glsl: Make is_16bit_constant from i965 an ir_constant method.
[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 /**
91 * Base class of all IR instructions
92 */
93 class ir_instruction : public exec_node {
94 public:
95 enum ir_node_type ir_type;
96
97 /**
98 * GCC 4.7+ and clang warn when deleting an ir_instruction unless
99 * there's a virtual destructor present. Because we almost
100 * universally use ralloc for our memory management of
101 * ir_instructions, the destructor doesn't need to do any work.
102 */
103 virtual ~ir_instruction()
104 {
105 }
106
107 /** ir_print_visitor helper for debugging. */
108 void print(void) const;
109 void fprint(FILE *f) const;
110
111 virtual void accept(ir_visitor *) = 0;
112 virtual ir_visitor_status accept(ir_hierarchical_visitor *) = 0;
113 virtual ir_instruction *clone(void *mem_ctx,
114 struct hash_table *ht) const = 0;
115
116 /**
117 * \name IR instruction downcast functions
118 *
119 * These functions either cast the object to a derived class or return
120 * \c NULL if the object's type does not match the specified derived class.
121 * Additional downcast functions will be added as needed.
122 */
123 /*@{*/
124 virtual class ir_variable * as_variable() { return NULL; }
125 virtual class ir_function * as_function() { return NULL; }
126 virtual class ir_dereference * as_dereference() { return NULL; }
127 virtual class ir_dereference_array * as_dereference_array() { return NULL; }
128 virtual class ir_dereference_variable *as_dereference_variable() { return NULL; }
129 virtual class ir_dereference_record *as_dereference_record() { return NULL; }
130 virtual class ir_expression * as_expression() { return NULL; }
131 virtual class ir_rvalue * as_rvalue() { return NULL; }
132 virtual class ir_loop * as_loop() { return NULL; }
133 virtual class ir_assignment * as_assignment() { return NULL; }
134 virtual class ir_call * as_call() { return NULL; }
135 virtual class ir_return * as_return() { return NULL; }
136 virtual class ir_if * as_if() { return NULL; }
137 virtual class ir_swizzle * as_swizzle() { return NULL; }
138 virtual class ir_texture * as_texture() { return NULL; }
139 virtual class ir_constant * as_constant() { return NULL; }
140 virtual class ir_discard * as_discard() { return NULL; }
141 virtual class ir_jump * as_jump() { return NULL; }
142 /*@}*/
143
144 /**
145 * IR equality method: Return true if the referenced instruction would
146 * return the same value as this one.
147 *
148 * This intended to be used for CSE and algebraic optimizations, on rvalues
149 * in particular. No support for other instruction types (assignments,
150 * jumps, calls, etc.) is planned.
151 */
152 virtual bool equals(ir_instruction *ir, enum ir_node_type ignore = ir_type_unset);
153
154 protected:
155 ir_instruction()
156 {
157 ir_type = ir_type_unset;
158 }
159 };
160
161
162 /**
163 * The base class for all "values"/expression trees.
164 */
165 class ir_rvalue : public ir_instruction {
166 public:
167 const struct glsl_type *type;
168
169 virtual ir_rvalue *clone(void *mem_ctx, struct hash_table *) const;
170
171 virtual void accept(ir_visitor *v)
172 {
173 v->visit(this);
174 }
175
176 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
177
178 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
179
180 virtual ir_rvalue * as_rvalue()
181 {
182 return this;
183 }
184
185 ir_rvalue *as_rvalue_to_saturate();
186
187 virtual bool is_lvalue() const
188 {
189 return false;
190 }
191
192 /**
193 * Get the variable that is ultimately referenced by an r-value
194 */
195 virtual ir_variable *variable_referenced() const
196 {
197 return NULL;
198 }
199
200
201 /**
202 * If an r-value is a reference to a whole variable, get that variable
203 *
204 * \return
205 * Pointer to a variable that is completely dereferenced by the r-value. If
206 * the r-value is not a dereference or the dereference does not access the
207 * entire variable (i.e., it's just one array element, struct field), \c NULL
208 * is returned.
209 */
210 virtual ir_variable *whole_variable_referenced()
211 {
212 return NULL;
213 }
214
215 /**
216 * Determine if an r-value has the value zero
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 * zero (or \c false for booleans).
222 *
223 * \sa ir_constant::has_value, ir_rvalue::is_one, ir_rvalue::is_negative_one,
224 * ir_constant::is_basis
225 */
226 virtual bool is_zero() const;
227
228 /**
229 * Determine if an r-value has the value 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 * one (or \c true for booleans).
235 *
236 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_negative_one,
237 * ir_constant::is_basis
238 */
239 virtual bool is_one() const;
240
241 /**
242 * Determine if an r-value has the value negative one
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 all elements set to the value
247 * negative one. For boolean types, the result is always \c false.
248 *
249 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one
250 * ir_constant::is_basis
251 */
252 virtual bool is_negative_one() const;
253
254 /**
255 * Determine if an r-value is a basis vector
256 *
257 * The base implementation of this function always returns \c false. The
258 * \c ir_constant class over-rides this function to return \c true \b only
259 * for vector and scalar types that have one element set to the value one,
260 * and the other elements set to the value zero. For boolean types, the
261 * result is always \c false.
262 *
263 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one,
264 * is_constant::is_negative_one
265 */
266 virtual bool is_basis() const;
267
268 /**
269 * Determine if an r-value is an unsigned integer constant which can be
270 * stored in 16 bits.
271 *
272 * \sa ir_constant::is_uint16_constant.
273 */
274 virtual bool is_uint16_constant() const { return false; }
275
276 /**
277 * Return a generic value of error_type.
278 *
279 * Allocation will be performed with 'mem_ctx' as ralloc owner.
280 */
281 static ir_rvalue *error_value(void *mem_ctx);
282
283 protected:
284 ir_rvalue();
285 };
286
287
288 /**
289 * Variable storage classes
290 */
291 enum ir_variable_mode {
292 ir_var_auto = 0, /**< Function local variables and globals. */
293 ir_var_uniform, /**< Variable declared as a uniform. */
294 ir_var_shader_in,
295 ir_var_shader_out,
296 ir_var_function_in,
297 ir_var_function_out,
298 ir_var_function_inout,
299 ir_var_const_in, /**< "in" param that must be a constant expression */
300 ir_var_system_value, /**< Ex: front-face, instance-id, etc. */
301 ir_var_temporary, /**< Temporary variable generated during compilation. */
302 ir_var_mode_count /**< Number of variable modes */
303 };
304
305 /**
306 * Enum keeping track of how a variable was declared. For error checking of
307 * the gl_PerVertex redeclaration rules.
308 */
309 enum ir_var_declaration_type {
310 /**
311 * Normal declaration (for most variables, this means an explicit
312 * declaration. Exception: temporaries are always implicitly declared, but
313 * they still use ir_var_declared_normally).
314 *
315 * Note: an ir_variable that represents a named interface block uses
316 * ir_var_declared_normally.
317 */
318 ir_var_declared_normally = 0,
319
320 /**
321 * Variable was explicitly declared (or re-declared) in an unnamed
322 * interface block.
323 */
324 ir_var_declared_in_block,
325
326 /**
327 * Variable is an implicitly declared built-in that has not been explicitly
328 * re-declared by the shader.
329 */
330 ir_var_declared_implicitly,
331 };
332
333 /**
334 * \brief Layout qualifiers for gl_FragDepth.
335 *
336 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
337 * with a layout qualifier.
338 */
339 enum ir_depth_layout {
340 ir_depth_layout_none, /**< No depth layout is specified. */
341 ir_depth_layout_any,
342 ir_depth_layout_greater,
343 ir_depth_layout_less,
344 ir_depth_layout_unchanged
345 };
346
347 /**
348 * \brief Convert depth layout qualifier to string.
349 */
350 const char*
351 depth_layout_string(ir_depth_layout layout);
352
353 /**
354 * Description of built-in state associated with a uniform
355 *
356 * \sa ir_variable::state_slots
357 */
358 struct ir_state_slot {
359 int tokens[5];
360 int swizzle;
361 };
362
363
364 /**
365 * Get the string value for an interpolation qualifier
366 *
367 * \return The string that would be used in a shader to specify \c
368 * mode will be returned.
