin;
707
708 friend class ir_function;
709
710 /**
711 * Helper function to run a list of instructions for constant
712 * expression evaluation.
713 *
714 * The hash table represents the values of the visible variables.
715 * There are no scoping issues because the table is indexed on
716 * ir_variable pointers, not variable names.
717 *
718 * Returns false if the expression is not constant, true otherwise,
719 * and the value in *result if result is non-NULL.
720 */
721 bool constant_expression_evaluate_expression_list(const struct exec_list &body,
722 struct hash_table *variable_context,
723 ir_constant **result);
724 };
725
726
727 /**
728 * Header for tracking multiple overloaded functions with the same name.
729 * Contains a list of ir_function_signatures representing each of the
730 * actual functions.
731 */
732 class ir_function : public ir_instruction {
733 public:
734 ir_function(const char *name);
735
736 virtual ir_function *clone(void *mem_ctx, struct hash_table *ht) const;
737
738 virtual ir_function *as_function()
739 {
740 return this;
741 }
742
743 virtual void accept(ir_visitor *v)
744 {
745 v->visit(this);
746 }
747
748 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
749
750 void add_signature(ir_function_signature *sig)
751 {
752 sig->_function = this;
753 this->signatures.push_tail(sig);
754 }
755
756 /**
757 * Get an iterator for the set of function signatures
758 */
759 exec_list_iterator iterator()
760 {
761 return signatures.iterator();
762 }
763
764 /**
765 * Find a signature that matches a set of actual parameters, taking implicit
766 * conversions into account. Also flags whether the match was exact.
767 */
768 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
769 const exec_list *actual_param,
770 bool *match_is_exact);
771
772 /**
773 * Find a signature that matches a set of actual parameters, taking implicit
774 * conversions into account.
775 */
776 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
777 const exec_list *actual_param);
778
779 /**
780 * Find a signature that exactly matches a set of actual parameters without
781 * any implicit type conversions.
782 */
783 ir_function_signature *exact_matching_signature(_mesa_glsl_parse_state *state,
784 const exec_list *actual_ps);
785
786 /**
787 * Name of the function.
788 */
789 const char *name;
790
791 /** Whether or not this function has a signature that isn't a built-in. */
792 bool has_user_signature();
793
794 /**
795 * List of ir_function_signature for each overloaded function with this name.
796 */
797 struct exec_list signatures;
798 };
799
800 inline const char *ir_function_signature::function_name() const
801 {
802 return this->_function->name;
803 }
804 /*@}*/
805
806
807 /**
808 * IR instruction representing high-level if-statements
809 */
810 class ir_if : public ir_instruction {
811 public:
812 ir_if(ir_rvalue *condition)
813 : condition(condition)
814 {
815 ir_type = ir_type_if;
816 }
817
818 virtual ir_if *clone(void *mem_ctx, struct hash_table *ht) const;
819
820 virtual ir_if *as_if()
821 {
822 return this;
823 }
824
825 virtual void accept(ir_visitor *v)
826 {
827 v->visit(this);
828 }
829
830 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
831
832 ir_rvalue *condition;
833 /** List of ir_instruction for the body of the then branch */
834 exec_list then_instructions;
835 /** List of ir_instruction for the body of the else branch */
836 exec_list else_instructions;
837 };
838
839
840 /**
841 * IR instruction representing a high-level loop structure.
842 */
843 class ir_loop : public ir_instruction {
844 public:
845 ir_loop();
846
847 virtual ir_loop *clone(void *mem_ctx, struct hash_table *ht) const;
848
849 virtual void accept(ir_visitor *v)
850 {
851 v->visit(this);
852 }
853
854 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
855
856 virtual ir_loop *as_loop()
857 {
858 return this;
859 }
860
861 /**
862 * Get an iterator for the instructions of the loop body
863 */
864 exec_list_iterator iterator()
865 {
866 return body_instructions.iterator();
867 }
868
869 /** List of ir_instruction that make up the body of the loop. */
870 exec_list body_instructions;
871
872 /**
873 * \name Loop counter and controls
874 *
875 * Represents a loop like a FORTRAN \c do-loop.
876 *
877 * \note
878 * If \c from and \c to are the same value, the loop will execute once.
879 */
880 /*@{*/
881 ir_rvalue *from; /** Value of the loop counter on the first
882 * iteration of the loop.
883 */
884 ir_rvalue *to; /** Value of the loop counter on the last
885 * iteration of the loop.
886 */
887 ir_rvalue *increment;
888 ir_variable *counter;
889
890 /**
891 * Comparison operation in the loop terminator.
892 *
893 * If any of the loop control fields are non-\c NULL, this field must be
894 * one of \c ir_binop_less, \c ir_binop_greater, \c ir_binop_lequal,
895 * \c ir_binop_gequal, \c ir_binop_equal, or \c ir_binop_nequal.
896 */
897 int cmp;
898 /*@}*/
899 };
900
901
902 class ir_assignment : public ir_instruction {
903 public:
904 ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs, ir_rvalue *condition = NULL);
905
906 /**
907 * Construct an assignment with an explicit write mask
908 *
909 * \note
910 * Since a write mask is supplied, the LHS must already be a bare
911 * \c ir_dereference. The cannot be any swizzles in the LHS.
