glsl: Add an ir_variable::reinit_interface_type() function.
[mesa.git] / src / glsl / ir.h
1 /* -*- c++ -*- */
2 /*
3 * Copyright © 2010 Intel Corporation
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
14 * Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
21 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
22 * DEALINGS IN THE SOFTWARE.
23 */
24
25 #pragma once
26 #ifndef IR_H
27 #define IR_H
28
29 #include <stdio.h>
30 #include <stdlib.h>
31
32 #include "ralloc.h"
33 #include "glsl_types.h"
34 #include "list.h"
35 #include "ir_visitor.h"
36 #include "ir_hierarchical_visitor.h"
37 #include "main/mtypes.h"
38
39 #ifdef __cplusplus
40
41 /**
42 * \defgroup IR Intermediate representation nodes
43 *
44 * @{
45 */
46
47 /**
48 * Class tags
49 *
50 * Each concrete class derived from \c ir_instruction has a value in this
51 * enumerant. The value for the type is stored in \c ir_instruction::ir_type
52 * by the constructor. While using type tags is not very C++, it is extremely
53 * convenient. For example, during debugging you can simply inspect
54 * \c ir_instruction::ir_type to find out the actual type of the object.
55 *
56 * In addition, it is possible to use a switch-statement based on \c
57 * \c ir_instruction::ir_type to select different behavior for different object
58 * types. For functions that have only slight differences for several object
59 * types, this allows writing very straightforward, readable code.
60 */
61 enum ir_node_type {
62 /**
63 * Zero is unused so that the IR validator can detect cases where
64 * \c ir_instruction::ir_type has not been initialized.
65 */
66 ir_type_unset,
67 ir_type_variable,
68 ir_type_assignment,
69 ir_type_call,
70 ir_type_constant,
71 ir_type_dereference_array,
72 ir_type_dereference_record,
73 ir_type_dereference_variable,
74 ir_type_discard,
75 ir_type_expression,
76 ir_type_function,
77 ir_type_function_signature,
78 ir_type_if,
79 ir_type_loop,
80 ir_type_loop_jump,
81 ir_type_return,
82 ir_type_swizzle,
83 ir_type_texture,
84 ir_type_emit_vertex,
85 ir_type_end_primitive,
86 ir_type_max /**< maximum ir_type enum number, for validation */
87 };
88
89 /**
90 * Base class of all IR instructions
91 */
92 class ir_instruction : public exec_node {
93 public:
94 enum ir_node_type ir_type;
95
96 /**
97 * GCC 4.7+ and clang warn when deleting an ir_instruction unless
98 * there's a virtual destructor present. Because we almost
99 * universally use ralloc for our memory management of
100 * ir_instructions, the destructor doesn't need to do any work.
101 */
102 virtual ~ir_instruction()
103 {
104 }
105
106 /** ir_print_visitor helper for debugging. */
107 void print(void) const;
108
109 virtual void accept(ir_visitor *) = 0;
110 virtual ir_visitor_status accept(ir_hierarchical_visitor *) = 0;
111 virtual ir_instruction *clone(void *mem_ctx,
112 struct hash_table *ht) const = 0;
113
114 /**
115 * \name IR instruction downcast functions
116 *
117 * These functions either cast the object to a derived class or return
118 * \c NULL if the object's type does not match the specified derived class.
119 * Additional downcast functions will be added as needed.
120 */
121 /*@{*/
122 virtual class ir_variable * as_variable() { return NULL; }
123 virtual class ir_function * as_function() { return NULL; }
124 virtual class ir_dereference * as_dereference() { return NULL; }
125 virtual class ir_dereference_array * as_dereference_array() { return NULL; }
126 virtual class ir_dereference_variable *as_dereference_variable() { return NULL; }
127 virtual class ir_dereference_record *as_dereference_record() { return NULL; }
128 virtual class ir_expression * as_expression() { return NULL; }
129 virtual class ir_rvalue * as_rvalue() { return NULL; }
130 virtual class ir_loop * as_loop() { return NULL; }
131 virtual class ir_assignment * as_assignment() { return NULL; }
132 virtual class ir_call * as_call() { return NULL; }
133 virtual class ir_return * as_return() { return NULL; }
134 virtual class ir_if * as_if() { return NULL; }
135 virtual class ir_swizzle * as_swizzle() { return NULL; }
136 virtual class ir_constant * as_constant() { return NULL; }
137 virtual class ir_discard * as_discard() { return NULL; }
138 virtual class ir_jump * as_jump() { return NULL; }
139 /*@}*/
140
141 protected:
142 ir_instruction()
143 {
144 ir_type = ir_type_unset;
145 }
146 };
147
148
149 /**
150 * The base class for all "values"/expression trees.
151 */
152 class ir_rvalue : public ir_instruction {
153 public:
154 const struct glsl_type *type;
155
156 virtual ir_rvalue *clone(void *mem_ctx, struct hash_table *) const;
157
158 virtual void accept(ir_visitor *v)
159 {
160 v->visit(this);
161 }
162
163 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
164
165 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
166
167 virtual ir_rvalue * as_rvalue()
168 {
169 return this;
170 }
171
172 ir_rvalue *as_rvalue_to_saturate();
173
174 virtual bool is_lvalue() const
175 {
176 return false;
177 }
178
179 /**
180 * Get the variable that is ultimately referenced by an r-value
181 */
182 virtual ir_variable *variable_referenced() const
183 {
184 return NULL;
185 }
186
187
188 /**
189 * If an r-value is a reference to a whole variable, get that variable
190 *
191 * \return
192 * Pointer to a variable that is completely dereferenced by the r-value. If
193 * the r-value is not a dereference or the dereference does not access the
194 * entire variable (i.e., it's just one array element, struct field), \c NULL
195 * is returned.
196 */
197 virtual ir_variable *whole_variable_referenced()
198 {
199 return NULL;
200 }
201
202 /**
203 * Determine if an r-value has the value zero
204 *
205 * The base implementation of this function always returns \c false. The
206 * \c ir_constant class over-rides this function to return \c true \b only
207 * for vector and scalar types that have all elements set to the value
208 * zero (or \c false for booleans).
209 *
210 * \sa ir_constant::has_value, ir_rvalue::is_one, ir_rvalue::is_negative_one,
211 * ir_constant::is_basis
212 */
213 virtual bool is_zero() const;
214
215 /**
216 * Determine if an r-value has the value one
217 *
218 * The base implementation of this function always returns \c false. The
219 * \c ir_constant class over-rides this function to return \c true \b only
220 * for vector and scalar types that have all elements set to the value
221 * one (or \c true for booleans).
222 *
223 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_negative_one,
224 * ir_constant::is_basis
225 */
226 virtual bool is_one() const;
227
228 /**
229 * Determine if an r-value has the value negative one
230 *
231 * The base implementation of this function always returns \c false. The
232 * \c ir_constant class over-rides this function to return \c true \b only
233 * for vector and scalar types that have all elements set to the value
234 * negative one. For boolean types, the result is always \c false.
235 *
236 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one
237 * ir_constant::is_basis
238 */
239 virtual bool is_negative_one() const;
240
241 /**
242 * Determine if an r-value is a basis vector
243 *
244 * The base implementation of this function always returns \c false. The
245 * \c ir_constant class over-rides this function to return \c true \b only
246 * for vector and scalar types that have one element set to the value one,
247 * and the other elements set to the value zero. For boolean types, the
248 * result is always \c false.
249 *
250 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one,
251 * is_constant::is_negative_one
252 */
253 virtual bool is_basis() const;
254
255
256 /**
257 * Return a generic value of error_type.
258 *
259 * Allocation will be performed with 'mem_ctx' as ralloc owner.
260 */
261 static ir_rvalue *error_value(void *mem_ctx);
262
263 protected:
264 ir_rvalue();
265 };
266
267
268 /**
269 * Variable storage classes
270 */
271 enum ir_variable_mode {
272 ir_var_auto = 0, /**< Function local variables and globals. */
273 ir_var_uniform, /**< Variable declared as a uniform. */
274 ir_var_shader_in,
275 ir_var_shader_out,
276 ir_var_function_in,
277 ir_var_function_out,
278 ir_var_function_inout,
279 ir_var_const_in, /**< "in" param that must be a constant expression */
280 ir_var_system_value, /**< Ex: front-face, instance-id, etc. */
281 ir_var_temporary, /**< Temporary variable generated during compilation. */
282 ir_var_mode_count /**< Number of variable modes */
283 };
284
285 /**
286 * \brief Layout qualifiers for gl_FragDepth.
