c4c00552ac31c2f138e0594ae497e3b3d90d3ce2
[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
316 /**
317 * Get the string value for an interpolation qualifier
318 *
319 * \return The string that would be used in a shader to specify \c
320 * mode will be returned.
321 *
322 * This function is used to generate error messages of the form "shader
323 * uses %s interpolation qualifier", so in the case where there is no
324 * interpolation qualifier, it returns "no".
325 *
326 * This function should only be used on a shader input or output variable.
327 */
328 const char *interpolation_string(unsigned interpolation);
329
330
331 class ir_variable : public ir_instruction {
332 public:
333 ir_variable(const struct glsl_type *, const char *, ir_variable_mode);
334
335 virtual ir_variable *clone(void *mem_ctx, struct hash_table *ht) const;
336
337 virtual ir_variable *as_variable()
338 {
339 return this;
340 }
341
342 virtual void accept(ir_visitor *v)
343 {
344 v->visit(this);
345 }
346
347 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
348
349
350 /**
351 * Determine how this variable should be interpolated based on its
352 * interpolation qualifier (if present), whether it is gl_Color or
353 * gl_SecondaryColor, and whether flatshading is enabled in the current GL
354 * state.
355 *
356 * The return value will always be either INTERP_QUALIFIER_SMOOTH,
357 * INTERP_QUALIFIER_NOPERSPECTIVE, or INTERP_QUALIFIER_FLAT.
358 */
359 glsl_interp_qualifier determine_interpolation_mode(bool flat_shade);
360
361 /**
362 * Determine whether or not a variable is part of a uniform block.
363 */
364 inline bool is_in_uniform_block() const
365 {
366 return this->mode == ir_var_uniform && this->interface_type != NULL;
367 }
368
369 /**
370 * Determine whether or not a variable is the declaration of an interface
371 * block
372 *
373 * For the first declaration below, there will be an \c ir_variable named
374 * "instance" whose type and whose instance_type will be the same
375 * \cglsl_type. For the second declaration, there will be an \c ir_variable
376 * named "f" whose type is float and whose instance_type is B2.
377 *
378 * "instance" is an interface instance variable, but "f" is not.
379 *
380 * uniform B1 {
381 * float f;
382 * } instance;
383 *
384 * uniform B2 {
385 * float f;
386 * };
387 */
388 inline bool is_interface_instance() const
389 {
390 const glsl_type *const t = this->type;
391
392 return (t == this->interface_type)
393 || (t->is_array() && t->fields.array == this->interface_type);
394 }
395
396 /**
397 * Set this->interface_type on a newly created variable.
398 */
399 void init_interface_type(const struct glsl_type *type)
400 {
401 assert(this->interface_type == NULL);
402 this->interface_type = type;
403 if (this->is_interface_instance()) {
404 this->max_ifc_array_access =
405 rzalloc_array(this, unsigned, type->length);
406 }
407 }
408
409 /**
410 * Change this->interface_type on a variable that previously had a
411 * different, but compatible, interface_type. This is used during linking
412 * to set the size of arrays in interface blocks.
413 */
414 void change_interface_type(const struct glsl_type *type)
415 {
416 if (this->max_ifc_array_access != NULL) {
417 /* max_ifc_array_access has already been allocated, so make sure the
418 * new interface has the same number of fields as the old one.
419 */
420 assert(this->interface_type->length == type->length);
421 }
422 this->interface_type = type;
423 }
424
425 /**
426 * Change this->interface_type on a variable that previously had a
427 * different, and incompatible, interface_type. This is used during
428 * compilation to handle redeclaration of the built-in gl_PerVertex
429 * interface block.
430 */
431 void reinit_interface_type(const struct glsl_type *type)
432 {
433 if (this->max_ifc_array_access != NULL) {
434 #ifndef _NDEBUG
435 /* Redeclaring gl_PerVertex is only allowed if none of the built-ins
436 * it defines have been accessed yet; so it's safe to throw away the
437 * old max_ifc_array_access pointer, since all of its values are
438 * zero.
439 */
440 for (unsigned i = 0; i < this->interface_type->length; i++)
441 assert(this->max_ifc_array_access[i] == 0);
442 #endif
443 ralloc_free(this->max_ifc_array_access);
444 this->max_ifc_array_access = NULL;
445 }
446 this->interface_type = NULL;
447 init_interface_type(type);
448 }
449
450 const glsl_type *get_interface_type() const
451 {
452 return this->interface_type;
453 }
454
455 /**
456 * Declared type of the variable
457 */
458 const struct glsl_type *type;
459
460 /**
461 * Declared name of the variable
462 */
463 const char *name;
464
465 /**
466 * Highest element accessed with a constant expression array index
467 *
468 * Not used for non-array variables.
469 */
470 unsigned max_array_access;
471
472 /**
473 * For variables which satisfy the is_interface_instance() predicate, this
474 * points to an array of integers such that if the ith member of the
475 * interface block is an array, max_ifc_array_access[i] is the maximum
476 * array element of that member that has been accessed. If the ith member
477 * of the interface block is not an array, max_ifc_array_access[i] is
478 * unused.
479 *
480 * For variables whose type is not an interface block, this pointer is
481 * NULL.
482 */
483 unsigned *max_ifc_array_access;
484
485 /**
486 * Is the variable read-only?
487 *
488 * This is set for variables declared as \c const, shader inputs,
489 * and uniforms.
490 */
491 unsigned read_only:1;
492 unsigned centroid:1;
493 unsigned invariant:1;
494
495 /**
496 * Has this variable been used for reading or writing?
497 *
498 * Several GLSL semantic checks require knowledge of whether or not a
499 * variable has been used. For example, it is an error to redeclare a
500 * variable as invariant after it has been used.
501 *
502 * This is only maintained in the ast_to_hir.cpp path, not in
503 * Mesa's fixed function or ARB program paths.
504 */
505 unsigned used:1;
506
507 /**
508 * Has this variable been statically assigned?
509 *
510 * This answers whether the variable was assigned in any path of
511 * the shader during ast_to_hir. This doesn't answer whether it is
512 * still written after dead code removal, nor is it maintained in
513 * non-ast_to_hir.cpp (GLSL parsing) paths.
514 */
515 unsigned assigned:1;
516
517 /**
518 * Storage class of the variable.
519 *
520 * \sa ir_variable_mode
521 */
522 unsigned mode:4;
523
524 /**
525 * Interpolation mode for shader inputs / outputs
526 *
527 * \sa ir_variable_interpolation
528 */
529 unsigned interpolation:2;
530
531 /**
532 * \name ARB_fragment_coord_conventions
533 * @{
534 */
535 unsigned origin_upper_left:1;
536 unsigned pixel_center_integer:1;
537 /*@}*/
538
539 /**
540 * Was the location explicitly set in the shader?
541 *
542 * If the location is explicitly set in the shader, it \b cannot be changed
543 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
544 * no effect).
545 */
546 unsigned explicit_location:1;
547 unsigned explicit_index:1;
548
549 /**
550 * Was an initial binding explicitly set in the shader?
551 *
552 * If so, constant_value contains an integer ir_constant representing the
553 * initial binding point.
