Revert "glsl: move xfb BufferStride into gl_transform_feedback_info"
[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 ralloc_array(this, int, type->length);
488 for (unsigned i = 0; i < type->length; i++) {
489 this->u.max_ifc_array_access[i] = -1;
490 }
491 }
492 }
493
494 /**
495 * Change this->interface_type on a variable that previously had a
496 * different, but compatible, interface_type. This is used during linking
497 * to set the size of arrays in interface blocks.
498 */
499 void change_interface_type(const struct glsl_type *type)
500 {
501 if (this->u.max_ifc_array_access != NULL) {
502 /* max_ifc_array_access has already been allocated, so make sure the
503 * new interface has the same number of fields as the old one.
504 */
505 assert(this->interface_type->length == type->length);
506 }
507 this->interface_type = type;
508 }
509
510 /**
511 * Change this->interface_type on a variable that previously had a
512 * different, and incompatible, interface_type. This is used during
513 * compilation to handle redeclaration of the built-in gl_PerVertex
514 * interface block.
515 */
516 void reinit_interface_type(const struct glsl_type *type)
517 {
518 if (this->u.max_ifc_array_access != NULL) {
519 #ifndef NDEBUG
520 /* Redeclaring gl_PerVertex is only allowed if none of the built-ins
521 * it defines have been accessed yet; so it's safe to throw away the
522 * old max_ifc_array_access pointer, since all of its values are
523 * zero.
524 */
525 for (unsigned i = 0; i < this->interface_type->length; i++)
526 assert(this->u.max_ifc_array_access[i] == -1);
527 #endif
528 ralloc_free(this->u.max_ifc_array_access);
529 this->u.max_ifc_array_access = NULL;
530 }
531 this->interface_type = NULL;
532 init_interface_type(type);
533 }
534
535 const glsl_type *get_interface_type() const
536 {
537 return this->interface_type;
538 }
539
540 enum glsl_interface_packing get_interface_type_packing() const
541 {
542 return this->interface_type->get_interface_packing();
543 }
544 /**
545 * Get the max_ifc_array_access pointer
546 *
547 * A "set" function is not needed because the array is dynmically allocated
548 * as necessary.
549 */
550 inline int *get_max_ifc_array_access()
551 {
552 assert(this->data._num_state_slots == 0);
553 return this->u.max_ifc_array_access;
554 }
555
556 inline unsigned get_num_state_slots() const
557 {
558 assert(!this->is_interface_instance()
559 || this->data._num_state_slots == 0);
560 return this->data._num_state_slots;
561 }
562
563 inline void set_num_state_slots(unsigned n)
564 {
565 assert(!this->is_interface_instance()
566 || n == 0);
567 this->data._num_state_slots = n;
568 }
569
570 inline ir_state_slot *get_state_slots()
571 {
572 return this->is_interface_instance() ? NULL : this->u.state_slots;
573 }
574
575 inline const ir_state_slot *get_state_slots() const
576 {
577 return this->is_interface_instance() ? NULL : this->u.state_slots;
578 }
579
580 inline ir_state_slot *allocate_state_slots(unsigned n)
581 {
582 assert(!this->is_interface_instance());
583
584 this->u.state_slots = ralloc_array(this, ir_state_slot, n);
585 this->data._num_state_slots = 0;
586
587 if (this->u.state_slots != NULL)
588 this->data._num_state_slots = n;
589
590 return this->u.state_slots;
591 }
592
593 inline bool is_interpolation_flat() const
594 {
595 return this->data.interpolation == INTERP_MODE_FLAT ||
596 this->type->contains_integer() ||
597 this->type->contains_double();
598 }
599
600 inline bool is_name_ralloced() const
601 {
602 return this->name != ir_variable::tmp_name;
603 }
604
605 /**
606 * Enable emitting extension warnings for this variable
607 */
608 void enable_extension_warning(const char *extension);
609
610 /**
611 * Get the extension warning string for this variable
612 *
613 * If warnings are not enabled, \c NULL is returned.
614 */
615 const char *get_extension_warning() const;
616
617 /**
618 * Declared type of the variable
619 */
620 const struct glsl_type *type;
621
622 /**
623 * Declared name of the variable
624 */
625 const char *name;
626
627 struct ir_variable_data {
628
629 /**
630 * Is the variable read-only?
631 *
632 * This is set for variables declared as \c const, shader inputs,
633 * and uniforms.
634 */
635 unsigned read_only:1;
636 unsigned centroid:1;
637 unsigned sample:1;
638 unsigned patch:1;
639 unsigned invariant:1;
640 unsigned precise:1;
641
642 /**
643 * Has this variable been used for reading or writing?
644 *
645 * Several GLSL semantic checks require knowledge of whether or not a
646 * variable has been used. For example, it is an error to redeclare a
647 * variable as invariant after it has been used.
648 *
649 * This is only maintained in the ast_to_hir.cpp path, not in
650 * Mesa's fixed function or ARB program paths.
651 */
652 unsigned used:1;
653
654 /**
655 * Has this variable been statically assigned?
656 *
657 * This answers whether the variable was assigned in any path of
658 * the shader during ast_to_hir. This doesn't answer whether it is
659 * still written after dead code removal, nor is it maintained in
660 * non-ast_to_hir.cpp (GLSL parsing) paths.
661 */
662 unsigned assigned:1;
663
664 /**
665 * When separate shader programs are enabled, only input/outputs between
666 * the stages of a multi-stage separate program can be safely removed
667 * from the shader interface. Other input/outputs must remains active.
668 */
669 unsigned always_active_io:1;
670
671 /**
672 * Enum indicating how the variable was declared. See
673 * ir_var_declaration_type.
674 *
675 * This is used to detect certain kinds of illegal variable redeclarations.
676 */
677 unsigned how_declared:2;
678
679 /**
680 * Storage class of the variable.
681 *
682 * \sa ir_variable_mode
683 */
684 unsigned mode:4;
685
686 /**
687 * Interpolation mode for shader inputs / outputs
688 *
689 * \sa glsl_interp_mode
690 */
691 unsigned interpolation:2;
692
693 /**
694 * \name ARB_fragment_coord_conventions
695 * @{
696 */
697 unsigned origin_upper_left:1;
698 unsigned pixel_center_integer:1;
699 /*@}*/
700
701 /**
702 * Was the location explicitly set in the shader?
703 *
704 * If the location is explicitly set in the shader, it \b cannot be changed
705 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
706 * no effect).
707 */
708 unsigned explicit_location:1;
709 unsigned explicit_index:1;
710
711 /**
712 * Was an initial binding explicitly set in the shader?
713 *
714 * If so, constant_value contains an integer ir_constant representing the
715 * initial binding point.
716 */
717 unsigned explicit_binding:1;
718
719 /**
720 * Was an initial component explicitly set in the shader?
721 */
722 unsigned explicit_component:1;
723
724 /**
725 * Does this variable have an initializer?
