Fix 'make check' in src/mapi/glapi/tests when builddir != srcdir
[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_texture * as_texture() { return NULL; }
137 virtual class ir_constant * as_constant() { return NULL; }
138 virtual class ir_discard * as_discard() { return NULL; }
139 virtual class ir_jump * as_jump() { return NULL; }
140 /*@}*/
141
142 /**
143 * IR equality method: Return true if the referenced instruction would
144 * return the same value as this one.
145 *
146 * This intended to be used for CSE and algebraic optimizations, on rvalues
147 * in particular. No support for other instruction types (assignments,
148 * jumps, calls, etc.) is planned.
149 */
150 virtual bool equals(ir_instruction *ir);
151
152 protected:
153 ir_instruction()
154 {
155 ir_type = ir_type_unset;
156 }
157 };
158
159
160 /**
161 * The base class for all "values"/expression trees.
162 */
163 class ir_rvalue : public ir_instruction {
164 public:
165 const struct glsl_type *type;
166
167 virtual ir_rvalue *clone(void *mem_ctx, struct hash_table *) const;
168
169 virtual void accept(ir_visitor *v)
170 {
171 v->visit(this);
172 }
173
174 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
175
176 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
177
178 virtual ir_rvalue * as_rvalue()
179 {
180 return this;
181 }
182
183 ir_rvalue *as_rvalue_to_saturate();
184
185 virtual bool is_lvalue() const
186 {
187 return false;
188 }
189
190 /**
191 * Get the variable that is ultimately referenced by an r-value
192 */
193 virtual ir_variable *variable_referenced() const
194 {
195 return NULL;
196 }
197
198
199 /**
200 * If an r-value is a reference to a whole variable, get that variable
201 *
202 * \return
203 * Pointer to a variable that is completely dereferenced by the r-value. If
204 * the r-value is not a dereference or the dereference does not access the
205 * entire variable (i.e., it's just one array element, struct field), \c NULL
206 * is returned.
207 */
208 virtual ir_variable *whole_variable_referenced()
209 {
210 return NULL;
211 }
212
213 /**
214 * Determine if an r-value has the value zero
215 *
216 * The base implementation of this function always returns \c false. The
217 * \c ir_constant class over-rides this function to return \c true \b only
218 * for vector and scalar types that have all elements set to the value
219 * zero (or \c false for booleans).
220 *
221 * \sa ir_constant::has_value, ir_rvalue::is_one, ir_rvalue::is_negative_one,
222 * ir_constant::is_basis
223 */
224 virtual bool is_zero() const;
225
226 /**
227 * Determine if an r-value has the value one
228 *
229 * The base implementation of this function always returns \c false. The
230 * \c ir_constant class over-rides this function to return \c true \b only
231 * for vector and scalar types that have all elements set to the value
232 * one (or \c true for booleans).
233 *
234 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_negative_one,
235 * ir_constant::is_basis
236 */
237 virtual bool is_one() const;
238
239 /**
240 * Determine if an r-value has the value negative one
241 *
242 * The base implementation of this function always returns \c false. The
243 * \c ir_constant class over-rides this function to return \c true \b only
244 * for vector and scalar types that have all elements set to the value
245 * negative one. For boolean types, the result is always \c false.
246 *
247 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one
248 * ir_constant::is_basis
249 */
250 virtual bool is_negative_one() const;
251
252 /**
253 * Determine if an r-value is a basis vector
254 *
255 * The base implementation of this function always returns \c false. The
256 * \c ir_constant class over-rides this function to return \c true \b only
257 * for vector and scalar types that have one element set to the value one,
258 * and the other elements set to the value zero. For boolean types, the
259 * result is always \c false.
260 *
261 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one,
262 * is_constant::is_negative_one
263 */
264 virtual bool is_basis() const;
265
266
267 /**
268 * Return a generic value of error_type.
269 *
270 * Allocation will be performed with 'mem_ctx' as ralloc owner.
271 */
272 static ir_rvalue *error_value(void *mem_ctx);
273
274 protected:
275 ir_rvalue();
276 };
277
278
279 /**
280 * Variable storage classes
281 */
282 enum ir_variable_mode {
283 ir_var_auto = 0, /**< Function local variables and globals. */
284 ir_var_uniform, /**< Variable declared as a uniform. */
285 ir_var_shader_in,
286 ir_var_shader_out,
287 ir_var_function_in,
288 ir_var_function_out,
289 ir_var_function_inout,
290 ir_var_const_in, /**< "in" param that must be a constant expression */
291 ir_var_system_value, /**< Ex: front-face, instance-id, etc. */
292 ir_var_temporary, /**< Temporary variable generated during compilation. */
293 ir_var_mode_count /**< Number of variable modes */
294 };
295
296 /**
297 * Enum keeping track of how a variable was declared. For error checking of
298 * the gl_PerVertex redeclaration rules.
299 */
300 enum ir_var_declaration_type {
301 /**
302 * Normal declaration (for most variables, this means an explicit
303 * declaration. Exception: temporaries are always implicitly declared, but
304 * they still use ir_var_declared_normally).
305 *
306 * Note: an ir_variable that represents a named interface block uses
307 * ir_var_declared_normally.
308 */
309 ir_var_declared_normally = 0,
310
311 /**
312 * Variable was explicitly declared (or re-declared) in an unnamed
313 * interface block.
314 */
315 ir_var_declared_in_block,
316
317 /**
318 * Variable is an implicitly declared built-in that has not been explicitly
319 * re-declared by the shader.
320 */
321 ir_var_declared_implicitly,
322 };
323
324 /**
325 * \brief Layout qualifiers for gl_FragDepth.
326 *
327 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
328 * with a layout qualifier.
329 */
330 enum ir_depth_layout {
331 ir_depth_layout_none, /**< No depth layout is specified. */
332 ir_depth_layout_any,
333 ir_depth_layout_greater,
334 ir_depth_layout_less,
335 ir_depth_layout_unchanged
336 };
337
338 /**
339 * \brief Convert depth layout qualifier to string.
340 */
341 const char*
342 depth_layout_string(ir_depth_layout layout);
343
344 /**
345 * Description of built-in state associated with a uniform
346 *
347 * \sa ir_variable::state_slots
348 */
349 struct ir_state_slot {
350 int tokens[5];
351 int swizzle;
352 };
353
354
355 /**
356 * Get the string value for an interpolation qualifier
357 *
358 * \return The string that would be used in a shader to specify \c
359 * mode will be returned.
360 *
361 * This function is used to generate error messages of the form "shader
362 * uses %s interpolation qualifier", so in the case where there is no
363 * interpolation qualifier, it returns "no".
364 *
365 * This function should only be used on a shader input or output variable.
366 */
367 const char *interpolation_string(unsigned interpolation);
368
369
370 class ir_variable : public ir_instruction {
371 public:
372 ir_variable(const struct glsl_type *, const char *, ir_variable_mode);
373
374 virtual ir_variable *clone(void *mem_ctx, struct hash_table *ht) const;
375
376 virtual ir_variable *as_variable()
377 {
378 return this;
379 }
380
381 virtual void accept(ir_visitor *v)
382 {
383 v->visit(this);
384 }
385
386 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
387
388
389 /**
390 * Determine how this variable should be interpolated based on its
391 * interpolation qualifier (if present), whether it is gl_Color or
392 * gl_SecondaryColor, and whether flatshading is enabled in the current GL
393 * state.
394 *
395 * The return value will always be either INTERP_QUALIFIER_SMOOTH,
396 * INTERP_QUALIFIER_NOPERSPECTIVE, or INTERP_QUALIFIER_FLAT.
397 */
398 glsl_interp_qualifier determine_interpolation_mode(bool flat_shade);
399
400 /**
401 * Determine whether or not a variable is part of a uniform block.
402 */
403 inline bool is_in_uniform_block() const
404 {
405 return this->mode == ir_var_uniform && this->interface_type != NULL;
406 }
407
408 /**
409 * Determine whether or not a variable is the declaration of an interface
410 * block
411 *
412 * For the first declaration below, there will be an \c ir_variable named
413 * "instance" whose type and whose instance_type will be the same
414 * \cglsl_type. For the second declaration, there will be an \c ir_variable
415 * named "f" whose type is float and whose instance_type is B2.