554 */
555 unsigned explicit_binding:1;
556
557 /**
558 * Does this variable have an initializer?
559 *
560 * This is used by the linker to cross-validiate initializers of global
561 * variables.
562 */
563 unsigned has_initializer:1;
564
565 /**
566 * Is this variable a generic output or input that has not yet been matched
567 * up to a variable in another stage of the pipeline?
568 *
569 * This is used by the linker as scratch storage while assigning locations
570 * to generic inputs and outputs.
571 */
572 unsigned is_unmatched_generic_inout:1;
573
574 /**
575 * If non-zero, then this variable may be packed along with other variables
576 * into a single varying slot, so this offset should be applied when
577 * accessing components. For example, an offset of 1 means that the x
578 * component of this variable is actually stored in component y of the
579 * location specified by \c location.
580 */
581 unsigned location_frac:2;
582
583 /**
584 * \brief Layout qualifier for gl_FragDepth.
585 *
586 * This is not equal to \c ir_depth_layout_none if and only if this
587 * variable is \c gl_FragDepth and a layout qualifier is specified.
588 */
589 ir_depth_layout depth_layout;
590
591 /**
592 * Storage location of the base of this variable
593 *
594 * The precise meaning of this field depends on the nature of the variable.
595 *
596 * - Vertex shader input: one of the values from \c gl_vert_attrib.
597 * - Vertex shader output: one of the values from \c gl_varying_slot.
598 * - Geometry shader input: one of the values from \c gl_varying_slot.
599 * - Geometry shader output: one of the values from \c gl_varying_slot.
600 * - Fragment shader input: one of the values from \c gl_varying_slot.
601 * - Fragment shader output: one of the values from \c gl_frag_result.
602 * - Uniforms: Per-stage uniform slot number for default uniform block.
603 * - Uniforms: Index within the uniform block definition for UBO members.
604 * - Other: This field is not currently used.
605 *
606 * If the variable is a uniform, shader input, or shader output, and the
607 * slot has not been assigned, the value will be -1.
608 */
609 int location;
610
611 /**
612 * output index for dual source blending.
613 */
614 int index;
615
616 /**
617 * Initial binding point for a sampler or UBO.
618 *
619 * For array types, this represents the binding point for the first element.
620 */
621 int binding;
622
623 /**
624 * Built-in state that backs this uniform
625 *
626 * Once set at variable creation, \c state_slots must remain invariant.
627 * This is because, ideally, this array would be shared by all clones of
628 * this variable in the IR tree. In other words, we'd really like for it
629 * to be a fly-weight.
630 *
631 * If the variable is not a uniform, \c num_state_slots will be zero and
632 * \c state_slots will be \c NULL.
633 */
634 /*@{*/
635 unsigned num_state_slots; /**< Number of state slots used */
636 ir_state_slot *state_slots; /**< State descriptors. */
637 /*@}*/
638
639 /**
640 * Emit a warning if this variable is accessed.
641 */
642 const char *warn_extension;
643
644 /**
645 * Value assigned in the initializer of a variable declared "const"
646 */
647 ir_constant *constant_value;
648
649 /**
650 * Constant expression assigned in the initializer of the variable
651 *
652 * \warning
653 * This field and \c ::constant_value are distinct. Even if the two fields
654 * refer to constants with the same value, they must point to separate
655 * objects.
656 */
657 ir_constant *constant_initializer;
658
659 private:
660 /**
661 * For variables that are in an interface block or are an instance of an
662 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
663 *
664 * \sa ir_variable::location
665 */
666 const glsl_type *interface_type;
667 };
668
669 /**
670 * A function that returns whether a built-in function is available in the
671 * current shading language (based on version, ES or desktop, and extensions).
672 */
673 typedef bool (*builtin_available_predicate)(const _mesa_glsl_parse_state *);
674
675 /*@{*/
676 /**
677 * The representation of a function instance; may be the full definition or
678 * simply a prototype.
679 */
680 class ir_function_signature : public ir_instruction {
681 /* An ir_function_signature will be part of the list of signatures in
682 * an ir_function.
683 */
684 public:
685 ir_function_signature(const glsl_type *return_type,
686 builtin_available_predicate builtin_avail = NULL);
687
688 virtual ir_function_signature *clone(void *mem_ctx,
689 struct hash_table *ht) const;
690 ir_function_signature *clone_prototype(void *mem_ctx,
691 struct hash_table *ht) const;
692
693 virtual void accept(ir_visitor *v)
694 {
695 v->visit(this);
696 }
697
698 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
699
700 /**
701 * Attempt to evaluate this function as a constant expression,
702 * given a list of the actual parameters and the variable context.
703 * Returns NULL for non-built-ins.
704 */
705 ir_constant *constant_expression_value(exec_list *actual_parameters, struct hash_table *variable_context);
706
707 /**
708 * Get the name of the function for which this is a signature
709 */
710 const char *function_name() const;
711
712 /**
713 * Get a handle to the function for which this is a signature
714 *
715 * There is no setter function, this function returns a \c const pointer,
716 * and \c ir_function_signature::_function is private for a reason. The
717 * only way to make a connection between a function and function signature
718 * is via \c ir_function::add_signature. This helps ensure that certain
719 * invariants (i.e., a function signature is in the list of signatures for
720 * its \c _function) are met.
721 *
722 * \sa ir_function::add_signature
723 */
724 inline const class ir_function *function() const
725 {
726 return this->_function;
727 }
728
729 /**
730 * Check whether the qualifiers match between this signature's parameters
731 * and the supplied parameter list. If not, returns the name of the first
732 * parameter with mismatched qualifiers (for use in error messages).
733 */
734 const char *qualifiers_match(exec_list *params);
735
736 /**
737 * Replace the current parameter list with the given one. This is useful
738 * if the current information came from a prototype, and either has invalid
739 * or missing parameter names.
740 */
741 void replace_parameters(exec_list *new_params);
742
743 /**
744 * Function return type.
745 *
746 * \note This discards the optional precision qualifier.
747 */
748 const struct glsl_type *return_type;
749
750 /**
751 * List of ir_variable of function parameters.
752 *
753 * This represents the storage. The paramaters passed in a particular
754 * call will be in ir_call::actual_paramaters.
755 */
756 struct exec_list parameters;
757
758 /** Whether or not this function has a body (which may be empty). */
759 unsigned is_defined:1;
760
761 /** Whether or not this function signature is a built-in. */
762 bool is_builtin() const;
763
764 /** Whether or not a built-in is available for this shader. */
765 bool is_builtin_available(const _mesa_glsl_parse_state *state) const;
766
767 /** Body of instructions in the function. */
768 struct exec_list body;
769
770 private:
771 /**
772 * A function pointer to a predicate that answers whether a built-in
773 * function is available in the current shader. NULL if not a built-in.
774 */
775 builtin_available_predicate builtin_avail;
776
777 /** Function of which this signature is one overload. */
778 class ir_function *_function;
779
780 /** Function signature of which this one is a prototype clone */
781 const ir_function_signature *origin;
782
783 friend class ir_function;
784
785 /**
786 * Helper function to run a list of instructions for constant
787 * expression evaluation.
788 *
789 * The hash table represents the values of the visible variables.