287 *
288 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
289 * with a layout qualifier.
290 */
291 enum ir_depth_layout {
292 ir_depth_layout_none, /**< No depth layout is specified. */
293 ir_depth_layout_any,
294 ir_depth_layout_greater,
295 ir_depth_layout_less,
296 ir_depth_layout_unchanged
297 };
298
299 /**
300 * \brief Convert depth layout qualifier to string.
301 */
302 const char*
303 depth_layout_string(ir_depth_layout layout);
304
305 /**
306 * Description of built-in state associated with a uniform
307 *
308 * \sa ir_variable::state_slots
309 */
310 struct ir_state_slot {
311 int tokens[5];
312 int swizzle;
313 };
314
315 class ir_variable : public ir_instruction {
316 public:
317 ir_variable(const struct glsl_type *, const char *, ir_variable_mode);
318
319 virtual ir_variable *clone(void *mem_ctx, struct hash_table *ht) const;
320
321 virtual ir_variable *as_variable()
322 {
323 return this;
324 }
325
326 virtual void accept(ir_visitor *v)
327 {
328 v->visit(this);
329 }
330
331 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
332
333
334 /**
335 * Get the string value for the interpolation qualifier
336 *
337 * \return The string that would be used in a shader to specify \c
338 * mode will be returned.
339 *
340 * This function is used to generate error messages of the form "shader
341 * uses %s interpolation qualifier", so in the case where there is no
342 * interpolation qualifier, it returns "no".
343 *
344 * This function should only be used on a shader input or output variable.
345 */
346 const char *interpolation_string() const;
347
348 /**
349 * Determine how this variable should be interpolated based on its
350 * interpolation qualifier (if present), whether it is gl_Color or
351 * gl_SecondaryColor, and whether flatshading is enabled in the current GL
352 * state.
353 *
354 * The return value will always be either INTERP_QUALIFIER_SMOOTH,
355 * INTERP_QUALIFIER_NOPERSPECTIVE, or INTERP_QUALIFIER_FLAT.
356 */
357 glsl_interp_qualifier determine_interpolation_mode(bool flat_shade);
358
359 /**
360 * Determine whether or not a variable is part of a uniform block.
361 */
362 inline bool is_in_uniform_block() const
363 {
364 return this->mode == ir_var_uniform && this->interface_type != NULL;
365 }
366
367 /**
368 * Determine whether or not a variable is the declaration of an interface
369 * block
370 *
371 * For the first declaration below, there will be an \c ir_variable named
372 * "instance" whose type and whose instance_type will be the same
373 * \cglsl_type. For the second declaration, there will be an \c ir_variable
374 * named "f" whose type is float and whose instance_type is B2.
375 *
376 * "instance" is an interface instance variable, but "f" is not.
377 *
378 * uniform B1 {
379 * float f;
380 * } instance;
381 *
382 * uniform B2 {
383 * float f;
384 * };
385 */
386 inline bool is_interface_instance() const
387 {
388 const glsl_type *const t = this->type;
389
390 return (t == this->interface_type)
391 || (t->is_array() && t->fields.array == this->interface_type);
392 }
393
394 /**
395 * Set this->interface_type on a newly created variable.
396 */
397 void init_interface_type(const struct glsl_type *type)
398 {
399 assert(this->interface_type == NULL);
400 this->interface_type = type;
401 if (this->is_interface_instance()) {
402 this->max_ifc_array_access =
403 rzalloc_array(this, unsigned, type->length);
404 }
405 }
406
407 /**
408 * Change this->interface_type on a variable that previously had a
409 * different, but compatible, interface_type. This is used during linking
410 * to set the size of arrays in interface blocks.
411 */
412 void change_interface_type(const struct glsl_type *type)
413 {
414 if (this->max_ifc_array_access != NULL) {
415 /* max_ifc_array_access has already been allocated, so make sure the
416 * new interface has the same number of fields as the old one.
417 */
418 assert(this->interface_type->length == type->length);
419 }
420 this->interface_type = type;
421 }
422
423 /**
424 * Change this->interface_type on a variable that previously had a
425 * different, and incompatible, interface_type. This is used during
426 * compilation to handle redeclaration of the built-in gl_PerVertex
427 * interface block.
428 */
429 void reinit_interface_type(const struct glsl_type *type)
430 {
431 if (this->max_ifc_array_access != NULL) {
432 #ifndef _NDEBUG
433 /* Redeclaring gl_PerVertex is only allowed if none of the built-ins
434 * it defines have been accessed yet; so it's safe to throw away the
435 * old max_ifc_array_access pointer, since all of its values are
436 * zero.
437 */
438 for (unsigned i = 0; i < this->interface_type->length; i++)
439 assert(this->max_ifc_array_access[i] == 0);
440 #endif
441 ralloc_free(this->max_ifc_array_access);
442 this->max_ifc_array_access = NULL;
443 }
444 this->interface_type = NULL;
445 init_interface_type(type);
446 }
447
448 const glsl_type *get_interface_type() const
449 {
450 return this->interface_type;
451 }
452
453 /**
454 * Declared type of the variable
455 */
456 const struct glsl_type *type;
457
458 /**
459 * Declared name of the variable
460 */
461 const char *name;
462
463 /**
464 * Highest element accessed with a constant expression array index
465 *
466 * Not used for non-array variables.
467 */
468 unsigned max_array_access;
469
470 /**
471 * For variables which satisfy the is_interface_instance() predicate, this
472 * points to an array of integers such that if the ith member of the
473 * interface block is an array, max_ifc_array_access[i] is the maximum
474 * array element of that member that has been accessed. If the ith member
475 * of the interface block is not an array, max_ifc_array_access[i] is
476 * unused.
477 *
478 * For variables whose type is not an interface block, this pointer is
479 * NULL.
480 */
481 unsigned *max_ifc_array_access;
482
483 /**
484 * Is the variable read-only?
485 *
486 * This is set for variables declared as \c const, shader inputs,
487 * and uniforms.
488 */
489 unsigned read_only:1;
490 unsigned centroid:1;
491 unsigned invariant:1;
492
493 /**
494 * Has this variable been used for reading or writing?
495 *
496 * Several GLSL semantic checks require knowledge of whether or not a
497 * variable has been used. For example, it is an error to redeclare a
498 * variable as invariant after it has been used.
499 *
500 * This is only maintained in the ast_to_hir.cpp path, not in
501 * Mesa's fixed function or ARB program paths.
502 */
503 unsigned used:1;
504
505 /**
506 * Has this variable been statically assigned?
507 *
508 * This answers whether the variable was assigned in any path of
509 * the shader during ast_to_hir. This doesn't answer whether it is
510 * still written after dead code removal, nor is it maintained in
511 * non-ast_to_hir.cpp (GLSL parsing) paths.
512 */
513 unsigned assigned:1;
514
515 /**
516 * Storage class of the variable.
517 *
518 * \sa ir_variable_mode
519 */
520 unsigned mode:4;
521
522 /**
523 * Interpolation mode for shader inputs / outputs
524 *
525 * \sa ir_variable_interpolation
526 */
527 unsigned interpolation:2;
528
529 /**
530 * \name ARB_fragment_coord_conventions
531 * @{
532 */
533 unsigned origin_upper_left:1;
534 unsigned pixel_center_integer:1;
535 /*@}*/
536
537 /**
538 * Was the location explicitly set in the shader?
539 *
540 * If the location is explicitly set in the shader, it \b cannot be changed
541 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
542 * no effect).
543 */
544 unsigned explicit_location:1;
545 unsigned explicit_index:1;
546
547 /**
548 * Was an initial binding explicitly set in the shader?
549 *
550 * If so, constant_value contains an integer ir_constant representing the
551 * initial binding point.
552 */
553 unsigned explicit_binding:1;
554
555 /**
556 * Does this variable have an initializer?
557 *
558 * This is used by the linker to cross-validiate initializers of global
559 * variables.
560 */
561 unsigned has_initializer:1;
562
563 /**
564 * Is this variable a generic output or input that has not yet been matched
565 * up to a variable in another stage of the pipeline?
