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.
32 #include "util/ralloc.h"
33 #include "compiler/glsl_types.h"
35 #include "ir_visitor.h"
36 #include "ir_hierarchical_visitor.h"
37 #include "main/mtypes.h"
42 * \defgroup IR Intermediate representation nodes
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.
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.
62 ir_type_dereference_array,
63 ir_type_dereference_record,
64 ir_type_dereference_variable,
73 ir_type_function_signature,
80 ir_type_end_primitive,
82 ir_type_max, /**< maximum ir_type enum number, for validation */
83 ir_type_unset = ir_type_max
88 * Base class of all IR instructions
90 class ir_instruction : public exec_node {
92 enum ir_node_type ir_type;
95 * GCC 4.7+ and clang warn when deleting an ir_instruction unless
96 * there's a virtual destructor present. Because we almost
97 * universally use ralloc for our memory management of
98 * ir_instructions, the destructor doesn't need to do any work.
100 virtual ~ir_instruction()
104 /** ir_print_visitor helper for debugging. */
105 void print(void) const;
106 void fprint(FILE *f) const;
108 virtual void accept(ir_visitor *) = 0;
109 virtual ir_visitor_status accept(ir_hierarchical_visitor *) = 0;
110 virtual ir_instruction *clone(void *mem_ctx,
111 struct hash_table *ht) const = 0;
113 bool is_rvalue() const
115 return ir_type == ir_type_dereference_array ||
116 ir_type == ir_type_dereference_record ||
117 ir_type == ir_type_dereference_variable ||
118 ir_type == ir_type_constant ||
119 ir_type == ir_type_expression ||
120 ir_type == ir_type_swizzle ||
121 ir_type == ir_type_texture;
124 bool is_dereference() const
126 return ir_type == ir_type_dereference_array ||
127 ir_type == ir_type_dereference_record ||
128 ir_type == ir_type_dereference_variable;
133 return ir_type == ir_type_loop_jump ||
134 ir_type == ir_type_return ||
135 ir_type == ir_type_discard;
139 * \name IR instruction downcast functions
141 * These functions either cast the object to a derived class or return
142 * \c NULL if the object's type does not match the specified derived class.
143 * Additional downcast functions will be added as needed.
146 #define AS_BASE(TYPE) \
147 class ir_##TYPE *as_##TYPE() \
149 assume(this != NULL); \
150 return is_##TYPE() ? (ir_##TYPE *) this : NULL; \
152 const class ir_##TYPE *as_##TYPE() const \
154 assume(this != NULL); \
155 return is_##TYPE() ? (ir_##TYPE *) this : NULL; \
163 #define AS_CHILD(TYPE) \
164 class ir_##TYPE * as_##TYPE() \
166 assume(this != NULL); \
167 return ir_type == ir_type_##TYPE ? (ir_##TYPE *) this : NULL; \
169 const class ir_##TYPE * as_##TYPE() const \
171 assume(this != NULL); \
172 return ir_type == ir_type_##TYPE ? (const ir_##TYPE *) this : NULL; \
176 AS_CHILD(dereference_array)
177 AS_CHILD(dereference_variable)
178 AS_CHILD(dereference_record)
193 * IR equality method: Return true if the referenced instruction would
194 * return the same value as this one.
196 * This intended to be used for CSE and algebraic optimizations, on rvalues
197 * in particular. No support for other instruction types (assignments,
198 * jumps, calls, etc.) is planned.
200 virtual bool equals(const ir_instruction *ir,
201 enum ir_node_type ignore = ir_type_unset) const;
204 ir_instruction(enum ir_node_type t)
212 assert(!"Should not get here.");
218 * The base class for all "values"/expression trees.
220 class ir_rvalue : public ir_instruction {
222 const struct glsl_type *type;
224 virtual ir_rvalue *clone(void *mem_ctx, struct hash_table *) const;
226 virtual void accept(ir_visitor *v)
231 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
233 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
235 ir_rvalue *as_rvalue_to_saturate();
237 virtual bool is_lvalue() const
243 * Get the variable that is ultimately referenced by an r-value
245 virtual ir_variable *variable_referenced() const
252 * If an r-value is a reference to a whole variable, get that variable
255 * Pointer to a variable that is completely dereferenced by the r-value. If
256 * the r-value is not a dereference or the dereference does not access the
257 * entire variable (i.e., it's just one array element, struct field), \c NULL
260 virtual ir_variable *whole_variable_referenced()
266 * Determine if an r-value has the value zero
268 * The base implementation of this function always returns \c false. The
269 * \c ir_constant class over-rides this function to return \c true \b only
270 * for vector and scalar types that have all elements set to the value
271 * zero (or \c false for booleans).
273 * \sa ir_constant::has_value, ir_rvalue::is_one, ir_rvalue::is_negative_one
275 virtual bool is_zero() const;
278 * Determine if an r-value has the value one
280 * The base implementation of this function always returns \c false. The
281 * \c ir_constant class over-rides this function to return \c true \b only
282 * for vector and scalar types that have all elements set to the value
283 * one (or \c true for booleans).
285 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_negative_one
287 virtual bool is_one() const;
290 * Determine if an r-value has the value negative one
292 * The base implementation of this function always returns \c false. The
293 * \c ir_constant class over-rides this function to return \c true \b only
294 * for vector and scalar types that have all elements set to the value
295 * negative one. For boolean types, the result is always \c false.
297 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one
299 virtual bool is_negative_one() const;
302 * Determine if an r-value is an unsigned integer constant which can be
305 * \sa ir_constant::is_uint16_constant.
307 virtual bool is_uint16_constant() const { return false; }
310 * Return a generic value of error_type.
312 * Allocation will be performed with 'mem_ctx' as ralloc owner.
314 static ir_rvalue *error_value(void *mem_ctx);
317 ir_rvalue(enum ir_node_type t);
322 * Variable storage classes
324 enum ir_variable_mode {
325 ir_var_auto = 0, /**< Function local variables and globals. */
326 ir_var_uniform, /**< Variable declared as a uniform. */
327 ir_var_shader_storage, /**< Variable declared as an ssbo. */
328 ir_var_shader_shared, /**< Variable declared as shared. */
333 ir_var_function_inout,
334 ir_var_const_in, /**< "in" param that must be a constant expression */
335 ir_var_system_value, /**< Ex: front-face, instance-id, etc. */
336 ir_var_temporary, /**< Temporary variable generated during compilation. */
337 ir_var_mode_count /**< Number of variable modes */
341 * Enum keeping track of how a variable was declared. For error checking of
342 * the gl_PerVertex redeclaration rules.
344 enum ir_var_declaration_type {
346 * Normal declaration (for most variables, this means an explicit
347 * declaration. Exception: temporaries are always implicitly declared, but
348 * they still use ir_var_declared_normally).
350 * Note: an ir_variable that represents a named interface block uses
351 * ir_var_declared_normally.
353 ir_var_declared_normally = 0,
356 * Variable was explicitly declared (or re-declared) in an unnamed
359 ir_var_declared_in_block,
362 * Variable is an implicitly declared built-in that has not been explicitly
363 * re-declared by the shader.
365 ir_var_declared_implicitly,
368 * Variable is implicitly generated by the compiler and should not be
369 * visible via the API.
375 * \brief Layout qualifiers for gl_FragDepth.