369 *
370 * This function is used to generate error messages of the form "shader
371 * uses %s interpolation qualifier", so in the case where there is no
372 * interpolation qualifier, it returns "no".
373 *
374 * This function should only be used on a shader input or output variable.
375 */
376 const char *interpolation_string(unsigned interpolation);
377
378
379 class ir_variable : public ir_instruction {
380 public:
381 ir_variable(const struct glsl_type *, const char *, ir_variable_mode);
382
383 virtual ir_variable *clone(void *mem_ctx, struct hash_table *ht) const;
384
385 virtual ir_variable *as_variable()
386 {
387 return this;
388 }
389
390 virtual void accept(ir_visitor *v)
391 {
392 v->visit(this);
393 }
394
395 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
396
397
398 /**
399 * Determine how this variable should be interpolated based on its
400 * interpolation qualifier (if present), whether it is gl_Color or
401 * gl_SecondaryColor, and whether flatshading is enabled in the current GL
402 * state.
403 *
404 * The return value will always be either INTERP_QUALIFIER_SMOOTH,
405 * INTERP_QUALIFIER_NOPERSPECTIVE, or INTERP_QUALIFIER_FLAT.
406 */
407 glsl_interp_qualifier determine_interpolation_mode(bool flat_shade);
408
409 /**
410 * Determine whether or not a variable is part of a uniform block.
411 */
412 inline bool is_in_uniform_block() const
413 {
414 return this->data.mode == ir_var_uniform && this->interface_type != NULL;
415 }
416
417 /**
418 * Determine whether or not a variable is the declaration of an interface
419 * block
420 *
421 * For the first declaration below, there will be an \c ir_variable named
422 * "instance" whose type and whose instance_type will be the same
423 * \cglsl_type. For the second declaration, there will be an \c ir_variable
424 * named "f" whose type is float and whose instance_type is B2.
425 *
426 * "instance" is an interface instance variable, but "f" is not.
427 *
428 * uniform B1 {
429 * float f;
430 * } instance;
431 *
432 * uniform B2 {
433 * float f;
434 * };
435 */
436 inline bool is_interface_instance() const
437 {
438 const glsl_type *const t = this->type;
439
440 return (t == this->interface_type)
441 || (t->is_array() && t->fields.array == this->interface_type);
442 }
443
444 /**
445 * Set this->interface_type on a newly created variable.
446 */
447 void init_interface_type(const struct glsl_type *type)
448 {
449 assert(this->interface_type == NULL);
450 this->interface_type = type;
451 if (this->is_interface_instance()) {
452 this->max_ifc_array_access =
453 rzalloc_array(this, unsigned, type->length);
454 }
455 }
456
457 /**
458 * Change this->interface_type on a variable that previously had a
459 * different, but compatible, interface_type. This is used during linking
460 * to set the size of arrays in interface blocks.
461 */
462 void change_interface_type(const struct glsl_type *type)
463 {
464 if (this->max_ifc_array_access != NULL) {
465 /* max_ifc_array_access has already been allocated, so make sure the
466 * new interface has the same number of fields as the old one.
467 */
468 assert(this->interface_type->length == type->length);
469 }
470 this->interface_type = type;
471 }
472
473 /**
474 * Change this->interface_type on a variable that previously had a
475 * different, and incompatible, interface_type. This is used during
476 * compilation to handle redeclaration of the built-in gl_PerVertex
477 * interface block.
478 */
479 void reinit_interface_type(const struct glsl_type *type)
480 {
481 if (this->max_ifc_array_access != NULL) {
482 #ifndef NDEBUG
483 /* Redeclaring gl_PerVertex is only allowed if none of the built-ins
484 * it defines have been accessed yet; so it's safe to throw away the
485 * old max_ifc_array_access pointer, since all of its values are
486 * zero.
487 */
488 for (unsigned i = 0; i < this->interface_type->length; i++)
489 assert(this->max_ifc_array_access[i] == 0);
490 #endif
491 ralloc_free(this->max_ifc_array_access);
492 this->max_ifc_array_access = NULL;
493 }
494 this->interface_type = NULL;
495 init_interface_type(type);
496 }
497
498 const glsl_type *get_interface_type() const
499 {
500 return this->interface_type;
501 }
502
503 /**
504 * Declared type of the variable
505 */
506 const struct glsl_type *type;
507
508 /**
509 * Declared name of the variable
510 */
511 const char *name;
512
513 /**
514 * For variables which satisfy the is_interface_instance() predicate, this
515 * points to an array of integers such that if the ith member of the
516 * interface block is an array, max_ifc_array_access[i] is the maximum
517 * array element of that member that has been accessed. If the ith member
518 * of the interface block is not an array, max_ifc_array_access[i] is
519 * unused.
520 *
521 * For variables whose type is not an interface block, this pointer is
522 * NULL.
523 */
524 unsigned *max_ifc_array_access;
525
526 struct ir_variable_data {
527
528 /**
529 * Is the variable read-only?
530 *
531 * This is set for variables declared as \c const, shader inputs,
532 * and uniforms.
533 */
534 unsigned read_only:1;
535 unsigned centroid:1;
536 unsigned sample:1;
537 unsigned invariant:1;
538
539 /**
540 * Has this variable been used for reading or writing?
541 *
542 * Several GLSL semantic checks require knowledge of whether or not a
543 * variable has been used. For example, it is an error to redeclare a
544 * variable as invariant after it has been used.
545 *
546 * This is only maintained in the ast_to_hir.cpp path, not in
547 * Mesa's fixed function or ARB program paths.
548 */
549 unsigned used:1;
550
551 /**
552 * Has this variable been statically assigned?
553 *
554 * This answers whether the variable was assigned in any path of
555 * the shader during ast_to_hir. This doesn't answer whether it is
556 * still written after dead code removal, nor is it maintained in
557 * non-ast_to_hir.cpp (GLSL parsing) paths.
558 */
559 unsigned assigned:1;
560
561 /**
562 * Enum indicating how the variable was declared. See
563 * ir_var_declaration_type.
564 *
565 * This is used to detect certain kinds of illegal variable redeclarations.
566 */
567 unsigned how_declared:2;
568
569 /**
570 * Storage class of the variable.
571 *
572 * \sa ir_variable_mode
573 */
574 unsigned mode:4;
575
576 /**
577 * Interpolation mode for shader inputs / outputs
578 *
579 * \sa ir_variable_interpolation
580 */
581 unsigned interpolation:2;
582
583 /**
584 * \name ARB_fragment_coord_conventions
585 * @{
586 */
587 unsigned origin_upper_left:1;
588 unsigned pixel_center_integer:1;
589 /*@}*/
590
591 /**
592 * Was the location explicitly set in the shader?
593 *
594 * If the location is explicitly set in the shader, it \b cannot be changed
595 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
596 * no effect).
597 */
598 unsigned explicit_location:1;
599 unsigned explicit_index:1;
600
601 /**
602 * Was an initial binding explicitly set in the shader?
603 *
604 * If so, constant_value contains an integer ir_constant representing the
605 * initial binding point.
606 */
607 unsigned explicit_binding:1;
608
609 /**
610 * Does this variable have an initializer?
611 *
612 * This is used by the linker to cross-validiate initializers of global
613 * variables.
614 */
615 unsigned has_initializer:1;
616
617 /**
618 * Is this variable a generic output or input that has not yet been matched
619 * up to a variable in another stage of the pipeline?
620 *
621 * This is used by the linker as scratch storage while assigning locations
622 * to generic inputs and outputs.
623 */
624 unsigned is_unmatched_generic_inout:1;
625
626 /**
627 * If non-zero, then this variable may be packed along with other variables
628 * into a single varying slot, so this offset should be applied when
629 * accessing components. For example, an offset of 1 means that the x
630 * component of this variable is actually stored in component y of the
631 * location specified by \c location.
632 */
633 unsigned location_frac:2;
634
635 /**
636 * Non-zero if this variable was created by lowering a named interface
637 * block which was not an array.
638 *
639 * Note that this variable and \c from_named_ifc_block_array will never
640 * both be non-zero.
641 */
642 unsigned from_named_ifc_block_nonarray:1;
643
644 /**
645 * Non-zero if this variable was created by lowering a named interface
646 * block which was an array.
647 *
648 * Note that this variable and \c from_named_ifc_block_nonarray will never
649 * both be non-zero.
