3 * Copyright © 2010 Intel Corporation
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:
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
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.
31 #include "util/ralloc.h"
32 #include "compiler/glsl_types.h"
34 #include "ir_visitor.h"
35 #include "ir_hierarchical_visitor.h"
40 * \defgroup IR Intermediate representation nodes
48 * Each concrete class derived from \c ir_instruction has a value in this
49 * enumerant. The value for the type is stored in \c ir_instruction::ir_type
50 * by the constructor. While using type tags is not very C++, it is extremely
51 * convenient. For example, during debugging you can simply inspect
52 * \c ir_instruction::ir_type to find out the actual type of the object.
54 * In addition, it is possible to use a switch-statement based on \c
55 * \c ir_instruction::ir_type to select different behavior for different object
56 * types. For functions that have only slight differences for several object
57 * types, this allows writing very straightforward, readable code.
60 ir_type_dereference_array,
61 ir_type_dereference_record,
62 ir_type_dereference_variable,
71 ir_type_function_signature,
78 ir_type_end_primitive,
80 ir_type_max, /**< maximum ir_type enum number, for validation */
81 ir_type_unset = ir_type_max
86 * Base class of all IR instructions
88 class ir_instruction : public exec_node {
90 enum ir_node_type ir_type;
93 * GCC 4.7+ and clang warn when deleting an ir_instruction unless
94 * there's a virtual destructor present. Because we almost
95 * universally use ralloc for our memory management of
96 * ir_instructions, the destructor doesn't need to do any work.
98 virtual ~ir_instruction()
102 /** ir_print_visitor helper for debugging. */
103 void print(void) const;
104 void fprint(FILE *f) const;
106 virtual void accept(ir_visitor *) = 0;
107 virtual ir_visitor_status accept(ir_hierarchical_visitor *) = 0;
108 virtual ir_instruction *clone(void *mem_ctx,
109 struct hash_table *ht) const = 0;
111 bool is_rvalue() const
113 return ir_type == ir_type_dereference_array ||
114 ir_type == ir_type_dereference_record ||
115 ir_type == ir_type_dereference_variable ||
116 ir_type == ir_type_constant ||
117 ir_type == ir_type_expression ||
118 ir_type == ir_type_swizzle ||
119 ir_type == ir_type_texture;
122 bool is_dereference() const
124 return ir_type == ir_type_dereference_array ||
125 ir_type == ir_type_dereference_record ||
126 ir_type == ir_type_dereference_variable;
131 return ir_type == ir_type_loop_jump ||
132 ir_type == ir_type_return ||
133 ir_type == ir_type_discard;
137 * \name IR instruction downcast functions
139 * These functions either cast the object to a derived class or return
140 * \c NULL if the object's type does not match the specified derived class.
141 * Additional downcast functions will be added as needed.
144 #define AS_BASE(TYPE) \
145 class ir_##TYPE *as_##TYPE() \
147 assume(this != NULL); \
148 return is_##TYPE() ? (ir_##TYPE *) this : NULL; \
150 const class ir_##TYPE *as_##TYPE() const \
152 assume(this != NULL); \
153 return is_##TYPE() ? (ir_##TYPE *) this : NULL; \
161 #define AS_CHILD(TYPE) \
162 class ir_##TYPE * as_##TYPE() \
164 assume(this != NULL); \
165 return ir_type == ir_type_##TYPE ? (ir_##TYPE *) this : NULL; \
167 const class ir_##TYPE * as_##TYPE() const \
169 assume(this != NULL); \
170 return ir_type == ir_type_##TYPE ? (const ir_##TYPE *) this : NULL; \
174 AS_CHILD(dereference_array)
175 AS_CHILD(dereference_variable)
176 AS_CHILD(dereference_record)
191 * IR equality method: Return true if the referenced instruction would
192 * return the same value as this one.
194 * This intended to be used for CSE and algebraic optimizations, on rvalues
195 * in particular. No support for other instruction types (assignments,
196 * jumps, calls, etc.) is planned.
198 virtual bool equals(const ir_instruction *ir,
199 enum ir_node_type ignore = ir_type_unset) const;
202 ir_instruction(enum ir_node_type t)
210 assert(!"Should not get here.");
216 * The base class for all "values"/expression trees.
218 class ir_rvalue : public ir_instruction {
220 const struct glsl_type *type;
222 virtual ir_rvalue *clone(void *mem_ctx, struct hash_table *) const;
224 virtual void accept(ir_visitor *v)
229 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
231 virtual ir_constant *constant_expression_value(void *mem_ctx,
232 struct hash_table *variable_context = NULL);
234 ir_rvalue *as_rvalue_to_saturate();
236 virtual bool is_lvalue(const struct _mesa_glsl_parse_state *state = NULL) const
242 * Get the variable that is ultimately referenced by an r-value
244 virtual ir_variable *variable_referenced() const
251 * If an r-value is a reference to a whole variable, get that variable
254 * Pointer to a variable that is completely dereferenced by the r-value. If
255 * the r-value is not a dereference or the dereference does not access the
256 * entire variable (i.e., it's just one array element, struct field), \c NULL
259 virtual ir_variable *whole_variable_referenced()
265 * Determine if an r-value has the value zero
267 * The base implementation of this function always returns \c false. The
268 * \c ir_constant class over-rides this function to return \c true \b only
269 * for vector and scalar types that have all elements set to the value
270 * zero (or \c false for booleans).
272 * \sa ir_constant::has_value, ir_rvalue::is_one, ir_rvalue::is_negative_one
274 virtual bool is_zero() const;
277 * Determine if an r-value has the value one
279 * The base implementation of this function always returns \c false. The
280 * \c ir_constant class over-rides this function to return \c true \b only
281 * for vector and scalar types that have all elements set to the value
282 * one (or \c true for booleans).
284 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_negative_one
286 virtual bool is_one() const;
289 * Determine if an r-value has the value negative one
291 * The base implementation of this function always returns \c false. The
292 * \c ir_constant class over-rides this function to return \c true \b only
293 * for vector and scalar types that have all elements set to the value
294 * negative one. For boolean types, the result is always \c false.
296 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one
298 virtual bool is_negative_one() const;
301 * Determine if an r-value is an unsigned integer constant which can be
304 * \sa ir_constant::is_uint16_constant.
306 virtual bool is_uint16_constant() const { return false; }
309 * Return a generic value of error_type.
311 * Allocation will be performed with 'mem_ctx' as ralloc owner.
313 static ir_rvalue *error_value(void *mem_ctx);
316 ir_rvalue(enum ir_node_type t);
321 * Variable storage classes
323 enum ir_variable_mode {
324 ir_var_auto = 0, /**< Function local variables and globals. */
325 ir_var_uniform, /**< Variable declared as a uniform. */
326 ir_var_shader_storage, /**< Variable declared as an ssbo. */
327 ir_var_shader_shared, /**< Variable declared as shared. */
332 ir_var_function_inout,
333 ir_var_const_in, /**< "in" param that must be a constant expression */
334 ir_var_system_value, /**< Ex: front-face, instance-id, etc. */
335 ir_var_temporary, /**< Temporary variable generated during compilation. */
336 ir_var_mode_count /**< Number of variable modes */
340 * Enum keeping track of how a variable was declared. For error checking of
341 * the gl_PerVertex redeclaration rules.
343 enum ir_var_declaration_type {
345 * Normal declaration (for most variables, this means an explicit
346 * declaration. Exception: temporaries are always implicitly declared, but
347 * they still use ir_var_declared_normally).