377 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
378 * with a layout qualifier.
380 enum ir_depth_layout {
381 ir_depth_layout_none, /**< No depth layout is specified. */
383 ir_depth_layout_greater,
384 ir_depth_layout_less,
385 ir_depth_layout_unchanged
389 * \brief Convert depth layout qualifier to string.
392 depth_layout_string(ir_depth_layout layout);
395 * Description of built-in state associated with a uniform
397 * \sa ir_variable::state_slots
399 struct ir_state_slot {
406 * Get the string value for an interpolation qualifier
408 * \return The string that would be used in a shader to specify \c
409 * mode will be returned.
411 * This function is used to generate error messages of the form "shader
412 * uses %s interpolation qualifier", so in the case where there is no
413 * interpolation qualifier, it returns "no".
415 * This function should only be used on a shader input or output variable.
417 const char *interpolation_string(unsigned interpolation);
420 class ir_variable : public ir_instruction {
422 ir_variable(const struct glsl_type *, const char *, ir_variable_mode);
424 virtual ir_variable *clone(void *mem_ctx, struct hash_table *ht) const;
426 virtual void accept(ir_visitor *v)
431 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
435 * Determine whether or not a variable is part of a uniform or
436 * shader storage block.
438 inline bool is_in_buffer_block() const
440 return (this->data.mode == ir_var_uniform ||
441 this->data.mode == ir_var_shader_storage) &&
442 this->interface_type != NULL;
446 * Determine whether or not a variable is part of a shader storage block.
448 inline bool is_in_shader_storage_block() const
450 return this->data.mode == ir_var_shader_storage &&
451 this->interface_type != NULL;
455 * Determine whether or not a variable is the declaration of an interface
458 * For the first declaration below, there will be an \c ir_variable named
459 * "instance" whose type and whose instance_type will be the same
460 * \cglsl_type. For the second declaration, there will be an \c ir_variable
461 * named "f" whose type is float and whose instance_type is B2.
463 * "instance" is an interface instance variable, but "f" is not.
473 inline bool is_interface_instance() const
475 return this->type->without_array() == this->interface_type;
479 * Set this->interface_type on a newly created variable.
481 void init_interface_type(const struct glsl_type *type)
483 assert(this->interface_type == NULL);
484 this->interface_type = type;
485 if (this->is_interface_instance()) {
486 this->u.max_ifc_array_access =
487 ralloc_array(this, int, type->length);
488 for (unsigned i = 0; i < type->length; i++) {
489 this->u.max_ifc_array_access[i] = -1;
495 * Change this->interface_type on a variable that previously had a
496 * different, but compatible, interface_type. This is used during linking
497 * to set the size of arrays in interface blocks.
499 void change_interface_type(const struct glsl_type *type)
501 if (this->u.max_ifc_array_access != NULL) {
502 /* max_ifc_array_access has already been allocated, so make sure the
503 * new interface has the same number of fields as the old one.
505 assert(this->interface_type->length == type->length);
507 this->interface_type = type;
511 * Change this->interface_type on a variable that previously had a
512 * different, and incompatible, interface_type. This is used during
513 * compilation to handle redeclaration of the built-in gl_PerVertex
516 void reinit_interface_type(const struct glsl_type *type)
518 if (this->u.max_ifc_array_access != NULL) {
520 /* Redeclaring gl_PerVertex is only allowed if none of the built-ins
521 * it defines have been accessed yet; so it's safe to throw away the
522 * old max_ifc_array_access pointer, since all of its values are
525 for (unsigned i = 0; i < this->interface_type->length; i++)
526 assert(this->u.max_ifc_array_access[i] == -1);
528 ralloc_free(this->u.max_ifc_array_access);
529 this->u.max_ifc_array_access = NULL;
531 this->interface_type = NULL;
532 init_interface_type(type);
535 const glsl_type *get_interface_type() const
537 return this->interface_type;
541 * Get the max_ifc_array_access pointer
543 * A "set" function is not needed because the array is dynmically allocated
546 inline int *get_max_ifc_array_access()
548 assert(this->data._num_state_slots == 0);
549 return this->u.max_ifc_array_access;
552 inline unsigned get_num_state_slots() const
554 assert(!this->is_interface_instance()
555 || this->data._num_state_slots == 0);
556 return this->data._num_state_slots;
559 inline void set_num_state_slots(unsigned n)
561 assert(!this->is_interface_instance()
563 this->data._num_state_slots = n;
566 inline ir_state_slot *get_state_slots()
568 return this->is_interface_instance() ? NULL : this->u.state_slots;
571 inline const ir_state_slot *get_state_slots() const
573 return this->is_interface_instance() ? NULL : this->u.state_slots;
576 inline ir_state_slot *allocate_state_slots(unsigned n)
578 assert(!this->is_interface_instance());
580 this->u.state_slots = ralloc_array(this, ir_state_slot, n);
581 this->data._num_state_slots = 0;
583 if (this->u.state_slots != NULL)
584 this->data._num_state_slots = n;
586 return this->u.state_slots;
589 inline bool is_name_ralloced() const
591 return this->name != ir_variable::tmp_name;
595 * Enable emitting extension warnings for this variable
597 void enable_extension_warning(const char *extension);
600 * Get the extension warning string for this variable
602 * If warnings are not enabled, \c NULL is returned.
604 const char *get_extension_warning() const;
607 * Declared type of the variable
609 const struct glsl_type *type;
612 * Declared name of the variable
616 struct ir_variable_data {
619 * Is the variable read-only?
621 * This is set for variables declared as \c const, shader inputs,
624 unsigned read_only:1;
628 unsigned invariant:1;
632 * Has this variable been used for reading or writing?
634 * Several GLSL semantic checks require knowledge of whether or not a
635 * variable has been used. For example, it is an error to redeclare a
636 * variable as invariant after it has been used.
638 * This is only maintained in the ast_to_hir.cpp path, not in
639 * Mesa's fixed function or ARB program paths.
644 * Has this variable been statically assigned?
646 * This answers whether the variable was assigned in any path of
647 * the shader during ast_to_hir. This doesn't answer whether it is
648 * still written after dead code removal, nor is it maintained in
649 * non-ast_to_hir.cpp (GLSL parsing) paths.
654 * When separate shader programs are enabled, only input/outputs between
655 * the stages of a multi-stage separate program can be safely removed
656 * from the shader interface. Other input/outputs must remains active.
658 unsigned always_active_io:1;
661 * Enum indicating how the variable was declared. See
662 * ir_var_declaration_type.
664 * This is used to detect certain kinds of illegal variable redeclarations.
666 unsigned how_declared:2;
669 * Storage class of the variable.
671 * \sa ir_variable_mode
676 * Interpolation mode for shader inputs / outputs
678 * \sa ir_variable_interpolation
680 unsigned interpolation:2;
683 * \name ARB_fragment_coord_conventions
686 unsigned origin_upper_left:1;
687 unsigned pixel_center_integer:1;
691 * Was the location explicitly set in the shader?
693 * If the location is explicitly set in the shader, it \b cannot be changed
694 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
697 unsigned explicit_location:1;
698 unsigned explicit_index:1;
701 * Was an initial binding explicitly set in the shader?
703 * If so, constant_value contains an integer ir_constant representing the
704 * initial binding point.