650 */
651 unsigned from_named_ifc_block_array:1;
652
653 /**
654 * \brief Layout qualifier for gl_FragDepth.
655 *
656 * This is not equal to \c ir_depth_layout_none if and only if this
657 * variable is \c gl_FragDepth and a layout qualifier is specified.
658 */
659 ir_depth_layout depth_layout;
660
661 /**
662 * Storage location of the base of this variable
663 *
664 * The precise meaning of this field depends on the nature of the variable.
665 *
666 * - Vertex shader input: one of the values from \c gl_vert_attrib.
667 * - Vertex shader output: one of the values from \c gl_varying_slot.
668 * - Geometry shader input: one of the values from \c gl_varying_slot.
669 * - Geometry shader output: one of the values from \c gl_varying_slot.
670 * - Fragment shader input: one of the values from \c gl_varying_slot.
671 * - Fragment shader output: one of the values from \c gl_frag_result.
672 * - Uniforms: Per-stage uniform slot number for default uniform block.
673 * - Uniforms: Index within the uniform block definition for UBO members.
674 * - Other: This field is not currently used.
675 *
676 * If the variable is a uniform, shader input, or shader output, and the
677 * slot has not been assigned, the value will be -1.
678 */
679 int location;
680
681 /**
682 * output index for dual source blending.
683 */
684 int index;
685
686 /**
687 * Initial binding point for a sampler or UBO.
688 *
689 * For array types, this represents the binding point for the first element.
690 */
691 int binding;
692
693 /**
694 * Location an atomic counter is stored at.
695 */
696 struct {
697 unsigned buffer_index;
698 unsigned offset;
699 } atomic;
700
701 /**
702 * ARB_shader_image_load_store qualifiers.
703 */
704 struct {
705 bool read_only; /**< "readonly" qualifier. */
706 bool write_only; /**< "writeonly" qualifier. */
707 bool coherent;
708 bool _volatile;
709 bool restrict_flag;
710
711 /** Image internal format if specified explicitly, otherwise GL_NONE. */
712 GLenum format;
713 } image;
714
715 /**
716 * Highest element accessed with a constant expression array index
717 *
718 * Not used for non-array variables.
719 */
720 unsigned max_array_access;
721
722 } data;
723
724 /**
725 * Built-in state that backs this uniform
726 *
727 * Once set at variable creation, \c state_slots must remain invariant.
728 * This is because, ideally, this array would be shared by all clones of
729 * this variable in the IR tree. In other words, we'd really like for it
730 * to be a fly-weight.
731 *
732 * If the variable is not a uniform, \c num_state_slots will be zero and
733 * \c state_slots will be \c NULL.
734 */
735 /*@{*/
736 unsigned num_state_slots; /**< Number of state slots used */
737 ir_state_slot *state_slots; /**< State descriptors. */
738 /*@}*/
739
740 /**
741 * Emit a warning if this variable is accessed.
742 */
743 const char *warn_extension;
744
745 /**
746 * Value assigned in the initializer of a variable declared "const"
747 */
748 ir_constant *constant_value;
749
750 /**
751 * Constant expression assigned in the initializer of the variable
752 *
753 * \warning
754 * This field and \c ::constant_value are distinct. Even if the two fields
755 * refer to constants with the same value, they must point to separate
756 * objects.
757 */
758 ir_constant *constant_initializer;
759
760 private:
761 /**
762 * For variables that are in an interface block or are an instance of an
763 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
764 *
765 * \sa ir_variable::location
766 */
767 const glsl_type *interface_type;
768 };
769
770 /**
771 * A function that returns whether a built-in function is available in the
772 * current shading language (based on version, ES or desktop, and extensions).
773 */
774 typedef bool (*builtin_available_predicate)(const _mesa_glsl_parse_state *);
775
776 /*@{*/
777 /**
778 * The representation of a function instance; may be the full definition or
779 * simply a prototype.
780 */
781 class ir_function_signature : public ir_instruction {
782 /* An ir_function_signature will be part of the list of signatures in
783 * an ir_function.
784 */
785 public:
786 ir_function_signature(const glsl_type *return_type,
787 builtin_available_predicate builtin_avail = NULL);
788
789 virtual ir_function_signature *clone(void *mem_ctx,
790 struct hash_table *ht) const;
791 ir_function_signature *clone_prototype(void *mem_ctx,
792 struct hash_table *ht) const;
793
794 virtual void accept(ir_visitor *v)
795 {
796 v->visit(this);
797 }
798
799 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
800
801 /**
802 * Attempt to evaluate this function as a constant expression,
803 * given a list of the actual parameters and the variable context.
804 * Returns NULL for non-built-ins.
805 */
806 ir_constant *constant_expression_value(exec_list *actual_parameters, struct hash_table *variable_context);
807
808 /**
809 * Get the name of the function for which this is a signature
810 */
811 const char *function_name() const;
812
813 /**
814 * Get a handle to the function for which this is a signature
815 *
816 * There is no setter function, this function returns a \c const pointer,
817 * and \c ir_function_signature::_function is private for a reason. The
818 * only way to make a connection between a function and function signature
819 * is via \c ir_function::add_signature. This helps ensure that certain
820 * invariants (i.e., a function signature is in the list of signatures for
821 * its \c _function) are met.
822 *
823 * \sa ir_function::add_signature
824 */
825 inline const class ir_function *function() const
826 {
827 return this->_function;
828 }
829
830 /**
831 * Check whether the qualifiers match between this signature's parameters
832 * and the supplied parameter list. If not, returns the name of the first
833 * parameter with mismatched qualifiers (for use in error messages).
834 */
835 const char *qualifiers_match(exec_list *params);
836
837 /**
838 * Replace the current parameter list with the given one. This is useful
839 * if the current information came from a prototype, and either has invalid
840 * or missing parameter names.
841 */
842 void replace_parameters(exec_list *new_params);
843
844 /**
845 * Function return type.
846 *
847 * \note This discards the optional precision qualifier.
848 */
849 const struct glsl_type *return_type;
850
851 /**
852 * List of ir_variable of function parameters.
853 *
854 * This represents the storage. The paramaters passed in a particular
855 * call will be in ir_call::actual_paramaters.
856 */
857 struct exec_list parameters;
858
859 /** Whether or not this function has a body (which may be empty). */
860 unsigned is_defined:1;
861
862 /** Whether or not this function signature is a built-in. */
863 bool is_builtin() const;
864
865 /**
866 * Whether or not this function is an intrinsic to be implemented
867 * by the driver.
868 */
869 bool is_intrinsic;
870
871 /** Whether or not a built-in is available for this shader. */
872 bool is_builtin_available(const _mesa_glsl_parse_state *state) const;
873
874 /** Body of instructions in the function. */
875 struct exec_list body;
876
877 private:
878 /**
879 * A function pointer to a predicate that answers whether a built-in
880 * function is available in the current shader. NULL if not a built-in.
881 */
882 builtin_available_predicate builtin_avail;
883
884 /** Function of which this signature is one overload. */
885 class ir_function *_function;
886
887 /** Function signature of which this one is a prototype clone */
888 const ir_function_signature *origin;
889
890 friend class ir_function;
891
892 /**
893 * Helper function to run a list of instructions for constant
894 * expression evaluation.
895 *
896 * The hash table represents the values of the visible variables.
897 * There are no scoping issues because the table is indexed on
898 * ir_variable pointers, not variable names.
899 *
900 * Returns false if the expression is not constant, true otherwise,
901 * and the value in *result if result is non-NULL.
902 */
903 bool constant_expression_evaluate_expression_list(const struct exec_list &body,
904 struct hash_table *variable_context,
905 ir_constant **result);
906 };
907
908
909 /**
910 * Header for tracking multiple overloaded functions with the same name.
911 * Contains a list of ir_function_signatures representing each of the
912 * actual functions.
913 */
914 class ir_function : public ir_instruction {
915 public:
916 ir_function(const char *name);
917
918 virtual ir_function *clone(void *mem_ctx, struct hash_table *ht) const;
919
920 virtual ir_function *as_function()
921 {
922 return this;
923 }
924
925 virtual void accept(ir_visitor *v)
926 {
927 v->visit(this);
928 }
929
930 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
931
932 void add_signature(ir_function_signature *sig)
933 {
934 sig->_function = this;
935 this->signatures.push_tail(sig);
936 }
937
938 /**
939 * Find a signature that matches a set of actual parameters, taking implicit
940 * conversions into account. Also flags whether the match was exact.
