5b30fe59b91ba2f378a6f64abc0d6215fb6a6586
[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 * Non-zero if this variable was created by lowering a named interface
585 * block which was not an array.
586 *
587 * Note that this variable and \c from_named_ifc_block_array will never
588 * both be non-zero.
589 */
590 unsigned from_named_ifc_block_nonarray:1;
591
592 /**
593 * Non-zero if this variable was created by lowering a named interface
594 * block which was an array.
595 *
596 * Note that this variable and \c from_named_ifc_block_nonarray will never
597 * both be non-zero.
598 */
599 unsigned from_named_ifc_block_array:1;
600
601 /**
602 * \brief Layout qualifier for gl_FragDepth.
603 *
604 * This is not equal to \c ir_depth_layout_none if and only if this
605 * variable is \c gl_FragDepth and a layout qualifier is specified.
606 */
607 ir_depth_layout depth_layout;
608
609 /**
610 * Storage location of the base of this variable
611 *
612 * The precise meaning of this field depends on the nature of the variable.
613 *
614 * - Vertex shader input: one of the values from \c gl_vert_attrib.
615 * - Vertex shader output: one of the values from \c gl_varying_slot.
616 * - Geometry shader input: one of the values from \c gl_varying_slot.
617 * - Geometry shader output: one of the values from \c gl_varying_slot.
618 * - Fragment shader input: one of the values from \c gl_varying_slot.
619 * - Fragment shader output: one of the values from \c gl_frag_result.
620 * - Uniforms: Per-stage uniform slot number for default uniform block.
621 * - Uniforms: Index within the uniform block definition for UBO members.
622 * - Other: This field is not currently used.
623 *
624 * If the variable is a uniform, shader input, or shader output, and the
625 * slot has not been assigned, the value will be -1.
626 */
627 int location;
628
629 /**
630 * output index for dual source blending.
631 */
632 int index;
633
634 /**
635 * Initial binding point for a sampler or UBO.
636 *
637 * For array types, this represents the binding point for the first element.
638 */
639 int binding;
640
641 /**
642 * Location an atomic counter is stored at.
643 */
644 struct {
645 unsigned buffer_index;
646 unsigned offset;
647 } atomic;
648
649 /**
650 * Built-in state that backs this uniform
651 *
652 * Once set at variable creation, \c state_slots must remain invariant.
653 * This is because, ideally, this array would be shared by all clones of
654 * this variable in the IR tree. In other words, we'd really like for it
655 * to be a fly-weight.
656 *
657 * If the variable is not a uniform, \c num_state_slots will be zero and
658 * \c state_slots will be \c NULL.
659 */
660 /*@{*/
661 unsigned num_state_slots; /**< Number of state slots used */
662 ir_state_slot *state_slots; /**< State descriptors. */
663 /*@}*/
664
665 /**
666 * Emit a warning if this variable is accessed.
667 */
668 const char *warn_extension;
669
670 /**
671 * Value assigned in the initializer of a variable declared "const"
672 */
673 ir_constant *constant_value;
674
675 /**
676 * Constant expression assigned in the initializer of the variable
677 *
678 * \warning
679 * This field and \c ::constant_value are distinct. Even if the two fields
680 * refer to constants with the same value, they must point to separate
681 * objects.
682 */
683 ir_constant *constant_initializer;
684
685 private:
686 /**
687 * For variables that are in an interface block or are an instance of an
688 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
689 *
690 * \sa ir_variable::location
691 */
692 const glsl_type *interface_type;
693 };
694
695 /**
696 * A function that returns whether a built-in function is available in the
697 * current shading language (based on version, ES or desktop, and extensions).
698 */
699 typedef bool (*builtin_available_predicate)(const _mesa_glsl_parse_state *);
700
701 /*@{*/
702 /**
703 * The representation of a function instance; may be the full definition or
704 * simply a prototype.
705 */
706 class ir_function_signature : public ir_instruction {
707 /* An ir_function_signature will be part of the list of signatures in
708 * an ir_function.
709 */
710 public:
711 ir_function_signature(const glsl_type *return_type,
712 builtin_available_predicate builtin_avail = NULL);
713
714 virtual ir_function_signature *clone(void *mem_ctx,
715 struct hash_table *ht) const;
716 ir_function_signature *clone_prototype(void *mem_ctx,
717 struct hash_table *ht) const;
718
719 virtual void accept(ir_visitor *v)
720 {
721 v->visit(this);
722 }
723
724 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
725
726 /**
727 * Attempt to evaluate this function as a constant expression,
728 * given a list of the actual parameters and the variable context.
729 * Returns NULL for non-built-ins.
730 */
731 ir_constant *constant_expression_value(exec_list *actual_parameters, struct hash_table *variable_context);
732
733 /**
734 * Get the name of the function for which this is a signature
735 */
736 const char *function_name() const;
737
738 /**
739 * Get a handle to the function for which this is a signature
740 *
741 * There is no setter function, this function returns a \c const pointer,
742 * and \c ir_function_signature::_function is private for a reason. The
743 * only way to make a connection between a function and function signature
744 * is via \c ir_function::add_signature. This helps ensure that certain
745 * invariants (i.e., a function signature is in the list of signatures for
746 * its \c _function) are met.
747 *
748 * \sa ir_function::add_signature
749 */
750 inline const class ir_function *function() const
751 {
752 return this->_function;
753 }
754
755 /**
756 * Check whether the qualifiers match between this signature's parameters
757 * and the supplied parameter list. If not, returns the name of the first
758 * parameter with mismatched qualifiers (for use in error messages).
759 */
760 const char *qualifiers_match(exec_list *params);
761
762 /**
763 * Replace the current parameter list with the given one. This is useful
764 * if the current information came from a prototype, and either has invalid
765 * or missing parameter names.
766 */
767 void replace_parameters(exec_list *new_params);
768
769 /**
770 * Function return type.
771 *
772 * \note This discards the optional precision qualifier.
773 */
774 const struct glsl_type *return_type;
775
776 /**
777 * List of ir_variable of function parameters.
778 *
779 * This represents the storage. The paramaters passed in a particular
780 * call will be in ir_call::actual_paramaters.
781 */
782 struct exec_list parameters;
783
784 /** Whether or not this function has a body (which may be empty). */
785 unsigned is_defined:1;
786
787 /** Whether or not this function signature is a built-in. */
788 bool is_builtin() const;
789
790 /**
791 * Whether or not this function is an intrinsic to be implemented
792 * by the driver.
