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 #ifndef IR_H
26 #define IR_H
27
28 #include <stdio.h>
29 #include <stdlib.h>
30
31 #include "util/ralloc.h"
32 #include "util/format/u_format.h"
33 #include "util/half_float.h"
34 #include "compiler/glsl_types.h"
35 #include "list.h"
36 #include "ir_visitor.h"
37 #include "ir_hierarchical_visitor.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_demote,
80 ir_type_emit_vertex,
81 ir_type_end_primitive,
82 ir_type_barrier,
83 ir_type_max, /**< maximum ir_type enum number, for validation */
84 ir_type_unset = ir_type_max
85 };
86
87
88 /**
89 * Base class of all IR instructions
90 */
91 class ir_instruction : public exec_node {
92 public:
93 enum ir_node_type ir_type;
94
95 /**
96 * GCC 4.7+ and clang warn when deleting an ir_instruction unless
97 * there's a virtual destructor present. Because we almost
98 * universally use ralloc for our memory management of
99 * ir_instructions, the destructor doesn't need to do any work.
100 */
~ir_instruction()101 virtual ~ir_instruction()
102 {
103 }
104
105 /** ir_print_visitor helper for debugging. */
106 void print(void) const;
107 void fprint(FILE *f) const;
108
109 virtual void accept(ir_visitor *) = 0;
110 virtual ir_visitor_status accept(ir_hierarchical_visitor *) = 0;
111 virtual ir_instruction *clone(void *mem_ctx,
112 struct hash_table *ht) const = 0;
113
is_rvalue()114 bool is_rvalue() const
115 {
116 return ir_type == ir_type_dereference_array ||
117 ir_type == ir_type_dereference_record ||
118 ir_type == ir_type_dereference_variable ||
119 ir_type == ir_type_constant ||
120 ir_type == ir_type_expression ||
121 ir_type == ir_type_swizzle ||
122 ir_type == ir_type_texture;
123 }
124
is_dereference()125 bool is_dereference() const
126 {
127 return ir_type == ir_type_dereference_array ||
128 ir_type == ir_type_dereference_record ||
129 ir_type == ir_type_dereference_variable;
130 }
131
is_jump()132 bool is_jump() const
133 {
134 return ir_type == ir_type_loop_jump ||
135 ir_type == ir_type_return ||
136 ir_type == ir_type_discard;
137 }
138
139 /**
140 * \name IR instruction downcast functions
141 *
142 * These functions either cast the object to a derived class or return
143 * \c NULL if the object's type does not match the specified derived class.
144 * Additional downcast functions will be added as needed.
145 */
146 /*@{*/
147 #define AS_BASE(TYPE) \
148 class ir_##TYPE *as_##TYPE() \
149 { \
150 assume(this != NULL); \
151 return is_##TYPE() ? (ir_##TYPE *) this : NULL; \
152 } \
153 const class ir_##TYPE *as_##TYPE() const \
154 { \
155 assume(this != NULL); \
156 return is_##TYPE() ? (ir_##TYPE *) this : NULL; \
157 }
158
159 AS_BASE(rvalue)
160 AS_BASE(dereference)
161 AS_BASE(jump)
162 #undef AS_BASE
163
164 #define AS_CHILD(TYPE) \
165 class ir_##TYPE * as_##TYPE() \
166 { \
167 assume(this != NULL); \
168 return ir_type == ir_type_##TYPE ? (ir_##TYPE *) this : NULL; \
169 } \
170 const class ir_##TYPE * as_##TYPE() const \
171 { \
172 assume(this != NULL); \
173 return ir_type == ir_type_##TYPE ? (const ir_##TYPE *) this : NULL; \
174 }
175 AS_CHILD(variable)
176 AS_CHILD(function)
177 AS_CHILD(dereference_array)
178 AS_CHILD(dereference_variable)
179 AS_CHILD(dereference_record)
180 AS_CHILD(expression)
181 AS_CHILD(loop)
182 AS_CHILD(assignment)
183 AS_CHILD(call)
184 AS_CHILD(return)
185 AS_CHILD(if)
186 AS_CHILD(swizzle)
187 AS_CHILD(texture)
188 AS_CHILD(constant)
189 AS_CHILD(discard)
190 #undef AS_CHILD
191 /*@}*/
192
193 /**
194 * IR equality method: Return true if the referenced instruction would
195 * return the same value as this one.
196 *
197 * This intended to be used for CSE and algebraic optimizations, on rvalues
198 * in particular. No support for other instruction types (assignments,
199 * jumps, calls, etc.) is planned.
200 */
201 virtual bool equals(const ir_instruction *ir,
202 enum ir_node_type ignore = ir_type_unset) const;
203
204 protected:
ir_instruction(enum ir_node_type t)205 ir_instruction(enum ir_node_type t)
206 : ir_type(t)
207 {
208 }
209
210 private:
ir_instruction()211 ir_instruction()
212 {
213 assert(!"Should not get here.");
214 }
215 };
216
217
218 /**
219 * The base class for all "values"/expression trees.
220 */
221 class ir_rvalue : public ir_instruction {
222 public:
223 const struct glsl_type *type;
224
225 virtual ir_rvalue *clone(void *mem_ctx, struct hash_table *) const;
226
accept(ir_visitor * v)227 virtual void accept(ir_visitor *v)
228 {
229 v->visit(this);
230 }
231
232 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
233
234 virtual ir_constant *constant_expression_value(void *mem_ctx,
235 struct hash_table *variable_context = NULL);
236
237 ir_rvalue *as_rvalue_to_saturate();
238
239 virtual bool is_lvalue(const struct _mesa_glsl_parse_state * = NULL) const
240 {
241 return false;
242 }
243
244 /**
245 * Get the variable that is ultimately referenced by an r-value
246 */
variable_referenced()247 virtual ir_variable *variable_referenced() const
248 {
249 return NULL;
250 }
251
252
253 /**
254 * If an r-value is a reference to a whole variable, get that variable
255 *
256 * \return
257 * Pointer to a variable that is completely dereferenced by the r-value. If
258 * the r-value is not a dereference or the dereference does not access the
259 * entire variable (i.e., it's just one array element, struct field), \c NULL
260 * is returned.
261 */
whole_variable_referenced()262 virtual ir_variable *whole_variable_referenced()
263 {
264 return NULL;
265 }
266
267 /**
268 * Determine if an r-value has the value zero
269 *
270 * The base implementation of this function always returns \c false. The
271 * \c ir_constant class over-rides this function to return \c true \b only
272 * for vector and scalar types that have all elements set to the value
273 * zero (or \c false for booleans).
274 *
275 * \sa ir_constant::has_value, ir_rvalue::is_one, ir_rvalue::is_negative_one
276 */
277 virtual bool is_zero() const;
278
279 /**
280 * Determine if an r-value has the value one
281 *
282 * The base implementation of this function always returns \c false. The
283 * \c ir_constant class over-rides this function to return \c true \b only
284 * for vector and scalar types that have all elements set to the value
285 * one (or \c true for booleans).
286 *
287 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_negative_one
288 */
289 virtual bool is_one() const;
290
291 /**
292 * Determine if an r-value has the value negative one
293 *
294 * The base implementation of this function always returns \c false. The
295 * \c ir_constant class over-rides this function to return \c true \b only
296 * for vector and scalar types that have all elements set to the value
297 * negative one. For boolean types, the result is always \c false.
298 *
299 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one
300 */
301 virtual bool is_negative_one() const;
302
303 /**
304 * Determine if an r-value is an unsigned integer constant which can be
305 * stored in 16 bits.
306 *
307 * \sa ir_constant::is_uint16_constant.
308 */
is_uint16_constant()309 virtual bool is_uint16_constant() const { return false; }
310
311 /**
312 * Return a generic value of error_type.
313 *
314 * Allocation will be performed with 'mem_ctx' as ralloc owner.
315 */
316 static ir_rvalue *error_value(void *mem_ctx);
317
318 protected:
319 ir_rvalue(enum ir_node_type t);
320 };
321
322
323 /**
324 * Variable storage classes
325 */
326 enum ir_variable_mode {
327 ir_var_auto = 0, /**< Function local variables and globals. */
328 ir_var_uniform, /**< Variable declared as a uniform. */
329 ir_var_shader_storage, /**< Variable declared as an ssbo. */
330 ir_var_shader_shared, /**< Variable declared as shared. */
331 ir_var_shader_in,
332 ir_var_shader_out,
333 ir_var_function_in,
334 ir_var_function_out,
335 ir_var_function_inout,
336 ir_var_const_in, /**< "in" param that must be a constant expression */
337 ir_var_system_value, /**< Ex: front-face, instance-id, etc. */
338 ir_var_temporary, /**< Temporary variable generated during compilation. */
339 ir_var_mode_count /**< Number of variable modes */
340 };
341
342 /**
343 * Enum keeping track of how a variable was declared. For error checking of
344 * the gl_PerVertex redeclaration rules.
345 */
346 enum ir_var_declaration_type {
347 /**
348 * Normal declaration (for most variables, this means an explicit
349 * declaration. Exception: temporaries are always implicitly declared, but
350 * they still use ir_var_declared_normally).
351 *
352 * Note: an ir_variable that represents a named interface block uses
353 * ir_var_declared_normally.
354 */
355 ir_var_declared_normally = 0,
356
357 /**
358 * Variable was explicitly declared (or re-declared) in an unnamed
359 * interface block.
360 */
361 ir_var_declared_in_block,
362
363 /**
364 * Variable is an implicitly declared built-in that has not been explicitly
365 * re-declared by the shader.
366 */
367 ir_var_declared_implicitly,
368
369 /**
370 * Variable is implicitly generated by the compiler and should not be
371 * visible via the API.
372 */
373 ir_var_hidden,
374 };
375
376 /**
377 * \brief Layout qualifiers for gl_FragDepth.
378 *
379 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
380 * with a layout qualifier.
381 */
382 enum ir_depth_layout {
383 ir_depth_layout_none, /**< No depth layout is specified. */
384 ir_depth_layout_any,
385 ir_depth_layout_greater,
386 ir_depth_layout_less,
387 ir_depth_layout_unchanged
388 };
389
390 /**
391 * \brief Convert depth layout qualifier to string.
