glsl: add missing null check in tfeedback_decl::init()
[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 "util/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 ir_type_dereference_array,
63 ir_type_dereference_record,
64 ir_type_dereference_variable,
65 ir_type_constant,
66 ir_type_expression,
67 ir_type_swizzle,
68 ir_type_texture,
69 ir_type_variable,
70 ir_type_assignment,
71 ir_type_call,
72 ir_type_function,
73 ir_type_function_signature,
74 ir_type_if,
75 ir_type_loop,
76 ir_type_loop_jump,
77 ir_type_return,
78 ir_type_discard,
79 ir_type_emit_vertex,
80 ir_type_end_primitive,
81 ir_type_max, /**< maximum ir_type enum number, for validation */
82 ir_type_unset = ir_type_max
83 };
84
85
86 /**
87 * Base class of all IR instructions
88 */
89 class ir_instruction : public exec_node {
90 public:
91 enum ir_node_type ir_type;
92
93 /**
94 * GCC 4.7+ and clang warn when deleting an ir_instruction unless
95 * there's a virtual destructor present. Because we almost
96 * universally use ralloc for our memory management of
97 * ir_instructions, the destructor doesn't need to do any work.
98 */
99 virtual ~ir_instruction()
100 {
101 }
102
103 /** ir_print_visitor helper for debugging. */
104 void print(void) const;
105 void fprint(FILE *f) const;
106
107 virtual void accept(ir_visitor *) = 0;
108 virtual ir_visitor_status accept(ir_hierarchical_visitor *) = 0;
109 virtual ir_instruction *clone(void *mem_ctx,
110 struct hash_table *ht) const = 0;
111
112 /**
113 * \name IR instruction downcast functions
114 *
115 * These functions either cast the object to a derived class or return
116 * \c NULL if the object's type does not match the specified derived class.
117 * Additional downcast functions will be added as needed.
118 */
119 /*@{*/
120 class ir_rvalue *as_rvalue()
121 {
122 if (ir_type == ir_type_dereference_array ||
123 ir_type == ir_type_dereference_record ||
124 ir_type == ir_type_dereference_variable ||
125 ir_type == ir_type_constant ||
126 ir_type == ir_type_expression ||
127 ir_type == ir_type_swizzle ||
128 ir_type == ir_type_texture)
129 return (class ir_rvalue *) this;
130 return NULL;
131 }
132
133 class ir_dereference *as_dereference()
134 {
135 if (ir_type == ir_type_dereference_array ||
136 ir_type == ir_type_dereference_record ||
137 ir_type == ir_type_dereference_variable)
138 return (class ir_dereference *) this;
139 return NULL;
140 }
141
142 class ir_jump *as_jump()
143 {
144 if (ir_type == ir_type_loop_jump ||
145 ir_type == ir_type_return ||
146 ir_type == ir_type_discard)
147 return (class ir_jump *) this;
148 return NULL;
149 }
150
151 #define AS_CHILD(TYPE) \
152 class ir_##TYPE * as_##TYPE() \
153 { \
154 return ir_type == ir_type_##TYPE ? (ir_##TYPE *) this : NULL; \
155 }
156 AS_CHILD(variable)
157 AS_CHILD(function)
158 AS_CHILD(dereference_array)
159 AS_CHILD(dereference_variable)
160 AS_CHILD(dereference_record)
161 AS_CHILD(expression)
162 AS_CHILD(loop)
163 AS_CHILD(assignment)
164 AS_CHILD(call)
165 AS_CHILD(return)
166 AS_CHILD(if)
167 AS_CHILD(swizzle)
168 AS_CHILD(texture)
169 AS_CHILD(constant)
170 AS_CHILD(discard)
171 #undef AS_CHILD
172 /*@}*/
173
174 /**
175 * IR equality method: Return true if the referenced instruction would
176 * return the same value as this one.
177 *
178 * This intended to be used for CSE and algebraic optimizations, on rvalues
179 * in particular. No support for other instruction types (assignments,
180 * jumps, calls, etc.) is planned.
181 */
182 virtual bool equals(ir_instruction *ir, enum ir_node_type ignore = ir_type_unset);
183
184 protected:
185 ir_instruction(enum ir_node_type t)
186 : ir_type(t)
187 {
188 }
189
190 private:
191 ir_instruction()
192 {
193 assert(!"Should not get here.");
194 }
195 };
196
197
198 /**
199 * The base class for all "values"/expression trees.
200 */
201 class ir_rvalue : public ir_instruction {
202 public:
203 const struct glsl_type *type;
204
205 virtual ir_rvalue *clone(void *mem_ctx, struct hash_table *) const;
206
207 virtual void accept(ir_visitor *v)
208 {
209 v->visit(this);
210 }
211
212 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
213
214 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
215
216 ir_rvalue *as_rvalue_to_saturate();
217
218 virtual bool is_lvalue() const
219 {
220 return false;
221 }
222
223 /**
224 * Get the variable that is ultimately referenced by an r-value
225 */
226 virtual ir_variable *variable_referenced() const
227 {
228 return NULL;
229 }
230
231
232 /**
233 * If an r-value is a reference to a whole variable, get that variable
234 *
235 * \return
236 * Pointer to a variable that is completely dereferenced by the r-value. If
237 * the r-value is not a dereference or the dereference does not access the
238 * entire variable (i.e., it's just one array element, struct field), \c NULL
239 * is returned.
240 */
241 virtual ir_variable *whole_variable_referenced()
242 {
243 return NULL;
244 }
245
246 /**
247 * Determine if an r-value has the value zero
248 *
249 * The base implementation of this function always returns \c false. The
250 * \c ir_constant class over-rides this function to return \c true \b only
251 * for vector and scalar types that have all elements set to the value
252 * zero (or \c false for booleans).
253 *
254 * \sa ir_constant::has_value, ir_rvalue::is_one, ir_rvalue::is_negative_one,
255 * ir_constant::is_basis
256 */
257 virtual bool is_zero() const;
258
259 /**
260 * Determine if an r-value has the value one
261 *
262 * The base implementation of this function always returns \c false. The
263 * \c ir_constant class over-rides this function to return \c true \b only
264 * for vector and scalar types that have all elements set to the value
265 * one (or \c true for booleans).
266 *
267 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_negative_one,
268 * ir_constant::is_basis
269 */
270 virtual bool is_one() const;
271
272 /**
273 * Determine if an r-value has the value negative one
274 *
275 * The base implementation of this function always returns \c false. The
276 * \c ir_constant class over-rides this function to return \c true \b only
277 * for vector and scalar types that have all elements set to the value
278 * negative one. For boolean types, the result is always \c false.
279 *
280 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one
281 * ir_constant::is_basis
282 */
283 virtual bool is_negative_one() const;
284
285 /**
286 * Determine if an r-value is a basis vector
287 *
288 * The base implementation of this function always returns \c false. The
289 * \c ir_constant class over-rides this function to return \c true \b only
290 * for vector and scalar types that have one element set to the value one,
291 * and the other elements set to the value zero. For boolean types, the
292 * result is always \c false.
293 *
294 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one,
295 * is_constant::is_negative_one
296 */
297 virtual bool is_basis() const;
298
299 /**
300 * Determine if an r-value is an unsigned integer constant which can be
301 * stored in 16 bits.
302 *
303 * \sa ir_constant::is_uint16_constant.
304 */
305 virtual bool is_uint16_constant() const { return false; }
306
307 /**
308 * Return a generic value of error_type.
309 *
310 * Allocation will be performed with 'mem_ctx' as ralloc owner.
311 */
312 static ir_rvalue *error_value(void *mem_ctx);
313
314 protected:
315 ir_rvalue(enum ir_node_type t);
316 };
317
318
319 /**
320 * Variable storage classes
321 */
322 enum ir_variable_mode {
323 ir_var_auto = 0, /**< Function local variables and globals. */
324 ir_var_uniform, /**< Variable declared as a uniform. */
325 ir_var_shader_in,
326 ir_var_shader_out,
327 ir_var_function_in,
328 ir_var_function_out,
329 ir_var_function_inout,
330 ir_var_const_in, /**< "in" param that must be a constant expression */
331 ir_var_system_value, /**< Ex: front-face, instance-id, etc. */
332 ir_var_temporary, /**< Temporary variable generated during compilation. */
333 ir_var_mode_count /**< Number of variable modes */
334 };
335
336 /**
337 * Enum keeping track of how a variable was declared. For error checking of
338 * the gl_PerVertex redeclaration rules.
