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.
33 #include "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
,
81 ir_type_max
, /**< maximum ir_type enum number, for validation */
82 ir_type_unset
= ir_type_max
87 * Base class of all IR instructions
89 class ir_instruction
: public exec_node
{
91 enum ir_node_type ir_type
;
94 * GCC 4.7+ and clang warn when deleting an ir_instruction unless
95 * there's a virtual destructor present. Because we almost
96 * universally use ralloc for our memory management of
97 * ir_instructions, the destructor doesn't need to do any work.
99 virtual ~ir_instruction()
103 /** ir_print_visitor helper for debugging. */
104 void print(void) const;
105 void fprint(FILE *f
) const;
107 virtual void accept(ir_visitor
*) = 0;
108 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*) = 0;
109 virtual ir_instruction
*clone(void *mem_ctx
,
110 struct hash_table
*ht
) const = 0;
113 * \name IR instruction downcast functions
115 * These functions either cast the object to a derived class or return
116 * \c NULL if the object's type does not match the specified derived class.
117 * Additional downcast functions will be added as needed.
120 class ir_rvalue
*as_rvalue()
122 if (ir_type
== ir_type_dereference_array
||
123 ir_type
== ir_type_dereference_record
||
124 ir_type
== ir_type_dereference_variable
||
125 ir_type
== ir_type_constant
||
126 ir_type
== ir_type_expression
||
127 ir_type
== ir_type_swizzle
||
128 ir_type
== ir_type_texture
)
129 return (class ir_rvalue
*) this;
133 class ir_dereference
*as_dereference()
135 if (ir_type
== ir_type_dereference_array
||
136 ir_type
== ir_type_dereference_record
||
137 ir_type
== ir_type_dereference_variable
)
138 return (class ir_dereference
*) this;
142 class ir_jump
*as_jump()
144 if (ir_type
== ir_type_loop_jump
||
145 ir_type
== ir_type_return
||
146 ir_type
== ir_type_discard
)
147 return (class ir_jump
*) this;
151 #define AS_CHILD(TYPE) \
152 class ir_##TYPE * as_##TYPE() \
154 return ir_type == ir_type_##TYPE ? (ir_##TYPE *) this : NULL; \
158 AS_CHILD(dereference_array
)
159 AS_CHILD(dereference_variable
)
160 AS_CHILD(dereference_record
)
175 * IR equality method: Return true if the referenced instruction would
176 * return the same value as this one.
178 * This intended to be used for CSE and algebraic optimizations, on rvalues
179 * in particular. No support for other instruction types (assignments,
180 * jumps, calls, etc.) is planned.
182 virtual bool equals(ir_instruction
*ir
, enum ir_node_type ignore
= ir_type_unset
);
185 ir_instruction(enum ir_node_type t
)
193 assert(!"Should not get here.");
199 * The base class for all "values"/expression trees.
201 class ir_rvalue
: public ir_instruction
{
203 const struct glsl_type
*type
;
205 virtual ir_rvalue
*clone(void *mem_ctx
, struct hash_table
*) const;
207 virtual void accept(ir_visitor
*v
)
212 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
214 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
216 ir_rvalue
*as_rvalue_to_saturate();
218 virtual bool is_lvalue() const
224 * Get the variable that is ultimately referenced by an r-value
226 virtual ir_variable
*variable_referenced() const
233 * If an r-value is a reference to a whole variable, get that variable
236 * Pointer to a variable that is completely dereferenced by the r-value. If
237 * the r-value is not a dereference or the dereference does not access the
238 * entire variable (i.e., it's just one array element, struct field), \c NULL
241 virtual ir_variable
*whole_variable_referenced()
247 * Determine if an r-value has the value zero
249 * The base implementation of this function always returns \c false. The
250 * \c ir_constant class over-rides this function to return \c true \b only
251 * for vector and scalar types that have all elements set to the value
252 * zero (or \c false for booleans).
254 * \sa ir_constant::has_value, ir_rvalue::is_one, ir_rvalue::is_negative_one,
255 * ir_constant::is_basis
257 virtual bool is_zero() const;
260 * Determine if an r-value has the value one
262 * The base implementation of this function always returns \c false. The
263 * \c ir_constant class over-rides this function to return \c true \b only
264 * for vector and scalar types that have all elements set to the value
265 * one (or \c true for booleans).