726 *
727 * This is used by the linker to cross-validiate initializers of global
728 * variables.
729 */
730 unsigned has_initializer:1;
731
732 /**
733 * Is this variable a generic output or input that has not yet been matched
734 * up to a variable in another stage of the pipeline?
735 *
736 * This is used by the linker as scratch storage while assigning locations
737 * to generic inputs and outputs.
738 */
739 unsigned is_unmatched_generic_inout:1;
740
741 /**
742 * Is this varying used only by transform feedback?
743 *
744 * This is used by the linker to decide if its safe to pack the varying.
745 */
746 unsigned is_xfb_only:1;
747
748 /**
749 * Was a transfor feedback buffer set in the shader?
750 */
751 unsigned explicit_xfb_buffer:1;
752
753 /**
754 * Was a transfor feedback offset set in the shader?
755 */
756 unsigned explicit_xfb_offset:1;
757
758 /**
759 * Was a transfor feedback stride set in the shader?
760 */
761 unsigned explicit_xfb_stride:1;
762
763 /**
764 * If non-zero, then this variable may be packed along with other variables
765 * into a single varying slot, so this offset should be applied when
766 * accessing components. For example, an offset of 1 means that the x
767 * component of this variable is actually stored in component y of the
768 * location specified by \c location.
769 */
770 unsigned location_frac:2;
771
772 /**
773 * Layout of the matrix. Uses glsl_matrix_layout values.
774 */
775 unsigned matrix_layout:2;
776
777 /**
778 * Non-zero if this variable was created by lowering a named interface
779 * block.
780 */
781 unsigned from_named_ifc_block:1;
782
783 /**
784 * Non-zero if the variable must be a shader input. This is useful for
785 * constraints on function parameters.
786 */
787 unsigned must_be_shader_input:1;
788
789 /**
790 * Output index for dual source blending.
791 *
792 * \note
793 * The GLSL spec only allows the values 0 or 1 for the index in \b dual
794 * source blending.
795 */
796 unsigned index:1;
797
798 /**
799 * Precision qualifier.
800 *
801 * In desktop GLSL we do not care about precision qualifiers at all, in
802 * fact, the spec says that precision qualifiers are ignored.
803 *
804 * To make things easy, we make it so that this field is always
805 * GLSL_PRECISION_NONE on desktop shaders. This way all the variables
806 * have the same precision value and the checks we add in the compiler
807 * for this field will never break a desktop shader compile.
808 */
809 unsigned precision:2;
810
811 /**
812 * \brief Layout qualifier for gl_FragDepth.
813 *
814 * This is not equal to \c ir_depth_layout_none if and only if this
815 * variable is \c gl_FragDepth and a layout qualifier is specified.
816 */
817 ir_depth_layout depth_layout:3;
818
819 /**
820 * ARB_shader_image_load_store qualifiers.
821 */
822 unsigned image_read_only:1; /**< "readonly" qualifier. */
823 unsigned image_write_only:1; /**< "writeonly" qualifier. */
824 unsigned image_coherent:1;
825 unsigned image_volatile:1;
826 unsigned image_restrict:1;
827
828 /**
829 * ARB_shader_storage_buffer_object
830 */
831 unsigned from_ssbo_unsized_array:1; /**< unsized array buffer variable. */
832
833 unsigned implicit_sized_array:1;
834
835 /**
836 * Whether this is a fragment shader output implicitly initialized with
837 * the previous contents of the specified render target at the
838 * framebuffer location corresponding to this shader invocation.
839 */
840 unsigned fb_fetch_output:1;
841
842 /**
843 * Emit a warning if this variable is accessed.
844 */
845 private:
846 uint8_t warn_extension_index;
847
848 public:
849 /** Image internal format if specified explicitly, otherwise GL_NONE. */
850 uint16_t image_format;
851
852 private:
853 /**
854 * Number of state slots used
855 *
856 * \note
857 * This could be stored in as few as 7-bits, if necessary. If it is made
858 * smaller, add an assertion to \c ir_variable::allocate_state_slots to
859 * be safe.
860 */
861 uint16_t _num_state_slots;
862
863 public:
864 /**
865 * Initial binding point for a sampler, atomic, or UBO.
866 *
867 * For array types, this represents the binding point for the first element.
868 */
869 int16_t binding;
870
871 /**
872 * Storage location of the base of this variable
873 *
874 * The precise meaning of this field depends on the nature of the variable.
875 *
876 * - Vertex shader input: one of the values from \c gl_vert_attrib.
877 * - Vertex shader output: one of the values from \c gl_varying_slot.
878 * - Geometry shader input: one of the values from \c gl_varying_slot.
879 * - Geometry shader output: one of the values from \c gl_varying_slot.
880 * - Fragment shader input: one of the values from \c gl_varying_slot.
881 * - Fragment shader output: one of the values from \c gl_frag_result.
882 * - Uniforms: Per-stage uniform slot number for default uniform block.
883 * - Uniforms: Index within the uniform block definition for UBO members.
884 * - Non-UBO Uniforms: explicit location until linking then reused to
885 * store uniform slot number.
886 * - Other: This field is not currently used.
887 *
888 * If the variable is a uniform, shader input, or shader output, and the
889 * slot has not been assigned, the value will be -1.
890 */
891 int location;
892
893 /**
894 * for glsl->tgsi/mesa IR we need to store the index into the
895 * parameters for uniforms, initially the code overloaded location
896 * but this causes problems with indirect samplers and AoA.
897 * This is assigned in _mesa_generate_parameters_list_for_uniforms.
898 */
899 int param_index;
900
901 /**
902 * Vertex stream output identifier.
903 */
904 unsigned stream;
905
906 /**
907 * Atomic, transform feedback or block member offset.
908 */
909 unsigned offset;
910
911 /**
912 * Highest element accessed with a constant expression array index
913 *
914 * Not used for non-array variables. -1 is never accessed.
915 */
916 int max_array_access;
917
918 /**
919 * Transform feedback buffer.
920 */
921 unsigned xfb_buffer;
922
923 /**
924 * Transform feedback stride.
925 */
926 unsigned xfb_stride;
927
928 /**
929 * Allow (only) ir_variable direct access private members.
930 */
931 friend class ir_variable;
932 } data;
933
934 /**
935 * Value assigned in the initializer of a variable declared "const"
936 */
937 ir_constant *constant_value;
938
939 /**
940 * Constant expression assigned in the initializer of the variable
941 *
942 * \warning
943 * This field and \c ::constant_value are distinct. Even if the two fields
944 * refer to constants with the same value, they must point to separate
945 * objects.
946 */
947 ir_constant *constant_initializer;
948
949 private:
950 static const char *const warn_extension_table[];
951
952 union {
953 /**
954 * For variables which satisfy the is_interface_instance() predicate,
955 * this points to an array of integers such that if the ith member of
956 * the interface block is an array, max_ifc_array_access[i] is the
957 * maximum array element of that member that has been accessed. If the
958 * ith member of the interface block is not an array,
959 * max_ifc_array_access[i] is unused.