790 * There are no scoping issues because the table is indexed on
791 * ir_variable pointers, not variable names.
792 *
793 * Returns false if the expression is not constant, true otherwise,
794 * and the value in *result if result is non-NULL.
795 */
796 bool constant_expression_evaluate_expression_list(const struct exec_list &body,
797 struct hash_table *variable_context,
798 ir_constant **result);
799 };
800
801
802 /**
803 * Header for tracking multiple overloaded functions with the same name.
804 * Contains a list of ir_function_signatures representing each of the
805 * actual functions.
806 */
807 class ir_function : public ir_instruction {
808 public:
809 ir_function(const char *name);
810
811 virtual ir_function *clone(void *mem_ctx, struct hash_table *ht) const;
812
813 virtual ir_function *as_function()
814 {
815 return this;
816 }
817
818 virtual void accept(ir_visitor *v)
819 {
820 v->visit(this);
821 }
822
823 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
824
825 void add_signature(ir_function_signature *sig)
826 {
827 sig->_function = this;
828 this->signatures.push_tail(sig);
829 }
830
831 /**
832 * Get an iterator for the set of function signatures
833 */
834 exec_list_iterator iterator()
835 {
836 return signatures.iterator();
837 }
838
839 /**
840 * Find a signature that matches a set of actual parameters, taking implicit
841 * conversions into account. Also flags whether the match was exact.
842 */
843 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
844 const exec_list *actual_param,
845 bool *match_is_exact);
846
847 /**
848 * Find a signature that matches a set of actual parameters, taking implicit
849 * conversions into account.
850 */
851 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
852 const exec_list *actual_param);
853
854 /**
855 * Find a signature that exactly matches a set of actual parameters without
856 * any implicit type conversions.
857 */
858 ir_function_signature *exact_matching_signature(_mesa_glsl_parse_state *state,
859 const exec_list *actual_ps);
860
861 /**
862 * Name of the function.
863 */
864 const char *name;
865
866 /** Whether or not this function has a signature that isn't a built-in. */
867 bool has_user_signature();
868
869 /**
870 * List of ir_function_signature for each overloaded function with this name.
871 */
872 struct exec_list signatures;
873 };
874
875 inline const char *ir_function_signature::function_name() const
876 {
877 return this->_function->name;
878 }
879 /*@}*/
880
881
882 /**
883 * IR instruction representing high-level if-statements
884 */
885 class ir_if : public ir_instruction {
886 public:
887 ir_if(ir_rvalue *condition)
888 : condition(condition)
889 {
890 ir_type = ir_type_if;
891 }
892
893 virtual ir_if *clone(void *mem_ctx, struct hash_table *ht) const;
894
895 virtual ir_if *as_if()
896 {
897 return this;
898 }
899
900 virtual void accept(ir_visitor *v)
901 {
902 v->visit(this);
903 }
904
905 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
906
907 ir_rvalue *condition;
908 /** List of ir_instruction for the body of the then branch */
909 exec_list then_instructions;
910 /** List of ir_instruction for the body of the else branch */
911 exec_list else_instructions;
912 };
913
914
915 /**
916 * IR instruction representing a high-level loop structure.
917 */
918 class ir_loop : public ir_instruction {
919 public:
920 ir_loop();
921
922 virtual ir_loop *clone(void *mem_ctx, struct hash_table *ht) const;
923
924 virtual void accept(ir_visitor *v)
925 {
926 v->visit(this);
927 }
928
929 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
930
931 virtual ir_loop *as_loop()
932 {
933 return this;
934 }
935
936 /**
937 * Get an iterator for the instructions of the loop body
938 */
939 exec_list_iterator iterator()
940 {
941 return body_instructions.iterator();
942 }
943
944 /** List of ir_instruction that make up the body of the loop. */
945 exec_list body_instructions;
946
947 /**
948 * \name Loop counter and controls
949 *
950 * Represents a loop like a FORTRAN \c do-loop.
951 *
952 * \note
953 * If \c from and \c to are the same value, the loop will execute once.
954 */
955 /*@{*/
956 ir_rvalue *from; /** Value of the loop counter on the first
957 * iteration of the loop.
958 */
959 ir_rvalue *to; /** Value of the loop counter on the last
960 * iteration of the loop.
961 */
962 ir_rvalue *increment;
963 ir_variable *counter;
964
965 /**
966 * Comparison operation in the loop terminator.
967 *
968 * If any of the loop control fields are non-\c NULL, this field must be
969 * one of \c ir_binop_less, \c ir_binop_greater, \c ir_binop_lequal,
970 * \c ir_binop_gequal, \c ir_binop_equal, or \c ir_binop_nequal.
971 */
972 int cmp;
973 /*@}*/
974 };
975
976
977 class ir_assignment : public ir_instruction {
978 public:
979 ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs, ir_rvalue *condition = NULL);
980
981 /**
982 * Construct an assignment with an explicit write mask
983 *
984 * \note
985 * Since a write mask is supplied, the LHS must already be a bare
986 * \c ir_dereference. The cannot be any swizzles in the LHS.
987 */
988 ir_assignment(ir_dereference *lhs, ir_rvalue *rhs, ir_rvalue *condition,
989 unsigned write_mask);
990
991 virtual ir_assignment *clone(void *mem_ctx, struct hash_table *ht) const;
992
993 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
994
995 virtual void accept(ir_visitor *v)
996 {
997 v->visit(this);
998 }
999
1000 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1001
1002 virtual ir_assignment * as_assignment()
1003 {
1004 return this;
1005 }
1006
1007 /**
1008 * Get a whole variable written by an assignment
1009 *
1010 * If the LHS of the assignment writes a whole variable, the variable is
1011 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1012 * assignment are:
1013 *
1014 * - Assigning to a scalar
1015 * - Assigning to all components of a vector
1016 * - Whole array (or matrix) assignment
1017 * - Whole structure assignment
1018 */
1019 ir_variable *whole_variable_written();
1020
1021 /**
1022 * Set the LHS of an assignment
1023 */
1024 void set_lhs(ir_rvalue *lhs);
1025
1026 /**
1027 * Left-hand side of the assignment.
1028 *
1029 * This should be treated as read only. If you need to set the LHS of an
1030 * assignment, use \c ir_assignment::set_lhs.
1031 */
1032 ir_dereference *lhs;
1033
1034 /**
1035 * Value being assigned
1036 */
1037 ir_rvalue *rhs;
1038
1039 /**
1040 * Optional condition for the assignment.
1041 */
1042 ir_rvalue *condition;
1043
1044
1045 /**
1046 * Component mask written
1047 *
1048 * For non-vector types in the LHS, this field will be zero. For vector
1049 * types, a bit will be set for each component that is written. Note that
1050 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1051 *
1052 * A partially-set write mask means that each enabled channel gets
1053 * the value from a consecutive channel of the rhs. For example,
1054 * to write just .xyw of gl_FrontColor with color:
1055 *
1056 * (assign (constant bool (1)) (xyw)
1057 * (var_ref gl_FragColor)
1058 * (swiz xyw (var_ref color)))
1059 */
1060 unsigned write_mask:4;
1061 };
1062
1063 /* Update ir_expression::get_num_operands() and operator_strs when
1064 * updating this list.