392 */
393 const char*
394 depth_layout_string(ir_depth_layout layout);
395
396 /**
397 * Description of built-in state associated with a uniform
398 *
399 * \sa ir_variable::state_slots
400 */
401 struct ir_state_slot {
402 gl_state_index16 tokens[STATE_LENGTH];
403 int swizzle;
404 };
405
406
407 /**
408 * Get the string value for an interpolation qualifier
409 *
410 * \return The string that would be used in a shader to specify \c
411 * mode will be returned.
412 *
413 * This function is used to generate error messages of the form "shader
414 * uses %s interpolation qualifier", so in the case where there is no
415 * interpolation qualifier, it returns "no".
416 *
417 * This function should only be used on a shader input or output variable.
418 */
419 const char *interpolation_string(unsigned interpolation);
420
421
422 class ir_variable : public ir_instruction {
423 public:
424 ir_variable(const struct glsl_type *, const char *, ir_variable_mode);
425
426 virtual ir_variable *clone(void *mem_ctx, struct hash_table *ht) const;
427
accept(ir_visitor * v)428 virtual void accept(ir_visitor *v)
429 {
430 v->visit(this);
431 }
432
433 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
434
435
436 /**
437 * Determine whether or not a variable is part of a uniform or
438 * shader storage block.
439 */
is_in_buffer_block()440 inline bool is_in_buffer_block() const
441 {
442 return (this->data.mode == ir_var_uniform ||
443 this->data.mode == ir_var_shader_storage) &&
444 this->interface_type != NULL;
445 }
446
447 /**
448 * Determine whether or not a variable is part of a shader storage block.
449 */
is_in_shader_storage_block()450 inline bool is_in_shader_storage_block() const
451 {
452 return this->data.mode == ir_var_shader_storage &&
453 this->interface_type != NULL;
454 }
455
456 /**
457 * Determine whether or not a variable is the declaration of an interface
458 * block
459 *
460 * For the first declaration below, there will be an \c ir_variable named
461 * "instance" whose type and whose instance_type will be the same
462 * \c glsl_type. For the second declaration, there will be an \c ir_variable
463 * named "f" whose type is float and whose instance_type is B2.
464 *
465 * "instance" is an interface instance variable, but "f" is not.
466 *
467 * uniform B1 {
468 * float f;
469 * } instance;
470 *
471 * uniform B2 {
472 * float f;
473 * };
474 */
is_interface_instance()475 inline bool is_interface_instance() const
476 {
477 return this->type->without_array() == this->interface_type;
478 }
479
480 /**
481 * Return whether this variable contains a bindless sampler/image.
482 */
contains_bindless()483 inline bool contains_bindless() const
484 {
485 if (!this->type->contains_sampler() && !this->type->contains_image())
486 return false;
487
488 return this->data.bindless || this->data.mode != ir_var_uniform;
489 }
490
491 /**
492 * Set this->interface_type on a newly created variable.
493 */
init_interface_type(const struct glsl_type * type)494 void init_interface_type(const struct glsl_type *type)
495 {
496 assert(this->interface_type == NULL);
497 this->interface_type = type;
498 if (this->is_interface_instance()) {
499 this->u.max_ifc_array_access =
500 ralloc_array(this, int, type->length);
501 for (unsigned i = 0; i < type->length; i++) {
502 this->u.max_ifc_array_access[i] = -1;
503 }
504 }
505 }
506
507 /**
508 * Change this->interface_type on a variable that previously had a
509 * different, but compatible, interface_type. This is used during linking
510 * to set the size of arrays in interface blocks.
511 */
change_interface_type(const struct glsl_type * type)512 void change_interface_type(const struct glsl_type *type)
513 {
514 if (this->u.max_ifc_array_access != NULL) {
515 /* max_ifc_array_access has already been allocated, so make sure the
516 * new interface has the same number of fields as the old one.
517 */
518 assert(this->interface_type->length == type->length);
519 }
520 this->interface_type = type;
521 }
522
523 /**
524 * Change this->interface_type on a variable that previously had a
525 * different, and incompatible, interface_type. This is used during
526 * compilation to handle redeclaration of the built-in gl_PerVertex
527 * interface block.
528 */
reinit_interface_type(const struct glsl_type * type)529 void reinit_interface_type(const struct glsl_type *type)
530 {
531 if (this->u.max_ifc_array_access != NULL) {
532 #ifndef NDEBUG
533 /* Redeclaring gl_PerVertex is only allowed if none of the built-ins
534 * it defines have been accessed yet; so it's safe to throw away the
535 * old max_ifc_array_access pointer, since all of its values are
536 * zero.
537 */
538 for (unsigned i = 0; i < this->interface_type->length; i++)
539 assert(this->u.max_ifc_array_access[i] == -1);
540 #endif
541 ralloc_free(this->u.max_ifc_array_access);
542 this->u.max_ifc_array_access = NULL;
543 }
544 this->interface_type = NULL;
545 init_interface_type(type);
546 }
547
get_interface_type()548 const glsl_type *get_interface_type() const
549 {
550 return this->interface_type;
551 }
552
get_interface_type_packing()553 enum glsl_interface_packing get_interface_type_packing() const
554 {
555 return this->interface_type->get_interface_packing();
556 }
557 /**
558 * Get the max_ifc_array_access pointer
559 *
560 * A "set" function is not needed because the array is dynamically allocated
561 * as necessary.
562 */
get_max_ifc_array_access()563 inline int *get_max_ifc_array_access()
564 {
565 assert(this->data._num_state_slots == 0);
566 return this->u.max_ifc_array_access;
567 }
568
get_num_state_slots()569 inline unsigned get_num_state_slots() const
570 {
571 assert(!this->is_interface_instance()
572 || this->data._num_state_slots == 0);
573 return this->data._num_state_slots;
574 }
575
set_num_state_slots(unsigned n)576 inline void set_num_state_slots(unsigned n)
577 {
578 assert(!this->is_interface_instance()
579 || n == 0);
580 this->data._num_state_slots = n;
581 }
582
get_state_slots()583 inline ir_state_slot *get_state_slots()
584 {
585 return this->is_interface_instance() ? NULL : this->u.state_slots;
586 }
587
get_state_slots()588 inline const ir_state_slot *get_state_slots() const
589 {
590 return this->is_interface_instance() ? NULL : this->u.state_slots;
591 }
592
allocate_state_slots(unsigned n)593 inline ir_state_slot *allocate_state_slots(unsigned n)
594 {
595 assert(!this->is_interface_instance());
596
597 this->u.state_slots = ralloc_array(this, ir_state_slot, n);
598 this->data._num_state_slots = 0;
599
600 if (this->u.state_slots != NULL)
601 this->data._num_state_slots = n;
602
603 return this->u.state_slots;
604 }
605
is_interpolation_flat()606 inline bool is_interpolation_flat() const
607 {
608 return this->data.interpolation == INTERP_MODE_FLAT ||
609 this->type->contains_integer() ||
610 this->type->contains_double();
611 }
612
is_name_ralloced()613 inline bool is_name_ralloced() const
614 {
615 return this->name != ir_variable::tmp_name &&
616 this->name != this->name_storage;
617 }
618
619 /**
620 * Enable emitting extension warnings for this variable
621 */
622 void enable_extension_warning(const char *extension);
623
624 /**
625 * Get the extension warning string for this variable
626 *
627 * If warnings are not enabled, \c NULL is returned.
628 */
629 const char *get_extension_warning() const;
630
631 /**
632 * Declared type of the variable
633 */
634 const struct glsl_type *type;
635
636 /**
637 * Declared name of the variable
638 */
639 const char *name;
640
641 private:
642 /**
643 * If the name length fits into name_storage, it's used, otherwise
644 * the name is ralloc'd. shader-db mining showed that 70% of variables
645 * fit here. This is a win over ralloc where only ralloc_header has
646 * 20 bytes on 64-bit (28 bytes with DEBUG), and we can also skip malloc.
647 */
648 char name_storage[16];
649
650 public:
651 struct ir_variable_data {
652
653 /**
654 * Is the variable read-only?
655 *
656 * This is set for variables declared as \c const, shader inputs,
657 * and uniforms.
658 */
659 unsigned read_only:1;
660 unsigned centroid:1;
661 unsigned sample:1;
662 unsigned patch:1;
663 /**
664 * Was an 'invariant' qualifier explicitly set in the shader?
665 *
666 * This is used to cross validate qualifiers.
667 */
668 unsigned explicit_invariant:1;
669 /**
670 * Is the variable invariant?
671 *
672 * It can happen either by having the 'invariant' qualifier
673 * explicitly set in the shader or by being used in calculations
674 * of other invariant variables.
675 */
676 unsigned invariant:1;
677 unsigned precise:1;
678
679 /**
680 * Has this variable been used for reading or writing?
681 *
682 * Several GLSL semantic checks require knowledge of whether or not a
683 * variable has been used. For example, it is an error to redeclare a
684 * variable as invariant after it has been used.
685 *
686 * This is maintained in the ast_to_hir.cpp path and during linking,
687 * but not in Mesa's fixed function or ARB program paths.
688 */
689 unsigned used:1;
690
691 /**
692 * Has this variable been statically assigned?
693 *
694 * This answers whether the variable was assigned in any path of
695 * the shader during ast_to_hir. This doesn't answer whether it is
696 * still written after dead code removal, nor is it maintained in
697 * non-ast_to_hir.cpp (GLSL parsing) paths.
698 */
699 unsigned assigned:1;
700
701 /**
702 * When separate shader programs are enabled, only input/outputs between
703 * the stages of a multi-stage separate program can be safely removed
704 * from the shader interface. Other input/outputs must remains active.
705 */
706 unsigned always_active_io:1;
707
708 /**
709 * Enum indicating how the variable was declared. See
710 * ir_var_declaration_type.
711 *
712 * This is used to detect certain kinds of illegal variable redeclarations.
713 */
714 unsigned how_declared:2;
715
716 /**
717 * Storage class of the variable.