416 *
417 * "instance" is an interface instance variable, but "f" is not.
418 *
419 * uniform B1 {
420 * float f;
421 * } instance;
422 *
423 * uniform B2 {
424 * float f;
425 * };
426 */
427 inline bool is_interface_instance() const
428 {
429 const glsl_type *const t = this->type;
430
431 return (t == this->interface_type)
432 || (t->is_array() && t->fields.array == this->interface_type);
433 }
434
435 /**
436 * Set this->interface_type on a newly created variable.
437 */
438 void init_interface_type(const struct glsl_type *type)
439 {
440 assert(this->interface_type == NULL);
441 this->interface_type = type;
442 if (this->is_interface_instance()) {
443 this->max_ifc_array_access =
444 rzalloc_array(this, unsigned, type->length);
445 }
446 }
447
448 /**
449 * Change this->interface_type on a variable that previously had a
450 * different, but compatible, interface_type. This is used during linking
451 * to set the size of arrays in interface blocks.
452 */
453 void change_interface_type(const struct glsl_type *type)
454 {
455 if (this->max_ifc_array_access != NULL) {
456 /* max_ifc_array_access has already been allocated, so make sure the
457 * new interface has the same number of fields as the old one.
458 */
459 assert(this->interface_type->length == type->length);
460 }
461 this->interface_type = type;
462 }
463
464 /**
465 * Change this->interface_type on a variable that previously had a
466 * different, and incompatible, interface_type. This is used during
467 * compilation to handle redeclaration of the built-in gl_PerVertex
468 * interface block.
469 */
470 void reinit_interface_type(const struct glsl_type *type)
471 {
472 if (this->max_ifc_array_access != NULL) {
473 #ifndef _NDEBUG
474 /* Redeclaring gl_PerVertex is only allowed if none of the built-ins
475 * it defines have been accessed yet; so it's safe to throw away the
476 * old max_ifc_array_access pointer, since all of its values are
477 * zero.
478 */
479 for (unsigned i = 0; i < this->interface_type->length; i++)
480 assert(this->max_ifc_array_access[i] == 0);
481 #endif
482 ralloc_free(this->max_ifc_array_access);
483 this->max_ifc_array_access = NULL;
484 }
485 this->interface_type = NULL;
486 init_interface_type(type);
487 }
488
489 const glsl_type *get_interface_type() const
490 {
491 return this->interface_type;
492 }
493
494 /**
495 * Declared type of the variable
496 */
497 const struct glsl_type *type;
498
499 /**
500 * Declared name of the variable
501 */
502 const char *name;
503
504 /**
505 * Highest element accessed with a constant expression array index
506 *
507 * Not used for non-array variables.
508 */
509 unsigned max_array_access;
510
511 /**
512 * For variables which satisfy the is_interface_instance() predicate, this
513 * points to an array of integers such that if the ith member of the
514 * interface block is an array, max_ifc_array_access[i] is the maximum
515 * array element of that member that has been accessed. If the ith member
516 * of the interface block is not an array, max_ifc_array_access[i] is
517 * unused.
518 *
519 * For variables whose type is not an interface block, this pointer is
520 * NULL.
521 */
522 unsigned *max_ifc_array_access;
523
524 /**
525 * Is the variable read-only?
526 *
527 * This is set for variables declared as \c const, shader inputs,
528 * and uniforms.
529 */
530 unsigned read_only:1;
531 unsigned centroid:1;
532 unsigned invariant:1;
533
534 /**
535 * Has this variable been used for reading or writing?
536 *
537 * Several GLSL semantic checks require knowledge of whether or not a
538 * variable has been used. For example, it is an error to redeclare a
539 * variable as invariant after it has been used.
540 *
541 * This is only maintained in the ast_to_hir.cpp path, not in
542 * Mesa's fixed function or ARB program paths.
543 */
544 unsigned used:1;
545
546 /**
547 * Has this variable been statically assigned?
548 *
549 * This answers whether the variable was assigned in any path of
550 * the shader during ast_to_hir. This doesn't answer whether it is
551 * still written after dead code removal, nor is it maintained in
552 * non-ast_to_hir.cpp (GLSL parsing) paths.
553 */
554 unsigned assigned:1;
555
556 /**
557 * Enum indicating how the variable was declared. See
558 * ir_var_declaration_type.
559 *
560 * This is used to detect certain kinds of illegal variable redeclarations.
561 */
562 unsigned how_declared:2;
563
564 /**
565 * Storage class of the variable.
566 *
567 * \sa ir_variable_mode
568 */
569 unsigned mode:4;
570
571 /**
572 * Interpolation mode for shader inputs / outputs
573 *
574 * \sa ir_variable_interpolation
575 */
576 unsigned interpolation:2;
577
578 /**
579 * \name ARB_fragment_coord_conventions
580 * @{
581 */
582 unsigned origin_upper_left:1;
583 unsigned pixel_center_integer:1;
584 /*@}*/
585
586 /**
587 * Was the location explicitly set in the shader?
588 *
589 * If the location is explicitly set in the shader, it \b cannot be changed
590 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
591 * no effect).
592 */
593 unsigned explicit_location:1;
594 unsigned explicit_index:1;
595
596 /**
597 * Was an initial binding explicitly set in the shader?
598 *
599 * If so, constant_value contains an integer ir_constant representing the
600 * initial binding point.
601 */
602 unsigned explicit_binding:1;
603
604 /**
605 * Does this variable have an initializer?
606 *
607 * This is used by the linker to cross-validiate initializers of global
608 * variables.
609 */
610 unsigned has_initializer:1;
611
612 /**
613 * Is this variable a generic output or input that has not yet been matched
614 * up to a variable in another stage of the pipeline?
615 *
616 * This is used by the linker as scratch storage while assigning locations
617 * to generic inputs and outputs.
618 */
619 unsigned is_unmatched_generic_inout:1;
620
621 /**
622 * If non-zero, then this variable may be packed along with other variables
623 * into a single varying slot, so this offset should be applied when
624 * accessing components. For example, an offset of 1 means that the x
625 * component of this variable is actually stored in component y of the
626 * location specified by \c location.
627 */
628 unsigned location_frac:2;
629
630 /**
631 * Non-zero if this variable was created by lowering a named interface
632 * block which was not an array.
633 *
634 * Note that this variable and \c from_named_ifc_block_array will never
635 * both be non-zero.
636 */
637 unsigned from_named_ifc_block_nonarray:1;
638
639 /**
640 * Non-zero if this variable was created by lowering a named interface
641 * block which was an array.
642 *
643 * Note that this variable and \c from_named_ifc_block_nonarray will never
644 * both be non-zero.
645 */
646 unsigned from_named_ifc_block_array:1;
647
648 /**
649 * \brief Layout qualifier for gl_FragDepth.
650 *
651 * This is not equal to \c ir_depth_layout_none if and only if this
652 * variable is \c gl_FragDepth and a layout qualifier is specified.
653 */
654 ir_depth_layout depth_layout;
655
656 /**
657 * Storage location of the base of this variable
658 *
659 * The precise meaning of this field depends on the nature of the variable.
660 *
661 * - Vertex shader input: one of the values from \c gl_vert_attrib.
662 * - Vertex shader output: one of the values from \c gl_varying_slot.
663 * - Geometry shader input: one of the values from \c gl_varying_slot.
664 * - Geometry shader output: one of the values from \c gl_varying_slot.
665 * - Fragment shader input: one of the values from \c gl_varying_slot.
666 * - Fragment shader output: one of the values from \c gl_frag_result.
667 * - Uniforms: Per-stage uniform slot number for default uniform block.
668 * - Uniforms: Index within the uniform block definition for UBO members.
669 * - Other: This field is not currently used.
670 *
671 * If the variable is a uniform, shader input, or shader output, and the
672 * slot has not been assigned, the value will be -1.
673 */
674 int location;
675
676 /**
677 * output index for dual source blending.
678 */
679 int index;
680
681 /**
682 * Initial binding point for a sampler or UBO.
683 *
684 * For array types, this represents the binding point for the first element.
685 */
686 int binding;
687
688 /**
689 * Location an atomic counter is stored at.