566 *
567 * This is used by the linker as scratch storage while assigning locations
568 * to generic inputs and outputs.
569 */
570 unsigned is_unmatched_generic_inout:1;
571
572 /**
573 * If non-zero, then this variable may be packed along with other variables
574 * into a single varying slot, so this offset should be applied when
575 * accessing components. For example, an offset of 1 means that the x
576 * component of this variable is actually stored in component y of the
577 * location specified by \c location.
578 */
579 unsigned location_frac:2;
580
581 /**
582 * \brief Layout qualifier for gl_FragDepth.
583 *
584 * This is not equal to \c ir_depth_layout_none if and only if this
585 * variable is \c gl_FragDepth and a layout qualifier is specified.
586 */
587 ir_depth_layout depth_layout;
588
589 /**
590 * Storage location of the base of this variable
591 *
592 * The precise meaning of this field depends on the nature of the variable.
593 *
594 * - Vertex shader input: one of the values from \c gl_vert_attrib.
595 * - Vertex shader output: one of the values from \c gl_varying_slot.
596 * - Geometry shader input: one of the values from \c gl_varying_slot.
597 * - Geometry shader output: one of the values from \c gl_varying_slot.
598 * - Fragment shader input: one of the values from \c gl_varying_slot.
599 * - Fragment shader output: one of the values from \c gl_frag_result.
600 * - Uniforms: Per-stage uniform slot number for default uniform block.
601 * - Uniforms: Index within the uniform block definition for UBO members.
602 * - Other: This field is not currently used.
603 *
604 * If the variable is a uniform, shader input, or shader output, and the
605 * slot has not been assigned, the value will be -1.
606 */
607 int location;
608
609 /**
610 * output index for dual source blending.
611 */
612 int index;
613
614 /**
615 * Initial binding point for a sampler or UBO.
616 *
617 * For array types, this represents the binding point for the first element.
618 */
619 int binding;
620
621 /**
622 * Built-in state that backs this uniform
623 *
624 * Once set at variable creation, \c state_slots must remain invariant.
625 * This is because, ideally, this array would be shared by all clones of
626 * this variable in the IR tree. In other words, we'd really like for it
627 * to be a fly-weight.
628 *
629 * If the variable is not a uniform, \c num_state_slots will be zero and
630 * \c state_slots will be \c NULL.
631 */
632 /*@{*/
633 unsigned num_state_slots; /**< Number of state slots used */
634 ir_state_slot *state_slots; /**< State descriptors. */
635 /*@}*/
636
637 /**
638 * Emit a warning if this variable is accessed.
639 */
640 const char *warn_extension;
641
642 /**
643 * Value assigned in the initializer of a variable declared "const"
644 */
645 ir_constant *constant_value;
646
647 /**
648 * Constant expression assigned in the initializer of the variable
649 *
650 * \warning
651 * This field and \c ::constant_value are distinct. Even if the two fields
652 * refer to constants with the same value, they must point to separate
653 * objects.
654 */
655 ir_constant *constant_initializer;
656
657 private:
658 /**
659 * For variables that are in an interface block or are an instance of an
660 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
661 *
662 * \sa ir_variable::location
663 */
664 const glsl_type *interface_type;
665 };
666
667 /**
668 * A function that returns whether a built-in function is available in the
669 * current shading language (based on version, ES or desktop, and extensions).
670 */
671 typedef bool (*builtin_available_predicate)(const _mesa_glsl_parse_state *);
672
673 /*@{*/
674 /**
675 * The representation of a function instance; may be the full definition or
676 * simply a prototype.
677 */
678 class ir_function_signature : public ir_instruction {
679 /* An ir_function_signature will be part of the list of signatures in
680 * an ir_function.
681 */
682 public:
683 ir_function_signature(const glsl_type *return_type,
684 builtin_available_predicate builtin_avail = NULL);
685
686 virtual ir_function_signature *clone(void *mem_ctx,
687 struct hash_table *ht) const;
688 ir_function_signature *clone_prototype(void *mem_ctx,
689 struct hash_table *ht) const;
690
691 virtual void accept(ir_visitor *v)
692 {
693 v->visit(this);
694 }
695
696 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
697
698 /**
699 * Attempt to evaluate this function as a constant expression,
700 * given a list of the actual parameters and the variable context.
701 * Returns NULL for non-built-ins.
702 */
703 ir_constant *constant_expression_value(exec_list *actual_parameters, struct hash_table *variable_context);
704
705 /**
706 * Get the name of the function for which this is a signature
707 */
708 const char *function_name() const;
709
710 /**
711 * Get a handle to the function for which this is a signature
712 *
713 * There is no setter function, this function returns a \c const pointer,
714 * and \c ir_function_signature::_function is private for a reason. The
715 * only way to make a connection between a function and function signature
716 * is via \c ir_function::add_signature. This helps ensure that certain
717 * invariants (i.e., a function signature is in the list of signatures for
718 * its \c _function) are met.
719 *
720 * \sa ir_function::add_signature
721 */
722 inline const class ir_function *function() const
723 {
724 return this->_function;
725 }
726
727 /**
728 * Check whether the qualifiers match between this signature's parameters
729 * and the supplied parameter list. If not, returns the name of the first
730 * parameter with mismatched qualifiers (for use in error messages).
731 */
732 const char *qualifiers_match(exec_list *params);
733
734 /**
735 * Replace the current parameter list with the given one. This is useful
736 * if the current information came from a prototype, and either has invalid
737 * or missing parameter names.
738 */
739 void replace_parameters(exec_list *new_params);
740
741 /**
742 * Function return type.
743 *
744 * \note This discards the optional precision qualifier.
745 */
746 const struct glsl_type *return_type;
747
748 /**
749 * List of ir_variable of function parameters.
750 *
751 * This represents the storage. The paramaters passed in a particular
752 * call will be in ir_call::actual_paramaters.
753 */
754 struct exec_list parameters;
755
756 /** Whether or not this function has a body (which may be empty). */
757 unsigned is_defined:1;
758
759 /** Whether or not this function signature is a built-in. */
760 bool is_builtin() const;
761
762 /** Whether or not a built-in is available for this shader. */
763 bool is_builtin_available(const _mesa_glsl_parse_state *state) const;
764
765 /** Body of instructions in the function. */
766 struct exec_list body;
767
768 private:
769 /**
770 * A function pointer to a predicate that answers whether a built-in
771 * function is available in the current shader. NULL if not a built-in.
772 */
773 builtin_available_predicate builtin_avail;
774
775 /** Function of which this signature is one overload. */
776 class ir_function *_function;
777
778 /** Function signature of which this one is a prototype clone */
779 const ir_function_signature *origin;
780
781 friend class ir_function;
782
783 /**
784 * Helper function to run a list of instructions for constant
785 * expression evaluation.
786 *
787 * The hash table represents the values of the visible variables.
788 * There are no scoping issues because the table is indexed on
789 * ir_variable pointers, not variable names.
790 *
791 * Returns false if the expression is not constant, true otherwise,
792 * and the value in *result if result is non-NULL.
793 */
794 bool constant_expression_evaluate_expression_list(const struct exec_list &body,
795 struct hash_table *variable_context,
796 ir_constant **result);
797 };
798
799
800 /**
801 * Header for tracking multiple overloaded functions with the same name.
802 * Contains a list of ir_function_signatures representing each of the
803 * actual functions.
804 */
805 class ir_function : public ir_instruction {
806 public:
807 ir_function(const char *name);
808
809 virtual ir_function *clone(void *mem_ctx, struct hash_table *ht) const;
810
811 virtual ir_function *as_function()
812 {
813 return this;
814 }
815
816 virtual void accept(ir_visitor *v)
817 {
818 v->visit(this);
819 }
820
821 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
822
823 void add_signature(ir_function_signature *sig)
824 {
825 sig->_function = this;
826 this->signatures.push_tail(sig);
827 }
828
829 /**
830 * Get an iterator for the set of function signatures
831 */
832 exec_list_iterator iterator()
833 {
834 return signatures.iterator();
835 }
836
837 /**
838 * Find a signature that matches a set of actual parameters, taking implicit
839 * conversions into account. Also flags whether the match was exact.
840 */
841 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
842 const exec_list *actual_param,
843 bool *match_is_exact);
844
845 /**
846 * Find a signature that matches a set of actual parameters, taking implicit
847 * conversions into account.
