Merge remote-tracking branch 'origin/master' into vulkan
[mesa.git] / src / compiler / 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 "compiler/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_barrier,
82 ir_type_max, /**< maximum ir_type enum number, for validation */
83 ir_type_unset = ir_type_max
84 };
85
86
87 /**
88 * Base class of all IR instructions
89 */
90 class ir_instruction : public exec_node {
91 public:
92 enum ir_node_type ir_type;
93
94 /**
95 * GCC 4.7+ and clang warn when deleting an ir_instruction unless
96 * there's a virtual destructor present. Because we almost
97 * universally use ralloc for our memory management of
98 * ir_instructions, the destructor doesn't need to do any work.
99 */
100 virtual ~ir_instruction()
101 {
102 }
103
104 /** ir_print_visitor helper for debugging. */
105 void print(void) const;
106 void fprint(FILE *f) const;
107
108 virtual void accept(ir_visitor *) = 0;
109 virtual ir_visitor_status accept(ir_hierarchical_visitor *) = 0;
110 virtual ir_instruction *clone(void *mem_ctx,
111 struct hash_table *ht) const = 0;
112
113 bool is_rvalue() const
114 {
115 return ir_type == ir_type_dereference_array ||
116 ir_type == ir_type_dereference_record ||
117 ir_type == ir_type_dereference_variable ||
118 ir_type == ir_type_constant ||
119 ir_type == ir_type_expression ||
120 ir_type == ir_type_swizzle ||
121 ir_type == ir_type_texture;
122 }
123
124 bool is_dereference() const
125 {
126 return ir_type == ir_type_dereference_array ||
127 ir_type == ir_type_dereference_record ||
128 ir_type == ir_type_dereference_variable;
129 }
130
131 bool is_jump() const
132 {
133 return ir_type == ir_type_loop_jump ||
134 ir_type == ir_type_return ||
135 ir_type == ir_type_discard;
136 }
137
138 /**
139 * \name IR instruction downcast functions
140 *
141 * These functions either cast the object to a derived class or return
142 * \c NULL if the object's type does not match the specified derived class.
143 * Additional downcast functions will be added as needed.
144 */
145 /*@{*/
146 #define AS_BASE(TYPE) \
147 class ir_##TYPE *as_##TYPE() \
148 { \
149 assume(this != NULL); \
150 return is_##TYPE() ? (ir_##TYPE *) this : NULL; \
151 } \
152 const class ir_##TYPE *as_##TYPE() const \
153 { \
154 assume(this != NULL); \
155 return is_##TYPE() ? (ir_##TYPE *) this : NULL; \
156 }
157
158 AS_BASE(rvalue)
159 AS_BASE(dereference)
160 AS_BASE(jump)
161 #undef AS_BASE
162
163 #define AS_CHILD(TYPE) \
164 class ir_##TYPE * as_##TYPE() \
165 { \
166 assume(this != NULL); \
167 return ir_type == ir_type_##TYPE ? (ir_##TYPE *) this : NULL; \
168 } \
169 const class ir_##TYPE * as_##TYPE() const \
170 { \
171 assume(this != NULL); \
172 return ir_type == ir_type_##TYPE ? (const ir_##TYPE *) this : NULL; \
173 }
174 AS_CHILD(variable)
175 AS_CHILD(function)
176 AS_CHILD(dereference_array)
177 AS_CHILD(dereference_variable)
178 AS_CHILD(dereference_record)
179 AS_CHILD(expression)
180 AS_CHILD(loop)
181 AS_CHILD(assignment)
182 AS_CHILD(call)
183 AS_CHILD(return)
184 AS_CHILD(if)
185 AS_CHILD(swizzle)
186 AS_CHILD(texture)
187 AS_CHILD(constant)
188 AS_CHILD(discard)
189 #undef AS_CHILD
190 /*@}*/
191
192 /**
193 * IR equality method: Return true if the referenced instruction would
194 * return the same value as this one.
195 *
196 * This intended to be used for CSE and algebraic optimizations, on rvalues
197 * in particular. No support for other instruction types (assignments,
198 * jumps, calls, etc.) is planned.
199 */
200 virtual bool equals(const ir_instruction *ir,
201 enum ir_node_type ignore = ir_type_unset) const;
202
203 protected:
204 ir_instruction(enum ir_node_type t)
205 : ir_type(t)
206 {
207 }
208
209 private:
210 ir_instruction()
211 {
212 assert(!"Should not get here.");
213 }
214 };
215
216
217 /**
218 * The base class for all "values"/expression trees.
219 */
220 class ir_rvalue : public ir_instruction {
221 public:
222 const struct glsl_type *type;
223
224 virtual ir_rvalue *clone(void *mem_ctx, struct hash_table *) const;
225
226 virtual void accept(ir_visitor *v)
227 {
228 v->visit(this);
229 }
230
231 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
232
233 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
234
235 ir_rvalue *as_rvalue_to_saturate();
236
237 virtual bool is_lvalue() const
238 {
239 return false;
240 }
241
242 /**
243 * Get the variable that is ultimately referenced by an r-value
244 */
245 virtual ir_variable *variable_referenced() const
246 {
247 return NULL;
248 }
249
250
251 /**
252 * If an r-value is a reference to a whole variable, get that variable
253 *
254 * \return
255 * Pointer to a variable that is completely dereferenced by the r-value. If
256 * the r-value is not a dereference or the dereference does not access the
257 * entire variable (i.e., it's just one array element, struct field), \c NULL
258 * is returned.
259 */
260 virtual ir_variable *whole_variable_referenced()
261 {
262 return NULL;
263 }
264
265 /**
266 * Determine if an r-value has the value zero
267 *
268 * The base implementation of this function always returns \c false. The
269 * \c ir_constant class over-rides this function to return \c true \b only
270 * for vector and scalar types that have all elements set to the value
271 * zero (or \c false for booleans).
272 *
273 * \sa ir_constant::has_value, ir_rvalue::is_one, ir_rvalue::is_negative_one
274 */
275 virtual bool is_zero() const;
276
277 /**
278 * Determine if an r-value has the value one
279 *
280 * The base implementation of this function always returns \c false. The
281 * \c ir_constant class over-rides this function to return \c true \b only
282 * for vector and scalar types that have all elements set to the value
283 * one (or \c true for booleans).
284 *
285 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_negative_one
286 */
287 virtual bool is_one() const;
288
289 /**
290 * Determine if an r-value has the value negative one
291 *
292 * The base implementation of this function always returns \c false. The
293 * \c ir_constant class over-rides this function to return \c true \b only
294 * for vector and scalar types that have all elements set to the value
295 * negative one. For boolean types, the result is always \c false.
296 *
297 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one
298 */
299 virtual bool is_negative_one() const;
300
301 /**
302 * Determine if an r-value is an unsigned integer constant which can be
303 * stored in 16 bits.
304 *
305 * \sa ir_constant::is_uint16_constant.
306 */
307 virtual bool is_uint16_constant() const { return false; }
308
309 /**
310 * Return a generic value of error_type.
311 *
312 * Allocation will be performed with 'mem_ctx' as ralloc owner.
313 */
314 static ir_rvalue *error_value(void *mem_ctx);
315
316 protected:
317 ir_rvalue(enum ir_node_type t);
318 };
319
320
321 /**
322 * Variable storage classes
323 */
324 enum ir_variable_mode {
325 ir_var_auto = 0, /**< Function local variables and globals. */
326 ir_var_uniform, /**< Variable declared as a uniform. */
327 ir_var_shader_storage, /**< Variable declared as an ssbo. */
328 ir_var_shader_shared, /**< Variable declared as shared. */
329 ir_var_shader_in,
330 ir_var_shader_out,
331 ir_var_function_in,
332 ir_var_function_out,
333 ir_var_function_inout,
334 ir_var_const_in, /**< "in" param that must be a constant expression */
335 ir_var_system_value, /**< Ex: front-face, instance-id, etc. */
336 ir_var_temporary, /**< Temporary variable generated during compilation. */
337 ir_var_mode_count /**< Number of variable modes */
338 };
339
340 /**
341 * Enum keeping track of how a variable was declared. For error checking of
342 * the gl_PerVertex redeclaration rules.
343 */
344 enum ir_var_declaration_type {
345 /**
346 * Normal declaration (for most variables, this means an explicit
347 * declaration. Exception: temporaries are always implicitly declared, but
348 * they still use ir_var_declared_normally).
349 *
350 * Note: an ir_variable that represents a named interface block uses
351 * ir_var_declared_normally.
352 */
353 ir_var_declared_normally = 0,
354
355 /**
356 * Variable was explicitly declared (or re-declared) in an unnamed
357 * interface block.
358 */
359 ir_var_declared_in_block,
360
361 /**
362 * Variable is an implicitly declared built-in that has not been explicitly
363 * re-declared by the shader.
364 */
365 ir_var_declared_implicitly,
366
367 /**
368 * Variable is implicitly generated by the compiler and should not be
369 * visible via the API.
370 */
371 ir_var_hidden,
372 };
373
374 /**
375 * \brief Layout qualifiers for gl_FragDepth.
376 *
377 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
378 * with a layout qualifier.
379 */
380 enum ir_depth_layout {
381 ir_depth_layout_none, /**< No depth layout is specified. */
382 ir_depth_layout_any,
383 ir_depth_layout_greater,
384 ir_depth_layout_less,
385 ir_depth_layout_unchanged
386 };
387
388 /**
389 * \brief Convert depth layout qualifier to string.