718 *
719 * \sa ir_variable_mode
720 */
721 unsigned mode:4;
722
723 /**
724 * Interpolation mode for shader inputs / outputs
725 *
726 * \sa glsl_interp_mode
727 */
728 unsigned interpolation:2;
729
730 /**
731 * Was the location explicitly set in the shader?
732 *
733 * If the location is explicitly set in the shader, it \b cannot be changed
734 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
735 * no effect).
736 */
737 unsigned explicit_location:1;
738 unsigned explicit_index:1;
739
740 /**
741 * Was an initial binding explicitly set in the shader?
742 *
743 * If so, constant_value contains an integer ir_constant representing the
744 * initial binding point.
745 */
746 unsigned explicit_binding:1;
747
748 /**
749 * Was an initial component explicitly set in the shader?
750 */
751 unsigned explicit_component:1;
752
753 /**
754 * Does this variable have an initializer?
755 *
756 * This is used by the linker to cross-validiate initializers of global
757 * variables.
758 */
759 unsigned has_initializer:1;
760
761 /**
762 * Is the initializer created by the compiler (glsl_zero_init)
763 */
764 unsigned is_implicit_initializer:1;
765
766 /**
767 * Is this variable a generic output or input that has not yet been matched
768 * up to a variable in another stage of the pipeline?
769 *
770 * This is used by the linker as scratch storage while assigning locations
771 * to generic inputs and outputs.
772 */
773 unsigned is_unmatched_generic_inout:1;
774
775 /**
776 * Is this varying used by transform feedback?
777 *
778 * This is used by the linker to decide if it's safe to pack the varying.
779 */
780 unsigned is_xfb:1;
781
782 /**
783 * Is this varying used only by transform feedback?
784 *
785 * This is used by the linker to decide if its safe to pack the varying.
786 */
787 unsigned is_xfb_only:1;
788
789 /**
790 * Was a transform feedback buffer set in the shader?
791 */
792 unsigned explicit_xfb_buffer:1;
793
794 /**
795 * Was a transform feedback offset set in the shader?
796 */
797 unsigned explicit_xfb_offset:1;
798
799 /**
800 * Was a transform feedback stride set in the shader?
801 */
802 unsigned explicit_xfb_stride:1;
803
804 /**
805 * If non-zero, then this variable may be packed along with other variables
806 * into a single varying slot, so this offset should be applied when
807 * accessing components. For example, an offset of 1 means that the x
808 * component of this variable is actually stored in component y of the
809 * location specified by \c location.
810 */
811 unsigned location_frac:2;
812
813 /**
814 * Layout of the matrix. Uses glsl_matrix_layout values.
815 */
816 unsigned matrix_layout:2;
817
818 /**
819 * Non-zero if this variable was created by lowering a named interface
820 * block.
821 */
822 unsigned from_named_ifc_block:1;
823
824 /**
825 * Non-zero if the variable must be a shader input. This is useful for
826 * constraints on function parameters.
827 */
828 unsigned must_be_shader_input:1;
829
830 /**
831 * Output index for dual source blending.
832 *
833 * \note
834 * The GLSL spec only allows the values 0 or 1 for the index in \b dual
835 * source blending.
836 */
837 unsigned index:1;
838
839 /**
840 * Precision qualifier.
841 *
842 * In desktop GLSL we do not care about precision qualifiers at all, in
843 * fact, the spec says that precision qualifiers are ignored.
844 *
845 * To make things easy, we make it so that this field is always
846 * GLSL_PRECISION_NONE on desktop shaders. This way all the variables
847 * have the same precision value and the checks we add in the compiler
848 * for this field will never break a desktop shader compile.
849 */
850 unsigned precision:2;
851
852 /**
853 * \brief Layout qualifier for gl_FragDepth.
854 *
855 * This is not equal to \c ir_depth_layout_none if and only if this
856 * variable is \c gl_FragDepth and a layout qualifier is specified.
857 */
858 ir_depth_layout depth_layout:3;
859
860 /**
861 * Memory qualifiers.
862 */
863 unsigned memory_read_only:1; /**< "readonly" qualifier. */
864 unsigned memory_write_only:1; /**< "writeonly" qualifier. */
865 unsigned memory_coherent:1;
866 unsigned memory_volatile:1;
867 unsigned memory_restrict:1;
868
869 /**
870 * ARB_shader_storage_buffer_object
871 */
872 unsigned from_ssbo_unsized_array:1; /**< unsized array buffer variable. */
873
874 unsigned implicit_sized_array:1;
875
876 /**
877 * Whether this is a fragment shader output implicitly initialized with
878 * the previous contents of the specified render target at the
879 * framebuffer location corresponding to this shader invocation.
880 */
881 unsigned fb_fetch_output:1;
882
883 /**
884 * Non-zero if this variable is considered bindless as defined by
885 * ARB_bindless_texture.
886 */
887 unsigned bindless:1;
888
889 /**
890 * Non-zero if this variable is considered bound as defined by
891 * ARB_bindless_texture.
892 */
893 unsigned bound:1;
894
895 /**
896 * Non-zero if the variable shall not be implicitly converted during
897 * functions matching.
898 */
899 unsigned implicit_conversion_prohibited:1;
900
901 /**
902 * Emit a warning if this variable is accessed.
903 */
904 private:
905 uint8_t warn_extension_index;
906
907 public:
908 /**
909 * Image internal format if specified explicitly, otherwise
910 * PIPE_FORMAT_NONE.
911 */
912 enum pipe_format image_format;
913
914 private:
915 /**
916 * Number of state slots used
917 *
918 * \note
919 * This could be stored in as few as 7-bits, if necessary. If it is made
920 * smaller, add an assertion to \c ir_variable::allocate_state_slots to
921 * be safe.
922 */
923 uint16_t _num_state_slots;
924
925 public:
926 /**
927 * Initial binding point for a sampler, atomic, or UBO.
928 *
929 * For array types, this represents the binding point for the first element.
930 */
931 uint16_t binding;
932
933 /**
934 * Storage location of the base of this variable
935 *
936 * The precise meaning of this field depends on the nature of the variable.
937 *
938 * - Vertex shader input: one of the values from \c gl_vert_attrib.
939 * - Vertex shader output: one of the values from \c gl_varying_slot.
940 * - Geometry shader input: one of the values from \c gl_varying_slot.
941 * - Geometry shader output: one of the values from \c gl_varying_slot.
942 * - Fragment shader input: one of the values from \c gl_varying_slot.
943 * - Fragment shader output: one of the values from \c gl_frag_result.
944 * - Uniforms: Per-stage uniform slot number for default uniform block.
945 * - Uniforms: Index within the uniform block definition for UBO members.
946 * - Non-UBO Uniforms: explicit location until linking then reused to
947 * store uniform slot number.
948 * - Other: This field is not currently used.
949 *
950 * If the variable is a uniform, shader input, or shader output, and the
951 * slot has not been assigned, the value will be -1.
952 */
953 int location;
954
955 /**
956 * for glsl->tgsi/mesa IR we need to store the index into the
957 * parameters for uniforms, initially the code overloaded location
958 * but this causes problems with indirect samplers and AoA.
959 * This is assigned in _mesa_generate_parameters_list_for_uniforms.
960 */
961 int param_index;
962
963 /**
964 * Vertex stream output identifier.
965 *
966 * For packed outputs, bit 31 is set and bits [2*i+1,2*i] indicate the
967 * stream of the i-th component.
968 */
969 unsigned stream;
970
971 /**
972 * Atomic, transform feedback or block member offset.
973 */
974 unsigned offset;
975
976 /**
977 * Highest element accessed with a constant expression array index
978 *
979 * Not used for non-array variables. -1 is never accessed.
980 */
981 int max_array_access;
982
983 /**
984 * Transform feedback buffer.
985 */
986 unsigned xfb_buffer;
987
988 /**
989 * Transform feedback stride.
990 */
991 unsigned xfb_stride;
992
993 /**
994 * Allow (only) ir_variable direct access private members.
995 */
996 friend class ir_variable;
997 } data;
998
999 /**
1000 * Value assigned in the initializer of a variable declared "const"
1001 */
1002 ir_constant *constant_value;
1003
1004 /**
1005 * Constant expression assigned in the initializer of the variable
1006 *
1007 * \warning
1008 * This field and \c ::constant_value are distinct. Even if the two fields
1009 * refer to constants with the same value, they must point to separate
1010 * objects.
1011 */
1012 ir_constant *constant_initializer;
1013
1014 private:
1015 static const char *const warn_extension_table[];
1016
1017 union {
1018 /**
1019 * For variables which satisfy the is_interface_instance() predicate,
1020 * this points to an array of integers such that if the ith member of
1021 * the interface block is an array, max_ifc_array_access[i] is the
1022 * maximum array element of that member that has been accessed. If the
1023 * ith member of the interface block is not an array,
1024 * max_ifc_array_access[i] is unused.
1025 *
1026 * For variables whose type is not an interface block, this pointer is
1027 * NULL.
1028 */
1029 int *max_ifc_array_access;
1030
1031 /**
1032 * Built-in state that backs this uniform
1033 *
1034 * Once set at variable creation, \c state_slots must remain invariant.
1035 *
1036 * If the variable is not a uniform, \c _num_state_slots will be zero
1037 * and \c state_slots will be \c NULL.
1038 */
1039 ir_state_slot *state_slots;
1040 } u;
1041
1042 /**
1043 * For variables that are in an interface block or are an instance of an
1044 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
1045 *
1046 * \sa ir_variable::location
1047 */
1048 const glsl_type *interface_type;
1049
1050 /**
1051 * Name used for anonymous compiler temporaries
1052 */
1053 static const char tmp_name[];
1054
1055 public:
1056 /**
1057 * Should the construct keep names for ir_var_temporary variables?
1058 *
1059 * When this global is false, names passed to the constructor for
1060 * \c ir_var_temporary variables will be dropped. Instead, the variable will
1061 * be named "compiler_temp". This name will be in static storage.
1062 *
1063 * \warning
1064 * \b NEVER change the mode of an \c ir_var_temporary.