339 */
340 enum ir_var_declaration_type {
341 /**
342 * Normal declaration (for most variables, this means an explicit
343 * declaration. Exception: temporaries are always implicitly declared, but
344 * they still use ir_var_declared_normally).
345 *
346 * Note: an ir_variable that represents a named interface block uses
347 * ir_var_declared_normally.
348 */
349 ir_var_declared_normally = 0,
350
351 /**
352 * Variable was explicitly declared (or re-declared) in an unnamed
353 * interface block.
354 */
355 ir_var_declared_in_block,
356
357 /**
358 * Variable is an implicitly declared built-in that has not been explicitly
359 * re-declared by the shader.
360 */
361 ir_var_declared_implicitly,
362 };
363
364 /**
365 * \brief Layout qualifiers for gl_FragDepth.
366 *
367 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
368 * with a layout qualifier.
369 */
370 enum ir_depth_layout {
371 ir_depth_layout_none, /**< No depth layout is specified. */
372 ir_depth_layout_any,
373 ir_depth_layout_greater,
374 ir_depth_layout_less,
375 ir_depth_layout_unchanged
376 };
377
378 /**
379 * \brief Convert depth layout qualifier to string.
380 */
381 const char*
382 depth_layout_string(ir_depth_layout layout);
383
384 /**
385 * Description of built-in state associated with a uniform
386 *
387 * \sa ir_variable::state_slots
388 */
389 struct ir_state_slot {
390 int tokens[5];
391 int swizzle;
392 };
393
394
395 /**
396 * Get the string value for an interpolation qualifier
397 *
398 * \return The string that would be used in a shader to specify \c
399 * mode will be returned.
400 *
401 * This function is used to generate error messages of the form "shader
402 * uses %s interpolation qualifier", so in the case where there is no
403 * interpolation qualifier, it returns "no".
404 *
405 * This function should only be used on a shader input or output variable.
406 */
407 const char *interpolation_string(unsigned interpolation);
408
409
410 class ir_variable : public ir_instruction {
411 public:
412 ir_variable(const struct glsl_type *, const char *, ir_variable_mode);
413
414 virtual ir_variable *clone(void *mem_ctx, struct hash_table *ht) const;
415
416 virtual void accept(ir_visitor *v)
417 {
418 v->visit(this);
419 }
420
421 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
422
423
424 /**
425 * Determine how this variable should be interpolated based on its
426 * interpolation qualifier (if present), whether it is gl_Color or
427 * gl_SecondaryColor, and whether flatshading is enabled in the current GL
428 * state.
429 *
430 * The return value will always be either INTERP_QUALIFIER_SMOOTH,
431 * INTERP_QUALIFIER_NOPERSPECTIVE, or INTERP_QUALIFIER_FLAT.
432 */
433 glsl_interp_qualifier determine_interpolation_mode(bool flat_shade);
434
435 /**
436 * Determine whether or not a variable is part of a uniform block.
437 */
438 inline bool is_in_uniform_block() const
439 {
440 return this->data.mode == ir_var_uniform && this->interface_type != NULL;
441 }
442
443 /**
444 * Determine whether or not a variable is the declaration of an interface
445 * block
446 *
447 * For the first declaration below, there will be an \c ir_variable named
448 * "instance" whose type and whose instance_type will be the same
449 * \cglsl_type. For the second declaration, there will be an \c ir_variable
450 * named "f" whose type is float and whose instance_type is B2.
451 *
452 * "instance" is an interface instance variable, but "f" is not.
453 *
454 * uniform B1 {
455 * float f;
456 * } instance;
457 *
458 * uniform B2 {
459 * float f;
460 * };
461 */
462 inline bool is_interface_instance() const
463 {
464 const glsl_type *const t = this->type;
465
466 return (t == this->interface_type)
467 || (t->is_array() && t->fields.array == this->interface_type);
468 }
469
470 /**
471 * Set this->interface_type on a newly created variable.
472 */
473 void init_interface_type(const struct glsl_type *type)
474 {
475 assert(this->interface_type == NULL);
476 this->interface_type = type;
477 if (this->is_interface_instance()) {
478 this->max_ifc_array_access =
479 rzalloc_array(this, unsigned, type->length);
480 }
481 }
482
483 /**
484 * Change this->interface_type on a variable that previously had a
485 * different, but compatible, interface_type. This is used during linking
486 * to set the size of arrays in interface blocks.
487 */
488 void change_interface_type(const struct glsl_type *type)
489 {
490 if (this->max_ifc_array_access != NULL) {
491 /* max_ifc_array_access has already been allocated, so make sure the
492 * new interface has the same number of fields as the old one.
493 */
494 assert(this->interface_type->length == type->length);
495 }
496 this->interface_type = type;
497 }
498
499 /**
500 * Change this->interface_type on a variable that previously had a
501 * different, and incompatible, interface_type. This is used during
502 * compilation to handle redeclaration of the built-in gl_PerVertex
503 * interface block.
504 */
505 void reinit_interface_type(const struct glsl_type *type)
506 {
507 if (this->max_ifc_array_access != NULL) {
508 #ifndef NDEBUG
509 /* Redeclaring gl_PerVertex is only allowed if none of the built-ins
510 * it defines have been accessed yet; so it's safe to throw away the
511 * old max_ifc_array_access pointer, since all of its values are
512 * zero.
513 */
514 for (unsigned i = 0; i < this->interface_type->length; i++)
515 assert(this->max_ifc_array_access[i] == 0);
516 #endif
517 ralloc_free(this->max_ifc_array_access);
518 this->max_ifc_array_access = NULL;
519 }
520 this->interface_type = NULL;
521 init_interface_type(type);
522 }
523
524 const glsl_type *get_interface_type() const
525 {
526 return this->interface_type;
527 }
528
529 /**
530 * Declared type of the variable
531 */
532 const struct glsl_type *type;
533
534 /**
535 * Declared name of the variable
536 */
537 const char *name;
538
539 /**
540 * For variables which satisfy the is_interface_instance() predicate, this
541 * points to an array of integers such that if the ith member of the
542 * interface block is an array, max_ifc_array_access[i] is the maximum
543 * array element of that member that has been accessed. If the ith member
544 * of the interface block is not an array, max_ifc_array_access[i] is
545 * unused.
546 *
547 * For variables whose type is not an interface block, this pointer is
548 * NULL.
549 */
550 unsigned *max_ifc_array_access;
551
552 struct ir_variable_data {
553
554 /**
555 * Is the variable read-only?
556 *
557 * This is set for variables declared as \c const, shader inputs,
558 * and uniforms.
559 */
560 unsigned read_only:1;
561 unsigned centroid:1;
562 unsigned sample:1;
563 unsigned invariant:1;
564 unsigned precise:1;
565
566 /**
567 * Has this variable been used for reading or writing?
568 *
569 * Several GLSL semantic checks require knowledge of whether or not a
570 * variable has been used. For example, it is an error to redeclare a
571 * variable as invariant after it has been used.
572 *
573 * This is only maintained in the ast_to_hir.cpp path, not in
574 * Mesa's fixed function or ARB program paths.
575 */
576 unsigned used:1;
577
578 /**
579 * Has this variable been statically assigned?
580 *
581 * This answers whether the variable was assigned in any path of
582 * the shader during ast_to_hir. This doesn't answer whether it is
583 * still written after dead code removal, nor is it maintained in
584 * non-ast_to_hir.cpp (GLSL parsing) paths.
585 */
586 unsigned assigned:1;
587
588 /**
589 * Enum indicating how the variable was declared. See
590 * ir_var_declaration_type.
591 *
592 * This is used to detect certain kinds of illegal variable redeclarations.
593 */
594 unsigned how_declared:2;
595
596 /**
597 * Storage class of the variable.
