glsl: Add parsing support for multi-stream output in geometry shaders.
[mesa.git] / src / glsl / ir.h
1 /* -*- c++ -*- */
2 /*
3 * Copyright © 2010 Intel Corporation
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
14 * Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
21 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
22 * DEALINGS IN THE SOFTWARE.
23 */
24
25 #pragma once
26 #ifndef IR_H
27 #define IR_H
28
29 #include <stdio.h>
30 #include <stdlib.h>
31
32 #include "ralloc.h"
33 #include "glsl_types.h"
34 #include "list.h"
35 #include "ir_visitor.h"
36 #include "ir_hierarchical_visitor.h"
37 #include "main/mtypes.h"
38
39 #ifdef __cplusplus
40
41 /**
42 * \defgroup IR Intermediate representation nodes
43 *
44 * @{
45 */
46
47 /**
48 * Class tags
49 *
50 * Each concrete class derived from \c ir_instruction has a value in this
51 * enumerant. The value for the type is stored in \c ir_instruction::ir_type
52 * by the constructor. While using type tags is not very C++, it is extremely
53 * convenient. For example, during debugging you can simply inspect
54 * \c ir_instruction::ir_type to find out the actual type of the object.
55 *
56 * In addition, it is possible to use a switch-statement based on \c
57 * \c ir_instruction::ir_type to select different behavior for different object
58 * types. For functions that have only slight differences for several object
59 * types, this allows writing very straightforward, readable code.
60 */
61 enum ir_node_type {
62 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 * Non-zero if this variable was created by lowering a named interface
664 * block which was not an array.
665 *
666 * Note that this variable and \c from_named_ifc_block_array will never
667 * both be non-zero.
668 */
669 unsigned from_named_ifc_block_nonarray:1;
670
671 /**
672 * Non-zero if this variable was created by lowering a named interface
673 * block which was an array.
674 *
675 * Note that this variable and \c from_named_ifc_block_nonarray will never
676 * both be non-zero.
677 */
678 unsigned from_named_ifc_block_array:1;
679
680 /**
681 * \brief Layout qualifier for gl_FragDepth.
682 *
683 * This is not equal to \c ir_depth_layout_none if and only if this
684 * variable is \c gl_FragDepth and a layout qualifier is specified.
685 */
686 ir_depth_layout depth_layout;
687
688 /**
689 * Storage location of the base of this variable
690 *
691 * The precise meaning of this field depends on the nature of the variable.
692 *
693 * - Vertex shader input: one of the values from \c gl_vert_attrib.
694 * - Vertex shader output: one of the values from \c gl_varying_slot.
695 * - Geometry shader input: one of the values from \c gl_varying_slot.
696 * - Geometry shader output: one of the values from \c gl_varying_slot.
697 * - Fragment shader input: one of the values from \c gl_varying_slot.
698 * - Fragment shader output: one of the values from \c gl_frag_result.
699 * - Uniforms: Per-stage uniform slot number for default uniform block.
700 * - Uniforms: Index within the uniform block definition for UBO members.
701 * - Other: This field is not currently used.
702 *
703 * If the variable is a uniform, shader input, or shader output, and the
704 * slot has not been assigned, the value will be -1.
705 */
706 int location;
707
708 /**
709 * Vertex stream output identifier.
710 */
711 unsigned stream;
712
713 /**
714 * output index for dual source blending.
715 */
716 int index;
717
718 /**
719 * Initial binding point for a sampler or UBO.
720 *
721 * For array types, this represents the binding point for the first element.
722 */
723 int binding;
724
725 /**
726 * Location an atomic counter is stored at.
727 */
728 struct {
729 unsigned buffer_index;
730 unsigned offset;
731 } atomic;
732
733 /**
734 * ARB_shader_image_load_store qualifiers.
735 */
736 struct {
737 bool read_only; /**< "readonly" qualifier. */
738 bool write_only; /**< "writeonly" qualifier. */
739 bool coherent;
740 bool _volatile;
741 bool restrict_flag;
742
743 /** Image internal format if specified explicitly, otherwise GL_NONE. */
744 GLenum format;
745 } image;
746
747 /**
748 * Highest element accessed with a constant expression array index
749 *
750 * Not used for non-array variables.
751 */
752 unsigned max_array_access;
753
754 } data;
755
756 /**
757 * Built-in state that backs this uniform
758 *
759 * Once set at variable creation, \c state_slots must remain invariant.
760 * This is because, ideally, this array would be shared by all clones of
761 * this variable in the IR tree. In other words, we'd really like for it
762 * to be a fly-weight.
763 *
764 * If the variable is not a uniform, \c num_state_slots will be zero and
765 * \c state_slots will be \c NULL.
766 */
767 /*@{*/
768 unsigned num_state_slots; /**< Number of state slots used */
769 ir_state_slot *state_slots; /**< State descriptors. */
770 /*@}*/
771
772 /**
773 * Emit a warning if this variable is accessed.
774 */
775 const char *warn_extension;
776
777 /**
778 * Value assigned in the initializer of a variable declared "const"
779 */
780 ir_constant *constant_value;
781
782 /**
783 * Constant expression assigned in the initializer of the variable
784 *
785 * \warning
786 * This field and \c ::constant_value are distinct. Even if the two fields
787 * refer to constants with the same value, they must point to separate
788 * objects.
789 */
790 ir_constant *constant_initializer;
791
792 private:
793 /**
794 * For variables that are in an interface block or are an instance of an
795 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
796 *
797 * \sa ir_variable::location
798 */
799 const glsl_type *interface_type;
800 };
801
802 /**
803 * A function that returns whether a built-in function is available in the
804 * current shading language (based on version, ES or desktop, and extensions).
805 */
806 typedef bool (*builtin_available_predicate)(const _mesa_glsl_parse_state *);
807
808 /*@{*/
809 /**
810 * The representation of a function instance; may be the full definition or
811 * simply a prototype.
812 */
813 class ir_function_signature : public ir_instruction {
814 /* An ir_function_signature will be part of the list of signatures in
815 * an ir_function.
816 */
817 public:
818 ir_function_signature(const glsl_type *return_type,
819 builtin_available_predicate builtin_avail = NULL);
820
821 virtual ir_function_signature *clone(void *mem_ctx,
822 struct hash_table *ht) const;
823 ir_function_signature *clone_prototype(void *mem_ctx,
824 struct hash_table *ht) const;
825
826 virtual void accept(ir_visitor *v)
827 {
828 v->visit(this);
829 }
830
831 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
832
833 /**
834 * Attempt to evaluate this function as a constant expression,
835 * given a list of the actual parameters and the variable context.