793 */
794 bool is_intrinsic;
795
796 /** Whether or not a built-in is available for this shader. */
797 bool is_builtin_available(const _mesa_glsl_parse_state *state) const;
798
799 /** Body of instructions in the function. */
800 struct exec_list body;
801
802 private:
803 /**
804 * A function pointer to a predicate that answers whether a built-in
805 * function is available in the current shader. NULL if not a built-in.
806 */
807 builtin_available_predicate builtin_avail;
808
809 /** Function of which this signature is one overload. */
810 class ir_function *_function;
811
812 /** Function signature of which this one is a prototype clone */
813 const ir_function_signature *origin;
814
815 friend class ir_function;
816
817 /**
818 * Helper function to run a list of instructions for constant
819 * expression evaluation.
820 *
821 * The hash table represents the values of the visible variables.
822 * There are no scoping issues because the table is indexed on
823 * ir_variable pointers, not variable names.
824 *
825 * Returns false if the expression is not constant, true otherwise,
826 * and the value in *result if result is non-NULL.
827 */
828 bool constant_expression_evaluate_expression_list(const struct exec_list &body,
829 struct hash_table *variable_context,
830 ir_constant **result);
831 };
832
833
834 /**
835 * Header for tracking multiple overloaded functions with the same name.
836 * Contains a list of ir_function_signatures representing each of the
837 * actual functions.
838 */
839 class ir_function : public ir_instruction {
840 public:
841 ir_function(const char *name);
842
843 virtual ir_function *clone(void *mem_ctx, struct hash_table *ht) const;
844
845 virtual ir_function *as_function()
846 {
847 return this;
848 }
849
850 virtual void accept(ir_visitor *v)
851 {
852 v->visit(this);
853 }
854
855 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
856
857 void add_signature(ir_function_signature *sig)
858 {
859 sig->_function = this;
860 this->signatures.push_tail(sig);
861 }
862
863 /**
864 * Get an iterator for the set of function signatures
865 */
866 exec_list_iterator iterator()
867 {
868 return signatures.iterator();
869 }
870
871 /**
872 * Find a signature that matches a set of actual parameters, taking implicit
873 * conversions into account. Also flags whether the match was exact.
874 */
875 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
876 const exec_list *actual_param,
877 bool *match_is_exact);
878
879 /**
880 * Find a signature that matches a set of actual parameters, taking implicit
881 * conversions into account.
882 */
883 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
884 const exec_list *actual_param);
885
886 /**
887 * Find a signature that exactly matches a set of actual parameters without
888 * any implicit type conversions.
889 */
890 ir_function_signature *exact_matching_signature(_mesa_glsl_parse_state *state,
891 const exec_list *actual_ps);
892
893 /**
894 * Name of the function.
895 */
896 const char *name;
897
898 /** Whether or not this function has a signature that isn't a built-in. */
899 bool has_user_signature();
900
901 /**
902 * List of ir_function_signature for each overloaded function with this name.
903 */
904 struct exec_list signatures;
905 };
906
907 inline const char *ir_function_signature::function_name() const
908 {
909 return this->_function->name;
910 }
911 /*@}*/
912
913
914 /**
915 * IR instruction representing high-level if-statements
916 */
917 class ir_if : public ir_instruction {
918 public:
919 ir_if(ir_rvalue *condition)
920 : condition(condition)
921 {
922 ir_type = ir_type_if;
923 }
924
925 virtual ir_if *clone(void *mem_ctx, struct hash_table *ht) const;
926
927 virtual ir_if *as_if()
928 {
929 return this;
930 }
931
932 virtual void accept(ir_visitor *v)
933 {
934 v->visit(this);
935 }
936
937 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
938
939 ir_rvalue *condition;
940 /** List of ir_instruction for the body of the then branch */
941 exec_list then_instructions;
942 /** List of ir_instruction for the body of the else branch */
943 exec_list else_instructions;
944 };
945
946
947 /**
948 * IR instruction representing a high-level loop structure.
949 */
950 class ir_loop : public ir_instruction {
951 public:
952 ir_loop();
953
954 virtual ir_loop *clone(void *mem_ctx, struct hash_table *ht) const;
955
956 virtual void accept(ir_visitor *v)
957 {
958 v->visit(this);
959 }
960
961 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
962
963 virtual ir_loop *as_loop()
964 {
965 return this;
966 }
967
968 /**
969 * Get an iterator for the instructions of the loop body
970 */
971 exec_list_iterator iterator()
972 {
973 return body_instructions.iterator();
974 }
975
976 /** List of ir_instruction that make up the body of the loop. */
977 exec_list body_instructions;
978
979 /**
980 * \name Loop counter and controls
981 *
982 * Represents a loop like a FORTRAN \c do-loop.
983 *
984 * \note
985 * If \c from and \c to are the same value, the loop will execute once.
986 */
987 /*@{*/
988 ir_rvalue *from; /** Value of the loop counter on the first
989 * iteration of the loop.
990 */
991 ir_rvalue *to; /** Value of the loop counter on the last
992 * iteration of the loop.
993 */
994 ir_rvalue *increment;
995 ir_variable *counter;
996
997 /**
998 * Comparison operation in the loop terminator.
999 *
1000 * If any of the loop control fields are non-\c NULL, this field must be
1001 * one of \c ir_binop_less, \c ir_binop_greater, \c ir_binop_lequal,
1002 * \c ir_binop_gequal, \c ir_binop_equal, or \c ir_binop_nequal.
1003 */
1004 int cmp;
1005 /*@}*/
1006 };
1007
1008
1009 class ir_assignment : public ir_instruction {
1010 public:
1011 ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs, ir_rvalue *condition = NULL);
1012
1013 /**
1014 * Construct an assignment with an explicit write mask
1015 *
1016 * \note
1017 * Since a write mask is supplied, the LHS must already be a bare
1018 * \c ir_dereference. The cannot be any swizzles in the LHS.