848 */
849 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
850 const exec_list *actual_param);
851
852 /**
853 * Find a signature that exactly matches a set of actual parameters without
854 * any implicit type conversions.
855 */
856 ir_function_signature *exact_matching_signature(_mesa_glsl_parse_state *state,
857 const exec_list *actual_ps);
858
859 /**
860 * Name of the function.
861 */
862 const char *name;
863
864 /** Whether or not this function has a signature that isn't a built-in. */
865 bool has_user_signature();
866
867 /**
868 * List of ir_function_signature for each overloaded function with this name.
869 */
870 struct exec_list signatures;
871 };
872
873 inline const char *ir_function_signature::function_name() const
874 {
875 return this->_function->name;
876 }
877 /*@}*/
878
879
880 /**
881 * IR instruction representing high-level if-statements
882 */
883 class ir_if : public ir_instruction {
884 public:
885 ir_if(ir_rvalue *condition)
886 : condition(condition)
887 {
888 ir_type = ir_type_if;
889 }
890
891 virtual ir_if *clone(void *mem_ctx, struct hash_table *ht) const;
892
893 virtual ir_if *as_if()
894 {
895 return this;
896 }
897
898 virtual void accept(ir_visitor *v)
899 {
900 v->visit(this);
901 }
902
903 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
904
905 ir_rvalue *condition;
906 /** List of ir_instruction for the body of the then branch */
907 exec_list then_instructions;
908 /** List of ir_instruction for the body of the else branch */
909 exec_list else_instructions;
910 };
911
912
913 /**
914 * IR instruction representing a high-level loop structure.
915 */
916 class ir_loop : public ir_instruction {
917 public:
918 ir_loop();
919
920 virtual ir_loop *clone(void *mem_ctx, struct hash_table *ht) const;
921
922 virtual void accept(ir_visitor *v)
923 {
924 v->visit(this);
925 }
926
927 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
928
929 virtual ir_loop *as_loop()
930 {
931 return this;
932 }
933
934 /**
935 * Get an iterator for the instructions of the loop body
936 */
937 exec_list_iterator iterator()
938 {
939 return body_instructions.iterator();
940 }
941
942 /** List of ir_instruction that make up the body of the loop. */
943 exec_list body_instructions;
944
945 /**
946 * \name Loop counter and controls
947 *
948 * Represents a loop like a FORTRAN \c do-loop.
949 *
950 * \note
951 * If \c from and \c to are the same value, the loop will execute once.
952 */
953 /*@{*/
954 ir_rvalue *from; /** Value of the loop counter on the first
955 * iteration of the loop.
956 */
957 ir_rvalue *to; /** Value of the loop counter on the last
958 * iteration of the loop.
959 */
960 ir_rvalue *increment;
961 ir_variable *counter;
962
963 /**
964 * Comparison operation in the loop terminator.
965 *
966 * If any of the loop control fields are non-\c NULL, this field must be
967 * one of \c ir_binop_less, \c ir_binop_greater, \c ir_binop_lequal,
968 * \c ir_binop_gequal, \c ir_binop_equal, or \c ir_binop_nequal.
969 */
970 int cmp;
971 /*@}*/
972 };
973
974
975 class ir_assignment : public ir_instruction {
976 public:
977 ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs, ir_rvalue *condition = NULL);
978
979 /**
980 * Construct an assignment with an explicit write mask
981 *
982 * \note
983 * Since a write mask is supplied, the LHS must already be a bare
984 * \c ir_dereference. The cannot be any swizzles in the LHS.
985 */
986 ir_assignment(ir_dereference *lhs, ir_rvalue *rhs, ir_rvalue *condition,
987 unsigned write_mask);
988
989 virtual ir_assignment *clone(void *mem_ctx, struct hash_table *ht) const;
990
991 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
992
993 virtual void accept(ir_visitor *v)
994 {
995 v->visit(this);
996 }
997
998 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
999
1000 virtual ir_assignment * as_assignment()
1001 {
1002 return this;
1003 }
1004
1005 /**
1006 * Get a whole variable written by an assignment
1007 *
1008 * If the LHS of the assignment writes a whole variable, the variable is
1009 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1010 * assignment are:
1011 *
1012 * - Assigning to a scalar
1013 * - Assigning to all components of a vector
1014 * - Whole array (or matrix) assignment
1015 * - Whole structure assignment
1016 */
1017 ir_variable *whole_variable_written();
1018
1019 /**
1020 * Set the LHS of an assignment
1021 */
1022 void set_lhs(ir_rvalue *lhs);
1023
1024 /**
1025 * Left-hand side of the assignment.
1026 *
1027 * This should be treated as read only. If you need to set the LHS of an
1028 * assignment, use \c ir_assignment::set_lhs.
1029 */
1030 ir_dereference *lhs;
1031
1032 /**
1033 * Value being assigned
1034 */
1035 ir_rvalue *rhs;
1036
1037 /**
1038 * Optional condition for the assignment.
1039 */
1040 ir_rvalue *condition;
1041
1042
1043 /**
1044 * Component mask written
1045 *
1046 * For non-vector types in the LHS, this field will be zero. For vector
1047 * types, a bit will be set for each component that is written. Note that
1048 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1049 *
1050 * A partially-set write mask means that each enabled channel gets
1051 * the value from a consecutive channel of the rhs. For example,
1052 * to write just .xyw of gl_FrontColor with color:
1053 *
1054 * (assign (constant bool (1)) (xyw)
1055 * (var_ref gl_FragColor)
1056 * (swiz xyw (var_ref color)))
1057 */
1058 unsigned write_mask:4;
1059 };
1060
1061 /* Update ir_expression::get_num_operands() and operator_strs when
1062 * updating this list.
1063 */
1064 enum ir_expression_operation {
1065 ir_unop_bit_not,
1066 ir_unop_logic_not,
1067 ir_unop_neg,
1068 ir_unop_abs,
1069 ir_unop_sign,
1070 ir_unop_rcp,
1071 ir_unop_rsq,
1072 ir_unop_sqrt,
1073 ir_unop_exp, /**< Log base e on gentype */
1074 ir_unop_log, /**< Natural log on gentype */
1075 ir_unop_exp2,
1076 ir_unop_log2,
1077 ir_unop_f2i, /**< Float-to-integer conversion. */
1078 ir_unop_f2u, /**< Float-to-unsigned conversion. */
1079 ir_unop_i2f, /**< Integer-to-float conversion. */
1080 ir_unop_f2b, /**< Float-to-boolean conversion */
1081 ir_unop_b2f, /**< Boolean-to-float conversion */
1082 ir_unop_i2b, /**< int-to-boolean conversion */
1083 ir_unop_b2i, /**< Boolean-to-int conversion */
1084 ir_unop_u2f, /**< Unsigned-to-float conversion. */
1085 ir_unop_i2u, /**< Integer-to-unsigned conversion. */
1086 ir_unop_u2i, /**< Unsigned-to-integer conversion. */
1087 ir_unop_bitcast_i2f, /**< Bit-identical int-to-float "conversion" */
1088 ir_unop_bitcast_f2i, /**< Bit-identical float-to-int "conversion" */
1089 ir_unop_bitcast_u2f, /**< Bit-identical uint-to-float "conversion" */
1090 ir_unop_bitcast_f2u, /**< Bit-identical float-to-uint "conversion" */
1091 ir_unop_any,
1092
1093 /**
1094 * \name Unary floating-point rounding operations.
1095 */
1096 /*@{*/
1097 ir_unop_trunc,
1098 ir_unop_ceil,
1099 ir_unop_floor,
1100 ir_unop_fract,
1101 ir_unop_round_even,
1102 /*@}*/
1103
1104 /**
1105 * \name Trigonometric operations.
1106 */
1107 /*@{*/
1108 ir_unop_sin,
1109 ir_unop_cos,
1110 ir_unop_sin_reduced, /**< Reduced range sin. [-pi, pi] */
1111 ir_unop_cos_reduced, /**< Reduced range cos. [-pi, pi] */
1112 /*@}*/
1113
1114 /**
1115 * \name Partial derivatives.
1116 */
1117 /*@{*/
1118 ir_unop_dFdx,
1119 ir_unop_dFdy,
1120 /*@}*/
1121
1122 /**
1123 * \name Floating point pack and unpack operations.