690 */
691 struct {
692 unsigned buffer_index;
693 unsigned offset;
694 } atomic;
695
696 /**
697 * Built-in state that backs this uniform
698 *
699 * Once set at variable creation, \c state_slots must remain invariant.
700 * This is because, ideally, this array would be shared by all clones of
701 * this variable in the IR tree. In other words, we'd really like for it
702 * to be a fly-weight.
703 *
704 * If the variable is not a uniform, \c num_state_slots will be zero and
705 * \c state_slots will be \c NULL.
706 */
707 /*@{*/
708 unsigned num_state_slots; /**< Number of state slots used */
709 ir_state_slot *state_slots; /**< State descriptors. */
710 /*@}*/
711
712 /**
713 * Emit a warning if this variable is accessed.
714 */
715 const char *warn_extension;
716
717 /**
718 * Value assigned in the initializer of a variable declared "const"
719 */
720 ir_constant *constant_value;
721
722 /**
723 * Constant expression assigned in the initializer of the variable
724 *
725 * \warning
726 * This field and \c ::constant_value are distinct. Even if the two fields
727 * refer to constants with the same value, they must point to separate
728 * objects.
729 */
730 ir_constant *constant_initializer;
731
732 private:
733 /**
734 * For variables that are in an interface block or are an instance of an
735 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
736 *
737 * \sa ir_variable::location
738 */
739 const glsl_type *interface_type;
740 };
741
742 /**
743 * A function that returns whether a built-in function is available in the
744 * current shading language (based on version, ES or desktop, and extensions).
745 */
746 typedef bool (*builtin_available_predicate)(const _mesa_glsl_parse_state *);
747
748 /*@{*/
749 /**
750 * The representation of a function instance; may be the full definition or
751 * simply a prototype.
752 */
753 class ir_function_signature : public ir_instruction {
754 /* An ir_function_signature will be part of the list of signatures in
755 * an ir_function.
756 */
757 public:
758 ir_function_signature(const glsl_type *return_type,
759 builtin_available_predicate builtin_avail = NULL);
760
761 virtual ir_function_signature *clone(void *mem_ctx,
762 struct hash_table *ht) const;
763 ir_function_signature *clone_prototype(void *mem_ctx,
764 struct hash_table *ht) const;
765
766 virtual void accept(ir_visitor *v)
767 {
768 v->visit(this);
769 }
770
771 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
772
773 /**
774 * Attempt to evaluate this function as a constant expression,
775 * given a list of the actual parameters and the variable context.
776 * Returns NULL for non-built-ins.
777 */
778 ir_constant *constant_expression_value(exec_list *actual_parameters, struct hash_table *variable_context);
779
780 /**
781 * Get the name of the function for which this is a signature
782 */
783 const char *function_name() const;
784
785 /**
786 * Get a handle to the function for which this is a signature
787 *
788 * There is no setter function, this function returns a \c const pointer,
789 * and \c ir_function_signature::_function is private for a reason. The
790 * only way to make a connection between a function and function signature
791 * is via \c ir_function::add_signature. This helps ensure that certain
792 * invariants (i.e., a function signature is in the list of signatures for
793 * its \c _function) are met.
794 *
795 * \sa ir_function::add_signature
796 */
797 inline const class ir_function *function() const
798 {
799 return this->_function;
800 }
801
802 /**
803 * Check whether the qualifiers match between this signature's parameters
804 * and the supplied parameter list. If not, returns the name of the first
805 * parameter with mismatched qualifiers (for use in error messages).
806 */
807 const char *qualifiers_match(exec_list *params);
808
809 /**
810 * Replace the current parameter list with the given one. This is useful
811 * if the current information came from a prototype, and either has invalid
812 * or missing parameter names.
813 */
814 void replace_parameters(exec_list *new_params);
815
816 /**
817 * Function return type.
818 *
819 * \note This discards the optional precision qualifier.
820 */
821 const struct glsl_type *return_type;
822
823 /**
824 * List of ir_variable of function parameters.
825 *
826 * This represents the storage. The paramaters passed in a particular
827 * call will be in ir_call::actual_paramaters.
828 */
829 struct exec_list parameters;
830
831 /** Whether or not this function has a body (which may be empty). */
832 unsigned is_defined:1;
833
834 /** Whether or not this function signature is a built-in. */
835 bool is_builtin() const;
836
837 /**
838 * Whether or not this function is an intrinsic to be implemented
839 * by the driver.
840 */
841 bool is_intrinsic;
842
843 /** Whether or not a built-in is available for this shader. */
844 bool is_builtin_available(const _mesa_glsl_parse_state *state) const;
845
846 /** Body of instructions in the function. */
847 struct exec_list body;
848
849 private:
850 /**
851 * A function pointer to a predicate that answers whether a built-in
852 * function is available in the current shader. NULL if not a built-in.
853 */
854 builtin_available_predicate builtin_avail;
855
856 /** Function of which this signature is one overload. */
857 class ir_function *_function;
858
859 /** Function signature of which this one is a prototype clone */
860 const ir_function_signature *origin;
861
862 friend class ir_function;
863
864 /**
865 * Helper function to run a list of instructions for constant
866 * expression evaluation.
867 *
868 * The hash table represents the values of the visible variables.
869 * There are no scoping issues because the table is indexed on
870 * ir_variable pointers, not variable names.
871 *
872 * Returns false if the expression is not constant, true otherwise,
873 * and the value in *result if result is non-NULL.
874 */
875 bool constant_expression_evaluate_expression_list(const struct exec_list &body,
876 struct hash_table *variable_context,
877 ir_constant **result);
878 };
879
880
881 /**
882 * Header for tracking multiple overloaded functions with the same name.
883 * Contains a list of ir_function_signatures representing each of the
884 * actual functions.
885 */
886 class ir_function : public ir_instruction {
887 public:
888 ir_function(const char *name);
889
890 virtual ir_function *clone(void *mem_ctx, struct hash_table *ht) const;
891
892 virtual ir_function *as_function()
893 {
894 return this;
895 }
896
897 virtual void accept(ir_visitor *v)
898 {
899 v->visit(this);
900 }
901
902 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
903
904 void add_signature(ir_function_signature *sig)
905 {
906 sig->_function = this;
907 this->signatures.push_tail(sig);
908 }
909
910 /**
911 * Get an iterator for the set of function signatures
912 */
913 exec_list_iterator iterator()
914 {
915 return signatures.iterator();
916 }
917
918 /**
919 * Find a signature that matches a set of actual parameters, taking implicit
920 * conversions into account. Also flags whether the match was exact.
921 */
922 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
923 const exec_list *actual_param,
924 bool *match_is_exact);
925
926 /**
927 * Find a signature that matches a set of actual parameters, taking implicit
928 * conversions into account.
929 */
930 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
931 const exec_list *actual_param);
932
933 /**
934 * Find a signature that exactly matches a set of actual parameters without
935 * any implicit type conversions.
936 */
937 ir_function_signature *exact_matching_signature(_mesa_glsl_parse_state *state,
938 const exec_list *actual_ps);
939
940 /**
941 * Name of the function.
942 */
943 const char *name;
944
945 /** Whether or not this function has a signature that isn't a built-in. */
946 bool has_user_signature();
947
948 /**
949 * List of ir_function_signature for each overloaded function with this name.
950 */
951 struct exec_list signatures;
952 };
953
954 inline const char *ir_function_signature::function_name() const
955 {
956 return this->_function->name;
957 }
958 /*@}*/
959
960
961 /**
962 * IR instruction representing high-level if-statements
963 */
964 class ir_if : public ir_instruction {
965 public:
966 ir_if(ir_rvalue *condition)
967 : condition(condition)
968 {
969 ir_type = ir_type_if;
970 }
971
972 virtual ir_if *clone(void *mem_ctx, struct hash_table *ht) const;
973
974 virtual ir_if *as_if()
975 {
976 return this;
977 }
978
979 virtual void accept(ir_visitor *v)
980 {
981 v->visit(this);
982 }
983
984 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
985
986 ir_rvalue *condition;
987 /** List of ir_instruction for the body of the then branch */
988 exec_list then_instructions;
989 /** List of ir_instruction for the body of the else branch */
990 exec_list else_instructions;
991 };
992
993
994 /**
995 * IR instruction representing a high-level loop structure.