349 * Note: an ir_variable that represents a named interface block uses
350 * ir_var_declared_normally.
352 ir_var_declared_normally = 0,
355 * Variable was explicitly declared (or re-declared) in an unnamed
358 ir_var_declared_in_block,
361 * Variable is an implicitly declared built-in that has not been explicitly
362 * re-declared by the shader.
364 ir_var_declared_implicitly,
367 * Variable is implicitly generated by the compiler and should not be
368 * visible via the API.
374 * \brief Layout qualifiers for gl_FragDepth.
376 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
377 * with a layout qualifier.
379 enum ir_depth_layout {
380 ir_depth_layout_none, /**< No depth layout is specified. */
382 ir_depth_layout_greater,
383 ir_depth_layout_less,
384 ir_depth_layout_unchanged
388 * \brief Convert depth layout qualifier to string.
391 depth_layout_string(ir_depth_layout layout);
394 * Description of built-in state associated with a uniform
396 * \sa ir_variable::state_slots
398 struct ir_state_slot {
399 gl_state_index16 tokens[STATE_LENGTH];
405 * Get the string value for an interpolation qualifier
407 * \return The string that would be used in a shader to specify \c
408 * mode will be returned.
410 * This function is used to generate error messages of the form "shader
411 * uses %s interpolation qualifier", so in the case where there is no
412 * interpolation qualifier, it returns "no".
414 * This function should only be used on a shader input or output variable.
416 const char *interpolation_string(unsigned interpolation);
419 class ir_variable : public ir_instruction {
421 ir_variable(const struct glsl_type *, const char *, ir_variable_mode);
423 virtual ir_variable *clone(void *mem_ctx, struct hash_table *ht) const;
425 virtual void accept(ir_visitor *v)
430 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
434 * Determine whether or not a variable is part of a uniform or
435 * shader storage block.
437 inline bool is_in_buffer_block() const
439 return (this->data.mode == ir_var_uniform ||
440 this->data.mode == ir_var_shader_storage) &&
441 this->interface_type != NULL;
445 * Determine whether or not a variable is part of a shader storage block.
447 inline bool is_in_shader_storage_block() const
449 return this->data.mode == ir_var_shader_storage &&
450 this->interface_type != NULL;
454 * Determine whether or not a variable is the declaration of an interface
457 * For the first declaration below, there will be an \c ir_variable named
458 * "instance" whose type and whose instance_type will be the same
459 * \c glsl_type. For the second declaration, there will be an \c ir_variable
460 * named "f" whose type is float and whose instance_type is B2.
462 * "instance" is an interface instance variable, but "f" is not.
472 inline bool is_interface_instance() const
474 return this->type->without_array() == this->interface_type;
478 * Return whether this variable contains a bindless sampler/image.
480 inline bool contains_bindless() const
482 if (!this->type->contains_sampler() && !this->type->contains_image())
485 return this->data.bindless || this->data.mode != ir_var_uniform;
489 * Set this->interface_type on a newly created variable.
491 void init_interface_type(const struct glsl_type *type)
493 assert(this->interface_type == NULL);
494 this->interface_type = type;
495 if (this->is_interface_instance()) {
496 this->u.max_ifc_array_access =
497 ralloc_array(this, int, type->length);
498 for (unsigned i = 0; i < type->length; i++) {
499 this->u.max_ifc_array_access[i] = -1;
505 * Change this->interface_type on a variable that previously had a
506 * different, but compatible, interface_type. This is used during linking
507 * to set the size of arrays in interface blocks.
509 void change_interface_type(const struct glsl_type *type)
511 if (this->u.max_ifc_array_access != NULL) {
512 /* max_ifc_array_access has already been allocated, so make sure the
513 * new interface has the same number of fields as the old one.
515 assert(this->interface_type->length == type->length);
517 this->interface_type = type;
521 * Change this->interface_type on a variable that previously had a
522 * different, and incompatible, interface_type. This is used during
523 * compilation to handle redeclaration of the built-in gl_PerVertex
526 void reinit_interface_type(const struct glsl_type *type)
528 if (this->u.max_ifc_array_access != NULL) {
530 /* Redeclaring gl_PerVertex is only allowed if none of the built-ins
531 * it defines have been accessed yet; so it's safe to throw away the
532 * old max_ifc_array_access pointer, since all of its values are
535 for (unsigned i = 0; i < this->interface_type->length; i++)
536 assert(this->u.max_ifc_array_access[i] == -1);
538 ralloc_free(this->u.max_ifc_array_access);
539 this->u.max_ifc_array_access = NULL;
541 this->interface_type = NULL;
542 init_interface_type(type);
545 const glsl_type *get_interface_type() const
547 return this->interface_type;
550 enum glsl_interface_packing get_interface_type_packing() const
552 return this->interface_type->get_interface_packing();
555 * Get the max_ifc_array_access pointer
557 * A "set" function is not needed because the array is dynmically allocated
560 inline int *get_max_ifc_array_access()
562 assert(this->data._num_state_slots == 0);
563 return this->u.max_ifc_array_access;
566 inline unsigned get_num_state_slots() const
568 assert(!this->is_interface_instance()
569 || this->data._num_state_slots == 0);
570 return this->data._num_state_slots;
573 inline void set_num_state_slots(unsigned n)
575 assert(!this->is_interface_instance()
577 this->data._num_state_slots = n;
580 inline ir_state_slot *get_state_slots()
582 return this->is_interface_instance() ? NULL : this->u.state_slots;
585 inline const ir_state_slot *get_state_slots() const
587 return this->is_interface_instance() ? NULL : this->u.state_slots;
590 inline ir_state_slot *allocate_state_slots(unsigned n)
592 assert(!this->is_interface_instance());
594 this->u.state_slots = ralloc_array(this, ir_state_slot, n);
595 this->data._num_state_slots = 0;
597 if (this->u.state_slots != NULL)
598 this->data._num_state_slots = n;
600 return this->u.state_slots;
603 inline bool is_interpolation_flat() const
605 return this->data.interpolation == INTERP_MODE_FLAT ||
606 this->type->contains_integer() ||
607 this->type->contains_double();
610 inline bool is_name_ralloced() const
612 return this->name != ir_variable::tmp_name &&
613 this->name != this->name_storage;
617 * Enable emitting extension warnings for this variable
619 void enable_extension_warning(const char *extension);
622 * Get the extension warning string for this variable
624 * If warnings are not enabled, \c NULL is returned.
626 const char *get_extension_warning() const;
629 * Declared type of the variable
631 const struct glsl_type *type;
634 * Declared name of the variable
640 * If the name length fits into name_storage, it's used, otherwise
641 * the name is ralloc'd. shader-db mining showed that 70% of variables
642 * fit here. This is a win over ralloc where only ralloc_header has
643 * 20 bytes on 64-bit (28 bytes with DEBUG), and we can also skip malloc.
645 char name_storage[16];
648 struct ir_variable_data {
651 * Is the variable read-only?
653 * This is set for variables declared as \c const, shader inputs,
656 unsigned read_only:1;
660 unsigned invariant:1;
664 * Has this variable been used for reading or writing?
666 * Several GLSL semantic checks require knowledge of whether or not a
667 * variable has been used. For example, it is an error to redeclare a
668 * variable as invariant after it has been used.
670 * This is only maintained in the ast_to_hir.cpp path, not in
671 * Mesa's fixed function or ARB program paths.
676 * Has this variable been statically assigned?
678 * This answers whether the variable was assigned in any path of
679 * the shader during ast_to_hir. This doesn't answer whether it is
680 * still written after dead code removal, nor is it maintained in
681 * non-ast_to_hir.cpp (GLSL parsing) paths.