960 *
961 * For variables whose type is not an interface block, this pointer is
962 * NULL.
963 */
964 int *max_ifc_array_access;
965
966 /**
967 * Built-in state that backs this uniform
968 *
969 * Once set at variable creation, \c state_slots must remain invariant.
970 *
971 * If the variable is not a uniform, \c _num_state_slots will be zero
972 * and \c state_slots will be \c NULL.
973 */
974 ir_state_slot *state_slots;
975 } u;
976
977 /**
978 * For variables that are in an interface block or are an instance of an
979 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
980 *
981 * \sa ir_variable::location
982 */
983 const glsl_type *interface_type;
984
985 /**
986 * Name used for anonymous compiler temporaries
987 */
988 static const char tmp_name[];
989
990 public:
991 /**
992 * Should the construct keep names for ir_var_temporary variables?
993 *
994 * When this global is false, names passed to the constructor for
995 * \c ir_var_temporary variables will be dropped. Instead, the variable will
996 * be named "compiler_temp". This name will be in static storage.
997 *
998 * \warning
999 * \b NEVER change the mode of an \c ir_var_temporary.
1000 *
1001 * \warning
1002 * This variable is \b not thread-safe. It is global, \b not
1003 * per-context. It begins life false. A context can, at some point, make
1004 * it true. From that point on, it will be true forever. This should be
1005 * okay since it will only be set true while debugging.
1006 */
1007 static bool temporaries_allocate_names;
1008 };
1009
1010 /**
1011 * A function that returns whether a built-in function is available in the
1012 * current shading language (based on version, ES or desktop, and extensions).
1013 */
1014 typedef bool (*builtin_available_predicate)(const _mesa_glsl_parse_state *);
1015
1016 /*@{*/
1017 /**
1018 * The representation of a function instance; may be the full definition or
1019 * simply a prototype.
1020 */
1021 class ir_function_signature : public ir_instruction {
1022 /* An ir_function_signature will be part of the list of signatures in
1023 * an ir_function.
1024 */
1025 public:
1026 ir_function_signature(const glsl_type *return_type,
1027 builtin_available_predicate builtin_avail = NULL);
1028
1029 virtual ir_function_signature *clone(void *mem_ctx,
1030 struct hash_table *ht) const;
1031 ir_function_signature *clone_prototype(void *mem_ctx,
1032 struct hash_table *ht) const;
1033
1034 virtual void accept(ir_visitor *v)
1035 {
1036 v->visit(this);
1037 }
1038
1039 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1040
1041 /**
1042 * Attempt to evaluate this function as a constant expression,
1043 * given a list of the actual parameters and the variable context.
1044 * Returns NULL for non-built-ins.
1045 */
1046 ir_constant *constant_expression_value(exec_list *actual_parameters, struct hash_table *variable_context);
1047
1048 /**
1049 * Get the name of the function for which this is a signature
1050 */
1051 const char *function_name() const;
1052
1053 /**
1054 * Get a handle to the function for which this is a signature
1055 *
1056 * There is no setter function, this function returns a \c const pointer,
1057 * and \c ir_function_signature::_function is private for a reason. The
1058 * only way to make a connection between a function and function signature
1059 * is via \c ir_function::add_signature. This helps ensure that certain
1060 * invariants (i.e., a function signature is in the list of signatures for
1061 * its \c _function) are met.
1062 *
1063 * \sa ir_function::add_signature
1064 */
1065 inline const class ir_function *function() const
1066 {
1067 return this->_function;
1068 }
1069
1070 /**
1071 * Check whether the qualifiers match between this signature's parameters
1072 * and the supplied parameter list. If not, returns the name of the first
1073 * parameter with mismatched qualifiers (for use in error messages).
1074 */
1075 const char *qualifiers_match(exec_list *params);
1076
1077 /**
1078 * Replace the current parameter list with the given one. This is useful
1079 * if the current information came from a prototype, and either has invalid
1080 * or missing parameter names.
1081 */
1082 void replace_parameters(exec_list *new_params);
1083
1084 /**
1085 * Function return type.
1086 *
1087 * \note This discards the optional precision qualifier.
1088 */
1089 const struct glsl_type *return_type;
1090
1091 /**
1092 * List of ir_variable of function parameters.
1093 *
1094 * This represents the storage. The paramaters passed in a particular
1095 * call will be in ir_call::actual_paramaters.
1096 */
1097 struct exec_list parameters;
1098
1099 /** Whether or not this function has a body (which may be empty). */
1100 unsigned is_defined:1;
1101
1102 /** Whether or not this function signature is a built-in. */
1103 bool is_builtin() const;
1104
1105 /**
1106 * Whether or not this function is an intrinsic to be implemented
1107 * by the driver.
1108 */
1109 bool is_intrinsic;
1110
1111 /** Whether or not a built-in is available for this shader. */
1112 bool is_builtin_available(const _mesa_glsl_parse_state *state) const;
1113
1114 /** Body of instructions in the function. */
1115 struct exec_list body;
1116
1117 private:
1118 /**
1119 * A function pointer to a predicate that answers whether a built-in
1120 * function is available in the current shader. NULL if not a built-in.
1121 */
1122 builtin_available_predicate builtin_avail;
1123
1124 /** Function of which this signature is one overload. */
1125 class ir_function *_function;
1126
1127 /** Function signature of which this one is a prototype clone */
1128 const ir_function_signature *origin;
1129
1130 friend class ir_function;
1131
1132 /**
1133 * Helper function to run a list of instructions for constant
1134 * expression evaluation.
1135 *
1136 * The hash table represents the values of the visible variables.
1137 * There are no scoping issues because the table is indexed on
1138 * ir_variable pointers, not variable names.
1139 *
1140 * Returns false if the expression is not constant, true otherwise,
1141 * and the value in *result if result is non-NULL.
1142 */
1143 bool constant_expression_evaluate_expression_list(const struct exec_list &body,
1144 struct hash_table *variable_context,
1145 ir_constant **result);
1146 };
1147
1148
1149 /**
1150 * Header for tracking multiple overloaded functions with the same name.
1151 * Contains a list of ir_function_signatures representing each of the
1152 * actual functions.
1153 */
1154 class ir_function : public ir_instruction {
1155 public:
1156 ir_function(const char *name);
1157
1158 virtual ir_function *clone(void *mem_ctx, struct hash_table *ht) const;
1159
1160 virtual void accept(ir_visitor *v)
1161 {
1162 v->visit(this);
1163 }
1164
1165 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1166
1167 void add_signature(ir_function_signature *sig)
1168 {
1169 sig->_function = this;
1170 this->signatures.push_tail(sig);
1171 }
1172
1173 /**
1174 * Find a signature that matches a set of actual parameters, taking implicit
1175 * conversions into account. Also flags whether the match was exact.