390 */
391 const char*
392 depth_layout_string(ir_depth_layout layout);
393
394 /**
395 * Description of built-in state associated with a uniform
396 *
397 * \sa ir_variable::state_slots
398 */
399 struct ir_state_slot {
400 int tokens[5];
401 int swizzle;
402 };
403
404
405 /**
406 * Get the string value for an interpolation qualifier
407 *
408 * \return The string that would be used in a shader to specify \c
409 * mode will be returned.
410 *
411 * This function is used to generate error messages of the form "shader
412 * uses %s interpolation qualifier", so in the case where there is no
413 * interpolation qualifier, it returns "no".
414 *
415 * This function should only be used on a shader input or output variable.
416 */
417 const char *interpolation_string(unsigned interpolation);
418
419
420 class ir_variable : public ir_instruction {
421 public:
422 ir_variable(const struct glsl_type *, const char *, ir_variable_mode);
423
424 virtual ir_variable *clone(void *mem_ctx, struct hash_table *ht) const;
425
426 virtual void accept(ir_visitor *v)
427 {
428 v->visit(this);
429 }
430
431 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
432
433
434 /**
435 * Determine whether or not a variable is part of a uniform or
436 * shader storage block.
437 */
438 inline bool is_in_buffer_block() const
439 {
440 return (this->data.mode == ir_var_uniform ||
441 this->data.mode == ir_var_shader_storage) &&
442 this->interface_type != NULL;
443 }
444
445 /**
446 * Determine whether or not a variable is part of a shader storage block.
447 */
448 inline bool is_in_shader_storage_block() const
449 {
450 return this->data.mode == ir_var_shader_storage &&
451 this->interface_type != NULL;
452 }
453
454 /**
455 * Determine whether or not a variable is the declaration of an interface
456 * block
457 *
458 * For the first declaration below, there will be an \c ir_variable named
459 * "instance" whose type and whose instance_type will be the same
460 * \cglsl_type. For the second declaration, there will be an \c ir_variable
461 * named "f" whose type is float and whose instance_type is B2.
462 *
463 * "instance" is an interface instance variable, but "f" is not.
464 *
465 * uniform B1 {
466 * float f;
467 * } instance;
468 *
469 * uniform B2 {
470 * float f;
471 * };
472 */
473 inline bool is_interface_instance() const
474 {
475 return this->type->without_array() == this->interface_type;
476 }
477
478 /**
479 * Set this->interface_type on a newly created variable.
480 */
481 void init_interface_type(const struct glsl_type *type)
482 {
483 assert(this->interface_type == NULL);
484 this->interface_type = type;
485 if (this->is_interface_instance()) {
486 this->u.max_ifc_array_access =
487 rzalloc_array(this, unsigned, type->length);
488 }
489 }
490
491 /**
492 * Change this->interface_type on a variable that previously had a
493 * different, but compatible, interface_type. This is used during linking
494 * to set the size of arrays in interface blocks.
495 */
496 void change_interface_type(const struct glsl_type *type)
497 {
498 if (this->u.max_ifc_array_access != NULL) {
499 /* max_ifc_array_access has already been allocated, so make sure the
500 * new interface has the same number of fields as the old one.
501 */
502 assert(this->interface_type->length == type->length);
503 }
504 this->interface_type = type;
505 }
506
507 /**
508 * Change this->interface_type on a variable that previously had a
509 * different, and incompatible, interface_type. This is used during
510 * compilation to handle redeclaration of the built-in gl_PerVertex
511 * interface block.
512 */
513 void reinit_interface_type(const struct glsl_type *type)
514 {
515 if (this->u.max_ifc_array_access != NULL) {
516 #ifndef NDEBUG
517 /* Redeclaring gl_PerVertex is only allowed if none of the built-ins
518 * it defines have been accessed yet; so it's safe to throw away the
519 * old max_ifc_array_access pointer, since all of its values are
520 * zero.
521 */
522 for (unsigned i = 0; i < this->interface_type->length; i++)
523 assert(this->u.max_ifc_array_access[i] == 0);
524 #endif
525 ralloc_free(this->u.max_ifc_array_access);
526 this->u.max_ifc_array_access = NULL;
527 }
528 this->interface_type = NULL;
529 init_interface_type(type);
530 }
531
532 const glsl_type *get_interface_type() const
533 {
534 return this->interface_type;
535 }
536
537 /**
538 * Get the max_ifc_array_access pointer
539 *
540 * A "set" function is not needed because the array is dynmically allocated
541 * as necessary.
542 */
543 inline unsigned *get_max_ifc_array_access()
544 {
545 assert(this->data._num_state_slots == 0);
546 return this->u.max_ifc_array_access;
547 }
548
549 inline unsigned get_num_state_slots() const
550 {
551 assert(!this->is_interface_instance()
552 || this->data._num_state_slots == 0);
553 return this->data._num_state_slots;
554 }
555
556 inline void set_num_state_slots(unsigned n)
557 {
558 assert(!this->is_interface_instance()
559 || n == 0);
560 this->data._num_state_slots = n;
561 }
562
563 inline ir_state_slot *get_state_slots()
564 {
565 return this->is_interface_instance() ? NULL : this->u.state_slots;
566 }
567
568 inline const ir_state_slot *get_state_slots() const
569 {
570 return this->is_interface_instance() ? NULL : this->u.state_slots;
571 }
572
573 inline ir_state_slot *allocate_state_slots(unsigned n)
574 {
575 assert(!this->is_interface_instance());
576
577 this->u.state_slots = ralloc_array(this, ir_state_slot, n);
578 this->data._num_state_slots = 0;
579
580 if (this->u.state_slots != NULL)
581 this->data._num_state_slots = n;
582
583 return this->u.state_slots;
584 }
585
586 inline bool is_name_ralloced() const
587 {
588 return this->name != ir_variable::tmp_name;
589 }
590
591 /**
592 * Enable emitting extension warnings for this variable
593 */
594 void enable_extension_warning(const char *extension);
595
596 /**
597 * Get the extension warning string for this variable
598 *
599 * If warnings are not enabled, \c NULL is returned.
600 */
601 const char *get_extension_warning() const;
602
603 /**
604 * Declared type of the variable
605 */
606 const struct glsl_type *type;
607
608 /**
609 * Declared name of the variable
610 */
611 const char *name;
612
613 struct ir_variable_data {
614
615 /**
616 * Is the variable read-only?
617 *
618 * This is set for variables declared as \c const, shader inputs,
619 * and uniforms.
620 */
621 unsigned read_only:1;
622 unsigned centroid:1;
623 unsigned sample:1;
624 unsigned patch:1;
625 unsigned invariant:1;
626 unsigned precise:1;
627
628 /**
629 * Has this variable been used for reading or writing?
630 *
631 * Several GLSL semantic checks require knowledge of whether or not a
632 * variable has been used. For example, it is an error to redeclare a
633 * variable as invariant after it has been used.
634 *
635 * This is only maintained in the ast_to_hir.cpp path, not in
636 * Mesa's fixed function or ARB program paths.
637 */
638 unsigned used:1;
639
640 /**
641 * Has this variable been statically assigned?
642 *
643 * This answers whether the variable was assigned in any path of
644 * the shader during ast_to_hir. This doesn't answer whether it is
645 * still written after dead code removal, nor is it maintained in
646 * non-ast_to_hir.cpp (GLSL parsing) paths.
647 */
648 unsigned assigned:1;
649
650 /**
651 * When separate shader programs are enabled, only input/outputs between
652 * the stages of a multi-stage separate program can be safely removed
653 * from the shader interface. Other input/outputs must remains active.
654 */
655 unsigned always_active_io:1;
656
657 /**
658 * Enum indicating how the variable was declared. See
659 * ir_var_declaration_type.
660 *
661 * This is used to detect certain kinds of illegal variable redeclarations.
662 */
663 unsigned how_declared:2;
664
665 /**
666 * Storage class of the variable.
667 *
668 * \sa ir_variable_mode
669 */
670 unsigned mode:4;
671
672 /**
673 * Interpolation mode for shader inputs / outputs
674 *
675 * \sa ir_variable_interpolation
676 */
677 unsigned interpolation:2;
678
679 /**
680 * \name ARB_fragment_coord_conventions
681 * @{
682 */
683 unsigned origin_upper_left:1;
684 unsigned pixel_center_integer:1;
685 /*@}*/
686
687 /**
688 * Was the location explicitly set in the shader?
689 *
690 * If the location is explicitly set in the shader, it \b cannot be changed
691 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
692 * no effect).
693 */
694 unsigned explicit_location:1;
695 unsigned explicit_index:1;
696
697 /**
698 * Was an initial binding explicitly set in the shader?
699 *
700 * If so, constant_value contains an integer ir_constant representing the
701 * initial binding point.
702 */
703 unsigned explicit_binding:1;
704
705 /**
706 * Does this variable have an initializer?
707 *
708 * This is used by the linker to cross-validiate initializers of global
709 * variables.
710 */
711 unsigned has_initializer:1;
712
713 /**
714 * Is this variable a generic output or input that has not yet been matched
715 * up to a variable in another stage of the pipeline?
716 *
717 * This is used by the linker as scratch storage while assigning locations
718 * to generic inputs and outputs.
719 */
720 unsigned is_unmatched_generic_inout:1;
721
722 /**
723 * If non-zero, then this variable may be packed along with other variables
724 * into a single varying slot, so this offset should be applied when
725 * accessing components. For example, an offset of 1 means that the x
726 * component of this variable is actually stored in component y of the
727 * location specified by \c location.