686 * When separate shader programs are enabled, only input/outputs between
687 * the stages of a multi-stage separate program can be safely removed
688 * from the shader interface. Other input/outputs must remains active.
690 unsigned always_active_io:1;
693 * Enum indicating how the variable was declared. See
694 * ir_var_declaration_type.
696 * This is used to detect certain kinds of illegal variable redeclarations.
698 unsigned how_declared:2;
701 * Storage class of the variable.
703 * \sa ir_variable_mode
708 * Interpolation mode for shader inputs / outputs
710 * \sa glsl_interp_mode
712 unsigned interpolation:2;
715 * \name ARB_fragment_coord_conventions
718 unsigned origin_upper_left:1;
719 unsigned pixel_center_integer:1;
723 * Was the location explicitly set in the shader?
725 * If the location is explicitly set in the shader, it \b cannot be changed
726 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
729 unsigned explicit_location:1;
730 unsigned explicit_index:1;
733 * Was an initial binding explicitly set in the shader?
735 * If so, constant_value contains an integer ir_constant representing the
736 * initial binding point.
738 unsigned explicit_binding:1;
741 * Was an initial component explicitly set in the shader?
743 unsigned explicit_component:1;
746 * Does this variable have an initializer?
748 * This is used by the linker to cross-validiate initializers of global
751 unsigned has_initializer:1;
754 * Is this variable a generic output or input that has not yet been matched
755 * up to a variable in another stage of the pipeline?
757 * This is used by the linker as scratch storage while assigning locations
758 * to generic inputs and outputs.
760 unsigned is_unmatched_generic_inout:1;
763 * Is this varying used only by transform feedback?
765 * This is used by the linker to decide if its safe to pack the varying.
767 unsigned is_xfb_only:1;
770 * Was a transfor feedback buffer set in the shader?
772 unsigned explicit_xfb_buffer:1;
775 * Was a transfor feedback offset set in the shader?
777 unsigned explicit_xfb_offset:1;
780 * Was a transfor feedback stride set in the shader?
782 unsigned explicit_xfb_stride:1;
785 * If non-zero, then this variable may be packed along with other variables
786 * into a single varying slot, so this offset should be applied when
787 * accessing components. For example, an offset of 1 means that the x
788 * component of this variable is actually stored in component y of the
789 * location specified by \c location.
791 unsigned location_frac:2;
794 * Layout of the matrix. Uses glsl_matrix_layout values.
796 unsigned matrix_layout:2;
799 * Non-zero if this variable was created by lowering a named interface
802 unsigned from_named_ifc_block:1;
805 * Non-zero if the variable must be a shader input. This is useful for
806 * constraints on function parameters.
808 unsigned must_be_shader_input:1;
811 * Output index for dual source blending.
814 * The GLSL spec only allows the values 0 or 1 for the index in \b dual
820 * Precision qualifier.
822 * In desktop GLSL we do not care about precision qualifiers at all, in
823 * fact, the spec says that precision qualifiers are ignored.
825 * To make things easy, we make it so that this field is always
826 * GLSL_PRECISION_NONE on desktop shaders. This way all the variables
827 * have the same precision value and the checks we add in the compiler
828 * for this field will never break a desktop shader compile.
830 unsigned precision:2;
833 * \brief Layout qualifier for gl_FragDepth.
835 * This is not equal to \c ir_depth_layout_none if and only if this
836 * variable is \c gl_FragDepth and a layout qualifier is specified.
838 ir_depth_layout depth_layout:3;
843 unsigned memory_read_only:1; /**< "readonly" qualifier. */
844 unsigned memory_write_only:1; /**< "writeonly" qualifier. */
845 unsigned memory_coherent:1;
846 unsigned memory_volatile:1;
847 unsigned memory_restrict:1;
850 * ARB_shader_storage_buffer_object
852 unsigned from_ssbo_unsized_array:1; /**< unsized array buffer variable. */
854 unsigned implicit_sized_array:1;
857 * Whether this is a fragment shader output implicitly initialized with
858 * the previous contents of the specified render target at the
859 * framebuffer location corresponding to this shader invocation.
861 unsigned fb_fetch_output:1;
864 * Non-zero if this variable is considered bindless as defined by
865 * ARB_bindless_texture.
870 * Non-zero if this variable is considered bound as defined by
871 * ARB_bindless_texture.
876 * Emit a warning if this variable is accessed.
879 uint8_t warn_extension_index;
882 /** Image internal format if specified explicitly, otherwise GL_NONE. */
883 uint16_t image_format;
887 * Number of state slots used
890 * This could be stored in as few as 7-bits, if necessary. If it is made
891 * smaller, add an assertion to \c ir_variable::allocate_state_slots to
894 uint16_t _num_state_slots;
898 * Initial binding point for a sampler, atomic, or UBO.
900 * For array types, this represents the binding point for the first element.
905 * Storage location of the base of this variable
907 * The precise meaning of this field depends on the nature of the variable.
909 * - Vertex shader input: one of the values from \c gl_vert_attrib.
910 * - Vertex shader output: one of the values from \c gl_varying_slot.
911 * - Geometry shader input: one of the values from \c gl_varying_slot.
912 * - Geometry shader output: one of the values from \c gl_varying_slot.
913 * - Fragment shader input: one of the values from \c gl_varying_slot.
914 * - Fragment shader output: one of the values from \c gl_frag_result.
915 * - Uniforms: Per-stage uniform slot number for default uniform block.
916 * - Uniforms: Index within the uniform block definition for UBO members.
917 * - Non-UBO Uniforms: explicit location until linking then reused to
918 * store uniform slot number.
919 * - Other: This field is not currently used.
921 * If the variable is a uniform, shader input, or shader output, and the
922 * slot has not been assigned, the value will be -1.
927 * for glsl->tgsi/mesa IR we need to store the index into the
928 * parameters for uniforms, initially the code overloaded location
929 * but this causes problems with indirect samplers and AoA.
930 * This is assigned in _mesa_generate_parameters_list_for_uniforms.
935 * Vertex stream output identifier.
937 * For packed outputs, bit 31 is set and bits [2*i+1,2*i] indicate the
938 * stream of the i-th component.
943 * Atomic, transform feedback or block member offset.
948 * Highest element accessed with a constant expression array index
950 * Not used for non-array variables. -1 is never accessed.
952 int max_array_access;
955 * Transform feedback buffer.
960 * Transform feedback stride.
965 * Allow (only) ir_variable direct access private members.
967 friend class ir_variable;
971 * Value assigned in the initializer of a variable declared "const"
973 ir_constant *constant_value;
976 * Constant expression assigned in the initializer of the variable
979 * This field and \c ::constant_value are distinct. Even if the two fields
980 * refer to constants with the same value, they must point to separate
983 ir_constant *constant_initializer;
986 static const char *const warn_extension_table[];
990 * For variables which satisfy the is_interface_instance() predicate,
991 * this points to an array of integers such that if the ith member of
992 * the interface block is an array, max_ifc_array_access[i] is the
993 * maximum array element of that member that has been accessed. If the
994 * ith member of the interface block is not an array,
995 * max_ifc_array_access[i] is unused.
997 * For variables whose type is not an interface block, this pointer is
1000 int *max_ifc_array_access;
1003 * Built-in state that backs this uniform
1005 * Once set at variable creation, \c state_slots must remain invariant.
1007 * If the variable is not a uniform, \c _num_state_slots will be zero
1008 * and \c state_slots will be \c NULL.