1176 */
1177 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
1178 const exec_list *actual_param,
1179 bool allow_builtins,
1180 bool *match_is_exact);
1181
1182 /**
1183 * Find a signature that matches a set of actual parameters, taking implicit
1184 * conversions into account.
1185 */
1186 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
1187 const exec_list *actual_param,
1188 bool allow_builtins);
1189
1190 /**
1191 * Find a signature that exactly matches a set of actual parameters without
1192 * any implicit type conversions.
1193 */
1194 ir_function_signature *exact_matching_signature(_mesa_glsl_parse_state *state,
1195 const exec_list *actual_ps);
1196
1197 /**
1198 * Name of the function.
1199 */
1200 const char *name;
1201
1202 /** Whether or not this function has a signature that isn't a built-in. */
1203 bool has_user_signature();
1204
1205 /**
1206 * List of ir_function_signature for each overloaded function with this name.
1207 */
1208 struct exec_list signatures;
1209
1210 /**
1211 * is this function a subroutine type declaration
1212 * e.g. subroutine void type1(float arg1);
1213 */
1214 bool is_subroutine;
1215
1216 /**
1217 * is this function associated to a subroutine type
1218 * e.g. subroutine (type1, type2) function_name { function_body };
1219 * would have num_subroutine_types 2,
1220 * and pointers to the type1 and type2 types.
1221 */
1222 int num_subroutine_types;
1223 const struct glsl_type **subroutine_types;
1224
1225 int subroutine_index;
1226 };
1227
1228 inline const char *ir_function_signature::function_name() const
1229 {
1230 return this->_function->name;
1231 }
1232 /*@}*/
1233
1234
1235 /**
1236 * IR instruction representing high-level if-statements
1237 */
1238 class ir_if : public ir_instruction {
1239 public:
1240 ir_if(ir_rvalue *condition)
1241 : ir_instruction(ir_type_if), condition(condition)
1242 {
1243 }
1244
1245 virtual ir_if *clone(void *mem_ctx, struct hash_table *ht) const;
1246
1247 virtual void accept(ir_visitor *v)
1248 {
1249 v->visit(this);
1250 }
1251
1252 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1253
1254 ir_rvalue *condition;
1255 /** List of ir_instruction for the body of the then branch */
1256 exec_list then_instructions;
1257 /** List of ir_instruction for the body of the else branch */
1258 exec_list else_instructions;
1259 };
1260
1261
1262 /**
1263 * IR instruction representing a high-level loop structure.
1264 */
1265 class ir_loop : public ir_instruction {
1266 public:
1267 ir_loop();
1268
1269 virtual ir_loop *clone(void *mem_ctx, struct hash_table *ht) const;
1270
1271 virtual void accept(ir_visitor *v)
1272 {
1273 v->visit(this);
1274 }
1275
1276 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1277
1278 /** List of ir_instruction that make up the body of the loop. */
1279 exec_list body_instructions;
1280 };
1281
1282
1283 class ir_assignment : public ir_instruction {
1284 public:
1285 ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs, ir_rvalue *condition = NULL);
1286
1287 /**
1288 * Construct an assignment with an explicit write mask
1289 *
1290 * \note
1291 * Since a write mask is supplied, the LHS must already be a bare
1292 * \c ir_dereference. The cannot be any swizzles in the LHS.
1293 */
1294 ir_assignment(ir_dereference *lhs, ir_rvalue *rhs, ir_rvalue *condition,
1295 unsigned write_mask);
1296
1297 virtual ir_assignment *clone(void *mem_ctx, struct hash_table *ht) const;
1298
1299 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1300
1301 virtual void accept(ir_visitor *v)
1302 {
1303 v->visit(this);
1304 }
1305
1306 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1307
1308 /**
1309 * Get a whole variable written by an assignment
1310 *
1311 * If the LHS of the assignment writes a whole variable, the variable is
1312 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1313 * assignment are:
1314 *
1315 * - Assigning to a scalar
1316 * - Assigning to all components of a vector
1317 * - Whole array (or matrix) assignment
1318 * - Whole structure assignment
1319 */
1320 ir_variable *whole_variable_written();
1321
1322 /**
1323 * Set the LHS of an assignment
1324 */
1325 void set_lhs(ir_rvalue *lhs);
1326
1327 /**
1328 * Left-hand side of the assignment.
1329 *
1330 * This should be treated as read only. If you need to set the LHS of an
1331 * assignment, use \c ir_assignment::set_lhs.
1332 */
1333 ir_dereference *lhs;
1334
1335 /**
1336 * Value being assigned
1337 */
1338 ir_rvalue *rhs;
1339
1340 /**
1341 * Optional condition for the assignment.
1342 */
1343 ir_rvalue *condition;
1344
1345
1346 /**
1347 * Component mask written
1348 *
1349 * For non-vector types in the LHS, this field will be zero. For vector
1350 * types, a bit will be set for each component that is written. Note that
1351 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1352 *
1353 * A partially-set write mask means that each enabled channel gets
1354 * the value from a consecutive channel of the rhs. For example,
1355 * to write just .xyw of gl_FrontColor with color:
1356 *
1357 * (assign (constant bool (1)) (xyw)
1358 * (var_ref gl_FragColor)
1359 * (swiz xyw (var_ref color)))
1360 */
1361 unsigned write_mask:4;
1362 };
1363
1364 #include "ir_expression_operation.h"
1365
1366 extern const char *const ir_expression_operation_strings[ir_last_opcode + 1];
1367
1368 class ir_expression : public ir_rvalue {
1369 public:
1370 ir_expression(int op, const struct glsl_type *type,
1371 ir_rvalue *op0, ir_rvalue *op1 = NULL,
1372 ir_rvalue *op2 = NULL, ir_rvalue *op3 = NULL);
1373
1374 /**
1375 * Constructor for unary operation expressions
1376 */
1377 ir_expression(int op, ir_rvalue *);
1378
1379 /**
1380 * Constructor for binary operation expressions
1381 */
1382 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1);
1383
1384 /**
1385 * Constructor for ternary operation expressions
1386 */
1387 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1, ir_rvalue *op2);
1388
1389 virtual bool equals(const ir_instruction *ir,
1390 enum ir_node_type ignore = ir_type_unset) const;
1391
1392 virtual ir_expression *clone(void *mem_ctx, struct hash_table *ht) const;
1393
1394 /**
1395 * Attempt to constant-fold the expression
1396 *
1397 * The "variable_context" hash table links ir_variable * to ir_constant *
1398 * that represent the variables' values. \c NULL represents an empty
1399 * context.
1400 *
1401 * If the expression cannot be constant folded, this method will return
1402 * \c NULL.
1403 */
1404 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1405
1406 /**
1407 * Determine the number of operands used by an expression
1408 */
1409 static unsigned int get_num_operands(ir_expression_operation);
1410
1411 /**
1412 * Determine the number of operands used by an expression
1413 */
1414 unsigned int get_num_operands() const
1415 {
1416 return (this->operation == ir_quadop_vector)
1417 ? this->type->vector_elements : get_num_operands(operation);
1418 }
1419
1420 /**
1421 * Return whether the expression operates on vectors horizontally.