728 */
729 unsigned location_frac:2;
730
731 /**
732 * Layout of the matrix. Uses glsl_matrix_layout values.
733 */
734 unsigned matrix_layout:2;
735
736 /**
737 * Non-zero if this variable was created by lowering a named interface
738 * block which was not an array.
739 *
740 * Note that this variable and \c from_named_ifc_block_array will never
741 * both be non-zero.
742 */
743 unsigned from_named_ifc_block_nonarray:1;
744
745 /**
746 * Non-zero if this variable was created by lowering a named interface
747 * block which was an array.
748 *
749 * Note that this variable and \c from_named_ifc_block_nonarray will never
750 * both be non-zero.
751 */
752 unsigned from_named_ifc_block_array:1;
753
754 /**
755 * Non-zero if the variable must be a shader input. This is useful for
756 * constraints on function parameters.
757 */
758 unsigned must_be_shader_input:1;
759
760 /**
761 * Output index for dual source blending.
762 *
763 * \note
764 * The GLSL spec only allows the values 0 or 1 for the index in \b dual
765 * source blending.
766 */
767 unsigned index:1;
768
769 /**
770 * Precision qualifier.
771 *
772 * In desktop GLSL we do not care about precision qualifiers at all, in
773 * fact, the spec says that precision qualifiers are ignored.
774 *
775 * To make things easy, we make it so that this field is always
776 * GLSL_PRECISION_NONE on desktop shaders. This way all the variables
777 * have the same precision value and the checks we add in the compiler
778 * for this field will never break a desktop shader compile.
779 */
780 unsigned precision:2;
781
782 /**
783 * \brief Layout qualifier for gl_FragDepth.
784 *
785 * This is not equal to \c ir_depth_layout_none if and only if this
786 * variable is \c gl_FragDepth and a layout qualifier is specified.
787 */
788 ir_depth_layout depth_layout:3;
789
790 /**
791 * ARB_shader_image_load_store qualifiers.
792 */
793 unsigned image_read_only:1; /**< "readonly" qualifier. */
794 unsigned image_write_only:1; /**< "writeonly" qualifier. */
795 unsigned image_coherent:1;
796 unsigned image_volatile:1;
797 unsigned image_restrict:1;
798
799 /**
800 * ARB_shader_storage_buffer_object
801 */
802 unsigned from_ssbo_unsized_array:1; /**< unsized array buffer variable. */
803
804 /**
805 * Emit a warning if this variable is accessed.
806 */
807 private:
808 uint8_t warn_extension_index;
809
810 public:
811 /** Image internal format if specified explicitly, otherwise GL_NONE. */
812 uint16_t image_format;
813
814 private:
815 /**
816 * Number of state slots used
817 *
818 * \note
819 * This could be stored in as few as 7-bits, if necessary. If it is made
820 * smaller, add an assertion to \c ir_variable::allocate_state_slots to
821 * be safe.
822 */
823 uint16_t _num_state_slots;
824
825 public:
826 /**
827 * Initial binding point for a sampler, atomic, or UBO.
828 *
829 * For array types, this represents the binding point for the first element.
830 */
831 int16_t binding;
832
833 /**
834 * Storage location of the base of this variable
835 *
836 * The precise meaning of this field depends on the nature of the variable.
837 *
838 * - Vertex shader input: one of the values from \c gl_vert_attrib.
839 * - Vertex shader output: one of the values from \c gl_varying_slot.
840 * - Geometry shader input: one of the values from \c gl_varying_slot.
841 * - Geometry shader output: one of the values from \c gl_varying_slot.
842 * - Fragment shader input: one of the values from \c gl_varying_slot.
843 * - Fragment shader output: one of the values from \c gl_frag_result.
844 * - Uniforms: Per-stage uniform slot number for default uniform block.
845 * - Uniforms: Index within the uniform block definition for UBO members.
846 * - Non-UBO Uniforms: explicit location until linking then reused to
847 * store uniform slot number.
848 * - Other: This field is not currently used.
849 *
850 * If the variable is a uniform, shader input, or shader output, and the
851 * slot has not been assigned, the value will be -1.
852 */
853 int location;
854
855 /**
856 * for glsl->tgsi/mesa IR we need to store the index into the
857 * parameters for uniforms, initially the code overloaded location
858 * but this causes problems with indirect samplers and AoA.
859 * This is assigned in _mesa_generate_parameters_list_for_uniforms.
860 */
861 int param_index;
862
863 /**
864 * Vertex stream output identifier.
865 */
866 unsigned stream;
867
868 /**
869 * Location an atomic counter is stored at.
870 */
871 unsigned offset;
872
873 /**
874 * Highest element accessed with a constant expression array index
875 *
876 * Not used for non-array variables.
877 */
878 unsigned max_array_access;
879
880 /**
881 * Allow (only) ir_variable direct access private members.
882 */
883 friend class ir_variable;
884 } data;
885
886 /**
887 * Value assigned in the initializer of a variable declared "const"
888 */
889 ir_constant *constant_value;
890
891 /**
892 * Constant expression assigned in the initializer of the variable
893 *
894 * \warning
895 * This field and \c ::constant_value are distinct. Even if the two fields
896 * refer to constants with the same value, they must point to separate
897 * objects.
898 */
899 ir_constant *constant_initializer;
900
901 private:
902 static const char *const warn_extension_table[];
903
904 union {
905 /**
906 * For variables which satisfy the is_interface_instance() predicate,
907 * this points to an array of integers such that if the ith member of
908 * the interface block is an array, max_ifc_array_access[i] is the
909 * maximum array element of that member that has been accessed. If the
910 * ith member of the interface block is not an array,
911 * max_ifc_array_access[i] is unused.
912 *
913 * For variables whose type is not an interface block, this pointer is
914 * NULL.
915 */
916 unsigned *max_ifc_array_access;
917
918 /**
919 * Built-in state that backs this uniform
920 *
921 * Once set at variable creation, \c state_slots must remain invariant.
922 *
923 * If the variable is not a uniform, \c _num_state_slots will be zero
924 * and \c state_slots will be \c NULL.
925 */
926 ir_state_slot *state_slots;
927 } u;
928
929 /**
930 * For variables that are in an interface block or are an instance of an
931 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
932 *
933 * \sa ir_variable::location
934 */
935 const glsl_type *interface_type;
936
937 /**
938 * Name used for anonymous compiler temporaries
939 */
940 static const char tmp_name[];
941
942 public:
943 /**
944 * Should the construct keep names for ir_var_temporary variables?
945 *
946 * When this global is false, names passed to the constructor for
947 * \c ir_var_temporary variables will be dropped. Instead, the variable will
948 * be named "compiler_temp". This name will be in static storage.
949 *
950 * \warning
951 * \b NEVER change the mode of an \c ir_var_temporary.
952 *
953 * \warning
954 * This variable is \b not thread-safe. It is global, \b not
955 * per-context. It begins life false. A context can, at some point, make
956 * it true. From that point on, it will be true forever. This should be
957 * okay since it will only be set true while debugging.
958 */
959 static bool temporaries_allocate_names;
960 };
961
962 /**
963 * A function that returns whether a built-in function is available in the
964 * current shading language (based on version, ES or desktop, and extensions).
965 */
966 typedef bool (*builtin_available_predicate)(const _mesa_glsl_parse_state *);
967
968 /*@{*/
969 /**
970 * The representation of a function instance; may be the full definition or
971 * simply a prototype.
972 */
973 class ir_function_signature : public ir_instruction {
974 /* An ir_function_signature will be part of the list of signatures in
975 * an ir_function.
976 */
977 public:
978 ir_function_signature(const glsl_type *return_type,
979 builtin_available_predicate builtin_avail = NULL);
980
981 virtual ir_function_signature *clone(void *mem_ctx,
982 struct hash_table *ht) const;
983 ir_function_signature *clone_prototype(void *mem_ctx,
984 struct hash_table *ht) const;
985
986 virtual void accept(ir_visitor *v)
987 {
988 v->visit(this);
989 }
990
991 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
992
993 /**
994 * Attempt to evaluate this function as a constant expression,
995 * given a list of the actual parameters and the variable context.
996 * Returns NULL for non-built-ins.
997 */
998 ir_constant *constant_expression_value(exec_list *actual_parameters, struct hash_table *variable_context);
999
1000 /**
1001 * Get the name of the function for which this is a signature
1002 */
1003 const char *function_name() const;
1004
1005 /**
1006 * Get a handle to the function for which this is a signature
1007 *
1008 * There is no setter function, this function returns a \c const pointer,
1009 * and \c ir_function_signature::_function is private for a reason. The
1010 * only way to make a connection between a function and function signature
1011 * is via \c ir_function::add_signature. This helps ensure that certain
1012 * invariants (i.e., a function signature is in the list of signatures for
1013 * its \c _function) are met.
1014 *
1015 * \sa ir_function::add_signature
1016 */
1017 inline const class ir_function *function() const
1018 {
1019 return this->_function;
1020 }
1021
1022 /**
1023 * Check whether the qualifiers match between this signature's parameters
1024 * and the supplied parameter list. If not, returns the name of the first
1025 * parameter with mismatched qualifiers (for use in error messages).
1026 */
1027 const char *qualifiers_match(exec_list *params);
1028
1029 /**
1030 * Replace the current parameter list with the given one. This is useful
1031 * if the current information came from a prototype, and either has invalid
1032 * or missing parameter names.
1033 */
1034 void replace_parameters(exec_list *new_params);
1035
1036 /**
1037 * Function return type.