836 * Returns NULL for non-built-ins.
837 */
838 ir_constant *constant_expression_value(exec_list *actual_parameters, struct hash_table *variable_context);
839
840 /**
841 * Get the name of the function for which this is a signature
842 */
843 const char *function_name() const;
844
845 /**
846 * Get a handle to the function for which this is a signature
847 *
848 * There is no setter function, this function returns a \c const pointer,
849 * and \c ir_function_signature::_function is private for a reason. The
850 * only way to make a connection between a function and function signature
851 * is via \c ir_function::add_signature. This helps ensure that certain
852 * invariants (i.e., a function signature is in the list of signatures for
853 * its \c _function) are met.
854 *
855 * \sa ir_function::add_signature
856 */
857 inline const class ir_function *function() const
858 {
859 return this->_function;
860 }
861
862 /**
863 * Check whether the qualifiers match between this signature's parameters
864 * and the supplied parameter list. If not, returns the name of the first
865 * parameter with mismatched qualifiers (for use in error messages).
866 */
867 const char *qualifiers_match(exec_list *params);
868
869 /**
870 * Replace the current parameter list with the given one. This is useful
871 * if the current information came from a prototype, and either has invalid
872 * or missing parameter names.
873 */
874 void replace_parameters(exec_list *new_params);
875
876 /**
877 * Function return type.
878 *
879 * \note This discards the optional precision qualifier.
880 */
881 const struct glsl_type *return_type;
882
883 /**
884 * List of ir_variable of function parameters.
885 *
886 * This represents the storage. The paramaters passed in a particular
887 * call will be in ir_call::actual_paramaters.
888 */
889 struct exec_list parameters;
890
891 /** Whether or not this function has a body (which may be empty). */
892 unsigned is_defined:1;
893
894 /** Whether or not this function signature is a built-in. */
895 bool is_builtin() const;
896
897 /**
898 * Whether or not this function is an intrinsic to be implemented
899 * by the driver.
900 */
901 bool is_intrinsic;
902
903 /** Whether or not a built-in is available for this shader. */
904 bool is_builtin_available(const _mesa_glsl_parse_state *state) const;
905
906 /** Body of instructions in the function. */
907 struct exec_list body;
908
909 private:
910 /**
911 * A function pointer to a predicate that answers whether a built-in
912 * function is available in the current shader. NULL if not a built-in.
913 */
914 builtin_available_predicate builtin_avail;
915
916 /** Function of which this signature is one overload. */
917 class ir_function *_function;
918
919 /** Function signature of which this one is a prototype clone */
920 const ir_function_signature *origin;
921
922 friend class ir_function;
923
924 /**
925 * Helper function to run a list of instructions for constant
926 * expression evaluation.
927 *
928 * The hash table represents the values of the visible variables.
929 * There are no scoping issues because the table is indexed on
930 * ir_variable pointers, not variable names.
931 *
932 * Returns false if the expression is not constant, true otherwise,
933 * and the value in *result if result is non-NULL.
934 */
935 bool constant_expression_evaluate_expression_list(const struct exec_list &body,
936 struct hash_table *variable_context,
937 ir_constant **result);
938 };
939
940
941 /**
942 * Header for tracking multiple overloaded functions with the same name.
943 * Contains a list of ir_function_signatures representing each of the
944 * actual functions.
945 */
946 class ir_function : public ir_instruction {
947 public:
948 ir_function(const char *name);
949
950 virtual ir_function *clone(void *mem_ctx, struct hash_table *ht) const;
951
952 virtual void accept(ir_visitor *v)
953 {
954 v->visit(this);
955 }
956
957 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
958
959 void add_signature(ir_function_signature *sig)
960 {
961 sig->_function = this;
962 this->signatures.push_tail(sig);
963 }
964
965 /**
966 * Find a signature that matches a set of actual parameters, taking implicit
967 * conversions into account. Also flags whether the match was exact.
968 */
969 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
970 const exec_list *actual_param,
971 bool *match_is_exact);
972
973 /**
974 * Find a signature that matches a set of actual parameters, taking implicit
975 * conversions into account.
976 */
977 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
978 const exec_list *actual_param);
979
980 /**
981 * Find a signature that exactly matches a set of actual parameters without
982 * any implicit type conversions.
983 */
984 ir_function_signature *exact_matching_signature(_mesa_glsl_parse_state *state,
985 const exec_list *actual_ps);
986
987 /**
988 * Name of the function.
989 */
990 const char *name;
991
992 /** Whether or not this function has a signature that isn't a built-in. */
993 bool has_user_signature();
994
995 /**
996 * List of ir_function_signature for each overloaded function with this name.
997 */
998 struct exec_list signatures;
999 };
1000
1001 inline const char *ir_function_signature::function_name() const
1002 {
1003 return this->_function->name;
1004 }
1005 /*@}*/
1006
1007
1008 /**
1009 * IR instruction representing high-level if-statements
1010 */
1011 class ir_if : public ir_instruction {
1012 public:
1013 ir_if(ir_rvalue *condition)
1014 : ir_instruction(ir_type_if), condition(condition)
1015 {
1016 }
1017
1018 virtual ir_if *clone(void *mem_ctx, struct hash_table *ht) const;
1019
1020 virtual void accept(ir_visitor *v)
1021 {
1022 v->visit(this);
1023 }
1024
1025 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1026
1027 ir_rvalue *condition;
1028 /** List of ir_instruction for the body of the then branch */
1029 exec_list then_instructions;
1030 /** List of ir_instruction for the body of the else branch */
1031 exec_list else_instructions;
1032 };
1033
1034
1035 /**
1036 * IR instruction representing a high-level loop structure.
1037 */
1038 class ir_loop : public ir_instruction {
1039 public:
1040 ir_loop();
1041
1042 virtual ir_loop *clone(void *mem_ctx, struct hash_table *ht) const;
1043
1044 virtual void accept(ir_visitor *v)
1045 {
1046 v->visit(this);
1047 }
1048
1049 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1050
1051 /** List of ir_instruction that make up the body of the loop. */
1052 exec_list body_instructions;
1053 };
1054
1055
1056 class ir_assignment : public ir_instruction {
1057 public:
1058 ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs, ir_rvalue *condition = NULL);
1059
1060 /**
1061 * Construct an assignment with an explicit write mask
1062 *
1063 * \note
1064 * Since a write mask is supplied, the LHS must already be a bare
1065 * \c ir_dereference. The cannot be any swizzles in the LHS.