1065 *
1066 * \warning
1067 * This variable is \b not thread-safe. It is global, \b not
1068 * per-context. It begins life false. A context can, at some point, make
1069 * it true. From that point on, it will be true forever. This should be
1070 * okay since it will only be set true while debugging.
1071 */
1072 static bool temporaries_allocate_names;
1073 };
1074
1075 /**
1076 * A function that returns whether a built-in function is available in the
1077 * current shading language (based on version, ES or desktop, and extensions).
1078 */
1079 typedef bool (*builtin_available_predicate)(const _mesa_glsl_parse_state *);
1080
1081 #define MAKE_INTRINSIC_FOR_TYPE(op, t) \
1082 ir_intrinsic_generic_ ## op - ir_intrinsic_generic_load + ir_intrinsic_ ## t ## _ ## load
1083
1084 #define MAP_INTRINSIC_TO_TYPE(i, t) \
1085 ir_intrinsic_id(int(i) - int(ir_intrinsic_generic_load) + int(ir_intrinsic_ ## t ## _ ## load))
1086
1087 enum ir_intrinsic_id {
1088 ir_intrinsic_invalid = 0,
1089
1090 /**
1091 * \name Generic intrinsics
1092 *
1093 * Each of these intrinsics has a specific version for shared variables and
1094 * SSBOs.
1095 */
1096 /*@{*/
1097 ir_intrinsic_generic_load,
1098 ir_intrinsic_generic_store,
1099 ir_intrinsic_generic_atomic_add,
1100 ir_intrinsic_generic_atomic_and,
1101 ir_intrinsic_generic_atomic_or,
1102 ir_intrinsic_generic_atomic_xor,
1103 ir_intrinsic_generic_atomic_min,
1104 ir_intrinsic_generic_atomic_max,
1105 ir_intrinsic_generic_atomic_exchange,
1106 ir_intrinsic_generic_atomic_comp_swap,
1107 /*@}*/
1108
1109 ir_intrinsic_atomic_counter_read,
1110 ir_intrinsic_atomic_counter_increment,
1111 ir_intrinsic_atomic_counter_predecrement,
1112 ir_intrinsic_atomic_counter_add,
1113 ir_intrinsic_atomic_counter_and,
1114 ir_intrinsic_atomic_counter_or,
1115 ir_intrinsic_atomic_counter_xor,
1116 ir_intrinsic_atomic_counter_min,
1117 ir_intrinsic_atomic_counter_max,
1118 ir_intrinsic_atomic_counter_exchange,
1119 ir_intrinsic_atomic_counter_comp_swap,
1120
1121 ir_intrinsic_image_load,
1122 ir_intrinsic_image_store,
1123 ir_intrinsic_image_atomic_add,
1124 ir_intrinsic_image_atomic_and,
1125 ir_intrinsic_image_atomic_or,
1126 ir_intrinsic_image_atomic_xor,
1127 ir_intrinsic_image_atomic_min,
1128 ir_intrinsic_image_atomic_max,
1129 ir_intrinsic_image_atomic_exchange,
1130 ir_intrinsic_image_atomic_comp_swap,
1131 ir_intrinsic_image_size,
1132 ir_intrinsic_image_samples,
1133 ir_intrinsic_image_atomic_inc_wrap,
1134 ir_intrinsic_image_atomic_dec_wrap,
1135
1136 ir_intrinsic_ssbo_load,
1137 ir_intrinsic_ssbo_store = MAKE_INTRINSIC_FOR_TYPE(store, ssbo),
1138 ir_intrinsic_ssbo_atomic_add = MAKE_INTRINSIC_FOR_TYPE(atomic_add, ssbo),
1139 ir_intrinsic_ssbo_atomic_and = MAKE_INTRINSIC_FOR_TYPE(atomic_and, ssbo),
1140 ir_intrinsic_ssbo_atomic_or = MAKE_INTRINSIC_FOR_TYPE(atomic_or, ssbo),
1141 ir_intrinsic_ssbo_atomic_xor = MAKE_INTRINSIC_FOR_TYPE(atomic_xor, ssbo),
1142 ir_intrinsic_ssbo_atomic_min = MAKE_INTRINSIC_FOR_TYPE(atomic_min, ssbo),
1143 ir_intrinsic_ssbo_atomic_max = MAKE_INTRINSIC_FOR_TYPE(atomic_max, ssbo),
1144 ir_intrinsic_ssbo_atomic_exchange = MAKE_INTRINSIC_FOR_TYPE(atomic_exchange, ssbo),
1145 ir_intrinsic_ssbo_atomic_comp_swap = MAKE_INTRINSIC_FOR_TYPE(atomic_comp_swap, ssbo),
1146
1147 ir_intrinsic_memory_barrier,
1148 ir_intrinsic_shader_clock,
1149 ir_intrinsic_group_memory_barrier,
1150 ir_intrinsic_memory_barrier_atomic_counter,
1151 ir_intrinsic_memory_barrier_buffer,
1152 ir_intrinsic_memory_barrier_image,
1153 ir_intrinsic_memory_barrier_shared,
1154 ir_intrinsic_begin_invocation_interlock,
1155 ir_intrinsic_end_invocation_interlock,
1156
1157 ir_intrinsic_vote_all,
1158 ir_intrinsic_vote_any,
1159 ir_intrinsic_vote_eq,
1160 ir_intrinsic_ballot,
1161 ir_intrinsic_read_invocation,
1162 ir_intrinsic_read_first_invocation,
1163
1164 ir_intrinsic_helper_invocation,
1165
1166 ir_intrinsic_shared_load,
1167 ir_intrinsic_shared_store = MAKE_INTRINSIC_FOR_TYPE(store, shared),
1168 ir_intrinsic_shared_atomic_add = MAKE_INTRINSIC_FOR_TYPE(atomic_add, shared),
1169 ir_intrinsic_shared_atomic_and = MAKE_INTRINSIC_FOR_TYPE(atomic_and, shared),
1170 ir_intrinsic_shared_atomic_or = MAKE_INTRINSIC_FOR_TYPE(atomic_or, shared),
1171 ir_intrinsic_shared_atomic_xor = MAKE_INTRINSIC_FOR_TYPE(atomic_xor, shared),
1172 ir_intrinsic_shared_atomic_min = MAKE_INTRINSIC_FOR_TYPE(atomic_min, shared),
1173 ir_intrinsic_shared_atomic_max = MAKE_INTRINSIC_FOR_TYPE(atomic_max, shared),
1174 ir_intrinsic_shared_atomic_exchange = MAKE_INTRINSIC_FOR_TYPE(atomic_exchange, shared),
1175 ir_intrinsic_shared_atomic_comp_swap = MAKE_INTRINSIC_FOR_TYPE(atomic_comp_swap, shared),
1176 };
1177
1178 /*@{*/
1179 /**
1180 * The representation of a function instance; may be the full definition or
1181 * simply a prototype.
1182 */
1183 class ir_function_signature : public ir_instruction {
1184 /* An ir_function_signature will be part of the list of signatures in
1185 * an ir_function.
1186 */
1187 public:
1188 ir_function_signature(const glsl_type *return_type,
1189 builtin_available_predicate builtin_avail = NULL);
1190
1191 virtual ir_function_signature *clone(void *mem_ctx,
1192 struct hash_table *ht) const;
1193 ir_function_signature *clone_prototype(void *mem_ctx,
1194 struct hash_table *ht) const;
1195
accept(ir_visitor * v)1196 virtual void accept(ir_visitor *v)
1197 {
1198 v->visit(this);
1199 }
1200
1201 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1202
1203 /**
1204 * Attempt to evaluate this function as a constant expression,
1205 * given a list of the actual parameters and the variable context.
1206 * Returns NULL for non-built-ins.
1207 */
1208 ir_constant *constant_expression_value(void *mem_ctx,
1209 exec_list *actual_parameters,
1210 struct hash_table *variable_context);
1211
1212 /**
1213 * Get the name of the function for which this is a signature
1214 */
1215 const char *function_name() const;
1216
1217 /**
1218 * Get a handle to the function for which this is a signature
1219 *
1220 * There is no setter function, this function returns a \c const pointer,
1221 * and \c ir_function_signature::_function is private for a reason. The
1222 * only way to make a connection between a function and function signature
1223 * is via \c ir_function::add_signature. This helps ensure that certain
1224 * invariants (i.e., a function signature is in the list of signatures for
1225 * its \c _function) are met.
1226 *
1227 * \sa ir_function::add_signature
1228 */
function()1229 inline const class ir_function *function() const
1230 {
1231 return this->_function;
1232 }
1233
1234 /**
1235 * Check whether the qualifiers match between this signature's parameters
1236 * and the supplied parameter list. If not, returns the name of the first
1237 * parameter with mismatched qualifiers (for use in error messages).
1238 */
1239 const char *qualifiers_match(exec_list *params);
1240
1241 /**
1242 * Replace the current parameter list with the given one. This is useful
1243 * if the current information came from a prototype, and either has invalid
1244 * or missing parameter names.
1245 */
1246 void replace_parameters(exec_list *new_params);
1247
1248 /**
1249 * Function return type.
1250 *
1251 * \note The precision qualifier is stored separately in return_precision.
1252 */
1253 const struct glsl_type *return_type;
1254
1255 /**
1256 * List of ir_variable of function parameters.
1257 *
1258 * This represents the storage. The paramaters passed in a particular
1259 * call will be in ir_call::actual_paramaters.
1260 */
1261 struct exec_list parameters;
1262
1263 /** Whether or not this function has a body (which may be empty). */
1264 unsigned is_defined:1;
1265
1266 /*
1267 * Precision qualifier for the return type.
1268 *
1269 * See the comment for ir_variable_data::precision for more details.
1270 */
1271 unsigned return_precision:2;
1272
1273 /** Whether or not this function signature is a built-in. */
1274 bool is_builtin() const;
1275
1276 /**
1277 * Whether or not this function is an intrinsic to be implemented
1278 * by the driver.