1010 ir_state_slot *state_slots;
1014 * For variables that are in an interface block or are an instance of an
1015 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
1017 * \sa ir_variable::location
1019 const glsl_type *interface_type;
1022 * Name used for anonymous compiler temporaries
1024 static const char tmp_name[];
1028 * Should the construct keep names for ir_var_temporary variables?
1030 * When this global is false, names passed to the constructor for
1031 * \c ir_var_temporary variables will be dropped. Instead, the variable will
1032 * be named "compiler_temp". This name will be in static storage.
1035 * \b NEVER change the mode of an \c ir_var_temporary.
1038 * This variable is \b not thread-safe. It is global, \b not
1039 * per-context. It begins life false. A context can, at some point, make
1040 * it true. From that point on, it will be true forever. This should be
1041 * okay since it will only be set true while debugging.
1043 static bool temporaries_allocate_names;
1047 * A function that returns whether a built-in function is available in the
1048 * current shading language (based on version, ES or desktop, and extensions).
1050 typedef bool (*builtin_available_predicate)(const _mesa_glsl_parse_state *);
1052 #define MAKE_INTRINSIC_FOR_TYPE(op, t) \
1053 ir_intrinsic_generic_ ## op - ir_intrinsic_generic_load + ir_intrinsic_ ## t ## _ ## load
1055 #define MAP_INTRINSIC_TO_TYPE(i, t) \
1056 ir_intrinsic_id(int(i) - int(ir_intrinsic_generic_load) + int(ir_intrinsic_ ## t ## _ ## load))
1058 enum ir_intrinsic_id {
1059 ir_intrinsic_invalid = 0,
1062 * \name Generic intrinsics
1064 * Each of these intrinsics has a specific version for shared variables and
1068 ir_intrinsic_generic_load,
1069 ir_intrinsic_generic_store,
1070 ir_intrinsic_generic_atomic_add,
1071 ir_intrinsic_generic_atomic_and,
1072 ir_intrinsic_generic_atomic_or,
1073 ir_intrinsic_generic_atomic_xor,
1074 ir_intrinsic_generic_atomic_min,
1075 ir_intrinsic_generic_atomic_max,
1076 ir_intrinsic_generic_atomic_exchange,
1077 ir_intrinsic_generic_atomic_comp_swap,
1080 ir_intrinsic_atomic_counter_read,
1081 ir_intrinsic_atomic_counter_increment,
1082 ir_intrinsic_atomic_counter_predecrement,
1083 ir_intrinsic_atomic_counter_add,
1084 ir_intrinsic_atomic_counter_and,
1085 ir_intrinsic_atomic_counter_or,
1086 ir_intrinsic_atomic_counter_xor,
1087 ir_intrinsic_atomic_counter_min,
1088 ir_intrinsic_atomic_counter_max,
1089 ir_intrinsic_atomic_counter_exchange,
1090 ir_intrinsic_atomic_counter_comp_swap,
1092 ir_intrinsic_image_load,
1093 ir_intrinsic_image_store,
1094 ir_intrinsic_image_atomic_add,
1095 ir_intrinsic_image_atomic_and,
1096 ir_intrinsic_image_atomic_or,
1097 ir_intrinsic_image_atomic_xor,
1098 ir_intrinsic_image_atomic_min,
1099 ir_intrinsic_image_atomic_max,
1100 ir_intrinsic_image_atomic_exchange,
1101 ir_intrinsic_image_atomic_comp_swap,
1102 ir_intrinsic_image_size,
1103 ir_intrinsic_image_samples,
1105 ir_intrinsic_ssbo_load,
1106 ir_intrinsic_ssbo_store = MAKE_INTRINSIC_FOR_TYPE(store, ssbo),
1107 ir_intrinsic_ssbo_atomic_add = MAKE_INTRINSIC_FOR_TYPE(atomic_add, ssbo),
1108 ir_intrinsic_ssbo_atomic_and = MAKE_INTRINSIC_FOR_TYPE(atomic_and, ssbo),
1109 ir_intrinsic_ssbo_atomic_or = MAKE_INTRINSIC_FOR_TYPE(atomic_or, ssbo),
1110 ir_intrinsic_ssbo_atomic_xor = MAKE_INTRINSIC_FOR_TYPE(atomic_xor, ssbo),
1111 ir_intrinsic_ssbo_atomic_min = MAKE_INTRINSIC_FOR_TYPE(atomic_min, ssbo),
1112 ir_intrinsic_ssbo_atomic_max = MAKE_INTRINSIC_FOR_TYPE(atomic_max, ssbo),
1113 ir_intrinsic_ssbo_atomic_exchange = MAKE_INTRINSIC_FOR_TYPE(atomic_exchange, ssbo),
1114 ir_intrinsic_ssbo_atomic_comp_swap = MAKE_INTRINSIC_FOR_TYPE(atomic_comp_swap, ssbo),
1116 ir_intrinsic_memory_barrier,
1117 ir_intrinsic_shader_clock,
1118 ir_intrinsic_group_memory_barrier,
1119 ir_intrinsic_memory_barrier_atomic_counter,
1120 ir_intrinsic_memory_barrier_buffer,
1121 ir_intrinsic_memory_barrier_image,
1122 ir_intrinsic_memory_barrier_shared,
1124 ir_intrinsic_vote_all,
1125 ir_intrinsic_vote_any,
1126 ir_intrinsic_vote_eq,
1127 ir_intrinsic_ballot,
1128 ir_intrinsic_read_invocation,
1129 ir_intrinsic_read_first_invocation,
1131 ir_intrinsic_shared_load,
1132 ir_intrinsic_shared_store = MAKE_INTRINSIC_FOR_TYPE(store, shared),
1133 ir_intrinsic_shared_atomic_add = MAKE_INTRINSIC_FOR_TYPE(atomic_add, shared),
1134 ir_intrinsic_shared_atomic_and = MAKE_INTRINSIC_FOR_TYPE(atomic_and, shared),
1135 ir_intrinsic_shared_atomic_or = MAKE_INTRINSIC_FOR_TYPE(atomic_or, shared),
1136 ir_intrinsic_shared_atomic_xor = MAKE_INTRINSIC_FOR_TYPE(atomic_xor, shared),
1137 ir_intrinsic_shared_atomic_min = MAKE_INTRINSIC_FOR_TYPE(atomic_min, shared),
1138 ir_intrinsic_shared_atomic_max = MAKE_INTRINSIC_FOR_TYPE(atomic_max, shared),
1139 ir_intrinsic_shared_atomic_exchange = MAKE_INTRINSIC_FOR_TYPE(atomic_exchange, shared),
1140 ir_intrinsic_shared_atomic_comp_swap = MAKE_INTRINSIC_FOR_TYPE(atomic_comp_swap, shared),
1145 * The representation of a function instance; may be the full definition or
1146 * simply a prototype.
1148 class ir_function_signature : public ir_instruction {
1149 /* An ir_function_signature will be part of the list of signatures in
1153 ir_function_signature(const glsl_type *return_type,
1154 builtin_available_predicate builtin_avail = NULL);
1156 virtual ir_function_signature *clone(void *mem_ctx,
1157 struct hash_table *ht) const;
1158 ir_function_signature *clone_prototype(void *mem_ctx,
1159 struct hash_table *ht) const;
1161 virtual void accept(ir_visitor *v)
1166 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1169 * Attempt to evaluate this function as a constant expression,
1170 * given a list of the actual parameters and the variable context.
1171 * Returns NULL for non-built-ins.