1066 */
1067 ir_assignment(ir_dereference *lhs, ir_rvalue *rhs, ir_rvalue *condition,
1068 unsigned write_mask);
1069
1070 virtual ir_assignment *clone(void *mem_ctx, struct hash_table *ht) const;
1071
1072 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1073
1074 virtual void accept(ir_visitor *v)
1075 {
1076 v->visit(this);
1077 }
1078
1079 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1080
1081 /**
1082 * Get a whole variable written by an assignment
1083 *
1084 * If the LHS of the assignment writes a whole variable, the variable is
1085 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1086 * assignment are:
1087 *
1088 * - Assigning to a scalar
1089 * - Assigning to all components of a vector
1090 * - Whole array (or matrix) assignment
1091 * - Whole structure assignment
1092 */
1093 ir_variable *whole_variable_written();
1094
1095 /**
1096 * Set the LHS of an assignment
1097 */
1098 void set_lhs(ir_rvalue *lhs);
1099
1100 /**
1101 * Left-hand side of the assignment.
1102 *
1103 * This should be treated as read only. If you need to set the LHS of an
1104 * assignment, use \c ir_assignment::set_lhs.
1105 */
1106 ir_dereference *lhs;
1107
1108 /**
1109 * Value being assigned
1110 */
1111 ir_rvalue *rhs;
1112
1113 /**
1114 * Optional condition for the assignment.
1115 */
1116 ir_rvalue *condition;
1117
1118
1119 /**
1120 * Component mask written
1121 *
1122 * For non-vector types in the LHS, this field will be zero. For vector
1123 * types, a bit will be set for each component that is written. Note that
1124 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1125 *
1126 * A partially-set write mask means that each enabled channel gets
1127 * the value from a consecutive channel of the rhs. For example,
1128 * to write just .xyw of gl_FrontColor with color:
1129 *
1130 * (assign (constant bool (1)) (xyw)
1131 * (var_ref gl_FragColor)
1132 * (swiz xyw (var_ref color)))
1133 */
1134 unsigned write_mask:4;
1135 };
1136
1137 /* Update ir_expression::get_num_operands() and operator_strs when
1138 * updating this list.
1139 */
1140 enum ir_expression_operation {
1141 ir_unop_bit_not,
1142 ir_unop_logic_not,
1143 ir_unop_neg,
1144 ir_unop_abs,
1145 ir_unop_sign,
1146 ir_unop_rcp,
1147 ir_unop_rsq,
1148 ir_unop_sqrt,
1149 ir_unop_exp, /**< Log base e on gentype */
1150 ir_unop_log, /**< Natural log on gentype */
1151 ir_unop_exp2,
1152 ir_unop_log2,
1153 ir_unop_f2i, /**< Float-to-integer conversion. */
1154 ir_unop_f2u, /**< Float-to-unsigned conversion. */
1155 ir_unop_i2f, /**< Integer-to-float conversion. */
1156 ir_unop_f2b, /**< Float-to-boolean conversion */
1157 ir_unop_b2f, /**< Boolean-to-float conversion */
1158 ir_unop_i2b, /**< int-to-boolean conversion */
1159 ir_unop_b2i, /**< Boolean-to-int conversion */
1160 ir_unop_u2f, /**< Unsigned-to-float conversion. */
1161 ir_unop_i2u, /**< Integer-to-unsigned conversion. */
1162 ir_unop_u2i, /**< Unsigned-to-integer conversion. */
1163 ir_unop_bitcast_i2f, /**< Bit-identical int-to-float "conversion" */
1164 ir_unop_bitcast_f2i, /**< Bit-identical float-to-int "conversion" */
1165 ir_unop_bitcast_u2f, /**< Bit-identical uint-to-float "conversion" */
1166 ir_unop_bitcast_f2u, /**< Bit-identical float-to-uint "conversion" */
1167 ir_unop_any,
1168
1169 /**
1170 * \name Unary floating-point rounding operations.
1171 */
1172 /*@{*/
1173 ir_unop_trunc,
1174 ir_unop_ceil,
1175 ir_unop_floor,
1176 ir_unop_fract,
1177 ir_unop_round_even,
1178 /*@}*/
1179
1180 /**
1181 * \name Trigonometric operations.
1182 */
1183 /*@{*/
1184 ir_unop_sin,
1185 ir_unop_cos,
1186 ir_unop_sin_reduced, /**< Reduced range sin. [-pi, pi] */
1187 ir_unop_cos_reduced, /**< Reduced range cos. [-pi, pi] */
1188 /*@}*/
1189
1190 /**
1191 * \name Partial derivatives.
1192 */
1193 /*@{*/
1194 ir_unop_dFdx,
1195 ir_unop_dFdy,
1196 /*@}*/
1197
1198 /**
1199 * \name Floating point pack and unpack operations.
1200 */
1201 /*@{*/
1202 ir_unop_pack_snorm_2x16,
1203 ir_unop_pack_snorm_4x8,
1204 ir_unop_pack_unorm_2x16,
1205 ir_unop_pack_unorm_4x8,
1206 ir_unop_pack_half_2x16,
1207 ir_unop_unpack_snorm_2x16,
1208 ir_unop_unpack_snorm_4x8,
1209 ir_unop_unpack_unorm_2x16,
1210 ir_unop_unpack_unorm_4x8,
1211 ir_unop_unpack_half_2x16,
1212 /*@}*/
1213
1214 /**
1215 * \name Lowered floating point unpacking operations.
1216 *
1217 * \see lower_packing_builtins_visitor::split_unpack_half_2x16
1218 */
1219 /*@{*/
1220 ir_unop_unpack_half_2x16_split_x,
1221 ir_unop_unpack_half_2x16_split_y,
1222 /*@}*/
1223
1224 /**
1225 * \name Bit operations, part of ARB_gpu_shader5.
1226 */
1227 /*@{*/
1228 ir_unop_bitfield_reverse,
1229 ir_unop_bit_count,
1230 ir_unop_find_msb,
1231 ir_unop_find_lsb,
1232 /*@}*/
1233
1234 ir_unop_noise,
1235
1236 /**
1237 * A sentinel marking the last of the unary operations.