912 */
913 ir_assignment(ir_dereference *lhs, ir_rvalue *rhs, ir_rvalue *condition,
914 unsigned write_mask);
915
916 virtual ir_assignment *clone(void *mem_ctx, struct hash_table *ht) const;
917
918 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
919
920 virtual void accept(ir_visitor *v)
921 {
922 v->visit(this);
923 }
924
925 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
926
927 virtual ir_assignment * as_assignment()
928 {
929 return this;
930 }
931
932 /**
933 * Get a whole variable written by an assignment
934 *
935 * If the LHS of the assignment writes a whole variable, the variable is
936 * returned. Otherwise \c NULL is returned. Examples of whole-variable
937 * assignment are:
938 *
939 * - Assigning to a scalar
940 * - Assigning to all components of a vector
941 * - Whole array (or matrix) assignment
942 * - Whole structure assignment
943 */
944 ir_variable *whole_variable_written();
945
946 /**
947 * Set the LHS of an assignment
948 */
949 void set_lhs(ir_rvalue *lhs);
950
951 /**
952 * Left-hand side of the assignment.
953 *
954 * This should be treated as read only. If you need to set the LHS of an
955 * assignment, use \c ir_assignment::set_lhs.
956 */
957 ir_dereference *lhs;
958
959 /**
960 * Value being assigned
961 */
962 ir_rvalue *rhs;
963
964 /**
965 * Optional condition for the assignment.
966 */
967 ir_rvalue *condition;
968
969
970 /**
971 * Component mask written
972 *
973 * For non-vector types in the LHS, this field will be zero. For vector
974 * types, a bit will be set for each component that is written. Note that
975 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
976 *
977 * A partially-set write mask means that each enabled channel gets
978 * the value from a consecutive channel of the rhs. For example,
979 * to write just .xyw of gl_FrontColor with color:
980 *
981 * (assign (constant bool (1)) (xyw)
982 * (var_ref gl_FragColor)
983 * (swiz xyw (var_ref color)))
984 */
985 unsigned write_mask:4;
986 };
987
988 /* Update ir_expression::get_num_operands() and operator_strs when
989 * updating this list.
990 */
991 enum ir_expression_operation {
992 ir_unop_bit_not,
993 ir_unop_logic_not,
994 ir_unop_neg,
995 ir_unop_abs,
996 ir_unop_sign,
997 ir_unop_rcp,
998 ir_unop_rsq,
999 ir_unop_sqrt,
1000 ir_unop_exp, /**< Log base e on gentype */
1001 ir_unop_log, /**< Natural log on gentype */
1002 ir_unop_exp2,
1003 ir_unop_log2,
1004 ir_unop_f2i, /**< Float-to-integer conversion. */
1005 ir_unop_f2u, /**< Float-to-unsigned conversion. */
1006 ir_unop_i2f, /**< Integer-to-float conversion. */
1007 ir_unop_f2b, /**< Float-to-boolean conversion */
1008 ir_unop_b2f, /**< Boolean-to-float conversion */
1009 ir_unop_i2b, /**< int-to-boolean conversion */
1010 ir_unop_b2i, /**< Boolean-to-int conversion */
1011 ir_unop_u2f, /**< Unsigned-to-float conversion. */
1012 ir_unop_i2u, /**< Integer-to-unsigned conversion. */
1013 ir_unop_u2i, /**< Unsigned-to-integer conversion. */
1014 ir_unop_bitcast_i2f, /**< Bit-identical int-to-float "conversion" */
1015 ir_unop_bitcast_f2i, /**< Bit-identical float-to-int "conversion" */
1016 ir_unop_bitcast_u2f, /**< Bit-identical uint-to-float "conversion" */
1017 ir_unop_bitcast_f2u, /**< Bit-identical float-to-uint "conversion" */
1018 ir_unop_any,
1019
1020 /**
1021 * \name Unary floating-point rounding operations.
1022 */
1023 /*@{*/
1024 ir_unop_trunc,
1025 ir_unop_ceil,
1026 ir_unop_floor,
1027 ir_unop_fract,
1028 ir_unop_round_even,
1029 /*@}*/
1030
1031 /**
1032 * \name Trigonometric operations.
1033 */
1034 /*@{*/
1035 ir_unop_sin,
1036 ir_unop_cos,
1037 ir_unop_sin_reduced, /**< Reduced range sin. [-pi, pi] */
1038 ir_unop_cos_reduced, /**< Reduced range cos. [-pi, pi] */
1039 /*@}*/
1040
1041 /**
1042 * \name Partial derivatives.
1043 */
1044 /*@{*/
1045 ir_unop_dFdx,
1046 ir_unop_dFdy,
1047 /*@}*/
1048
1049 /**
1050 * \name Floating point pack and unpack operations.
1051 */
1052 /*@{*/
1053 ir_unop_pack_snorm_2x16,
1054 ir_unop_pack_snorm_4x8,
1055 ir_unop_pack_unorm_2x16,
1056 ir_unop_pack_unorm_4x8,
1057 ir_unop_pack_half_2x16,
1058 ir_unop_unpack_snorm_2x16,
1059 ir_unop_unpack_snorm_4x8,
1060 ir_unop_unpack_unorm_2x16,
1061 ir_unop_unpack_unorm_4x8,
1062 ir_unop_unpack_half_2x16,
1063 /*@}*/
1064
1065 /**
1066 * \name Lowered floating point unpacking operations.