996 */
997 class ir_loop : public ir_instruction {
998 public:
999 ir_loop();
1000
1001 virtual ir_loop *clone(void *mem_ctx, struct hash_table *ht) const;
1002
1003 virtual void accept(ir_visitor *v)
1004 {
1005 v->visit(this);
1006 }
1007
1008 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1009
1010 virtual ir_loop *as_loop()
1011 {
1012 return this;
1013 }
1014
1015 /**
1016 * Get an iterator for the instructions of the loop body
1017 */
1018 exec_list_iterator iterator()
1019 {
1020 return body_instructions.iterator();
1021 }
1022
1023 /** List of ir_instruction that make up the body of the loop. */
1024 exec_list body_instructions;
1025
1026 /**
1027 * \name Loop counter and controls
1028 *
1029 * Represents a loop like a FORTRAN \c do-loop.
1030 *
1031 * \note
1032 * If \c from and \c to are the same value, the loop will execute once.
1033 */
1034 /*@{*/
1035
1036 /**
1037 * Value which should be assigned to \c counter before the first iteration
1038 * of the loop. Must be non-null whenever \c counter is non-null, and vice
1039 * versa.
1040 */
1041 ir_rvalue *from;
1042
1043 /**
1044 * Value which \c counter should be compared to in order to determine
1045 * whether to exit the loop. Must be non-null whenever \c counter is
1046 * non-null, and vice versa.
1047 */
1048 ir_rvalue *to;
1049
1050 /**
1051 * Value which should be added to \c counter at the end of each loop
1052 * iteration. Must be non-null whenever \c counter is non-null, and vice
1053 * versa.
1054 */
1055 ir_rvalue *increment;
1056
1057 /**
1058 * Variable which counts loop iterations. This is a brand new ir_variable
1059 * declaration (not a reference to a previously declared ir_variable, as in
1060 * ir_dereference_variable).
1061 */
1062 ir_variable *counter;
1063
1064 /**
1065 * Comparison operation in the loop terminator.
1066 *
1067 * If any of the loop control fields are non-\c NULL, this field must be
1068 * one of \c ir_binop_less, \c ir_binop_greater, \c ir_binop_lequal,
1069 * \c ir_binop_gequal, \c ir_binop_equal, or \c ir_binop_nequal.
1070 *
1071 * Ignored if \c counter is NULL.
1072 */
1073 int cmp;
1074 /*@}*/
1075 };
1076
1077
1078 class ir_assignment : public ir_instruction {
1079 public:
1080 ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs, ir_rvalue *condition = NULL);
1081
1082 /**
1083 * Construct an assignment with an explicit write mask
1084 *
1085 * \note
1086 * Since a write mask is supplied, the LHS must already be a bare
1087 * \c ir_dereference. The cannot be any swizzles in the LHS.
1088 */
1089 ir_assignment(ir_dereference *lhs, ir_rvalue *rhs, ir_rvalue *condition,
1090 unsigned write_mask);
1091
1092 virtual ir_assignment *clone(void *mem_ctx, struct hash_table *ht) const;
1093
1094 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1095
1096 virtual void accept(ir_visitor *v)
1097 {
1098 v->visit(this);
1099 }
1100
1101 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1102
1103 virtual ir_assignment * as_assignment()
1104 {
1105 return this;
1106 }
1107
1108 /**
1109 * Get a whole variable written by an assignment
1110 *
1111 * If the LHS of the assignment writes a whole variable, the variable is
1112 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1113 * assignment are:
1114 *
1115 * - Assigning to a scalar
1116 * - Assigning to all components of a vector
1117 * - Whole array (or matrix) assignment
1118 * - Whole structure assignment
1119 */
1120 ir_variable *whole_variable_written();
1121
1122 /**
1123 * Set the LHS of an assignment
1124 */
1125 void set_lhs(ir_rvalue *lhs);
1126
1127 /**
1128 * Left-hand side of the assignment.
1129 *
1130 * This should be treated as read only. If you need to set the LHS of an
1131 * assignment, use \c ir_assignment::set_lhs.
1132 */
1133 ir_dereference *lhs;
1134
1135 /**
1136 * Value being assigned
1137 */
1138 ir_rvalue *rhs;
1139
1140 /**
1141 * Optional condition for the assignment.
1142 */
1143 ir_rvalue *condition;
1144
1145
1146 /**
1147 * Component mask written
1148 *
1149 * For non-vector types in the LHS, this field will be zero. For vector
1150 * types, a bit will be set for each component that is written. Note that
1151 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1152 *
1153 * A partially-set write mask means that each enabled channel gets
1154 * the value from a consecutive channel of the rhs. For example,
1155 * to write just .xyw of gl_FrontColor with color:
1156 *
1157 * (assign (constant bool (1)) (xyw)
1158 * (var_ref gl_FragColor)
1159 * (swiz xyw (var_ref color)))
1160 */
1161 unsigned write_mask:4;
1162 };
1163
1164 /* Update ir_expression::get_num_operands() and operator_strs when
1165 * updating this list.
1166 */
1167 enum ir_expression_operation {
1168 ir_unop_bit_not,
1169 ir_unop_logic_not,
1170 ir_unop_neg,
1171 ir_unop_abs,
1172 ir_unop_sign,
1173 ir_unop_rcp,
1174 ir_unop_rsq,
1175 ir_unop_sqrt,
1176 ir_unop_exp, /**< Log base e on gentype */
1177 ir_unop_log, /**< Natural log on gentype */
1178 ir_unop_exp2,
1179 ir_unop_log2,
1180 ir_unop_f2i, /**< Float-to-integer conversion. */
1181 ir_unop_f2u, /**< Float-to-unsigned conversion. */
1182 ir_unop_i2f, /**< Integer-to-float conversion. */
1183 ir_unop_f2b, /**< Float-to-boolean conversion */
1184 ir_unop_b2f, /**< Boolean-to-float conversion */
1185 ir_unop_i2b, /**< int-to-boolean conversion */
1186 ir_unop_b2i, /**< Boolean-to-int conversion */
1187 ir_unop_u2f, /**< Unsigned-to-float conversion. */
1188 ir_unop_i2u, /**< Integer-to-unsigned conversion. */
1189 ir_unop_u2i, /**< Unsigned-to-integer conversion. */
1190 ir_unop_bitcast_i2f, /**< Bit-identical int-to-float "conversion" */
1191 ir_unop_bitcast_f2i, /**< Bit-identical float-to-int "conversion" */
1192 ir_unop_bitcast_u2f, /**< Bit-identical uint-to-float "conversion" */
1193 ir_unop_bitcast_f2u, /**< Bit-identical float-to-uint "conversion" */
1194 ir_unop_any,
1195
1196 /**
1197 * \name Unary floating-point rounding operations.
1198 */
1199 /*@{*/
1200 ir_unop_trunc,
1201 ir_unop_ceil,
1202 ir_unop_floor,
1203 ir_unop_fract,
1204 ir_unop_round_even,
1205 /*@}*/
1206
1207 /**
1208 * \name Trigonometric operations.
1209 */
1210 /*@{*/
1211 ir_unop_sin,
1212 ir_unop_cos,
1213 ir_unop_sin_reduced, /**< Reduced range sin. [-pi, pi] */
1214 ir_unop_cos_reduced, /**< Reduced range cos. [-pi, pi] */
1215 /*@}*/
1216
1217 /**
1218 * \name Partial derivatives.
1219 */
1220 /*@{*/
1221 ir_unop_dFdx,
1222 ir_unop_dFdy,
1223 /*@}*/
1224
1225 /**
1226 * \name Floating point pack and unpack operations.
1227 */
1228 /*@{*/
1229 ir_unop_pack_snorm_2x16,
1230 ir_unop_pack_snorm_4x8,
1231 ir_unop_pack_unorm_2x16,
1232 ir_unop_pack_unorm_4x8,
1233 ir_unop_pack_half_2x16,
1234 ir_unop_unpack_snorm_2x16,
1235 ir_unop_unpack_snorm_4x8,
1236 ir_unop_unpack_unorm_2x16,
1237 ir_unop_unpack_unorm_4x8,
1238 ir_unop_unpack_half_2x16,
1239 /*@}*/
1240
1241 /**
1242 * \name Lowered floating point unpacking operations.