1422 */
1423 bool is_horizontal() const
1424 {
1425 return operation == ir_binop_all_equal ||
1426 operation == ir_binop_any_nequal ||
1427 operation == ir_binop_dot ||
1428 operation == ir_binop_vector_extract ||
1429 operation == ir_triop_vector_insert ||
1430 operation == ir_binop_ubo_load ||
1431 operation == ir_quadop_vector;
1432 }
1433
1434 /**
1435 * Do a reverse-lookup to translate the given string into an operator.
1436 */
1437 static ir_expression_operation get_operator(const char *);
1438
1439 virtual void accept(ir_visitor *v)
1440 {
1441 v->visit(this);
1442 }
1443
1444 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1445
1446 virtual ir_variable *variable_referenced() const;
1447
1448 ir_expression_operation operation;
1449 ir_rvalue *operands[4];
1450 };
1451
1452
1453 /**
1454 * HIR instruction representing a high-level function call, containing a list
1455 * of parameters and returning a value in the supplied temporary.
1456 */
1457 class ir_call : public ir_instruction {
1458 public:
1459 ir_call(ir_function_signature *callee,
1460 ir_dereference_variable *return_deref,
1461 exec_list *actual_parameters)
1462 : ir_instruction(ir_type_call), return_deref(return_deref), callee(callee), sub_var(NULL), array_idx(NULL)
1463 {
1464 assert(callee->return_type != NULL);
1465 actual_parameters->move_nodes_to(& this->actual_parameters);
1466 this->use_builtin = callee->is_builtin();
1467 }
1468
1469 ir_call(ir_function_signature *callee,
1470 ir_dereference_variable *return_deref,
1471 exec_list *actual_parameters,
1472 ir_variable *var, ir_rvalue *array_idx)
1473 : ir_instruction(ir_type_call), return_deref(return_deref), callee(callee), sub_var(var), array_idx(array_idx)
1474 {
1475 assert(callee->return_type != NULL);
1476 actual_parameters->move_nodes_to(& this->actual_parameters);
1477 this->use_builtin = callee->is_builtin();
1478 }
1479
1480 virtual ir_call *clone(void *mem_ctx, struct hash_table *ht) const;
1481
1482 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1483
1484 virtual void accept(ir_visitor *v)
1485 {
1486 v->visit(this);
1487 }
1488
1489 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1490
1491 /**
1492 * Get the name of the function being called.
1493 */
1494 const char *callee_name() const
1495 {
1496 return callee->function_name();
1497 }
1498
1499 /**
1500 * Generates an inline version of the function before @ir,
1501 * storing the return value in return_deref.
1502 */
1503 void generate_inline(ir_instruction *ir);
1504
1505 /**
1506 * Storage for the function's return value.
1507 * This must be NULL if the return type is void.
1508 */
1509 ir_dereference_variable *return_deref;
1510
1511 /**
1512 * The specific function signature being called.
1513 */
1514 ir_function_signature *callee;
1515
1516 /* List of ir_rvalue of paramaters passed in this call. */
1517 exec_list actual_parameters;
1518
1519 /** Should this call only bind to a built-in function? */
1520 bool use_builtin;
1521
1522 /*
1523 * ARB_shader_subroutine support -
1524 * the subroutine uniform variable and array index
1525 * rvalue to be used in the lowering pass later.
1526 */
1527 ir_variable *sub_var;
1528 ir_rvalue *array_idx;
1529 };
1530
1531
1532 /**
1533 * \name Jump-like IR instructions.
1534 *
1535 * These include \c break, \c continue, \c return, and \c discard.
1536 */
1537 /*@{*/
1538 class ir_jump : public ir_instruction {
1539 protected:
1540 ir_jump(enum ir_node_type t)
1541 : ir_instruction(t)
1542 {
1543 }
1544 };
1545
1546 class ir_return : public ir_jump {
1547 public:
1548 ir_return()
1549 : ir_jump(ir_type_return), value(NULL)
1550 {
1551 }
1552
1553 ir_return(ir_rvalue *value)
1554 : ir_jump(ir_type_return), value(value)
1555 {
1556 }
1557
1558 virtual ir_return *clone(void *mem_ctx, struct hash_table *) const;
1559
1560 ir_rvalue *get_value() const
1561 {
1562 return value;
1563 }
1564
1565 virtual void accept(ir_visitor *v)
1566 {
1567 v->visit(this);
1568 }
1569
1570 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1571
1572 ir_rvalue *value;
1573 };
1574
1575
1576 /**
1577 * Jump instructions used inside loops
1578 *
1579 * These include \c break and \c continue. The \c break within a loop is
1580 * different from the \c break within a switch-statement.
1581 *
1582 * \sa ir_switch_jump
1583 */
1584 class ir_loop_jump : public ir_jump {
1585 public:
1586 enum jump_mode {
1587 jump_break,
1588 jump_continue
1589 };
1590
1591 ir_loop_jump(jump_mode mode)
1592 : ir_jump(ir_type_loop_jump)
1593 {
1594 this->mode = mode;
1595 }
1596
1597 virtual ir_loop_jump *clone(void *mem_ctx, struct hash_table *) const;
1598
1599 virtual void accept(ir_visitor *v)
1600 {
1601 v->visit(this);
1602 }
1603
1604 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1605
1606 bool is_break() const
1607 {
1608 return mode == jump_break;
1609 }
1610
1611 bool is_continue() const
1612 {
1613 return mode == jump_continue;
1614 }
1615
1616 /** Mode selector for the jump instruction. */
1617 enum jump_mode mode;
1618 };
1619
1620 /**
1621 * IR instruction representing discard statements.