598 *
599 * \sa ir_variable_mode
600 */
601 unsigned mode:4;
602
603 /**
604 * Interpolation mode for shader inputs / outputs
605 *
606 * \sa ir_variable_interpolation
607 */
608 unsigned interpolation:2;
609
610 /**
611 * \name ARB_fragment_coord_conventions
612 * @{
613 */
614 unsigned origin_upper_left:1;
615 unsigned pixel_center_integer:1;
616 /*@}*/
617
618 /**
619 * Was the location explicitly set in the shader?
620 *
621 * If the location is explicitly set in the shader, it \b cannot be changed
622 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
623 * no effect).
624 */
625 unsigned explicit_location:1;
626 unsigned explicit_index:1;
627
628 /**
629 * Was an initial binding explicitly set in the shader?
630 *
631 * If so, constant_value contains an integer ir_constant representing the
632 * initial binding point.
633 */
634 unsigned explicit_binding:1;
635
636 /**
637 * Does this variable have an initializer?
638 *
639 * This is used by the linker to cross-validiate initializers of global
640 * variables.
641 */
642 unsigned has_initializer:1;
643
644 /**
645 * Is this variable a generic output or input that has not yet been matched
646 * up to a variable in another stage of the pipeline?
647 *
648 * This is used by the linker as scratch storage while assigning locations
649 * to generic inputs and outputs.
650 */
651 unsigned is_unmatched_generic_inout:1;
652
653 /**
654 * If non-zero, then this variable may be packed along with other variables
655 * into a single varying slot, so this offset should be applied when
656 * accessing components. For example, an offset of 1 means that the x
657 * component of this variable is actually stored in component y of the
658 * location specified by \c location.
659 */
660 unsigned location_frac:2;
661
662 /**
663 * Layout of the matrix. Uses glsl_matrix_layout values.
664 */
665 unsigned matrix_layout:2;
666
667 /**
668 * Non-zero if this variable was created by lowering a named interface
669 * block which was not an array.
670 *
671 * Note that this variable and \c from_named_ifc_block_array will never
672 * both be non-zero.
673 */
674 unsigned from_named_ifc_block_nonarray:1;
675
676 /**
677 * Non-zero if this variable was created by lowering a named interface
678 * block which was an array.
679 *
680 * Note that this variable and \c from_named_ifc_block_nonarray will never
681 * both be non-zero.
682 */
683 unsigned from_named_ifc_block_array:1;
684
685 /**
686 * Non-zero if the variable must be a shader input. This is useful for
687 * constraints on function parameters.
688 */
689 unsigned must_be_shader_input:1;
690
691 /**
692 * Output index for dual source blending.
693 *
694 * \note
695 * The GLSL spec only allows the values 0 or 1 for the index in \b dual
696 * source blending.
697 */
698 unsigned index:1;
699
700
701 /**
702 * ARB_shader_image_load_store qualifiers.
703 */
704 unsigned image_read_only:1; /**< "readonly" qualifier. */
705 unsigned image_write_only:1; /**< "writeonly" qualifier. */
706 unsigned image_coherent:1;
707 unsigned image_volatile:1;
708 unsigned image_restrict:1;
709
710 /** Image internal format if specified explicitly, otherwise GL_NONE. */
711 uint16_t image_format;
712
713 /**
714 * \brief Layout qualifier for gl_FragDepth.
715 *
716 * This is not equal to \c ir_depth_layout_none if and only if this
717 * variable is \c gl_FragDepth and a layout qualifier is specified.
718 */
719 ir_depth_layout depth_layout;
720
721 /**
722 * Storage location of the base of this variable
723 *
724 * The precise meaning of this field depends on the nature of the variable.
725 *
726 * - Vertex shader input: one of the values from \c gl_vert_attrib.
727 * - Vertex shader output: one of the values from \c gl_varying_slot.
728 * - Geometry shader input: one of the values from \c gl_varying_slot.
729 * - Geometry shader output: one of the values from \c gl_varying_slot.
730 * - Fragment shader input: one of the values from \c gl_varying_slot.
731 * - Fragment shader output: one of the values from \c gl_frag_result.
732 * - Uniforms: Per-stage uniform slot number for default uniform block.
733 * - Uniforms: Index within the uniform block definition for UBO members.
734 * - Other: This field is not currently used.
735 *
736 * If the variable is a uniform, shader input, or shader output, and the
737 * slot has not been assigned, the value will be -1.
738 */
739 int location;
740
741 /**
742 * Vertex stream output identifier.
743 */
744 unsigned stream;
745
746 /**
747 * Initial binding point for a sampler, atomic, or UBO.
748 *
749 * For array types, this represents the binding point for the first element.
750 */
751 int binding;
752
753 /**
754 * Location an atomic counter is stored at.
755 */
756 struct {
757 unsigned offset;
758 } atomic;
759
760 /**
761 * Highest element accessed with a constant expression array index
762 *
763 * Not used for non-array variables.
764 */
765 unsigned max_array_access;
766
767 } data;
768
769 /**
770 * Built-in state that backs this uniform
771 *
772 * Once set at variable creation, \c state_slots must remain invariant.
773 * This is because, ideally, this array would be shared by all clones of
774 * this variable in the IR tree. In other words, we'd really like for it
775 * to be a fly-weight.
776 *
777 * If the variable is not a uniform, \c num_state_slots will be zero and
778 * \c state_slots will be \c NULL.
779 */
780 /*@{*/
781 unsigned num_state_slots; /**< Number of state slots used */
782 ir_state_slot *state_slots; /**< State descriptors. */
783 /*@}*/
784
785 /**
786 * Emit a warning if this variable is accessed.
787 */
788 const char *warn_extension;
789
790 /**
791 * Value assigned in the initializer of a variable declared "const"
792 */
793 ir_constant *constant_value;
794
795 /**
796 * Constant expression assigned in the initializer of the variable
797 *
798 * \warning
799 * This field and \c ::constant_value are distinct. Even if the two fields
800 * refer to constants with the same value, they must point to separate
801 * objects.
802 */
803 ir_constant *constant_initializer;
804
805 private:
806 /**
807 * For variables that are in an interface block or are an instance of an
808 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
809 *
810 * \sa ir_variable::location
811 */
812 const glsl_type *interface_type;
813 };
814
815 /**
816 * A function that returns whether a built-in function is available in the
817 * current shading language (based on version, ES or desktop, and extensions).
818 */
819 typedef bool (*builtin_available_predicate)(const _mesa_glsl_parse_state *);
820
821 /*@{*/
822 /**
823 * The representation of a function instance; may be the full definition or
824 * simply a prototype.
825 */
826 class ir_function_signature : public ir_instruction {
827 /* An ir_function_signature will be part of the list of signatures in
828 * an ir_function.
829 */
830 public:
831 ir_function_signature(const glsl_type *return_type,
832 builtin_available_predicate builtin_avail = NULL);
833
834 virtual ir_function_signature *clone(void *mem_ctx,
835 struct hash_table *ht) const;
836 ir_function_signature *clone_prototype(void *mem_ctx,
837 struct hash_table *ht) const;
838
839 virtual void accept(ir_visitor *v)
840 {
841 v->visit(this);
842 }
843
844 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
845
846 /**
847 * Attempt to evaluate this function as a constant expression,
848 * given a list of the actual parameters and the variable context.
849 * Returns NULL for non-built-ins.
850 */
851 ir_constant *constant_expression_value(exec_list *actual_parameters, struct hash_table *variable_context);
852
853 /**
854 * Get the name of the function for which this is a signature
855 */
856 const char *function_name() const;
857
858 /**
859 * Get a handle to the function for which this is a signature
860 *
861 * There is no setter function, this function returns a \c const pointer,
862 * and \c ir_function_signature::_function is private for a reason. The
863 * only way to make a connection between a function and function signature
864 * is via \c ir_function::add_signature. This helps ensure that certain
865 * invariants (i.e., a function signature is in the list of signatures for
866 * its \c _function) are met.
867 *
868 * \sa ir_function::add_signature
869 */
870 inline const class ir_function *function() const
871 {
872 return this->_function;
873 }
874
875 /**
876 * Check whether the qualifiers match between this signature's parameters
877 * and the supplied parameter list. If not, returns the name of the first
878 * parameter with mismatched qualifiers (for use in error messages).
879 */
880 const char *qualifiers_match(exec_list *params);
881
882 /**
883 * Replace the current parameter list with the given one. This is useful
884 * if the current information came from a prototype, and either has invalid
885 * or missing parameter names.