1243 *
1244 * \see lower_packing_builtins_visitor::split_unpack_half_2x16
1245 */
1246 /*@{*/
1247 ir_unop_unpack_half_2x16_split_x,
1248 ir_unop_unpack_half_2x16_split_y,
1249 /*@}*/
1250
1251 /**
1252 * \name Bit operations, part of ARB_gpu_shader5.
1253 */
1254 /*@{*/
1255 ir_unop_bitfield_reverse,
1256 ir_unop_bit_count,
1257 ir_unop_find_msb,
1258 ir_unop_find_lsb,
1259 /*@}*/
1260
1261 ir_unop_noise,
1262
1263 /**
1264 * A sentinel marking the last of the unary operations.
1265 */
1266 ir_last_unop = ir_unop_noise,
1267
1268 ir_binop_add,
1269 ir_binop_sub,
1270 ir_binop_mul, /**< Floating-point or low 32-bit integer multiply. */
1271 ir_binop_imul_high, /**< Calculates the high 32-bits of a 64-bit multiply. */
1272 ir_binop_div,
1273
1274 /**
1275 * Returns the carry resulting from the addition of the two arguments.
1276 */
1277 /*@{*/
1278 ir_binop_carry,
1279 /*@}*/
1280
1281 /**
1282 * Returns the borrow resulting from the subtraction of the second argument
1283 * from the first argument.
1284 */
1285 /*@{*/
1286 ir_binop_borrow,
1287 /*@}*/
1288
1289 /**
1290 * Takes one of two combinations of arguments:
1291 *
1292 * - mod(vecN, vecN)
1293 * - mod(vecN, float)
1294 *
1295 * Does not take integer types.
1296 */
1297 ir_binop_mod,
1298
1299 /**
1300 * \name Binary comparison operators which return a boolean vector.
1301 * The type of both operands must be equal.
1302 */
1303 /*@{*/
1304 ir_binop_less,
1305 ir_binop_greater,
1306 ir_binop_lequal,
1307 ir_binop_gequal,
1308 ir_binop_equal,
1309 ir_binop_nequal,
1310 /**
1311 * Returns single boolean for whether all components of operands[0]
1312 * equal the components of operands[1].
1313 */
1314 ir_binop_all_equal,
1315 /**
1316 * Returns single boolean for whether any component of operands[0]
1317 * is not equal to the corresponding component of operands[1].
1318 */
1319 ir_binop_any_nequal,
1320 /*@}*/
1321
1322 /**
1323 * \name Bit-wise binary operations.
1324 */
1325 /*@{*/
1326 ir_binop_lshift,
1327 ir_binop_rshift,
1328 ir_binop_bit_and,
1329 ir_binop_bit_xor,
1330 ir_binop_bit_or,
1331 /*@}*/
1332
1333 ir_binop_logic_and,
1334 ir_binop_logic_xor,
1335 ir_binop_logic_or,
1336
1337 ir_binop_dot,
1338 ir_binop_min,
1339 ir_binop_max,
1340
1341 ir_binop_pow,
1342
1343 /**
1344 * \name Lowered floating point packing operations.
1345 *
1346 * \see lower_packing_builtins_visitor::split_pack_half_2x16
1347 */
1348 /*@{*/
1349 ir_binop_pack_half_2x16_split,
1350 /*@}*/
1351
1352 /**
1353 * \name First half of a lowered bitfieldInsert() operation.
1354 *
1355 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1356 */
1357 /*@{*/
1358 ir_binop_bfm,
1359 /*@}*/
1360
1361 /**
1362 * Load a value the size of a given GLSL type from a uniform block.
1363 *
1364 * operand0 is the ir_constant uniform block index in the linked shader.
1365 * operand1 is a byte offset within the uniform block.
1366 */
1367 ir_binop_ubo_load,
1368
1369 /**
1370 * \name Multiplies a number by two to a power, part of ARB_gpu_shader5.
1371 */
1372 /*@{*/
1373 ir_binop_ldexp,
1374 /*@}*/
1375
1376 /**
1377 * Extract a scalar from a vector
1378 *
1379 * operand0 is the vector
1380 * operand1 is the index of the field to read from operand0
1381 */
1382 ir_binop_vector_extract,
1383
1384 /**
1385 * A sentinel marking the last of the binary operations.
1386 */
1387 ir_last_binop = ir_binop_vector_extract,
1388
1389 /**
1390 * \name Fused floating-point multiply-add, part of ARB_gpu_shader5.
1391 */
1392 /*@{*/
1393 ir_triop_fma,
1394 /*@}*/
1395
1396 ir_triop_lrp,
1397
1398 /**
1399 * \name Conditional Select
1400 *
1401 * A vector conditional select instruction (like ?:, but operating per-
1402 * component on vectors).
1403 *
1404 * \see lower_instructions_visitor::ldexp_to_arith
1405 */
1406 /*@{*/
1407 ir_triop_csel,
1408 /*@}*/
1409
1410 /**
1411 * \name Second half of a lowered bitfieldInsert() operation.
1412 *
1413 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1414 */
1415 /*@{*/
1416 ir_triop_bfi,
1417 /*@}*/
1418
1419 ir_triop_bitfield_extract,
1420
1421 /**
1422 * Generate a value with one field of a vector changed
1423 *
1424 * operand0 is the vector
1425 * operand1 is the value to write into the vector result
1426 * operand2 is the index in operand0 to be modified
1427 */
1428 ir_triop_vector_insert,
1429
1430 /**
1431 * A sentinel marking the last of the ternary operations.
1432 */
1433 ir_last_triop = ir_triop_vector_insert,
1434
1435 ir_quadop_bitfield_insert,
1436
1437 ir_quadop_vector,
1438
1439 /**
1440 * A sentinel marking the last of the ternary operations.
1441 */
1442 ir_last_quadop = ir_quadop_vector,
1443
1444 /**
1445 * A sentinel marking the last of all operations.
1446 */
1447 ir_last_opcode = ir_quadop_vector
1448 };
1449
1450 class ir_expression : public ir_rvalue {
1451 public:
1452 ir_expression(int op, const struct glsl_type *type,
1453 ir_rvalue *op0, ir_rvalue *op1 = NULL,
1454 ir_rvalue *op2 = NULL, ir_rvalue *op3 = NULL);
1455
1456 /**
1457 * Constructor for unary operation expressions
1458 */
1459 ir_expression(int op, ir_rvalue *);
1460
1461 /**
1462 * Constructor for binary operation expressions
1463 */
1464 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1);
1465
1466 /**
1467 * Constructor for ternary operation expressions
1468 */
1469 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1, ir_rvalue *op2);
1470
1471 virtual ir_expression *as_expression()
1472 {
1473 return this;
1474 }
1475
1476 virtual bool equals(ir_instruction *ir);
1477
1478 virtual ir_expression *clone(void *mem_ctx, struct hash_table *ht) const;
1479
1480 /**
1481 * Attempt to constant-fold the expression
1482 *
1483 * The "variable_context" hash table links ir_variable * to ir_constant *
1484 * that represent the variables' values. \c NULL represents an empty
1485 * context.
1486 *
1487 * If the expression cannot be constant folded, this method will return
1488 * \c NULL.
1489 */
1490 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1491
1492 /**
1493 * Determine the number of operands used by an expression
1494 */
1495 static unsigned int get_num_operands(ir_expression_operation);
1496
1497 /**
1498 * Determine the number of operands used by an expression
1499 */
1500 unsigned int get_num_operands() const
1501 {
1502 return (this->operation == ir_quadop_vector)
1503 ? this->type->vector_elements : get_num_operands(operation);
1504 }
1505
1506 /**
1507 * Return a string representing this expression's operator.
1508 */
1509 const char *operator_string();
1510
1511 /**
1512 * Return a string representing this expression's operator.
1513 */
1514 static const char *operator_string(ir_expression_operation);
1515
1516
1517 /**
1518 * Do a reverse-lookup to translate the given string into an operator.
1519 */
1520 static ir_expression_operation get_operator(const char *);
1521
1522 virtual void accept(ir_visitor *v)
1523 {
1524 v->visit(this);
1525 }
1526
1527 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1528
1529 ir_expression_operation operation;
1530 ir_rvalue *operands[4];
1531 };
1532
1533
1534 /**
1535 * HIR instruction representing a high-level function call, containing a list
1536 * of parameters and returning a value in the supplied temporary.