1622 */
1623 class ir_discard : public ir_jump {
1624 public:
1625 ir_discard()
1626 : ir_jump(ir_type_discard)
1627 {
1628 this->condition = NULL;
1629 }
1630
1631 ir_discard(ir_rvalue *cond)
1632 : ir_jump(ir_type_discard)
1633 {
1634 this->condition = cond;
1635 }
1636
1637 virtual ir_discard *clone(void *mem_ctx, struct hash_table *ht) const;
1638
1639 virtual void accept(ir_visitor *v)
1640 {
1641 v->visit(this);
1642 }
1643
1644 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1645
1646 ir_rvalue *condition;
1647 };
1648 /*@}*/
1649
1650
1651 /**
1652 * Texture sampling opcodes used in ir_texture
1653 */
1654 enum ir_texture_opcode {
1655 ir_tex, /**< Regular texture look-up */
1656 ir_txb, /**< Texture look-up with LOD bias */
1657 ir_txl, /**< Texture look-up with explicit LOD */
1658 ir_txd, /**< Texture look-up with partial derivatvies */
1659 ir_txf, /**< Texel fetch with explicit LOD */
1660 ir_txf_ms, /**< Multisample texture fetch */
1661 ir_txs, /**< Texture size */
1662 ir_lod, /**< Texture lod query */
1663 ir_tg4, /**< Texture gather */
1664 ir_query_levels, /**< Texture levels query */
1665 ir_texture_samples, /**< Texture samples query */
1666 ir_samples_identical, /**< Query whether all samples are definitely identical. */
1667 };
1668
1669
1670 /**
1671 * IR instruction to sample a texture
1672 *
1673 * The specific form of the IR instruction depends on the \c mode value
1674 * selected from \c ir_texture_opcodes. In the printed IR, these will
1675 * appear as:
1676 *
1677 * Texel offset (0 or an expression)
1678 * | Projection divisor
1679 * | | Shadow comparitor
1680 * | | |
1681 * v v v
1682 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1683 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1684 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1685 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1686 * (txf <type> <sampler> <coordinate> 0 <lod>)
1687 * (txf_ms
1688 * <type> <sampler> <coordinate> <sample_index>)
1689 * (txs <type> <sampler> <lod>)
1690 * (lod <type> <sampler> <coordinate>)
1691 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
1692 * (query_levels <type> <sampler>)
1693 * (samples_identical <sampler> <coordinate>)
1694 */
1695 class ir_texture : public ir_rvalue {
1696 public:
1697 ir_texture(enum ir_texture_opcode op)
1698 : ir_rvalue(ir_type_texture),
1699 op(op), sampler(NULL), coordinate(NULL), projector(NULL),
1700 shadow_comparitor(NULL), offset(NULL)
1701 {
1702 memset(&lod_info, 0, sizeof(lod_info));
1703 }
1704
1705 virtual ir_texture *clone(void *mem_ctx, struct hash_table *) const;
1706
1707 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1708
1709 virtual void accept(ir_visitor *v)
1710 {
1711 v->visit(this);
1712 }
1713
1714 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1715
1716 virtual bool equals(const ir_instruction *ir,
1717 enum ir_node_type ignore = ir_type_unset) const;
1718
1719 /**
1720 * Return a string representing the ir_texture_opcode.
1721 */
1722 const char *opcode_string();
1723
1724 /** Set the sampler and type. */
1725 void set_sampler(ir_dereference *sampler, const glsl_type *type);
1726
1727 /**
1728 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1729 */
1730 static ir_texture_opcode get_opcode(const char *);
1731
1732 enum ir_texture_opcode op;
1733
1734 /** Sampler to use for the texture access. */
1735 ir_dereference *sampler;
1736
1737 /** Texture coordinate to sample */
1738 ir_rvalue *coordinate;
1739
1740 /**
1741 * Value used for projective divide.
1742 *
1743 * If there is no projective divide (the common case), this will be
1744 * \c NULL. Optimization passes should check for this to point to a constant
1745 * of 1.0 and replace that with \c NULL.
1746 */
1747 ir_rvalue *projector;
1748
1749 /**
1750 * Coordinate used for comparison on shadow look-ups.
1751 *
1752 * If there is no shadow comparison, this will be \c NULL. For the
1753 * \c ir_txf opcode, this *must* be \c NULL.
1754 */
1755 ir_rvalue *shadow_comparitor;
1756
1757 /** Texel offset. */
1758 ir_rvalue *offset;
1759
1760 union {
1761 ir_rvalue *lod; /**< Floating point LOD */
1762 ir_rvalue *bias; /**< Floating point LOD bias */
1763 ir_rvalue *sample_index; /**< MSAA sample index */
1764 ir_rvalue *component; /**< Gather component selector */
1765 struct {
1766 ir_rvalue *dPdx; /**< Partial derivative of coordinate wrt X */
1767 ir_rvalue *dPdy; /**< Partial derivative of coordinate wrt Y */
1768 } grad;
1769 } lod_info;
1770 };
1771
1772
1773 struct ir_swizzle_mask {
1774 unsigned x:2;
1775 unsigned y:2;
1776 unsigned z:2;
1777 unsigned w:2;
1778
1779 /**
1780 * Number of components in the swizzle.
1781 */
1782 unsigned num_components:3;
1783
1784 /**
1785 * Does the swizzle contain duplicate components?
1786 *
1787 * L-value swizzles cannot contain duplicate components.
1788 */
1789 unsigned has_duplicates:1;
1790 };
1791
1792
1793 class ir_swizzle : public ir_rvalue {
1794 public:
1795 ir_swizzle(ir_rvalue *, unsigned x, unsigned y, unsigned z, unsigned w,
1796 unsigned count);
1797
1798 ir_swizzle(ir_rvalue *val, const unsigned *components, unsigned count);
1799
1800 ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask);
1801
1802 virtual ir_swizzle *clone(void *mem_ctx, struct hash_table *) const;
1803
1804 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1805
1806 /**
1807 * Construct an ir_swizzle from the textual representation. Can fail.
1808 */
1809 static ir_swizzle *create(ir_rvalue *, const char *, unsigned vector_length);
1810
1811 virtual void accept(ir_visitor *v)
1812 {
1813 v->visit(this);
1814 }
1815
1816 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1817
1818 virtual bool equals(const ir_instruction *ir,
1819 enum ir_node_type ignore = ir_type_unset) const;
1820
1821 bool is_lvalue() const
1822 {
1823 return val->is_lvalue() && !mask.has_duplicates;
1824 }
1825
1826 /**
1827 * Get the variable that is ultimately referenced by an r-value
1828 */
1829 virtual ir_variable *variable_referenced() const;
1830
1831 ir_rvalue *val;
1832 ir_swizzle_mask mask;
1833
1834 private:
1835 /**
1836 * Initialize the mask component of a swizzle
1837 *
1838 * This is used by the \c ir_swizzle constructors.
1839 */
1840 void init_mask(const unsigned *components, unsigned count);
1841 };
1842
1843
1844 class ir_dereference : public ir_rvalue {
1845 public:
1846 virtual ir_dereference *clone(void *mem_ctx, struct hash_table *) const = 0;
1847
1848 bool is_lvalue() const;
1849
1850 /**
1851 * Get the variable that is ultimately referenced by an r-value
1852 */
1853 virtual ir_variable *variable_referenced() const = 0;
1854
1855 protected:
1856 ir_dereference(enum ir_node_type t)
1857 : ir_rvalue(t)
1858 {
1859 }
1860 };
1861
1862
1863 class ir_dereference_variable : public ir_dereference {
1864 public:
1865 ir_dereference_variable(ir_variable *var);
1866
1867 virtual ir_dereference_variable *clone(void *mem_ctx,
1868 struct hash_table *) const;
1869
1870 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1871
1872 virtual bool equals(const ir_instruction *ir,
1873 enum ir_node_type ignore = ir_type_unset) const;
1874
1875 /**
1876 * Get the variable that is ultimately referenced by an r-value
1877 */
1878 virtual ir_variable *variable_referenced() const
1879 {
1880 return this->var;
1881 }
1882
1883 virtual ir_variable *whole_variable_referenced()
1884 {
1885 /* ir_dereference_variable objects always dereference the entire
1886 * variable. However, if this dereference is dereferenced by anything
1887 * else, the complete deferefernce chain is not a whole-variable
1888 * dereference. This method should only be called on the top most
1889 * ir_rvalue in a dereference chain.