941 */
942 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
943 const exec_list *actual_param,
944 bool *match_is_exact);
945
946 /**
947 * Find a signature that matches a set of actual parameters, taking implicit
948 * conversions into account.
949 */
950 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
951 const exec_list *actual_param);
952
953 /**
954 * Find a signature that exactly matches a set of actual parameters without
955 * any implicit type conversions.
956 */
957 ir_function_signature *exact_matching_signature(_mesa_glsl_parse_state *state,
958 const exec_list *actual_ps);
959
960 /**
961 * Name of the function.
962 */
963 const char *name;
964
965 /** Whether or not this function has a signature that isn't a built-in. */
966 bool has_user_signature();
967
968 /**
969 * List of ir_function_signature for each overloaded function with this name.
970 */
971 struct exec_list signatures;
972 };
973
974 inline const char *ir_function_signature::function_name() const
975 {
976 return this->_function->name;
977 }
978 /*@}*/
979
980
981 /**
982 * IR instruction representing high-level if-statements
983 */
984 class ir_if : public ir_instruction {
985 public:
986 ir_if(ir_rvalue *condition)
987 : condition(condition)
988 {
989 ir_type = ir_type_if;
990 }
991
992 virtual ir_if *clone(void *mem_ctx, struct hash_table *ht) const;
993
994 virtual ir_if *as_if()
995 {
996 return this;
997 }
998
999 virtual void accept(ir_visitor *v)
1000 {
1001 v->visit(this);
1002 }
1003
1004 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1005
1006 ir_rvalue *condition;
1007 /** List of ir_instruction for the body of the then branch */
1008 exec_list then_instructions;
1009 /** List of ir_instruction for the body of the else branch */
1010 exec_list else_instructions;
1011 };
1012
1013
1014 /**
1015 * IR instruction representing a high-level loop structure.
1016 */
1017 class ir_loop : public ir_instruction {
1018 public:
1019 ir_loop();
1020
1021 virtual ir_loop *clone(void *mem_ctx, struct hash_table *ht) const;
1022
1023 virtual void accept(ir_visitor *v)
1024 {
1025 v->visit(this);
1026 }
1027
1028 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1029
1030 virtual ir_loop *as_loop()
1031 {
1032 return this;
1033 }
1034
1035 /** List of ir_instruction that make up the body of the loop. */
1036 exec_list body_instructions;
1037 };
1038
1039
1040 class ir_assignment : public ir_instruction {
1041 public:
1042 ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs, ir_rvalue *condition = NULL);
1043
1044 /**
1045 * Construct an assignment with an explicit write mask
1046 *
1047 * \note
1048 * Since a write mask is supplied, the LHS must already be a bare
1049 * \c ir_dereference. The cannot be any swizzles in the LHS.
1050 */
1051 ir_assignment(ir_dereference *lhs, ir_rvalue *rhs, ir_rvalue *condition,
1052 unsigned write_mask);
1053
1054 virtual ir_assignment *clone(void *mem_ctx, struct hash_table *ht) const;
1055
1056 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1057
1058 virtual void accept(ir_visitor *v)
1059 {
1060 v->visit(this);
1061 }
1062
1063 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1064
1065 virtual ir_assignment * as_assignment()
1066 {
1067 return this;
1068 }
1069
1070 /**
1071 * Get a whole variable written by an assignment
1072 *
1073 * If the LHS of the assignment writes a whole variable, the variable is
1074 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1075 * assignment are:
1076 *
1077 * - Assigning to a scalar
1078 * - Assigning to all components of a vector
1079 * - Whole array (or matrix) assignment
1080 * - Whole structure assignment
1081 */
1082 ir_variable *whole_variable_written();
1083
1084 /**
1085 * Set the LHS of an assignment
1086 */
1087 void set_lhs(ir_rvalue *lhs);
1088
1089 /**
1090 * Left-hand side of the assignment.
1091 *
1092 * This should be treated as read only. If you need to set the LHS of an
1093 * assignment, use \c ir_assignment::set_lhs.
1094 */
1095 ir_dereference *lhs;
1096
1097 /**
1098 * Value being assigned
1099 */
1100 ir_rvalue *rhs;
1101
1102 /**
1103 * Optional condition for the assignment.
1104 */
1105 ir_rvalue *condition;
1106
1107
1108 /**
1109 * Component mask written
1110 *
1111 * For non-vector types in the LHS, this field will be zero. For vector
1112 * types, a bit will be set for each component that is written. Note that
1113 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1114 *
1115 * A partially-set write mask means that each enabled channel gets
1116 * the value from a consecutive channel of the rhs. For example,
1117 * to write just .xyw of gl_FrontColor with color:
1118 *
1119 * (assign (constant bool (1)) (xyw)
1120 * (var_ref gl_FragColor)
1121 * (swiz xyw (var_ref color)))
1122 */
1123 unsigned write_mask:4;
1124 };
1125
1126 /* Update ir_expression::get_num_operands() and operator_strs when
1127 * updating this list.
1128 */
1129 enum ir_expression_operation {
1130 ir_unop_bit_not,
1131 ir_unop_logic_not,
1132 ir_unop_neg,
1133 ir_unop_abs,
1134 ir_unop_sign,
1135 ir_unop_rcp,
1136 ir_unop_rsq,
1137 ir_unop_sqrt,
1138 ir_unop_exp, /**< Log base e on gentype */
1139 ir_unop_log, /**< Natural log on gentype */
1140 ir_unop_exp2,
1141 ir_unop_log2,
1142 ir_unop_f2i, /**< Float-to-integer conversion. */
1143 ir_unop_f2u, /**< Float-to-unsigned conversion. */
1144 ir_unop_i2f, /**< Integer-to-float conversion. */
1145 ir_unop_f2b, /**< Float-to-boolean conversion */
1146 ir_unop_b2f, /**< Boolean-to-float conversion */
1147 ir_unop_i2b, /**< int-to-boolean conversion */
1148 ir_unop_b2i, /**< Boolean-to-int conversion */
1149 ir_unop_u2f, /**< Unsigned-to-float conversion. */
1150 ir_unop_i2u, /**< Integer-to-unsigned conversion. */
1151 ir_unop_u2i, /**< Unsigned-to-integer conversion. */
1152 ir_unop_bitcast_i2f, /**< Bit-identical int-to-float "conversion" */
1153 ir_unop_bitcast_f2i, /**< Bit-identical float-to-int "conversion" */
1154 ir_unop_bitcast_u2f, /**< Bit-identical uint-to-float "conversion" */
1155 ir_unop_bitcast_f2u, /**< Bit-identical float-to-uint "conversion" */
1156 ir_unop_any,
1157
1158 /**
1159 * \name Unary floating-point rounding operations.
1160 */
1161 /*@{*/
1162 ir_unop_trunc,
1163 ir_unop_ceil,
1164 ir_unop_floor,
1165 ir_unop_fract,
1166 ir_unop_round_even,
1167 /*@}*/
1168
1169 /**
1170 * \name Trigonometric operations.
1171 */
1172 /*@{*/
1173 ir_unop_sin,
1174 ir_unop_cos,
1175 ir_unop_sin_reduced, /**< Reduced range sin. [-pi, pi] */
1176 ir_unop_cos_reduced, /**< Reduced range cos. [-pi, pi] */
1177 /*@}*/
1178
1179 /**
1180 * \name Partial derivatives.
1181 */
1182 /*@{*/
1183 ir_unop_dFdx,
1184 ir_unop_dFdy,
1185 /*@}*/
1186
1187 /**
1188 * \name Floating point pack and unpack operations.
1189 */
1190 /*@{*/
1191 ir_unop_pack_snorm_2x16,
1192 ir_unop_pack_snorm_4x8,
1193 ir_unop_pack_unorm_2x16,
1194 ir_unop_pack_unorm_4x8,
1195 ir_unop_pack_half_2x16,
1196 ir_unop_unpack_snorm_2x16,
1197 ir_unop_unpack_snorm_4x8,
1198 ir_unop_unpack_unorm_2x16,
1199 ir_unop_unpack_unorm_4x8,
1200 ir_unop_unpack_half_2x16,
1201 /*@}*/
1202
1203 /**
1204 * \name Lowered floating point unpacking operations.