1537 */
1538 class ir_call : public ir_instruction {
1539 public:
1540 ir_call(ir_function_signature *callee,
1541 ir_dereference_variable *return_deref,
1542 exec_list *actual_parameters)
1543 : return_deref(return_deref), callee(callee)
1544 {
1545 ir_type = ir_type_call;
1546 assert(callee->return_type != NULL);
1547 actual_parameters->move_nodes_to(& this->actual_parameters);
1548 this->use_builtin = callee->is_builtin();
1549 }
1550
1551 virtual ir_call *clone(void *mem_ctx, struct hash_table *ht) const;
1552
1553 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1554
1555 virtual ir_call *as_call()
1556 {
1557 return this;
1558 }
1559
1560 virtual void accept(ir_visitor *v)
1561 {
1562 v->visit(this);
1563 }
1564
1565 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1566
1567 /**
1568 * Get an iterator for the set of acutal parameters
1569 */
1570 exec_list_iterator iterator()
1571 {
1572 return actual_parameters.iterator();
1573 }
1574
1575 /**
1576 * Get the name of the function being called.
1577 */
1578 const char *callee_name() const
1579 {
1580 return callee->function_name();
1581 }
1582
1583 /**
1584 * Generates an inline version of the function before @ir,
1585 * storing the return value in return_deref.
1586 */
1587 void generate_inline(ir_instruction *ir);
1588
1589 /**
1590 * Storage for the function's return value.
1591 * This must be NULL if the return type is void.
1592 */
1593 ir_dereference_variable *return_deref;
1594
1595 /**
1596 * The specific function signature being called.
1597 */
1598 ir_function_signature *callee;
1599
1600 /* List of ir_rvalue of paramaters passed in this call. */
1601 exec_list actual_parameters;
1602
1603 /** Should this call only bind to a built-in function? */
1604 bool use_builtin;
1605 };
1606
1607
1608 /**
1609 * \name Jump-like IR instructions.
1610 *
1611 * These include \c break, \c continue, \c return, and \c discard.
1612 */
1613 /*@{*/
1614 class ir_jump : public ir_instruction {
1615 protected:
1616 ir_jump()
1617 {
1618 ir_type = ir_type_unset;
1619 }
1620
1621 public:
1622 virtual ir_jump *as_jump()
1623 {
1624 return this;
1625 }
1626 };
1627
1628 class ir_return : public ir_jump {
1629 public:
1630 ir_return()
1631 : value(NULL)
1632 {
1633 this->ir_type = ir_type_return;
1634 }
1635
1636 ir_return(ir_rvalue *value)
1637 : value(value)
1638 {
1639 this->ir_type = ir_type_return;
1640 }
1641
1642 virtual ir_return *clone(void *mem_ctx, struct hash_table *) const;
1643
1644 virtual ir_return *as_return()
1645 {
1646 return this;
1647 }
1648
1649 ir_rvalue *get_value() const
1650 {
1651 return value;
1652 }
1653
1654 virtual void accept(ir_visitor *v)
1655 {
1656 v->visit(this);
1657 }
1658
1659 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1660
1661 ir_rvalue *value;
1662 };
1663
1664
1665 /**
1666 * Jump instructions used inside loops
1667 *
1668 * These include \c break and \c continue. The \c break within a loop is
1669 * different from the \c break within a switch-statement.
1670 *
1671 * \sa ir_switch_jump
1672 */
1673 class ir_loop_jump : public ir_jump {
1674 public:
1675 enum jump_mode {
1676 jump_break,
1677 jump_continue
1678 };
1679
1680 ir_loop_jump(jump_mode mode)
1681 {
1682 this->ir_type = ir_type_loop_jump;
1683 this->mode = mode;
1684 }
1685
1686 virtual ir_loop_jump *clone(void *mem_ctx, struct hash_table *) const;
1687
1688 virtual void accept(ir_visitor *v)
1689 {
1690 v->visit(this);
1691 }
1692
1693 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1694
1695 bool is_break() const
1696 {
1697 return mode == jump_break;
1698 }
1699
1700 bool is_continue() const
1701 {
1702 return mode == jump_continue;
1703 }
1704
1705 /** Mode selector for the jump instruction. */
1706 enum jump_mode mode;
1707 };
1708
1709 /**
1710 * IR instruction representing discard statements.
1711 */
1712 class ir_discard : public ir_jump {
1713 public:
1714 ir_discard()
1715 {
1716 this->ir_type = ir_type_discard;
1717 this->condition = NULL;
1718 }
1719
1720 ir_discard(ir_rvalue *cond)
1721 {
1722 this->ir_type = ir_type_discard;
1723 this->condition = cond;
1724 }
1725
1726 virtual ir_discard *clone(void *mem_ctx, struct hash_table *ht) const;
1727
1728 virtual void accept(ir_visitor *v)
1729 {
1730 v->visit(this);
1731 }
1732
1733 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1734
1735 virtual ir_discard *as_discard()
1736 {
1737 return this;
1738 }
1739
1740 ir_rvalue *condition;
1741 };
1742 /*@}*/
1743
1744
1745 /**
1746 * Texture sampling opcodes used in ir_texture
1747 */
1748 enum ir_texture_opcode {
1749 ir_tex, /**< Regular texture look-up */
1750 ir_txb, /**< Texture look-up with LOD bias */
1751 ir_txl, /**< Texture look-up with explicit LOD */
1752 ir_txd, /**< Texture look-up with partial derivatvies */
1753 ir_txf, /**< Texel fetch with explicit LOD */
1754 ir_txf_ms, /**< Multisample texture fetch */
1755 ir_txs, /**< Texture size */
1756 ir_lod, /**< Texture lod query */
1757 ir_tg4, /**< Texture gather */
1758 ir_query_levels /**< Texture levels query */
1759 };
1760
1761
1762 /**
1763 * IR instruction to sample a texture
1764 *
1765 * The specific form of the IR instruction depends on the \c mode value
1766 * selected from \c ir_texture_opcodes. In the printed IR, these will
1767 * appear as:
1768 *
1769 * Texel offset (0 or an expression)
1770 * | Projection divisor
1771 * | | Shadow comparitor
1772 * | | |
1773 * v v v
1774 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1775 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1776 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1777 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1778 * (txf <type> <sampler> <coordinate> 0 <lod>)
1779 * (txf_ms
1780 * <type> <sampler> <coordinate> <sample_index>)
1781 * (txs <type> <sampler> <lod>)
1782 * (lod <type> <sampler> <coordinate>)
1783 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
1784 * (query_levels <type> <sampler>)
1785 */
1786 class ir_texture : public ir_rvalue {
1787 public:
1788 ir_texture(enum ir_texture_opcode op)
1789 : op(op), sampler(NULL), coordinate(NULL), projector(NULL),
1790 shadow_comparitor(NULL), offset(NULL)
1791 {
1792 this->ir_type = ir_type_texture;
1793 memset(&lod_info, 0, sizeof(lod_info));
1794 }
1795
1796 virtual ir_texture *clone(void *mem_ctx, struct hash_table *) const;
1797
1798 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1799
1800 virtual void accept(ir_visitor *v)
1801 {
1802 v->visit(this);
1803 }
1804
1805 virtual ir_texture *as_texture()
1806 {
1807 return this;
1808 }
1809
1810 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1811
1812 virtual bool equals(ir_instruction *ir);
1813
1814 /**
1815 * Return a string representing the ir_texture_opcode.
1816 */
1817 const char *opcode_string();
1818
1819 /** Set the sampler and type. */
1820 void set_sampler(ir_dereference *sampler, const glsl_type *type);
1821
1822 /**
1823 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1824 */
1825 static ir_texture_opcode get_opcode(const char *);
1826
1827 enum ir_texture_opcode op;
1828
1829 /** Sampler to use for the texture access. */
1830 ir_dereference *sampler;
1831
1832 /** Texture coordinate to sample */
1833 ir_rvalue *coordinate;
1834
1835 /**
1836 * Value used for projective divide.
1837 *
1838 * If there is no projective divide (the common case), this will be
1839 * \c NULL. Optimization passes should check for this to point to a constant
1840 * of 1.0 and replace that with \c NULL.