1067 *
1068 * \see lower_packing_builtins_visitor::split_unpack_half_2x16
1069 */
1070 /*@{*/
1071 ir_unop_unpack_half_2x16_split_x,
1072 ir_unop_unpack_half_2x16_split_y,
1073 /*@}*/
1074
1075 /**
1076 * \name Bit operations, part of ARB_gpu_shader5.
1077 */
1078 /*@{*/
1079 ir_unop_bitfield_reverse,
1080 ir_unop_bit_count,
1081 ir_unop_find_msb,
1082 ir_unop_find_lsb,
1083 /*@}*/
1084
1085 ir_unop_noise,
1086
1087 /**
1088 * A sentinel marking the last of the unary operations.
1089 */
1090 ir_last_unop = ir_unop_noise,
1091
1092 ir_binop_add,
1093 ir_binop_sub,
1094 ir_binop_mul,
1095 ir_binop_div,
1096
1097 /**
1098 * Takes one of two combinations of arguments:
1099 *
1100 * - mod(vecN, vecN)
1101 * - mod(vecN, float)
1102 *
1103 * Does not take integer types.
1104 */
1105 ir_binop_mod,
1106
1107 /**
1108 * \name Binary comparison operators which return a boolean vector.
1109 * The type of both operands must be equal.
1110 */
1111 /*@{*/
1112 ir_binop_less,
1113 ir_binop_greater,
1114 ir_binop_lequal,
1115 ir_binop_gequal,
1116 ir_binop_equal,
1117 ir_binop_nequal,
1118 /**
1119 * Returns single boolean for whether all components of operands[0]
1120 * equal the components of operands[1].
1121 */
1122 ir_binop_all_equal,
1123 /**
1124 * Returns single boolean for whether any component of operands[0]
1125 * is not equal to the corresponding component of operands[1].
1126 */
1127 ir_binop_any_nequal,
1128 /*@}*/
1129
1130 /**
1131 * \name Bit-wise binary operations.
1132 */
1133 /*@{*/
1134 ir_binop_lshift,
1135 ir_binop_rshift,
1136 ir_binop_bit_and,
1137 ir_binop_bit_xor,
1138 ir_binop_bit_or,
1139 /*@}*/
1140
1141 ir_binop_logic_and,
1142 ir_binop_logic_xor,
1143 ir_binop_logic_or,
1144
1145 ir_binop_dot,
1146 ir_binop_min,
1147 ir_binop_max,
1148
1149 ir_binop_pow,
1150
1151 /**
1152 * \name Lowered floating point packing operations.
1153 *
1154 * \see lower_packing_builtins_visitor::split_pack_half_2x16
1155 */
1156 /*@{*/
1157 ir_binop_pack_half_2x16_split,
1158 /*@}*/
1159
1160 /**
1161 * \name First half of a lowered bitfieldInsert() operation.
1162 *
1163 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1164 */
1165 /*@{*/
1166 ir_binop_bfm,
1167 /*@}*/
1168
1169 /**
1170 * Load a value the size of a given GLSL type from a uniform block.
1171 *
1172 * operand0 is the ir_constant uniform block index in the linked shader.
1173 * operand1 is a byte offset within the uniform block.
1174 */
1175 ir_binop_ubo_load,
1176
1177 /**
1178 * \name Multiplies a number by two to a power, part of ARB_gpu_shader5.
1179 */
1180 /*@{*/
1181 ir_binop_ldexp,
1182 /*@}*/
1183
1184 /**
1185 * Extract a scalar from a vector
1186 *
1187 * operand0 is the vector
1188 * operand1 is the index of the field to read from operand0
1189 */
1190 ir_binop_vector_extract,
1191
1192 /**
1193 * A sentinel marking the last of the binary operations.
1194 */
1195 ir_last_binop = ir_binop_vector_extract,
1196
1197 /**
1198 * \name Fused floating-point multiply-add, part of ARB_gpu_shader5.
1199 */
1200 /*@{*/
1201 ir_triop_fma,
1202 /*@}*/
1203
1204 ir_triop_lrp,
1205
1206 /**
1207 * \name Conditional Select
1208 *
1209 * A vector conditional select instruction (like ?:, but operating per-
1210 * component on vectors).
1211 *
1212 * \see lower_instructions_visitor::ldexp_to_arith
1213 */
1214 /*@{*/
1215 ir_triop_csel,
1216 /*@}*/
1217
1218 /**
1219 * \name Second half of a lowered bitfieldInsert() operation.
1220 *
1221 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1222 */
1223 /*@{*/
1224 ir_triop_bfi,
1225 /*@}*/
1226
1227 ir_triop_bitfield_extract,
1228
1229 /**
1230 * Generate a value with one field of a vector changed
1231 *
1232 * operand0 is the vector
1233 * operand1 is the value to write into the vector result
1234 * operand2 is the index in operand0 to be modified
1235 */
1236 ir_triop_vector_insert,
1237
1238 /**
1239 * A sentinel marking the last of the ternary operations.
1240 */
1241 ir_last_triop = ir_triop_vector_insert,
1242
1243 ir_quadop_bitfield_insert,
1244
1245 ir_quadop_vector,
1246
1247 /**
1248 * A sentinel marking the last of the ternary operations.
1249 */
1250 ir_last_quadop = ir_quadop_vector,
1251
1252 /**
1253 * A sentinel marking the last of all operations.