1173 ir_constant *constant_expression_value(void *mem_ctx,
1174 exec_list *actual_parameters,
1175 struct hash_table *variable_context);
1178 * Get the name of the function for which this is a signature
1180 const char *function_name() const;
1183 * Get a handle to the function for which this is a signature
1185 * There is no setter function, this function returns a \c const pointer,
1186 * and \c ir_function_signature::_function is private for a reason. The
1187 * only way to make a connection between a function and function signature
1188 * is via \c ir_function::add_signature. This helps ensure that certain
1189 * invariants (i.e., a function signature is in the list of signatures for
1190 * its \c _function) are met.
1192 * \sa ir_function::add_signature
1194 inline const class ir_function *function() const
1196 return this->_function;
1200 * Check whether the qualifiers match between this signature's parameters
1201 * and the supplied parameter list. If not, returns the name of the first
1202 * parameter with mismatched qualifiers (for use in error messages).
1204 const char *qualifiers_match(exec_list *params);
1207 * Replace the current parameter list with the given one. This is useful
1208 * if the current information came from a prototype, and either has invalid
1209 * or missing parameter names.
1211 void replace_parameters(exec_list *new_params);
1214 * Function return type.
1216 * \note This discards the optional precision qualifier.
1218 const struct glsl_type *return_type;
1221 * List of ir_variable of function parameters.
1223 * This represents the storage. The paramaters passed in a particular
1224 * call will be in ir_call::actual_paramaters.
1226 struct exec_list parameters;
1228 /** Whether or not this function has a body (which may be empty). */
1229 unsigned is_defined:1;
1231 /** Whether or not this function signature is a built-in. */
1232 bool is_builtin() const;
1235 * Whether or not this function is an intrinsic to be implemented
1238 inline bool is_intrinsic() const
1240 return intrinsic_id != ir_intrinsic_invalid;
1243 /** Indentifier for this intrinsic. */
1244 enum ir_intrinsic_id intrinsic_id;
1246 /** Whether or not a built-in is available for this shader. */
1247 bool is_builtin_available(const _mesa_glsl_parse_state *state) const;
1249 /** Body of instructions in the function. */
1250 struct exec_list body;
1254 * A function pointer to a predicate that answers whether a built-in
1255 * function is available in the current shader. NULL if not a built-in.
1257 builtin_available_predicate builtin_avail;
1259 /** Function of which this signature is one overload. */
1260 class ir_function *_function;
1262 /** Function signature of which this one is a prototype clone */
1263 const ir_function_signature *origin;
1265 friend class ir_function;
1268 * Helper function to run a list of instructions for constant
1269 * expression evaluation.
1271 * The hash table represents the values of the visible variables.
1272 * There are no scoping issues because the table is indexed on
1273 * ir_variable pointers, not variable names.
1275 * Returns false if the expression is not constant, true otherwise,
1276 * and the value in *result if result is non-NULL.
1278 bool constant_expression_evaluate_expression_list(void *mem_ctx,
1279 const struct exec_list &body,
1280 struct hash_table *variable_context,
1281 ir_constant **result);
1286 * Header for tracking multiple overloaded functions with the same name.
1287 * Contains a list of ir_function_signatures representing each of the
1290 class ir_function : public ir_instruction {
1292 ir_function(const char *name);
1294 virtual ir_function *clone(void *mem_ctx, struct hash_table *ht) const;
1296 virtual void accept(ir_visitor *v)
1301 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1303 void add_signature(ir_function_signature *sig)
1305 sig->_function = this;
1306 this->signatures.push_tail(sig);
1310 * Find a signature that matches a set of actual parameters, taking implicit
1311 * conversions into account. Also flags whether the match was exact.
1313 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
1314 const exec_list *actual_param,
1315 bool allow_builtins,
1316 bool *match_is_exact);
1319 * Find a signature that matches a set of actual parameters, taking implicit
1320 * conversions into account.
1322 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
1323 const exec_list *actual_param,
1324 bool allow_builtins);
1327 * Find a signature that exactly matches a set of actual parameters without
1328 * any implicit type conversions.
1330 ir_function_signature *exact_matching_signature(_mesa_glsl_parse_state *state,
1331 const exec_list *actual_ps);
1334 * Name of the function.
1338 /** Whether or not this function has a signature that isn't a built-in. */
1339 bool has_user_signature();
1342 * List of ir_function_signature for each overloaded function with this name.
1344 struct exec_list signatures;
1347 * is this function a subroutine type declaration
1348 * e.g. subroutine void type1(float arg1);
1353 * is this function associated to a subroutine type
1354 * e.g. subroutine (type1, type2) function_name { function_body };
1355 * would have num_subroutine_types 2,
1356 * and pointers to the type1 and type2 types.
1358 int num_subroutine_types;
1359 const struct glsl_type **subroutine_types;
1361 int subroutine_index;
1364 inline const char *ir_function_signature::function_name() const
1366 return this->_function->name;
1372 * IR instruction representing high-level if-statements
1374 class ir_if : public ir_instruction {
1376 ir_if(ir_rvalue *condition)
1377 : ir_instruction(ir_type_if), condition(condition)
1381 virtual ir_if *clone(void *mem_ctx, struct hash_table *ht) const;
1383 virtual void accept(ir_visitor *v)
1388 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1390 ir_rvalue *condition;
1391 /** List of ir_instruction for the body of the then branch */
1392 exec_list then_instructions;
1393 /** List of ir_instruction for the body of the else branch */
1394 exec_list else_instructions;
1399 * IR instruction representing a high-level loop structure.
1401 class ir_loop : public ir_instruction {
1405 virtual ir_loop *clone(void *mem_ctx, struct hash_table *ht) const;
1407 virtual void accept(ir_visitor *v)
1412 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1414 /** List of ir_instruction that make up the body of the loop. */
1415 exec_list body_instructions;
1419 class ir_assignment : public ir_instruction {
1421 ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs, ir_rvalue *condition = NULL);
1424 * Construct an assignment with an explicit write mask
1427 * Since a write mask is supplied, the LHS must already be a bare
1428 * \c ir_dereference. The cannot be any swizzles in the LHS.
1430 ir_assignment(ir_dereference *lhs, ir_rvalue *rhs, ir_rvalue *condition,
1431 unsigned write_mask);
1433 virtual ir_assignment *clone(void *mem_ctx, struct hash_table *ht) const;
1435 virtual ir_constant *constant_expression_value(void *mem_ctx,
1436 struct hash_table *variable_context = NULL);
1438 virtual void accept(ir_visitor *v)
1443 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1446 * Get a whole variable written by an assignment
1448 * If the LHS of the assignment writes a whole variable, the variable is
1449 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1452 * - Assigning to a scalar
1453 * - Assigning to all components of a vector
1454 * - Whole array (or matrix) assignment
1455 * - Whole structure assignment
1457 ir_variable *whole_variable_written();
1460 * Set the LHS of an assignment
1462 void set_lhs(ir_rvalue *lhs);
1465 * Left-hand side of the assignment.
1467 * This should be treated as read only. If you need to set the LHS of an
1468 * assignment, use \c ir_assignment::set_lhs.
1470 ir_dereference *lhs;
1473 * Value being assigned
1478 * Optional condition for the assignment.