1841 */
1842 ir_rvalue *projector;
1843
1844 /**
1845 * Coordinate used for comparison on shadow look-ups.
1846 *
1847 * If there is no shadow comparison, this will be \c NULL. For the
1848 * \c ir_txf opcode, this *must* be \c NULL.
1849 */
1850 ir_rvalue *shadow_comparitor;
1851
1852 /** Texel offset. */
1853 ir_rvalue *offset;
1854
1855 union {
1856 ir_rvalue *lod; /**< Floating point LOD */
1857 ir_rvalue *bias; /**< Floating point LOD bias */
1858 ir_rvalue *sample_index; /**< MSAA sample index */
1859 ir_rvalue *component; /**< Gather component selector */
1860 struct {
1861 ir_rvalue *dPdx; /**< Partial derivative of coordinate wrt X */
1862 ir_rvalue *dPdy; /**< Partial derivative of coordinate wrt Y */
1863 } grad;
1864 } lod_info;
1865 };
1866
1867
1868 struct ir_swizzle_mask {
1869 unsigned x:2;
1870 unsigned y:2;
1871 unsigned z:2;
1872 unsigned w:2;
1873
1874 /**
1875 * Number of components in the swizzle.
1876 */
1877 unsigned num_components:3;
1878
1879 /**
1880 * Does the swizzle contain duplicate components?
1881 *
1882 * L-value swizzles cannot contain duplicate components.
1883 */
1884 unsigned has_duplicates:1;
1885 };
1886
1887
1888 class ir_swizzle : public ir_rvalue {
1889 public:
1890 ir_swizzle(ir_rvalue *, unsigned x, unsigned y, unsigned z, unsigned w,
1891 unsigned count);
1892
1893 ir_swizzle(ir_rvalue *val, const unsigned *components, unsigned count);
1894
1895 ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask);
1896
1897 virtual ir_swizzle *clone(void *mem_ctx, struct hash_table *) const;
1898
1899 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1900
1901 virtual ir_swizzle *as_swizzle()
1902 {
1903 return this;
1904 }
1905
1906 /**
1907 * Construct an ir_swizzle from the textual representation. Can fail.
1908 */
1909 static ir_swizzle *create(ir_rvalue *, const char *, unsigned vector_length);
1910
1911 virtual void accept(ir_visitor *v)
1912 {
1913 v->visit(this);
1914 }
1915
1916 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1917
1918 virtual bool equals(ir_instruction *ir);
1919
1920 bool is_lvalue() const
1921 {
1922 return val->is_lvalue() && !mask.has_duplicates;
1923 }
1924
1925 /**
1926 * Get the variable that is ultimately referenced by an r-value
1927 */
1928 virtual ir_variable *variable_referenced() const;
1929
1930 ir_rvalue *val;
1931 ir_swizzle_mask mask;
1932
1933 private:
1934 /**
1935 * Initialize the mask component of a swizzle
1936 *
1937 * This is used by the \c ir_swizzle constructors.
1938 */
1939 void init_mask(const unsigned *components, unsigned count);
1940 };
1941
1942
1943 class ir_dereference : public ir_rvalue {
1944 public:
1945 virtual ir_dereference *clone(void *mem_ctx, struct hash_table *) const = 0;
1946
1947 virtual ir_dereference *as_dereference()
1948 {
1949 return this;
1950 }
1951
1952 bool is_lvalue() const;
1953
1954 /**
1955 * Get the variable that is ultimately referenced by an r-value
1956 */
1957 virtual ir_variable *variable_referenced() const = 0;
1958
1959 /**
1960 * Get the constant that is ultimately referenced by an r-value,
1961 * in a constant expression evaluation context.
1962 *
1963 * The offset is used when the reference is to a specific column of
1964 * a matrix.
1965 */
1966 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const = 0;
1967 };
1968
1969
1970 class ir_dereference_variable : public ir_dereference {
1971 public:
1972 ir_dereference_variable(ir_variable *var);
1973
1974 virtual ir_dereference_variable *clone(void *mem_ctx,
1975 struct hash_table *) const;
1976
1977 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1978
1979 virtual ir_dereference_variable *as_dereference_variable()
1980 {
1981 return this;
1982 }
1983
1984 virtual bool equals(ir_instruction *ir);
1985
1986 /**
1987 * Get the variable that is ultimately referenced by an r-value
1988 */
1989 virtual ir_variable *variable_referenced() const
1990 {
1991 return this->var;
1992 }
1993
1994 /**
1995 * Get the constant that is ultimately referenced by an r-value,
1996 * in a constant expression evaluation context.
1997 *
1998 * The offset is used when the reference is to a specific column of
1999 * a matrix.
2000 */
2001 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
2002
2003 virtual ir_variable *whole_variable_referenced()
2004 {
2005 /* ir_dereference_variable objects always dereference the entire
2006 * variable. However, if this dereference is dereferenced by anything
2007 * else, the complete deferefernce chain is not a whole-variable
2008 * dereference. This method should only be called on the top most
2009 * ir_rvalue in a dereference chain.
2010 */
2011 return this->var;
2012 }
2013
2014 virtual void accept(ir_visitor *v)
2015 {
2016 v->visit(this);
2017 }
2018
2019 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2020
2021 /**
2022 * Object being dereferenced.
2023 */
2024 ir_variable *var;
2025 };
2026
2027
2028 class ir_dereference_array : public ir_dereference {
2029 public:
2030 ir_dereference_array(ir_rvalue *value, ir_rvalue *array_index);
2031
2032 ir_dereference_array(ir_variable *var, ir_rvalue *array_index);
2033
2034 virtual ir_dereference_array *clone(void *mem_ctx,
2035 struct hash_table *) const;
2036
2037 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
2038
2039 virtual ir_dereference_array *as_dereference_array()
2040 {
2041 return this;
2042 }
2043
2044 virtual bool equals(ir_instruction *ir);
2045
2046 /**
2047 * Get the variable that is ultimately referenced by an r-value
2048 */
2049 virtual ir_variable *variable_referenced() const
2050 {
2051 return this->array->variable_referenced();
2052 }
2053
2054 /**
2055 * Get the constant that is ultimately referenced by an r-value,
2056 * in a constant expression evaluation context.
2057 *
2058 * The offset is used when the reference is to a specific column of
2059 * a matrix.
2060 */
2061 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
2062
2063 virtual void accept(ir_visitor *v)
2064 {
2065 v->visit(this);
2066 }
2067
2068 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2069
2070 ir_rvalue *array;
2071 ir_rvalue *array_index;
2072
2073 private:
2074 void set_array(ir_rvalue *value);
2075 };
2076
2077
2078 class ir_dereference_record : public ir_dereference {
2079 public:
2080 ir_dereference_record(ir_rvalue *value, const char *field);
2081
2082 ir_dereference_record(ir_variable *var, const char *field);
2083
2084 virtual ir_dereference_record *clone(void *mem_ctx,
2085 struct hash_table *) const;
2086
2087 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
2088
2089 virtual ir_dereference_record *as_dereference_record()
2090 {
2091 return this;
2092 }
2093
2094 /**
2095 * Get the variable that is ultimately referenced by an r-value
2096 */
2097 virtual ir_variable *variable_referenced() const
2098 {
2099 return this->record->variable_referenced();
2100 }
2101
2102 /**
2103 * Get the constant that is ultimately referenced by an r-value,
2104 * in a constant expression evaluation context.
2105 *
2106 * The offset is used when the reference is to a specific column of
2107 * a matrix.
2108 */
2109 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
2110
2111 virtual void accept(ir_visitor *v)
2112 {
2113 v->visit(this);
2114 }
2115
2116 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2117
2118 ir_rvalue *record;
2119 const char *field;
2120 };
2121
2122
2123 /**
2124 * Data stored in an ir_constant
2125 */
2126 union ir_constant_data {
2127 unsigned u[16];
2128 int i[16];
2129 float f[16];
2130 bool b[16];
2131 };
2132
2133
2134 class ir_constant : public ir_rvalue {
2135 public:
2136 ir_constant(const struct glsl_type *type, const ir_constant_data *data);
2137 ir_constant(bool b, unsigned vector_elements=1);
2138 ir_constant(unsigned int u, unsigned vector_elements=1);
2139 ir_constant(int i, unsigned vector_elements=1);
2140 ir_constant(float f, unsigned vector_elements=1);
2141
2142 /**
2143 * Construct an ir_constant from a list of ir_constant values
2144 */
2145 ir_constant(const struct glsl_type *type, exec_list *values);
2146
2147 /**
2148 * Construct an ir_constant from a scalar component of another ir_constant
2149 *
2150 * The new \c ir_constant inherits the type of the component from the
2151 * source constant.