706 unsigned explicit_binding:1;
709 * Was an initial component explicitly set in the shader?
711 unsigned explicit_component:1;
714 * Does this variable have an initializer?
716 * This is used by the linker to cross-validiate initializers of global
719 unsigned has_initializer:1;
722 * Is this variable a generic output or input that has not yet been matched
723 * up to a variable in another stage of the pipeline?
725 * This is used by the linker as scratch storage while assigning locations
726 * to generic inputs and outputs.
728 unsigned is_unmatched_generic_inout:1;
731 * Is this varying used only by transform feedback?
733 * This is used by the linker to decide if its safe to pack the varying.
735 unsigned is_xfb_only:1;
738 * Was a transfor feedback buffer set in the shader?
740 unsigned explicit_xfb_buffer:1;
743 * Was a transfor feedback offset set in the shader?
745 unsigned explicit_xfb_offset:1;
748 * Was a transfor feedback stride set in the shader?
750 unsigned explicit_xfb_stride:1;
753 * If non-zero, then this variable may be packed along with other variables
754 * into a single varying slot, so this offset should be applied when
755 * accessing components. For example, an offset of 1 means that the x
756 * component of this variable is actually stored in component y of the
757 * location specified by \c location.
759 unsigned location_frac:2;
762 * Layout of the matrix. Uses glsl_matrix_layout values.
764 unsigned matrix_layout:2;
767 * Non-zero if this variable was created by lowering a named interface
770 unsigned from_named_ifc_block:1;
773 * Non-zero if the variable must be a shader input. This is useful for
774 * constraints on function parameters.
776 unsigned must_be_shader_input:1;
779 * Output index for dual source blending.
782 * The GLSL spec only allows the values 0 or 1 for the index in \b dual
788 * Precision qualifier.
790 * In desktop GLSL we do not care about precision qualifiers at all, in
791 * fact, the spec says that precision qualifiers are ignored.
793 * To make things easy, we make it so that this field is always
794 * GLSL_PRECISION_NONE on desktop shaders. This way all the variables
795 * have the same precision value and the checks we add in the compiler
796 * for this field will never break a desktop shader compile.
798 unsigned precision:2;
801 * \brief Layout qualifier for gl_FragDepth.
803 * This is not equal to \c ir_depth_layout_none if and only if this
804 * variable is \c gl_FragDepth and a layout qualifier is specified.
806 ir_depth_layout depth_layout:3;
809 * ARB_shader_image_load_store qualifiers.
811 unsigned image_read_only:1; /**< "readonly" qualifier. */
812 unsigned image_write_only:1; /**< "writeonly" qualifier. */
813 unsigned image_coherent:1;
814 unsigned image_volatile:1;
815 unsigned image_restrict:1;
818 * ARB_shader_storage_buffer_object
820 unsigned from_ssbo_unsized_array:1; /**< unsized array buffer variable. */
822 unsigned implicit_sized_array:1;
824 * Emit a warning if this variable is accessed.
827 uint8_t warn_extension_index;
830 /** Image internal format if specified explicitly, otherwise GL_NONE. */
831 uint16_t image_format;
835 * Number of state slots used
838 * This could be stored in as few as 7-bits, if necessary. If it is made
839 * smaller, add an assertion to \c ir_variable::allocate_state_slots to
842 uint16_t _num_state_slots;
846 * Initial binding point for a sampler, atomic, or UBO.
848 * For array types, this represents the binding point for the first element.
853 * Storage location of the base of this variable
855 * The precise meaning of this field depends on the nature of the variable.
857 * - Vertex shader input: one of the values from \c gl_vert_attrib.
858 * - Vertex shader output: one of the values from \c gl_varying_slot.
859 * - Geometry shader input: one of the values from \c gl_varying_slot.
860 * - Geometry shader output: one of the values from \c gl_varying_slot.
861 * - Fragment shader input: one of the values from \c gl_varying_slot.
862 * - Fragment shader output: one of the values from \c gl_frag_result.
863 * - Uniforms: Per-stage uniform slot number for default uniform block.
864 * - Uniforms: Index within the uniform block definition for UBO members.
865 * - Non-UBO Uniforms: explicit location until linking then reused to
866 * store uniform slot number.
867 * - Other: This field is not currently used.
869 * If the variable is a uniform, shader input, or shader output, and the
870 * slot has not been assigned, the value will be -1.
875 * for glsl->tgsi/mesa IR we need to store the index into the
876 * parameters for uniforms, initially the code overloaded location
877 * but this causes problems with indirect samplers and AoA.
878 * This is assigned in _mesa_generate_parameters_list_for_uniforms.
883 * Vertex stream output identifier.
888 * Atomic, transform feedback or block member offset.
893 * Highest element accessed with a constant expression array index
895 * Not used for non-array variables. -1 is never accessed.
897 int max_array_access;
900 * Transform feedback buffer.
905 * Transform feedback stride.
910 * Allow (only) ir_variable direct access private members.
912 friend class ir_variable;
916 * Value assigned in the initializer of a variable declared "const"
918 ir_constant *constant_value;
921 * Constant expression assigned in the initializer of the variable
924 * This field and \c ::constant_value are distinct. Even if the two fields
925 * refer to constants with the same value, they must point to separate
928 ir_constant *constant_initializer;
931 static const char *const warn_extension_table[];
935 * For variables which satisfy the is_interface_instance() predicate,
936 * this points to an array of integers such that if the ith member of
937 * the interface block is an array, max_ifc_array_access[i] is the
938 * maximum array element of that member that has been accessed. If the
939 * ith member of the interface block is not an array,
940 * max_ifc_array_access[i] is unused.
942 * For variables whose type is not an interface block, this pointer is
945 int *max_ifc_array_access;
948 * Built-in state that backs this uniform
950 * Once set at variable creation, \c state_slots must remain invariant.
952 * If the variable is not a uniform, \c _num_state_slots will be zero
953 * and \c state_slots will be \c NULL.
955 ir_state_slot *state_slots;
959 * For variables that are in an interface block or are an instance of an
960 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
962 * \sa ir_variable::location
964 const glsl_type *interface_type;
967 * Name used for anonymous compiler temporaries
969 static const char tmp_name[];
973 * Should the construct keep names for ir_var_temporary variables?
975 * When this global is false, names passed to the constructor for
976 * \c ir_var_temporary variables will be dropped. Instead, the variable will
977 * be named "compiler_temp". This name will be in static storage.
980 * \b NEVER change the mode of an \c ir_var_temporary.
983 * This variable is \b not thread-safe. It is global, \b not
984 * per-context. It begins life false. A context can, at some point, make
985 * it true. From that point on, it will be true forever. This should be
986 * okay since it will only be set true while debugging.
988 static bool temporaries_allocate_names;
992 * A function that returns whether a built-in function is available in the
993 * current shading language (based on version, ES or desktop, and extensions).
995 typedef bool (*builtin_available_predicate)(const _mesa_glsl_parse_state *);
999 * The representation of a function instance; may be the full definition or
1000 * simply a prototype.