1065 */
1066 enum ir_expression_operation {
1067 ir_unop_bit_not,
1068 ir_unop_logic_not,
1069 ir_unop_neg,
1070 ir_unop_abs,
1071 ir_unop_sign,
1072 ir_unop_rcp,
1073 ir_unop_rsq,
1074 ir_unop_sqrt,
1075 ir_unop_exp, /**< Log base e on gentype */
1076 ir_unop_log, /**< Natural log on gentype */
1077 ir_unop_exp2,
1078 ir_unop_log2,
1079 ir_unop_f2i, /**< Float-to-integer conversion. */
1080 ir_unop_f2u, /**< Float-to-unsigned conversion. */
1081 ir_unop_i2f, /**< Integer-to-float conversion. */
1082 ir_unop_f2b, /**< Float-to-boolean conversion */
1083 ir_unop_b2f, /**< Boolean-to-float conversion */
1084 ir_unop_i2b, /**< int-to-boolean conversion */
1085 ir_unop_b2i, /**< Boolean-to-int conversion */
1086 ir_unop_u2f, /**< Unsigned-to-float conversion. */
1087 ir_unop_i2u, /**< Integer-to-unsigned conversion. */
1088 ir_unop_u2i, /**< Unsigned-to-integer conversion. */
1089 ir_unop_bitcast_i2f, /**< Bit-identical int-to-float "conversion" */
1090 ir_unop_bitcast_f2i, /**< Bit-identical float-to-int "conversion" */
1091 ir_unop_bitcast_u2f, /**< Bit-identical uint-to-float "conversion" */
1092 ir_unop_bitcast_f2u, /**< Bit-identical float-to-uint "conversion" */
1093 ir_unop_any,
1094
1095 /**
1096 * \name Unary floating-point rounding operations.
1097 */
1098 /*@{*/
1099 ir_unop_trunc,
1100 ir_unop_ceil,
1101 ir_unop_floor,
1102 ir_unop_fract,
1103 ir_unop_round_even,
1104 /*@}*/
1105
1106 /**
1107 * \name Trigonometric operations.
1108 */
1109 /*@{*/
1110 ir_unop_sin,
1111 ir_unop_cos,
1112 ir_unop_sin_reduced, /**< Reduced range sin. [-pi, pi] */
1113 ir_unop_cos_reduced, /**< Reduced range cos. [-pi, pi] */
1114 /*@}*/
1115
1116 /**
1117 * \name Partial derivatives.
1118 */
1119 /*@{*/
1120 ir_unop_dFdx,
1121 ir_unop_dFdy,
1122 /*@}*/
1123
1124 /**
1125 * \name Floating point pack and unpack operations.
1126 */
1127 /*@{*/
1128 ir_unop_pack_snorm_2x16,
1129 ir_unop_pack_snorm_4x8,
1130 ir_unop_pack_unorm_2x16,
1131 ir_unop_pack_unorm_4x8,
1132 ir_unop_pack_half_2x16,
1133 ir_unop_unpack_snorm_2x16,
1134 ir_unop_unpack_snorm_4x8,
1135 ir_unop_unpack_unorm_2x16,
1136 ir_unop_unpack_unorm_4x8,
1137 ir_unop_unpack_half_2x16,
1138 /*@}*/
1139
1140 /**
1141 * \name Lowered floating point unpacking operations.
1142 *
1143 * \see lower_packing_builtins_visitor::split_unpack_half_2x16
1144 */
1145 /*@{*/
1146 ir_unop_unpack_half_2x16_split_x,
1147 ir_unop_unpack_half_2x16_split_y,
1148 /*@}*/
1149
1150 /**
1151 * \name Bit operations, part of ARB_gpu_shader5.
1152 */
1153 /*@{*/
1154 ir_unop_bitfield_reverse,
1155 ir_unop_bit_count,
1156 ir_unop_find_msb,
1157 ir_unop_find_lsb,
1158 /*@}*/
1159
1160 ir_unop_noise,
1161
1162 /**
1163 * A sentinel marking the last of the unary operations.
1164 */
1165 ir_last_unop = ir_unop_noise,
1166
1167 ir_binop_add,
1168 ir_binop_sub,
1169 ir_binop_mul, /**< Floating-point or low 32-bit integer multiply. */
1170 ir_binop_imul_high, /**< Calculates the high 32-bits of a 64-bit multiply. */
1171 ir_binop_div,
1172
1173 /**
1174 * Returns the carry resulting from the addition of the two arguments.
1175 */
1176 /*@{*/
1177 ir_binop_carry,
1178 /*@}*/
1179
1180 /**
1181 * Returns the borrow resulting from the subtraction of the second argument
1182 * from the first argument.
1183 */
1184 /*@{*/
1185 ir_binop_borrow,
1186 /*@}*/
1187
1188 /**
1189 * Takes one of two combinations of arguments:
1190 *
1191 * - mod(vecN, vecN)
1192 * - mod(vecN, float)
1193 *
1194 * Does not take integer types.
1195 */
1196 ir_binop_mod,
1197
1198 /**
1199 * \name Binary comparison operators which return a boolean vector.
1200 * The type of both operands must be equal.
1201 */
1202 /*@{*/
1203 ir_binop_less,
1204 ir_binop_greater,
1205 ir_binop_lequal,
1206 ir_binop_gequal,
1207 ir_binop_equal,
1208 ir_binop_nequal,
1209 /**
1210 * Returns single boolean for whether all components of operands[0]
1211 * equal the components of operands[1].
1212 */
1213 ir_binop_all_equal,
1214 /**
1215 * Returns single boolean for whether any component of operands[0]
1216 * is not equal to the corresponding component of operands[1].
1217 */
1218 ir_binop_any_nequal,
1219 /*@}*/
1220
1221 /**
1222 * \name Bit-wise binary operations.
1223 */
1224 /*@{*/
1225 ir_binop_lshift,
1226 ir_binop_rshift,
1227 ir_binop_bit_and,
1228 ir_binop_bit_xor,
1229 ir_binop_bit_or,
1230 /*@}*/
1231
1232 ir_binop_logic_and,
1233 ir_binop_logic_xor,
1234 ir_binop_logic_or,
1235
1236 ir_binop_dot,
1237 ir_binop_min,
1238 ir_binop_max,
1239
1240 ir_binop_pow,
1241
1242 /**
1243 * \name Lowered floating point packing operations.
1244 *
1245 * \see lower_packing_builtins_visitor::split_pack_half_2x16
1246 */
1247 /*@{*/
1248 ir_binop_pack_half_2x16_split,
1249 /*@}*/
1250
1251 /**
1252 * \name First half of a lowered bitfieldInsert() operation.
1253 *
1254 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1255 */
1256 /*@{*/
1257 ir_binop_bfm,
1258 /*@}*/
1259
1260 /**
1261 * Load a value the size of a given GLSL type from a uniform block.
1262 *
1263 * operand0 is the ir_constant uniform block index in the linked shader.
1264 * operand1 is a byte offset within the uniform block.
1265 */
1266 ir_binop_ubo_load,
1267
1268 /**
1269 * \name Multiplies a number by two to a power, part of ARB_gpu_shader5.