1124 */
1125 /*@{*/
1126 ir_unop_pack_snorm_2x16,
1127 ir_unop_pack_snorm_4x8,
1128 ir_unop_pack_unorm_2x16,
1129 ir_unop_pack_unorm_4x8,
1130 ir_unop_pack_half_2x16,
1131 ir_unop_unpack_snorm_2x16,
1132 ir_unop_unpack_snorm_4x8,
1133 ir_unop_unpack_unorm_2x16,
1134 ir_unop_unpack_unorm_4x8,
1135 ir_unop_unpack_half_2x16,
1136 /*@}*/
1137
1138 /**
1139 * \name Lowered floating point unpacking operations.
1140 *
1141 * \see lower_packing_builtins_visitor::split_unpack_half_2x16
1142 */
1143 /*@{*/
1144 ir_unop_unpack_half_2x16_split_x,
1145 ir_unop_unpack_half_2x16_split_y,
1146 /*@}*/
1147
1148 /**
1149 * \name Bit operations, part of ARB_gpu_shader5.
1150 */
1151 /*@{*/
1152 ir_unop_bitfield_reverse,
1153 ir_unop_bit_count,
1154 ir_unop_find_msb,
1155 ir_unop_find_lsb,
1156 /*@}*/
1157
1158 ir_unop_noise,
1159
1160 /**
1161 * A sentinel marking the last of the unary operations.
1162 */
1163 ir_last_unop = ir_unop_noise,
1164
1165 ir_binop_add,
1166 ir_binop_sub,
1167 ir_binop_mul, /**< Floating-point or low 32-bit integer multiply. */
1168 ir_binop_imul_high, /**< Calculates the high 32-bits of a 64-bit multiply. */
1169 ir_binop_div,
1170
1171 /**
1172 * Returns the carry resulting from the addition of the two arguments.
1173 */
1174 /*@{*/
1175 ir_binop_carry,
1176 /*@}*/
1177
1178 /**
1179 * Returns the borrow resulting from the subtraction of the second argument
1180 * from the first argument.
1181 */
1182 /*@{*/
1183 ir_binop_borrow,
1184 /*@}*/
1185
1186 /**
1187 * Takes one of two combinations of arguments:
1188 *
1189 * - mod(vecN, vecN)
1190 * - mod(vecN, float)
1191 *
1192 * Does not take integer types.
1193 */
1194 ir_binop_mod,
1195
1196 /**
1197 * \name Binary comparison operators which return a boolean vector.
1198 * The type of both operands must be equal.
1199 */
1200 /*@{*/
1201 ir_binop_less,
1202 ir_binop_greater,
1203 ir_binop_lequal,
1204 ir_binop_gequal,
1205 ir_binop_equal,
1206 ir_binop_nequal,
1207 /**
1208 * Returns single boolean for whether all components of operands[0]
1209 * equal the components of operands[1].
1210 */
1211 ir_binop_all_equal,
1212 /**
1213 * Returns single boolean for whether any component of operands[0]
1214 * is not equal to the corresponding component of operands[1].
1215 */
1216 ir_binop_any_nequal,
1217 /*@}*/
1218
1219 /**
1220 * \name Bit-wise binary operations.
1221 */
1222 /*@{*/
1223 ir_binop_lshift,
1224 ir_binop_rshift,
1225 ir_binop_bit_and,
1226 ir_binop_bit_xor,
1227 ir_binop_bit_or,
1228 /*@}*/
1229
1230 ir_binop_logic_and,
1231 ir_binop_logic_xor,
1232 ir_binop_logic_or,
1233
1234 ir_binop_dot,
1235 ir_binop_min,
1236 ir_binop_max,
1237
1238 ir_binop_pow,
1239
1240 /**
1241 * \name Lowered floating point packing operations.
1242 *
1243 * \see lower_packing_builtins_visitor::split_pack_half_2x16
1244 */
1245 /*@{*/
1246 ir_binop_pack_half_2x16_split,
1247 /*@}*/
1248
1249 /**
1250 * \name First half of a lowered bitfieldInsert() operation.
1251 *
1252 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1253 */
1254 /*@{*/
1255 ir_binop_bfm,
1256 /*@}*/
1257
1258 /**
1259 * Load a value the size of a given GLSL type from a uniform block.
1260 *
1261 * operand0 is the ir_constant uniform block index in the linked shader.
1262 * operand1 is a byte offset within the uniform block.
1263 */
1264 ir_binop_ubo_load,
1265
1266 /**
1267 * \name Multiplies a number by two to a power, part of ARB_gpu_shader5.
1268 */
1269 /*@{*/
1270 ir_binop_ldexp,
1271 /*@}*/
1272
1273 /**
1274 * Extract a scalar from a vector
1275 *
1276 * operand0 is the vector
1277 * operand1 is the index of the field to read from operand0
1278 */
1279 ir_binop_vector_extract,
1280
1281 /**
1282 * A sentinel marking the last of the binary operations.
1283 */
1284 ir_last_binop = ir_binop_vector_extract,
1285
1286 /**
1287 * \name Fused floating-point multiply-add, part of ARB_gpu_shader5.
1288 */
1289 /*@{*/
1290 ir_triop_fma,
1291 /*@}*/
1292
1293 ir_triop_lrp,
1294
1295 /**
1296 * \name Conditional Select
1297 *
1298 * A vector conditional select instruction (like ?:, but operating per-
1299 * component on vectors).
1300 *
1301 * \see lower_instructions_visitor::ldexp_to_arith
1302 */
1303 /*@{*/
1304 ir_triop_csel,
1305 /*@}*/
1306
1307 /**
1308 * \name Second half of a lowered bitfieldInsert() operation.
1309 *
1310 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1311 */
1312 /*@{*/
1313 ir_triop_bfi,
1314 /*@}*/
1315
1316 ir_triop_bitfield_extract,
1317
1318 /**
1319 * Generate a value with one field of a vector changed
1320 *
1321 * operand0 is the vector
1322 * operand1 is the value to write into the vector result
1323 * operand2 is the index in operand0 to be modified
1324 */
1325 ir_triop_vector_insert,
1326
1327 /**
1328 * A sentinel marking the last of the ternary operations.
1329 */
1330 ir_last_triop = ir_triop_vector_insert,
1331
1332 ir_quadop_bitfield_insert,
1333
1334 ir_quadop_vector,
1335
1336 /**
1337 * A sentinel marking the last of the ternary operations.
1338 */
1339 ir_last_quadop = ir_quadop_vector,
1340
1341 /**
1342 * A sentinel marking the last of all operations.
1343 */
1344 ir_last_opcode = ir_quadop_vector
1345 };
1346
1347 class ir_expression : public ir_rvalue {
1348 public:
1349 ir_expression(int op, const struct glsl_type *type,
1350 ir_rvalue *op0, ir_rvalue *op1 = NULL,
1351 ir_rvalue *op2 = NULL, ir_rvalue *op3 = NULL);
1352
1353 /**
1354 * Constructor for unary operation expressions
1355 */
1356 ir_expression(int op, ir_rvalue *);
1357
1358 /**
1359 * Constructor for binary operation expressions
1360 */
1361 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1);
1362
1363 /**
1364 * Constructor for ternary operation expressions
1365 */
1366 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1, ir_rvalue *op2);
1367
1368 virtual ir_expression *as_expression()
1369 {
1370 return this;
1371 }
1372
1373 virtual ir_expression *clone(void *mem_ctx, struct hash_table *ht) const;
1374
1375 /**
1376 * Attempt to constant-fold the expression
1377 *
1378 * The "variable_context" hash table links ir_variable * to ir_constant *
1379 * that represent the variables' values. \c NULL represents an empty
1380 * context.
1381 *
1382 * If the expression cannot be constant folded, this method will return
1383 * \c NULL.
1384 */
1385 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1386
1387 /**
1388 * Determine the number of operands used by an expression
1389 */
1390 static unsigned int get_num_operands(ir_expression_operation);
1391
1392 /**
1393 * Determine the number of operands used by an expression
1394 */
1395 unsigned int get_num_operands() const
1396 {
1397 return (this->operation == ir_quadop_vector)
1398 ? this->type->vector_elements : get_num_operands(operation);
1399 }
1400
1401 /**
1402 * Return a string representing this expression's operator.