1019 */
1020 ir_assignment(ir_dereference *lhs, ir_rvalue *rhs, ir_rvalue *condition,
1021 unsigned write_mask);
1022
1023 virtual ir_assignment *clone(void *mem_ctx, struct hash_table *ht) const;
1024
1025 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1026
1027 virtual void accept(ir_visitor *v)
1028 {
1029 v->visit(this);
1030 }
1031
1032 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1033
1034 virtual ir_assignment * as_assignment()
1035 {
1036 return this;
1037 }
1038
1039 /**
1040 * Get a whole variable written by an assignment
1041 *
1042 * If the LHS of the assignment writes a whole variable, the variable is
1043 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1044 * assignment are:
1045 *
1046 * - Assigning to a scalar
1047 * - Assigning to all components of a vector
1048 * - Whole array (or matrix) assignment
1049 * - Whole structure assignment
1050 */
1051 ir_variable *whole_variable_written();
1052
1053 /**
1054 * Set the LHS of an assignment
1055 */
1056 void set_lhs(ir_rvalue *lhs);
1057
1058 /**
1059 * Left-hand side of the assignment.
1060 *
1061 * This should be treated as read only. If you need to set the LHS of an
1062 * assignment, use \c ir_assignment::set_lhs.
1063 */
1064 ir_dereference *lhs;
1065
1066 /**
1067 * Value being assigned
1068 */
1069 ir_rvalue *rhs;
1070
1071 /**
1072 * Optional condition for the assignment.
1073 */
1074 ir_rvalue *condition;
1075
1076
1077 /**
1078 * Component mask written
1079 *
1080 * For non-vector types in the LHS, this field will be zero. For vector
1081 * types, a bit will be set for each component that is written. Note that
1082 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1083 *
1084 * A partially-set write mask means that each enabled channel gets
1085 * the value from a consecutive channel of the rhs. For example,
1086 * to write just .xyw of gl_FrontColor with color:
1087 *
1088 * (assign (constant bool (1)) (xyw)
1089 * (var_ref gl_FragColor)
1090 * (swiz xyw (var_ref color)))
1091 */
1092 unsigned write_mask:4;
1093 };
1094
1095 /* Update ir_expression::get_num_operands() and operator_strs when
1096 * updating this list.
1097 */
1098 enum ir_expression_operation {
1099 ir_unop_bit_not,
1100 ir_unop_logic_not,
1101 ir_unop_neg,
1102 ir_unop_abs,
1103 ir_unop_sign,
1104 ir_unop_rcp,
1105 ir_unop_rsq,
1106 ir_unop_sqrt,
1107 ir_unop_exp, /**< Log base e on gentype */
1108 ir_unop_log, /**< Natural log on gentype */
1109 ir_unop_exp2,
1110 ir_unop_log2,
1111 ir_unop_f2i, /**< Float-to-integer conversion. */
1112 ir_unop_f2u, /**< Float-to-unsigned conversion. */
1113 ir_unop_i2f, /**< Integer-to-float conversion. */
1114 ir_unop_f2b, /**< Float-to-boolean conversion */
1115 ir_unop_b2f, /**< Boolean-to-float conversion */
1116 ir_unop_i2b, /**< int-to-boolean conversion */
1117 ir_unop_b2i, /**< Boolean-to-int conversion */
1118 ir_unop_u2f, /**< Unsigned-to-float conversion. */
1119 ir_unop_i2u, /**< Integer-to-unsigned conversion. */
1120 ir_unop_u2i, /**< Unsigned-to-integer conversion. */
1121 ir_unop_bitcast_i2f, /**< Bit-identical int-to-float "conversion" */
1122 ir_unop_bitcast_f2i, /**< Bit-identical float-to-int "conversion" */
1123 ir_unop_bitcast_u2f, /**< Bit-identical uint-to-float "conversion" */
1124 ir_unop_bitcast_f2u, /**< Bit-identical float-to-uint "conversion" */
1125 ir_unop_any,
1126
1127 /**
1128 * \name Unary floating-point rounding operations.
1129 */
1130 /*@{*/
1131 ir_unop_trunc,
1132 ir_unop_ceil,
1133 ir_unop_floor,
1134 ir_unop_fract,
1135 ir_unop_round_even,
1136 /*@}*/
1137
1138 /**
1139 * \name Trigonometric operations.
1140 */
1141 /*@{*/
1142 ir_unop_sin,
1143 ir_unop_cos,
1144 ir_unop_sin_reduced, /**< Reduced range sin. [-pi, pi] */
1145 ir_unop_cos_reduced, /**< Reduced range cos. [-pi, pi] */
1146 /*@}*/
1147
1148 /**
1149 * \name Partial derivatives.
1150 */
1151 /*@{*/
1152 ir_unop_dFdx,
1153 ir_unop_dFdy,
1154 /*@}*/
1155
1156 /**
1157 * \name Floating point pack and unpack operations.
1158 */
1159 /*@{*/
1160 ir_unop_pack_snorm_2x16,
1161 ir_unop_pack_snorm_4x8,
1162 ir_unop_pack_unorm_2x16,
1163 ir_unop_pack_unorm_4x8,
1164 ir_unop_pack_half_2x16,
1165 ir_unop_unpack_snorm_2x16,
1166 ir_unop_unpack_snorm_4x8,
1167 ir_unop_unpack_unorm_2x16,
1168 ir_unop_unpack_unorm_4x8,
1169 ir_unop_unpack_half_2x16,
1170 /*@}*/
1171
1172 /**
1173 * \name Lowered floating point unpacking operations.
1174 *
1175 * \see lower_packing_builtins_visitor::split_unpack_half_2x16
1176 */
1177 /*@{*/
1178 ir_unop_unpack_half_2x16_split_x,
1179 ir_unop_unpack_half_2x16_split_y,
1180 /*@}*/
1181
1182 /**
1183 * \name Bit operations, part of ARB_gpu_shader5.
1184 */
1185 /*@{*/
1186 ir_unop_bitfield_reverse,
1187 ir_unop_bit_count,
1188 ir_unop_find_msb,
1189 ir_unop_find_lsb,
1190 /*@}*/
1191
1192 ir_unop_noise,
1193
1194 /**
1195 * A sentinel marking the last of the unary operations.
1196 */
1197 ir_last_unop = ir_unop_noise,
1198
1199 ir_binop_add,
1200 ir_binop_sub,
1201 ir_binop_mul, /**< Floating-point or low 32-bit integer multiply. */
1202 ir_binop_imul_high, /**< Calculates the high 32-bits of a 64-bit multiply. */
1203 ir_binop_div,
1204
1205 /**
1206 * Returns the carry resulting from the addition of the two arguments.