1480 ir_rvalue *condition;
1484 * Component mask written
1486 * For non-vector types in the LHS, this field will be zero. For vector
1487 * types, a bit will be set for each component that is written. Note that
1488 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1490 * A partially-set write mask means that each enabled channel gets
1491 * the value from a consecutive channel of the rhs. For example,
1492 * to write just .xyw of gl_FrontColor with color:
1494 * (assign (constant bool (1)) (xyw)
1495 * (var_ref gl_FragColor)
1496 * (swiz xyw (var_ref color)))
1498 unsigned write_mask:4;
1501 #include "ir_expression_operation.h"
1503 extern const char *const ir_expression_operation_strings[ir_last_opcode + 1];
1504 extern const char *const ir_expression_operation_enum_strings[ir_last_opcode + 1];
1506 class ir_expression : public ir_rvalue {
1508 ir_expression(int op, const struct glsl_type *type,
1509 ir_rvalue *op0, ir_rvalue *op1 = NULL,
1510 ir_rvalue *op2 = NULL, ir_rvalue *op3 = NULL);
1513 * Constructor for unary operation expressions
1515 ir_expression(int op, ir_rvalue *);
1518 * Constructor for binary operation expressions
1520 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1);
1523 * Constructor for ternary operation expressions
1525 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1, ir_rvalue *op2);
1527 virtual bool equals(const ir_instruction *ir,
1528 enum ir_node_type ignore = ir_type_unset) const;
1530 virtual ir_expression *clone(void *mem_ctx, struct hash_table *ht) const;
1533 * Attempt to constant-fold the expression
1535 * The "variable_context" hash table links ir_variable * to ir_constant *
1536 * that represent the variables' values. \c NULL represents an empty
1539 * If the expression cannot be constant folded, this method will return
1542 virtual ir_constant *constant_expression_value(void *mem_ctx,
1543 struct hash_table *variable_context = NULL);
1546 * This is only here for ir_reader to used for testing purposes please use
1547 * the precomputed num_operands field if you need the number of operands.
1549 static unsigned get_num_operands(ir_expression_operation);
1552 * Return whether the expression operates on vectors horizontally.
1554 bool is_horizontal() const
1556 return operation == ir_binop_all_equal ||
1557 operation == ir_binop_any_nequal ||
1558 operation == ir_binop_dot ||
1559 operation == ir_binop_vector_extract ||
1560 operation == ir_triop_vector_insert ||
1561 operation == ir_binop_ubo_load ||
1562 operation == ir_quadop_vector;
1566 * Do a reverse-lookup to translate the given string into an operator.
1568 static ir_expression_operation get_operator(const char *);
1570 virtual void accept(ir_visitor *v)
1575 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1577 virtual ir_variable *variable_referenced() const;
1580 * Determine the number of operands used by an expression
1582 void init_num_operands()
1584 if (operation == ir_quadop_vector) {
1585 num_operands = this->type->vector_elements;
1587 num_operands = get_num_operands(operation);
1591 ir_expression_operation operation;
1592 ir_rvalue *operands[4];
1593 uint8_t num_operands;
1598 * HIR instruction representing a high-level function call, containing a list
1599 * of parameters and returning a value in the supplied temporary.
1601 class ir_call : public ir_instruction {
1603 ir_call(ir_function_signature *callee,
1604 ir_dereference_variable *return_deref,
1605 exec_list *actual_parameters)
1606 : ir_instruction(ir_type_call), return_deref(return_deref), callee(callee), sub_var(NULL), array_idx(NULL)
1608 assert(callee->return_type != NULL);
1609 actual_parameters->move_nodes_to(& this->actual_parameters);
1612 ir_call(ir_function_signature *callee,
1613 ir_dereference_variable *return_deref,
1614 exec_list *actual_parameters,
1615 ir_variable *var, ir_rvalue *array_idx)
1616 : ir_instruction(ir_type_call), return_deref(return_deref), callee(callee), sub_var(var), array_idx(array_idx)
1618 assert(callee->return_type != NULL);
1619 actual_parameters->move_nodes_to(& this->actual_parameters);
1622 virtual ir_call *clone(void *mem_ctx, struct hash_table *ht) const;
1624 virtual ir_constant *constant_expression_value(void *mem_ctx,
1625 struct hash_table *variable_context = NULL);
1627 virtual void accept(ir_visitor *v)
1632 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1635 * Get the name of the function being called.
1637 const char *callee_name() const
1639 return callee->function_name();
1643 * Generates an inline version of the function before @ir,
1644 * storing the return value in return_deref.
1646 void generate_inline(ir_instruction *ir);
1649 * Storage for the function's return value.
1650 * This must be NULL if the return type is void.
1652 ir_dereference_variable *return_deref;
1655 * The specific function signature being called.
1657 ir_function_signature *callee;
1659 /* List of ir_rvalue of paramaters passed in this call. */
1660 exec_list actual_parameters;
1663 * ARB_shader_subroutine support -
1664 * the subroutine uniform variable and array index
1665 * rvalue to be used in the lowering pass later.
1667 ir_variable *sub_var;
1668 ir_rvalue *array_idx;
1673 * \name Jump-like IR instructions.
1675 * These include \c break, \c continue, \c return, and \c discard.
1678 class ir_jump : public ir_instruction {
1680 ir_jump(enum ir_node_type t)
1686 class ir_return : public ir_jump {
1689 : ir_jump(ir_type_return), value(NULL)
1693 ir_return(ir_rvalue *value)
1694 : ir_jump(ir_type_return), value(value)
1698 virtual ir_return *clone(void *mem_ctx, struct hash_table *) const;
1700 ir_rvalue *get_value() const
1705 virtual void accept(ir_visitor *v)
1710 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1717 * Jump instructions used inside loops
1719 * These include \c break and \c continue. The \c break within a loop is
1720 * different from the \c break within a switch-statement.
1722 * \sa ir_switch_jump
1724 class ir_loop_jump : public ir_jump {
1731 ir_loop_jump(jump_mode mode)
1732 : ir_jump(ir_type_loop_jump)
1737 virtual ir_loop_jump *clone(void *mem_ctx, struct hash_table *) const;
1739 virtual void accept(ir_visitor *v)
1744 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1746 bool is_break() const
1748 return mode == jump_break;
1751 bool is_continue() const
1753 return mode == jump_continue;
1756 /** Mode selector for the jump instruction. */
1757 enum jump_mode mode;
1761 * IR instruction representing discard statements.
1763 class ir_discard : public ir_jump {
1766 : ir_jump(ir_type_discard)
1768 this->condition = NULL;
1771 ir_discard(ir_rvalue *cond)
1772 : ir_jump(ir_type_discard)
1774 this->condition = cond;
1777 virtual ir_discard *clone(void *mem_ctx, struct hash_table *ht) const;
1779 virtual void accept(ir_visitor *v)
1784 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1786 ir_rvalue *condition;
1792 * Texture sampling opcodes used in ir_texture
1794 enum ir_texture_opcode {
1795 ir_tex, /**< Regular texture look-up */
1796 ir_txb, /**< Texture look-up with LOD bias */
1797 ir_txl, /**< Texture look-up with explicit LOD */
1798 ir_txd, /**< Texture look-up with partial derivatvies */
1799 ir_txf, /**< Texel fetch with explicit LOD */
1800 ir_txf_ms, /**< Multisample texture fetch */
1801 ir_txs, /**< Texture size */
1802 ir_lod, /**< Texture lod query */
1803 ir_tg4, /**< Texture gather */
1804 ir_query_levels, /**< Texture levels query */
1805 ir_texture_samples, /**< Texture samples query */
1806 ir_samples_identical, /**< Query whether all samples are definitely identical. */
1811 * IR instruction to sample a texture
1813 * The specific form of the IR instruction depends on the \c mode value
1814 * selected from \c ir_texture_opcodes. In the printed IR, these will
1817 * Texel offset (0 or an expression)
1818 * | Projection divisor
1819 * | | Shadow comparator
1822 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1823 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1824 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1825 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1826 * (txf <type> <sampler> <coordinate> 0 <lod>)
1828 * <type> <sampler> <coordinate> <sample_index>)
1829 * (txs <type> <sampler> <lod>)
1830 * (lod <type> <sampler> <coordinate>)
1831 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
1832 * (query_levels <type> <sampler>)
1833 * (samples_identical <sampler> <coordinate>)
1835 class ir_texture : public ir_rvalue {
1837 ir_texture(enum ir_texture_opcode op)
1838 : ir_rvalue(ir_type_texture),
1839 op(op), sampler(NULL), coordinate(NULL), projector(NULL),
1840 shadow_comparator(NULL), offset(NULL)
1842 memset(&lod_info, 0, sizeof(lod_info));
1845 virtual ir_texture *clone(void *mem_ctx, struct hash_table *) const;
1847 virtual ir_constant *constant_expression_value(void *mem_ctx,
1848 struct hash_table *variable_context = NULL);
1850 virtual void accept(ir_visitor *v)
1855 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1857 virtual bool equals(const ir_instruction *ir,
1858 enum ir_node_type ignore = ir_type_unset) const;
1861 * Return a string representing the ir_texture_opcode.