1238 */
1239 ir_last_unop = ir_unop_noise,
1240
1241 ir_binop_add,
1242 ir_binop_sub,
1243 ir_binop_mul, /**< Floating-point or low 32-bit integer multiply. */
1244 ir_binop_imul_high, /**< Calculates the high 32-bits of a 64-bit multiply. */
1245 ir_binop_div,
1246
1247 /**
1248 * Returns the carry resulting from the addition of the two arguments.
1249 */
1250 /*@{*/
1251 ir_binop_carry,
1252 /*@}*/
1253
1254 /**
1255 * Returns the borrow resulting from the subtraction of the second argument
1256 * from the first argument.
1257 */
1258 /*@{*/
1259 ir_binop_borrow,
1260 /*@}*/
1261
1262 /**
1263 * Takes one of two combinations of arguments:
1264 *
1265 * - mod(vecN, vecN)
1266 * - mod(vecN, float)
1267 *
1268 * Does not take integer types.
1269 */
1270 ir_binop_mod,
1271
1272 /**
1273 * \name Binary comparison operators which return a boolean vector.
1274 * The type of both operands must be equal.
1275 */
1276 /*@{*/
1277 ir_binop_less,
1278 ir_binop_greater,
1279 ir_binop_lequal,
1280 ir_binop_gequal,
1281 ir_binop_equal,
1282 ir_binop_nequal,
1283 /**
1284 * Returns single boolean for whether all components of operands[0]
1285 * equal the components of operands[1].
1286 */
1287 ir_binop_all_equal,
1288 /**
1289 * Returns single boolean for whether any component of operands[0]
1290 * is not equal to the corresponding component of operands[1].
1291 */
1292 ir_binop_any_nequal,
1293 /*@}*/
1294
1295 /**
1296 * \name Bit-wise binary operations.
1297 */
1298 /*@{*/
1299 ir_binop_lshift,
1300 ir_binop_rshift,
1301 ir_binop_bit_and,
1302 ir_binop_bit_xor,
1303 ir_binop_bit_or,
1304 /*@}*/
1305
1306 ir_binop_logic_and,
1307 ir_binop_logic_xor,
1308 ir_binop_logic_or,
1309
1310 ir_binop_dot,
1311 ir_binop_min,
1312 ir_binop_max,
1313
1314 ir_binop_pow,
1315
1316 /**
1317 * \name Lowered floating point packing operations.
1318 *
1319 * \see lower_packing_builtins_visitor::split_pack_half_2x16
1320 */
1321 /*@{*/
1322 ir_binop_pack_half_2x16_split,
1323 /*@}*/
1324
1325 /**
1326 * \name First half of a lowered bitfieldInsert() operation.
1327 *
1328 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1329 */
1330 /*@{*/
1331 ir_binop_bfm,
1332 /*@}*/
1333
1334 /**
1335 * Load a value the size of a given GLSL type from a uniform block.
1336 *
1337 * operand0 is the ir_constant uniform block index in the linked shader.
1338 * operand1 is a byte offset within the uniform block.
1339 */
1340 ir_binop_ubo_load,
1341
1342 /**
1343 * \name Multiplies a number by two to a power, part of ARB_gpu_shader5.
1344 */
1345 /*@{*/
1346 ir_binop_ldexp,
1347 /*@}*/
1348
1349 /**
1350 * Extract a scalar from a vector
1351 *
1352 * operand0 is the vector
1353 * operand1 is the index of the field to read from operand0
1354 */
1355 ir_binop_vector_extract,
1356
1357 /**
1358 * A sentinel marking the last of the binary operations.
1359 */
1360 ir_last_binop = ir_binop_vector_extract,
1361
1362 /**
1363 * \name Fused floating-point multiply-add, part of ARB_gpu_shader5.
1364 */
1365 /*@{*/
1366 ir_triop_fma,
1367 /*@}*/
1368
1369 ir_triop_lrp,
1370
1371 /**
1372 * \name Conditional Select
1373 *
1374 * A vector conditional select instruction (like ?:, but operating per-
1375 * component on vectors).
1376 *
1377 * \see lower_instructions_visitor::ldexp_to_arith
1378 */
1379 /*@{*/
1380 ir_triop_csel,
1381 /*@}*/
1382
1383 /**
1384 * \name Second half of a lowered bitfieldInsert() operation.
1385 *
1386 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1387 */
1388 /*@{*/
1389 ir_triop_bfi,
1390 /*@}*/
1391
1392 ir_triop_bitfield_extract,
1393
1394 /**
1395 * Generate a value with one field of a vector changed
1396 *
1397 * operand0 is the vector
1398 * operand1 is the value to write into the vector result
1399 * operand2 is the index in operand0 to be modified
1400 */
1401 ir_triop_vector_insert,
1402
1403 /**
1404 * A sentinel marking the last of the ternary operations.
1405 */
1406 ir_last_triop = ir_triop_vector_insert,
1407
1408 ir_quadop_bitfield_insert,
1409
1410 ir_quadop_vector,
1411
1412 /**
1413 * A sentinel marking the last of the ternary operations.
1414 */
1415 ir_last_quadop = ir_quadop_vector,
1416
1417 /**
1418 * A sentinel marking the last of all operations.
1419 */
1420 ir_last_opcode = ir_quadop_vector
1421 };
1422
1423 class ir_expression : public ir_rvalue {
1424 public:
1425 ir_expression(int op, const struct glsl_type *type,
1426 ir_rvalue *op0, ir_rvalue *op1 = NULL,
1427 ir_rvalue *op2 = NULL, ir_rvalue *op3 = NULL);
1428
1429 /**
1430 * Constructor for unary operation expressions
1431 */
1432 ir_expression(int op, ir_rvalue *);
1433
1434 /**
1435 * Constructor for binary operation expressions
1436 */
1437 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1);
1438
1439 /**
1440 * Constructor for ternary operation expressions
1441 */
1442 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1, ir_rvalue *op2);
1443
1444 virtual bool equals(ir_instruction *ir, enum ir_node_type ignore = ir_type_unset);
1445
1446 virtual ir_expression *clone(void *mem_ctx, struct hash_table *ht) const;
1447
1448 /**
1449 * Attempt to constant-fold the expression
1450 *
1451 * The "variable_context" hash table links ir_variable * to ir_constant *
1452 * that represent the variables' values. \c NULL represents an empty
1453 * context.