1207 */
1208 /*@{*/
1209 ir_binop_carry,
1210 /*@}*/
1211
1212 /**
1213 * Returns the borrow resulting from the subtraction of the second argument
1214 * from the first argument.
1215 */
1216 /*@{*/
1217 ir_binop_borrow,
1218 /*@}*/
1219
1220 /**
1221 * Takes one of two combinations of arguments:
1222 *
1223 * - mod(vecN, vecN)
1224 * - mod(vecN, float)
1225 *
1226 * Does not take integer types.
1227 */
1228 ir_binop_mod,
1229
1230 /**
1231 * \name Binary comparison operators which return a boolean vector.
1232 * The type of both operands must be equal.
1233 */
1234 /*@{*/
1235 ir_binop_less,
1236 ir_binop_greater,
1237 ir_binop_lequal,
1238 ir_binop_gequal,
1239 ir_binop_equal,
1240 ir_binop_nequal,
1241 /**
1242 * Returns single boolean for whether all components of operands[0]
1243 * equal the components of operands[1].
1244 */
1245 ir_binop_all_equal,
1246 /**
1247 * Returns single boolean for whether any component of operands[0]
1248 * is not equal to the corresponding component of operands[1].
1249 */
1250 ir_binop_any_nequal,
1251 /*@}*/
1252
1253 /**
1254 * \name Bit-wise binary operations.
1255 */
1256 /*@{*/
1257 ir_binop_lshift,
1258 ir_binop_rshift,
1259 ir_binop_bit_and,
1260 ir_binop_bit_xor,
1261 ir_binop_bit_or,
1262 /*@}*/
1263
1264 ir_binop_logic_and,
1265 ir_binop_logic_xor,
1266 ir_binop_logic_or,
1267
1268 ir_binop_dot,
1269 ir_binop_min,
1270 ir_binop_max,
1271
1272 ir_binop_pow,
1273
1274 /**
1275 * \name Lowered floating point packing operations.
1276 *
1277 * \see lower_packing_builtins_visitor::split_pack_half_2x16
1278 */
1279 /*@{*/
1280 ir_binop_pack_half_2x16_split,
1281 /*@}*/
1282
1283 /**
1284 * \name First half of a lowered bitfieldInsert() operation.
1285 *
1286 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1287 */
1288 /*@{*/
1289 ir_binop_bfm,
1290 /*@}*/
1291
1292 /**
1293 * Load a value the size of a given GLSL type from a uniform block.
1294 *
1295 * operand0 is the ir_constant uniform block index in the linked shader.
1296 * operand1 is a byte offset within the uniform block.
1297 */
1298 ir_binop_ubo_load,
1299
1300 /**
1301 * \name Multiplies a number by two to a power, part of ARB_gpu_shader5.
1302 */
1303 /*@{*/
1304 ir_binop_ldexp,
1305 /*@}*/
1306
1307 /**
1308 * Extract a scalar from a vector
1309 *
1310 * operand0 is the vector
1311 * operand1 is the index of the field to read from operand0
1312 */
1313 ir_binop_vector_extract,
1314
1315 /**
1316 * A sentinel marking the last of the binary operations.
1317 */
1318 ir_last_binop = ir_binop_vector_extract,
1319
1320 /**
1321 * \name Fused floating-point multiply-add, part of ARB_gpu_shader5.
1322 */
1323 /*@{*/
1324 ir_triop_fma,
1325 /*@}*/
1326
1327 ir_triop_lrp,
1328
1329 /**
1330 * \name Conditional Select
1331 *
1332 * A vector conditional select instruction (like ?:, but operating per-
1333 * component on vectors).
1334 *
1335 * \see lower_instructions_visitor::ldexp_to_arith
1336 */
1337 /*@{*/
1338 ir_triop_csel,
1339 /*@}*/
1340
1341 /**
1342 * \name Second half of a lowered bitfieldInsert() operation.
1343 *
1344 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1345 */
1346 /*@{*/
1347 ir_triop_bfi,
1348 /*@}*/
1349
1350 ir_triop_bitfield_extract,
1351
1352 /**
1353 * Generate a value with one field of a vector changed
1354 *
1355 * operand0 is the vector
1356 * operand1 is the value to write into the vector result
1357 * operand2 is the index in operand0 to be modified
1358 */
1359 ir_triop_vector_insert,
1360
1361 /**
1362 * A sentinel marking the last of the ternary operations.
1363 */
1364 ir_last_triop = ir_triop_vector_insert,
1365
1366 ir_quadop_bitfield_insert,
1367
1368 ir_quadop_vector,
1369
1370 /**
1371 * A sentinel marking the last of the ternary operations.
1372 */
1373 ir_last_quadop = ir_quadop_vector,
1374
1375 /**
1376 * A sentinel marking the last of all operations.
1377 */
1378 ir_last_opcode = ir_quadop_vector
1379 };
1380
1381 class ir_expression : public ir_rvalue {
1382 public:
1383 ir_expression(int op, const struct glsl_type *type,
1384 ir_rvalue *op0, ir_rvalue *op1 = NULL,
1385 ir_rvalue *op2 = NULL, ir_rvalue *op3 = NULL);
1386
1387 /**
1388 * Constructor for unary operation expressions
1389 */
1390 ir_expression(int op, ir_rvalue *);
1391
1392 /**
1393 * Constructor for binary operation expressions
1394 */
1395 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1);
1396
1397 /**
1398 * Constructor for ternary operation expressions
1399 */
1400 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1, ir_rvalue *op2);
1401
1402 virtual ir_expression *as_expression()
1403 {
1404 return this;
1405 }
1406
1407 virtual ir_expression *clone(void *mem_ctx, struct hash_table *ht) const;
1408
1409 /**
1410 * Attempt to constant-fold the expression
1411 *
1412 * The "variable_context" hash table links ir_variable * to ir_constant *
1413 * that represent the variables' values. \c NULL represents an empty
1414 * context.
1415 *
1416 * If the expression cannot be constant folded, this method will return
1417 * \c NULL.