886 */
887 void replace_parameters(exec_list *new_params);
888
889 /**
890 * Function return type.
891 *
892 * \note This discards the optional precision qualifier.
893 */
894 const struct glsl_type *return_type;
895
896 /**
897 * List of ir_variable of function parameters.
898 *
899 * This represents the storage. The paramaters passed in a particular
900 * call will be in ir_call::actual_paramaters.
901 */
902 struct exec_list parameters;
903
904 /** Whether or not this function has a body (which may be empty). */
905 unsigned is_defined:1;
906
907 /** Whether or not this function signature is a built-in. */
908 bool is_builtin() const;
909
910 /**
911 * Whether or not this function is an intrinsic to be implemented
912 * by the driver.
913 */
914 bool is_intrinsic;
915
916 /** Whether or not a built-in is available for this shader. */
917 bool is_builtin_available(const _mesa_glsl_parse_state *state) const;
918
919 /** Body of instructions in the function. */
920 struct exec_list body;
921
922 private:
923 /**
924 * A function pointer to a predicate that answers whether a built-in
925 * function is available in the current shader. NULL if not a built-in.
926 */
927 builtin_available_predicate builtin_avail;
928
929 /** Function of which this signature is one overload. */
930 class ir_function *_function;
931
932 /** Function signature of which this one is a prototype clone */
933 const ir_function_signature *origin;
934
935 friend class ir_function;
936
937 /**
938 * Helper function to run a list of instructions for constant
939 * expression evaluation.
940 *
941 * The hash table represents the values of the visible variables.
942 * There are no scoping issues because the table is indexed on
943 * ir_variable pointers, not variable names.
944 *
945 * Returns false if the expression is not constant, true otherwise,
946 * and the value in *result if result is non-NULL.
947 */
948 bool constant_expression_evaluate_expression_list(const struct exec_list &body,
949 struct hash_table *variable_context,
950 ir_constant **result);
951 };
952
953
954 /**
955 * Header for tracking multiple overloaded functions with the same name.
956 * Contains a list of ir_function_signatures representing each of the
957 * actual functions.
958 */
959 class ir_function : public ir_instruction {
960 public:
961 ir_function(const char *name);
962
963 virtual ir_function *clone(void *mem_ctx, struct hash_table *ht) const;
964
965 virtual void accept(ir_visitor *v)
966 {
967 v->visit(this);
968 }
969
970 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
971
972 void add_signature(ir_function_signature *sig)
973 {
974 sig->_function = this;
975 this->signatures.push_tail(sig);
976 }
977
978 /**
979 * Find a signature that matches a set of actual parameters, taking implicit
980 * conversions into account. Also flags whether the match was exact.
981 */
982 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
983 const exec_list *actual_param,
984 bool allow_builtins,
985 bool *match_is_exact);
986
987 /**
988 * Find a signature that matches a set of actual parameters, taking implicit
989 * conversions into account.
990 */
991 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
992 const exec_list *actual_param,
993 bool allow_builtins);
994
995 /**
996 * Find a signature that exactly matches a set of actual parameters without
997 * any implicit type conversions.
998 */
999 ir_function_signature *exact_matching_signature(_mesa_glsl_parse_state *state,
1000 const exec_list *actual_ps);
1001
1002 /**
1003 * Name of the function.
1004 */
1005 const char *name;
1006
1007 /** Whether or not this function has a signature that isn't a built-in. */
1008 bool has_user_signature();
1009
1010 /**
1011 * List of ir_function_signature for each overloaded function with this name.
1012 */
1013 struct exec_list signatures;
1014 };
1015
1016 inline const char *ir_function_signature::function_name() const
1017 {
1018 return this->_function->name;
1019 }
1020 /*@}*/
1021
1022
1023 /**
1024 * IR instruction representing high-level if-statements
1025 */
1026 class ir_if : public ir_instruction {
1027 public:
1028 ir_if(ir_rvalue *condition)
1029 : ir_instruction(ir_type_if), condition(condition)
1030 {
1031 }
1032
1033 virtual ir_if *clone(void *mem_ctx, struct hash_table *ht) const;
1034
1035 virtual void accept(ir_visitor *v)
1036 {
1037 v->visit(this);
1038 }
1039
1040 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1041
1042 ir_rvalue *condition;
1043 /** List of ir_instruction for the body of the then branch */
1044 exec_list then_instructions;
1045 /** List of ir_instruction for the body of the else branch */
1046 exec_list else_instructions;
1047 };
1048
1049
1050 /**
1051 * IR instruction representing a high-level loop structure.
1052 */
1053 class ir_loop : public ir_instruction {
1054 public:
1055 ir_loop();
1056
1057 virtual ir_loop *clone(void *mem_ctx, struct hash_table *ht) const;
1058
1059 virtual void accept(ir_visitor *v)
1060 {
1061 v->visit(this);
1062 }
1063
1064 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1065
1066 /** List of ir_instruction that make up the body of the loop. */
1067 exec_list body_instructions;
1068 };
1069
1070
1071 class ir_assignment : public ir_instruction {
1072 public:
1073 ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs, ir_rvalue *condition = NULL);
1074
1075 /**
1076 * Construct an assignment with an explicit write mask
1077 *
1078 * \note
1079 * Since a write mask is supplied, the LHS must already be a bare
1080 * \c ir_dereference. The cannot be any swizzles in the LHS.
1081 */
1082 ir_assignment(ir_dereference *lhs, ir_rvalue *rhs, ir_rvalue *condition,
1083 unsigned write_mask);
1084
1085 virtual ir_assignment *clone(void *mem_ctx, struct hash_table *ht) const;
1086
1087 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1088
1089 virtual void accept(ir_visitor *v)
1090 {
1091 v->visit(this);
1092 }
1093
1094 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1095
1096 /**
1097 * Get a whole variable written by an assignment
1098 *
1099 * If the LHS of the assignment writes a whole variable, the variable is
1100 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1101 * assignment are:
1102 *
1103 * - Assigning to a scalar
1104 * - Assigning to all components of a vector
1105 * - Whole array (or matrix) assignment
1106 * - Whole structure assignment
1107 */
1108 ir_variable *whole_variable_written();
1109
1110 /**
1111 * Set the LHS of an assignment
1112 */
1113 void set_lhs(ir_rvalue *lhs);
1114
1115 /**
1116 * Left-hand side of the assignment.
1117 *
1118 * This should be treated as read only. If you need to set the LHS of an
1119 * assignment, use \c ir_assignment::set_lhs.
1120 */
1121 ir_dereference *lhs;
1122
1123 /**
1124 * Value being assigned
1125 */
1126 ir_rvalue *rhs;
1127
1128 /**
1129 * Optional condition for the assignment.
1130 */
1131 ir_rvalue *condition;
1132
1133
1134 /**
1135 * Component mask written
1136 *
1137 * For non-vector types in the LHS, this field will be zero. For vector
1138 * types, a bit will be set for each component that is written. Note that
1139 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1140 *
1141 * A partially-set write mask means that each enabled channel gets
1142 * the value from a consecutive channel of the rhs. For example,
1143 * to write just .xyw of gl_FrontColor with color:
1144 *
1145 * (assign (constant bool (1)) (xyw)
1146 * (var_ref gl_FragColor)
1147 * (swiz xyw (var_ref color)))
1148 */
1149 unsigned write_mask:4;
1150 };
1151
1152 /* Update ir_expression::get_num_operands() and operator_strs when
1153 * updating this list.