1254 */
1255 ir_last_opcode = ir_quadop_vector
1256 };
1257
1258 class ir_expression : public ir_rvalue {
1259 public:
1260 ir_expression(int op, const struct glsl_type *type,
1261 ir_rvalue *op0, ir_rvalue *op1 = NULL,
1262 ir_rvalue *op2 = NULL, ir_rvalue *op3 = NULL);
1263
1264 /**
1265 * Constructor for unary operation expressions
1266 */
1267 ir_expression(int op, ir_rvalue *);
1268
1269 /**
1270 * Constructor for binary operation expressions
1271 */
1272 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1);
1273
1274 /**
1275 * Constructor for ternary operation expressions
1276 */
1277 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1, ir_rvalue *op2);
1278
1279 virtual ir_expression *as_expression()
1280 {
1281 return this;
1282 }
1283
1284 virtual ir_expression *clone(void *mem_ctx, struct hash_table *ht) const;
1285
1286 /**
1287 * Attempt to constant-fold the expression
1288 *
1289 * The "variable_context" hash table links ir_variable * to ir_constant *
1290 * that represent the variables' values. \c NULL represents an empty
1291 * context.
1292 *
1293 * If the expression cannot be constant folded, this method will return
1294 * \c NULL.
1295 */
1296 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1297
1298 /**
1299 * Determine the number of operands used by an expression
1300 */
1301 static unsigned int get_num_operands(ir_expression_operation);
1302
1303 /**
1304 * Determine the number of operands used by an expression
1305 */
1306 unsigned int get_num_operands() const
1307 {
1308 return (this->operation == ir_quadop_vector)
1309 ? this->type->vector_elements : get_num_operands(operation);
1310 }
1311
1312 /**
1313 * Return a string representing this expression's operator.
1314 */
1315 const char *operator_string();
1316
1317 /**
1318 * Return a string representing this expression's operator.
1319 */
1320 static const char *operator_string(ir_expression_operation);
1321
1322
1323 /**
1324 * Do a reverse-lookup to translate the given string into an operator.
1325 */
1326 static ir_expression_operation get_operator(const char *);
1327
1328 virtual void accept(ir_visitor *v)
1329 {
1330 v->visit(this);
1331 }
1332
1333 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1334
1335 ir_expression_operation operation;
1336 ir_rvalue *operands[4];
1337 };
1338
1339
1340 /**
1341 * HIR instruction representing a high-level function call, containing a list
1342 * of parameters and returning a value in the supplied temporary.
1343 */
1344 class ir_call : public ir_instruction {
1345 public:
1346 ir_call(ir_function_signature *callee,
1347 ir_dereference_variable *return_deref,
1348 exec_list *actual_parameters)
1349 : return_deref(return_deref), callee(callee)
1350 {
1351 ir_type = ir_type_call;
1352 assert(callee->return_type != NULL);
1353 actual_parameters->move_nodes_to(& this->actual_parameters);
1354 this->use_builtin = callee->is_builtin();
1355 }
1356
1357 virtual ir_call *clone(void *mem_ctx, struct hash_table *ht) const;
1358
1359 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1360
1361 virtual ir_call *as_call()
1362 {
1363 return this;
1364 }
1365
1366 virtual void accept(ir_visitor *v)
1367 {
1368 v->visit(this);
1369 }
1370
1371 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1372
1373 /**
1374 * Get an iterator for the set of acutal parameters
1375 */
1376 exec_list_iterator iterator()
1377 {
1378 return actual_parameters.iterator();
1379 }
1380
1381 /**
1382 * Get the name of the function being called.
1383 */
1384 const char *callee_name() const
1385 {
1386 return callee->function_name();
1387 }
1388
1389 /**
1390 * Generates an inline version of the function before @ir,
1391 * storing the return value in return_deref.
1392 */
1393 void generate_inline(ir_instruction *ir);
1394
1395 /**
1396 * Storage for the function's return value.
1397 * This must be NULL if the return type is void.
1398 */
1399 ir_dereference_variable *return_deref;
1400
1401 /**
1402 * The specific function signature being called.
1403 */
1404 ir_function_signature *callee;
1405
1406 /* List of ir_rvalue of paramaters passed in this call. */
1407 exec_list actual_parameters;
1408
1409 /** Should this call only bind to a built-in function? */
1410 bool use_builtin;
1411 };
1412
1413
1414 /**
1415 * \name Jump-like IR instructions.
1416 *
1417 * These include \c break, \c continue, \c return, and \c discard.
1418 */
1419 /*@{*/
1420 class ir_jump : public ir_instruction {
1421 protected:
1422 ir_jump()
1423 {
1424 ir_type = ir_type_unset;
1425 }
1426
1427 public:
1428 virtual ir_jump *as_jump()
1429 {
1430 return this;
1431 }
1432 };
1433
1434 class ir_return : public ir_jump {
1435 public:
1436 ir_return()
1437 : value(NULL)
1438 {
1439 this->ir_type = ir_type_return;
1440 }
1441
1442 ir_return(ir_rvalue *value)
1443 : value(value)
1444 {
1445 this->ir_type = ir_type_return;
1446 }
1447
1448 virtual ir_return *clone(void *mem_ctx, struct hash_table *) const;
1449
1450 virtual ir_return *as_return()
1451 {
1452 return this;
1453 }
1454
1455 ir_rvalue *get_value() const
1456 {
1457 return value;
1458 }
1459
1460 virtual void accept(ir_visitor *v)
1461 {
1462 v->visit(this);
1463 }
1464
1465 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1466
1467 ir_rvalue *value;
1468 };
1469
1470
1471 /**
1472 * Jump instructions used inside loops
1473 *
1474 * These include \c break and \c continue. The \c break within a loop is
1475 * different from the \c break within a switch-statement.