1863 const char *opcode_string();
1865 /** Set the sampler and type. */
1866 void set_sampler(ir_dereference *sampler, const glsl_type *type);
1869 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1871 static ir_texture_opcode get_opcode(const char *);
1873 enum ir_texture_opcode op;
1875 /** Sampler to use for the texture access. */
1876 ir_dereference *sampler;
1878 /** Texture coordinate to sample */
1879 ir_rvalue *coordinate;
1882 * Value used for projective divide.
1884 * If there is no projective divide (the common case), this will be
1885 * \c NULL. Optimization passes should check for this to point to a constant
1886 * of 1.0 and replace that with \c NULL.
1888 ir_rvalue *projector;
1891 * Coordinate used for comparison on shadow look-ups.
1893 * If there is no shadow comparison, this will be \c NULL. For the
1894 * \c ir_txf opcode, this *must* be \c NULL.
1896 ir_rvalue *shadow_comparator;
1898 /** Texel offset. */
1902 ir_rvalue *lod; /**< Floating point LOD */
1903 ir_rvalue *bias; /**< Floating point LOD bias */
1904 ir_rvalue *sample_index; /**< MSAA sample index */
1905 ir_rvalue *component; /**< Gather component selector */
1907 ir_rvalue *dPdx; /**< Partial derivative of coordinate wrt X */
1908 ir_rvalue *dPdy; /**< Partial derivative of coordinate wrt Y */
1914 struct ir_swizzle_mask {
1921 * Number of components in the swizzle.
1923 unsigned num_components:3;
1926 * Does the swizzle contain duplicate components?
1928 * L-value swizzles cannot contain duplicate components.
1930 unsigned has_duplicates:1;
1934 class ir_swizzle : public ir_rvalue {
1936 ir_swizzle(ir_rvalue *, unsigned x, unsigned y, unsigned z, unsigned w,
1939 ir_swizzle(ir_rvalue *val, const unsigned *components, unsigned count);
1941 ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask);
1943 virtual ir_swizzle *clone(void *mem_ctx, struct hash_table *) const;
1945 virtual ir_constant *constant_expression_value(void *mem_ctx,
1946 struct hash_table *variable_context = NULL);
1949 * Construct an ir_swizzle from the textual representation. Can fail.
1951 static ir_swizzle *create(ir_rvalue *, const char *, unsigned vector_length);
1953 virtual void accept(ir_visitor *v)
1958 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1960 virtual bool equals(const ir_instruction *ir,
1961 enum ir_node_type ignore = ir_type_unset) const;
1963 bool is_lvalue(const struct _mesa_glsl_parse_state *state) const
1965 return val->is_lvalue(state) && !mask.has_duplicates;
1969 * Get the variable that is ultimately referenced by an r-value
1971 virtual ir_variable *variable_referenced() const;
1974 ir_swizzle_mask mask;
1978 * Initialize the mask component of a swizzle
1980 * This is used by the \c ir_swizzle constructors.
1982 void init_mask(const unsigned *components, unsigned count);
1986 class ir_dereference : public ir_rvalue {
1988 virtual ir_dereference *clone(void *mem_ctx, struct hash_table *) const = 0;
1990 bool is_lvalue(const struct _mesa_glsl_parse_state *state) const;
1993 * Get the variable that is ultimately referenced by an r-value
1995 virtual ir_variable *variable_referenced() const = 0;
1998 ir_dereference(enum ir_node_type t)
2005 class ir_dereference_variable : public ir_dereference {
2007 ir_dereference_variable(ir_variable *var);
2009 virtual ir_dereference_variable *clone(void *mem_ctx,
2010 struct hash_table *) const;
2012 virtual ir_constant *constant_expression_value(void *mem_ctx,
2013 struct hash_table *variable_context = NULL);
2015 virtual bool equals(const ir_instruction *ir,
2016 enum ir_node_type ignore = ir_type_unset) const;
2019 * Get the variable that is ultimately referenced by an r-value
2021 virtual ir_variable *variable_referenced() const
2026 virtual ir_variable *whole_variable_referenced()
2028 /* ir_dereference_variable objects always dereference the entire
2029 * variable. However, if this dereference is dereferenced by anything
2030 * else, the complete deferefernce chain is not a whole-variable
2031 * dereference. This method should only be called on the top most
2032 * ir_rvalue in a dereference chain.
2037 virtual void accept(ir_visitor *v)
2042 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2045 * Object being dereferenced.
2051 class ir_dereference_array : public ir_dereference {
2053 ir_dereference_array(ir_rvalue *value, ir_rvalue *array_index);
2055 ir_dereference_array(ir_variable *var, ir_rvalue *array_index);
2057 virtual ir_dereference_array *clone(void *mem_ctx,
2058 struct hash_table *) const;
2060 virtual ir_constant *constant_expression_value(void *mem_ctx,
2061 struct hash_table *variable_context = NULL);
2063 virtual bool equals(const ir_instruction *ir,
2064 enum ir_node_type ignore = ir_type_unset) const;
2067 * Get the variable that is ultimately referenced by an r-value
2069 virtual ir_variable *variable_referenced() const
2071 return this->array->variable_referenced();
2074 virtual void accept(ir_visitor *v)
2079 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2082 ir_rvalue *array_index;
2085 void set_array(ir_rvalue *value);
2089 class ir_dereference_record : public ir_dereference {
2091 ir_dereference_record(ir_rvalue *value, const char *field);
2093 ir_dereference_record(ir_variable *var, const char *field);
2095 virtual ir_dereference_record *clone(void *mem_ctx,
2096 struct hash_table *) const;
2098 virtual ir_constant *constant_expression_value(void *mem_ctx,
2099 struct hash_table *variable_context = NULL);
2102 * Get the variable that is ultimately referenced by an r-value
2104 virtual ir_variable *variable_referenced() const
2106 return this->record->variable_referenced();
2109 virtual void accept(ir_visitor *v)
2114 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2122 * Data stored in an ir_constant
2124 union ir_constant_data {
2135 class ir_constant : public ir_rvalue {
2137 ir_constant(const struct glsl_type *type, const ir_constant_data *data);
2138 ir_constant(bool b, unsigned vector_elements=1);
2139 ir_constant(unsigned int u, unsigned vector_elements=1);
2140 ir_constant(int i, unsigned vector_elements=1);
2141 ir_constant(float f, unsigned vector_elements=1);
2142 ir_constant(double d, unsigned vector_elements=1);
2143 ir_constant(uint64_t u64, unsigned vector_elements=1);
2144 ir_constant(int64_t i64, unsigned vector_elements=1);
2147 * Construct an ir_constant from a list of ir_constant values
2149 ir_constant(const struct glsl_type *type, exec_list *values);
2152 * Construct an ir_constant from a scalar component of another ir_constant
2154 * The new \c ir_constant inherits the type of the component from the
2158 * In the case of a matrix constant, the new constant is a scalar, \b not
2161 ir_constant(const ir_constant *c, unsigned i);
2164 * Return a new ir_constant of the specified type containing all zeros.