1038 *
1039 * \note This discards the optional precision qualifier.
1040 */
1041 const struct glsl_type *return_type;
1042
1043 /**
1044 * List of ir_variable of function parameters.
1045 *
1046 * This represents the storage. The paramaters passed in a particular
1047 * call will be in ir_call::actual_paramaters.
1048 */
1049 struct exec_list parameters;
1050
1051 /** Whether or not this function has a body (which may be empty). */
1052 unsigned is_defined:1;
1053
1054 /** Whether or not this function signature is a built-in. */
1055 bool is_builtin() const;
1056
1057 /**
1058 * Whether or not this function is an intrinsic to be implemented
1059 * by the driver.
1060 */
1061 bool is_intrinsic;
1062
1063 /** Whether or not a built-in is available for this shader. */
1064 bool is_builtin_available(const _mesa_glsl_parse_state *state) const;
1065
1066 /** Body of instructions in the function. */
1067 struct exec_list body;
1068
1069 private:
1070 /**
1071 * A function pointer to a predicate that answers whether a built-in
1072 * function is available in the current shader. NULL if not a built-in.
1073 */
1074 builtin_available_predicate builtin_avail;
1075
1076 /** Function of which this signature is one overload. */
1077 class ir_function *_function;
1078
1079 /** Function signature of which this one is a prototype clone */
1080 const ir_function_signature *origin;
1081
1082 friend class ir_function;
1083
1084 /**
1085 * Helper function to run a list of instructions for constant
1086 * expression evaluation.
1087 *
1088 * The hash table represents the values of the visible variables.
1089 * There are no scoping issues because the table is indexed on
1090 * ir_variable pointers, not variable names.
1091 *
1092 * Returns false if the expression is not constant, true otherwise,
1093 * and the value in *result if result is non-NULL.
1094 */
1095 bool constant_expression_evaluate_expression_list(const struct exec_list &body,
1096 struct hash_table *variable_context,
1097 ir_constant **result);
1098 };
1099
1100
1101 /**
1102 * Header for tracking multiple overloaded functions with the same name.
1103 * Contains a list of ir_function_signatures representing each of the
1104 * actual functions.
1105 */
1106 class ir_function : public ir_instruction {
1107 public:
1108 ir_function(const char *name);
1109
1110 virtual ir_function *clone(void *mem_ctx, struct hash_table *ht) const;
1111
1112 virtual void accept(ir_visitor *v)
1113 {
1114 v->visit(this);
1115 }
1116
1117 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1118
1119 void add_signature(ir_function_signature *sig)
1120 {
1121 sig->_function = this;
1122 this->signatures.push_tail(sig);
1123 }
1124
1125 /**
1126 * Find a signature that matches a set of actual parameters, taking implicit
1127 * conversions into account. Also flags whether the match was exact.
1128 */
1129 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
1130 const exec_list *actual_param,
1131 bool allow_builtins,
1132 bool *match_is_exact);
1133
1134 /**
1135 * Find a signature that matches a set of actual parameters, taking implicit
1136 * conversions into account.
1137 */
1138 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
1139 const exec_list *actual_param,
1140 bool allow_builtins);
1141
1142 /**
1143 * Find a signature that exactly matches a set of actual parameters without
1144 * any implicit type conversions.
1145 */
1146 ir_function_signature *exact_matching_signature(_mesa_glsl_parse_state *state,
1147 const exec_list *actual_ps);
1148
1149 /**
1150 * Name of the function.
1151 */
1152 const char *name;
1153
1154 /** Whether or not this function has a signature that isn't a built-in. */
1155 bool has_user_signature();
1156
1157 /**
1158 * List of ir_function_signature for each overloaded function with this name.
1159 */
1160 struct exec_list signatures;
1161
1162 /**
1163 * is this function a subroutine type declaration
1164 * e.g. subroutine void type1(float arg1);
1165 */
1166 bool is_subroutine;
1167
1168 /**
1169 * is this function associated to a subroutine type
1170 * e.g. subroutine (type1, type2) function_name { function_body };
1171 * would have num_subroutine_types 2,
1172 * and pointers to the type1 and type2 types.
1173 */
1174 int num_subroutine_types;
1175 const struct glsl_type **subroutine_types;
1176
1177 int subroutine_index;
1178 };
1179
1180 inline const char *ir_function_signature::function_name() const
1181 {
1182 return this->_function->name;
1183 }
1184 /*@}*/
1185
1186
1187 /**
1188 * IR instruction representing high-level if-statements
1189 */
1190 class ir_if : public ir_instruction {
1191 public:
1192 ir_if(ir_rvalue *condition)
1193 : ir_instruction(ir_type_if), condition(condition)
1194 {
1195 }
1196
1197 virtual ir_if *clone(void *mem_ctx, struct hash_table *ht) const;
1198
1199 virtual void accept(ir_visitor *v)
1200 {
1201 v->visit(this);
1202 }
1203
1204 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1205
1206 ir_rvalue *condition;
1207 /** List of ir_instruction for the body of the then branch */
1208 exec_list then_instructions;
1209 /** List of ir_instruction for the body of the else branch */
1210 exec_list else_instructions;
1211 };
1212
1213
1214 /**
1215 * IR instruction representing a high-level loop structure.
1216 */
1217 class ir_loop : public ir_instruction {
1218 public:
1219 ir_loop();
1220
1221 virtual ir_loop *clone(void *mem_ctx, struct hash_table *ht) const;
1222
1223 virtual void accept(ir_visitor *v)
1224 {
1225 v->visit(this);
1226 }
1227
1228 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1229
1230 /** List of ir_instruction that make up the body of the loop. */
1231 exec_list body_instructions;
1232 };
1233
1234
1235 class ir_assignment : public ir_instruction {
1236 public:
1237 ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs, ir_rvalue *condition = NULL);
1238
1239 /**
1240 * Construct an assignment with an explicit write mask
1241 *
1242 * \note
1243 * Since a write mask is supplied, the LHS must already be a bare
1244 * \c ir_dereference. The cannot be any swizzles in the LHS.
1245 */
1246 ir_assignment(ir_dereference *lhs, ir_rvalue *rhs, ir_rvalue *condition,
1247 unsigned write_mask);
1248
1249 virtual ir_assignment *clone(void *mem_ctx, struct hash_table *ht) const;
1250
1251 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1252
1253 virtual void accept(ir_visitor *v)
1254 {
1255 v->visit(this);
1256 }
1257
1258 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1259
1260 /**
1261 * Get a whole variable written by an assignment
1262 *
1263 * If the LHS of the assignment writes a whole variable, the variable is
1264 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1265 * assignment are:
1266 *
1267 * - Assigning to a scalar
1268 * - Assigning to all components of a vector
1269 * - Whole array (or matrix) assignment
1270 * - Whole structure assignment
1271 */
1272 ir_variable *whole_variable_written();
1273
1274 /**
1275 * Set the LHS of an assignment
1276 */
1277 void set_lhs(ir_rvalue *lhs);
1278
1279 /**
1280 * Left-hand side of the assignment.
1281 *
1282 * This should be treated as read only. If you need to set the LHS of an
1283 * assignment, use \c ir_assignment::set_lhs.
1284 */
1285 ir_dereference *lhs;
1286
1287 /**
1288 * Value being assigned
1289 */
1290 ir_rvalue *rhs;
1291
1292 /**
1293 * Optional condition for the assignment.
1294 */
1295 ir_rvalue *condition;
1296
1297
1298 /**
1299 * Component mask written
1300 *
1301 * For non-vector types in the LHS, this field will be zero. For vector
1302 * types, a bit will be set for each component that is written. Note that
1303 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1304 *
1305 * A partially-set write mask means that each enabled channel gets
1306 * the value from a consecutive channel of the rhs. For example,
1307 * to write just .xyw of gl_FrontColor with color:
1308 *
1309 * (assign (constant bool (1)) (xyw)
1310 * (var_ref gl_FragColor)
1311 * (swiz xyw (var_ref color)))
1312 */
1313 unsigned write_mask:4;
1314 };
1315
1316 /* Update ir_expression::get_num_operands() and operator_strs when
1317 * updating this list.
1318 */
1319 enum ir_expression_operation {
1320 ir_unop_bit_not,
1321 ir_unop_logic_not,
1322 ir_unop_neg,
1323 ir_unop_abs,
1324 ir_unop_sign,
1325 ir_unop_rcp,
1326 ir_unop_rsq,
1327 ir_unop_sqrt,
1328 ir_unop_exp, /**< Log base e on gentype */
1329 ir_unop_log, /**< Natural log on gentype */
1330 ir_unop_exp2,
1331 ir_unop_log2,
1332 ir_unop_f2i, /**< Float-to-integer conversion. */
1333 ir_unop_f2u, /**< Float-to-unsigned conversion. */
1334 ir_unop_i2f, /**< Integer-to-float conversion. */
1335 ir_unop_f2b, /**< Float-to-boolean conversion */
1336 ir_unop_b2f, /**< Boolean-to-float conversion */
1337 ir_unop_i2b, /**< int-to-boolean conversion */
1338 ir_unop_b2i, /**< Boolean-to-int conversion */
1339 ir_unop_u2f, /**< Unsigned-to-float conversion. */
1340 ir_unop_i2u, /**< Integer-to-unsigned conversion. */
1341 ir_unop_u2i, /**< Unsigned-to-integer conversion. */
1342 ir_unop_d2f, /**< Double-to-float conversion. */
1343 ir_unop_f2d, /**< Float-to-double conversion. */
1344 ir_unop_d2i, /**< Double-to-integer conversion. */
1345 ir_unop_i2d, /**< Integer-to-double conversion. */
1346 ir_unop_d2u, /**< Double-to-unsigned conversion. */
1347 ir_unop_u2d, /**< Unsigned-to-double conversion. */
1348 ir_unop_d2b, /**< Double-to-boolean conversion. */
1349 ir_unop_bitcast_i2f, /**< Bit-identical int-to-float "conversion" */
1350 ir_unop_bitcast_f2i, /**< Bit-identical float-to-int "conversion" */
1351 ir_unop_bitcast_u2f, /**< Bit-identical uint-to-float "conversion" */
1352 ir_unop_bitcast_f2u, /**< Bit-identical float-to-uint "conversion" */
1353
1354 /**
1355 * \name Unary floating-point rounding operations.