1205 *
1206 * \see lower_packing_builtins_visitor::split_unpack_half_2x16
1207 */
1208 /*@{*/
1209 ir_unop_unpack_half_2x16_split_x,
1210 ir_unop_unpack_half_2x16_split_y,
1211 /*@}*/
1212
1213 /**
1214 * \name Bit operations, part of ARB_gpu_shader5.
1215 */
1216 /*@{*/
1217 ir_unop_bitfield_reverse,
1218 ir_unop_bit_count,
1219 ir_unop_find_msb,
1220 ir_unop_find_lsb,
1221 /*@}*/
1222
1223 ir_unop_noise,
1224
1225 /**
1226 * A sentinel marking the last of the unary operations.
1227 */
1228 ir_last_unop = ir_unop_noise,
1229
1230 ir_binop_add,
1231 ir_binop_sub,
1232 ir_binop_mul, /**< Floating-point or low 32-bit integer multiply. */
1233 ir_binop_imul_high, /**< Calculates the high 32-bits of a 64-bit multiply. */
1234 ir_binop_div,
1235
1236 /**
1237 * Returns the carry resulting from the addition of the two arguments.
1238 */
1239 /*@{*/
1240 ir_binop_carry,
1241 /*@}*/
1242
1243 /**
1244 * Returns the borrow resulting from the subtraction of the second argument
1245 * from the first argument.
1246 */
1247 /*@{*/
1248 ir_binop_borrow,
1249 /*@}*/
1250
1251 /**
1252 * Takes one of two combinations of arguments:
1253 *
1254 * - mod(vecN, vecN)
1255 * - mod(vecN, float)
1256 *
1257 * Does not take integer types.
1258 */
1259 ir_binop_mod,
1260
1261 /**
1262 * \name Binary comparison operators which return a boolean vector.
1263 * The type of both operands must be equal.
1264 */
1265 /*@{*/
1266 ir_binop_less,
1267 ir_binop_greater,
1268 ir_binop_lequal,
1269 ir_binop_gequal,
1270 ir_binop_equal,
1271 ir_binop_nequal,
1272 /**
1273 * Returns single boolean for whether all components of operands[0]
1274 * equal the components of operands[1].
1275 */
1276 ir_binop_all_equal,
1277 /**
1278 * Returns single boolean for whether any component of operands[0]
1279 * is not equal to the corresponding component of operands[1].
1280 */
1281 ir_binop_any_nequal,
1282 /*@}*/
1283
1284 /**
1285 * \name Bit-wise binary operations.
1286 */
1287 /*@{*/
1288 ir_binop_lshift,
1289 ir_binop_rshift,
1290 ir_binop_bit_and,
1291 ir_binop_bit_xor,
1292 ir_binop_bit_or,
1293 /*@}*/
1294
1295 ir_binop_logic_and,
1296 ir_binop_logic_xor,
1297 ir_binop_logic_or,
1298
1299 ir_binop_dot,
1300 ir_binop_min,
1301 ir_binop_max,
1302
1303 ir_binop_pow,
1304
1305 /**
1306 * \name Lowered floating point packing operations.
1307 *
1308 * \see lower_packing_builtins_visitor::split_pack_half_2x16
1309 */
1310 /*@{*/
1311 ir_binop_pack_half_2x16_split,
1312 /*@}*/
1313
1314 /**
1315 * \name First half of a lowered bitfieldInsert() operation.
1316 *
1317 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1318 */
1319 /*@{*/
1320 ir_binop_bfm,
1321 /*@}*/
1322
1323 /**
1324 * Load a value the size of a given GLSL type from a uniform block.
1325 *
1326 * operand0 is the ir_constant uniform block index in the linked shader.
1327 * operand1 is a byte offset within the uniform block.
1328 */
1329 ir_binop_ubo_load,
1330
1331 /**
1332 * \name Multiplies a number by two to a power, part of ARB_gpu_shader5.
1333 */
1334 /*@{*/
1335 ir_binop_ldexp,
1336 /*@}*/
1337
1338 /**
1339 * Extract a scalar from a vector
1340 *
1341 * operand0 is the vector
1342 * operand1 is the index of the field to read from operand0
1343 */
1344 ir_binop_vector_extract,
1345
1346 /**
1347 * A sentinel marking the last of the binary operations.
1348 */
1349 ir_last_binop = ir_binop_vector_extract,
1350
1351 /**
1352 * \name Fused floating-point multiply-add, part of ARB_gpu_shader5.
1353 */
1354 /*@{*/
1355 ir_triop_fma,
1356 /*@}*/
1357
1358 ir_triop_lrp,
1359
1360 /**
1361 * \name Conditional Select
1362 *
1363 * A vector conditional select instruction (like ?:, but operating per-
1364 * component on vectors).
1365 *
1366 * \see lower_instructions_visitor::ldexp_to_arith
1367 */
1368 /*@{*/
1369 ir_triop_csel,
1370 /*@}*/
1371
1372 /**
1373 * \name Second half of a lowered bitfieldInsert() operation.
1374 *
1375 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1376 */
1377 /*@{*/
1378 ir_triop_bfi,
1379 /*@}*/
1380
1381 ir_triop_bitfield_extract,
1382
1383 /**
1384 * Generate a value with one field of a vector changed
1385 *
1386 * operand0 is the vector
1387 * operand1 is the value to write into the vector result
1388 * operand2 is the index in operand0 to be modified
1389 */
1390 ir_triop_vector_insert,
1391
1392 /**
1393 * A sentinel marking the last of the ternary operations.
1394 */
1395 ir_last_triop = ir_triop_vector_insert,
1396
1397 ir_quadop_bitfield_insert,
1398
1399 ir_quadop_vector,
1400
1401 /**
1402 * A sentinel marking the last of the ternary operations.
1403 */
1404 ir_last_quadop = ir_quadop_vector,
1405
1406 /**
1407 * A sentinel marking the last of all operations.
1408 */
1409 ir_last_opcode = ir_quadop_vector
1410 };
1411
1412 class ir_expression : public ir_rvalue {
1413 public:
1414 ir_expression(int op, const struct glsl_type *type,
1415 ir_rvalue *op0, ir_rvalue *op1 = NULL,
1416 ir_rvalue *op2 = NULL, ir_rvalue *op3 = NULL);
1417
1418 /**
1419 * Constructor for unary operation expressions
1420 */
1421 ir_expression(int op, ir_rvalue *);
1422
1423 /**
1424 * Constructor for binary operation expressions
1425 */
1426 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1);
1427
1428 /**
1429 * Constructor for ternary operation expressions
1430 */
1431 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1, ir_rvalue *op2);
1432
1433 virtual ir_expression *as_expression()
1434 {
1435 return this;
1436 }
1437
1438 virtual bool equals(ir_instruction *ir, enum ir_node_type ignore = ir_type_unset);
1439
1440 virtual ir_expression *clone(void *mem_ctx, struct hash_table *ht) const;
1441
1442 /**
1443 * Attempt to constant-fold the expression
1444 *
1445 * The "variable_context" hash table links ir_variable * to ir_constant *
1446 * that represent the variables' values. \c NULL represents an empty
1447 * context.
1448 *
1449 * If the expression cannot be constant folded, this method will return
1450 * \c NULL.
1451 */
1452 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1453
1454 /**
1455 * Determine the number of operands used by an expression
1456 */
1457 static unsigned int get_num_operands(ir_expression_operation);
1458
1459 /**
1460 * Determine the number of operands used by an expression
1461 */
1462 unsigned int get_num_operands() const
1463 {
1464 return (this->operation == ir_quadop_vector)
1465 ? this->type->vector_elements : get_num_operands(operation);
1466 }
1467
1468 /**
1469 * Return whether the expression operates on vectors horizontally.
1470 */
1471 bool is_horizontal() const
1472 {
1473 return operation == ir_binop_all_equal ||
1474 operation == ir_binop_any_nequal ||
1475 operation == ir_unop_any ||
1476 operation == ir_binop_dot ||
1477 operation == ir_quadop_vector;
1478 }
1479
1480 /**
1481 * Return a string representing this expression's operator.