1154 */
1155 enum ir_expression_operation {
1156 ir_unop_bit_not,
1157 ir_unop_logic_not,
1158 ir_unop_neg,
1159 ir_unop_abs,
1160 ir_unop_sign,
1161 ir_unop_rcp,
1162 ir_unop_rsq,
1163 ir_unop_sqrt,
1164 ir_unop_exp, /**< Log base e on gentype */
1165 ir_unop_log, /**< Natural log on gentype */
1166 ir_unop_exp2,
1167 ir_unop_log2,
1168 ir_unop_f2i, /**< Float-to-integer conversion. */
1169 ir_unop_f2u, /**< Float-to-unsigned conversion. */
1170 ir_unop_i2f, /**< Integer-to-float conversion. */
1171 ir_unop_f2b, /**< Float-to-boolean conversion */
1172 ir_unop_b2f, /**< Boolean-to-float conversion */
1173 ir_unop_i2b, /**< int-to-boolean conversion */
1174 ir_unop_b2i, /**< Boolean-to-int conversion */
1175 ir_unop_u2f, /**< Unsigned-to-float conversion. */
1176 ir_unop_i2u, /**< Integer-to-unsigned conversion. */
1177 ir_unop_u2i, /**< Unsigned-to-integer conversion. */
1178 ir_unop_bitcast_i2f, /**< Bit-identical int-to-float "conversion" */
1179 ir_unop_bitcast_f2i, /**< Bit-identical float-to-int "conversion" */
1180 ir_unop_bitcast_u2f, /**< Bit-identical uint-to-float "conversion" */
1181 ir_unop_bitcast_f2u, /**< Bit-identical float-to-uint "conversion" */
1182 ir_unop_any,
1183
1184 /**
1185 * \name Unary floating-point rounding operations.
1186 */
1187 /*@{*/
1188 ir_unop_trunc,
1189 ir_unop_ceil,
1190 ir_unop_floor,
1191 ir_unop_fract,
1192 ir_unop_round_even,
1193 /*@}*/
1194
1195 /**
1196 * \name Trigonometric operations.
1197 */
1198 /*@{*/
1199 ir_unop_sin,
1200 ir_unop_cos,
1201 ir_unop_sin_reduced, /**< Reduced range sin. [-pi, pi] */
1202 ir_unop_cos_reduced, /**< Reduced range cos. [-pi, pi] */
1203 /*@}*/
1204
1205 /**
1206 * \name Partial derivatives.
1207 */
1208 /*@{*/
1209 ir_unop_dFdx,
1210 ir_unop_dFdx_coarse,
1211 ir_unop_dFdx_fine,
1212 ir_unop_dFdy,
1213 ir_unop_dFdy_coarse,
1214 ir_unop_dFdy_fine,
1215 /*@}*/
1216
1217 /**
1218 * \name Floating point pack and unpack operations.
1219 */
1220 /*@{*/
1221 ir_unop_pack_snorm_2x16,
1222 ir_unop_pack_snorm_4x8,
1223 ir_unop_pack_unorm_2x16,
1224 ir_unop_pack_unorm_4x8,
1225 ir_unop_pack_half_2x16,
1226 ir_unop_unpack_snorm_2x16,
1227 ir_unop_unpack_snorm_4x8,
1228 ir_unop_unpack_unorm_2x16,
1229 ir_unop_unpack_unorm_4x8,
1230 ir_unop_unpack_half_2x16,
1231 /*@}*/
1232
1233 /**
1234 * \name Lowered floating point unpacking operations.
1235 *
1236 * \see lower_packing_builtins_visitor::split_unpack_half_2x16
1237 */
1238 /*@{*/
1239 ir_unop_unpack_half_2x16_split_x,
1240 ir_unop_unpack_half_2x16_split_y,
1241 /*@}*/
1242
1243 /**
1244 * \name Bit operations, part of ARB_gpu_shader5.
1245 */
1246 /*@{*/
1247 ir_unop_bitfield_reverse,
1248 ir_unop_bit_count,
1249 ir_unop_find_msb,
1250 ir_unop_find_lsb,
1251 /*@}*/
1252
1253 ir_unop_saturate,
1254 ir_unop_noise,
1255
1256 /**
1257 * Interpolate fs input at centroid
1258 *
1259 * operand0 is the fs input.
1260 */
1261 ir_unop_interpolate_at_centroid,
1262
1263 /**
1264 * A sentinel marking the last of the unary operations.
1265 */
1266 ir_last_unop = ir_unop_interpolate_at_centroid,
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 * Interpolate fs input at offset
1386 *
1387 * operand0 is the fs input
1388 * operand1 is the offset from the pixel center
1389 */
1390 ir_binop_interpolate_at_offset,
1391
1392 /**
1393 * Interpolate fs input at sample position
1394 *
1395 * operand0 is the fs input
1396 * operand1 is the sample ID
1397 */
1398 ir_binop_interpolate_at_sample,
1399
1400 /**
1401 * A sentinel marking the last of the binary operations.
1402 */
1403 ir_last_binop = ir_binop_interpolate_at_sample,
1404
1405 /**
1406 * \name Fused floating-point multiply-add, part of ARB_gpu_shader5.
1407 */
1408 /*@{*/
1409 ir_triop_fma,
1410 /*@}*/
1411
1412 ir_triop_lrp,
1413
1414 /**
1415 * \name Conditional Select
1416 *
1417 * A vector conditional select instruction (like ?:, but operating per-
1418 * component on vectors).
1419 *
1420 * \see lower_instructions_visitor::ldexp_to_arith
1421 */
1422 /*@{*/
1423 ir_triop_csel,
1424 /*@}*/
1425
1426 /**
1427 * \name Second half of a lowered bitfieldInsert() operation.
1428 *
1429 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1430 */
1431 /*@{*/
1432 ir_triop_bfi,
1433 /*@}*/
1434
1435 ir_triop_bitfield_extract,
1436
1437 /**
1438 * Generate a value with one field of a vector changed
1439 *
1440 * operand0 is the vector
1441 * operand1 is the value to write into the vector result
1442 * operand2 is the index in operand0 to be modified
1443 */
1444 ir_triop_vector_insert,
1445
1446 /**
1447 * A sentinel marking the last of the ternary operations.
1448 */
1449 ir_last_triop = ir_triop_vector_insert,
1450
1451 ir_quadop_bitfield_insert,
1452
1453 ir_quadop_vector,
1454
1455 /**
1456 * A sentinel marking the last of the ternary operations.
1457 */
1458 ir_last_quadop = ir_quadop_vector,
1459
1460 /**
1461 * A sentinel marking the last of all operations.
1462 */
1463 ir_last_opcode = ir_quadop_vector
1464 };
1465
1466 class ir_expression : public ir_rvalue {
1467 public:
1468 ir_expression(int op, const struct glsl_type *type,
1469 ir_rvalue *op0, ir_rvalue *op1 = NULL,
1470 ir_rvalue *op2 = NULL, ir_rvalue *op3 = NULL);
1471
1472 /**
1473 * Constructor for unary operation expressions
1474 */
1475 ir_expression(int op, ir_rvalue *);
1476
1477 /**
1478 * Constructor for binary operation expressions
1479 */
1480 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1);
1481
1482 /**
1483 * Constructor for ternary operation expressions
1484 */
1485 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1, ir_rvalue *op2);
1486
1487 virtual bool equals(ir_instruction *ir, enum ir_node_type ignore = ir_type_unset);
1488
1489 virtual ir_expression *clone(void *mem_ctx, struct hash_table *ht) const;
1490
1491 /**
1492 * Attempt to constant-fold the expression
1493 *
1494 * The "variable_context" hash table links ir_variable * to ir_constant *
1495 * that represent the variables' values. \c NULL represents an empty
1496 * context.
1497 *
1498 * If the expression cannot be constant folded, this method will return
1499 * \c NULL.
1500 */
1501 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1502
1503 /**
1504 * Determine the number of operands used by an expression
1505 */
1506 static unsigned int get_num_operands(ir_expression_operation);
1507
1508 /**
1509 * Determine the number of operands used by an expression
1510 */
1511 unsigned int get_num_operands() const
1512 {
1513 return (this->operation == ir_quadop_vector)
1514 ? this->type->vector_elements : get_num_operands(operation);
1515 }
1516
1517 /**
1518 * Return whether the expression operates on vectors horizontally.
1519 */
1520 bool is_horizontal() const
1521 {
1522 return operation == ir_binop_all_equal ||
1523 operation == ir_binop_any_nequal ||
1524 operation == ir_unop_any ||
1525 operation == ir_binop_dot ||
1526 operation == ir_quadop_vector;
1527 }
1528
1529 /**
1530 * Return a string representing this expression's operator.
1531 */
1532 const char *operator_string();
1533
1534 /**
1535 * Return a string representing this expression's operator.
1536 */
1537 static const char *operator_string(ir_expression_operation);
1538
1539
1540 /**
1541 * Do a reverse-lookup to translate the given string into an operator.