1279 */
is_intrinsic()1280 inline bool is_intrinsic() const
1281 {
1282 return intrinsic_id != ir_intrinsic_invalid;
1283 }
1284
1285 /** Identifier for this intrinsic. */
1286 enum ir_intrinsic_id intrinsic_id;
1287
1288 /** Whether or not a built-in is available for this shader. */
1289 bool is_builtin_available(const _mesa_glsl_parse_state *state) const;
1290
1291 /** Body of instructions in the function. */
1292 struct exec_list body;
1293
1294 private:
1295 /**
1296 * A function pointer to a predicate that answers whether a built-in
1297 * function is available in the current shader. NULL if not a built-in.
1298 */
1299 builtin_available_predicate builtin_avail;
1300
1301 /** Function of which this signature is one overload. */
1302 class ir_function *_function;
1303
1304 /** Function signature of which this one is a prototype clone */
1305 const ir_function_signature *origin;
1306
1307 friend class ir_function;
1308
1309 /**
1310 * Helper function to run a list of instructions for constant
1311 * expression evaluation.
1312 *
1313 * The hash table represents the values of the visible variables.
1314 * There are no scoping issues because the table is indexed on
1315 * ir_variable pointers, not variable names.
1316 *
1317 * Returns false if the expression is not constant, true otherwise,
1318 * and the value in *result if result is non-NULL.
1319 */
1320 bool constant_expression_evaluate_expression_list(void *mem_ctx,
1321 const struct exec_list &body,
1322 struct hash_table *variable_context,
1323 ir_constant **result);
1324 };
1325
1326
1327 /**
1328 * Header for tracking multiple overloaded functions with the same name.
1329 * Contains a list of ir_function_signatures representing each of the
1330 * actual functions.
1331 */
1332 class ir_function : public ir_instruction {
1333 public:
1334 ir_function(const char *name);
1335
1336 virtual ir_function *clone(void *mem_ctx, struct hash_table *ht) const;
1337
accept(ir_visitor * v)1338 virtual void accept(ir_visitor *v)
1339 {
1340 v->visit(this);
1341 }
1342
1343 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1344
add_signature(ir_function_signature * sig)1345 void add_signature(ir_function_signature *sig)
1346 {
1347 sig->_function = this;
1348 this->signatures.push_tail(sig);
1349 }
1350
1351 /**
1352 * Find a signature that matches a set of actual parameters, taking implicit
1353 * conversions into account. Also flags whether the match was exact.
1354 */
1355 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
1356 const exec_list *actual_param,
1357 bool allow_builtins,
1358 bool *match_is_exact);
1359
1360 /**
1361 * Find a signature that matches a set of actual parameters, taking implicit
1362 * conversions into account.
1363 */
1364 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
1365 const exec_list *actual_param,
1366 bool allow_builtins);
1367
1368 /**
1369 * Find a signature that exactly matches a set of actual parameters without
1370 * any implicit type conversions.
1371 */
1372 ir_function_signature *exact_matching_signature(_mesa_glsl_parse_state *state,
1373 const exec_list *actual_ps);
1374
1375 /**
1376 * Name of the function.
1377 */
1378 const char *name;
1379
1380 /** Whether or not this function has a signature that isn't a built-in. */
1381 bool has_user_signature();
1382
1383 /**
1384 * List of ir_function_signature for each overloaded function with this name.
1385 */
1386 struct exec_list signatures;
1387
1388 /**
1389 * is this function a subroutine type declaration
1390 * e.g. subroutine void type1(float arg1);
1391 */
1392 bool is_subroutine;
1393
1394 /**
1395 * is this function associated to a subroutine type
1396 * e.g. subroutine (type1, type2) function_name { function_body };
1397 * would have num_subroutine_types 2,
1398 * and pointers to the type1 and type2 types.
1399 */
1400 int num_subroutine_types;
1401 const struct glsl_type **subroutine_types;
1402
1403 int subroutine_index;
1404 };
1405
function_name()1406 inline const char *ir_function_signature::function_name() const
1407 {
1408 return this->_function->name;
1409 }
1410 /*@}*/
1411
1412
1413 /**
1414 * IR instruction representing high-level if-statements
1415 */
1416 class ir_if : public ir_instruction {
1417 public:
ir_if(ir_rvalue * condition)1418 ir_if(ir_rvalue *condition)
1419 : ir_instruction(ir_type_if), condition(condition)
1420 {
1421 }
1422
1423 virtual ir_if *clone(void *mem_ctx, struct hash_table *ht) const;
1424
accept(ir_visitor * v)1425 virtual void accept(ir_visitor *v)
1426 {
1427 v->visit(this);
1428 }
1429
1430 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1431
1432 ir_rvalue *condition;
1433 /** List of ir_instruction for the body of the then branch */
1434 exec_list then_instructions;
1435 /** List of ir_instruction for the body of the else branch */
1436 exec_list else_instructions;
1437 };
1438
1439
1440 /**
1441 * IR instruction representing a high-level loop structure.
1442 */
1443 class ir_loop : public ir_instruction {
1444 public:
1445 ir_loop();
1446
1447 virtual ir_loop *clone(void *mem_ctx, struct hash_table *ht) const;
1448
accept(ir_visitor * v)1449 virtual void accept(ir_visitor *v)
1450 {
1451 v->visit(this);
1452 }
1453
1454 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1455
1456 /** List of ir_instruction that make up the body of the loop. */
1457 exec_list body_instructions;
1458 };
1459
1460
1461 class ir_assignment : public ir_instruction {
1462 public:
1463 ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs, ir_rvalue *condition = NULL);
1464
1465 /**
1466 * Construct an assignment with an explicit write mask
1467 *
1468 * \note
1469 * Since a write mask is supplied, the LHS must already be a bare
1470 * \c ir_dereference. The cannot be any swizzles in the LHS.
1471 */
1472 ir_assignment(ir_dereference *lhs, ir_rvalue *rhs, ir_rvalue *condition,
1473 unsigned write_mask);
1474
1475 virtual ir_assignment *clone(void *mem_ctx, struct hash_table *ht) const;
1476
1477 virtual ir_constant *constant_expression_value(void *mem_ctx,
1478 struct hash_table *variable_context = NULL);
1479
accept(ir_visitor * v)1480 virtual void accept(ir_visitor *v)
1481 {
1482 v->visit(this);
1483 }
1484
1485 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1486
1487 /**
1488 * Get a whole variable written by an assignment
1489 *
1490 * If the LHS of the assignment writes a whole variable, the variable is
1491 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1492 * assignment are:
1493 *
1494 * - Assigning to a scalar
1495 * - Assigning to all components of a vector
1496 * - Whole array (or matrix) assignment
1497 * - Whole structure assignment
1498 */
1499 ir_variable *whole_variable_written();
1500
1501 /**
1502 * Set the LHS of an assignment
1503 */
1504 void set_lhs(ir_rvalue *lhs);
1505
1506 /**
1507 * Left-hand side of the assignment.
1508 *
1509 * This should be treated as read only. If you need to set the LHS of an
1510 * assignment, use \c ir_assignment::set_lhs.
1511 */
1512 ir_dereference *lhs;
1513
1514 /**
1515 * Value being assigned
1516 */
1517 ir_rvalue *rhs;
1518
1519 /**
1520 * Optional condition for the assignment.
1521 */
1522 ir_rvalue *condition;
1523
1524
1525 /**
1526 * Component mask written
1527 *
1528 * For non-vector types in the LHS, this field will be zero. For vector
1529 * types, a bit will be set for each component that is written. Note that
1530 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1531 *
1532 * A partially-set write mask means that each enabled channel gets
1533 * the value from a consecutive channel of the rhs. For example,
1534 * to write just .xyw of gl_FrontColor with color:
1535 *
1536 * (assign (constant bool (1)) (xyw)
1537 * (var_ref gl_FragColor)
1538 * (swiz xyw (var_ref color)))
1539 */
1540 unsigned write_mask:4;
1541 };
1542
1543 #include "ir_expression_operation.h"
1544
1545 extern const char *const ir_expression_operation_strings[ir_last_opcode + 1];
1546 extern const char *const ir_expression_operation_enum_strings[ir_last_opcode + 1];
1547
1548 class ir_expression : public ir_rvalue {
1549 public:
1550 ir_expression(int op, const struct glsl_type *type,
1551 ir_rvalue *op0, ir_rvalue *op1 = NULL,
1552 ir_rvalue *op2 = NULL, ir_rvalue *op3 = NULL);
1553
1554 /**
1555 * Constructor for unary operation expressions
1556 */
1557 ir_expression(int op, ir_rvalue *);
1558
1559 /**
1560 * Constructor for binary operation expressions
1561 */
1562 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1);
1563
1564 /**
1565 * Constructor for ternary operation expressions
1566 */
1567 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1, ir_rvalue *op2);
1568
1569 virtual bool equals(const ir_instruction *ir,
1570 enum ir_node_type ignore = ir_type_unset) const;
1571
1572 virtual ir_expression *clone(void *mem_ctx, struct hash_table *ht) const;
1573
1574 /**
1575 * Attempt to constant-fold the expression
1576 *
1577 * The "variable_context" hash table links ir_variable * to ir_constant *
1578 * that represent the variables' values. \c NULL represents an empty
1579 * context.
1580 *
1581 * If the expression cannot be constant folded, this method will return
1582 * \c NULL.
1583 */
1584 virtual ir_constant *constant_expression_value(void *mem_ctx,
1585 struct hash_table *variable_context = NULL);
1586
1587 /**
1588 * This is only here for ir_reader to used for testing purposes please use
1589 * the precomputed num_operands field if you need the number of operands.
1590 */
1591 static unsigned get_num_operands(ir_expression_operation);
1592
1593 /**
1594 * Return whether the expression operates on vectors horizontally.
1595 */
is_horizontal()1596 bool is_horizontal() const
1597 {
1598 return operation == ir_binop_all_equal ||
1599 operation == ir_binop_any_nequal ||
1600 operation == ir_binop_dot ||
1601 operation == ir_binop_vector_extract ||
1602 operation == ir_triop_vector_insert ||
1603 operation == ir_binop_ubo_load ||
1604 operation == ir_quadop_vector;
1605 }
1606
1607 /**
1608 * Do a reverse-lookup to translate the given string into an operator.