1002 class ir_function_signature : public ir_instruction {
1003 /* An ir_function_signature will be part of the list of signatures in
1007 ir_function_signature(const glsl_type *return_type,
1008 builtin_available_predicate builtin_avail = NULL);
1010 virtual ir_function_signature *clone(void *mem_ctx,
1011 struct hash_table *ht) const;
1012 ir_function_signature *clone_prototype(void *mem_ctx,
1013 struct hash_table *ht) const;
1015 virtual void accept(ir_visitor *v)
1020 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1023 * Attempt to evaluate this function as a constant expression,
1024 * given a list of the actual parameters and the variable context.
1025 * Returns NULL for non-built-ins.
1027 ir_constant *constant_expression_value(exec_list *actual_parameters, struct hash_table *variable_context);
1030 * Get the name of the function for which this is a signature
1032 const char *function_name() const;
1035 * Get a handle to the function for which this is a signature
1037 * There is no setter function, this function returns a \c const pointer,
1038 * and \c ir_function_signature::_function is private for a reason. The
1039 * only way to make a connection between a function and function signature
1040 * is via \c ir_function::add_signature. This helps ensure that certain
1041 * invariants (i.e., a function signature is in the list of signatures for
1042 * its \c _function) are met.
1044 * \sa ir_function::add_signature
1046 inline const class ir_function *function() const
1048 return this->_function;
1052 * Check whether the qualifiers match between this signature's parameters
1053 * and the supplied parameter list. If not, returns the name of the first
1054 * parameter with mismatched qualifiers (for use in error messages).
1056 const char *qualifiers_match(exec_list *params);
1059 * Replace the current parameter list with the given one. This is useful
1060 * if the current information came from a prototype, and either has invalid
1061 * or missing parameter names.
1063 void replace_parameters(exec_list *new_params);
1066 * Function return type.
1068 * \note This discards the optional precision qualifier.
1070 const struct glsl_type *return_type;
1073 * List of ir_variable of function parameters.
1075 * This represents the storage. The paramaters passed in a particular
1076 * call will be in ir_call::actual_paramaters.
1078 struct exec_list parameters;
1080 /** Whether or not this function has a body (which may be empty). */
1081 unsigned is_defined:1;
1083 /** Whether or not this function signature is a built-in. */
1084 bool is_builtin() const;
1087 * Whether or not this function is an intrinsic to be implemented
1092 /** Whether or not a built-in is available for this shader. */
1093 bool is_builtin_available(const _mesa_glsl_parse_state *state) const;
1095 /** Body of instructions in the function. */
1096 struct exec_list body;
1100 * A function pointer to a predicate that answers whether a built-in
1101 * function is available in the current shader. NULL if not a built-in.
1103 builtin_available_predicate builtin_avail;
1105 /** Function of which this signature is one overload. */
1106 class ir_function *_function;
1108 /** Function signature of which this one is a prototype clone */
1109 const ir_function_signature *origin;
1111 friend class ir_function;
1114 * Helper function to run a list of instructions for constant
1115 * expression evaluation.
1117 * The hash table represents the values of the visible variables.
1118 * There are no scoping issues because the table is indexed on
1119 * ir_variable pointers, not variable names.
1121 * Returns false if the expression is not constant, true otherwise,
1122 * and the value in *result if result is non-NULL.
1124 bool constant_expression_evaluate_expression_list(const struct exec_list &body,
1125 struct hash_table *variable_context,
1126 ir_constant **result);
1131 * Header for tracking multiple overloaded functions with the same name.
1132 * Contains a list of ir_function_signatures representing each of the
1135 class ir_function : public ir_instruction {
1137 ir_function(const char *name);
1139 virtual ir_function *clone(void *mem_ctx, struct hash_table *ht) const;
1141 virtual void accept(ir_visitor *v)
1146 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1148 void add_signature(ir_function_signature *sig)
1150 sig->_function = this;
1151 this->signatures.push_tail(sig);
1155 * Find a signature that matches a set of actual parameters, taking implicit
1156 * conversions into account. Also flags whether the match was exact.
1158 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
1159 const exec_list *actual_param,
1160 bool allow_builtins,
1161 bool *match_is_exact);
1164 * Find a signature that matches a set of actual parameters, taking implicit
1165 * conversions into account.
1167 ir_function_signature *matching_signature(_mesa_glsl_parse_state *state,
1168 const exec_list *actual_param,
1169 bool allow_builtins);
1172 * Find a signature that exactly matches a set of actual parameters without
1173 * any implicit type conversions.
1175 ir_function_signature *exact_matching_signature(_mesa_glsl_parse_state *state,
1176 const exec_list *actual_ps);
1179 * Name of the function.
1183 /** Whether or not this function has a signature that isn't a built-in. */
1184 bool has_user_signature();
1187 * List of ir_function_signature for each overloaded function with this name.
1189 struct exec_list signatures;
1192 * is this function a subroutine type declaration
1193 * e.g. subroutine void type1(float arg1);
1198 * is this function associated to a subroutine type
1199 * e.g. subroutine (type1, type2) function_name { function_body };
1200 * would have num_subroutine_types 2,
1201 * and pointers to the type1 and type2 types.
1203 int num_subroutine_types;
1204 const struct glsl_type **subroutine_types;
1206 int subroutine_index;
1209 inline const char *ir_function_signature::function_name() const
1211 return this->_function->name;
1217 * IR instruction representing high-level if-statements
1219 class ir_if : public ir_instruction {
1221 ir_if(ir_rvalue *condition)
1222 : ir_instruction(ir_type_if), condition(condition)
1226 virtual ir_if *clone(void *mem_ctx, struct hash_table *ht) const;
1228 virtual void accept(ir_visitor *v)
1233 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1235 ir_rvalue *condition;
1236 /** List of ir_instruction for the body of the then branch */
1237 exec_list then_instructions;
1238 /** List of ir_instruction for the body of the else branch */
1239 exec_list else_instructions;
1244 * IR instruction representing a high-level loop structure.
1246 class ir_loop : public ir_instruction {
1250 virtual ir_loop *clone(void *mem_ctx, struct hash_table *ht) const;
1252 virtual void accept(ir_visitor *v)
1257 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1259 /** List of ir_instruction that make up the body of the loop. */
1260 exec_list body_instructions;
1264 class ir_assignment : public ir_instruction {
1266 ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs, ir_rvalue *condition = NULL);
1269 * Construct an assignment with an explicit write mask
1272 * Since a write mask is supplied, the LHS must already be a bare
1273 * \c ir_dereference. The cannot be any swizzles in the LHS.
1275 ir_assignment(ir_dereference *lhs, ir_rvalue *rhs, ir_rvalue *condition,
1276 unsigned write_mask);
1278 virtual ir_assignment *clone(void *mem_ctx, struct hash_table *ht) const;
1280 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1282 virtual void accept(ir_visitor *v)
1287 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1290 * Get a whole variable written by an assignment
1292 * If the LHS of the assignment writes a whole variable, the variable is
1293 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1296 * - Assigning to a scalar
1297 * - Assigning to all components of a vector
1298 * - Whole array (or matrix) assignment
1299 * - Whole structure assignment
1301 ir_variable *whole_variable_written();
1304 * Set the LHS of an assignment
1306 void set_lhs(ir_rvalue *lhs);
1309 * Left-hand side of the assignment.
1311 * This should be treated as read only. If you need to set the LHS of an
1312 * assignment, use \c ir_assignment::set_lhs.
1314 ir_dereference *lhs;
1317 * Value being assigned
1322 * Optional condition for the assignment.