1356 */
1357 /*@{*/
1358 ir_unop_trunc,
1359 ir_unop_ceil,
1360 ir_unop_floor,
1361 ir_unop_fract,
1362 ir_unop_round_even,
1363 /*@}*/
1364
1365 /**
1366 * \name Trigonometric operations.
1367 */
1368 /*@{*/
1369 ir_unop_sin,
1370 ir_unop_cos,
1371 /*@}*/
1372
1373 /**
1374 * \name Partial derivatives.
1375 */
1376 /*@{*/
1377 ir_unop_dFdx,
1378 ir_unop_dFdx_coarse,
1379 ir_unop_dFdx_fine,
1380 ir_unop_dFdy,
1381 ir_unop_dFdy_coarse,
1382 ir_unop_dFdy_fine,
1383 /*@}*/
1384
1385 /**
1386 * \name Floating point pack and unpack operations.
1387 */
1388 /*@{*/
1389 ir_unop_pack_snorm_2x16,
1390 ir_unop_pack_snorm_4x8,
1391 ir_unop_pack_unorm_2x16,
1392 ir_unop_pack_unorm_4x8,
1393 ir_unop_pack_half_2x16,
1394 ir_unop_unpack_snorm_2x16,
1395 ir_unop_unpack_snorm_4x8,
1396 ir_unop_unpack_unorm_2x16,
1397 ir_unop_unpack_unorm_4x8,
1398 ir_unop_unpack_half_2x16,
1399 /*@}*/
1400
1401 /**
1402 * \name Bit operations, part of ARB_gpu_shader5.
1403 */
1404 /*@{*/
1405 ir_unop_bitfield_reverse,
1406 ir_unop_bit_count,
1407 ir_unop_find_msb,
1408 ir_unop_find_lsb,
1409 /*@}*/
1410
1411 ir_unop_saturate,
1412
1413 /**
1414 * \name Double packing, part of ARB_gpu_shader_fp64.
1415 */
1416 /*@{*/
1417 ir_unop_pack_double_2x32,
1418 ir_unop_unpack_double_2x32,
1419 /*@}*/
1420
1421 ir_unop_frexp_sig,
1422 ir_unop_frexp_exp,
1423
1424 ir_unop_noise,
1425
1426 ir_unop_subroutine_to_int,
1427 /**
1428 * Interpolate fs input at centroid
1429 *
1430 * operand0 is the fs input.
1431 */
1432 ir_unop_interpolate_at_centroid,
1433
1434 /**
1435 * Ask the driver for the total size of a buffer block.
1436 *
1437 * operand0 is the ir_constant buffer block index in the linked shader.
1438 */
1439 ir_unop_get_buffer_size,
1440
1441 /**
1442 * Calculate length of an unsized array inside a buffer block.
1443 * This opcode is going to be replaced in a lowering pass inside
1444 * the linker.
1445 *
1446 * operand0 is the unsized array's ir_value for the calculation
1447 * of its length.
1448 */
1449 ir_unop_ssbo_unsized_array_length,
1450
1451 /**
1452 * A sentinel marking the last of the unary operations.
1453 */
1454 ir_last_unop = ir_unop_ssbo_unsized_array_length,
1455
1456 ir_binop_add,
1457 ir_binop_sub,
1458 ir_binop_mul, /**< Floating-point or low 32-bit integer multiply. */
1459 ir_binop_imul_high, /**< Calculates the high 32-bits of a 64-bit multiply. */
1460 ir_binop_div,
1461
1462 /**
1463 * Returns the carry resulting from the addition of the two arguments.
1464 */
1465 /*@{*/
1466 ir_binop_carry,
1467 /*@}*/
1468
1469 /**
1470 * Returns the borrow resulting from the subtraction of the second argument
1471 * from the first argument.
1472 */
1473 /*@{*/
1474 ir_binop_borrow,
1475 /*@}*/
1476
1477 /**
1478 * Takes one of two combinations of arguments:
1479 *
1480 * - mod(vecN, vecN)
1481 * - mod(vecN, float)
1482 *
1483 * Does not take integer types.
1484 */
1485 ir_binop_mod,
1486
1487 /**
1488 * \name Binary comparison operators which return a boolean vector.
1489 * The type of both operands must be equal.
1490 */
1491 /*@{*/
1492 ir_binop_less,
1493 ir_binop_greater,
1494 ir_binop_lequal,
1495 ir_binop_gequal,
1496 ir_binop_equal,
1497 ir_binop_nequal,
1498 /**
1499 * Returns single boolean for whether all components of operands[0]
1500 * equal the components of operands[1].
1501 */
1502 ir_binop_all_equal,
1503 /**
1504 * Returns single boolean for whether any component of operands[0]
1505 * is not equal to the corresponding component of operands[1].
1506 */
1507 ir_binop_any_nequal,
1508 /*@}*/
1509
1510 /**
1511 * \name Bit-wise binary operations.
1512 */
1513 /*@{*/
1514 ir_binop_lshift,
1515 ir_binop_rshift,
1516 ir_binop_bit_and,
1517 ir_binop_bit_xor,
1518 ir_binop_bit_or,
1519 /*@}*/
1520
1521 ir_binop_logic_and,
1522 ir_binop_logic_xor,
1523 ir_binop_logic_or,
1524
1525 ir_binop_dot,
1526 ir_binop_min,
1527 ir_binop_max,
1528
1529 ir_binop_pow,
1530
1531 /**
1532 * Load a value the size of a given GLSL type from a uniform block.
1533 *
1534 * operand0 is the ir_constant uniform block index in the linked shader.
1535 * operand1 is a byte offset within the uniform block.
1536 */
1537 ir_binop_ubo_load,
1538
1539 /**
1540 * \name Multiplies a number by two to a power, part of ARB_gpu_shader5.
1541 */
1542 /*@{*/
1543 ir_binop_ldexp,
1544 /*@}*/
1545
1546 /**
1547 * Extract a scalar from a vector
1548 *
1549 * operand0 is the vector
1550 * operand1 is the index of the field to read from operand0
1551 */
1552 ir_binop_vector_extract,
1553
1554 /**
1555 * Interpolate fs input at offset
1556 *
1557 * operand0 is the fs input
1558 * operand1 is the offset from the pixel center
1559 */
1560 ir_binop_interpolate_at_offset,
1561
1562 /**
1563 * Interpolate fs input at sample position
1564 *
1565 * operand0 is the fs input
1566 * operand1 is the sample ID
1567 */
1568 ir_binop_interpolate_at_sample,
1569
1570 /**
1571 * A sentinel marking the last of the binary operations.
1572 */
1573 ir_last_binop = ir_binop_interpolate_at_sample,
1574
1575 /**
1576 * \name Fused floating-point multiply-add, part of ARB_gpu_shader5.
1577 */
1578 /*@{*/
1579 ir_triop_fma,
1580 /*@}*/
1581
1582 ir_triop_lrp,
1583
1584 /**
1585 * \name Conditional Select
1586 *
1587 * A vector conditional select instruction (like ?:, but operating per-
1588 * component on vectors).
1589 *
1590 * \see lower_instructions_visitor::ldexp_to_arith
1591 */
1592 /*@{*/
1593 ir_triop_csel,
1594 /*@}*/
1595
1596 ir_triop_bitfield_extract,
1597
1598 /**
1599 * Generate a value with one field of a vector changed
1600 *
1601 * operand0 is the vector
1602 * operand1 is the value to write into the vector result
1603 * operand2 is the index in operand0 to be modified
1604 */
1605 ir_triop_vector_insert,
1606
1607 /**
1608 * A sentinel marking the last of the ternary operations.
1609 */
1610 ir_last_triop = ir_triop_vector_insert,
1611
1612 ir_quadop_bitfield_insert,
1613
1614 ir_quadop_vector,
1615
1616 /**
1617 * A sentinel marking the last of the ternary operations.
1618 */
1619 ir_last_quadop = ir_quadop_vector,
1620
1621 /**
1622 * A sentinel marking the last of all operations.
1623 */
1624 ir_last_opcode = ir_quadop_vector
1625 };
1626
1627 class ir_expression : public ir_rvalue {
1628 public:
1629 ir_expression(int op, const struct glsl_type *type,
1630 ir_rvalue *op0, ir_rvalue *op1 = NULL,
1631 ir_rvalue *op2 = NULL, ir_rvalue *op3 = NULL);
1632
1633 /**
1634 * Constructor for unary operation expressions
1635 */
1636 ir_expression(int op, ir_rvalue *);
1637
1638 /**
1639 * Constructor for binary operation expressions
1640 */
1641 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1);
1642
1643 /**
1644 * Constructor for ternary operation expressions
1645 */
1646 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1, ir_rvalue *op2);
1647
1648 virtual bool equals(const ir_instruction *ir,
1649 enum ir_node_type ignore = ir_type_unset) const;
1650
1651 virtual ir_expression *clone(void *mem_ctx, struct hash_table *ht) const;
1652
1653 /**
1654 * Attempt to constant-fold the expression
1655 *
1656 * The "variable_context" hash table links ir_variable * to ir_constant *
1657 * that represent the variables' values. \c NULL represents an empty
1658 * context.