1542 */
1543 static ir_expression_operation get_operator(const char *);
1544
1545 virtual void accept(ir_visitor *v)
1546 {
1547 v->visit(this);
1548 }
1549
1550 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1551
1552 ir_expression_operation operation;
1553 ir_rvalue *operands[4];
1554 };
1555
1556
1557 /**
1558 * HIR instruction representing a high-level function call, containing a list
1559 * of parameters and returning a value in the supplied temporary.
1560 */
1561 class ir_call : public ir_instruction {
1562 public:
1563 ir_call(ir_function_signature *callee,
1564 ir_dereference_variable *return_deref,
1565 exec_list *actual_parameters)
1566 : ir_instruction(ir_type_call), return_deref(return_deref), callee(callee)
1567 {
1568 assert(callee->return_type != NULL);
1569 actual_parameters->move_nodes_to(& this->actual_parameters);
1570 this->use_builtin = callee->is_builtin();
1571 }
1572
1573 virtual ir_call *clone(void *mem_ctx, struct hash_table *ht) const;
1574
1575 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1576
1577 virtual void accept(ir_visitor *v)
1578 {
1579 v->visit(this);
1580 }
1581
1582 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1583
1584 /**
1585 * Get the name of the function being called.
1586 */
1587 const char *callee_name() const
1588 {
1589 return callee->function_name();
1590 }
1591
1592 /**
1593 * Generates an inline version of the function before @ir,
1594 * storing the return value in return_deref.
1595 */
1596 void generate_inline(ir_instruction *ir);
1597
1598 /**
1599 * Storage for the function's return value.
1600 * This must be NULL if the return type is void.
1601 */
1602 ir_dereference_variable *return_deref;
1603
1604 /**
1605 * The specific function signature being called.
1606 */
1607 ir_function_signature *callee;
1608
1609 /* List of ir_rvalue of paramaters passed in this call. */
1610 exec_list actual_parameters;
1611
1612 /** Should this call only bind to a built-in function? */
1613 bool use_builtin;
1614 };
1615
1616
1617 /**
1618 * \name Jump-like IR instructions.
1619 *
1620 * These include \c break, \c continue, \c return, and \c discard.
1621 */
1622 /*@{*/
1623 class ir_jump : public ir_instruction {
1624 protected:
1625 ir_jump(enum ir_node_type t)
1626 : ir_instruction(t)
1627 {
1628 }
1629 };
1630
1631 class ir_return : public ir_jump {
1632 public:
1633 ir_return()
1634 : ir_jump(ir_type_return), value(NULL)
1635 {
1636 }
1637
1638 ir_return(ir_rvalue *value)
1639 : ir_jump(ir_type_return), value(value)
1640 {
1641 }
1642
1643 virtual ir_return *clone(void *mem_ctx, struct hash_table *) const;
1644
1645 ir_rvalue *get_value() const
1646 {
1647 return value;
1648 }
1649
1650 virtual void accept(ir_visitor *v)
1651 {
1652 v->visit(this);
1653 }
1654
1655 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1656
1657 ir_rvalue *value;
1658 };
1659
1660
1661 /**
1662 * Jump instructions used inside loops
1663 *
1664 * These include \c break and \c continue. The \c break within a loop is
1665 * different from the \c break within a switch-statement.
1666 *
1667 * \sa ir_switch_jump
1668 */
1669 class ir_loop_jump : public ir_jump {
1670 public:
1671 enum jump_mode {
1672 jump_break,
1673 jump_continue
1674 };
1675
1676 ir_loop_jump(jump_mode mode)
1677 : ir_jump(ir_type_loop_jump)
1678 {
1679 this->mode = mode;
1680 }
1681
1682 virtual ir_loop_jump *clone(void *mem_ctx, struct hash_table *) const;
1683
1684 virtual void accept(ir_visitor *v)
1685 {
1686 v->visit(this);
1687 }
1688
1689 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1690
1691 bool is_break() const
1692 {
1693 return mode == jump_break;
1694 }
1695
1696 bool is_continue() const
1697 {
1698 return mode == jump_continue;
1699 }
1700
1701 /** Mode selector for the jump instruction. */
1702 enum jump_mode mode;
1703 };
1704
1705 /**
1706 * IR instruction representing discard statements.
1707 */
1708 class ir_discard : public ir_jump {
1709 public:
1710 ir_discard()
1711 : ir_jump(ir_type_discard)
1712 {
1713 this->condition = NULL;
1714 }
1715
1716 ir_discard(ir_rvalue *cond)
1717 : ir_jump(ir_type_discard)
1718 {
1719 this->condition = cond;
1720 }
1721
1722 virtual ir_discard *clone(void *mem_ctx, struct hash_table *ht) const;
1723
1724 virtual void accept(ir_visitor *v)
1725 {
1726 v->visit(this);
1727 }
1728
1729 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1730
1731 ir_rvalue *condition;
1732 };
1733 /*@}*/
1734
1735
1736 /**
1737 * Texture sampling opcodes used in ir_texture
1738 */
1739 enum ir_texture_opcode {
1740 ir_tex, /**< Regular texture look-up */
1741 ir_txb, /**< Texture look-up with LOD bias */
1742 ir_txl, /**< Texture look-up with explicit LOD */
1743 ir_txd, /**< Texture look-up with partial derivatvies */
1744 ir_txf, /**< Texel fetch with explicit LOD */
1745 ir_txf_ms, /**< Multisample texture fetch */
1746 ir_txs, /**< Texture size */
1747 ir_lod, /**< Texture lod query */
1748 ir_tg4, /**< Texture gather */
1749 ir_query_levels /**< Texture levels query */
1750 };
1751
1752
1753 /**
1754 * IR instruction to sample a texture
1755 *
1756 * The specific form of the IR instruction depends on the \c mode value
1757 * selected from \c ir_texture_opcodes. In the printed IR, these will
1758 * appear as:
1759 *
1760 * Texel offset (0 or an expression)
1761 * | Projection divisor
1762 * | | Shadow comparitor
1763 * | | |
1764 * v v v
1765 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1766 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1767 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1768 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1769 * (txf <type> <sampler> <coordinate> 0 <lod>)
1770 * (txf_ms
1771 * <type> <sampler> <coordinate> <sample_index>)
1772 * (txs <type> <sampler> <lod>)
1773 * (lod <type> <sampler> <coordinate>)
1774 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
1775 * (query_levels <type> <sampler>)
1776 */
1777 class ir_texture : public ir_rvalue {
1778 public:
1779 ir_texture(enum ir_texture_opcode op)
1780 : ir_rvalue(ir_type_texture),
1781 op(op), sampler(NULL), coordinate(NULL), projector(NULL),
1782 shadow_comparitor(NULL), offset(NULL)
1783 {
1784 memset(&lod_info, 0, sizeof(lod_info));
1785 }
1786
1787 virtual ir_texture *clone(void *mem_ctx, struct hash_table *) const;
1788
1789 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1790
1791 virtual void accept(ir_visitor *v)
1792 {
1793 v->visit(this);
1794 }
1795
1796 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1797
1798 virtual bool equals(ir_instruction *ir, enum ir_node_type ignore = ir_type_unset);
1799
1800 /**
1801 * Return a string representing the ir_texture_opcode.
1802 */
1803 const char *opcode_string();
1804
1805 /** Set the sampler and type. */
1806 void set_sampler(ir_dereference *sampler, const glsl_type *type);
1807
1808 /**
1809 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1810 */
1811 static ir_texture_opcode get_opcode(const char *);
1812
1813 enum ir_texture_opcode op;
1814
1815 /** Sampler to use for the texture access. */
1816 ir_dereference *sampler;
1817
1818 /** Texture coordinate to sample */
1819 ir_rvalue *coordinate;
1820
1821 /**
1822 * Value used for projective divide.
1823 *
1824 * If there is no projective divide (the common case), this will be
1825 * \c NULL. Optimization passes should check for this to point to a constant
1826 * of 1.0 and replace that with \c NULL.
1827 */
1828 ir_rvalue *projector;
1829
1830 /**
1831 * Coordinate used for comparison on shadow look-ups.
1832 *
1833 * If there is no shadow comparison, this will be \c NULL. For the
1834 * \c ir_txf opcode, this *must* be \c NULL.