1403 */
1404 const char *operator_string();
1405
1406 /**
1407 * Return a string representing this expression's operator.
1408 */
1409 static const char *operator_string(ir_expression_operation);
1410
1411
1412 /**
1413 * Do a reverse-lookup to translate the given string into an operator.
1414 */
1415 static ir_expression_operation get_operator(const char *);
1416
1417 virtual void accept(ir_visitor *v)
1418 {
1419 v->visit(this);
1420 }
1421
1422 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1423
1424 ir_expression_operation operation;
1425 ir_rvalue *operands[4];
1426 };
1427
1428
1429 /**
1430 * HIR instruction representing a high-level function call, containing a list
1431 * of parameters and returning a value in the supplied temporary.
1432 */
1433 class ir_call : public ir_instruction {
1434 public:
1435 ir_call(ir_function_signature *callee,
1436 ir_dereference_variable *return_deref,
1437 exec_list *actual_parameters)
1438 : return_deref(return_deref), callee(callee)
1439 {
1440 ir_type = ir_type_call;
1441 assert(callee->return_type != NULL);
1442 actual_parameters->move_nodes_to(& this->actual_parameters);
1443 this->use_builtin = callee->is_builtin();
1444 }
1445
1446 virtual ir_call *clone(void *mem_ctx, struct hash_table *ht) const;
1447
1448 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1449
1450 virtual ir_call *as_call()
1451 {
1452 return this;
1453 }
1454
1455 virtual void accept(ir_visitor *v)
1456 {
1457 v->visit(this);
1458 }
1459
1460 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1461
1462 /**
1463 * Get an iterator for the set of acutal parameters
1464 */
1465 exec_list_iterator iterator()
1466 {
1467 return actual_parameters.iterator();
1468 }
1469
1470 /**
1471 * Get the name of the function being called.
1472 */
1473 const char *callee_name() const
1474 {
1475 return callee->function_name();
1476 }
1477
1478 /**
1479 * Generates an inline version of the function before @ir,
1480 * storing the return value in return_deref.
1481 */
1482 void generate_inline(ir_instruction *ir);
1483
1484 /**
1485 * Storage for the function's return value.
1486 * This must be NULL if the return type is void.
1487 */
1488 ir_dereference_variable *return_deref;
1489
1490 /**
1491 * The specific function signature being called.
1492 */
1493 ir_function_signature *callee;
1494
1495 /* List of ir_rvalue of paramaters passed in this call. */
1496 exec_list actual_parameters;
1497
1498 /** Should this call only bind to a built-in function? */
1499 bool use_builtin;
1500 };
1501
1502
1503 /**
1504 * \name Jump-like IR instructions.
1505 *
1506 * These include \c break, \c continue, \c return, and \c discard.
1507 */
1508 /*@{*/
1509 class ir_jump : public ir_instruction {
1510 protected:
1511 ir_jump()
1512 {
1513 ir_type = ir_type_unset;
1514 }
1515
1516 public:
1517 virtual ir_jump *as_jump()
1518 {
1519 return this;
1520 }
1521 };
1522
1523 class ir_return : public ir_jump {
1524 public:
1525 ir_return()
1526 : value(NULL)
1527 {
1528 this->ir_type = ir_type_return;
1529 }
1530
1531 ir_return(ir_rvalue *value)
1532 : value(value)
1533 {
1534 this->ir_type = ir_type_return;
1535 }
1536
1537 virtual ir_return *clone(void *mem_ctx, struct hash_table *) const;
1538
1539 virtual ir_return *as_return()
1540 {
1541 return this;
1542 }
1543
1544 ir_rvalue *get_value() const
1545 {
1546 return value;
1547 }
1548
1549 virtual void accept(ir_visitor *v)
1550 {
1551 v->visit(this);
1552 }
1553
1554 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1555
1556 ir_rvalue *value;
1557 };
1558
1559
1560 /**
1561 * Jump instructions used inside loops
1562 *
1563 * These include \c break and \c continue. The \c break within a loop is
1564 * different from the \c break within a switch-statement.
1565 *
1566 * \sa ir_switch_jump
1567 */
1568 class ir_loop_jump : public ir_jump {
1569 public:
1570 enum jump_mode {
1571 jump_break,
1572 jump_continue
1573 };
1574
1575 ir_loop_jump(jump_mode mode)
1576 {
1577 this->ir_type = ir_type_loop_jump;
1578 this->mode = mode;
1579 }
1580
1581 virtual ir_loop_jump *clone(void *mem_ctx, struct hash_table *) const;
1582
1583 virtual void accept(ir_visitor *v)
1584 {
1585 v->visit(this);
1586 }
1587
1588 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1589
1590 bool is_break() const
1591 {
1592 return mode == jump_break;
1593 }
1594
1595 bool is_continue() const
1596 {
1597 return mode == jump_continue;
1598 }
1599
1600 /** Mode selector for the jump instruction. */
1601 enum jump_mode mode;
1602 };
1603
1604 /**
1605 * IR instruction representing discard statements.
1606 */
1607 class ir_discard : public ir_jump {
1608 public:
1609 ir_discard()
1610 {
1611 this->ir_type = ir_type_discard;
1612 this->condition = NULL;
1613 }
1614
1615 ir_discard(ir_rvalue *cond)
1616 {
1617 this->ir_type = ir_type_discard;
1618 this->condition = cond;
1619 }
1620
1621 virtual ir_discard *clone(void *mem_ctx, struct hash_table *ht) const;
1622
1623 virtual void accept(ir_visitor *v)
1624 {
1625 v->visit(this);
1626 }
1627
1628 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1629
1630 virtual ir_discard *as_discard()
1631 {
1632 return this;
1633 }
1634
1635 ir_rvalue *condition;
1636 };
1637 /*@}*/
1638
1639
1640 /**
1641 * Texture sampling opcodes used in ir_texture
1642 */
1643 enum ir_texture_opcode {
1644 ir_tex, /**< Regular texture look-up */
1645 ir_txb, /**< Texture look-up with LOD bias */
1646 ir_txl, /**< Texture look-up with explicit LOD */
1647 ir_txd, /**< Texture look-up with partial derivatvies */
1648 ir_txf, /**< Texel fetch with explicit LOD */
1649 ir_txf_ms, /**< Multisample texture fetch */
1650 ir_txs, /**< Texture size */
1651 ir_lod, /**< Texture lod query */
1652 ir_tg4, /**< Texture gather */
1653 ir_query_levels /**< Texture levels query */
1654 };
1655
1656
1657 /**
1658 * IR instruction to sample a texture
1659 *
1660 * The specific form of the IR instruction depends on the \c mode value
1661 * selected from \c ir_texture_opcodes. In the printed IR, these will
1662 * appear as:
1663 *
1664 * Texel offset (0 or an expression)
1665 * | Projection divisor
1666 * | | Shadow comparitor
1667 * | | |
1668 * v v v
1669 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1670 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1671 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1672 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1673 * (txf <type> <sampler> <coordinate> 0 <lod>)
1674 * (txf_ms
1675 * <type> <sampler> <coordinate> <sample_index>)
1676 * (txs <type> <sampler> <lod>)
1677 * (lod <type> <sampler> <coordinate>)
1678 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
1679 * (query_levels <type> <sampler>)
1680 */
1681 class ir_texture : public ir_rvalue {
1682 public:
1683 ir_texture(enum ir_texture_opcode op)
1684 : op(op), sampler(NULL), coordinate(NULL), projector(NULL),
1685 shadow_comparitor(NULL), offset(NULL)
1686 {
1687 this->ir_type = ir_type_texture;
1688 memset(&lod_info, 0, sizeof(lod_info));
1689 }
1690
1691 virtual ir_texture *clone(void *mem_ctx, struct hash_table *) const;
1692
1693 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1694
1695 virtual void accept(ir_visitor *v)
1696 {
1697 v->visit(this);
1698 }
1699
1700 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1701
1702 /**
1703 * Return a string representing the ir_texture_opcode.
1704 */
1705 const char *opcode_string();
1706
1707 /** Set the sampler and type. */
1708 void set_sampler(ir_dereference *sampler, const glsl_type *type);
1709
1710 /**
1711 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1712 */
1713 static ir_texture_opcode get_opcode(const char *);
1714
1715 enum ir_texture_opcode op;
1716
1717 /** Sampler to use for the texture access. */
1718 ir_dereference *sampler;
1719
1720 /** Texture coordinate to sample */
1721 ir_rvalue *coordinate;
1722
1723 /**
1724 * Value used for projective divide.