1270 */
1271 /*@{*/
1272 ir_binop_ldexp,
1273 /*@}*/
1274
1275 /**
1276 * Extract a scalar from a vector
1277 *
1278 * operand0 is the vector
1279 * operand1 is the index of the field to read from operand0
1280 */
1281 ir_binop_vector_extract,
1282
1283 /**
1284 * A sentinel marking the last of the binary operations.
1285 */
1286 ir_last_binop = ir_binop_vector_extract,
1287
1288 /**
1289 * \name Fused floating-point multiply-add, part of ARB_gpu_shader5.
1290 */
1291 /*@{*/
1292 ir_triop_fma,
1293 /*@}*/
1294
1295 ir_triop_lrp,
1296
1297 /**
1298 * \name Conditional Select
1299 *
1300 * A vector conditional select instruction (like ?:, but operating per-
1301 * component on vectors).
1302 *
1303 * \see lower_instructions_visitor::ldexp_to_arith
1304 */
1305 /*@{*/
1306 ir_triop_csel,
1307 /*@}*/
1308
1309 /**
1310 * \name Second half of a lowered bitfieldInsert() operation.
1311 *
1312 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1313 */
1314 /*@{*/
1315 ir_triop_bfi,
1316 /*@}*/
1317
1318 ir_triop_bitfield_extract,
1319
1320 /**
1321 * Generate a value with one field of a vector changed
1322 *
1323 * operand0 is the vector
1324 * operand1 is the value to write into the vector result
1325 * operand2 is the index in operand0 to be modified
1326 */
1327 ir_triop_vector_insert,
1328
1329 /**
1330 * A sentinel marking the last of the ternary operations.
1331 */
1332 ir_last_triop = ir_triop_vector_insert,
1333
1334 ir_quadop_bitfield_insert,
1335
1336 ir_quadop_vector,
1337
1338 /**
1339 * A sentinel marking the last of the ternary operations.
1340 */
1341 ir_last_quadop = ir_quadop_vector,
1342
1343 /**
1344 * A sentinel marking the last of all operations.
1345 */
1346 ir_last_opcode = ir_quadop_vector
1347 };
1348
1349 class ir_expression : public ir_rvalue {
1350 public:
1351 ir_expression(int op, const struct glsl_type *type,
1352 ir_rvalue *op0, ir_rvalue *op1 = NULL,
1353 ir_rvalue *op2 = NULL, ir_rvalue *op3 = NULL);
1354
1355 /**
1356 * Constructor for unary operation expressions
1357 */
1358 ir_expression(int op, ir_rvalue *);
1359
1360 /**
1361 * Constructor for binary operation expressions
1362 */
1363 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1);
1364
1365 /**
1366 * Constructor for ternary operation expressions
1367 */
1368 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1, ir_rvalue *op2);
1369
1370 virtual ir_expression *as_expression()
1371 {
1372 return this;
1373 }
1374
1375 virtual ir_expression *clone(void *mem_ctx, struct hash_table *ht) const;
1376
1377 /**
1378 * Attempt to constant-fold the expression
1379 *
1380 * The "variable_context" hash table links ir_variable * to ir_constant *
1381 * that represent the variables' values. \c NULL represents an empty
1382 * context.
1383 *
1384 * If the expression cannot be constant folded, this method will return
1385 * \c NULL.
1386 */
1387 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1388
1389 /**
1390 * Determine the number of operands used by an expression
1391 */
1392 static unsigned int get_num_operands(ir_expression_operation);
1393
1394 /**
1395 * Determine the number of operands used by an expression
1396 */
1397 unsigned int get_num_operands() const
1398 {
1399 return (this->operation == ir_quadop_vector)
1400 ? this->type->vector_elements : get_num_operands(operation);
1401 }
1402
1403 /**
1404 * Return a string representing this expression's operator.
1405 */
1406 const char *operator_string();
1407
1408 /**
1409 * Return a string representing this expression's operator.
1410 */
1411 static const char *operator_string(ir_expression_operation);
1412
1413
1414 /**
1415 * Do a reverse-lookup to translate the given string into an operator.
1416 */
1417 static ir_expression_operation get_operator(const char *);
1418
1419 virtual void accept(ir_visitor *v)
1420 {
1421 v->visit(this);
1422 }
1423
1424 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1425
1426 ir_expression_operation operation;
1427 ir_rvalue *operands[4];
1428 };
1429
1430
1431 /**
1432 * HIR instruction representing a high-level function call, containing a list
1433 * of parameters and returning a value in the supplied temporary.
1434 */
1435 class ir_call : public ir_instruction {
1436 public:
1437 ir_call(ir_function_signature *callee,
1438 ir_dereference_variable *return_deref,
1439 exec_list *actual_parameters)
1440 : return_deref(return_deref), callee(callee)
1441 {
1442 ir_type = ir_type_call;
1443 assert(callee->return_type != NULL);
1444 actual_parameters->move_nodes_to(& this->actual_parameters);
1445 this->use_builtin = callee->is_builtin();
1446 }
1447
1448 virtual ir_call *clone(void *mem_ctx, struct hash_table *ht) const;
1449
1450 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1451
1452 virtual ir_call *as_call()
1453 {
1454 return this;
1455 }
1456
1457 virtual void accept(ir_visitor *v)
1458 {
1459 v->visit(this);
1460 }
1461
1462 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1463
1464 /**
1465 * Get an iterator for the set of acutal parameters
1466 */
1467 exec_list_iterator iterator()
1468 {
1469 return actual_parameters.iterator();
1470 }
1471
1472 /**
1473 * Get the name of the function being called.
1474 */
1475 const char *callee_name() const
1476 {
1477 return callee->function_name();
1478 }
1479
1480 /**
1481 * Generates an inline version of the function before @ir,
1482 * storing the return value in return_deref.
1483 */
1484 void generate_inline(ir_instruction *ir);
1485
1486 /**
1487 * Storage for the function's return value.
1488 * This must be NULL if the return type is void.
1489 */
1490 ir_dereference_variable *return_deref;
1491
1492 /**
1493 * The specific function signature being called.
1494 */
1495 ir_function_signature *callee;
1496
1497 /* List of ir_rvalue of paramaters passed in this call. */
1498 exec_list actual_parameters;
1499
1500 /** Should this call only bind to a built-in function? */
1501 bool use_builtin;
1502 };
1503
1504
1505 /**
1506 * \name Jump-like IR instructions.
1507 *
1508 * These include \c break, \c continue, \c return, and \c discard.