1659 *
1660 * If the expression cannot be constant folded, this method will return
1661 * \c NULL.
1662 */
1663 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1664
1665 /**
1666 * Determine the number of operands used by an expression
1667 */
1668 static unsigned int get_num_operands(ir_expression_operation);
1669
1670 /**
1671 * Determine the number of operands used by an expression
1672 */
1673 unsigned int get_num_operands() const
1674 {
1675 return (this->operation == ir_quadop_vector)
1676 ? this->type->vector_elements : get_num_operands(operation);
1677 }
1678
1679 /**
1680 * Return whether the expression operates on vectors horizontally.
1681 */
1682 bool is_horizontal() const
1683 {
1684 return operation == ir_binop_all_equal ||
1685 operation == ir_binop_any_nequal ||
1686 operation == ir_binop_dot ||
1687 operation == ir_binop_vector_extract ||
1688 operation == ir_triop_vector_insert ||
1689 operation == ir_quadop_vector;
1690 }
1691
1692 /**
1693 * Return a string representing this expression's operator.
1694 */
1695 const char *operator_string();
1696
1697 /**
1698 * Return a string representing this expression's operator.
1699 */
1700 static const char *operator_string(ir_expression_operation);
1701
1702
1703 /**
1704 * Do a reverse-lookup to translate the given string into an operator.
1705 */
1706 static ir_expression_operation get_operator(const char *);
1707
1708 virtual void accept(ir_visitor *v)
1709 {
1710 v->visit(this);
1711 }
1712
1713 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1714
1715 virtual ir_variable *variable_referenced() const;
1716
1717 ir_expression_operation operation;
1718 ir_rvalue *operands[4];
1719 };
1720
1721
1722 /**
1723 * HIR instruction representing a high-level function call, containing a list
1724 * of parameters and returning a value in the supplied temporary.
1725 */
1726 class ir_call : public ir_instruction {
1727 public:
1728 ir_call(ir_function_signature *callee,
1729 ir_dereference_variable *return_deref,
1730 exec_list *actual_parameters)
1731 : ir_instruction(ir_type_call), return_deref(return_deref), callee(callee), sub_var(NULL), array_idx(NULL)
1732 {
1733 assert(callee->return_type != NULL);
1734 actual_parameters->move_nodes_to(& this->actual_parameters);
1735 this->use_builtin = callee->is_builtin();
1736 }
1737
1738 ir_call(ir_function_signature *callee,
1739 ir_dereference_variable *return_deref,
1740 exec_list *actual_parameters,
1741 ir_variable *var, ir_rvalue *array_idx)
1742 : ir_instruction(ir_type_call), return_deref(return_deref), callee(callee), sub_var(var), array_idx(array_idx)
1743 {
1744 assert(callee->return_type != NULL);
1745 actual_parameters->move_nodes_to(& this->actual_parameters);
1746 this->use_builtin = callee->is_builtin();
1747 }
1748
1749 virtual ir_call *clone(void *mem_ctx, struct hash_table *ht) const;
1750
1751 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1752
1753 virtual void accept(ir_visitor *v)
1754 {
1755 v->visit(this);
1756 }
1757
1758 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1759
1760 /**
1761 * Get the name of the function being called.
1762 */
1763 const char *callee_name() const
1764 {
1765 return callee->function_name();
1766 }
1767
1768 /**
1769 * Generates an inline version of the function before @ir,
1770 * storing the return value in return_deref.
1771 */
1772 void generate_inline(ir_instruction *ir);
1773
1774 /**
1775 * Storage for the function's return value.
1776 * This must be NULL if the return type is void.
1777 */
1778 ir_dereference_variable *return_deref;
1779
1780 /**
1781 * The specific function signature being called.
1782 */
1783 ir_function_signature *callee;
1784
1785 /* List of ir_rvalue of paramaters passed in this call. */
1786 exec_list actual_parameters;
1787
1788 /** Should this call only bind to a built-in function? */
1789 bool use_builtin;
1790
1791 /*
1792 * ARB_shader_subroutine support -
1793 * the subroutine uniform variable and array index
1794 * rvalue to be used in the lowering pass later.
1795 */
1796 ir_variable *sub_var;
1797 ir_rvalue *array_idx;
1798 };
1799
1800
1801 /**
1802 * \name Jump-like IR instructions.
1803 *
1804 * These include \c break, \c continue, \c return, and \c discard.
1805 */
1806 /*@{*/
1807 class ir_jump : public ir_instruction {
1808 protected:
1809 ir_jump(enum ir_node_type t)
1810 : ir_instruction(t)
1811 {
1812 }
1813 };
1814
1815 class ir_return : public ir_jump {
1816 public:
1817 ir_return()
1818 : ir_jump(ir_type_return), value(NULL)
1819 {
1820 }
1821
1822 ir_return(ir_rvalue *value)
1823 : ir_jump(ir_type_return), value(value)
1824 {
1825 }
1826
1827 virtual ir_return *clone(void *mem_ctx, struct hash_table *) const;
1828
1829 ir_rvalue *get_value() const
1830 {
1831 return value;
1832 }
1833
1834 virtual void accept(ir_visitor *v)
1835 {
1836 v->visit(this);
1837 }
1838
1839 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1840
1841 ir_rvalue *value;
1842 };
1843
1844
1845 /**
1846 * Jump instructions used inside loops
1847 *
1848 * These include \c break and \c continue. The \c break within a loop is
1849 * different from the \c break within a switch-statement.
1850 *
1851 * \sa ir_switch_jump
1852 */
1853 class ir_loop_jump : public ir_jump {
1854 public:
1855 enum jump_mode {
1856 jump_break,
1857 jump_continue
1858 };
1859
1860 ir_loop_jump(jump_mode mode)
1861 : ir_jump(ir_type_loop_jump)
1862 {
1863 this->mode = mode;
1864 }
1865
1866 virtual ir_loop_jump *clone(void *mem_ctx, struct hash_table *) const;
1867
1868 virtual void accept(ir_visitor *v)
1869 {
1870 v->visit(this);
1871 }
1872
1873 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1874
1875 bool is_break() const
1876 {
1877 return mode == jump_break;
1878 }
1879
1880 bool is_continue() const
1881 {
1882 return mode == jump_continue;
1883 }
1884
1885 /** Mode selector for the jump instruction. */
1886 enum jump_mode mode;
1887 };
1888
1889 /**
1890 * IR instruction representing discard statements.
1891 */
1892 class ir_discard : public ir_jump {
1893 public:
1894 ir_discard()
1895 : ir_jump(ir_type_discard)
1896 {
1897 this->condition = NULL;
1898 }
1899
1900 ir_discard(ir_rvalue *cond)
1901 : ir_jump(ir_type_discard)
1902 {
1903 this->condition = cond;
1904 }
1905
1906 virtual ir_discard *clone(void *mem_ctx, struct hash_table *ht) const;
1907
1908 virtual void accept(ir_visitor *v)
1909 {
1910 v->visit(this);
1911 }
1912
1913 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1914
1915 ir_rvalue *condition;
1916 };
1917 /*@}*/
1918
1919
1920 /**
1921 * Texture sampling opcodes used in ir_texture
1922 */
1923 enum ir_texture_opcode {
1924 ir_tex, /**< Regular texture look-up */
1925 ir_txb, /**< Texture look-up with LOD bias */
1926 ir_txl, /**< Texture look-up with explicit LOD */
1927 ir_txd, /**< Texture look-up with partial derivatvies */
1928 ir_txf, /**< Texel fetch with explicit LOD */
1929 ir_txf_ms, /**< Multisample texture fetch */
1930 ir_txs, /**< Texture size */
1931 ir_lod, /**< Texture lod query */
1932 ir_tg4, /**< Texture gather */
1933 ir_query_levels, /**< Texture levels query */
1934 ir_texture_samples, /**< Texture samples query */
1935 ir_samples_identical, /**< Query whether all samples are definitely identical. */
1936 };
1937
1938
1939 /**
1940 * IR instruction to sample a texture
1941 *
1942 * The specific form of the IR instruction depends on the \c mode value
1943 * selected from \c ir_texture_opcodes. In the printed IR, these will
1944 * appear as:
1945 *
1946 * Texel offset (0 or an expression)
1947 * | Projection divisor
1948 * | | Shadow comparitor
1949 * | | |
1950 * v v v
1951 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1952 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1953 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1954 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1955 * (txf <type> <sampler> <coordinate> 0 <lod>)
1956 * (txf_ms
1957 * <type> <sampler> <coordinate> <sample_index>)
1958 * (txs <type> <sampler> <lod>)
1959 * (lod <type> <sampler> <coordinate>)
1960 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
1961 * (query_levels <type> <sampler>)
1962 * (samples_identical <sampler> <coordinate>)
1963 */
1964 class ir_texture : public ir_rvalue {
1965 public:
1966 ir_texture(enum ir_texture_opcode op)
1967 : ir_rvalue(ir_type_texture),
1968 op(op), sampler(NULL), coordinate(NULL), projector(NULL),
1969 shadow_comparitor(NULL), offset(NULL)
1970 {
1971 memset(&lod_info, 0, sizeof(lod_info));
1972 }
1973
1974 virtual ir_texture *clone(void *mem_ctx, struct hash_table *) const;
1975
1976 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1977
1978 virtual void accept(ir_visitor *v)
1979 {
1980 v->visit(this);
1981 }
1982
1983 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1984
1985 virtual bool equals(const ir_instruction *ir,
1986 enum ir_node_type ignore = ir_type_unset) const;
1987
1988 /**
1989 * Return a string representing the ir_texture_opcode.