1324 ir_rvalue *condition;
1328 * Component mask written
1330 * For non-vector types in the LHS, this field will be zero. For vector
1331 * types, a bit will be set for each component that is written. Note that
1332 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1334 * A partially-set write mask means that each enabled channel gets
1335 * the value from a consecutive channel of the rhs. For example,
1336 * to write just .xyw of gl_FrontColor with color:
1338 * (assign (constant bool (1)) (xyw)
1339 * (var_ref gl_FragColor)
1340 * (swiz xyw (var_ref color)))
1342 unsigned write_mask:4;
1345 /* Update ir_expression::get_num_operands() and operator_strs when
1346 * updating this list.
1348 enum ir_expression_operation {
1357 ir_unop_exp, /**< Log base e on gentype */
1358 ir_unop_log, /**< Natural log on gentype */
1361 ir_unop_f2i, /**< Float-to-integer conversion. */
1362 ir_unop_f2u, /**< Float-to-unsigned conversion. */
1363 ir_unop_i2f, /**< Integer-to-float conversion. */
1364 ir_unop_f2b, /**< Float-to-boolean conversion */
1365 ir_unop_b2f, /**< Boolean-to-float conversion */
1366 ir_unop_i2b, /**< int-to-boolean conversion */
1367 ir_unop_b2i, /**< Boolean-to-int conversion */
1368 ir_unop_u2f, /**< Unsigned-to-float conversion. */
1369 ir_unop_i2u, /**< Integer-to-unsigned conversion. */
1370 ir_unop_u2i, /**< Unsigned-to-integer conversion. */
1371 ir_unop_d2f, /**< Double-to-float conversion. */
1372 ir_unop_f2d, /**< Float-to-double conversion. */
1373 ir_unop_d2i, /**< Double-to-integer conversion. */
1374 ir_unop_i2d, /**< Integer-to-double conversion. */
1375 ir_unop_d2u, /**< Double-to-unsigned conversion. */
1376 ir_unop_u2d, /**< Unsigned-to-double conversion. */
1377 ir_unop_d2b, /**< Double-to-boolean conversion. */
1378 ir_unop_bitcast_i2f, /**< Bit-identical int-to-float "conversion" */
1379 ir_unop_bitcast_f2i, /**< Bit-identical float-to-int "conversion" */
1380 ir_unop_bitcast_u2f, /**< Bit-identical uint-to-float "conversion" */
1381 ir_unop_bitcast_f2u, /**< Bit-identical float-to-uint "conversion" */
1384 * \name Unary floating-point rounding operations.
1395 * \name Trigonometric operations.
1403 * \name Partial derivatives.
1407 ir_unop_dFdx_coarse,
1410 ir_unop_dFdy_coarse,
1415 * \name Floating point pack and unpack operations.
1418 ir_unop_pack_snorm_2x16,
1419 ir_unop_pack_snorm_4x8,
1420 ir_unop_pack_unorm_2x16,
1421 ir_unop_pack_unorm_4x8,
1422 ir_unop_pack_half_2x16,
1423 ir_unop_unpack_snorm_2x16,
1424 ir_unop_unpack_snorm_4x8,
1425 ir_unop_unpack_unorm_2x16,
1426 ir_unop_unpack_unorm_4x8,
1427 ir_unop_unpack_half_2x16,
1431 * \name Bit operations, part of ARB_gpu_shader5.
1434 ir_unop_bitfield_reverse,
1443 * \name Double packing, part of ARB_gpu_shader_fp64.
1446 ir_unop_pack_double_2x32,
1447 ir_unop_unpack_double_2x32,
1455 ir_unop_subroutine_to_int,
1457 * Interpolate fs input at centroid
1459 * operand0 is the fs input.
1461 ir_unop_interpolate_at_centroid,
1464 * Ask the driver for the total size of a buffer block.
1466 * operand0 is the ir_constant buffer block index in the linked shader.
1468 ir_unop_get_buffer_size,
1471 * Calculate length of an unsized array inside a buffer block.
1472 * This opcode is going to be replaced in a lowering pass inside
1475 * operand0 is the unsized array's ir_value for the calculation
1478 ir_unop_ssbo_unsized_array_length,
1481 * A sentinel marking the last of the unary operations.
1483 ir_last_unop = ir_unop_ssbo_unsized_array_length,
1487 ir_binop_mul, /**< Floating-point or low 32-bit integer multiply. */
1488 ir_binop_imul_high, /**< Calculates the high 32-bits of a 64-bit multiply. */
1492 * Returns the carry resulting from the addition of the two arguments.
1499 * Returns the borrow resulting from the subtraction of the second argument
1500 * from the first argument.
1507 * Takes one of two combinations of arguments:
1510 * - mod(vecN, float)
1512 * Does not take integer types.
1517 * \name Binary comparison operators which return a boolean vector.
1518 * The type of both operands must be equal.
1528 * Returns single boolean for whether all components of operands[0]
1529 * equal the components of operands[1].
1533 * Returns single boolean for whether any component of operands[0]
1534 * is not equal to the corresponding component of operands[1].
1536 ir_binop_any_nequal,
1540 * \name Bit-wise binary operations.
1561 * Load a value the size of a given GLSL type from a uniform block.
1563 * operand0 is the ir_constant uniform block index in the linked shader.
1564 * operand1 is a byte offset within the uniform block.
1569 * \name Multiplies a number by two to a power, part of ARB_gpu_shader5.
1576 * Extract a scalar from a vector
1578 * operand0 is the vector
1579 * operand1 is the index of the field to read from operand0
1581 ir_binop_vector_extract,
1584 * Interpolate fs input at offset
1586 * operand0 is the fs input
1587 * operand1 is the offset from the pixel center
1589 ir_binop_interpolate_at_offset,
1592 * Interpolate fs input at sample position
1594 * operand0 is the fs input
1595 * operand1 is the sample ID
1597 ir_binop_interpolate_at_sample,
1600 * A sentinel marking the last of the binary operations.
1602 ir_last_binop = ir_binop_interpolate_at_sample,
1605 * \name Fused floating-point multiply-add, part of ARB_gpu_shader5.
1614 * \name Conditional Select
1616 * A vector conditional select instruction (like ?:, but operating per-
1617 * component on vectors).
1619 * \see lower_instructions_visitor::ldexp_to_arith
1625 ir_triop_bitfield_extract,
1628 * Generate a value with one field of a vector changed
1630 * operand0 is the vector
1631 * operand1 is the value to write into the vector result
1632 * operand2 is the index in operand0 to be modified
1634 ir_triop_vector_insert,
1637 * A sentinel marking the last of the ternary operations.
1639 ir_last_triop = ir_triop_vector_insert,
1641 ir_quadop_bitfield_insert,
1646 * A sentinel marking the last of the ternary operations.
1648 ir_last_quadop = ir_quadop_vector,
1651 * A sentinel marking the last of all operations.