1509 */
1510 /*@{*/
1511 class ir_jump : public ir_instruction {
1512 protected:
1513 ir_jump()
1514 {
1515 ir_type = ir_type_unset;
1516 }
1517
1518 public:
1519 virtual ir_jump *as_jump()
1520 {
1521 return this;
1522 }
1523 };
1524
1525 class ir_return : public ir_jump {
1526 public:
1527 ir_return()
1528 : value(NULL)
1529 {
1530 this->ir_type = ir_type_return;
1531 }
1532
1533 ir_return(ir_rvalue *value)
1534 : value(value)
1535 {
1536 this->ir_type = ir_type_return;
1537 }
1538
1539 virtual ir_return *clone(void *mem_ctx, struct hash_table *) const;
1540
1541 virtual ir_return *as_return()
1542 {
1543 return this;
1544 }
1545
1546 ir_rvalue *get_value() const
1547 {
1548 return value;
1549 }
1550
1551 virtual void accept(ir_visitor *v)
1552 {
1553 v->visit(this);
1554 }
1555
1556 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1557
1558 ir_rvalue *value;
1559 };
1560
1561
1562 /**
1563 * Jump instructions used inside loops
1564 *
1565 * These include \c break and \c continue. The \c break within a loop is
1566 * different from the \c break within a switch-statement.
1567 *
1568 * \sa ir_switch_jump
1569 */
1570 class ir_loop_jump : public ir_jump {
1571 public:
1572 enum jump_mode {
1573 jump_break,
1574 jump_continue
1575 };
1576
1577 ir_loop_jump(jump_mode mode)
1578 {
1579 this->ir_type = ir_type_loop_jump;
1580 this->mode = mode;
1581 }
1582
1583 virtual ir_loop_jump *clone(void *mem_ctx, struct hash_table *) const;
1584
1585 virtual void accept(ir_visitor *v)
1586 {
1587 v->visit(this);
1588 }
1589
1590 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1591
1592 bool is_break() const
1593 {
1594 return mode == jump_break;
1595 }
1596
1597 bool is_continue() const
1598 {
1599 return mode == jump_continue;
1600 }
1601
1602 /** Mode selector for the jump instruction. */
1603 enum jump_mode mode;
1604 };
1605
1606 /**
1607 * IR instruction representing discard statements.
1608 */
1609 class ir_discard : public ir_jump {
1610 public:
1611 ir_discard()
1612 {
1613 this->ir_type = ir_type_discard;
1614 this->condition = NULL;
1615 }
1616
1617 ir_discard(ir_rvalue *cond)
1618 {
1619 this->ir_type = ir_type_discard;
1620 this->condition = cond;
1621 }
1622
1623 virtual ir_discard *clone(void *mem_ctx, struct hash_table *ht) const;
1624
1625 virtual void accept(ir_visitor *v)
1626 {
1627 v->visit(this);
1628 }
1629
1630 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1631
1632 virtual ir_discard *as_discard()
1633 {
1634 return this;
1635 }
1636
1637 ir_rvalue *condition;
1638 };
1639 /*@}*/
1640
1641
1642 /**
1643 * Texture sampling opcodes used in ir_texture
1644 */
1645 enum ir_texture_opcode {
1646 ir_tex, /**< Regular texture look-up */
1647 ir_txb, /**< Texture look-up with LOD bias */
1648 ir_txl, /**< Texture look-up with explicit LOD */
1649 ir_txd, /**< Texture look-up with partial derivatvies */
1650 ir_txf, /**< Texel fetch with explicit LOD */
1651 ir_txf_ms, /**< Multisample texture fetch */
1652 ir_txs, /**< Texture size */
1653 ir_lod, /**< Texture lod query */
1654 ir_tg4, /**< Texture gather */
1655 ir_query_levels /**< Texture levels query */
1656 };
1657
1658
1659 /**
1660 * IR instruction to sample a texture
1661 *
1662 * The specific form of the IR instruction depends on the \c mode value
1663 * selected from \c ir_texture_opcodes. In the printed IR, these will
1664 * appear as:
1665 *
1666 * Texel offset (0 or an expression)
1667 * | Projection divisor
1668 * | | Shadow comparitor
1669 * | | |
1670 * v v v
1671 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1672 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1673 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1674 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1675 * (txf <type> <sampler> <coordinate> 0 <lod>)
1676 * (txf_ms
1677 * <type> <sampler> <coordinate> <sample_index>)
1678 * (txs <type> <sampler> <lod>)
1679 * (lod <type> <sampler> <coordinate>)
1680 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
1681 * (query_levels <type> <sampler>)
1682 */
1683 class ir_texture : public ir_rvalue {
1684 public:
1685 ir_texture(enum ir_texture_opcode op)
1686 : op(op), sampler(NULL), coordinate(NULL), projector(NULL),
1687 shadow_comparitor(NULL), offset(NULL)
1688 {
1689 this->ir_type = ir_type_texture;
1690 memset(&lod_info, 0, sizeof(lod_info));
1691 }
1692
1693 virtual ir_texture *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 void accept(ir_visitor *v)
1698 {
1699 v->visit(this);
1700 }
1701
1702 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1703
1704 /**
1705 * Return a string representing the ir_texture_opcode.
1706 */
1707 const char *opcode_string();
1708
1709 /** Set the sampler and type. */
1710 void set_sampler(ir_dereference *sampler, const glsl_type *type);
1711
1712 /**
1713 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1714 */
1715 static ir_texture_opcode get_opcode(const char *);
1716
1717 enum ir_texture_opcode op;
1718
1719 /** Sampler to use for the texture access. */
1720 ir_dereference *sampler;
1721
1722 /** Texture coordinate to sample */
1723 ir_rvalue *coordinate;
1724
1725 /**
1726 * Value used for projective divide.
1727 *
1728 * If there is no projective divide (the common case), this will be
1729 * \c NULL. Optimization passes should check for this to point to a constant
1730 * of 1.0 and replace that with \c NULL.
1731 */
1732 ir_rvalue *projector;
1733
1734 /**
1735 * Coordinate used for comparison on shadow look-ups.
1736 *
1737 * If there is no shadow comparison, this will be \c NULL. For the
1738 * \c ir_txf opcode, this *must* be \c NULL.
1739 */
1740 ir_rvalue *shadow_comparitor;
1741
1742 /** Texel offset. */
1743 ir_rvalue *offset;
1744
1745 union {
1746 ir_rvalue *lod; /**< Floating point LOD */
1747 ir_rvalue *bias; /**< Floating point LOD bias */
1748 ir_rvalue *sample_index; /**< MSAA sample index */
1749 ir_rvalue *component; /**< Gather component selector */
1750 struct {
1751 ir_rvalue *dPdx; /**< Partial derivative of coordinate wrt X */
1752 ir_rvalue *dPdy; /**< Partial derivative of coordinate wrt Y */
1753 } grad;
1754 } lod_info;
1755 };
1756
1757
1758 struct ir_swizzle_mask {
1759 unsigned x:2;
1760 unsigned y:2;
1761 unsigned z:2;
1762 unsigned w:2;
1763
1764 /**
1765 * Number of components in the swizzle.
1766 */
1767 unsigned num_components:3;
1768
1769 /**
1770 * Does the swizzle contain duplicate components?
1771 *
1772 * L-value swizzles cannot contain duplicate components.
1773 */
1774 unsigned has_duplicates:1;
1775 };
1776
1777
1778 class ir_swizzle : public ir_rvalue {
1779 public:
1780 ir_swizzle(ir_rvalue *, unsigned x, unsigned y, unsigned z, unsigned w,
1781 unsigned count);
1782
1783 ir_swizzle(ir_rvalue *val, const unsigned *components, unsigned count);
1784
1785 ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask);
1786
1787 virtual ir_swizzle *clone(void *mem_ctx, struct hash_table *) const;
1788
1789 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1790
1791 virtual ir_swizzle *as_swizzle()
1792 {
1793 return this;
1794 }
1795
1796 /**
1797 * Construct an ir_swizzle from the textual representation. Can fail.