1418 */
1419 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1420
1421 /**
1422 * Determine the number of operands used by an expression
1423 */
1424 static unsigned int get_num_operands(ir_expression_operation);
1425
1426 /**
1427 * Determine the number of operands used by an expression
1428 */
1429 unsigned int get_num_operands() const
1430 {
1431 return (this->operation == ir_quadop_vector)
1432 ? this->type->vector_elements : get_num_operands(operation);
1433 }
1434
1435 /**
1436 * Return a string representing this expression's operator.
1437 */
1438 const char *operator_string();
1439
1440 /**
1441 * Return a string representing this expression's operator.
1442 */
1443 static const char *operator_string(ir_expression_operation);
1444
1445
1446 /**
1447 * Do a reverse-lookup to translate the given string into an operator.
1448 */
1449 static ir_expression_operation get_operator(const char *);
1450
1451 virtual void accept(ir_visitor *v)
1452 {
1453 v->visit(this);
1454 }
1455
1456 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1457
1458 ir_expression_operation operation;
1459 ir_rvalue *operands[4];
1460 };
1461
1462
1463 /**
1464 * HIR instruction representing a high-level function call, containing a list
1465 * of parameters and returning a value in the supplied temporary.
1466 */
1467 class ir_call : public ir_instruction {
1468 public:
1469 ir_call(ir_function_signature *callee,
1470 ir_dereference_variable *return_deref,
1471 exec_list *actual_parameters)
1472 : return_deref(return_deref), callee(callee)
1473 {
1474 ir_type = ir_type_call;
1475 assert(callee->return_type != NULL);
1476 actual_parameters->move_nodes_to(& this->actual_parameters);
1477 this->use_builtin = callee->is_builtin();
1478 }
1479
1480 virtual ir_call *clone(void *mem_ctx, struct hash_table *ht) const;
1481
1482 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1483
1484 virtual ir_call *as_call()
1485 {
1486 return this;
1487 }
1488
1489 virtual void accept(ir_visitor *v)
1490 {
1491 v->visit(this);
1492 }
1493
1494 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1495
1496 /**
1497 * Get an iterator for the set of acutal parameters
1498 */
1499 exec_list_iterator iterator()
1500 {
1501 return actual_parameters.iterator();
1502 }
1503
1504 /**
1505 * Get the name of the function being called.
1506 */
1507 const char *callee_name() const
1508 {
1509 return callee->function_name();
1510 }
1511
1512 /**
1513 * Generates an inline version of the function before @ir,
1514 * storing the return value in return_deref.
1515 */
1516 void generate_inline(ir_instruction *ir);
1517
1518 /**
1519 * Storage for the function's return value.
1520 * This must be NULL if the return type is void.
1521 */
1522 ir_dereference_variable *return_deref;
1523
1524 /**
1525 * The specific function signature being called.
1526 */
1527 ir_function_signature *callee;
1528
1529 /* List of ir_rvalue of paramaters passed in this call. */
1530 exec_list actual_parameters;
1531
1532 /** Should this call only bind to a built-in function? */
1533 bool use_builtin;
1534 };
1535
1536
1537 /**
1538 * \name Jump-like IR instructions.
1539 *
1540 * These include \c break, \c continue, \c return, and \c discard.
1541 */
1542 /*@{*/
1543 class ir_jump : public ir_instruction {
1544 protected:
1545 ir_jump()
1546 {
1547 ir_type = ir_type_unset;
1548 }
1549
1550 public:
1551 virtual ir_jump *as_jump()
1552 {
1553 return this;
1554 }
1555 };
1556
1557 class ir_return : public ir_jump {
1558 public:
1559 ir_return()
1560 : value(NULL)
1561 {
1562 this->ir_type = ir_type_return;
1563 }
1564
1565 ir_return(ir_rvalue *value)
1566 : value(value)
1567 {
1568 this->ir_type = ir_type_return;
1569 }
1570
1571 virtual ir_return *clone(void *mem_ctx, struct hash_table *) const;
1572
1573 virtual ir_return *as_return()
1574 {
1575 return this;
1576 }
1577
1578 ir_rvalue *get_value() const
1579 {
1580 return value;
1581 }
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 ir_rvalue *value;
1591 };
1592
1593
1594 /**
1595 * Jump instructions used inside loops
1596 *
1597 * These include \c break and \c continue. The \c break within a loop is
1598 * different from the \c break within a switch-statement.
1599 *
1600 * \sa ir_switch_jump
1601 */
1602 class ir_loop_jump : public ir_jump {
1603 public:
1604 enum jump_mode {
1605 jump_break,
1606 jump_continue
1607 };
1608
1609 ir_loop_jump(jump_mode mode)
1610 {
1611 this->ir_type = ir_type_loop_jump;
1612 this->mode = mode;
1613 }
1614
1615 virtual ir_loop_jump *clone(void *mem_ctx, struct hash_table *) const;
1616
1617 virtual void accept(ir_visitor *v)
1618 {
1619 v->visit(this);
1620 }
1621
1622 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1623
1624 bool is_break() const
1625 {
1626 return mode == jump_break;
1627 }
1628
1629 bool is_continue() const
1630 {
1631 return mode == jump_continue;
1632 }
1633
1634 /** Mode selector for the jump instruction. */
1635 enum jump_mode mode;
1636 };
1637
1638 /**
1639 * IR instruction representing discard statements.