1835 */
1836 ir_rvalue *shadow_comparitor;
1837
1838 /** Texel offset. */
1839 ir_rvalue *offset;
1840
1841 union {
1842 ir_rvalue *lod; /**< Floating point LOD */
1843 ir_rvalue *bias; /**< Floating point LOD bias */
1844 ir_rvalue *sample_index; /**< MSAA sample index */
1845 ir_rvalue *component; /**< Gather component selector */
1846 struct {
1847 ir_rvalue *dPdx; /**< Partial derivative of coordinate wrt X */
1848 ir_rvalue *dPdy; /**< Partial derivative of coordinate wrt Y */
1849 } grad;
1850 } lod_info;
1851 };
1852
1853
1854 struct ir_swizzle_mask {
1855 unsigned x:2;
1856 unsigned y:2;
1857 unsigned z:2;
1858 unsigned w:2;
1859
1860 /**
1861 * Number of components in the swizzle.
1862 */
1863 unsigned num_components:3;
1864
1865 /**
1866 * Does the swizzle contain duplicate components?
1867 *
1868 * L-value swizzles cannot contain duplicate components.
1869 */
1870 unsigned has_duplicates:1;
1871 };
1872
1873
1874 class ir_swizzle : public ir_rvalue {
1875 public:
1876 ir_swizzle(ir_rvalue *, unsigned x, unsigned y, unsigned z, unsigned w,
1877 unsigned count);
1878
1879 ir_swizzle(ir_rvalue *val, const unsigned *components, unsigned count);
1880
1881 ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask);
1882
1883 virtual ir_swizzle *clone(void *mem_ctx, struct hash_table *) const;
1884
1885 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1886
1887 /**
1888 * Construct an ir_swizzle from the textual representation. Can fail.
1889 */
1890 static ir_swizzle *create(ir_rvalue *, const char *, unsigned vector_length);
1891
1892 virtual void accept(ir_visitor *v)
1893 {
1894 v->visit(this);
1895 }
1896
1897 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1898
1899 virtual bool equals(ir_instruction *ir, enum ir_node_type ignore = ir_type_unset);
1900
1901 bool is_lvalue() const
1902 {
1903 return val->is_lvalue() && !mask.has_duplicates;
1904 }
1905
1906 /**
1907 * Get the variable that is ultimately referenced by an r-value
1908 */
1909 virtual ir_variable *variable_referenced() const;
1910
1911 ir_rvalue *val;
1912 ir_swizzle_mask mask;
1913
1914 private:
1915 /**
1916 * Initialize the mask component of a swizzle
1917 *
1918 * This is used by the \c ir_swizzle constructors.
1919 */
1920 void init_mask(const unsigned *components, unsigned count);
1921 };
1922
1923
1924 class ir_dereference : public ir_rvalue {
1925 public:
1926 virtual ir_dereference *clone(void *mem_ctx, struct hash_table *) const = 0;
1927
1928 bool is_lvalue() const;
1929
1930 /**
1931 * Get the variable that is ultimately referenced by an r-value
1932 */
1933 virtual ir_variable *variable_referenced() const = 0;
1934
1935 protected:
1936 ir_dereference(enum ir_node_type t)
1937 : ir_rvalue(t)
1938 {
1939 }
1940 };
1941
1942
1943 class ir_dereference_variable : public ir_dereference {
1944 public:
1945 ir_dereference_variable(ir_variable *var);
1946
1947 virtual ir_dereference_variable *clone(void *mem_ctx,
1948 struct hash_table *) const;
1949
1950 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1951
1952 virtual bool equals(ir_instruction *ir, enum ir_node_type ignore = ir_type_unset);
1953
1954 /**
1955 * Get the variable that is ultimately referenced by an r-value
1956 */
1957 virtual ir_variable *variable_referenced() const
1958 {
1959 return this->var;
1960 }
1961
1962 virtual ir_variable *whole_variable_referenced()
1963 {
1964 /* ir_dereference_variable objects always dereference the entire
1965 * variable. However, if this dereference is dereferenced by anything
1966 * else, the complete deferefernce chain is not a whole-variable
1967 * dereference. This method should only be called on the top most
1968 * ir_rvalue in a dereference chain.
1969 */
1970 return this->var;
1971 }
1972
1973 virtual void accept(ir_visitor *v)
1974 {
1975 v->visit(this);
1976 }
1977
1978 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1979
1980 /**
1981 * Object being dereferenced.
1982 */
1983 ir_variable *var;
1984 };
1985
1986
1987 class ir_dereference_array : public ir_dereference {
1988 public:
1989 ir_dereference_array(ir_rvalue *value, ir_rvalue *array_index);
1990
1991 ir_dereference_array(ir_variable *var, ir_rvalue *array_index);
1992
1993 virtual ir_dereference_array *clone(void *mem_ctx,
1994 struct hash_table *) const;
1995
1996 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1997
1998 virtual bool equals(ir_instruction *ir, enum ir_node_type ignore = ir_type_unset);
1999
2000 /**
2001 * Get the variable that is ultimately referenced by an r-value
2002 */
2003 virtual ir_variable *variable_referenced() const
2004 {
2005 return this->array->variable_referenced();
2006 }
2007
2008 virtual void accept(ir_visitor *v)
2009 {
2010 v->visit(this);
2011 }
2012
2013 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2014
2015 ir_rvalue *array;
2016 ir_rvalue *array_index;
2017
2018 private:
2019 void set_array(ir_rvalue *value);
2020 };
2021
2022
2023 class ir_dereference_record : public ir_dereference {
2024 public:
2025 ir_dereference_record(ir_rvalue *value, const char *field);
2026
2027 ir_dereference_record(ir_variable *var, const char *field);
2028
2029 virtual ir_dereference_record *clone(void *mem_ctx,
2030 struct hash_table *) const;
2031
2032 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
2033
2034 /**
2035 * Get the variable that is ultimately referenced by an r-value
2036 */
2037 virtual ir_variable *variable_referenced() const
2038 {
2039 return this->record->variable_referenced();
2040 }
2041
2042 virtual void accept(ir_visitor *v)
2043 {
2044 v->visit(this);
2045 }
2046
2047 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2048
2049 ir_rvalue *record;
2050 const char *field;
2051 };
2052
2053
2054 /**
2055 * Data stored in an ir_constant
2056 */
2057 union ir_constant_data {
2058 unsigned u[16];
2059 int i[16];
2060 float f[16];
2061 bool b[16];
2062 };
2063
2064
2065 class ir_constant : public ir_rvalue {
2066 public:
2067 ir_constant(const struct glsl_type *type, const ir_constant_data *data);
2068 ir_constant(bool b, unsigned vector_elements=1);
2069 ir_constant(unsigned int u, unsigned vector_elements=1);
2070 ir_constant(int i, unsigned vector_elements=1);
2071 ir_constant(float f, unsigned vector_elements=1);
2072
2073 /**
2074 * Construct an ir_constant from a list of ir_constant values
2075 */
2076 ir_constant(const struct glsl_type *type, exec_list *values);
2077
2078 /**
2079 * Construct an ir_constant from a scalar component of another ir_constant
2080 *
2081 * The new \c ir_constant inherits the type of the component from the
2082 * source constant.
2083 *
2084 * \note
2085 * In the case of a matrix constant, the new constant is a scalar, \b not
2086 * a vector.
2087 */
2088 ir_constant(const ir_constant *c, unsigned i);
2089
2090 /**
2091 * Return a new ir_constant of the specified type containing all zeros.
2092 */
2093 static ir_constant *zero(void *mem_ctx, const glsl_type *type);
2094
2095 virtual ir_constant *clone(void *mem_ctx, struct hash_table *) const;
2096
2097 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
2098
2099 virtual void accept(ir_visitor *v)
2100 {
2101 v->visit(this);
2102 }
2103
2104 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2105
2106 virtual bool equals(ir_instruction *ir, enum ir_node_type ignore = ir_type_unset);
2107
2108 /**
2109 * Get a particular component of a constant as a specific type
2110 *
2111 * This is useful, for example, to get a value from an integer constant
2112 * as a float or bool. This appears frequently when constructors are
2113 * called with all constant parameters.