1798 */
1799 static ir_swizzle *create(ir_rvalue *, const char *, unsigned vector_length);
1800
1801 virtual void accept(ir_visitor *v)
1802 {
1803 v->visit(this);
1804 }
1805
1806 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1807
1808 bool is_lvalue() const
1809 {
1810 return val->is_lvalue() && !mask.has_duplicates;
1811 }
1812
1813 /**
1814 * Get the variable that is ultimately referenced by an r-value
1815 */
1816 virtual ir_variable *variable_referenced() const;
1817
1818 ir_rvalue *val;
1819 ir_swizzle_mask mask;
1820
1821 private:
1822 /**
1823 * Initialize the mask component of a swizzle
1824 *
1825 * This is used by the \c ir_swizzle constructors.
1826 */
1827 void init_mask(const unsigned *components, unsigned count);
1828 };
1829
1830
1831 class ir_dereference : public ir_rvalue {
1832 public:
1833 virtual ir_dereference *clone(void *mem_ctx, struct hash_table *) const = 0;
1834
1835 virtual ir_dereference *as_dereference()
1836 {
1837 return this;
1838 }
1839
1840 bool is_lvalue() const;
1841
1842 /**
1843 * Get the variable that is ultimately referenced by an r-value
1844 */
1845 virtual ir_variable *variable_referenced() const = 0;
1846
1847 /**
1848 * Get the constant that is ultimately referenced by an r-value,
1849 * in a constant expression evaluation context.
1850 *
1851 * The offset is used when the reference is to a specific column of
1852 * a matrix.
1853 */
1854 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const = 0;
1855 };
1856
1857
1858 class ir_dereference_variable : public ir_dereference {
1859 public:
1860 ir_dereference_variable(ir_variable *var);
1861
1862 virtual ir_dereference_variable *clone(void *mem_ctx,
1863 struct hash_table *) const;
1864
1865 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1866
1867 virtual ir_dereference_variable *as_dereference_variable()
1868 {
1869 return this;
1870 }
1871
1872 /**
1873 * Get the variable that is ultimately referenced by an r-value
1874 */
1875 virtual ir_variable *variable_referenced() const
1876 {
1877 return this->var;
1878 }
1879
1880 /**
1881 * Get the constant that is ultimately referenced by an r-value,
1882 * in a constant expression evaluation context.
1883 *
1884 * The offset is used when the reference is to a specific column of
1885 * a matrix.
1886 */
1887 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
1888
1889 virtual ir_variable *whole_variable_referenced()
1890 {
1891 /* ir_dereference_variable objects always dereference the entire
1892 * variable. However, if this dereference is dereferenced by anything
1893 * else, the complete deferefernce chain is not a whole-variable
1894 * dereference. This method should only be called on the top most
1895 * ir_rvalue in a dereference chain.
1896 */
1897 return this->var;
1898 }
1899
1900 virtual void accept(ir_visitor *v)
1901 {
1902 v->visit(this);
1903 }
1904
1905 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1906
1907 /**
1908 * Object being dereferenced.
1909 */
1910 ir_variable *var;
1911 };
1912
1913
1914 class ir_dereference_array : public ir_dereference {
1915 public:
1916 ir_dereference_array(ir_rvalue *value, ir_rvalue *array_index);
1917
1918 ir_dereference_array(ir_variable *var, ir_rvalue *array_index);
1919
1920 virtual ir_dereference_array *clone(void *mem_ctx,
1921 struct hash_table *) const;
1922
1923 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1924
1925 virtual ir_dereference_array *as_dereference_array()
1926 {
1927 return this;
1928 }
1929
1930 /**
1931 * Get the variable that is ultimately referenced by an r-value
1932 */
1933 virtual ir_variable *variable_referenced() const
1934 {
1935 return this->array->variable_referenced();
1936 }
1937
1938 /**
1939 * Get the constant that is ultimately referenced by an r-value,
1940 * in a constant expression evaluation context.
1941 *
1942 * The offset is used when the reference is to a specific column of
1943 * a matrix.
1944 */
1945 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
1946
1947 virtual void accept(ir_visitor *v)
1948 {
1949 v->visit(this);
1950 }
1951
1952 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1953
1954 ir_rvalue *array;
1955 ir_rvalue *array_index;
1956
1957 private:
1958 void set_array(ir_rvalue *value);
1959 };
1960
1961
1962 class ir_dereference_record : public ir_dereference {
1963 public:
1964 ir_dereference_record(ir_rvalue *value, const char *field);
1965
1966 ir_dereference_record(ir_variable *var, const char *field);
1967
1968 virtual ir_dereference_record *clone(void *mem_ctx,
1969 struct hash_table *) const;
1970
1971 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1972
1973 virtual ir_dereference_record *as_dereference_record()
1974 {
1975 return this;
1976 }
1977
1978 /**
1979 * Get the variable that is ultimately referenced by an r-value
1980 */
1981 virtual ir_variable *variable_referenced() const
1982 {
1983 return this->record->variable_referenced();
1984 }
1985
1986 /**
1987 * Get the constant that is ultimately referenced by an r-value,
1988 * in a constant expression evaluation context.
1989 *
1990 * The offset is used when the reference is to a specific column of
1991 * a matrix.
1992 */
1993 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
1994
1995 virtual void accept(ir_visitor *v)
1996 {
1997 v->visit(this);
1998 }
1999
2000 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2001
2002 ir_rvalue *record;
2003 const char *field;
2004 };
2005
2006
2007 /**
2008 * Data stored in an ir_constant
2009 */
2010 union ir_constant_data {
2011 unsigned u[16];
2012 int i[16];
2013 float f[16];
2014 bool b[16];
2015 };
2016
2017
2018 class ir_constant : public ir_rvalue {
2019 public:
2020 ir_constant(const struct glsl_type *type, const ir_constant_data *data);
2021 ir_constant(bool b, unsigned vector_elements=1);
2022 ir_constant(unsigned int u, unsigned vector_elements=1);
2023 ir_constant(int i, unsigned vector_elements=1);
2024 ir_constant(float f, unsigned vector_elements=1);
2025
2026 /**
2027 * Construct an ir_constant from a list of ir_constant values
2028 */
2029 ir_constant(const struct glsl_type *type, exec_list *values);
2030
2031 /**
2032 * Construct an ir_constant from a scalar component of another ir_constant
2033 *
2034 * The new \c ir_constant inherits the type of the component from the
2035 * source constant.
2036 *
2037 * \note
2038 * In the case of a matrix constant, the new constant is a scalar, \b not
2039 * a vector.
2040 */
2041 ir_constant(const ir_constant *c, unsigned i);
2042
2043 /**
2044 * Return a new ir_constant of the specified type containing all zeros.