1609 */
1610 static ir_expression_operation get_operator(const char *);
1611
accept(ir_visitor * v)1612 virtual void accept(ir_visitor *v)
1613 {
1614 v->visit(this);
1615 }
1616
1617 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1618
1619 virtual ir_variable *variable_referenced() const;
1620
1621 /**
1622 * Determine the number of operands used by an expression
1623 */
init_num_operands()1624 void init_num_operands()
1625 {
1626 if (operation == ir_quadop_vector) {
1627 num_operands = this->type->vector_elements;
1628 } else {
1629 num_operands = get_num_operands(operation);
1630 }
1631 }
1632
1633 ir_expression_operation operation;
1634 ir_rvalue *operands[4];
1635 uint8_t num_operands;
1636 };
1637
1638
1639 /**
1640 * HIR instruction representing a high-level function call, containing a list
1641 * of parameters and returning a value in the supplied temporary.
1642 */
1643 class ir_call : public ir_instruction {
1644 public:
ir_call(ir_function_signature * callee,ir_dereference_variable * return_deref,exec_list * actual_parameters)1645 ir_call(ir_function_signature *callee,
1646 ir_dereference_variable *return_deref,
1647 exec_list *actual_parameters)
1648 : ir_instruction(ir_type_call), return_deref(return_deref), callee(callee), sub_var(NULL), array_idx(NULL)
1649 {
1650 assert(callee->return_type != NULL);
1651 actual_parameters->move_nodes_to(& this->actual_parameters);
1652 }
1653
ir_call(ir_function_signature * callee,ir_dereference_variable * return_deref,exec_list * actual_parameters,ir_variable * var,ir_rvalue * array_idx)1654 ir_call(ir_function_signature *callee,
1655 ir_dereference_variable *return_deref,
1656 exec_list *actual_parameters,
1657 ir_variable *var, ir_rvalue *array_idx)
1658 : ir_instruction(ir_type_call), return_deref(return_deref), callee(callee), sub_var(var), array_idx(array_idx)
1659 {
1660 assert(callee->return_type != NULL);
1661 actual_parameters->move_nodes_to(& this->actual_parameters);
1662 }
1663
1664 virtual ir_call *clone(void *mem_ctx, struct hash_table *ht) const;
1665
1666 virtual ir_constant *constant_expression_value(void *mem_ctx,
1667 struct hash_table *variable_context = NULL);
1668
accept(ir_visitor * v)1669 virtual void accept(ir_visitor *v)
1670 {
1671 v->visit(this);
1672 }
1673
1674 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1675
1676 /**
1677 * Get the name of the function being called.
1678 */
callee_name()1679 const char *callee_name() const
1680 {
1681 return callee->function_name();
1682 }
1683
1684 /**
1685 * Generates an inline version of the function before @ir,
1686 * storing the return value in return_deref.
1687 */
1688 void generate_inline(ir_instruction *ir);
1689
1690 /**
1691 * Storage for the function's return value.
1692 * This must be NULL if the return type is void.
1693 */
1694 ir_dereference_variable *return_deref;
1695
1696 /**
1697 * The specific function signature being called.
1698 */
1699 ir_function_signature *callee;
1700
1701 /* List of ir_rvalue of paramaters passed in this call. */
1702 exec_list actual_parameters;
1703
1704 /*
1705 * ARB_shader_subroutine support -
1706 * the subroutine uniform variable and array index
1707 * rvalue to be used in the lowering pass later.
1708 */
1709 ir_variable *sub_var;
1710 ir_rvalue *array_idx;
1711 };
1712
1713
1714 /**
1715 * \name Jump-like IR instructions.
1716 *
1717 * These include \c break, \c continue, \c return, and \c discard.
1718 */
1719 /*@{*/
1720 class ir_jump : public ir_instruction {
1721 protected:
ir_jump(enum ir_node_type t)1722 ir_jump(enum ir_node_type t)
1723 : ir_instruction(t)
1724 {
1725 }
1726 };
1727
1728 class ir_return : public ir_jump {
1729 public:
ir_return()1730 ir_return()
1731 : ir_jump(ir_type_return), value(NULL)
1732 {
1733 }
1734
ir_return(ir_rvalue * value)1735 ir_return(ir_rvalue *value)
1736 : ir_jump(ir_type_return), value(value)
1737 {
1738 }
1739
1740 virtual ir_return *clone(void *mem_ctx, struct hash_table *) const;
1741
get_value()1742 ir_rvalue *get_value() const
1743 {
1744 return value;
1745 }
1746
accept(ir_visitor * v)1747 virtual void accept(ir_visitor *v)
1748 {
1749 v->visit(this);
1750 }
1751
1752 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1753
1754 ir_rvalue *value;
1755 };
1756
1757
1758 /**
1759 * Jump instructions used inside loops
1760 *
1761 * These include \c break and \c continue. The \c break within a loop is
1762 * different from the \c break within a switch-statement.
1763 *
1764 * \sa ir_switch_jump
1765 */
1766 class ir_loop_jump : public ir_jump {
1767 public:
1768 enum jump_mode {
1769 jump_break,
1770 jump_continue
1771 };
1772
ir_loop_jump(jump_mode mode)1773 ir_loop_jump(jump_mode mode)
1774 : ir_jump(ir_type_loop_jump)
1775 {
1776 this->mode = mode;
1777 }
1778
1779 virtual ir_loop_jump *clone(void *mem_ctx, struct hash_table *) const;
1780
accept(ir_visitor * v)1781 virtual void accept(ir_visitor *v)
1782 {
1783 v->visit(this);
1784 }
1785
1786 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1787
is_break()1788 bool is_break() const
1789 {
1790 return mode == jump_break;
1791 }
1792
is_continue()1793 bool is_continue() const
1794 {
1795 return mode == jump_continue;
1796 }
1797
1798 /** Mode selector for the jump instruction. */
1799 enum jump_mode mode;
1800 };
1801
1802 /**
1803 * IR instruction representing discard statements.
1804 */
1805 class ir_discard : public ir_jump {
1806 public:
ir_discard()1807 ir_discard()
1808 : ir_jump(ir_type_discard)
1809 {
1810 this->condition = NULL;
1811 }
1812
ir_discard(ir_rvalue * cond)1813 ir_discard(ir_rvalue *cond)
1814 : ir_jump(ir_type_discard)
1815 {
1816 this->condition = cond;
1817 }
1818
1819 virtual ir_discard *clone(void *mem_ctx, struct hash_table *ht) const;
1820
accept(ir_visitor * v)1821 virtual void accept(ir_visitor *v)
1822 {
1823 v->visit(this);
1824 }
1825
1826 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1827
1828 ir_rvalue *condition;
1829 };
1830 /*@}*/
1831
1832
1833 /**
1834 * IR instruction representing demote statements from
1835 * GL_EXT_demote_to_helper_invocation.
1836 */
1837 class ir_demote : public ir_instruction {
1838 public:
ir_demote()1839 ir_demote()
1840 : ir_instruction(ir_type_demote)
1841 {
1842 }
1843
1844 virtual ir_demote *clone(void *mem_ctx, struct hash_table *ht) const;
1845
accept(ir_visitor * v)1846 virtual void accept(ir_visitor *v)
1847 {
1848 v->visit(this);
1849 }
1850
1851 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1852 };
1853
1854
1855 /**
1856 * Texture sampling opcodes used in ir_texture
1857 */
1858 enum ir_texture_opcode {
1859 ir_tex, /**< Regular texture look-up */
1860 ir_txb, /**< Texture look-up with LOD bias */
1861 ir_txl, /**< Texture look-up with explicit LOD */
1862 ir_txd, /**< Texture look-up with partial derivatives */
1863 ir_txf, /**< Texel fetch with explicit LOD */
1864 ir_txf_ms, /**< Multisample texture fetch */
1865 ir_txs, /**< Texture size */
1866 ir_lod, /**< Texture lod query */
1867 ir_tg4, /**< Texture gather */
1868 ir_query_levels, /**< Texture levels query */
1869 ir_texture_samples, /**< Texture samples query */
1870 ir_samples_identical, /**< Query whether all samples are definitely identical. */
1871 };
1872
1873
1874 /**
1875 * IR instruction to sample a texture
1876 *
1877 * The specific form of the IR instruction depends on the \c mode value
1878 * selected from \c ir_texture_opcodes. In the printed IR, these will
1879 * appear as:
1880 *
1881 * Texel offset (0 or an expression)
1882 * | Projection divisor
1883 * | | Shadow comparator
1884 * | | |
1885 * v v v
1886 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1887 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1888 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1889 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1890 * (txf <type> <sampler> <coordinate> 0 <lod>)
1891 * (txf_ms
1892 * <type> <sampler> <coordinate> <sample_index>)
1893 * (txs <type> <sampler> <lod>)
1894 * (lod <type> <sampler> <coordinate>)
1895 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
1896 * (query_levels <type> <sampler>)
1897 * (samples_identical <sampler> <coordinate>)
1898 */
1899 class ir_texture : public ir_rvalue {
1900 public:
ir_texture(enum ir_texture_opcode op)1901 ir_texture(enum ir_texture_opcode op)
1902 : ir_rvalue(ir_type_texture),
1903 op(op), sampler(NULL), coordinate(NULL), projector(NULL),
1904 shadow_comparator(NULL), offset(NULL)
1905 {
1906 memset(&lod_info, 0, sizeof(lod_info));
1907 }
1908
1909 virtual ir_texture *clone(void *mem_ctx, struct hash_table *) const;
1910
1911 virtual ir_constant *constant_expression_value(void *mem_ctx,
1912 struct hash_table *variable_context = NULL);
1913
accept(ir_visitor * v)1914 virtual void accept(ir_visitor *v)
1915 {
1916 v->visit(this);
1917 }
1918
1919 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1920
1921 virtual bool equals(const ir_instruction *ir,
1922 enum ir_node_type ignore = ir_type_unset) const;
1923
1924 /**
1925 * Return a string representing the ir_texture_opcode.