1476 *
1477 * \sa ir_switch_jump
1478 */
1479 class ir_loop_jump : public ir_jump {
1480 public:
1481 enum jump_mode {
1482 jump_break,
1483 jump_continue
1484 };
1485
1486 ir_loop_jump(jump_mode mode)
1487 {
1488 this->ir_type = ir_type_loop_jump;
1489 this->mode = mode;
1490 }
1491
1492 virtual ir_loop_jump *clone(void *mem_ctx, struct hash_table *) const;
1493
1494 virtual void accept(ir_visitor *v)
1495 {
1496 v->visit(this);
1497 }
1498
1499 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1500
1501 bool is_break() const
1502 {
1503 return mode == jump_break;
1504 }
1505
1506 bool is_continue() const
1507 {
1508 return mode == jump_continue;
1509 }
1510
1511 /** Mode selector for the jump instruction. */
1512 enum jump_mode mode;
1513 };
1514
1515 /**
1516 * IR instruction representing discard statements.
1517 */
1518 class ir_discard : public ir_jump {
1519 public:
1520 ir_discard()
1521 {
1522 this->ir_type = ir_type_discard;
1523 this->condition = NULL;
1524 }
1525
1526 ir_discard(ir_rvalue *cond)
1527 {
1528 this->ir_type = ir_type_discard;
1529 this->condition = cond;
1530 }
1531
1532 virtual ir_discard *clone(void *mem_ctx, struct hash_table *ht) const;
1533
1534 virtual void accept(ir_visitor *v)
1535 {
1536 v->visit(this);
1537 }
1538
1539 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1540
1541 virtual ir_discard *as_discard()
1542 {
1543 return this;
1544 }
1545
1546 ir_rvalue *condition;
1547 };
1548 /*@}*/
1549
1550
1551 /**
1552 * Texture sampling opcodes used in ir_texture
1553 */
1554 enum ir_texture_opcode {
1555 ir_tex, /**< Regular texture look-up */
1556 ir_txb, /**< Texture look-up with LOD bias */
1557 ir_txl, /**< Texture look-up with explicit LOD */
1558 ir_txd, /**< Texture look-up with partial derivatvies */
1559 ir_txf, /**< Texel fetch with explicit LOD */
1560 ir_txf_ms, /**< Multisample texture fetch */
1561 ir_txs, /**< Texture size */
1562 ir_lod, /**< Texture lod query */
1563 ir_tg4 /**< Texture gather */
1564 };
1565
1566
1567 /**
1568 * IR instruction to sample a texture
1569 *
1570 * The specific form of the IR instruction depends on the \c mode value
1571 * selected from \c ir_texture_opcodes. In the printed IR, these will
1572 * appear as:
1573 *
1574 * Texel offset (0 or an expression)
1575 * | Projection divisor
1576 * | | Shadow comparitor
1577 * | | |
1578 * v v v
1579 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1580 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1581 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1582 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1583 * (txf <type> <sampler> <coordinate> 0 <lod>)
1584 * (txf_ms
1585 * <type> <sampler> <coordinate> <sample_index>)
1586 * (txs <type> <sampler> <lod>)
1587 * (lod <type> <sampler> <coordinate>)
1588 * (tg4 <type> <sampler> <coordinate> 0)
1589 */
1590 class ir_texture : public ir_rvalue {
1591 public:
1592 ir_texture(enum ir_texture_opcode op)
1593 : op(op), sampler(NULL), coordinate(NULL), projector(NULL),
1594 shadow_comparitor(NULL), offset(NULL)
1595 {
1596 this->ir_type = ir_type_texture;
1597 memset(&lod_info, 0, sizeof(lod_info));
1598 }
1599
1600 virtual ir_texture *clone(void *mem_ctx, struct hash_table *) const;
1601
1602 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1603
1604 virtual void accept(ir_visitor *v)
1605 {
1606 v->visit(this);
1607 }
1608
1609 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1610
1611 /**
1612 * Return a string representing the ir_texture_opcode.
1613 */
1614 const char *opcode_string();
1615
1616 /** Set the sampler and type. */
1617 void set_sampler(ir_dereference *sampler, const glsl_type *type);
1618
1619 /**
1620 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1621 */
1622 static ir_texture_opcode get_opcode(const char *);
1623
1624 enum ir_texture_opcode op;
1625
1626 /** Sampler to use for the texture access. */
1627 ir_dereference *sampler;
1628
1629 /** Texture coordinate to sample */
1630 ir_rvalue *coordinate;
1631
1632 /**
1633 * Value used for projective divide.
1634 *
1635 * If there is no projective divide (the common case), this will be
1636 * \c NULL. Optimization passes should check for this to point to a constant
1637 * of 1.0 and replace that with \c NULL.
1638 */
1639 ir_rvalue *projector;
1640
1641 /**
1642 * Coordinate used for comparison on shadow look-ups.