1454 *
1455 * If the expression cannot be constant folded, this method will return
1456 * \c NULL.
1457 */
1458 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1459
1460 /**
1461 * Determine the number of operands used by an expression
1462 */
1463 static unsigned int get_num_operands(ir_expression_operation);
1464
1465 /**
1466 * Determine the number of operands used by an expression
1467 */
1468 unsigned int get_num_operands() const
1469 {
1470 return (this->operation == ir_quadop_vector)
1471 ? this->type->vector_elements : get_num_operands(operation);
1472 }
1473
1474 /**
1475 * Return whether the expression operates on vectors horizontally.
1476 */
1477 bool is_horizontal() const
1478 {
1479 return operation == ir_binop_all_equal ||
1480 operation == ir_binop_any_nequal ||
1481 operation == ir_unop_any ||
1482 operation == ir_binop_dot ||
1483 operation == ir_quadop_vector;
1484 }
1485
1486 /**
1487 * Return a string representing this expression's operator.
1488 */
1489 const char *operator_string();
1490
1491 /**
1492 * Return a string representing this expression's operator.
1493 */
1494 static const char *operator_string(ir_expression_operation);
1495
1496
1497 /**
1498 * Do a reverse-lookup to translate the given string into an operator.
1499 */
1500 static ir_expression_operation get_operator(const char *);
1501
1502 virtual void accept(ir_visitor *v)
1503 {
1504 v->visit(this);
1505 }
1506
1507 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1508
1509 ir_expression_operation operation;
1510 ir_rvalue *operands[4];
1511 };
1512
1513
1514 /**
1515 * HIR instruction representing a high-level function call, containing a list
1516 * of parameters and returning a value in the supplied temporary.
1517 */
1518 class ir_call : public ir_instruction {
1519 public:
1520 ir_call(ir_function_signature *callee,
1521 ir_dereference_variable *return_deref,
1522 exec_list *actual_parameters)
1523 : ir_instruction(ir_type_call), return_deref(return_deref), callee(callee)
1524 {
1525 assert(callee->return_type != NULL);
1526 actual_parameters->move_nodes_to(& this->actual_parameters);
1527 this->use_builtin = callee->is_builtin();
1528 }
1529
1530 virtual ir_call *clone(void *mem_ctx, struct hash_table *ht) const;
1531
1532 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1533
1534 virtual void accept(ir_visitor *v)
1535 {
1536 v->visit(this);
1537 }
1538
1539 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1540
1541 /**
1542 * Get the name of the function being called.
1543 */
1544 const char *callee_name() const
1545 {
1546 return callee->function_name();
1547 }
1548
1549 /**
1550 * Generates an inline version of the function before @ir,
1551 * storing the return value in return_deref.
1552 */
1553 void generate_inline(ir_instruction *ir);
1554
1555 /**
1556 * Storage for the function's return value.
1557 * This must be NULL if the return type is void.
1558 */
1559 ir_dereference_variable *return_deref;
1560
1561 /**
1562 * The specific function signature being called.
1563 */
1564 ir_function_signature *callee;
1565
1566 /* List of ir_rvalue of paramaters passed in this call. */
1567 exec_list actual_parameters;
1568
1569 /** Should this call only bind to a built-in function? */
1570 bool use_builtin;
1571 };
1572
1573
1574 /**
1575 * \name Jump-like IR instructions.
1576 *
1577 * These include \c break, \c continue, \c return, and \c discard.
1578 */
1579 /*@{*/
1580 class ir_jump : public ir_instruction {
1581 protected:
1582 ir_jump(enum ir_node_type t)
1583 : ir_instruction(t)
1584 {
1585 }
1586 };
1587
1588 class ir_return : public ir_jump {
1589 public:
1590 ir_return()
1591 : ir_jump(ir_type_return), value(NULL)
1592 {
1593 }
1594
1595 ir_return(ir_rvalue *value)
1596 : ir_jump(ir_type_return), value(value)
1597 {
1598 }
1599
1600 virtual ir_return *clone(void *mem_ctx, struct hash_table *) const;
1601
1602 ir_rvalue *get_value() const
1603 {
1604 return value;
1605 }
1606
1607 virtual void accept(ir_visitor *v)
1608 {
1609 v->visit(this);
1610 }
1611
1612 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1613
1614 ir_rvalue *value;
1615 };
1616
1617
1618 /**
1619 * Jump instructions used inside loops
1620 *
1621 * These include \c break and \c continue. The \c break within a loop is
1622 * different from the \c break within a switch-statement.
1623 *
1624 * \sa ir_switch_jump
1625 */
1626 class ir_loop_jump : public ir_jump {
1627 public:
1628 enum jump_mode {
1629 jump_break,
1630 jump_continue
1631 };
1632
1633 ir_loop_jump(jump_mode mode)
1634 : ir_jump(ir_type_loop_jump)
1635 {
1636 this->mode = mode;
1637 }
1638
1639 virtual ir_loop_jump *clone(void *mem_ctx, struct hash_table *) const;
1640
1641 virtual void accept(ir_visitor *v)
1642 {
1643 v->visit(this);
1644 }
1645
1646 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1647
1648 bool is_break() const
1649 {
1650 return mode == jump_break;
1651 }
1652
1653 bool is_continue() const
1654 {
1655 return mode == jump_continue;
1656 }
1657
1658 /** Mode selector for the jump instruction. */
1659 enum jump_mode mode;
1660 };
1661
1662 /**
1663 * IR instruction representing discard statements.