1653 ir_last_opcode = ir_quadop_vector
1656 class ir_expression : public ir_rvalue {
1658 ir_expression(int op, const struct glsl_type *type,
1659 ir_rvalue *op0, ir_rvalue *op1 = NULL,
1660 ir_rvalue *op2 = NULL, ir_rvalue *op3 = NULL);
1663 * Constructor for unary operation expressions
1665 ir_expression(int op, ir_rvalue *);
1668 * Constructor for binary operation expressions
1670 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1);
1673 * Constructor for ternary operation expressions
1675 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1, ir_rvalue *op2);
1677 virtual bool equals(const ir_instruction *ir,
1678 enum ir_node_type ignore = ir_type_unset) const;
1680 virtual ir_expression *clone(void *mem_ctx, struct hash_table *ht) const;
1683 * Attempt to constant-fold the expression
1685 * The "variable_context" hash table links ir_variable * to ir_constant *
1686 * that represent the variables' values. \c NULL represents an empty
1689 * If the expression cannot be constant folded, this method will return
1692 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1695 * Determine the number of operands used by an expression
1697 static unsigned int get_num_operands(ir_expression_operation);
1700 * Determine the number of operands used by an expression
1702 unsigned int get_num_operands() const
1704 return (this->operation == ir_quadop_vector)
1705 ? this->type->vector_elements : get_num_operands(operation);
1709 * Return whether the expression operates on vectors horizontally.
1711 bool is_horizontal() const
1713 return operation == ir_binop_all_equal ||
1714 operation == ir_binop_any_nequal ||
1715 operation == ir_binop_dot ||
1716 operation == ir_binop_vector_extract ||
1717 operation == ir_triop_vector_insert ||
1718 operation == ir_binop_ubo_load ||
1719 operation == ir_quadop_vector;
1723 * Return a string representing this expression's operator.
1725 const char *operator_string();
1728 * Return a string representing this expression's operator.
1730 static const char *operator_string(ir_expression_operation);
1734 * Do a reverse-lookup to translate the given string into an operator.
1736 static ir_expression_operation get_operator(const char *);
1738 virtual void accept(ir_visitor *v)
1743 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1745 virtual ir_variable *variable_referenced() const;
1747 ir_expression_operation operation;
1748 ir_rvalue *operands[4];
1753 * HIR instruction representing a high-level function call, containing a list
1754 * of parameters and returning a value in the supplied temporary.
1756 class ir_call : public ir_instruction {
1758 ir_call(ir_function_signature *callee,
1759 ir_dereference_variable *return_deref,
1760 exec_list *actual_parameters)
1761 : ir_instruction(ir_type_call), return_deref(return_deref), callee(callee), sub_var(NULL), array_idx(NULL)
1763 assert(callee->return_type != NULL);
1764 actual_parameters->move_nodes_to(& this->actual_parameters);
1765 this->use_builtin = callee->is_builtin();
1768 ir_call(ir_function_signature *callee,
1769 ir_dereference_variable *return_deref,
1770 exec_list *actual_parameters,
1771 ir_variable *var, ir_rvalue *array_idx)
1772 : ir_instruction(ir_type_call), return_deref(return_deref), callee(callee), sub_var(var), array_idx(array_idx)
1774 assert(callee->return_type != NULL);
1775 actual_parameters->move_nodes_to(& this->actual_parameters);
1776 this->use_builtin = callee->is_builtin();
1779 virtual ir_call *clone(void *mem_ctx, struct hash_table *ht) const;
1781 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1783 virtual void accept(ir_visitor *v)
1788 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1791 * Get the name of the function being called.
1793 const char *callee_name() const
1795 return callee->function_name();
1799 * Generates an inline version of the function before @ir,
1800 * storing the return value in return_deref.
1802 void generate_inline(ir_instruction *ir);
1805 * Storage for the function's return value.
1806 * This must be NULL if the return type is void.
1808 ir_dereference_variable *return_deref;
1811 * The specific function signature being called.
1813 ir_function_signature *callee;
1815 /* List of ir_rvalue of paramaters passed in this call. */
1816 exec_list actual_parameters;
1818 /** Should this call only bind to a built-in function? */
1822 * ARB_shader_subroutine support -
1823 * the subroutine uniform variable and array index
1824 * rvalue to be used in the lowering pass later.
1826 ir_variable *sub_var;
1827 ir_rvalue *array_idx;
1832 * \name Jump-like IR instructions.
1834 * These include \c break, \c continue, \c return, and \c discard.
1837 class ir_jump : public ir_instruction {
1839 ir_jump(enum ir_node_type t)
1845 class ir_return : public ir_jump {
1848 : ir_jump(ir_type_return), value(NULL)
1852 ir_return(ir_rvalue *value)
1853 : ir_jump(ir_type_return), value(value)
1857 virtual ir_return *clone(void *mem_ctx, struct hash_table *) const;
1859 ir_rvalue *get_value() const
1864 virtual void accept(ir_visitor *v)
1869 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1876 * Jump instructions used inside loops
1878 * These include \c break and \c continue. The \c break within a loop is
1879 * different from the \c break within a switch-statement.
1881 * \sa ir_switch_jump
1883 class ir_loop_jump : public ir_jump {
1890 ir_loop_jump(jump_mode mode)
1891 : ir_jump(ir_type_loop_jump)
1896 virtual ir_loop_jump *clone(void *mem_ctx, struct hash_table *) const;
1898 virtual void accept(ir_visitor *v)
1903 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1905 bool is_break() const
1907 return mode == jump_break;
1910 bool is_continue() const
1912 return mode == jump_continue;
1915 /** Mode selector for the jump instruction. */
1916 enum jump_mode mode;
1920 * IR instruction representing discard statements.
1922 class ir_discard : public ir_jump {
1925 : ir_jump(ir_type_discard)
1927 this->condition = NULL;
1930 ir_discard(ir_rvalue *cond)
1931 : ir_jump(ir_type_discard)
1933 this->condition = cond;
1936 virtual ir_discard *clone(void *mem_ctx, struct hash_table *ht) const;
1938 virtual void accept(ir_visitor *v)
1943 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1945 ir_rvalue *condition;
1951 * Texture sampling opcodes used in ir_texture
1953 enum ir_texture_opcode {
1954 ir_tex, /**< Regular texture look-up */
1955 ir_txb, /**< Texture look-up with LOD bias */
1956 ir_txl, /**< Texture look-up with explicit LOD */
1957 ir_txd, /**< Texture look-up with partial derivatvies */
1958 ir_txf, /**< Texel fetch with explicit LOD */
1959 ir_txf_ms, /**< Multisample texture fetch */
1960 ir_txs, /**< Texture size */
1961 ir_lod, /**< Texture lod query */
1962 ir_tg4, /**< Texture gather */
1963 ir_query_levels, /**< Texture levels query */
1964 ir_texture_samples, /**< Texture samples query */
1965 ir_samples_identical, /**< Query whether all samples are definitely identical. */
1970 * IR instruction to sample a texture
1972 * The specific form of the IR instruction depends on the \c mode value
1973 * selected from \c ir_texture_opcodes. In the printed IR, these will
1976 * Texel offset (0 or an expression)
1977 * | Projection divisor
1978 * | | Shadow comparitor
1981 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1982 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1983 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1984 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1985 * (txf <type> <sampler> <coordinate> 0 <lod>)
1987 * <type> <sampler> <coordinate> <sample_index>)
1988 * (txs <type> <sampler> <lod>)
1989 * (lod <type> <sampler> <coordinate>)
1990 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
1991 * (query_levels <type> <sampler>)
1992 * (samples_identical <sampler> <coordinate>)
1994 class ir_texture : public ir_rvalue {
1996 ir_texture(enum ir_texture_opcode op)
1997 : ir_rvalue(ir_type_texture),
1998 op(op), sampler(NULL), coordinate(NULL), projector(NULL),
1999 shadow_comparitor(NULL), offset(NULL)
2001 memset(&lod_info, 0, sizeof(lod_info));
2004 virtual ir_texture *clone(void *mem_ctx, struct hash_table *) const;
2006 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
2008 virtual void accept(ir_visitor *v)
2013 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2015 virtual bool equals(const ir_instruction *ir,
2016 enum ir_node_type ignore = ir_type_unset) const;
2019 * Return a string representing the ir_texture_opcode.