1890 */
1891 return this->var;
1892 }
1893
1894 virtual void accept(ir_visitor *v)
1895 {
1896 v->visit(this);
1897 }
1898
1899 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1900
1901 /**
1902 * Object being dereferenced.
1903 */
1904 ir_variable *var;
1905 };
1906
1907
1908 class ir_dereference_array : public ir_dereference {
1909 public:
1910 ir_dereference_array(ir_rvalue *value, ir_rvalue *array_index);
1911
1912 ir_dereference_array(ir_variable *var, ir_rvalue *array_index);
1913
1914 virtual ir_dereference_array *clone(void *mem_ctx,
1915 struct hash_table *) const;
1916
1917 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1918
1919 virtual bool equals(const ir_instruction *ir,
1920 enum ir_node_type ignore = ir_type_unset) const;
1921
1922 /**
1923 * Get the variable that is ultimately referenced by an r-value
1924 */
1925 virtual ir_variable *variable_referenced() const
1926 {
1927 return this->array->variable_referenced();
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 ir_rvalue *array;
1938 ir_rvalue *array_index;
1939
1940 private:
1941 void set_array(ir_rvalue *value);
1942 };
1943
1944
1945 class ir_dereference_record : public ir_dereference {
1946 public:
1947 ir_dereference_record(ir_rvalue *value, const char *field);
1948
1949 ir_dereference_record(ir_variable *var, const char *field);
1950
1951 virtual ir_dereference_record *clone(void *mem_ctx,
1952 struct hash_table *) const;
1953
1954 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1955
1956 /**
1957 * Get the variable that is ultimately referenced by an r-value
1958 */
1959 virtual ir_variable *variable_referenced() const
1960 {
1961 return this->record->variable_referenced();
1962 }
1963
1964 virtual void accept(ir_visitor *v)
1965 {
1966 v->visit(this);
1967 }
1968
1969 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1970
1971 ir_rvalue *record;
1972 const char *field;
1973 };
1974
1975
1976 /**
1977 * Data stored in an ir_constant
1978 */
1979 union ir_constant_data {
1980 unsigned u[16];
1981 int i[16];
1982 float f[16];
1983 bool b[16];
1984 double d[16];
1985 };
1986
1987
1988 class ir_constant : public ir_rvalue {
1989 public:
1990 ir_constant(const struct glsl_type *type, const ir_constant_data *data);
1991 ir_constant(bool b, unsigned vector_elements=1);
1992 ir_constant(unsigned int u, unsigned vector_elements=1);
1993 ir_constant(int i, unsigned vector_elements=1);
1994 ir_constant(float f, unsigned vector_elements=1);
1995 ir_constant(double d, unsigned vector_elements=1);
1996
1997 /**
1998 * Construct an ir_constant from a list of ir_constant values
1999 */
2000 ir_constant(const struct glsl_type *type, exec_list *values);
2001
2002 /**
2003 * Construct an ir_constant from a scalar component of another ir_constant
2004 *
2005 * The new \c ir_constant inherits the type of the component from the
2006 * source constant.
2007 *
2008 * \note
2009 * In the case of a matrix constant, the new constant is a scalar, \b not
2010 * a vector.
2011 */
2012 ir_constant(const ir_constant *c, unsigned i);
2013
2014 /**
2015 * Return a new ir_constant of the specified type containing all zeros.
2016 */
2017 static ir_constant *zero(void *mem_ctx, const glsl_type *type);
2018
2019 virtual ir_constant *clone(void *mem_ctx, struct hash_table *) const;
2020
2021 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
2022
2023 virtual void accept(ir_visitor *v)
2024 {
2025 v->visit(this);
2026 }
2027
2028 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2029
2030 virtual bool equals(const ir_instruction *ir,
2031 enum ir_node_type ignore = ir_type_unset) const;
2032
2033 /**
2034 * Get a particular component of a constant as a specific type
2035 *
2036 * This is useful, for example, to get a value from an integer constant
2037 * as a float or bool. This appears frequently when constructors are
2038 * called with all constant parameters.
2039 */
2040 /*@{*/
2041 bool get_bool_component(unsigned i) const;
2042 float get_float_component(unsigned i) const;
2043 double get_double_component(unsigned i) const;
2044 int get_int_component(unsigned i) const;
2045 unsigned get_uint_component(unsigned i) const;
2046 /*@}*/
2047
2048 ir_constant *get_array_element(unsigned i) const;
2049
2050 ir_constant *get_record_field(const char *name);
2051
2052 /**
2053 * Copy the values on another constant at a given offset.
2054 *
2055 * The offset is ignored for array or struct copies, it's only for
2056 * scalars or vectors into vectors or matrices.
2057 *
2058 * With identical types on both sides and zero offset it's clone()
2059 * without creating a new object.
2060 */
2061
2062 void copy_offset(ir_constant *src, int offset);
2063
2064 /**
2065 * Copy the values on another constant at a given offset and
2066 * following an assign-like mask.
2067 *
2068 * The mask is ignored for scalars.
2069 *
2070 * Note that this function only handles what assign can handle,
2071 * i.e. at most a vector as source and a column of a matrix as
2072 * destination.
2073 */
2074
2075 void copy_masked_offset(ir_constant *src, int offset, unsigned int mask);
2076
2077 /**
2078 * Determine whether a constant has the same value as another constant
2079 *
2080 * \sa ir_constant::is_zero, ir_constant::is_one,
2081 * ir_constant::is_negative_one
2082 */
2083 bool has_value(const ir_constant *) const;
2084
2085 /**
2086 * Return true if this ir_constant represents the given value.
2087 *
2088 * For vectors, this checks that each component is the given value.
2089 */
2090 virtual bool is_value(float f, int i) const;
2091 virtual bool is_zero() const;
2092 virtual bool is_one() const;
2093 virtual bool is_negative_one() const;
2094
2095 /**
2096 * Return true for constants that could be stored as 16-bit unsigned values.
2097 *
2098 * Note that this will return true even for signed integer ir_constants, as
2099 * long as the value is non-negative and fits in 16-bits.
2100 */
2101 virtual bool is_uint16_constant() const;
2102
2103 /**
2104 * Value of the constant.
2105 *
2106 * The field used to back the values supplied by the constant is determined
2107 * by the type associated with the \c ir_instruction. Constants may be
2108 * scalars, vectors, or matrices.