1926 */
1927 const char *opcode_string();
1928
1929 /** Set the sampler and type. */
1930 void set_sampler(ir_dereference *sampler, const glsl_type *type);
1931
1932 /**
1933 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1934 */
1935 static ir_texture_opcode get_opcode(const char *);
1936
1937 enum ir_texture_opcode op;
1938
1939 /** Sampler to use for the texture access. */
1940 ir_dereference *sampler;
1941
1942 /** Texture coordinate to sample */
1943 ir_rvalue *coordinate;
1944
1945 /**
1946 * Value used for projective divide.
1947 *
1948 * If there is no projective divide (the common case), this will be
1949 * \c NULL. Optimization passes should check for this to point to a constant
1950 * of 1.0 and replace that with \c NULL.
1951 */
1952 ir_rvalue *projector;
1953
1954 /**
1955 * Coordinate used for comparison on shadow look-ups.
1956 *
1957 * If there is no shadow comparison, this will be \c NULL. For the
1958 * \c ir_txf opcode, this *must* be \c NULL.
1959 */
1960 ir_rvalue *shadow_comparator;
1961
1962 /** Texel offset. */
1963 ir_rvalue *offset;
1964
1965 union {
1966 ir_rvalue *lod; /**< Floating point LOD */
1967 ir_rvalue *bias; /**< Floating point LOD bias */
1968 ir_rvalue *sample_index; /**< MSAA sample index */
1969 ir_rvalue *component; /**< Gather component selector */
1970 struct {
1971 ir_rvalue *dPdx; /**< Partial derivative of coordinate wrt X */
1972 ir_rvalue *dPdy; /**< Partial derivative of coordinate wrt Y */
1973 } grad;
1974 } lod_info;
1975 };
1976
1977
1978 struct ir_swizzle_mask {
1979 unsigned x:2;
1980 unsigned y:2;
1981 unsigned z:2;
1982 unsigned w:2;
1983
1984 /**
1985 * Number of components in the swizzle.
1986 */
1987 unsigned num_components:3;
1988
1989 /**
1990 * Does the swizzle contain duplicate components?
1991 *
1992 * L-value swizzles cannot contain duplicate components.
1993 */
1994 unsigned has_duplicates:1;
1995 };
1996
1997
1998 class ir_swizzle : public ir_rvalue {
1999 public:
2000 ir_swizzle(ir_rvalue *, unsigned x, unsigned y, unsigned z, unsigned w,
2001 unsigned count);
2002
2003 ir_swizzle(ir_rvalue *val, const unsigned *components, unsigned count);
2004
2005 ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask);
2006
2007 virtual ir_swizzle *clone(void *mem_ctx, struct hash_table *) const;
2008
2009 virtual ir_constant *constant_expression_value(void *mem_ctx,
2010 struct hash_table *variable_context = NULL);
2011
2012 /**
2013 * Construct an ir_swizzle from the textual representation. Can fail.
2014 */
2015 static ir_swizzle *create(ir_rvalue *, const char *, unsigned vector_length);
2016
accept(ir_visitor * v)2017 virtual void accept(ir_visitor *v)
2018 {
2019 v->visit(this);
2020 }
2021
2022 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2023
2024 virtual bool equals(const ir_instruction *ir,
2025 enum ir_node_type ignore = ir_type_unset) const;
2026
is_lvalue(const struct _mesa_glsl_parse_state * state)2027 bool is_lvalue(const struct _mesa_glsl_parse_state *state) const
2028 {
2029 return val->is_lvalue(state) && !mask.has_duplicates;
2030 }
2031
2032 /**
2033 * Get the variable that is ultimately referenced by an r-value
2034 */
2035 virtual ir_variable *variable_referenced() const;
2036
2037 ir_rvalue *val;
2038 ir_swizzle_mask mask;
2039
2040 private:
2041 /**
2042 * Initialize the mask component of a swizzle
2043 *
2044 * This is used by the \c ir_swizzle constructors.
2045 */
2046 void init_mask(const unsigned *components, unsigned count);
2047 };
2048
2049
2050 class ir_dereference : public ir_rvalue {
2051 public:
2052 virtual ir_dereference *clone(void *mem_ctx, struct hash_table *) const = 0;
2053
2054 bool is_lvalue(const struct _mesa_glsl_parse_state *state) const;
2055
2056 /**
2057 * Get the variable that is ultimately referenced by an r-value
2058 */
2059 virtual ir_variable *variable_referenced() const = 0;
2060
2061 /**
2062 * Get the precision. This can either come from the eventual variable that
2063 * is dereferenced, or from a record member.
2064 */
2065 virtual int precision() const = 0;
2066
2067 protected:
ir_dereference(enum ir_node_type t)2068 ir_dereference(enum ir_node_type t)
2069 : ir_rvalue(t)
2070 {
2071 }
2072 };
2073
2074
2075 class ir_dereference_variable : public ir_dereference {
2076 public:
2077 ir_dereference_variable(ir_variable *var);
2078
2079 virtual ir_dereference_variable *clone(void *mem_ctx,
2080 struct hash_table *) const;
2081
2082 virtual ir_constant *constant_expression_value(void *mem_ctx,
2083 struct hash_table *variable_context = NULL);
2084
2085 virtual bool equals(const ir_instruction *ir,
2086 enum ir_node_type ignore = ir_type_unset) const;
2087
2088 /**
2089 * Get the variable that is ultimately referenced by an r-value
2090 */
variable_referenced()2091 virtual ir_variable *variable_referenced() const
2092 {
2093 return this->var;
2094 }
2095
precision()2096 virtual int precision() const
2097 {
2098 return this->var->data.precision;
2099 }
2100
whole_variable_referenced()2101 virtual ir_variable *whole_variable_referenced()
2102 {
2103 /* ir_dereference_variable objects always dereference the entire
2104 * variable. However, if this dereference is dereferenced by anything
2105 * else, the complete dereference chain is not a whole-variable
2106 * dereference. This method should only be called on the top most
2107 * ir_rvalue in a dereference chain.
2108 */
2109 return this->var;
2110 }
2111
accept(ir_visitor * v)2112 virtual void accept(ir_visitor *v)
2113 {
2114 v->visit(this);
2115 }
2116
2117 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2118
2119 /**
2120 * Object being dereferenced.
2121 */
2122 ir_variable *var;
2123 };
2124
2125
2126 class ir_dereference_array : public ir_dereference {
2127 public:
2128 ir_dereference_array(ir_rvalue *value, ir_rvalue *array_index);
2129
2130 ir_dereference_array(ir_variable *var, ir_rvalue *array_index);
2131
2132 virtual ir_dereference_array *clone(void *mem_ctx,
2133 struct hash_table *) const;
2134
2135 virtual ir_constant *constant_expression_value(void *mem_ctx,
2136 struct hash_table *variable_context = NULL);
2137
2138 virtual bool equals(const ir_instruction *ir,
2139 enum ir_node_type ignore = ir_type_unset) const;
2140
2141 /**
2142 * Get the variable that is ultimately referenced by an r-value
2143 */
variable_referenced()2144 virtual ir_variable *variable_referenced() const
2145 {
2146 return this->array->variable_referenced();
2147 }
2148
precision()2149 virtual int precision() const
2150 {
2151 ir_dereference *deref = this->array->as_dereference();
2152
2153 if (deref == NULL)
2154 return GLSL_PRECISION_NONE;
2155 else
2156 return deref->precision();
2157 }
2158
accept(ir_visitor * v)2159 virtual void accept(ir_visitor *v)
2160 {
2161 v->visit(this);
2162 }
2163
2164 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2165
2166 ir_rvalue *array;
2167 ir_rvalue *array_index;
2168
2169 private:
2170 void set_array(ir_rvalue *value);
2171 };
2172
2173
2174 class ir_dereference_record : public ir_dereference {
2175 public:
2176 ir_dereference_record(ir_rvalue *value, const char *field);
2177
2178 ir_dereference_record(ir_variable *var, const char *field);
2179
2180 virtual ir_dereference_record *clone(void *mem_ctx,
2181 struct hash_table *) const;
2182
2183 virtual ir_constant *constant_expression_value(void *mem_ctx,
2184 struct hash_table *variable_context = NULL);
2185
2186 /**
2187 * Get the variable that is ultimately referenced by an r-value
2188 */
variable_referenced()2189 virtual ir_variable *variable_referenced() const
2190 {
2191 return this->record->variable_referenced();
2192 }
2193
precision()2194 virtual int precision() const
2195 {
2196 glsl_struct_field *field = record->type->fields.structure + field_idx;
2197
2198 return field->precision;
2199 }
2200
accept(ir_visitor * v)2201 virtual void accept(ir_visitor *v)
2202 {
2203 v->visit(this);
2204 }
2205
2206 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2207
2208 ir_rvalue *record;
2209 int field_idx;
2210 };
2211
2212
2213 /**
2214 * Data stored in an ir_constant
2215 */
2216 union ir_constant_data {
2217 unsigned u[16];
2218 int i[16];
2219 float f[16];
2220 bool b[16];
2221 double d[16];
2222 uint16_t f16[16];
2223 uint16_t u16[16];
2224 int16_t i16[16];
2225 uint64_t u64[16];
2226 int64_t i64[16];
2227 };
2228
2229
2230 class ir_constant : public ir_rvalue {
2231 public:
2232 ir_constant(const struct glsl_type *type, const ir_constant_data *data);
2233 ir_constant(bool b, unsigned vector_elements=1);
2234 ir_constant(int16_t i16, unsigned vector_elements=1);
2235 ir_constant(uint16_t u16, unsigned vector_elements=1);
2236 ir_constant(unsigned int u, unsigned vector_elements=1);
2237 ir_constant(int i, unsigned vector_elements=1);
2238 ir_constant(float16_t f16, unsigned vector_elements=1);
2239 ir_constant(float f, unsigned vector_elements=1);
2240 ir_constant(double d, unsigned vector_elements=1);
2241 ir_constant(uint64_t u64, unsigned vector_elements=1);
2242 ir_constant(int64_t i64, unsigned vector_elements=1);
2243
2244 /**
2245 * Construct an ir_constant from a list of ir_constant values
2246 */
2247 ir_constant(const struct glsl_type *type, exec_list *values);
2248
2249 /**
2250 * Construct an ir_constant from a scalar component of another ir_constant
2251 *
2252 * The new \c ir_constant inherits the type of the component from the
2253 * source constant.