1990 */
1991 const char *opcode_string();
1992
1993 /** Set the sampler and type. */
1994 void set_sampler(ir_dereference *sampler, const glsl_type *type);
1995
1996 /**
1997 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1998 */
1999 static ir_texture_opcode get_opcode(const char *);
2000
2001 enum ir_texture_opcode op;
2002
2003 /** Sampler to use for the texture access. */
2004 ir_dereference *sampler;
2005
2006 /** Texture coordinate to sample */
2007 ir_rvalue *coordinate;
2008
2009 /**
2010 * Value used for projective divide.
2011 *
2012 * If there is no projective divide (the common case), this will be
2013 * \c NULL. Optimization passes should check for this to point to a constant
2014 * of 1.0 and replace that with \c NULL.
2015 */
2016 ir_rvalue *projector;
2017
2018 /**
2019 * Coordinate used for comparison on shadow look-ups.
2020 *
2021 * If there is no shadow comparison, this will be \c NULL. For the
2022 * \c ir_txf opcode, this *must* be \c NULL.
2023 */
2024 ir_rvalue *shadow_comparitor;
2025
2026 /** Texel offset. */
2027 ir_rvalue *offset;
2028
2029 union {
2030 ir_rvalue *lod; /**< Floating point LOD */
2031 ir_rvalue *bias; /**< Floating point LOD bias */
2032 ir_rvalue *sample_index; /**< MSAA sample index */
2033 ir_rvalue *component; /**< Gather component selector */
2034 struct {
2035 ir_rvalue *dPdx; /**< Partial derivative of coordinate wrt X */
2036 ir_rvalue *dPdy; /**< Partial derivative of coordinate wrt Y */
2037 } grad;
2038 } lod_info;
2039 };
2040
2041
2042 struct ir_swizzle_mask {
2043 unsigned x:2;
2044 unsigned y:2;
2045 unsigned z:2;
2046 unsigned w:2;
2047
2048 /**
2049 * Number of components in the swizzle.
2050 */
2051 unsigned num_components:3;
2052
2053 /**
2054 * Does the swizzle contain duplicate components?
2055 *
2056 * L-value swizzles cannot contain duplicate components.
2057 */
2058 unsigned has_duplicates:1;
2059 };
2060
2061
2062 class ir_swizzle : public ir_rvalue {
2063 public:
2064 ir_swizzle(ir_rvalue *, unsigned x, unsigned y, unsigned z, unsigned w,
2065 unsigned count);
2066
2067 ir_swizzle(ir_rvalue *val, const unsigned *components, unsigned count);
2068
2069 ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask);
2070
2071 virtual ir_swizzle *clone(void *mem_ctx, struct hash_table *) const;
2072
2073 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
2074
2075 /**
2076 * Construct an ir_swizzle from the textual representation. Can fail.
2077 */
2078 static ir_swizzle *create(ir_rvalue *, const char *, unsigned vector_length);
2079
2080 virtual void accept(ir_visitor *v)
2081 {
2082 v->visit(this);
2083 }
2084
2085 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2086
2087 virtual bool equals(const ir_instruction *ir,
2088 enum ir_node_type ignore = ir_type_unset) const;
2089
2090 bool is_lvalue() const
2091 {
2092 return val->is_lvalue() && !mask.has_duplicates;
2093 }
2094
2095 /**
2096 * Get the variable that is ultimately referenced by an r-value
2097 */
2098 virtual ir_variable *variable_referenced() const;
2099
2100 ir_rvalue *val;
2101 ir_swizzle_mask mask;
2102
2103 private:
2104 /**
2105 * Initialize the mask component of a swizzle
2106 *
2107 * This is used by the \c ir_swizzle constructors.
2108 */
2109 void init_mask(const unsigned *components, unsigned count);
2110 };
2111
2112
2113 class ir_dereference : public ir_rvalue {
2114 public:
2115 virtual ir_dereference *clone(void *mem_ctx, struct hash_table *) const = 0;
2116
2117 bool is_lvalue() const;
2118
2119 /**
2120 * Get the variable that is ultimately referenced by an r-value
2121 */
2122 virtual ir_variable *variable_referenced() const = 0;
2123
2124 protected:
2125 ir_dereference(enum ir_node_type t)
2126 : ir_rvalue(t)
2127 {
2128 }
2129 };
2130
2131
2132 class ir_dereference_variable : public ir_dereference {
2133 public:
2134 ir_dereference_variable(ir_variable *var);
2135
2136 virtual ir_dereference_variable *clone(void *mem_ctx,
2137 struct hash_table *) const;
2138
2139 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
2140
2141 virtual bool equals(const ir_instruction *ir,
2142 enum ir_node_type ignore = ir_type_unset) const;
2143
2144 /**
2145 * Get the variable that is ultimately referenced by an r-value
2146 */
2147 virtual ir_variable *variable_referenced() const
2148 {
2149 return this->var;
2150 }
2151
2152 virtual ir_variable *whole_variable_referenced()
2153 {
2154 /* ir_dereference_variable objects always dereference the entire
2155 * variable. However, if this dereference is dereferenced by anything
2156 * else, the complete deferefernce chain is not a whole-variable
2157 * dereference. This method should only be called on the top most
2158 * ir_rvalue in a dereference chain.
2159 */
2160 return this->var;
2161 }
2162
2163 virtual void accept(ir_visitor *v)
2164 {
2165 v->visit(this);
2166 }
2167
2168 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2169
2170 /**
2171 * Object being dereferenced.
2172 */
2173 ir_variable *var;
2174 };
2175
2176
2177 class ir_dereference_array : public ir_dereference {
2178 public:
2179 ir_dereference_array(ir_rvalue *value, ir_rvalue *array_index);
2180
2181 ir_dereference_array(ir_variable *var, ir_rvalue *array_index);
2182
2183 virtual ir_dereference_array *clone(void *mem_ctx,
2184 struct hash_table *) const;
2185
2186 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
2187
2188 virtual bool equals(const ir_instruction *ir,
2189 enum ir_node_type ignore = ir_type_unset) const;
2190
2191 /**
2192 * Get the variable that is ultimately referenced by an r-value
2193 */
2194 virtual ir_variable *variable_referenced() const
2195 {
2196 return this->array->variable_referenced();
2197 }
2198
2199 virtual void accept(ir_visitor *v)
2200 {
2201 v->visit(this);
2202 }
2203
2204 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2205
2206 ir_rvalue *array;
2207 ir_rvalue *array_index;
2208
2209 private:
2210 void set_array(ir_rvalue *value);
2211 };
2212
2213
2214 class ir_dereference_record : public ir_dereference {
2215 public:
2216 ir_dereference_record(ir_rvalue *value, const char *field);
2217
2218 ir_dereference_record(ir_variable *var, const char *field);
2219
2220 virtual ir_dereference_record *clone(void *mem_ctx,
2221 struct hash_table *) const;
2222
2223 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
2224
2225 /**
2226 * Get the variable that is ultimately referenced by an r-value
2227 */
2228 virtual ir_variable *variable_referenced() const
2229 {
2230 return this->record->variable_referenced();
2231 }
2232
2233 virtual void accept(ir_visitor *v)
2234 {
2235 v->visit(this);
2236 }
2237
2238 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2239
2240 ir_rvalue *record;
2241 const char *field;
2242 };
2243
2244
2245 /**
2246 * Data stored in an ir_constant
2247 */
2248 union ir_constant_data {
2249 unsigned u[16];
2250 int i[16];
2251 float f[16];
2252 bool b[16];
2253 double d[16];
2254 };
2255
2256
2257 class ir_constant : public ir_rvalue {
2258 public:
2259 ir_constant(const struct glsl_type *type, const ir_constant_data *data);
2260 ir_constant(bool b, unsigned vector_elements=1);
2261 ir_constant(unsigned int u, unsigned vector_elements=1);
2262 ir_constant(int i, unsigned vector_elements=1);
2263 ir_constant(float f, unsigned vector_elements=1);
2264 ir_constant(double d, unsigned vector_elements=1);
2265
2266 /**
2267 * Construct an ir_constant from a list of ir_constant values
2268 */
2269 ir_constant(const struct glsl_type *type, exec_list *values);
2270
2271 /**
2272 * Construct an ir_constant from a scalar component of another ir_constant
2273 *
2274 * The new \c ir_constant inherits the type of the component from the
2275 * source constant.
2276 *
2277 * \note
2278 * In the case of a matrix constant, the new constant is a scalar, \b not
2279 * a vector.
2280 */
2281 ir_constant(const ir_constant *c, unsigned i);
2282
2283 /**
2284 * Return a new ir_constant of the specified type containing all zeros.