2152 *
2153 * \note
2154 * In the case of a matrix constant, the new constant is a scalar, \b not
2155 * a vector.
2156 */
2157 ir_constant(const ir_constant *c, unsigned i);
2158
2159 /**
2160 * Return a new ir_constant of the specified type containing all zeros.
2161 */
2162 static ir_constant *zero(void *mem_ctx, const glsl_type *type);
2163
2164 virtual ir_constant *clone(void *mem_ctx, struct hash_table *) const;
2165
2166 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
2167
2168 virtual ir_constant *as_constant()
2169 {
2170 return this;
2171 }
2172
2173 virtual void accept(ir_visitor *v)
2174 {
2175 v->visit(this);
2176 }
2177
2178 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2179
2180 virtual bool equals(ir_instruction *ir);
2181
2182 /**
2183 * Get a particular component of a constant as a specific type
2184 *
2185 * This is useful, for example, to get a value from an integer constant
2186 * as a float or bool. This appears frequently when constructors are
2187 * called with all constant parameters.
2188 */
2189 /*@{*/
2190 bool get_bool_component(unsigned i) const;
2191 float get_float_component(unsigned i) const;
2192 int get_int_component(unsigned i) const;
2193 unsigned get_uint_component(unsigned i) const;
2194 /*@}*/
2195
2196 ir_constant *get_array_element(unsigned i) const;
2197
2198 ir_constant *get_record_field(const char *name);
2199
2200 /**
2201 * Copy the values on another constant at a given offset.
2202 *
2203 * The offset is ignored for array or struct copies, it's only for
2204 * scalars or vectors into vectors or matrices.
2205 *
2206 * With identical types on both sides and zero offset it's clone()
2207 * without creating a new object.
2208 */
2209
2210 void copy_offset(ir_constant *src, int offset);
2211
2212 /**
2213 * Copy the values on another constant at a given offset and
2214 * following an assign-like mask.
2215 *
2216 * The mask is ignored for scalars.
2217 *
2218 * Note that this function only handles what assign can handle,
2219 * i.e. at most a vector as source and a column of a matrix as
2220 * destination.
2221 */
2222
2223 void copy_masked_offset(ir_constant *src, int offset, unsigned int mask);
2224
2225 /**
2226 * Determine whether a constant has the same value as another constant
2227 *
2228 * \sa ir_constant::is_zero, ir_constant::is_one,
2229 * ir_constant::is_negative_one, ir_constant::is_basis
2230 */
2231 bool has_value(const ir_constant *) const;
2232
2233 virtual bool is_zero() const;
2234 virtual bool is_one() const;
2235 virtual bool is_negative_one() const;
2236 virtual bool is_basis() const;
2237
2238 /**
2239 * Value of the constant.
2240 *
2241 * The field used to back the values supplied by the constant is determined
2242 * by the type associated with the \c ir_instruction. Constants may be
2243 * scalars, vectors, or matrices.
2244 */
2245 union ir_constant_data value;
2246
2247 /* Array elements */
2248 ir_constant **array_elements;
2249
2250 /* Structure fields */
2251 exec_list components;
2252
2253 private:
2254 /**
2255 * Parameterless constructor only used by the clone method
2256 */
2257 ir_constant(void);
2258 };
2259
2260 /*@}*/
2261
2262 /**
2263 * IR instruction to emit a vertex in a geometry shader.
2264 */
2265 class ir_emit_vertex : public ir_instruction {
2266 public:
2267 ir_emit_vertex()
2268 {
2269 ir_type = ir_type_emit_vertex;
2270 }
2271
2272 virtual void accept(ir_visitor *v)
2273 {
2274 v->visit(this);
2275 }
2276
2277 virtual ir_emit_vertex *clone(void *mem_ctx, struct hash_table *) const
2278 {
2279 return new(mem_ctx) ir_emit_vertex();
2280 }
2281
2282 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2283 };
2284
2285 /**
2286 * IR instruction to complete the current primitive and start a new one in a
2287 * geometry shader.
2288 */
2289 class ir_end_primitive : public ir_instruction {
2290 public:
2291 ir_end_primitive()
2292 {
2293 ir_type = ir_type_end_primitive;
2294 }
2295
2296 virtual void accept(ir_visitor *v)
2297 {
2298 v->visit(this);
2299 }
2300
2301 virtual ir_end_primitive *clone(void *mem_ctx, struct hash_table *) const
2302 {
2303 return new(mem_ctx) ir_end_primitive();
2304 }
2305
2306 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2307 };
2308
2309 /**
2310 * Apply a visitor to each IR node in a list
2311 */
2312 void
2313 visit_exec_list(exec_list *list, ir_visitor *visitor);
2314
2315 /**
2316 * Validate invariants on each IR node in a list
2317 */
2318 void validate_ir_tree(exec_list *instructions);
2319
2320 struct _mesa_glsl_parse_state;
2321 struct gl_shader_program;
2322
2323 /**
2324 * Detect whether an unlinked shader contains static recursion
2325 *
2326 * If the list of instructions is determined to contain static recursion,
2327 * \c _mesa_glsl_error will be called to emit error messages for each function
2328 * that is in the recursion cycle.
2329 */
2330 void
2331 detect_recursion_unlinked(struct _mesa_glsl_parse_state *state,
2332 exec_list *instructions);
2333
2334 /**
2335 * Detect whether a linked shader contains static recursion
2336 *
2337 * If the list of instructions is determined to contain static recursion,
2338 * \c link_error_printf will be called to emit error messages for each function
2339 * that is in the recursion cycle. In addition,
2340 * \c gl_shader_program::LinkStatus will be set to false.
2341 */
2342 void
2343 detect_recursion_linked(struct gl_shader_program *prog,
2344 exec_list *instructions);
2345
2346 /**
2347 * Make a clone of each IR instruction in a list
2348 *
2349 * \param in List of IR instructions that are to be cloned
2350 * \param out List to hold the cloned instructions
2351 */
2352 void
2353 clone_ir_list(void *mem_ctx, exec_list *out, const exec_list *in);
2354
2355 extern void
2356 _mesa_glsl_initialize_variables(exec_list *instructions,
2357 struct _mesa_glsl_parse_state *state);
2358
2359 extern void
2360 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state *state);
2361
2362 extern void
2363 _mesa_glsl_initialize_builtin_functions();
2364
2365 extern ir_function_signature *
2366 _mesa_glsl_find_builtin_function(_mesa_glsl_parse_state *state,
2367 const char *name, exec_list *actual_parameters);
2368
2369 extern void
2370 _mesa_glsl_release_functions(void);
2371
2372 extern void
2373 _mesa_glsl_release_builtin_functions(void);
2374
2375 extern void
2376 reparent_ir(exec_list *list, void *mem_ctx);
2377
2378 struct glsl_symbol_table;
2379
2380 extern void
2381 import_prototypes(const exec_list *source, exec_list *dest,
2382 struct glsl_symbol_table *symbols, void *mem_ctx);
2383
2384 extern bool
2385 ir_has_call(ir_instruction *ir);
2386
2387 extern void
2388 do_set_program_inouts(exec_list *instructions, struct gl_program *prog,
2389 GLenum shader_type);
2390
2391 extern char *
2392 prototype_string(const glsl_type *return_type, const char *name,
2393 exec_list *parameters);
2394
2395 const char *
2396 mode_string(const ir_variable *var);
2397
2398 extern "C" {
2399 #endif /* __cplusplus */
2400
2401 extern void _mesa_print_ir(struct exec_list *instructions,
2402 struct _mesa_glsl_parse_state *state);
2403
2404 #ifdef __cplusplus
2405 } /* extern "C" */
2406 #endif
2407
2408 unsigned
2409 vertices_per_prim(GLenum prim);
2410
2411 #endif /* IR_H */