1482 */
1483 const char *operator_string();
1484
1485 /**
1486 * Return a string representing this expression's operator.
1487 */
1488 static const char *operator_string(ir_expression_operation);
1489
1490
1491 /**
1492 * Do a reverse-lookup to translate the given string into an operator.
1493 */
1494 static ir_expression_operation get_operator(const char *);
1495
1496 virtual void accept(ir_visitor *v)
1497 {
1498 v->visit(this);
1499 }
1500
1501 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1502
1503 ir_expression_operation operation;
1504 ir_rvalue *operands[4];
1505 };
1506
1507
1508 /**
1509 * HIR instruction representing a high-level function call, containing a list
1510 * of parameters and returning a value in the supplied temporary.
1511 */
1512 class ir_call : public ir_instruction {
1513 public:
1514 ir_call(ir_function_signature *callee,
1515 ir_dereference_variable *return_deref,
1516 exec_list *actual_parameters)
1517 : return_deref(return_deref), callee(callee)
1518 {
1519 ir_type = ir_type_call;
1520 assert(callee->return_type != NULL);
1521 actual_parameters->move_nodes_to(& this->actual_parameters);
1522 this->use_builtin = callee->is_builtin();
1523 }
1524
1525 virtual ir_call *clone(void *mem_ctx, struct hash_table *ht) const;
1526
1527 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1528
1529 virtual ir_call *as_call()
1530 {
1531 return this;
1532 }
1533
1534 virtual void accept(ir_visitor *v)
1535 {
1536 v->visit(this);
1537 }
1538
1539 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1540
1541 /**
1542 * Get the name of the function being called.
1543 */
1544 const char *callee_name() const
1545 {
1546 return callee->function_name();
1547 }
1548
1549 /**
1550 * Generates an inline version of the function before @ir,
1551 * storing the return value in return_deref.
1552 */
1553 void generate_inline(ir_instruction *ir);
1554
1555 /**
1556 * Storage for the function's return value.
1557 * This must be NULL if the return type is void.
1558 */
1559 ir_dereference_variable *return_deref;
1560
1561 /**
1562 * The specific function signature being called.
1563 */
1564 ir_function_signature *callee;
1565
1566 /* List of ir_rvalue of paramaters passed in this call. */
1567 exec_list actual_parameters;
1568
1569 /** Should this call only bind to a built-in function? */
1570 bool use_builtin;
1571 };
1572
1573
1574 /**
1575 * \name Jump-like IR instructions.
1576 *
1577 * These include \c break, \c continue, \c return, and \c discard.
1578 */
1579 /*@{*/
1580 class ir_jump : public ir_instruction {
1581 protected:
1582 ir_jump()
1583 {
1584 ir_type = ir_type_unset;
1585 }
1586
1587 public:
1588 virtual ir_jump *as_jump()
1589 {
1590 return this;
1591 }
1592 };
1593
1594 class ir_return : public ir_jump {
1595 public:
1596 ir_return()
1597 : value(NULL)
1598 {
1599 this->ir_type = ir_type_return;
1600 }
1601
1602 ir_return(ir_rvalue *value)
1603 : value(value)
1604 {
1605 this->ir_type = ir_type_return;
1606 }
1607
1608 virtual ir_return *clone(void *mem_ctx, struct hash_table *) const;
1609
1610 virtual ir_return *as_return()
1611 {
1612 return this;
1613 }
1614
1615 ir_rvalue *get_value() const
1616 {
1617 return value;
1618 }
1619
1620 virtual void accept(ir_visitor *v)
1621 {
1622 v->visit(this);
1623 }
1624
1625 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1626
1627 ir_rvalue *value;
1628 };
1629
1630
1631 /**
1632 * Jump instructions used inside loops
1633 *
1634 * These include \c break and \c continue. The \c break within a loop is
1635 * different from the \c break within a switch-statement.
1636 *
1637 * \sa ir_switch_jump
1638 */
1639 class ir_loop_jump : public ir_jump {
1640 public:
1641 enum jump_mode {
1642 jump_break,
1643 jump_continue
1644 };
1645
1646 ir_loop_jump(jump_mode mode)
1647 {
1648 this->ir_type = ir_type_loop_jump;
1649 this->mode = mode;
1650 }
1651
1652 virtual ir_loop_jump *clone(void *mem_ctx, struct hash_table *) const;
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 bool is_break() const
1662 {
1663 return mode == jump_break;
1664 }
1665
1666 bool is_continue() const
1667 {
1668 return mode == jump_continue;
1669 }
1670
1671 /** Mode selector for the jump instruction. */
1672 enum jump_mode mode;
1673 };
1674
1675 /**
1676 * IR instruction representing discard statements.
1677 */
1678 class ir_discard : public ir_jump {
1679 public:
1680 ir_discard()
1681 {
1682 this->ir_type = ir_type_discard;
1683 this->condition = NULL;
1684 }
1685
1686 ir_discard(ir_rvalue *cond)
1687 {
1688 this->ir_type = ir_type_discard;
1689 this->condition = cond;
1690 }
1691
1692 virtual ir_discard *clone(void *mem_ctx, struct hash_table *ht) const;
1693
1694 virtual void accept(ir_visitor *v)
1695 {
1696 v->visit(this);
1697 }
1698
1699 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1700
1701 virtual ir_discard *as_discard()
1702 {
1703 return this;
1704 }
1705
1706 ir_rvalue *condition;
1707 };
1708 /*@}*/
1709
1710
1711 /**
1712 * Texture sampling opcodes used in ir_texture
1713 */
1714 enum ir_texture_opcode {
1715 ir_tex, /**< Regular texture look-up */
1716 ir_txb, /**< Texture look-up with LOD bias */
1717 ir_txl, /**< Texture look-up with explicit LOD */
1718 ir_txd, /**< Texture look-up with partial derivatvies */
1719 ir_txf, /**< Texel fetch with explicit LOD */
1720 ir_txf_ms, /**< Multisample texture fetch */
1721 ir_txs, /**< Texture size */
1722 ir_lod, /**< Texture lod query */
1723 ir_tg4, /**< Texture gather */
1724 ir_query_levels /**< Texture levels query */
1725 };
1726
1727
1728 /**
1729 * IR instruction to sample a texture
1730 *
1731 * The specific form of the IR instruction depends on the \c mode value
1732 * selected from \c ir_texture_opcodes. In the printed IR, these will
1733 * appear as:
1734 *
1735 * Texel offset (0 or an expression)
1736 * | Projection divisor
1737 * | | Shadow comparitor
1738 * | | |
1739 * v v v
1740 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1741 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1742 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1743 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1744 * (txf <type> <sampler> <coordinate> 0 <lod>)
1745 * (txf_ms
1746 * <type> <sampler> <coordinate> <sample_index>)
1747 * (txs <type> <sampler> <lod>)
1748 * (lod <type> <sampler> <coordinate>)
1749 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
1750 * (query_levels <type> <sampler>)
1751 */
1752 class ir_texture : public ir_rvalue {
1753 public:
1754 ir_texture(enum ir_texture_opcode op)
1755 : op(op), sampler(NULL), coordinate(NULL), projector(NULL),
1756 shadow_comparitor(NULL), offset(NULL)
1757 {
1758 this->ir_type = ir_type_texture;
1759 memset(&lod_info, 0, sizeof(lod_info));
1760 }
1761
1762 virtual ir_texture *clone(void *mem_ctx, struct hash_table *) const;
1763
1764 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1765
1766 virtual void accept(ir_visitor *v)
1767 {
1768 v->visit(this);
1769 }
1770
1771 virtual ir_texture *as_texture()
1772 {
1773 return this;
1774 }
1775
1776 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1777
1778 virtual bool equals(ir_instruction *ir, enum ir_node_type ignore = ir_type_unset);
1779
1780 /**
1781 * Return a string representing the ir_texture_opcode.
1782 */
1783 const char *opcode_string();
1784
1785 /** Set the sampler and type. */
1786 void set_sampler(ir_dereference *sampler, const glsl_type *type);
1787
1788 /**
1789 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1790 */
1791 static ir_texture_opcode get_opcode(const char *);
1792
1793 enum ir_texture_opcode op;
1794
1795 /** Sampler to use for the texture access. */
1796 ir_dereference *sampler;
1797
1798 /** Texture coordinate to sample */
1799 ir_rvalue *coordinate;
1800
1801 /**
1802 * Value used for projective divide.