2254 *
2255 * \note
2256 * In the case of a matrix constant, the new constant is a scalar, \b not
2257 * a vector.
2258 */
2259 ir_constant(const ir_constant *c, unsigned i);
2260
2261 /**
2262 * Return a new ir_constant of the specified type containing all zeros.
2263 */
2264 static ir_constant *zero(void *mem_ctx, const glsl_type *type);
2265
2266 virtual ir_constant *clone(void *mem_ctx, struct hash_table *) const;
2267
2268 virtual ir_constant *constant_expression_value(void *mem_ctx,
2269 struct hash_table *variable_context = NULL);
2270
accept(ir_visitor * v)2271 virtual void accept(ir_visitor *v)
2272 {
2273 v->visit(this);
2274 }
2275
2276 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2277
2278 virtual bool equals(const ir_instruction *ir,
2279 enum ir_node_type ignore = ir_type_unset) const;
2280
2281 /**
2282 * Get a particular component of a constant as a specific type
2283 *
2284 * This is useful, for example, to get a value from an integer constant
2285 * as a float or bool. This appears frequently when constructors are
2286 * called with all constant parameters.
2287 */
2288 /*@{*/
2289 bool get_bool_component(unsigned i) const;
2290 float get_float_component(unsigned i) const;
2291 uint16_t get_float16_component(unsigned i) const;
2292 double get_double_component(unsigned i) const;
2293 int16_t get_int16_component(unsigned i) const;
2294 uint16_t get_uint16_component(unsigned i) const;
2295 int get_int_component(unsigned i) const;
2296 unsigned get_uint_component(unsigned i) const;
2297 int64_t get_int64_component(unsigned i) const;
2298 uint64_t get_uint64_component(unsigned i) const;
2299 /*@}*/
2300
2301 ir_constant *get_array_element(unsigned i) const;
2302
2303 ir_constant *get_record_field(int idx);
2304
2305 /**
2306 * Copy the values on another constant at a given offset.
2307 *
2308 * The offset is ignored for array or struct copies, it's only for
2309 * scalars or vectors into vectors or matrices.
2310 *
2311 * With identical types on both sides and zero offset it's clone()
2312 * without creating a new object.
2313 */
2314
2315 void copy_offset(ir_constant *src, int offset);
2316
2317 /**
2318 * Copy the values on another constant at a given offset and
2319 * following an assign-like mask.
2320 *
2321 * The mask is ignored for scalars.
2322 *
2323 * Note that this function only handles what assign can handle,
2324 * i.e. at most a vector as source and a column of a matrix as
2325 * destination.
2326 */
2327
2328 void copy_masked_offset(ir_constant *src, int offset, unsigned int mask);
2329
2330 /**
2331 * Determine whether a constant has the same value as another constant
2332 *
2333 * \sa ir_constant::is_zero, ir_constant::is_one,
2334 * ir_constant::is_negative_one
2335 */
2336 bool has_value(const ir_constant *) const;
2337
2338 /**
2339 * Return true if this ir_constant represents the given value.
2340 *
2341 * For vectors, this checks that each component is the given value.
2342 */
2343 virtual bool is_value(float f, int i) const;
2344 virtual bool is_zero() const;
2345 virtual bool is_one() const;
2346 virtual bool is_negative_one() const;
2347
2348 /**
2349 * Return true for constants that could be stored as 16-bit unsigned values.
2350 *
2351 * Note that this will return true even for signed integer ir_constants, as
2352 * long as the value is non-negative and fits in 16-bits.
2353 */
2354 virtual bool is_uint16_constant() const;
2355
2356 /**
2357 * Value of the constant.
2358 *
2359 * The field used to back the values supplied by the constant is determined
2360 * by the type associated with the \c ir_instruction. Constants may be
2361 * scalars, vectors, or matrices.
2362 */
2363 union ir_constant_data value;
2364
2365 /* Array elements and structure fields */
2366 ir_constant **const_elements;
2367
2368 private:
2369 /**
2370 * Parameterless constructor only used by the clone method
2371 */
2372 ir_constant(void);
2373 };
2374
2375 /**
2376 * IR instruction to emit a vertex in a geometry shader.
2377 */
2378 class ir_emit_vertex : public ir_instruction {
2379 public:
ir_emit_vertex(ir_rvalue * stream)2380 ir_emit_vertex(ir_rvalue *stream)
2381 : ir_instruction(ir_type_emit_vertex),
2382 stream(stream)
2383 {
2384 assert(stream);
2385 }
2386
accept(ir_visitor * v)2387 virtual void accept(ir_visitor *v)
2388 {
2389 v->visit(this);
2390 }
2391
clone(void * mem_ctx,struct hash_table * ht)2392 virtual ir_emit_vertex *clone(void *mem_ctx, struct hash_table *ht) const
2393 {
2394 return new(mem_ctx) ir_emit_vertex(this->stream->clone(mem_ctx, ht));
2395 }
2396
2397 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2398
stream_id()2399 int stream_id() const
2400 {
2401 return stream->as_constant()->value.i[0];
2402 }
2403
2404 ir_rvalue *stream;
2405 };
2406
2407 /**
2408 * IR instruction to complete the current primitive and start a new one in a
2409 * geometry shader.
2410 */
2411 class ir_end_primitive : public ir_instruction {
2412 public:
ir_end_primitive(ir_rvalue * stream)2413 ir_end_primitive(ir_rvalue *stream)
2414 : ir_instruction(ir_type_end_primitive),
2415 stream(stream)
2416 {
2417 assert(stream);
2418 }
2419
accept(ir_visitor * v)2420 virtual void accept(ir_visitor *v)
2421 {
2422 v->visit(this);
2423 }
2424
clone(void * mem_ctx,struct hash_table * ht)2425 virtual ir_end_primitive *clone(void *mem_ctx, struct hash_table *ht) const
2426 {
2427 return new(mem_ctx) ir_end_primitive(this->stream->clone(mem_ctx, ht));
2428 }
2429
2430 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2431
stream_id()2432 int stream_id() const
2433 {
2434 return stream->as_constant()->value.i[0];
2435 }
2436
2437 ir_rvalue *stream;
2438 };
2439
2440 /**
2441 * IR instruction for tessellation control and compute shader barrier.
2442 */
2443 class ir_barrier : public ir_instruction {
2444 public:
ir_barrier()2445 ir_barrier()
2446 : ir_instruction(ir_type_barrier)
2447 {
2448 }
2449
accept(ir_visitor * v)2450 virtual void accept(ir_visitor *v)
2451 {
2452 v->visit(this);
2453 }
2454
clone(void * mem_ctx,struct hash_table *)2455 virtual ir_barrier *clone(void *mem_ctx, struct hash_table *) const
2456 {
2457 return new(mem_ctx) ir_barrier();
2458 }
2459
2460 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2461 };
2462
2463 /*@}*/
2464
2465 /**
2466 * Apply a visitor to each IR node in a list
2467 */
2468 void
2469 visit_exec_list(exec_list *list, ir_visitor *visitor);
2470
2471 /**
2472 * Validate invariants on each IR node in a list
2473 */
2474 void validate_ir_tree(exec_list *instructions);
2475
2476 struct _mesa_glsl_parse_state;
2477 struct gl_shader_program;
2478
2479 /**
2480 * Detect whether an unlinked shader contains static recursion
2481 *
2482 * If the list of instructions is determined to contain static recursion,
2483 * \c _mesa_glsl_error will be called to emit error messages for each function
2484 * that is in the recursion cycle.
2485 */
2486 void
2487 detect_recursion_unlinked(struct _mesa_glsl_parse_state *state,
2488 exec_list *instructions);
2489
2490 /**
2491 * Detect whether a linked shader contains static recursion
2492 *
2493 * If the list of instructions is determined to contain static recursion,
2494 * \c link_error_printf will be called to emit error messages for each function
2495 * that is in the recursion cycle. In addition,
2496 * \c gl_shader_program::LinkStatus will be set to false.
2497 */
2498 void
2499 detect_recursion_linked(struct gl_shader_program *prog,
2500 exec_list *instructions);
2501
2502 /**
2503 * Make a clone of each IR instruction in a list
2504 *
2505 * \param in List of IR instructions that are to be cloned
2506 * \param out List to hold the cloned instructions
2507 */
2508 void
2509 clone_ir_list(void *mem_ctx, exec_list *out, const exec_list *in);
2510
2511 extern void
2512 _mesa_glsl_initialize_variables(exec_list *instructions,
2513 struct _mesa_glsl_parse_state *state);
2514
2515 extern void
2516 reparent_ir(exec_list *list, void *mem_ctx);
2517
2518 extern void
2519 do_set_program_inouts(exec_list *instructions, struct gl_program *prog,
2520 gl_shader_stage shader_stage);
2521
2522 extern char *
2523 prototype_string(const glsl_type *return_type, const char *name,
2524 exec_list *parameters);
2525
2526 const char *
2527 mode_string(const ir_variable *var);
2528
2529 /**
2530 * Built-in / reserved GL variables names start with "gl_"
2531 */
2532 static inline bool
is_gl_identifier(const char * s)2533 is_gl_identifier(const char *s)
2534 {
2535 return s && s[0] == 'g' && s[1] == 'l' && s[2] == '_';
2536 }
2537
2538 extern "C" {
2539 #endif /* __cplusplus */
2540
2541 extern void _mesa_print_ir(FILE *f, struct exec_list *instructions,
2542 struct _mesa_glsl_parse_state *state);
2543
2544 extern void
2545 fprint_ir(FILE *f, const void *instruction);
2546
2547 extern const struct gl_builtin_uniform_desc *
2548 _mesa_glsl_get_builtin_uniform_desc(const char *name);
2549
2550 #ifdef __cplusplus
2551 } /* extern "C" */
2552 #endif
2553
2554 unsigned
2555 vertices_per_prim(GLenum prim);
2556
2557 #endif /* IR_H */
2558