1664 */
1665 class ir_discard : public ir_jump {
1666 public:
1667 ir_discard()
1668 : ir_jump(ir_type_discard)
1669 {
1670 this->condition = NULL;
1671 }
1672
1673 ir_discard(ir_rvalue *cond)
1674 : ir_jump(ir_type_discard)
1675 {
1676 this->condition = cond;
1677 }
1678
1679 virtual ir_discard *clone(void *mem_ctx, struct hash_table *ht) const;
1680
1681 virtual void accept(ir_visitor *v)
1682 {
1683 v->visit(this);
1684 }
1685
1686 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1687
1688 ir_rvalue *condition;
1689 };
1690 /*@}*/
1691
1692
1693 /**
1694 * Texture sampling opcodes used in ir_texture
1695 */
1696 enum ir_texture_opcode {
1697 ir_tex, /**< Regular texture look-up */
1698 ir_txb, /**< Texture look-up with LOD bias */
1699 ir_txl, /**< Texture look-up with explicit LOD */
1700 ir_txd, /**< Texture look-up with partial derivatvies */
1701 ir_txf, /**< Texel fetch with explicit LOD */
1702 ir_txf_ms, /**< Multisample texture fetch */
1703 ir_txs, /**< Texture size */
1704 ir_lod, /**< Texture lod query */
1705 ir_tg4, /**< Texture gather */
1706 ir_query_levels /**< Texture levels query */
1707 };
1708
1709
1710 /**
1711 * IR instruction to sample a texture
1712 *
1713 * The specific form of the IR instruction depends on the \c mode value
1714 * selected from \c ir_texture_opcodes. In the printed IR, these will
1715 * appear as:
1716 *
1717 * Texel offset (0 or an expression)
1718 * | Projection divisor
1719 * | | Shadow comparitor
1720 * | | |
1721 * v v v
1722 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1723 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1724 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1725 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1726 * (txf <type> <sampler> <coordinate> 0 <lod>)
1727 * (txf_ms
1728 * <type> <sampler> <coordinate> <sample_index>)
1729 * (txs <type> <sampler> <lod>)
1730 * (lod <type> <sampler> <coordinate>)
1731 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
1732 * (query_levels <type> <sampler>)
1733 */
1734 class ir_texture : public ir_rvalue {
1735 public:
1736 ir_texture(enum ir_texture_opcode op)
1737 : ir_rvalue(ir_type_texture),
1738 op(op), sampler(NULL), coordinate(NULL), projector(NULL),
1739 shadow_comparitor(NULL), offset(NULL)
1740 {
1741 memset(&lod_info, 0, sizeof(lod_info));
1742 }
1743
1744 virtual ir_texture *clone(void *mem_ctx, struct hash_table *) const;
1745
1746 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1747
1748 virtual void accept(ir_visitor *v)
1749 {
1750 v->visit(this);
1751 }
1752
1753 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1754
1755 virtual bool equals(ir_instruction *ir, enum ir_node_type ignore = ir_type_unset);
1756
1757 /**
1758 * Return a string representing the ir_texture_opcode.
1759 */
1760 const char *opcode_string();
1761
1762 /** Set the sampler and type. */
1763 void set_sampler(ir_dereference *sampler, const glsl_type *type);
1764
1765 /**
1766 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1767 */
1768 static ir_texture_opcode get_opcode(const char *);
1769
1770 enum ir_texture_opcode op;
1771
1772 /** Sampler to use for the texture access. */
1773 ir_dereference *sampler;
1774
1775 /** Texture coordinate to sample */
1776 ir_rvalue *coordinate;
1777
1778 /**
1779 * Value used for projective divide.
1780 *
1781 * If there is no projective divide (the common case), this will be
1782 * \c NULL. Optimization passes should check for this to point to a constant
1783 * of 1.0 and replace that with \c NULL.
1784 */
1785 ir_rvalue *projector;
1786
1787 /**
1788 * Coordinate used for comparison on shadow look-ups.
1789 *
1790 * If there is no shadow comparison, this will be \c NULL. For the
1791 * \c ir_txf opcode, this *must* be \c NULL.
1792 */
1793 ir_rvalue *shadow_comparitor;
1794
1795 /** Texel offset. */
1796 ir_rvalue *offset;
1797
1798 union {
1799 ir_rvalue *lod; /**< Floating point LOD */
1800 ir_rvalue *bias; /**< Floating point LOD bias */
1801 ir_rvalue *sample_index; /**< MSAA sample index */
1802 ir_rvalue *component; /**< Gather component selector */
1803 struct {
1804 ir_rvalue *dPdx; /**< Partial derivative of coordinate wrt X */
1805 ir_rvalue *dPdy; /**< Partial derivative of coordinate wrt Y */
1806 } grad;
1807 } lod_info;
1808 };
1809
1810
1811 struct ir_swizzle_mask {
1812 unsigned x:2;
1813 unsigned y:2;
1814 unsigned z:2;
1815 unsigned w:2;
1816
1817 /**
1818 * Number of components in the swizzle.
1819 */
1820 unsigned num_components:3;
1821
1822 /**
1823 * Does the swizzle contain duplicate components?
1824 *
1825 * L-value swizzles cannot contain duplicate components.
1826 */
1827 unsigned has_duplicates:1;
1828 };
1829
1830
1831 class ir_swizzle : public ir_rvalue {
1832 public:
1833 ir_swizzle(ir_rvalue *, unsigned x, unsigned y, unsigned z, unsigned w,
1834 unsigned count);
1835
1836 ir_swizzle(ir_rvalue *val, const unsigned *components, unsigned count);
1837
1838 ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask);
1839
1840 virtual ir_swizzle *clone(void *mem_ctx, struct hash_table *) const;
1841
1842 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1843
1844 /**
1845 * Construct an ir_swizzle from the textual representation. Can fail.
1846 */
1847 static ir_swizzle *create(ir_rvalue *, const char *, unsigned vector_length);
1848
1849 virtual void accept(ir_visitor *v)
1850 {
1851 v->visit(this);
1852 }
1853
1854 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1855
1856 virtual bool equals(ir_instruction *ir, enum ir_node_type ignore = ir_type_unset);
1857
1858 bool is_lvalue() const
1859 {
1860 return val->is_lvalue() && !mask.has_duplicates;
1861 }
1862
1863 /**
1864 * Get the variable that is ultimately referenced by an r-value
1865 */
1866 virtual ir_variable *variable_referenced() const;
1867
1868 ir_rvalue *val;
1869 ir_swizzle_mask mask;
1870
1871 private:
1872 /**
1873 * Initialize the mask component of a swizzle
1874 *
1875 * This is used by the \c ir_swizzle constructors.