1640 */
1641 class ir_discard : public ir_jump {
1642 public:
1643 ir_discard()
1644 {
1645 this->ir_type = ir_type_discard;
1646 this->condition = NULL;
1647 }
1648
1649 ir_discard(ir_rvalue *cond)
1650 {
1651 this->ir_type = ir_type_discard;
1652 this->condition = cond;
1653 }
1654
1655 virtual ir_discard *clone(void *mem_ctx, struct hash_table *ht) const;
1656
1657 virtual void accept(ir_visitor *v)
1658 {
1659 v->visit(this);
1660 }
1661
1662 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1663
1664 virtual ir_discard *as_discard()
1665 {
1666 return this;
1667 }
1668
1669 ir_rvalue *condition;
1670 };
1671 /*@}*/
1672
1673
1674 /**
1675 * Texture sampling opcodes used in ir_texture
1676 */
1677 enum ir_texture_opcode {
1678 ir_tex, /**< Regular texture look-up */
1679 ir_txb, /**< Texture look-up with LOD bias */
1680 ir_txl, /**< Texture look-up with explicit LOD */
1681 ir_txd, /**< Texture look-up with partial derivatvies */
1682 ir_txf, /**< Texel fetch with explicit LOD */
1683 ir_txf_ms, /**< Multisample texture fetch */
1684 ir_txs, /**< Texture size */
1685 ir_lod, /**< Texture lod query */
1686 ir_tg4, /**< Texture gather */
1687 ir_query_levels /**< Texture levels query */
1688 };
1689
1690
1691 /**
1692 * IR instruction to sample a texture
1693 *
1694 * The specific form of the IR instruction depends on the \c mode value
1695 * selected from \c ir_texture_opcodes. In the printed IR, these will
1696 * appear as:
1697 *
1698 * Texel offset (0 or an expression)
1699 * | Projection divisor
1700 * | | Shadow comparitor
1701 * | | |
1702 * v v v
1703 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1704 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1705 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1706 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1707 * (txf <type> <sampler> <coordinate> 0 <lod>)
1708 * (txf_ms
1709 * <type> <sampler> <coordinate> <sample_index>)
1710 * (txs <type> <sampler> <lod>)
1711 * (lod <type> <sampler> <coordinate>)
1712 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
1713 * (query_levels <type> <sampler>)
1714 */
1715 class ir_texture : public ir_rvalue {
1716 public:
1717 ir_texture(enum ir_texture_opcode op)
1718 : op(op), sampler(NULL), coordinate(NULL), projector(NULL),
1719 shadow_comparitor(NULL), offset(NULL)
1720 {
1721 this->ir_type = ir_type_texture;
1722 memset(&lod_info, 0, sizeof(lod_info));
1723 }
1724
1725 virtual ir_texture *clone(void *mem_ctx, struct hash_table *) const;
1726
1727 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1728
1729 virtual void accept(ir_visitor *v)
1730 {
1731 v->visit(this);
1732 }
1733
1734 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1735
1736 /**
1737 * Return a string representing the ir_texture_opcode.
1738 */
1739 const char *opcode_string();
1740
1741 /** Set the sampler and type. */
1742 void set_sampler(ir_dereference *sampler, const glsl_type *type);
1743
1744 /**
1745 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1746 */
1747 static ir_texture_opcode get_opcode(const char *);
1748
1749 enum ir_texture_opcode op;
1750
1751 /** Sampler to use for the texture access. */
1752 ir_dereference *sampler;
1753
1754 /** Texture coordinate to sample */
1755 ir_rvalue *coordinate;
1756
1757 /**
1758 * Value used for projective divide.
1759 *
1760 * If there is no projective divide (the common case), this will be
1761 * \c NULL. Optimization passes should check for this to point to a constant
1762 * of 1.0 and replace that with \c NULL.
1763 */
1764 ir_rvalue *projector;
1765
1766 /**
1767 * Coordinate used for comparison on shadow look-ups.
1768 *
1769 * If there is no shadow comparison, this will be \c NULL. For the
1770 * \c ir_txf opcode, this *must* be \c NULL.
1771 */
1772 ir_rvalue *shadow_comparitor;
1773
1774 /** Texel offset. */
1775 ir_rvalue *offset;
1776
1777 union {
1778 ir_rvalue *lod; /**< Floating point LOD */
1779 ir_rvalue *bias; /**< Floating point LOD bias */
1780 ir_rvalue *sample_index; /**< MSAA sample index */
1781 ir_rvalue *component; /**< Gather component selector */
1782 struct {
1783 ir_rvalue *dPdx; /**< Partial derivative of coordinate wrt X */
1784 ir_rvalue *dPdy; /**< Partial derivative of coordinate wrt Y */
1785 } grad;
1786 } lod_info;
1787 };
1788
1789
1790 struct ir_swizzle_mask {
1791 unsigned x:2;
1792 unsigned y:2;
1793 unsigned z:2;
1794 unsigned w:2;
1795
1796 /**
1797 * Number of components in the swizzle.
1798 */
1799 unsigned num_components:3;
1800
1801 /**
1802 * Does the swizzle contain duplicate components?
1803 *
1804 * L-value swizzles cannot contain duplicate components.
1805 */
1806 unsigned has_duplicates:1;
1807 };
1808
1809
1810 class ir_swizzle : public ir_rvalue {
1811 public:
1812 ir_swizzle(ir_rvalue *, unsigned x, unsigned y, unsigned z, unsigned w,
1813 unsigned count);
1814
1815 ir_swizzle(ir_rvalue *val, const unsigned *components, unsigned count);
1816
1817 ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask);
1818
1819 virtual ir_swizzle *clone(void *mem_ctx, struct hash_table *) const;
1820
1821 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1822
1823 virtual ir_swizzle *as_swizzle()
1824 {
1825 return this;
1826 }
1827
1828 /**
1829 * Construct an ir_swizzle from the textual representation. Can fail.
1830 */
1831 static ir_swizzle *create(ir_rvalue *, const char *, unsigned vector_length);
1832
1833 virtual void accept(ir_visitor *v)
1834 {
1835 v->visit(this);
1836 }
1837
1838 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1839
1840 bool is_lvalue() const
1841 {
1842 return val->is_lvalue() && !mask.has_duplicates;
1843 }
1844
1845 /**
1846 * Get the variable that is ultimately referenced by an r-value
1847 */
1848 virtual ir_variable *variable_referenced() const;
1849
1850 ir_rvalue *val;
1851 ir_swizzle_mask mask;
1852
1853 private:
1854 /**
1855 * Initialize the mask component of a swizzle
1856 *
1857 * This is used by the \c ir_swizzle constructors.
1858 */
1859 void init_mask(const unsigned *components, unsigned count);
1860 };
1861
1862
1863 class ir_dereference : public ir_rvalue {
1864 public:
1865 virtual ir_dereference *clone(void *mem_ctx, struct hash_table *) const = 0;
1866
1867 virtual ir_dereference *as_dereference()
1868 {
1869 return this;
1870 }
1871
1872 bool is_lvalue() const;
1873
1874 /**
1875 * Get the variable that is ultimately referenced by an r-value
1876 */
1877 virtual ir_variable *variable_referenced() const = 0;
1878
1879 /**
1880 * Get the constant that is ultimately referenced by an r-value,
1881 * in a constant expression evaluation context.
1882 *
1883 * The offset is used when the reference is to a specific column of
1884 * a matrix.