1725 *
1726 * If there is no projective divide (the common case), this will be
1727 * \c NULL. Optimization passes should check for this to point to a constant
1728 * of 1.0 and replace that with \c NULL.
1729 */
1730 ir_rvalue *projector;
1731
1732 /**
1733 * Coordinate used for comparison on shadow look-ups.
1734 *
1735 * If there is no shadow comparison, this will be \c NULL. For the
1736 * \c ir_txf opcode, this *must* be \c NULL.
1737 */
1738 ir_rvalue *shadow_comparitor;
1739
1740 /** Texel offset. */
1741 ir_rvalue *offset;
1742
1743 union {
1744 ir_rvalue *lod; /**< Floating point LOD */
1745 ir_rvalue *bias; /**< Floating point LOD bias */
1746 ir_rvalue *sample_index; /**< MSAA sample index */
1747 ir_rvalue *component; /**< Gather component selector */
1748 struct {
1749 ir_rvalue *dPdx; /**< Partial derivative of coordinate wrt X */
1750 ir_rvalue *dPdy; /**< Partial derivative of coordinate wrt Y */
1751 } grad;
1752 } lod_info;
1753 };
1754
1755
1756 struct ir_swizzle_mask {
1757 unsigned x:2;
1758 unsigned y:2;
1759 unsigned z:2;
1760 unsigned w:2;
1761
1762 /**
1763 * Number of components in the swizzle.
1764 */
1765 unsigned num_components:3;
1766
1767 /**
1768 * Does the swizzle contain duplicate components?
1769 *
1770 * L-value swizzles cannot contain duplicate components.
1771 */
1772 unsigned has_duplicates:1;
1773 };
1774
1775
1776 class ir_swizzle : public ir_rvalue {
1777 public:
1778 ir_swizzle(ir_rvalue *, unsigned x, unsigned y, unsigned z, unsigned w,
1779 unsigned count);
1780
1781 ir_swizzle(ir_rvalue *val, const unsigned *components, unsigned count);
1782
1783 ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask);
1784
1785 virtual ir_swizzle *clone(void *mem_ctx, struct hash_table *) const;
1786
1787 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1788
1789 virtual ir_swizzle *as_swizzle()
1790 {
1791 return this;
1792 }
1793
1794 /**
1795 * Construct an ir_swizzle from the textual representation. Can fail.
1796 */
1797 static ir_swizzle *create(ir_rvalue *, const char *, unsigned vector_length);
1798
1799 virtual void accept(ir_visitor *v)
1800 {
1801 v->visit(this);
1802 }
1803
1804 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1805
1806 bool is_lvalue() const
1807 {
1808 return val->is_lvalue() && !mask.has_duplicates;
1809 }
1810
1811 /**
1812 * Get the variable that is ultimately referenced by an r-value
1813 */
1814 virtual ir_variable *variable_referenced() const;
1815
1816 ir_rvalue *val;
1817 ir_swizzle_mask mask;
1818
1819 private:
1820 /**
1821 * Initialize the mask component of a swizzle
1822 *
1823 * This is used by the \c ir_swizzle constructors.
1824 */
1825 void init_mask(const unsigned *components, unsigned count);
1826 };
1827
1828
1829 class ir_dereference : public ir_rvalue {
1830 public:
1831 virtual ir_dereference *clone(void *mem_ctx, struct hash_table *) const = 0;
1832
1833 virtual ir_dereference *as_dereference()
1834 {
1835 return this;
1836 }
1837
1838 bool is_lvalue() const;
1839
1840 /**
1841 * Get the variable that is ultimately referenced by an r-value
1842 */
1843 virtual ir_variable *variable_referenced() const = 0;
1844
1845 /**
1846 * Get the constant that is ultimately referenced by an r-value,
1847 * in a constant expression evaluation context.
1848 *
1849 * The offset is used when the reference is to a specific column of
1850 * a matrix.
1851 */
1852 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const = 0;
1853 };
1854
1855
1856 class ir_dereference_variable : public ir_dereference {
1857 public:
1858 ir_dereference_variable(ir_variable *var);
1859
1860 virtual ir_dereference_variable *clone(void *mem_ctx,
1861 struct hash_table *) const;
1862
1863 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1864
1865 virtual ir_dereference_variable *as_dereference_variable()
1866 {
1867 return this;
1868 }
1869
1870 /**
1871 * Get the variable that is ultimately referenced by an r-value
1872 */
1873 virtual ir_variable *variable_referenced() const
1874 {
1875 return this->var;
1876 }
1877
1878 /**
1879 * Get the constant that is ultimately referenced by an r-value,
1880 * in a constant expression evaluation context.
1881 *
1882 * The offset is used when the reference is to a specific column of
1883 * a matrix.
1884 */
1885 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
1886
1887 virtual ir_variable *whole_variable_referenced()
1888 {
1889 /* ir_dereference_variable objects always dereference the entire
1890 * variable. However, if this dereference is dereferenced by anything
1891 * else, the complete deferefernce chain is not a whole-variable
1892 * dereference. This method should only be called on the top most
1893 * ir_rvalue in a dereference chain.
1894 */
1895 return this->var;
1896 }
1897
1898 virtual void accept(ir_visitor *v)
1899 {
1900 v->visit(this);
1901 }
1902
1903 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1904
1905 /**
1906 * Object being dereferenced.
1907 */
1908 ir_variable *var;
1909 };
1910
1911
1912 class ir_dereference_array : public ir_dereference {
1913 public:
1914 ir_dereference_array(ir_rvalue *value, ir_rvalue *array_index);
1915
1916 ir_dereference_array(ir_variable *var, ir_rvalue *array_index);
1917
1918 virtual ir_dereference_array *clone(void *mem_ctx,
1919 struct hash_table *) const;
1920
1921 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1922
1923 virtual ir_dereference_array *as_dereference_array()
1924 {
1925 return this;
1926 }
1927
1928 /**
1929 * Get the variable that is ultimately referenced by an r-value
1930 */
1931 virtual ir_variable *variable_referenced() const
1932 {
1933 return this->array->variable_referenced();
1934 }
1935
1936 /**
1937 * Get the constant that is ultimately referenced by an r-value,
1938 * in a constant expression evaluation context.
1939 *
1940 * The offset is used when the reference is to a specific column of
1941 * a matrix.
1942 */
1943 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
1944
1945 virtual void accept(ir_visitor *v)
1946 {
1947 v->visit(this);
1948 }
1949
1950 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1951
1952 ir_rvalue *array;
1953 ir_rvalue *array_index;
1954
1955 private:
1956 void set_array(ir_rvalue *value);
1957 };
1958
1959
1960 class ir_dereference_record : public ir_dereference {
1961 public:
1962 ir_dereference_record(ir_rvalue *value, const char *field);
1963
1964 ir_dereference_record(ir_variable *var, const char *field);
1965
1966 virtual ir_dereference_record *clone(void *mem_ctx,
1967 struct hash_table *) const;
1968
1969 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1970
1971 virtual ir_dereference_record *as_dereference_record()
1972 {
1973 return this;
1974 }
1975
1976 /**
1977 * Get the variable that is ultimately referenced by an r-value
1978 */
1979 virtual ir_variable *variable_referenced() const
1980 {
1981 return this->record->variable_referenced();
1982 }
1983
1984 /**
1985 * Get the constant that is ultimately referenced by an r-value,
1986 * in a constant expression evaluation context.
1987 *
1988 * The offset is used when the reference is to a specific column of
1989 * a matrix.