2021 const char *opcode_string();
2023 /** Set the sampler and type. */
2024 void set_sampler(ir_dereference *sampler, const glsl_type *type);
2027 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
2029 static ir_texture_opcode get_opcode(const char *);
2031 enum ir_texture_opcode op;
2033 /** Sampler to use for the texture access. */
2034 ir_dereference *sampler;
2036 /** Texture coordinate to sample */
2037 ir_rvalue *coordinate;
2040 * Value used for projective divide.
2042 * If there is no projective divide (the common case), this will be
2043 * \c NULL. Optimization passes should check for this to point to a constant
2044 * of 1.0 and replace that with \c NULL.
2046 ir_rvalue *projector;
2049 * Coordinate used for comparison on shadow look-ups.
2051 * If there is no shadow comparison, this will be \c NULL. For the
2052 * \c ir_txf opcode, this *must* be \c NULL.
2054 ir_rvalue *shadow_comparitor;
2056 /** Texel offset. */
2060 ir_rvalue *lod; /**< Floating point LOD */
2061 ir_rvalue *bias; /**< Floating point LOD bias */
2062 ir_rvalue *sample_index; /**< MSAA sample index */
2063 ir_rvalue *component; /**< Gather component selector */
2065 ir_rvalue *dPdx; /**< Partial derivative of coordinate wrt X */
2066 ir_rvalue *dPdy; /**< Partial derivative of coordinate wrt Y */
2072 struct ir_swizzle_mask {
2079 * Number of components in the swizzle.
2081 unsigned num_components:3;
2084 * Does the swizzle contain duplicate components?
2086 * L-value swizzles cannot contain duplicate components.
2088 unsigned has_duplicates:1;
2092 class ir_swizzle : public ir_rvalue {
2094 ir_swizzle(ir_rvalue *, unsigned x, unsigned y, unsigned z, unsigned w,
2097 ir_swizzle(ir_rvalue *val, const unsigned *components, unsigned count);
2099 ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask);
2101 virtual ir_swizzle *clone(void *mem_ctx, struct hash_table *) const;
2103 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
2106 * Construct an ir_swizzle from the textual representation. Can fail.
2108 static ir_swizzle *create(ir_rvalue *, const char *, unsigned vector_length);
2110 virtual void accept(ir_visitor *v)
2115 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2117 virtual bool equals(const ir_instruction *ir,
2118 enum ir_node_type ignore = ir_type_unset) const;
2120 bool is_lvalue() const
2122 return val->is_lvalue() && !mask.has_duplicates;
2126 * Get the variable that is ultimately referenced by an r-value
2128 virtual ir_variable *variable_referenced() const;
2131 ir_swizzle_mask mask;
2135 * Initialize the mask component of a swizzle
2137 * This is used by the \c ir_swizzle constructors.
2139 void init_mask(const unsigned *components, unsigned count);
2143 class ir_dereference : public ir_rvalue {
2145 virtual ir_dereference *clone(void *mem_ctx, struct hash_table *) const = 0;
2147 bool is_lvalue() const;
2150 * Get the variable that is ultimately referenced by an r-value
2152 virtual ir_variable *variable_referenced() const = 0;
2155 ir_dereference(enum ir_node_type t)
2162 class ir_dereference_variable : public ir_dereference {
2164 ir_dereference_variable(ir_variable *var);
2166 virtual ir_dereference_variable *clone(void *mem_ctx,
2167 struct hash_table *) const;
2169 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
2171 virtual bool equals(const ir_instruction *ir,
2172 enum ir_node_type ignore = ir_type_unset) const;
2175 * Get the variable that is ultimately referenced by an r-value
2177 virtual ir_variable *variable_referenced() const
2182 virtual ir_variable *whole_variable_referenced()
2184 /* ir_dereference_variable objects always dereference the entire
2185 * variable. However, if this dereference is dereferenced by anything
2186 * else, the complete deferefernce chain is not a whole-variable
2187 * dereference. This method should only be called on the top most
2188 * ir_rvalue in a dereference chain.
2193 virtual void accept(ir_visitor *v)
2198 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2201 * Object being dereferenced.
2207 class ir_dereference_array : public ir_dereference {
2209 ir_dereference_array(ir_rvalue *value, ir_rvalue *array_index);
2211 ir_dereference_array(ir_variable *var, ir_rvalue *array_index);
2213 virtual ir_dereference_array *clone(void *mem_ctx,
2214 struct hash_table *) const;
2216 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
2218 virtual bool equals(const ir_instruction *ir,
2219 enum ir_node_type ignore = ir_type_unset) const;
2222 * Get the variable that is ultimately referenced by an r-value
2224 virtual ir_variable *variable_referenced() const
2226 return this->array->variable_referenced();
2229 virtual void accept(ir_visitor *v)
2234 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2237 ir_rvalue *array_index;
2240 void set_array(ir_rvalue *value);
2244 class ir_dereference_record : public ir_dereference {
2246 ir_dereference_record(ir_rvalue *value, const char *field);
2248 ir_dereference_record(ir_variable *var, const char *field);
2250 virtual ir_dereference_record *clone(void *mem_ctx,
2251 struct hash_table *) const;
2253 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
2256 * Get the variable that is ultimately referenced by an r-value
2258 virtual ir_variable *variable_referenced() const
2260 return this->record->variable_referenced();
2263 virtual void accept(ir_visitor *v)
2268 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2276 * Data stored in an ir_constant
2278 union ir_constant_data {
2287 class ir_constant : public ir_rvalue {
2289 ir_constant(const struct glsl_type *type, const ir_constant_data *data);
2290 ir_constant(bool b, unsigned vector_elements=1);
2291 ir_constant(unsigned int u, unsigned vector_elements=1);
2292 ir_constant(int i, unsigned vector_elements=1);
2293 ir_constant(float f, unsigned vector_elements=1);
2294 ir_constant(double d, unsigned vector_elements=1);
2297 * Construct an ir_constant from a list of ir_constant values
2299 ir_constant(const struct glsl_type *type, exec_list *values);
2302 * Construct an ir_constant from a scalar component of another ir_constant
2304 * The new \c ir_constant inherits the type of the component from the
2308 * In the case of a matrix constant, the new constant is a scalar, \b not
2311 ir_constant(const ir_constant *c, unsigned i);
2314 * Return a new ir_constant of the specified type containing all zeros.