1643 *
1644 * If there is no shadow comparison, this will be \c NULL. For the
1645 * \c ir_txf opcode, this *must* be \c NULL.
1646 */
1647 ir_rvalue *shadow_comparitor;
1648
1649 /** Texel offset. */
1650 ir_rvalue *offset;
1651
1652 union {
1653 ir_rvalue *lod; /**< Floating point LOD */
1654 ir_rvalue *bias; /**< Floating point LOD bias */
1655 ir_rvalue *sample_index; /**< MSAA sample index */
1656 struct {
1657 ir_rvalue *dPdx; /**< Partial derivative of coordinate wrt X */
1658 ir_rvalue *dPdy; /**< Partial derivative of coordinate wrt Y */
1659 } grad;
1660 } lod_info;
1661 };
1662
1663
1664 struct ir_swizzle_mask {
1665 unsigned x:2;
1666 unsigned y:2;
1667 unsigned z:2;
1668 unsigned w:2;
1669
1670 /**
1671 * Number of components in the swizzle.
1672 */
1673 unsigned num_components:3;
1674
1675 /**
1676 * Does the swizzle contain duplicate components?
1677 *
1678 * L-value swizzles cannot contain duplicate components.
1679 */
1680 unsigned has_duplicates:1;
1681 };
1682
1683
1684 class ir_swizzle : public ir_rvalue {
1685 public:
1686 ir_swizzle(ir_rvalue *, unsigned x, unsigned y, unsigned z, unsigned w,
1687 unsigned count);
1688
1689 ir_swizzle(ir_rvalue *val, const unsigned *components, unsigned count);
1690
1691 ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask);
1692
1693 virtual ir_swizzle *clone(void *mem_ctx, struct hash_table *) const;
1694
1695 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1696
1697 virtual ir_swizzle *as_swizzle()
1698 {
1699 return this;
1700 }
1701
1702 /**
1703 * Construct an ir_swizzle from the textual representation. Can fail.
1704 */
1705 static ir_swizzle *create(ir_rvalue *, const char *, unsigned vector_length);
1706
1707 virtual void accept(ir_visitor *v)
1708 {
1709 v->visit(this);
1710 }
1711
1712 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1713
1714 bool is_lvalue() const
1715 {
1716 return val->is_lvalue() && !mask.has_duplicates;
1717 }
1718
1719 /**
1720 * Get the variable that is ultimately referenced by an r-value
1721 */
1722 virtual ir_variable *variable_referenced() const;
1723
1724 ir_rvalue *val;
1725 ir_swizzle_mask mask;
1726
1727 private:
1728 /**
1729 * Initialize the mask component of a swizzle
1730 *
1731 * This is used by the \c ir_swizzle constructors.
1732 */
1733 void init_mask(const unsigned *components, unsigned count);
1734 };
1735
1736
1737 class ir_dereference : public ir_rvalue {
1738 public:
1739 virtual ir_dereference *clone(void *mem_ctx, struct hash_table *) const = 0;
1740
1741 virtual ir_dereference *as_dereference()
1742 {
1743 return this;
1744 }
1745
1746 bool is_lvalue() const;
1747
1748 /**
1749 * Get the variable that is ultimately referenced by an r-value
1750 */
1751 virtual ir_variable *variable_referenced() const = 0;
1752
1753 /**
1754 * Get the constant that is ultimately referenced by an r-value,
1755 * in a constant expression evaluation context.
1756 *
1757 * The offset is used when the reference is to a specific column of
1758 * a matrix.
1759 */
1760 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const = 0;
1761 };
1762
1763
1764 class ir_dereference_variable : public ir_dereference {
1765 public:
1766 ir_dereference_variable(ir_variable *var);
1767
1768 virtual ir_dereference_variable *clone(void *mem_ctx,
1769 struct hash_table *) const;
1770
1771 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1772
1773 virtual ir_dereference_variable *as_dereference_variable()
1774 {
1775 return this;
1776 }
1777
1778 /**
1779 * Get the variable that is ultimately referenced by an r-value
1780 */
1781 virtual ir_variable *variable_referenced() const
1782 {
1783 return this->var;
1784 }
1785
1786 /**
1787 * Get the constant that is ultimately referenced by an r-value,
1788 * in a constant expression evaluation context.
1789 *
1790 * The offset is used when the reference is to a specific column of
1791 * a matrix.
1792 */
1793 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
1794
1795 virtual ir_variable *whole_variable_referenced()
1796 {
1797 /* ir_dereference_variable objects always dereference the entire
1798 * variable. However, if this dereference is dereferenced by anything
1799 * else, the complete deferefernce chain is not a whole-variable
1800 * dereference. This method should only be called on the top most
1801 * ir_rvalue in a dereference chain.
1802 */
1803 return this->var;
1804 }
1805
1806 virtual void accept(ir_visitor *v)
1807 {
1808 v->visit(this);
1809 }
1810
1811 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1812
1813 /**
1814 * Object being dereferenced.