1876 */
1877 void init_mask(const unsigned *components, unsigned count);
1878 };
1879
1880
1881 class ir_dereference : public ir_rvalue {
1882 public:
1883 virtual ir_dereference *clone(void *mem_ctx, struct hash_table *) const = 0;
1884
1885 bool is_lvalue() const;
1886
1887 /**
1888 * Get the variable that is ultimately referenced by an r-value
1889 */
1890 virtual ir_variable *variable_referenced() const = 0;
1891
1892 protected:
1893 ir_dereference(enum ir_node_type t)
1894 : ir_rvalue(t)
1895 {
1896 }
1897 };
1898
1899
1900 class ir_dereference_variable : public ir_dereference {
1901 public:
1902 ir_dereference_variable(ir_variable *var);
1903
1904 virtual ir_dereference_variable *clone(void *mem_ctx,
1905 struct hash_table *) const;
1906
1907 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1908
1909 virtual bool equals(ir_instruction *ir, enum ir_node_type ignore = ir_type_unset);
1910
1911 /**
1912 * Get the variable that is ultimately referenced by an r-value
1913 */
1914 virtual ir_variable *variable_referenced() const
1915 {
1916 return this->var;
1917 }
1918
1919 virtual ir_variable *whole_variable_referenced()
1920 {
1921 /* ir_dereference_variable objects always dereference the entire
1922 * variable. However, if this dereference is dereferenced by anything
1923 * else, the complete deferefernce chain is not a whole-variable
1924 * dereference. This method should only be called on the top most
1925 * ir_rvalue in a dereference chain.
1926 */
1927 return this->var;
1928 }
1929
1930 virtual void accept(ir_visitor *v)
1931 {
1932 v->visit(this);
1933 }
1934
1935 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1936
1937 /**
1938 * Object being dereferenced.
1939 */
1940 ir_variable *var;
1941 };
1942
1943
1944 class ir_dereference_array : public ir_dereference {
1945 public:
1946 ir_dereference_array(ir_rvalue *value, ir_rvalue *array_index);
1947
1948 ir_dereference_array(ir_variable *var, ir_rvalue *array_index);
1949
1950 virtual ir_dereference_array *clone(void *mem_ctx,
1951 struct hash_table *) const;
1952
1953 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1954
1955 virtual bool equals(ir_instruction *ir, enum ir_node_type ignore = ir_type_unset);
1956
1957 /**
1958 * Get the variable that is ultimately referenced by an r-value
1959 */
1960 virtual ir_variable *variable_referenced() const
1961 {
1962 return this->array->variable_referenced();
1963 }
1964
1965 virtual void accept(ir_visitor *v)
1966 {
1967 v->visit(this);
1968 }
1969
1970 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1971
1972 ir_rvalue *array;
1973 ir_rvalue *array_index;
1974
1975 private:
1976 void set_array(ir_rvalue *value);
1977 };
1978
1979
1980 class ir_dereference_record : public ir_dereference {
1981 public:
1982 ir_dereference_record(ir_rvalue *value, const char *field);
1983
1984 ir_dereference_record(ir_variable *var, const char *field);
1985
1986 virtual ir_dereference_record *clone(void *mem_ctx,
1987 struct hash_table *) const;
1988
1989 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1990
1991 /**
1992 * Get the variable that is ultimately referenced by an r-value
1993 */
1994 virtual ir_variable *variable_referenced() const
1995 {
1996 return this->record->variable_referenced();
1997 }
1998
1999 virtual void accept(ir_visitor *v)
2000 {
2001 v->visit(this);
2002 }
2003
2004 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2005
2006 ir_rvalue *record;
2007 const char *field;
2008 };
2009
2010
2011 /**
2012 * Data stored in an ir_constant
2013 */
2014 union ir_constant_data {
2015 unsigned u[16];
2016 int i[16];
2017 float f[16];
2018 bool b[16];
2019 };
2020
2021
2022 class ir_constant : public ir_rvalue {
2023 public:
2024 ir_constant(const struct glsl_type *type, const ir_constant_data *data);
2025 ir_constant(bool b, unsigned vector_elements=1);
2026 ir_constant(unsigned int u, unsigned vector_elements=1);
2027 ir_constant(int i, unsigned vector_elements=1);
2028 ir_constant(float f, unsigned vector_elements=1);
2029
2030 /**
2031 * Construct an ir_constant from a list of ir_constant values
2032 */
2033 ir_constant(const struct glsl_type *type, exec_list *values);
2034
2035 /**
2036 * Construct an ir_constant from a scalar component of another ir_constant
2037 *
2038 * The new \c ir_constant inherits the type of the component from the
2039 * source constant.
2040 *
2041 * \note
2042 * In the case of a matrix constant, the new constant is a scalar, \b not
2043 * a vector.
2044 */
2045 ir_constant(const ir_constant *c, unsigned i);
2046
2047 /**
2048 * Return a new ir_constant of the specified type containing all zeros.
2049 */
2050 static ir_constant *zero(void *mem_ctx, const glsl_type *type);
2051
2052 virtual ir_constant *clone(void *mem_ctx, struct hash_table *) const;
2053
2054 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
2055
2056 virtual void accept(ir_visitor *v)
2057 {
2058 v->visit(this);
2059 }
2060
2061 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2062
2063 virtual bool equals(ir_instruction *ir, enum ir_node_type ignore = ir_type_unset);
2064
2065 /**
2066 * Get a particular component of a constant as a specific type
2067 *
2068 * This is useful, for example, to get a value from an integer constant
2069 * as a float or bool. This appears frequently when constructors are
2070 * called with all constant parameters.
2071 */
2072 /*@{*/
2073 bool get_bool_component(unsigned i) const;
2074 float get_float_component(unsigned i) const;
2075 int get_int_component(unsigned i) const;
2076 unsigned get_uint_component(unsigned i) const;
2077 /*@}*/
2078
2079 ir_constant *get_array_element(unsigned i) const;
2080
2081 ir_constant *get_record_field(const char *name);
2082
2083 /**
2084 * Copy the values on another constant at a given offset.
2085 *
2086 * The offset is ignored for array or struct copies, it's only for
2087 * scalars or vectors into vectors or matrices.
2088 *
2089 * With identical types on both sides and zero offset it's clone()
2090 * without creating a new object.
2091 */
2092
2093 void copy_offset(ir_constant *src, int offset);
2094
2095 /**
2096 * Copy the values on another constant at a given offset and
2097 * following an assign-like mask.
2098 *
2099 * The mask is ignored for scalars.
2100 *
2101 * Note that this function only handles what assign can handle,
2102 * i.e. at most a vector as source and a column of a matrix as
2103 * destination.
2104 */
2105
2106 void copy_masked_offset(ir_constant *src, int offset, unsigned int mask);
2107
2108 /**
2109 * Determine whether a constant has the same value as another constant
2110 *
2111 * \sa ir_constant::is_zero, ir_constant::is_one,
2112 * ir_constant::is_negative_one, ir_constant::is_basis
2113 */
2114 bool has_value(const ir_constant *) const;
2115
2116 /**
2117 * Return true if this ir_constant represents the given value.