1885 */
1886 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const = 0;
1887 };
1888
1889
1890 class ir_dereference_variable : public ir_dereference {
1891 public:
1892 ir_dereference_variable(ir_variable *var);
1893
1894 virtual ir_dereference_variable *clone(void *mem_ctx,
1895 struct hash_table *) const;
1896
1897 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1898
1899 virtual ir_dereference_variable *as_dereference_variable()
1900 {
1901 return this;
1902 }
1903
1904 /**
1905 * Get the variable that is ultimately referenced by an r-value
1906 */
1907 virtual ir_variable *variable_referenced() const
1908 {
1909 return this->var;
1910 }
1911
1912 /**
1913 * Get the constant that is ultimately referenced by an r-value,
1914 * in a constant expression evaluation context.
1915 *
1916 * The offset is used when the reference is to a specific column of
1917 * a matrix.
1918 */
1919 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
1920
1921 virtual ir_variable *whole_variable_referenced()
1922 {
1923 /* ir_dereference_variable objects always dereference the entire
1924 * variable. However, if this dereference is dereferenced by anything
1925 * else, the complete deferefernce chain is not a whole-variable
1926 * dereference. This method should only be called on the top most
1927 * ir_rvalue in a dereference chain.
1928 */
1929 return this->var;
1930 }
1931
1932 virtual void accept(ir_visitor *v)
1933 {
1934 v->visit(this);
1935 }
1936
1937 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1938
1939 /**
1940 * Object being dereferenced.
1941 */
1942 ir_variable *var;
1943 };
1944
1945
1946 class ir_dereference_array : public ir_dereference {
1947 public:
1948 ir_dereference_array(ir_rvalue *value, ir_rvalue *array_index);
1949
1950 ir_dereference_array(ir_variable *var, ir_rvalue *array_index);
1951
1952 virtual ir_dereference_array *clone(void *mem_ctx,
1953 struct hash_table *) const;
1954
1955 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1956
1957 virtual ir_dereference_array *as_dereference_array()
1958 {
1959 return this;
1960 }
1961
1962 /**
1963 * Get the variable that is ultimately referenced by an r-value
1964 */
1965 virtual ir_variable *variable_referenced() const
1966 {
1967 return this->array->variable_referenced();
1968 }
1969
1970 /**
1971 * Get the constant that is ultimately referenced by an r-value,
1972 * in a constant expression evaluation context.
1973 *
1974 * The offset is used when the reference is to a specific column of
1975 * a matrix.
1976 */
1977 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
1978
1979 virtual void accept(ir_visitor *v)
1980 {
1981 v->visit(this);
1982 }
1983
1984 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1985
1986 ir_rvalue *array;
1987 ir_rvalue *array_index;
1988
1989 private:
1990 void set_array(ir_rvalue *value);
1991 };
1992
1993
1994 class ir_dereference_record : public ir_dereference {
1995 public:
1996 ir_dereference_record(ir_rvalue *value, const char *field);
1997
1998 ir_dereference_record(ir_variable *var, const char *field);
1999
2000 virtual ir_dereference_record *clone(void *mem_ctx,
2001 struct hash_table *) const;
2002
2003 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
2004
2005 virtual ir_dereference_record *as_dereference_record()
2006 {
2007 return this;
2008 }
2009
2010 /**
2011 * Get the variable that is ultimately referenced by an r-value
2012 */
2013 virtual ir_variable *variable_referenced() const
2014 {
2015 return this->record->variable_referenced();
2016 }
2017
2018 /**
2019 * Get the constant that is ultimately referenced by an r-value,
2020 * in a constant expression evaluation context.
2021 *
2022 * The offset is used when the reference is to a specific column of
2023 * a matrix.
2024 */
2025 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
2026
2027 virtual void accept(ir_visitor *v)
2028 {
2029 v->visit(this);
2030 }
2031
2032 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2033
2034 ir_rvalue *record;
2035 const char *field;
2036 };
2037
2038
2039 /**
2040 * Data stored in an ir_constant
2041 */
2042 union ir_constant_data {
2043 unsigned u[16];
2044 int i[16];
2045 float f[16];
2046 bool b[16];
2047 };
2048
2049
2050 class ir_constant : public ir_rvalue {
2051 public:
2052 ir_constant(const struct glsl_type *type, const ir_constant_data *data);
2053 ir_constant(bool b, unsigned vector_elements=1);
2054 ir_constant(unsigned int u, unsigned vector_elements=1);
2055 ir_constant(int i, unsigned vector_elements=1);
2056 ir_constant(float f, unsigned vector_elements=1);
2057
2058 /**
2059 * Construct an ir_constant from a list of ir_constant values
2060 */
2061 ir_constant(const struct glsl_type *type, exec_list *values);
2062
2063 /**
2064 * Construct an ir_constant from a scalar component of another ir_constant
2065 *
2066 * The new \c ir_constant inherits the type of the component from the
2067 * source constant.
2068 *
2069 * \note
2070 * In the case of a matrix constant, the new constant is a scalar, \b not
2071 * a vector.
2072 */
2073 ir_constant(const ir_constant *c, unsigned i);
2074
2075 /**
2076 * Return a new ir_constant of the specified type containing all zeros.
2077 */
2078 static ir_constant *zero(void *mem_ctx, const glsl_type *type);
2079
2080 virtual ir_constant *clone(void *mem_ctx, struct hash_table *) const;
2081
2082 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
2083
2084 virtual ir_constant *as_constant()
2085 {
2086 return this;
2087 }
2088
2089 virtual void accept(ir_visitor *v)
2090 {
2091 v->visit(this);
2092 }
2093
2094 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2095
2096 /**
2097 * Get a particular component of a constant as a specific type
2098 *
2099 * This is useful, for example, to get a value from an integer constant
2100 * as a float or bool. This appears frequently when constructors are
2101 * called with all constant parameters.
2102 */
2103 /*@{*/
2104 bool get_bool_component(unsigned i) const;
2105 float get_float_component(unsigned i) const;
2106 int get_int_component(unsigned i) const;
2107 unsigned get_uint_component(unsigned i) const;
2108 /*@}*/
2109
2110 ir_constant *get_array_element(unsigned i) const;
2111
2112 ir_constant *get_record_field(const char *name);
2113
2114 /**
2115 * Copy the values on another constant at a given offset.