1815 */
1816 ir_variable *var;
1817 };
1818
1819
1820 class ir_dereference_array : public ir_dereference {
1821 public:
1822 ir_dereference_array(ir_rvalue *value, ir_rvalue *array_index);
1823
1824 ir_dereference_array(ir_variable *var, ir_rvalue *array_index);
1825
1826 virtual ir_dereference_array *clone(void *mem_ctx,
1827 struct hash_table *) const;
1828
1829 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1830
1831 virtual ir_dereference_array *as_dereference_array()
1832 {
1833 return this;
1834 }
1835
1836 /**
1837 * Get the variable that is ultimately referenced by an r-value
1838 */
1839 virtual ir_variable *variable_referenced() const
1840 {
1841 return this->array->variable_referenced();
1842 }
1843
1844 /**
1845 * Get the constant that is ultimately referenced by an r-value,
1846 * in a constant expression evaluation context.
1847 *
1848 * The offset is used when the reference is to a specific column of
1849 * a matrix.
1850 */
1851 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
1852
1853 virtual void accept(ir_visitor *v)
1854 {
1855 v->visit(this);
1856 }
1857
1858 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1859
1860 ir_rvalue *array;
1861 ir_rvalue *array_index;
1862
1863 private:
1864 void set_array(ir_rvalue *value);
1865 };
1866
1867
1868 class ir_dereference_record : public ir_dereference {
1869 public:
1870 ir_dereference_record(ir_rvalue *value, const char *field);
1871
1872 ir_dereference_record(ir_variable *var, const char *field);
1873
1874 virtual ir_dereference_record *clone(void *mem_ctx,
1875 struct hash_table *) const;
1876
1877 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1878
1879 virtual ir_dereference_record *as_dereference_record()
1880 {
1881 return this;
1882 }
1883
1884 /**
1885 * Get the variable that is ultimately referenced by an r-value
1886 */
1887 virtual ir_variable *variable_referenced() const
1888 {
1889 return this->record->variable_referenced();
1890 }
1891
1892 /**
1893 * Get the constant that is ultimately referenced by an r-value,
1894 * in a constant expression evaluation context.
1895 *
1896 * The offset is used when the reference is to a specific column of
1897 * a matrix.
1898 */
1899 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
1900
1901 virtual void accept(ir_visitor *v)
1902 {
1903 v->visit(this);
1904 }
1905
1906 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1907
1908 ir_rvalue *record;
1909 const char *field;
1910 };
1911
1912
1913 /**
1914 * Data stored in an ir_constant
1915 */
1916 union ir_constant_data {
1917 unsigned u[16];
1918 int i[16];
1919 float f[16];
1920 bool b[16];
1921 };
1922
1923
1924 class ir_constant : public ir_rvalue {
1925 public:
1926 ir_constant(const struct glsl_type *type, const ir_constant_data *data);
1927 ir_constant(bool b, unsigned vector_elements=1);
1928 ir_constant(unsigned int u, unsigned vector_elements=1);
1929 ir_constant(int i, unsigned vector_elements=1);
1930 ir_constant(float f, unsigned vector_elements=1);
1931
1932 /**
1933 * Construct an ir_constant from a list of ir_constant values
1934 */
1935 ir_constant(const struct glsl_type *type, exec_list *values);
1936
1937 /**
1938 * Construct an ir_constant from a scalar component of another ir_constant
1939 *
1940 * The new \c ir_constant inherits the type of the component from the
1941 * source constant.
1942 *
1943 * \note
1944 * In the case of a matrix constant, the new constant is a scalar, \b not
1945 * a vector.
1946 */
1947 ir_constant(const ir_constant *c, unsigned i);
1948
1949 /**
1950 * Return a new ir_constant of the specified type containing all zeros.
1951 */
1952 static ir_constant *zero(void *mem_ctx, const glsl_type *type);
1953
1954 virtual ir_constant *clone(void *mem_ctx, struct hash_table *) const;
1955
1956 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1957
1958 virtual ir_constant *as_constant()
1959 {
1960 return this;
1961 }
1962
1963 virtual void accept(ir_visitor *v)
1964 {
1965 v->visit(this);
1966 }
1967
1968 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1969
1970 /**
1971 * Get a particular component of a constant as a specific type
1972 *
1973 * This is useful, for example, to get a value from an integer constant
1974 * as a float or bool. This appears frequently when constructors are
1975 * called with all constant parameters.
1976 */
1977 /*@{*/
1978 bool get_bool_component(unsigned i) const;
1979 float get_float_component(unsigned i) const;
1980 int get_int_component(unsigned i) const;
1981 unsigned get_uint_component(unsigned i) const;
1982 /*@}*/
1983
1984 ir_constant *get_array_element(unsigned i) const;
1985
1986 ir_constant *get_record_field(const char *name);
1987
1988 /**
1989 * Copy the values on another constant at a given offset.
1990 *
1991 * The offset is ignored for array or struct copies, it's only for
1992 * scalars or vectors into vectors or matrices.
1993 *
1994 * With identical types on both sides and zero offset it's clone()
1995 * without creating a new object.
1996 */
1997
1998 void copy_offset(ir_constant *src, int offset);
1999
2000 /**
2001 * Copy the values on another constant at a given offset and
2002 * following an assign-like mask.
2003 *
2004 * The mask is ignored for scalars.
2005 *
2006 * Note that this function only handles what assign can handle,
2007 * i.e. at most a vector as source and a column of a matrix as
2008 * destination.
2009 */
2010
2011 void copy_masked_offset(ir_constant *src, int offset, unsigned int mask);
2012
2013 /**
2014 * Determine whether a constant has the same value as another constant
2015 *
2016 * \sa ir_constant::is_zero, ir_constant::is_one,
2017 * ir_constant::is_negative_one, ir_constant::is_basis
2018 */
2019 bool has_value(const ir_constant *) const;
2020
2021 virtual bool is_zero() const;
2022 virtual bool is_one() const;
2023 virtual bool is_negative_one() const;
2024 virtual bool is_basis() const;
2025
2026 /**
2027 * Value of the constant.