2114 */
2115 /*@{*/
2116 bool get_bool_component(unsigned i) const;
2117 float get_float_component(unsigned i) const;
2118 int get_int_component(unsigned i) const;
2119 unsigned get_uint_component(unsigned i) const;
2120 /*@}*/
2121
2122 ir_constant *get_array_element(unsigned i) const;
2123
2124 ir_constant *get_record_field(const char *name);
2125
2126 /**
2127 * Copy the values on another constant at a given offset.
2128 *
2129 * The offset is ignored for array or struct copies, it's only for
2130 * scalars or vectors into vectors or matrices.
2131 *
2132 * With identical types on both sides and zero offset it's clone()
2133 * without creating a new object.
2134 */
2135
2136 void copy_offset(ir_constant *src, int offset);
2137
2138 /**
2139 * Copy the values on another constant at a given offset and
2140 * following an assign-like mask.
2141 *
2142 * The mask is ignored for scalars.
2143 *
2144 * Note that this function only handles what assign can handle,
2145 * i.e. at most a vector as source and a column of a matrix as
2146 * destination.
2147 */
2148
2149 void copy_masked_offset(ir_constant *src, int offset, unsigned int mask);
2150
2151 /**
2152 * Determine whether a constant has the same value as another constant
2153 *
2154 * \sa ir_constant::is_zero, ir_constant::is_one,
2155 * ir_constant::is_negative_one, ir_constant::is_basis
2156 */
2157 bool has_value(const ir_constant *) const;
2158
2159 /**
2160 * Return true if this ir_constant represents the given value.
2161 *
2162 * For vectors, this checks that each component is the given value.
2163 */
2164 virtual bool is_value(float f, int i) const;
2165 virtual bool is_zero() const;
2166 virtual bool is_one() const;
2167 virtual bool is_negative_one() const;
2168 virtual bool is_basis() const;
2169
2170 /**
2171 * Return true for constants that could be stored as 16-bit unsigned values.
2172 *
2173 * Note that this will return true even for signed integer ir_constants, as
2174 * long as the value is non-negative and fits in 16-bits.
2175 */
2176 virtual bool is_uint16_constant() const;
2177
2178 /**
2179 * Value of the constant.
2180 *
2181 * The field used to back the values supplied by the constant is determined
2182 * by the type associated with the \c ir_instruction. Constants may be
2183 * scalars, vectors, or matrices.
2184 */
2185 union ir_constant_data value;
2186
2187 /* Array elements */
2188 ir_constant **array_elements;
2189
2190 /* Structure fields */
2191 exec_list components;
2192
2193 private:
2194 /**
2195 * Parameterless constructor only used by the clone method
2196 */
2197 ir_constant(void);
2198 };
2199
2200 /**
2201 * IR instruction to emit a vertex in a geometry shader.
2202 */
2203 class ir_emit_vertex : public ir_instruction {
2204 public:
2205 ir_emit_vertex(ir_rvalue *stream)
2206 : ir_instruction(ir_type_emit_vertex),
2207 stream(stream)
2208 {
2209 assert(stream);
2210 }
2211
2212 virtual void accept(ir_visitor *v)
2213 {
2214 v->visit(this);
2215 }
2216
2217 virtual ir_emit_vertex *clone(void *mem_ctx, struct hash_table *ht) const
2218 {
2219 return new(mem_ctx) ir_emit_vertex(this->stream->clone(mem_ctx, ht));
2220 }
2221
2222 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2223
2224 int stream_id() const
2225 {
2226 return stream->as_constant()->value.i[0];
2227 }
2228
2229 ir_rvalue *stream;
2230 };
2231
2232 /**
2233 * IR instruction to complete the current primitive and start a new one in a
2234 * geometry shader.
2235 */
2236 class ir_end_primitive : public ir_instruction {
2237 public:
2238 ir_end_primitive(ir_rvalue *stream)
2239 : ir_instruction(ir_type_end_primitive),
2240 stream(stream)
2241 {
2242 assert(stream);
2243 }
2244
2245 virtual void accept(ir_visitor *v)
2246 {
2247 v->visit(this);
2248 }
2249
2250 virtual ir_end_primitive *clone(void *mem_ctx, struct hash_table *ht) const
2251 {
2252 return new(mem_ctx) ir_end_primitive(this->stream->clone(mem_ctx, ht));
2253 }
2254
2255 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2256
2257 int stream_id() const
2258 {
2259 return stream->as_constant()->value.i[0];
2260 }
2261
2262 ir_rvalue *stream;
2263 };
2264
2265 /*@}*/
2266
2267 /**
2268 * Apply a visitor to each IR node in a list
2269 */
2270 void
2271 visit_exec_list(exec_list *list, ir_visitor *visitor);
2272
2273 /**
2274 * Validate invariants on each IR node in a list
2275 */
2276 void validate_ir_tree(exec_list *instructions);
2277
2278 struct _mesa_glsl_parse_state;
2279 struct gl_shader_program;
2280
2281 /**
2282 * Detect whether an unlinked shader contains static recursion
2283 *
2284 * If the list of instructions is determined to contain static recursion,
2285 * \c _mesa_glsl_error will be called to emit error messages for each function
2286 * that is in the recursion cycle.
2287 */
2288 void
2289 detect_recursion_unlinked(struct _mesa_glsl_parse_state *state,
2290 exec_list *instructions);
2291
2292 /**
2293 * Detect whether a linked shader contains static recursion
2294 *
2295 * If the list of instructions is determined to contain static recursion,
2296 * \c link_error_printf will be called to emit error messages for each function
2297 * that is in the recursion cycle. In addition,
2298 * \c gl_shader_program::LinkStatus will be set to false.
2299 */
2300 void
2301 detect_recursion_linked(struct gl_shader_program *prog,
2302 exec_list *instructions);
2303
2304 /**
2305 * Make a clone of each IR instruction in a list
2306 *
2307 * \param in List of IR instructions that are to be cloned
2308 * \param out List to hold the cloned instructions
2309 */
2310 void
2311 clone_ir_list(void *mem_ctx, exec_list *out, const exec_list *in);
2312
2313 extern void
2314 _mesa_glsl_initialize_variables(exec_list *instructions,
2315 struct _mesa_glsl_parse_state *state);
2316
2317 extern void
2318 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state *state);
2319
2320 extern void
2321 _mesa_glsl_initialize_builtin_functions();
2322
2323 extern ir_function_signature *
2324 _mesa_glsl_find_builtin_function(_mesa_glsl_parse_state *state,
2325 const char *name, exec_list *actual_parameters);
2326
2327 extern gl_shader *
2328 _mesa_glsl_get_builtin_function_shader(void);
2329
2330 extern void
2331 _mesa_glsl_release_functions(void);
2332
2333 extern void
2334 _mesa_glsl_release_builtin_functions(void);
2335
2336 extern void
2337 reparent_ir(exec_list *list, void *mem_ctx);
2338
2339 struct glsl_symbol_table;
2340
2341 extern void
2342 import_prototypes(const exec_list *source, exec_list *dest,
2343 struct glsl_symbol_table *symbols, void *mem_ctx);
2344
2345 extern bool
2346 ir_has_call(ir_instruction *ir);
2347
2348 extern void
2349 do_set_program_inouts(exec_list *instructions, struct gl_program *prog,
2350 gl_shader_stage shader_stage);
2351
2352 extern char *
2353 prototype_string(const glsl_type *return_type, const char *name,
2354 exec_list *parameters);
2355
2356 const char *
2357 mode_string(const ir_variable *var);
2358
2359 /**
2360 * Built-in / reserved GL variables names start with "gl_"
2361 */
2362 static inline bool
2363 is_gl_identifier(const char *s)
2364 {
2365 return s && s[0] == 'g' && s[1] == 'l' && s[2] == '_';
2366 }
2367
2368 extern "C" {
2369 #endif /* __cplusplus */
2370
2371 extern void _mesa_print_ir(FILE *f, struct exec_list *instructions,
2372 struct _mesa_glsl_parse_state *state);
2373
2374 extern void
2375 fprint_ir(FILE *f, const void *instruction);
2376
2377 #ifdef __cplusplus
2378 } /* extern "C" */
2379 #endif
2380
2381 unsigned
2382 vertices_per_prim(GLenum prim);
2383
2384 #endif /* IR_H */