2116 *
2117 * The offset is ignored for array or struct copies, it's only for
2118 * scalars or vectors into vectors or matrices.
2119 *
2120 * With identical types on both sides and zero offset it's clone()
2121 * without creating a new object.
2122 */
2123
2124 void copy_offset(ir_constant *src, int offset);
2125
2126 /**
2127 * Copy the values on another constant at a given offset and
2128 * following an assign-like mask.
2129 *
2130 * The mask is ignored for scalars.
2131 *
2132 * Note that this function only handles what assign can handle,
2133 * i.e. at most a vector as source and a column of a matrix as
2134 * destination.
2135 */
2136
2137 void copy_masked_offset(ir_constant *src, int offset, unsigned int mask);
2138
2139 /**
2140 * Determine whether a constant has the same value as another constant
2141 *
2142 * \sa ir_constant::is_zero, ir_constant::is_one,
2143 * ir_constant::is_negative_one, ir_constant::is_basis
2144 */
2145 bool has_value(const ir_constant *) const;
2146
2147 virtual bool is_zero() const;
2148 virtual bool is_one() const;
2149 virtual bool is_negative_one() const;
2150 virtual bool is_basis() const;
2151
2152 /**
2153 * Value of the constant.
2154 *
2155 * The field used to back the values supplied by the constant is determined
2156 * by the type associated with the \c ir_instruction. Constants may be
2157 * scalars, vectors, or matrices.
2158 */
2159 union ir_constant_data value;
2160
2161 /* Array elements */
2162 ir_constant **array_elements;
2163
2164 /* Structure fields */
2165 exec_list components;
2166
2167 private:
2168 /**
2169 * Parameterless constructor only used by the clone method
2170 */
2171 ir_constant(void);
2172 };
2173
2174 /*@}*/
2175
2176 /**
2177 * IR instruction to emit a vertex in a geometry shader.
2178 */
2179 class ir_emit_vertex : public ir_instruction {
2180 public:
2181 ir_emit_vertex()
2182 {
2183 ir_type = ir_type_emit_vertex;
2184 }
2185
2186 virtual void accept(ir_visitor *v)
2187 {
2188 v->visit(this);
2189 }
2190
2191 virtual ir_emit_vertex *clone(void *mem_ctx, struct hash_table *) const
2192 {
2193 return new(mem_ctx) ir_emit_vertex();
2194 }
2195
2196 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2197 };
2198
2199 /**
2200 * IR instruction to complete the current primitive and start a new one in a
2201 * geometry shader.
2202 */
2203 class ir_end_primitive : public ir_instruction {
2204 public:
2205 ir_end_primitive()
2206 {
2207 ir_type = ir_type_end_primitive;
2208 }
2209
2210 virtual void accept(ir_visitor *v)
2211 {
2212 v->visit(this);
2213 }
2214
2215 virtual ir_end_primitive *clone(void *mem_ctx, struct hash_table *) const
2216 {
2217 return new(mem_ctx) ir_end_primitive();
2218 }
2219
2220 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2221 };
2222
2223 /**
2224 * Apply a visitor to each IR node in a list
2225 */
2226 void
2227 visit_exec_list(exec_list *list, ir_visitor *visitor);
2228
2229 /**
2230 * Validate invariants on each IR node in a list
2231 */
2232 void validate_ir_tree(exec_list *instructions);
2233
2234 struct _mesa_glsl_parse_state;
2235 struct gl_shader_program;
2236
2237 /**
2238 * Detect whether an unlinked shader contains static recursion
2239 *
2240 * If the list of instructions is determined to contain static recursion,
2241 * \c _mesa_glsl_error will be called to emit error messages for each function
2242 * that is in the recursion cycle.
2243 */
2244 void
2245 detect_recursion_unlinked(struct _mesa_glsl_parse_state *state,
2246 exec_list *instructions);
2247
2248 /**
2249 * Detect whether a linked shader contains static recursion
2250 *
2251 * If the list of instructions is determined to contain static recursion,
2252 * \c link_error_printf will be called to emit error messages for each function
2253 * that is in the recursion cycle. In addition,
2254 * \c gl_shader_program::LinkStatus will be set to false.
2255 */
2256 void
2257 detect_recursion_linked(struct gl_shader_program *prog,
2258 exec_list *instructions);
2259
2260 /**
2261 * Make a clone of each IR instruction in a list
2262 *
2263 * \param in List of IR instructions that are to be cloned
2264 * \param out List to hold the cloned instructions
2265 */
2266 void
2267 clone_ir_list(void *mem_ctx, exec_list *out, const exec_list *in);
2268
2269 extern void
2270 _mesa_glsl_initialize_variables(exec_list *instructions,
2271 struct _mesa_glsl_parse_state *state);
2272
2273 extern void
2274 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state *state);
2275
2276 extern void
2277 _mesa_glsl_initialize_builtin_functions();
2278
2279 extern ir_function_signature *
2280 _mesa_glsl_find_builtin_function(_mesa_glsl_parse_state *state,
2281 const char *name, exec_list *actual_parameters);
2282
2283 extern void
2284 _mesa_glsl_release_functions(void);
2285
2286 extern void
2287 _mesa_glsl_release_builtin_functions(void);
2288
2289 extern void
2290 reparent_ir(exec_list *list, void *mem_ctx);
2291
2292 struct glsl_symbol_table;
2293
2294 extern void
2295 import_prototypes(const exec_list *source, exec_list *dest,
2296 struct glsl_symbol_table *symbols, void *mem_ctx);
2297
2298 extern bool
2299 ir_has_call(ir_instruction *ir);
2300
2301 extern void
2302 do_set_program_inouts(exec_list *instructions, struct gl_program *prog,
2303 GLenum shader_type);
2304
2305 extern char *
2306 prototype_string(const glsl_type *return_type, const char *name,
2307 exec_list *parameters);
2308
2309 const char *
2310 mode_string(const ir_variable *var);
2311
2312 extern "C" {
2313 #endif /* __cplusplus */
2314
2315 extern void _mesa_print_ir(struct exec_list *instructions,
2316 struct _mesa_glsl_parse_state *state);
2317
2318 #ifdef __cplusplus
2319 } /* extern "C" */
2320 #endif
2321
2322 unsigned
2323 vertices_per_prim(GLenum prim);
2324
2325 #endif /* IR_H */