2028 *
2029 * The field used to back the values supplied by the constant is determined
2030 * by the type associated with the \c ir_instruction. Constants may be
2031 * scalars, vectors, or matrices.
2032 */
2033 union ir_constant_data value;
2034
2035 /* Array elements */
2036 ir_constant **array_elements;
2037
2038 /* Structure fields */
2039 exec_list components;
2040
2041 private:
2042 /**
2043 * Parameterless constructor only used by the clone method
2044 */
2045 ir_constant(void);
2046 };
2047
2048 /*@}*/
2049
2050 /**
2051 * IR instruction to emit a vertex in a geometry shader.
2052 */
2053 class ir_emit_vertex : public ir_instruction {
2054 public:
2055 ir_emit_vertex()
2056 {
2057 ir_type = ir_type_emit_vertex;
2058 }
2059
2060 virtual void accept(ir_visitor *v)
2061 {
2062 v->visit(this);
2063 }
2064
2065 virtual ir_emit_vertex *clone(void *mem_ctx, struct hash_table *) const
2066 {
2067 return new(mem_ctx) ir_emit_vertex();
2068 }
2069
2070 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2071 };
2072
2073 /**
2074 * IR instruction to complete the current primitive and start a new one in a
2075 * geometry shader.
2076 */
2077 class ir_end_primitive : public ir_instruction {
2078 public:
2079 ir_end_primitive()
2080 {
2081 ir_type = ir_type_end_primitive;
2082 }
2083
2084 virtual void accept(ir_visitor *v)
2085 {
2086 v->visit(this);
2087 }
2088
2089 virtual ir_end_primitive *clone(void *mem_ctx, struct hash_table *) const
2090 {
2091 return new(mem_ctx) ir_end_primitive();
2092 }
2093
2094 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2095 };
2096
2097 /**
2098 * Apply a visitor to each IR node in a list
2099 */
2100 void
2101 visit_exec_list(exec_list *list, ir_visitor *visitor);
2102
2103 /**
2104 * Validate invariants on each IR node in a list
2105 */
2106 void validate_ir_tree(exec_list *instructions);
2107
2108 struct _mesa_glsl_parse_state;
2109 struct gl_shader_program;
2110
2111 /**
2112 * Detect whether an unlinked shader contains static recursion
2113 *
2114 * If the list of instructions is determined to contain static recursion,
2115 * \c _mesa_glsl_error will be called to emit error messages for each function
2116 * that is in the recursion cycle.
2117 */
2118 void
2119 detect_recursion_unlinked(struct _mesa_glsl_parse_state *state,
2120 exec_list *instructions);
2121
2122 /**
2123 * Detect whether a linked shader contains static recursion
2124 *
2125 * If the list of instructions is determined to contain static recursion,
2126 * \c link_error_printf will be called to emit error messages for each function
2127 * that is in the recursion cycle. In addition,
2128 * \c gl_shader_program::LinkStatus will be set to false.
2129 */
2130 void
2131 detect_recursion_linked(struct gl_shader_program *prog,
2132 exec_list *instructions);
2133
2134 /**
2135 * Make a clone of each IR instruction in a list
2136 *
2137 * \param in List of IR instructions that are to be cloned
2138 * \param out List to hold the cloned instructions
2139 */
2140 void
2141 clone_ir_list(void *mem_ctx, exec_list *out, const exec_list *in);
2142
2143 extern void
2144 _mesa_glsl_initialize_variables(exec_list *instructions,
2145 struct _mesa_glsl_parse_state *state);
2146
2147 extern void
2148 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state *state);
2149
2150 extern void
2151 _mesa_glsl_initialize_builtin_functions();
2152
2153 extern ir_function_signature *
2154 _mesa_glsl_find_builtin_function(_mesa_glsl_parse_state *state,
2155 const char *name, exec_list *actual_parameters);
2156
2157 extern void
2158 _mesa_glsl_release_functions(void);
2159
2160 extern void
2161 _mesa_glsl_release_builtin_functions(void);
2162
2163 extern void
2164 reparent_ir(exec_list *list, void *mem_ctx);
2165
2166 struct glsl_symbol_table;
2167
2168 extern void
2169 import_prototypes(const exec_list *source, exec_list *dest,
2170 struct glsl_symbol_table *symbols, void *mem_ctx);
2171
2172 extern bool
2173 ir_has_call(ir_instruction *ir);
2174
2175 extern void
2176 do_set_program_inouts(exec_list *instructions, struct gl_program *prog,
2177 GLenum shader_type);
2178
2179 extern char *
2180 prototype_string(const glsl_type *return_type, const char *name,
2181 exec_list *parameters);
2182
2183 extern "C" {
2184 #endif /* __cplusplus */
2185
2186 extern void _mesa_print_ir(struct exec_list *instructions,
2187 struct _mesa_glsl_parse_state *state);
2188
2189 #ifdef __cplusplus
2190 } /* extern "C" */
2191 #endif
2192
2193 unsigned
2194 vertices_per_prim(GLenum prim);
2195
2196 #endif /* IR_H */