1803 *
1804 * If there is no projective divide (the common case), this will be
1805 * \c NULL. Optimization passes should check for this to point to a constant
1806 * of 1.0 and replace that with \c NULL.
1807 */
1808 ir_rvalue *projector;
1809
1810 /**
1811 * Coordinate used for comparison on shadow look-ups.
1812 *
1813 * If there is no shadow comparison, this will be \c NULL. For the
1814 * \c ir_txf opcode, this *must* be \c NULL.
1815 */
1816 ir_rvalue *shadow_comparitor;
1817
1818 /** Texel offset. */
1819 ir_rvalue *offset;
1820
1821 union {
1822 ir_rvalue *lod; /**< Floating point LOD */
1823 ir_rvalue *bias; /**< Floating point LOD bias */
1824 ir_rvalue *sample_index; /**< MSAA sample index */
1825 ir_rvalue *component; /**< Gather component selector */
1826 struct {
1827 ir_rvalue *dPdx; /**< Partial derivative of coordinate wrt X */
1828 ir_rvalue *dPdy; /**< Partial derivative of coordinate wrt Y */
1829 } grad;
1830 } lod_info;
1831 };
1832
1833
1834 struct ir_swizzle_mask {
1835 unsigned x:2;
1836 unsigned y:2;
1837 unsigned z:2;
1838 unsigned w:2;
1839
1840 /**
1841 * Number of components in the swizzle.
1842 */
1843 unsigned num_components:3;
1844
1845 /**
1846 * Does the swizzle contain duplicate components?
1847 *
1848 * L-value swizzles cannot contain duplicate components.
1849 */
1850 unsigned has_duplicates:1;
1851 };
1852
1853
1854 class ir_swizzle : public ir_rvalue {
1855 public:
1856 ir_swizzle(ir_rvalue *, unsigned x, unsigned y, unsigned z, unsigned w,
1857 unsigned count);
1858
1859 ir_swizzle(ir_rvalue *val, const unsigned *components, unsigned count);
1860
1861 ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask);
1862
1863 virtual ir_swizzle *clone(void *mem_ctx, struct hash_table *) const;
1864
1865 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1866
1867 virtual ir_swizzle *as_swizzle()
1868 {
1869 return this;
1870 }
1871
1872 /**
1873 * Construct an ir_swizzle from the textual representation. Can fail.
1874 */
1875 static ir_swizzle *create(ir_rvalue *, const char *, unsigned vector_length);
1876
1877 virtual void accept(ir_visitor *v)
1878 {
1879 v->visit(this);
1880 }
1881
1882 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1883
1884 virtual bool equals(ir_instruction *ir, enum ir_node_type ignore = ir_type_unset);
1885
1886 bool is_lvalue() const
1887 {
1888 return val->is_lvalue() && !mask.has_duplicates;
1889 }
1890
1891 /**
1892 * Get the variable that is ultimately referenced by an r-value
1893 */
1894 virtual ir_variable *variable_referenced() const;
1895
1896 ir_rvalue *val;
1897 ir_swizzle_mask mask;
1898
1899 private:
1900 /**
1901 * Initialize the mask component of a swizzle
1902 *
1903 * This is used by the \c ir_swizzle constructors.
1904 */
1905 void init_mask(const unsigned *components, unsigned count);
1906 };
1907
1908
1909 class ir_dereference : public ir_rvalue {
1910 public:
1911 virtual ir_dereference *clone(void *mem_ctx, struct hash_table *) const = 0;
1912
1913 virtual ir_dereference *as_dereference()
1914 {
1915 return this;
1916 }
1917
1918 bool is_lvalue() const;
1919
1920 /**
1921 * Get the variable that is ultimately referenced by an r-value
1922 */
1923 virtual ir_variable *variable_referenced() const = 0;
1924 };
1925
1926
1927 class ir_dereference_variable : public ir_dereference {
1928 public:
1929 ir_dereference_variable(ir_variable *var);
1930
1931 virtual ir_dereference_variable *clone(void *mem_ctx,
1932 struct hash_table *) const;
1933
1934 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1935
1936 virtual ir_dereference_variable *as_dereference_variable()
1937 {
1938 return this;
1939 }
1940
1941 virtual bool equals(ir_instruction *ir, enum ir_node_type ignore = ir_type_unset);
1942
1943 /**
1944 * Get the variable that is ultimately referenced by an r-value
1945 */
1946 virtual ir_variable *variable_referenced() const
1947 {
1948 return this->var;
1949 }
1950
1951 virtual ir_variable *whole_variable_referenced()
1952 {
1953 /* ir_dereference_variable objects always dereference the entire
1954 * variable. However, if this dereference is dereferenced by anything
1955 * else, the complete deferefernce chain is not a whole-variable
1956 * dereference. This method should only be called on the top most
1957 * ir_rvalue in a dereference chain.
1958 */
1959 return this->var;
1960 }
1961
1962 virtual void accept(ir_visitor *v)
1963 {
1964 v->visit(this);
1965 }
1966
1967 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1968
1969 /**
1970 * Object being dereferenced.
1971 */
1972 ir_variable *var;
1973 };
1974
1975
1976 class ir_dereference_array : public ir_dereference {
1977 public:
1978 ir_dereference_array(ir_rvalue *value, ir_rvalue *array_index);
1979
1980 ir_dereference_array(ir_variable *var, ir_rvalue *array_index);
1981
1982 virtual ir_dereference_array *clone(void *mem_ctx,
1983 struct hash_table *) const;
1984
1985 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1986
1987 virtual ir_dereference_array *as_dereference_array()
1988 {
1989 return this;
1990 }
1991
1992 virtual bool equals(ir_instruction *ir, enum ir_node_type ignore = ir_type_unset);
1993
1994 /**
1995 * Get the variable that is ultimately referenced by an r-value
1996 */
1997 virtual ir_variable *variable_referenced() const
1998 {
1999 return this->array->variable_referenced();
2000 }
2001
2002 virtual void accept(ir_visitor *v)
2003 {
2004 v->visit(this);
2005 }
2006
2007 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2008
2009 ir_rvalue *array;
2010 ir_rvalue *array_index;
2011
2012 private:
2013 void set_array(ir_rvalue *value);
2014 };
2015
2016
2017 class ir_dereference_record : public ir_dereference {
2018 public:
2019 ir_dereference_record(ir_rvalue *value, const char *field);
2020
2021 ir_dereference_record(ir_variable *var, const char *field);
2022
2023 virtual ir_dereference_record *clone(void *mem_ctx,
2024 struct hash_table *) const;
2025
2026 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
2027
2028 virtual ir_dereference_record *as_dereference_record()
2029 {
2030 return this;
2031 }
2032
2033 /**
2034 * Get the variable that is ultimately referenced by an r-value
2035 */
2036 virtual ir_variable *variable_referenced() const
2037 {
2038 return this->record->variable_referenced();
2039 }
2040
2041 virtual void accept(ir_visitor *v)
2042 {
2043 v->visit(this);
2044 }
2045
2046 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2047
2048 ir_rvalue *record;
2049 const char *field;
2050 };
2051
2052
2053 /**
2054 * Data stored in an ir_constant
2055 */
2056 union ir_constant_data {
2057 unsigned u[16];
2058 int i[16];
2059 float f[16];
2060 bool b[16];
2061 };
2062
2063
2064 class ir_constant : public ir_rvalue {
2065 public:
2066 ir_constant(const struct glsl_type *type, const ir_constant_data *data);
2067 ir_constant(bool b, unsigned vector_elements=1);
2068 ir_constant(unsigned int u, unsigned vector_elements=1);
2069 ir_constant(int i, unsigned vector_elements=1);
2070 ir_constant(float f, unsigned vector_elements=1);
2071
2072 /**
2073 * Construct an ir_constant from a list of ir_constant values
2074 */
2075 ir_constant(const struct glsl_type *type, exec_list *values);
2076
2077 /**
2078 * Construct an ir_constant from a scalar component of another ir_constant
2079 *
2080 * The new \c ir_constant inherits the type of the component from the
2081 * source constant.
2082 *
2083 * \note
2084 * In the case of a matrix constant, the new constant is a scalar, \b not
2085 * a vector.