2109 */
2110 union ir_constant_data value;
2111
2112 /* Array elements */
2113 ir_constant **array_elements;
2114
2115 /* Structure fields */
2116 exec_list components;
2117
2118 private:
2119 /**
2120 * Parameterless constructor only used by the clone method
2121 */
2122 ir_constant(void);
2123 };
2124
2125 /**
2126 * IR instruction to emit a vertex in a geometry shader.
2127 */
2128 class ir_emit_vertex : public ir_instruction {
2129 public:
2130 ir_emit_vertex(ir_rvalue *stream)
2131 : ir_instruction(ir_type_emit_vertex),
2132 stream(stream)
2133 {
2134 assert(stream);
2135 }
2136
2137 virtual void accept(ir_visitor *v)
2138 {
2139 v->visit(this);
2140 }
2141
2142 virtual ir_emit_vertex *clone(void *mem_ctx, struct hash_table *ht) const
2143 {
2144 return new(mem_ctx) ir_emit_vertex(this->stream->clone(mem_ctx, ht));
2145 }
2146
2147 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2148
2149 int stream_id() const
2150 {
2151 return stream->as_constant()->value.i[0];
2152 }
2153
2154 ir_rvalue *stream;
2155 };
2156
2157 /**
2158 * IR instruction to complete the current primitive and start a new one in a
2159 * geometry shader.
2160 */
2161 class ir_end_primitive : public ir_instruction {
2162 public:
2163 ir_end_primitive(ir_rvalue *stream)
2164 : ir_instruction(ir_type_end_primitive),
2165 stream(stream)
2166 {
2167 assert(stream);
2168 }
2169
2170 virtual void accept(ir_visitor *v)
2171 {
2172 v->visit(this);
2173 }
2174
2175 virtual ir_end_primitive *clone(void *mem_ctx, struct hash_table *ht) const
2176 {
2177 return new(mem_ctx) ir_end_primitive(this->stream->clone(mem_ctx, ht));
2178 }
2179
2180 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2181
2182 int stream_id() const
2183 {
2184 return stream->as_constant()->value.i[0];
2185 }
2186
2187 ir_rvalue *stream;
2188 };
2189
2190 /**
2191 * IR instruction for tessellation control and compute shader barrier.
2192 */
2193 class ir_barrier : public ir_instruction {
2194 public:
2195 ir_barrier()
2196 : ir_instruction(ir_type_barrier)
2197 {
2198 }
2199
2200 virtual void accept(ir_visitor *v)
2201 {
2202 v->visit(this);
2203 }
2204
2205 virtual ir_barrier *clone(void *mem_ctx, struct hash_table *) const
2206 {
2207 return new(mem_ctx) ir_barrier();
2208 }
2209
2210 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2211 };
2212
2213 /*@}*/
2214
2215 /**
2216 * Apply a visitor to each IR node in a list
2217 */
2218 void
2219 visit_exec_list(exec_list *list, ir_visitor *visitor);
2220
2221 /**
2222 * Validate invariants on each IR node in a list
2223 */
2224 void validate_ir_tree(exec_list *instructions);
2225
2226 struct _mesa_glsl_parse_state;
2227 struct gl_shader_program;
2228
2229 /**
2230 * Detect whether an unlinked shader contains static recursion
2231 *
2232 * If the list of instructions is determined to contain static recursion,
2233 * \c _mesa_glsl_error will be called to emit error messages for each function
2234 * that is in the recursion cycle.
2235 */
2236 void
2237 detect_recursion_unlinked(struct _mesa_glsl_parse_state *state,
2238 exec_list *instructions);
2239
2240 /**
2241 * Detect whether a linked shader contains static recursion
2242 *
2243 * If the list of instructions is determined to contain static recursion,
2244 * \c link_error_printf will be called to emit error messages for each function
2245 * that is in the recursion cycle. In addition,
2246 * \c gl_shader_program::LinkStatus will be set to false.
2247 */
2248 void
2249 detect_recursion_linked(struct gl_shader_program *prog,
2250 exec_list *instructions);
2251
2252 /**
2253 * Make a clone of each IR instruction in a list
2254 *
2255 * \param in List of IR instructions that are to be cloned
2256 * \param out List to hold the cloned instructions
2257 */
2258 void
2259 clone_ir_list(void *mem_ctx, exec_list *out, const exec_list *in);
2260
2261 extern void
2262 _mesa_glsl_initialize_variables(exec_list *instructions,
2263 struct _mesa_glsl_parse_state *state);
2264
2265 extern void
2266 _mesa_glsl_initialize_derived_variables(struct gl_context *ctx,
2267 gl_shader *shader);
2268
2269 extern void
2270 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state *state);
2271
2272 extern void
2273 _mesa_glsl_initialize_builtin_functions();
2274
2275 extern ir_function_signature *
2276 _mesa_glsl_find_builtin_function(_mesa_glsl_parse_state *state,
2277 const char *name, exec_list *actual_parameters);
2278
2279 extern ir_function *
2280 _mesa_glsl_find_builtin_function_by_name(const char *name);
2281
2282 extern gl_shader *
2283 _mesa_glsl_get_builtin_function_shader(void);
2284
2285 extern ir_function_signature *
2286 _mesa_get_main_function_signature(glsl_symbol_table *symbols);
2287
2288 extern void
2289 _mesa_glsl_release_functions(void);
2290
2291 extern void
2292 _mesa_glsl_release_builtin_functions(void);
2293
2294 extern void
2295 reparent_ir(exec_list *list, void *mem_ctx);
2296
2297 struct glsl_symbol_table;
2298
2299 extern void
2300 import_prototypes(const exec_list *source, exec_list *dest,
2301 struct glsl_symbol_table *symbols, void *mem_ctx);
2302
2303 extern bool
2304 ir_has_call(ir_instruction *ir);
2305
2306 extern void
2307 do_set_program_inouts(exec_list *instructions, struct gl_program *prog,
2308 gl_shader_stage shader_stage);
2309
2310 extern char *
2311 prototype_string(const glsl_type *return_type, const char *name,
2312 exec_list *parameters);
2313
2314 const char *
2315 mode_string(const ir_variable *var);
2316
2317 /**
2318 * Built-in / reserved GL variables names start with "gl_"
2319 */
2320 static inline bool
2321 is_gl_identifier(const char *s)
2322 {
2323 return s && s[0] == 'g' && s[1] == 'l' && s[2] == '_';
2324 }
2325
2326 extern "C" {
2327 #endif /* __cplusplus */
2328
2329 extern void _mesa_print_ir(FILE *f, struct exec_list *instructions,
2330 struct _mesa_glsl_parse_state *state);
2331
2332 extern void
2333 fprint_ir(FILE *f, const void *instruction);
2334
2335 extern const struct gl_builtin_uniform_desc *
2336 _mesa_glsl_get_builtin_uniform_desc(const char *name);
2337
2338 #ifdef __cplusplus
2339 } /* extern "C" */
2340 #endif
2341
2342 unsigned
2343 vertices_per_prim(GLenum prim);
2344
2345 #endif /* IR_H */