2045 */
2046 static ir_constant *zero(void *mem_ctx, const glsl_type *type);
2047
2048 virtual ir_constant *clone(void *mem_ctx, struct hash_table *) const;
2049
2050 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
2051
2052 virtual ir_constant *as_constant()
2053 {
2054 return this;
2055 }
2056
2057 virtual void accept(ir_visitor *v)
2058 {
2059 v->visit(this);
2060 }
2061
2062 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2063
2064 /**
2065 * Get a particular component of a constant as a specific type
2066 *
2067 * This is useful, for example, to get a value from an integer constant
2068 * as a float or bool. This appears frequently when constructors are
2069 * called with all constant parameters.
2070 */
2071 /*@{*/
2072 bool get_bool_component(unsigned i) const;
2073 float get_float_component(unsigned i) const;
2074 int get_int_component(unsigned i) const;
2075 unsigned get_uint_component(unsigned i) const;
2076 /*@}*/
2077
2078 ir_constant *get_array_element(unsigned i) const;
2079
2080 ir_constant *get_record_field(const char *name);
2081
2082 /**
2083 * Copy the values on another constant at a given offset.
2084 *
2085 * The offset is ignored for array or struct copies, it's only for
2086 * scalars or vectors into vectors or matrices.
2087 *
2088 * With identical types on both sides and zero offset it's clone()
2089 * without creating a new object.
2090 */
2091
2092 void copy_offset(ir_constant *src, int offset);
2093
2094 /**
2095 * Copy the values on another constant at a given offset and
2096 * following an assign-like mask.
2097 *
2098 * The mask is ignored for scalars.
2099 *
2100 * Note that this function only handles what assign can handle,
2101 * i.e. at most a vector as source and a column of a matrix as
2102 * destination.
2103 */
2104
2105 void copy_masked_offset(ir_constant *src, int offset, unsigned int mask);
2106
2107 /**
2108 * Determine whether a constant has the same value as another constant
2109 *
2110 * \sa ir_constant::is_zero, ir_constant::is_one,
2111 * ir_constant::is_negative_one, ir_constant::is_basis
2112 */
2113 bool has_value(const ir_constant *) const;
2114
2115 virtual bool is_zero() const;
2116 virtual bool is_one() const;
2117 virtual bool is_negative_one() const;
2118 virtual bool is_basis() const;
2119
2120 /**
2121 * Value of the constant.
2122 *
2123 * The field used to back the values supplied by the constant is determined
2124 * by the type associated with the \c ir_instruction. Constants may be
2125 * scalars, vectors, or matrices.
2126 */
2127 union ir_constant_data value;
2128
2129 /* Array elements */
2130 ir_constant **array_elements;
2131
2132 /* Structure fields */
2133 exec_list components;
2134
2135 private:
2136 /**
2137 * Parameterless constructor only used by the clone method
2138 */
2139 ir_constant(void);
2140 };
2141
2142 /*@}*/
2143
2144 /**
2145 * IR instruction to emit a vertex in a geometry shader.
2146 */
2147 class ir_emit_vertex : public ir_instruction {
2148 public:
2149 ir_emit_vertex()
2150 {
2151 ir_type = ir_type_emit_vertex;
2152 }
2153
2154 virtual void accept(ir_visitor *v)
2155 {
2156 v->visit(this);
2157 }
2158
2159 virtual ir_emit_vertex *clone(void *mem_ctx, struct hash_table *) const
2160 {
2161 return new(mem_ctx) ir_emit_vertex();
2162 }
2163
2164 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2165 };
2166
2167 /**
2168 * IR instruction to complete the current primitive and start a new one in a
2169 * geometry shader.
2170 */
2171 class ir_end_primitive : public ir_instruction {
2172 public:
2173 ir_end_primitive()
2174 {
2175 ir_type = ir_type_end_primitive;
2176 }
2177
2178 virtual void accept(ir_visitor *v)
2179 {
2180 v->visit(this);
2181 }
2182
2183 virtual ir_end_primitive *clone(void *mem_ctx, struct hash_table *) const
2184 {
2185 return new(mem_ctx) ir_end_primitive();
2186 }
2187
2188 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2189 };
2190
2191 /**
2192 * Apply a visitor to each IR node in a list
2193 */
2194 void
2195 visit_exec_list(exec_list *list, ir_visitor *visitor);
2196
2197 /**
2198 * Validate invariants on each IR node in a list
2199 */
2200 void validate_ir_tree(exec_list *instructions);
2201
2202 struct _mesa_glsl_parse_state;
2203 struct gl_shader_program;
2204
2205 /**
2206 * Detect whether an unlinked shader contains static recursion
2207 *
2208 * If the list of instructions is determined to contain static recursion,
2209 * \c _mesa_glsl_error will be called to emit error messages for each function
2210 * that is in the recursion cycle.
2211 */
2212 void
2213 detect_recursion_unlinked(struct _mesa_glsl_parse_state *state,
2214 exec_list *instructions);
2215
2216 /**
2217 * Detect whether a linked shader contains static recursion
2218 *
2219 * If the list of instructions is determined to contain static recursion,
2220 * \c link_error_printf will be called to emit error messages for each function
2221 * that is in the recursion cycle. In addition,
2222 * \c gl_shader_program::LinkStatus will be set to false.
2223 */
2224 void
2225 detect_recursion_linked(struct gl_shader_program *prog,
2226 exec_list *instructions);
2227
2228 /**
2229 * Make a clone of each IR instruction in a list
2230 *
2231 * \param in List of IR instructions that are to be cloned
2232 * \param out List to hold the cloned instructions
2233 */
2234 void
2235 clone_ir_list(void *mem_ctx, exec_list *out, const exec_list *in);
2236
2237 extern void
2238 _mesa_glsl_initialize_variables(exec_list *instructions,
2239 struct _mesa_glsl_parse_state *state);
2240
2241 extern void
2242 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state *state);
2243
2244 extern void
2245 _mesa_glsl_initialize_builtin_functions();
2246
2247 extern ir_function_signature *
2248 _mesa_glsl_find_builtin_function(_mesa_glsl_parse_state *state,
2249 const char *name, exec_list *actual_parameters);
2250
2251 extern void
2252 _mesa_glsl_release_functions(void);
2253
2254 extern void
2255 _mesa_glsl_release_builtin_functions(void);
2256
2257 extern void
2258 reparent_ir(exec_list *list, void *mem_ctx);
2259
2260 struct glsl_symbol_table;
2261
2262 extern void
2263 import_prototypes(const exec_list *source, exec_list *dest,
2264 struct glsl_symbol_table *symbols, void *mem_ctx);
2265
2266 extern bool
2267 ir_has_call(ir_instruction *ir);
2268
2269 extern void
2270 do_set_program_inouts(exec_list *instructions, struct gl_program *prog,
2271 GLenum shader_type);
2272
2273 extern char *
2274 prototype_string(const glsl_type *return_type, const char *name,
2275 exec_list *parameters);
2276
2277 extern "C" {
2278 #endif /* __cplusplus */
2279
2280 extern void _mesa_print_ir(struct exec_list *instructions,
2281 struct _mesa_glsl_parse_state *state);
2282
2283 #ifdef __cplusplus
2284 } /* extern "C" */
2285 #endif
2286
2287 unsigned
2288 vertices_per_prim(GLenum prim);
2289
2290 #endif /* IR_H */