2086 */
2087 ir_constant(const ir_constant *c, unsigned i);
2088
2089 /**
2090 * Return a new ir_constant of the specified type containing all zeros.
2091 */
2092 static ir_constant *zero(void *mem_ctx, const glsl_type *type);
2093
2094 virtual ir_constant *clone(void *mem_ctx, struct hash_table *) const;
2095
2096 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
2097
2098 virtual ir_constant *as_constant()
2099 {
2100 return this;
2101 }
2102
2103 virtual void accept(ir_visitor *v)
2104 {
2105 v->visit(this);
2106 }
2107
2108 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2109
2110 virtual bool equals(ir_instruction *ir, enum ir_node_type ignore = ir_type_unset);
2111
2112 /**
2113 * Get a particular component of a constant as a specific type
2114 *
2115 * This is useful, for example, to get a value from an integer constant
2116 * as a float or bool. This appears frequently when constructors are
2117 * called with all constant parameters.
2118 */
2119 /*@{*/
2120 bool get_bool_component(unsigned i) const;
2121 float get_float_component(unsigned i) const;
2122 int get_int_component(unsigned i) const;
2123 unsigned get_uint_component(unsigned i) const;
2124 /*@}*/
2125
2126 ir_constant *get_array_element(unsigned i) const;
2127
2128 ir_constant *get_record_field(const char *name);
2129
2130 /**
2131 * Copy the values on another constant at a given offset.
2132 *
2133 * The offset is ignored for array or struct copies, it's only for
2134 * scalars or vectors into vectors or matrices.
2135 *
2136 * With identical types on both sides and zero offset it's clone()
2137 * without creating a new object.
2138 */
2139
2140 void copy_offset(ir_constant *src, int offset);
2141
2142 /**
2143 * Copy the values on another constant at a given offset and
2144 * following an assign-like mask.
2145 *
2146 * The mask is ignored for scalars.
2147 *
2148 * Note that this function only handles what assign can handle,
2149 * i.e. at most a vector as source and a column of a matrix as
2150 * destination.
2151 */
2152
2153 void copy_masked_offset(ir_constant *src, int offset, unsigned int mask);
2154
2155 /**
2156 * Determine whether a constant has the same value as another constant
2157 *
2158 * \sa ir_constant::is_zero, ir_constant::is_one,
2159 * ir_constant::is_negative_one, ir_constant::is_basis
2160 */
2161 bool has_value(const ir_constant *) const;
2162
2163 /**
2164 * Return true if this ir_constant represents the given value.
2165 *
2166 * For vectors, this checks that each component is the given value.
2167 */
2168 virtual bool is_value(float f, int i) const;
2169 virtual bool is_zero() const;
2170 virtual bool is_one() const;
2171 virtual bool is_negative_one() const;
2172 virtual bool is_basis() const;
2173
2174 /**
2175 * Return true for constants that could be stored as 16-bit unsigned values.
2176 *
2177 * Note that this will return true even for signed integer ir_constants, as
2178 * long as the value is non-negative and fits in 16-bits.
2179 */
2180 virtual bool is_uint16_constant() const;
2181
2182 /**
2183 * Value of the constant.
2184 *
2185 * The field used to back the values supplied by the constant is determined
2186 * by the type associated with the \c ir_instruction. Constants may be
2187 * scalars, vectors, or matrices.
2188 */
2189 union ir_constant_data value;
2190
2191 /* Array elements */
2192 ir_constant **array_elements;
2193
2194 /* Structure fields */
2195 exec_list components;
2196
2197 private:
2198 /**
2199 * Parameterless constructor only used by the clone method
2200 */
2201 ir_constant(void);
2202 };
2203
2204 /**
2205 * IR instruction to emit a vertex in a geometry shader.
2206 */
2207 class ir_emit_vertex : public ir_instruction {
2208 public:
2209 ir_emit_vertex()
2210 {
2211 ir_type = ir_type_emit_vertex;
2212 }
2213
2214 virtual void accept(ir_visitor *v)
2215 {
2216 v->visit(this);
2217 }
2218
2219 virtual ir_emit_vertex *clone(void *mem_ctx, struct hash_table *) const
2220 {
2221 return new(mem_ctx) ir_emit_vertex();
2222 }
2223
2224 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2225 };
2226
2227 /**
2228 * IR instruction to complete the current primitive and start a new one in a
2229 * geometry shader.
2230 */
2231 class ir_end_primitive : public ir_instruction {
2232 public:
2233 ir_end_primitive()
2234 {
2235 ir_type = ir_type_end_primitive;
2236 }
2237
2238 virtual void accept(ir_visitor *v)
2239 {
2240 v->visit(this);
2241 }
2242
2243 virtual ir_end_primitive *clone(void *mem_ctx, struct hash_table *) const
2244 {
2245 return new(mem_ctx) ir_end_primitive();
2246 }
2247
2248 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2249 };
2250
2251 /*@}*/
2252
2253 /**
2254 * Apply a visitor to each IR node in a list
2255 */
2256 void
2257 visit_exec_list(exec_list *list, ir_visitor *visitor);
2258
2259 /**
2260 * Validate invariants on each IR node in a list
2261 */
2262 void validate_ir_tree(exec_list *instructions);
2263
2264 struct _mesa_glsl_parse_state;
2265 struct gl_shader_program;
2266
2267 /**
2268 * Detect whether an unlinked shader contains static recursion
2269 *
2270 * If the list of instructions is determined to contain static recursion,
2271 * \c _mesa_glsl_error will be called to emit error messages for each function
2272 * that is in the recursion cycle.
2273 */
2274 void
2275 detect_recursion_unlinked(struct _mesa_glsl_parse_state *state,
2276 exec_list *instructions);
2277
2278 /**
2279 * Detect whether a linked shader contains static recursion
2280 *
2281 * If the list of instructions is determined to contain static recursion,
2282 * \c link_error_printf will be called to emit error messages for each function
2283 * that is in the recursion cycle. In addition,
2284 * \c gl_shader_program::LinkStatus will be set to false.
2285 */
2286 void
2287 detect_recursion_linked(struct gl_shader_program *prog,
2288 exec_list *instructions);
2289
2290 /**
2291 * Make a clone of each IR instruction in a list
2292 *
2293 * \param in List of IR instructions that are to be cloned
2294 * \param out List to hold the cloned instructions
2295 */
2296 void
2297 clone_ir_list(void *mem_ctx, exec_list *out, const exec_list *in);
2298
2299 extern void
2300 _mesa_glsl_initialize_variables(exec_list *instructions,
2301 struct _mesa_glsl_parse_state *state);
2302
2303 extern void
2304 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state *state);
2305
2306 extern void
2307 _mesa_glsl_initialize_builtin_functions();
2308
2309 extern ir_function_signature *
2310 _mesa_glsl_find_builtin_function(_mesa_glsl_parse_state *state,
2311 const char *name, exec_list *actual_parameters);
2312
2313 extern gl_shader *
2314 _mesa_glsl_get_builtin_function_shader(void);
2315
2316 extern void
2317 _mesa_glsl_release_functions(void);
2318
2319 extern void
2320 _mesa_glsl_release_builtin_functions(void);
2321
2322 extern void
2323 reparent_ir(exec_list *list, void *mem_ctx);
2324
2325 struct glsl_symbol_table;
2326
2327 extern void
2328 import_prototypes(const exec_list *source, exec_list *dest,
2329 struct glsl_symbol_table *symbols, void *mem_ctx);
2330
2331 extern bool
2332 ir_has_call(ir_instruction *ir);
2333
2334 extern void
2335 do_set_program_inouts(exec_list *instructions, struct gl_program *prog,
2336 gl_shader_stage shader_stage);
2337
2338 extern char *
2339 prototype_string(const glsl_type *return_type, const char *name,
2340 exec_list *parameters);
2341
2342 const char *
2343 mode_string(const ir_variable *var);
2344
2345 extern "C" {
2346 #endif /* __cplusplus */
2347
2348 extern void _mesa_print_ir(FILE *f, struct exec_list *instructions,
2349 struct _mesa_glsl_parse_state *state);
2350
2351 #ifdef __cplusplus
2352 } /* extern "C" */
2353 #endif
2354
2355 unsigned
2356 vertices_per_prim(GLenum prim);
2357
2358 #endif /* IR_H */