2166 static ir_constant *zero(void *mem_ctx, const glsl_type *type);
2168 virtual ir_constant *clone(void *mem_ctx, struct hash_table *) const;
2170 virtual ir_constant *constant_expression_value(void *mem_ctx,
2171 struct hash_table *variable_context = NULL);
2173 virtual void accept(ir_visitor *v)
2178 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2180 virtual bool equals(const ir_instruction *ir,
2181 enum ir_node_type ignore = ir_type_unset) const;
2184 * Get a particular component of a constant as a specific type
2186 * This is useful, for example, to get a value from an integer constant
2187 * as a float or bool. This appears frequently when constructors are
2188 * called with all constant parameters.
2191 bool get_bool_component(unsigned i) const;
2192 float get_float_component(unsigned i) const;
2193 double get_double_component(unsigned i) const;
2194 int get_int_component(unsigned i) const;
2195 unsigned get_uint_component(unsigned i) const;
2196 int64_t get_int64_component(unsigned i) const;
2197 uint64_t get_uint64_component(unsigned i) const;
2200 ir_constant *get_array_element(unsigned i) const;
2202 ir_constant *get_record_field(int idx);
2205 * Copy the values on another constant at a given offset.
2207 * The offset is ignored for array or struct copies, it's only for
2208 * scalars or vectors into vectors or matrices.
2210 * With identical types on both sides and zero offset it's clone()
2211 * without creating a new object.
2214 void copy_offset(ir_constant *src, int offset);
2217 * Copy the values on another constant at a given offset and
2218 * following an assign-like mask.
2220 * The mask is ignored for scalars.
2222 * Note that this function only handles what assign can handle,
2223 * i.e. at most a vector as source and a column of a matrix as
2227 void copy_masked_offset(ir_constant *src, int offset, unsigned int mask);
2230 * Determine whether a constant has the same value as another constant
2232 * \sa ir_constant::is_zero, ir_constant::is_one,
2233 * ir_constant::is_negative_one
2235 bool has_value(const ir_constant *) const;
2238 * Return true if this ir_constant represents the given value.
2240 * For vectors, this checks that each component is the given value.
2242 virtual bool is_value(float f, int i) const;
2243 virtual bool is_zero() const;
2244 virtual bool is_one() const;
2245 virtual bool is_negative_one() const;
2248 * Return true for constants that could be stored as 16-bit unsigned values.
2250 * Note that this will return true even for signed integer ir_constants, as
2251 * long as the value is non-negative and fits in 16-bits.
2253 virtual bool is_uint16_constant() const;
2256 * Value of the constant.
2258 * The field used to back the values supplied by the constant is determined
2259 * by the type associated with the \c ir_instruction. Constants may be
2260 * scalars, vectors, or matrices.
2262 union ir_constant_data value;
2264 /* Array elements and structure fields */
2265 ir_constant **const_elements;
2269 * Parameterless constructor only used by the clone method
2275 * IR instruction to emit a vertex in a geometry shader.
2277 class ir_emit_vertex : public ir_instruction {
2279 ir_emit_vertex(ir_rvalue *stream)
2280 : ir_instruction(ir_type_emit_vertex),
2286 virtual void accept(ir_visitor *v)
2291 virtual ir_emit_vertex *clone(void *mem_ctx, struct hash_table *ht) const
2293 return new(mem_ctx) ir_emit_vertex(this->stream->clone(mem_ctx, ht));
2296 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2298 int stream_id() const
2300 return stream->as_constant()->value.i[0];
2307 * IR instruction to complete the current primitive and start a new one in a
2310 class ir_end_primitive : public ir_instruction {
2312 ir_end_primitive(ir_rvalue *stream)
2313 : ir_instruction(ir_type_end_primitive),
2319 virtual void accept(ir_visitor *v)
2324 virtual ir_end_primitive *clone(void *mem_ctx, struct hash_table *ht) const
2326 return new(mem_ctx) ir_end_primitive(this->stream->clone(mem_ctx, ht));
2329 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2331 int stream_id() const
2333 return stream->as_constant()->value.i[0];
2340 * IR instruction for tessellation control and compute shader barrier.
2342 class ir_barrier : public ir_instruction {
2345 : ir_instruction(ir_type_barrier)
2349 virtual void accept(ir_visitor *v)
2354 virtual ir_barrier *clone(void *mem_ctx, struct hash_table *) const
2356 return new(mem_ctx) ir_barrier();
2359 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2365 * Apply a visitor to each IR node in a list
2368 visit_exec_list(exec_list *list, ir_visitor *visitor);
2371 * Validate invariants on each IR node in a list
2373 void validate_ir_tree(exec_list *instructions);
2375 struct _mesa_glsl_parse_state;
2376 struct gl_shader_program;
2379 * Detect whether an unlinked shader contains static recursion
2381 * If the list of instructions is determined to contain static recursion,
2382 * \c _mesa_glsl_error will be called to emit error messages for each function
2383 * that is in the recursion cycle.
2386 detect_recursion_unlinked(struct _mesa_glsl_parse_state *state,
2387 exec_list *instructions);
2390 * Detect whether a linked shader contains static recursion
2392 * If the list of instructions is determined to contain static recursion,
2393 * \c link_error_printf will be called to emit error messages for each function
2394 * that is in the recursion cycle. In addition,
2395 * \c gl_shader_program::LinkStatus will be set to false.
2398 detect_recursion_linked(struct gl_shader_program *prog,
2399 exec_list *instructions);
2402 * Make a clone of each IR instruction in a list
2404 * \param in List of IR instructions that are to be cloned
2405 * \param out List to hold the cloned instructions
2408 clone_ir_list(void *mem_ctx, exec_list *out, const exec_list *in);
2411 _mesa_glsl_initialize_variables(exec_list *instructions,
2412 struct _mesa_glsl_parse_state *state);
2415 reparent_ir(exec_list *list, void *mem_ctx);
2418 do_set_program_inouts(exec_list *instructions, struct gl_program *prog,
2419 gl_shader_stage shader_stage);
2422 prototype_string(const glsl_type *return_type, const char *name,
2423 exec_list *parameters);
2426 mode_string(const ir_variable *var);
2429 * Built-in / reserved GL variables names start with "gl_"
2432 is_gl_identifier(const char *s)
2434 return s && s[0] == 'g' && s[1] == 'l' && s[2] == '_';
2438 #endif /* __cplusplus */
2440 extern void _mesa_print_ir(FILE *f, struct exec_list *instructions,
2441 struct _mesa_glsl_parse_state *state);
2444 fprint_ir(FILE *f, const void *instruction);
2446 extern const struct gl_builtin_uniform_desc *
2447 _mesa_glsl_get_builtin_uniform_desc(const char *name);
2454 vertices_per_prim(GLenum prim);