2285 */
2286 static ir_constant *zero(void *mem_ctx, const glsl_type *type);
2287
2288 virtual ir_constant *clone(void *mem_ctx, struct hash_table *) const;
2289
2290 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
2291
2292 virtual void accept(ir_visitor *v)
2293 {
2294 v->visit(this);
2295 }
2296
2297 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2298
2299 virtual bool equals(const ir_instruction *ir,
2300 enum ir_node_type ignore = ir_type_unset) const;
2301
2302 /**
2303 * Get a particular component of a constant as a specific type
2304 *
2305 * This is useful, for example, to get a value from an integer constant
2306 * as a float or bool. This appears frequently when constructors are
2307 * called with all constant parameters.
2308 */
2309 /*@{*/
2310 bool get_bool_component(unsigned i) const;
2311 float get_float_component(unsigned i) const;
2312 double get_double_component(unsigned i) const;
2313 int get_int_component(unsigned i) const;
2314 unsigned get_uint_component(unsigned i) const;
2315 /*@}*/
2316
2317 ir_constant *get_array_element(unsigned i) const;
2318
2319 ir_constant *get_record_field(const char *name);
2320
2321 /**
2322 * Copy the values on another constant at a given offset.
2323 *
2324 * The offset is ignored for array or struct copies, it's only for
2325 * scalars or vectors into vectors or matrices.
2326 *
2327 * With identical types on both sides and zero offset it's clone()
2328 * without creating a new object.
2329 */
2330
2331 void copy_offset(ir_constant *src, int offset);
2332
2333 /**
2334 * Copy the values on another constant at a given offset and
2335 * following an assign-like mask.
2336 *
2337 * The mask is ignored for scalars.
2338 *
2339 * Note that this function only handles what assign can handle,
2340 * i.e. at most a vector as source and a column of a matrix as
2341 * destination.
2342 */
2343
2344 void copy_masked_offset(ir_constant *src, int offset, unsigned int mask);
2345
2346 /**
2347 * Determine whether a constant has the same value as another constant
2348 *
2349 * \sa ir_constant::is_zero, ir_constant::is_one,
2350 * ir_constant::is_negative_one
2351 */
2352 bool has_value(const ir_constant *) const;
2353
2354 /**
2355 * Return true if this ir_constant represents the given value.
2356 *
2357 * For vectors, this checks that each component is the given value.
2358 */
2359 virtual bool is_value(float f, int i) const;
2360 virtual bool is_zero() const;
2361 virtual bool is_one() const;
2362 virtual bool is_negative_one() const;
2363
2364 /**
2365 * Return true for constants that could be stored as 16-bit unsigned values.
2366 *
2367 * Note that this will return true even for signed integer ir_constants, as
2368 * long as the value is non-negative and fits in 16-bits.
2369 */
2370 virtual bool is_uint16_constant() const;
2371
2372 /**
2373 * Value of the constant.
2374 *
2375 * The field used to back the values supplied by the constant is determined
2376 * by the type associated with the \c ir_instruction. Constants may be
2377 * scalars, vectors, or matrices.
2378 */
2379 union ir_constant_data value;
2380
2381 /* Array elements */
2382 ir_constant **array_elements;
2383
2384 /* Structure fields */
2385 exec_list components;
2386
2387 private:
2388 /**
2389 * Parameterless constructor only used by the clone method
2390 */
2391 ir_constant(void);
2392 };
2393
2394 /**
2395 * IR instruction to emit a vertex in a geometry shader.
2396 */
2397 class ir_emit_vertex : public ir_instruction {
2398 public:
2399 ir_emit_vertex(ir_rvalue *stream)
2400 : ir_instruction(ir_type_emit_vertex),
2401 stream(stream)
2402 {
2403 assert(stream);
2404 }
2405
2406 virtual void accept(ir_visitor *v)
2407 {
2408 v->visit(this);
2409 }
2410
2411 virtual ir_emit_vertex *clone(void *mem_ctx, struct hash_table *ht) const
2412 {
2413 return new(mem_ctx) ir_emit_vertex(this->stream->clone(mem_ctx, ht));
2414 }
2415
2416 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2417
2418 int stream_id() const
2419 {
2420 return stream->as_constant()->value.i[0];
2421 }
2422
2423 ir_rvalue *stream;
2424 };
2425
2426 /**
2427 * IR instruction to complete the current primitive and start a new one in a
2428 * geometry shader.
2429 */
2430 class ir_end_primitive : public ir_instruction {
2431 public:
2432 ir_end_primitive(ir_rvalue *stream)
2433 : ir_instruction(ir_type_end_primitive),
2434 stream(stream)
2435 {
2436 assert(stream);
2437 }
2438
2439 virtual void accept(ir_visitor *v)
2440 {
2441 v->visit(this);
2442 }
2443
2444 virtual ir_end_primitive *clone(void *mem_ctx, struct hash_table *ht) const
2445 {
2446 return new(mem_ctx) ir_end_primitive(this->stream->clone(mem_ctx, ht));
2447 }
2448
2449 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2450
2451 int stream_id() const
2452 {
2453 return stream->as_constant()->value.i[0];
2454 }
2455
2456 ir_rvalue *stream;
2457 };
2458
2459 /**
2460 * IR instruction for tessellation control and compute shader barrier.
2461 */
2462 class ir_barrier : public ir_instruction {
2463 public:
2464 ir_barrier()
2465 : ir_instruction(ir_type_barrier)
2466 {
2467 }
2468
2469 virtual void accept(ir_visitor *v)
2470 {
2471 v->visit(this);
2472 }
2473
2474 virtual ir_barrier *clone(void *mem_ctx, struct hash_table *) const
2475 {
2476 return new(mem_ctx) ir_barrier();
2477 }
2478
2479 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2480 };
2481
2482 /*@}*/
2483
2484 /**
2485 * Apply a visitor to each IR node in a list
2486 */
2487 void
2488 visit_exec_list(exec_list *list, ir_visitor *visitor);
2489
2490 /**
2491 * Validate invariants on each IR node in a list
2492 */
2493 void validate_ir_tree(exec_list *instructions);
2494
2495 struct _mesa_glsl_parse_state;
2496 struct gl_shader_program;
2497
2498 /**
2499 * Detect whether an unlinked shader contains static recursion
2500 *
2501 * If the list of instructions is determined to contain static recursion,
2502 * \c _mesa_glsl_error will be called to emit error messages for each function
2503 * that is in the recursion cycle.
2504 */
2505 void
2506 detect_recursion_unlinked(struct _mesa_glsl_parse_state *state,
2507 exec_list *instructions);
2508
2509 /**
2510 * Detect whether a linked shader contains static recursion
2511 *
2512 * If the list of instructions is determined to contain static recursion,
2513 * \c link_error_printf will be called to emit error messages for each function
2514 * that is in the recursion cycle. In addition,
2515 * \c gl_shader_program::LinkStatus will be set to false.
2516 */
2517 void
2518 detect_recursion_linked(struct gl_shader_program *prog,
2519 exec_list *instructions);
2520
2521 /**
2522 * Make a clone of each IR instruction in a list
2523 *
2524 * \param in List of IR instructions that are to be cloned
2525 * \param out List to hold the cloned instructions
2526 */
2527 void
2528 clone_ir_list(void *mem_ctx, exec_list *out, const exec_list *in);
2529
2530 extern void
2531 _mesa_glsl_initialize_variables(exec_list *instructions,
2532 struct _mesa_glsl_parse_state *state);
2533
2534 extern void
2535 _mesa_glsl_initialize_derived_variables(gl_shader *shader);
2536
2537 extern void
2538 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state *state);
2539
2540 extern void
2541 _mesa_glsl_initialize_builtin_functions();
2542
2543 extern ir_function_signature *
2544 _mesa_glsl_find_builtin_function(_mesa_glsl_parse_state *state,
2545 const char *name, exec_list *actual_parameters);
2546
2547 extern ir_function *
2548 _mesa_glsl_find_builtin_function_by_name(const char *name);
2549
2550 extern gl_shader *
2551 _mesa_glsl_get_builtin_function_shader(void);
2552
2553 extern ir_function_signature *
2554 _mesa_get_main_function_signature(gl_shader *sh);
2555
2556 extern void
2557 _mesa_glsl_release_functions(void);
2558
2559 extern void
2560 _mesa_glsl_release_builtin_functions(void);
2561
2562 extern void
2563 reparent_ir(exec_list *list, void *mem_ctx);
2564
2565 struct glsl_symbol_table;
2566
2567 extern void
2568 import_prototypes(const exec_list *source, exec_list *dest,
2569 struct glsl_symbol_table *symbols, void *mem_ctx);
2570
2571 extern bool
2572 ir_has_call(ir_instruction *ir);
2573
2574 extern void
2575 do_set_program_inouts(exec_list *instructions, struct gl_program *prog,
2576 gl_shader_stage shader_stage);
2577
2578 extern char *
2579 prototype_string(const glsl_type *return_type, const char *name,
2580 exec_list *parameters);
2581
2582 const char *
2583 mode_string(const ir_variable *var);
2584
2585 /**
2586 * Built-in / reserved GL variables names start with "gl_"
2587 */
2588 static inline bool
2589 is_gl_identifier(const char *s)
2590 {
2591 return s && s[0] == 'g' && s[1] == 'l' && s[2] == '_';
2592 }
2593
2594 extern "C" {
2595 #endif /* __cplusplus */
2596
2597 extern void _mesa_print_ir(FILE *f, struct exec_list *instructions,
2598 struct _mesa_glsl_parse_state *state);
2599
2600 extern void
2601 fprint_ir(FILE *f, const void *instruction);
2602
2603 #ifdef __cplusplus
2604 } /* extern "C" */
2605 #endif
2606
2607 unsigned
2608 vertices_per_prim(GLenum prim);
2609
2610 #endif /* IR_H */