2316 static ir_constant *zero(void *mem_ctx, const glsl_type *type);
2318 virtual ir_constant *clone(void *mem_ctx, struct hash_table *) const;
2320 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
2322 virtual void accept(ir_visitor *v)
2327 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2329 virtual bool equals(const ir_instruction *ir,
2330 enum ir_node_type ignore = ir_type_unset) const;
2333 * Get a particular component of a constant as a specific type
2335 * This is useful, for example, to get a value from an integer constant
2336 * as a float or bool. This appears frequently when constructors are
2337 * called with all constant parameters.
2340 bool get_bool_component(unsigned i) const;
2341 float get_float_component(unsigned i) const;
2342 double get_double_component(unsigned i) const;
2343 int get_int_component(unsigned i) const;
2344 unsigned get_uint_component(unsigned i) const;
2347 ir_constant *get_array_element(unsigned i) const;
2349 ir_constant *get_record_field(const char *name);
2352 * Copy the values on another constant at a given offset.
2354 * The offset is ignored for array or struct copies, it's only for
2355 * scalars or vectors into vectors or matrices.
2357 * With identical types on both sides and zero offset it's clone()
2358 * without creating a new object.
2361 void copy_offset(ir_constant *src, int offset);
2364 * Copy the values on another constant at a given offset and
2365 * following an assign-like mask.
2367 * The mask is ignored for scalars.
2369 * Note that this function only handles what assign can handle,
2370 * i.e. at most a vector as source and a column of a matrix as
2374 void copy_masked_offset(ir_constant *src, int offset, unsigned int mask);
2377 * Determine whether a constant has the same value as another constant
2379 * \sa ir_constant::is_zero, ir_constant::is_one,
2380 * ir_constant::is_negative_one
2382 bool has_value(const ir_constant *) const;
2385 * Return true if this ir_constant represents the given value.
2387 * For vectors, this checks that each component is the given value.
2389 virtual bool is_value(float f, int i) const;
2390 virtual bool is_zero() const;
2391 virtual bool is_one() const;
2392 virtual bool is_negative_one() const;
2395 * Return true for constants that could be stored as 16-bit unsigned values.
2397 * Note that this will return true even for signed integer ir_constants, as
2398 * long as the value is non-negative and fits in 16-bits.
2400 virtual bool is_uint16_constant() const;
2403 * Value of the constant.
2405 * The field used to back the values supplied by the constant is determined
2406 * by the type associated with the \c ir_instruction. Constants may be
2407 * scalars, vectors, or matrices.
2409 union ir_constant_data value;
2411 /* Array elements */
2412 ir_constant **array_elements;
2414 /* Structure fields */
2415 exec_list components;
2419 * Parameterless constructor only used by the clone method
2425 * IR instruction to emit a vertex in a geometry shader.
2427 class ir_emit_vertex : public ir_instruction {
2429 ir_emit_vertex(ir_rvalue *stream)
2430 : ir_instruction(ir_type_emit_vertex),
2436 virtual void accept(ir_visitor *v)
2441 virtual ir_emit_vertex *clone(void *mem_ctx, struct hash_table *ht) const
2443 return new(mem_ctx) ir_emit_vertex(this->stream->clone(mem_ctx, ht));
2446 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2448 int stream_id() const
2450 return stream->as_constant()->value.i[0];
2457 * IR instruction to complete the current primitive and start a new one in a
2460 class ir_end_primitive : public ir_instruction {
2462 ir_end_primitive(ir_rvalue *stream)
2463 : ir_instruction(ir_type_end_primitive),
2469 virtual void accept(ir_visitor *v)
2474 virtual ir_end_primitive *clone(void *mem_ctx, struct hash_table *ht) const
2476 return new(mem_ctx) ir_end_primitive(this->stream->clone(mem_ctx, ht));
2479 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2481 int stream_id() const
2483 return stream->as_constant()->value.i[0];
2490 * IR instruction for tessellation control and compute shader barrier.
2492 class ir_barrier : public ir_instruction {
2495 : ir_instruction(ir_type_barrier)
2499 virtual void accept(ir_visitor *v)
2504 virtual ir_barrier *clone(void *mem_ctx, struct hash_table *) const
2506 return new(mem_ctx) ir_barrier();
2509 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
2515 * Apply a visitor to each IR node in a list
2518 visit_exec_list(exec_list *list, ir_visitor *visitor);
2521 * Validate invariants on each IR node in a list
2523 void validate_ir_tree(exec_list *instructions);
2525 struct _mesa_glsl_parse_state;
2526 struct gl_shader_program;
2529 * Detect whether an unlinked shader contains static recursion
2531 * If the list of instructions is determined to contain static recursion,
2532 * \c _mesa_glsl_error will be called to emit error messages for each function
2533 * that is in the recursion cycle.
2536 detect_recursion_unlinked(struct _mesa_glsl_parse_state *state,
2537 exec_list *instructions);
2540 * Detect whether a linked shader contains static recursion
2542 * If the list of instructions is determined to contain static recursion,
2543 * \c link_error_printf will be called to emit error messages for each function
2544 * that is in the recursion cycle. In addition,
2545 * \c gl_shader_program::LinkStatus will be set to false.
2548 detect_recursion_linked(struct gl_shader_program *prog,
2549 exec_list *instructions);
2552 * Make a clone of each IR instruction in a list
2554 * \param in List of IR instructions that are to be cloned
2555 * \param out List to hold the cloned instructions
2558 clone_ir_list(void *mem_ctx, exec_list *out, const exec_list *in);
2561 _mesa_glsl_initialize_variables(exec_list *instructions,
2562 struct _mesa_glsl_parse_state *state);
2565 _mesa_glsl_initialize_derived_variables(struct gl_context *ctx,
2569 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state *state);
2572 _mesa_glsl_initialize_builtin_functions();
2574 extern ir_function_signature *
2575 _mesa_glsl_find_builtin_function(_mesa_glsl_parse_state *state,
2576 const char *name, exec_list *actual_parameters);
2578 extern ir_function *
2579 _mesa_glsl_find_builtin_function_by_name(const char *name);
2582 _mesa_glsl_get_builtin_function_shader(void);
2584 extern ir_function_signature *
2585 _mesa_get_main_function_signature(gl_shader *sh);
2588 _mesa_glsl_release_functions(void);
2591 _mesa_glsl_release_builtin_functions(void);
2594 reparent_ir(exec_list *list, void *mem_ctx);
2596 struct glsl_symbol_table;
2599 import_prototypes(const exec_list *source, exec_list *dest,
2600 struct glsl_symbol_table *symbols, void *mem_ctx);
2603 ir_has_call(ir_instruction *ir);
2606 do_set_program_inouts(exec_list *instructions, struct gl_program *prog,
2607 gl_shader_stage shader_stage);
2610 prototype_string(const glsl_type *return_type, const char *name,
2611 exec_list *parameters);
2614 mode_string(const ir_variable *var);
2617 * Built-in / reserved GL variables names start with "gl_"
2620 is_gl_identifier(const char *s)
2622 return s && s[0] == 'g' && s[1] == 'l' && s[2] == '_';
2626 #endif /* __cplusplus */
2628 extern void _mesa_print_ir(FILE *f, struct exec_list *instructions,
2629 struct _mesa_glsl_parse_state *state);
2632 fprint_ir(FILE *f, const void *instruction);
2634 extern const struct gl_builtin_uniform_desc *
2635 _mesa_glsl_get_builtin_uniform_desc(const char *name);
2642 vertices_per_prim(GLenum prim);