1990 */
1991 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
1992
1993 virtual void accept(ir_visitor *v)
1994 {
1995 v->visit(this);
1996 }
1997
1998 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1999
2000 ir_rvalue *record;
2001 const char *field;
2002 };
2003
2004
2005 /**
2006 * Data stored in an ir_constant
2007 */
2008 union ir_constant_data {
2009 unsigned u[16];
2010 int i[16];
2011 float f[16];
2012 bool b[16];
2013 };
2014
2015
2016 class ir_constant : public ir_rvalue {
2017 public:
2018 ir_constant(const struct glsl_type *type, const ir_constant_data *data);
2019 ir_constant(bool b, unsigned vector_elements=1);
2020 ir_constant(unsigned int u, unsigned vector_elements=1);
2021 ir_constant(int i, unsigned vector_elements=1);
2022 ir_constant(float f, unsigned vector_elements=1);
2023
2024 /**
2025 * Construct an ir_constant from a list of ir_constant values
2026 */
2027 ir_constant(const struct glsl_type *type, exec_list *values);
2028
2029 /**
2030 * Construct an ir_constant from a scalar component of another ir_constant
2031 *
2032 * The new \c ir_constant inherits the type of the component from the
2033 * source constant.
2034 *
2035 * \note
2036 * In the case of a matrix constant, the new constant is a scalar, \b not
2037 * a vector.
2038 */
2039 ir_constant(const ir_constant *c, unsigned i);
2040
2041 /**
2042 * Return a new ir_constant of the specified type containing all zeros.
2043 */
2044 static ir_constant *zero(void *mem_ctx, const glsl_type *type);
2045
2046 virtual ir_constant *clone(void *mem_ctx, struct hash_table *) const;
2047
2048 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
2049
2050 virtual ir_constant *as_constant()
2051 {
2052 return this;
2053 }
2054
2055 virtual void accept(ir_visitor *v)
2056 {
2057 v->visit(this);
2058 }
2059
2060 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2061
2062 /**
2063 * Get a particular component of a constant as a specific type
2064 *
2065 * This is useful, for example, to get a value from an integer constant
2066 * as a float or bool. This appears frequently when constructors are
2067 * called with all constant parameters.
2068 */
2069 /*@{*/
2070 bool get_bool_component(unsigned i) const;
2071 float get_float_component(unsigned i) const;
2072 int get_int_component(unsigned i) const;
2073 unsigned get_uint_component(unsigned i) const;
2074 /*@}*/
2075
2076 ir_constant *get_array_element(unsigned i) const;
2077
2078 ir_constant *get_record_field(const char *name);
2079
2080 /**
2081 * Copy the values on another constant at a given offset.
2082 *
2083 * The offset is ignored for array or struct copies, it's only for
2084 * scalars or vectors into vectors or matrices.
2085 *
2086 * With identical types on both sides and zero offset it's clone()
2087 * without creating a new object.
2088 */
2089
2090 void copy_offset(ir_constant *src, int offset);
2091
2092 /**
2093 * Copy the values on another constant at a given offset and
2094 * following an assign-like mask.
2095 *
2096 * The mask is ignored for scalars.
2097 *
2098 * Note that this function only handles what assign can handle,
2099 * i.e. at most a vector as source and a column of a matrix as
2100 * destination.
2101 */
2102
2103 void copy_masked_offset(ir_constant *src, int offset, unsigned int mask);
2104
2105 /**
2106 * Determine whether a constant has the same value as another constant
2107 *
2108 * \sa ir_constant::is_zero, ir_constant::is_one,
2109 * ir_constant::is_negative_one, ir_constant::is_basis
2110 */
2111 bool has_value(const ir_constant *) const;
2112
2113 virtual bool is_zero() const;
2114 virtual bool is_one() const;
2115 virtual bool is_negative_one() const;
2116 virtual bool is_basis() const;
2117
2118 /**
2119 * Value of the constant.
2120 *
2121 * The field used to back the values supplied by the constant is determined
2122 * by the type associated with the \c ir_instruction. Constants may be
2123 * scalars, vectors, or matrices.
2124 */
2125 union ir_constant_data value;
2126
2127 /* Array elements */
2128 ir_constant **array_elements;
2129
2130 /* Structure fields */
2131 exec_list components;
2132
2133 private:
2134 /**
2135 * Parameterless constructor only used by the clone method
2136 */
2137 ir_constant(void);
2138 };
2139
2140 /*@}*/
2141
2142 /**
2143 * IR instruction to emit a vertex in a geometry shader.
2144 */
2145 class ir_emit_vertex : public ir_instruction {
2146 public:
2147 ir_emit_vertex()
2148 {
2149 ir_type = ir_type_emit_vertex;
2150 }
2151
2152 virtual void accept(ir_visitor *v)
2153 {
2154 v->visit(this);
2155 }
2156
2157 virtual ir_emit_vertex *clone(void *mem_ctx, struct hash_table *) const
2158 {
2159 return new(mem_ctx) ir_emit_vertex();
2160 }
2161
2162 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2163 };
2164
2165 /**
2166 * IR instruction to complete the current primitive and start a new one in a
2167 * geometry shader.
2168 */
2169 class ir_end_primitive : public ir_instruction {
2170 public:
2171 ir_end_primitive()
2172 {
2173 ir_type = ir_type_end_primitive;
2174 }
2175
2176 virtual void accept(ir_visitor *v)
2177 {
2178 v->visit(this);
2179 }
2180
2181 virtual ir_end_primitive *clone(void *mem_ctx, struct hash_table *) const
2182 {
2183 return new(mem_ctx) ir_end_primitive();
2184 }
2185
2186 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2187 };
2188
2189 /**
2190 * Apply a visitor to each IR node in a list
2191 */
2192 void
2193 visit_exec_list(exec_list *list, ir_visitor *visitor);
2194
2195 /**
2196 * Validate invariants on each IR node in a list
2197 */
2198 void validate_ir_tree(exec_list *instructions);
2199
2200 struct _mesa_glsl_parse_state;
2201 struct gl_shader_program;
2202
2203 /**
2204 * Detect whether an unlinked shader contains static recursion
2205 *
2206 * If the list of instructions is determined to contain static recursion,
2207 * \c _mesa_glsl_error will be called to emit error messages for each function
2208 * that is in the recursion cycle.
2209 */
2210 void
2211 detect_recursion_unlinked(struct _mesa_glsl_parse_state *state,
2212 exec_list *instructions);
2213
2214 /**
2215 * Detect whether a linked shader contains static recursion
2216 *
2217 * If the list of instructions is determined to contain static recursion,
2218 * \c link_error_printf will be called to emit error messages for each function
2219 * that is in the recursion cycle. In addition,
2220 * \c gl_shader_program::LinkStatus will be set to false.
2221 */
2222 void
2223 detect_recursion_linked(struct gl_shader_program *prog,
2224 exec_list *instructions);
2225
2226 /**
2227 * Make a clone of each IR instruction in a list
2228 *
2229 * \param in List of IR instructions that are to be cloned
2230 * \param out List to hold the cloned instructions
2231 */
2232 void
2233 clone_ir_list(void *mem_ctx, exec_list *out, const exec_list *in);
2234
2235 extern void
2236 _mesa_glsl_initialize_variables(exec_list *instructions,
2237 struct _mesa_glsl_parse_state *state);
2238
2239 extern void
2240 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state *state);
2241
2242 extern void
2243 _mesa_glsl_initialize_builtin_functions();
2244
2245 extern ir_function_signature *
2246 _mesa_glsl_find_builtin_function(_mesa_glsl_parse_state *state,
2247 const char *name, exec_list *actual_parameters);
2248
2249 extern void
2250 _mesa_glsl_release_functions(void);
2251
2252 extern void
2253 _mesa_glsl_release_builtin_functions(void);
2254
2255 extern void
2256 reparent_ir(exec_list *list, void *mem_ctx);
2257
2258 struct glsl_symbol_table;
2259
2260 extern void
2261 import_prototypes(const exec_list *source, exec_list *dest,
2262 struct glsl_symbol_table *symbols, void *mem_ctx);
2263
2264 extern bool
2265 ir_has_call(ir_instruction *ir);
2266
2267 extern void
2268 do_set_program_inouts(exec_list *instructions, struct gl_program *prog,
2269 GLenum shader_type);
2270
2271 extern char *
2272 prototype_string(const glsl_type *return_type, const char *name,
2273 exec_list *parameters);
2274
2275 extern "C" {
2276 #endif /* __cplusplus */
2277
2278 extern void _mesa_print_ir(struct exec_list *instructions,
2279 struct _mesa_glsl_parse_state *state);
2280
2281 #ifdef __cplusplus
2282 } /* extern "C" */
2283 #endif
2284
2285 unsigned
2286 vertices_per_prim(GLenum prim);
2287
2288 #endif /* IR_H */