2118 *
2119 * For vectors, this checks that each component is the given value.
2120 */
2121 virtual bool is_value(float f, int i) const;
2122 virtual bool is_zero() const;
2123 virtual bool is_one() const;
2124 virtual bool is_negative_one() const;
2125 virtual bool is_basis() const;
2126
2127 /**
2128 * Return true for constants that could be stored as 16-bit unsigned values.
2129 *
2130 * Note that this will return true even for signed integer ir_constants, as
2131 * long as the value is non-negative and fits in 16-bits.
2132 */
2133 virtual bool is_uint16_constant() const;
2134
2135 /**
2136 * Value of the constant.
2137 *
2138 * The field used to back the values supplied by the constant is determined
2139 * by the type associated with the \c ir_instruction. Constants may be
2140 * scalars, vectors, or matrices.
2141 */
2142 union ir_constant_data value;
2143
2144 /* Array elements */
2145 ir_constant **array_elements;
2146
2147 /* Structure fields */
2148 exec_list components;
2149
2150 private:
2151 /**
2152 * Parameterless constructor only used by the clone method
2153 */
2154 ir_constant(void);
2155 };
2156
2157 /**
2158 * IR instruction to emit a vertex in a geometry shader.
2159 */
2160 class ir_emit_vertex : public ir_instruction {
2161 public:
2162 ir_emit_vertex()
2163 : ir_instruction(ir_type_emit_vertex)
2164 {
2165 }
2166
2167 virtual void accept(ir_visitor *v)
2168 {
2169 v->visit(this);
2170 }
2171
2172 virtual ir_emit_vertex *clone(void *mem_ctx, struct hash_table *) const
2173 {
2174 return new(mem_ctx) ir_emit_vertex();
2175 }
2176
2177 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2178 };
2179
2180 /**
2181 * IR instruction to complete the current primitive and start a new one in a
2182 * geometry shader.
2183 */
2184 class ir_end_primitive : public ir_instruction {
2185 public:
2186 ir_end_primitive()
2187 : ir_instruction(ir_type_end_primitive)
2188 {
2189 }
2190
2191 virtual void accept(ir_visitor *v)
2192 {
2193 v->visit(this);
2194 }
2195
2196 virtual ir_end_primitive *clone(void *mem_ctx, struct hash_table *) const
2197 {
2198 return new(mem_ctx) ir_end_primitive();
2199 }
2200
2201 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2202 };
2203
2204 /*@}*/
2205
2206 /**
2207 * Apply a visitor to each IR node in a list
2208 */
2209 void
2210 visit_exec_list(exec_list *list, ir_visitor *visitor);
2211
2212 /**
2213 * Validate invariants on each IR node in a list
2214 */
2215 void validate_ir_tree(exec_list *instructions);
2216
2217 struct _mesa_glsl_parse_state;
2218 struct gl_shader_program;
2219
2220 /**
2221 * Detect whether an unlinked shader contains static recursion
2222 *
2223 * If the list of instructions is determined to contain static recursion,
2224 * \c _mesa_glsl_error will be called to emit error messages for each function
2225 * that is in the recursion cycle.
2226 */
2227 void
2228 detect_recursion_unlinked(struct _mesa_glsl_parse_state *state,
2229 exec_list *instructions);
2230
2231 /**
2232 * Detect whether a linked shader contains static recursion
2233 *
2234 * If the list of instructions is determined to contain static recursion,
2235 * \c link_error_printf will be called to emit error messages for each function
2236 * that is in the recursion cycle. In addition,
2237 * \c gl_shader_program::LinkStatus will be set to false.
2238 */
2239 void
2240 detect_recursion_linked(struct gl_shader_program *prog,
2241 exec_list *instructions);
2242
2243 /**
2244 * Make a clone of each IR instruction in a list
2245 *
2246 * \param in List of IR instructions that are to be cloned
2247 * \param out List to hold the cloned instructions
2248 */
2249 void
2250 clone_ir_list(void *mem_ctx, exec_list *out, const exec_list *in);
2251
2252 extern void
2253 _mesa_glsl_initialize_variables(exec_list *instructions,
2254 struct _mesa_glsl_parse_state *state);
2255
2256 extern void
2257 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state *state);
2258
2259 extern void
2260 _mesa_glsl_initialize_builtin_functions();
2261
2262 extern ir_function_signature *
2263 _mesa_glsl_find_builtin_function(_mesa_glsl_parse_state *state,
2264 const char *name, exec_list *actual_parameters);
2265
2266 extern gl_shader *
2267 _mesa_glsl_get_builtin_function_shader(void);
2268
2269 extern void
2270 _mesa_glsl_release_functions(void);
2271
2272 extern void
2273 _mesa_glsl_release_builtin_functions(void);
2274
2275 extern void
2276 reparent_ir(exec_list *list, void *mem_ctx);
2277
2278 struct glsl_symbol_table;
2279
2280 extern void
2281 import_prototypes(const exec_list *source, exec_list *dest,
2282 struct glsl_symbol_table *symbols, void *mem_ctx);
2283
2284 extern bool
2285 ir_has_call(ir_instruction *ir);
2286
2287 extern void
2288 do_set_program_inouts(exec_list *instructions, struct gl_program *prog,
2289 gl_shader_stage shader_stage);
2290
2291 extern char *
2292 prototype_string(const glsl_type *return_type, const char *name,
2293 exec_list *parameters);
2294
2295 const char *
2296 mode_string(const ir_variable *var);
2297
2298 /**
2299 * Built-in / reserved GL variables names start with "gl_"
2300 */
2301 static inline bool
2302 is_gl_identifier(const char *s)
2303 {
2304 return s && s[0] == 'g' && s[1] == 'l' && s[2] == '_';
2305 }
2306
2307 extern "C" {
2308 #endif /* __cplusplus */
2309
2310 extern void _mesa_print_ir(FILE *f, struct exec_list *instructions,
2311 struct _mesa_glsl_parse_state *state);
2312
2313 extern void
2314 fprint_ir(FILE *f, const void *instruction);
2315
2316 #ifdef __cplusplus
2317 } /* extern "C" */
2318 #endif
2319
2320 unsigned
2321 vertices_per_prim(GLenum prim);
2322
2323 #endif /* IR_H */