267 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_negative_one,
268 * ir_constant::is_basis
270 virtual bool is_one() const;
273 * Determine if an r-value has the value negative one
275 * The base implementation of this function always returns \c false. The
276 * \c ir_constant class over-rides this function to return \c true \b only
277 * for vector and scalar types that have all elements set to the value
278 * negative one. For boolean types, the result is always \c false.
280 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one
281 * ir_constant::is_basis
283 virtual bool is_negative_one() const;
286 * Determine if an r-value is a basis vector
288 * The base implementation of this function always returns \c false. The
289 * \c ir_constant class over-rides this function to return \c true \b only
290 * for vector and scalar types that have one element set to the value one,
291 * and the other elements set to the value zero. For boolean types, the
292 * result is always \c false.
294 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one,
295 * is_constant::is_negative_one
297 virtual bool is_basis() const;
300 * Determine if an r-value is an unsigned integer constant which can be
303 * \sa ir_constant::is_uint16_constant.
305 virtual bool is_uint16_constant() const { return false; }
308 * Return a generic value of error_type.
310 * Allocation will be performed with 'mem_ctx' as ralloc owner.
312 static ir_rvalue
*error_value(void *mem_ctx
);
315 ir_rvalue(enum ir_node_type t
);
320 * Variable storage classes
322 enum ir_variable_mode
{
323 ir_var_auto
= 0, /**< Function local variables and globals. */
324 ir_var_uniform
, /**< Variable declared as a uniform. */
329 ir_var_function_inout
,
330 ir_var_const_in
, /**< "in" param that must be a constant expression */
331 ir_var_system_value
, /**< Ex: front-face, instance-id, etc. */
332 ir_var_temporary
, /**< Temporary variable generated during compilation. */
333 ir_var_mode_count
/**< Number of variable modes */
337 * Enum keeping track of how a variable was declared. For error checking of
338 * the gl_PerVertex redeclaration rules.
340 enum ir_var_declaration_type
{
342 * Normal declaration (for most variables, this means an explicit
343 * declaration. Exception: temporaries are always implicitly declared, but
344 * they still use ir_var_declared_normally).
346 * Note: an ir_variable that represents a named interface block uses
347 * ir_var_declared_normally.
349 ir_var_declared_normally
= 0,
352 * Variable was explicitly declared (or re-declared) in an unnamed
355 ir_var_declared_in_block
,
358 * Variable is an implicitly declared built-in that has not been explicitly
359 * re-declared by the shader.
361 ir_var_declared_implicitly
,
365 * \brief Layout qualifiers for gl_FragDepth.
367 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
368 * with a layout qualifier.
370 enum ir_depth_layout
{
371 ir_depth_layout_none
, /**< No depth layout is specified. */
373 ir_depth_layout_greater
,
374 ir_depth_layout_less
,
375 ir_depth_layout_unchanged
379 * \brief Convert depth layout qualifier to string.
382 depth_layout_string(ir_depth_layout layout
);
385 * Description of built-in state associated with a uniform
387 * \sa ir_variable::state_slots
389 struct ir_state_slot
{
396 * Get the string value for an interpolation qualifier
398 * \return The string that would be used in a shader to specify \c
399 * mode will be returned.
401 * This function is used to generate error messages of the form "shader
402 * uses %s interpolation qualifier", so in the case where there is no
403 * interpolation qualifier, it returns "no".
405 * This function should only be used on a shader input or output variable.
407 const char *interpolation_string(unsigned interpolation
);
410 class ir_variable
: public ir_instruction
{
412 ir_variable(const struct glsl_type
*, const char *, ir_variable_mode
);
414 virtual ir_variable
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
416 virtual void accept(ir_visitor
*v
)
421 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
425 * Determine how this variable should be interpolated based on its
426 * interpolation qualifier (if present), whether it is gl_Color or
427 * gl_SecondaryColor, and whether flatshading is enabled in the current GL
430 * The return value will always be either INTERP_QUALIFIER_SMOOTH,
431 * INTERP_QUALIFIER_NOPERSPECTIVE, or INTERP_QUALIFIER_FLAT.
433 glsl_interp_qualifier
determine_interpolation_mode(bool flat_shade
);
436 * Determine whether or not a variable is part of a uniform block.
438 inline bool is_in_uniform_block() const
440 return this->data
.mode
== ir_var_uniform
&& this->interface_type
!= NULL
;
444 * Determine whether or not a variable is the declaration of an interface
447 * For the first declaration below, there will be an \c ir_variable named
448 * "instance" whose type and whose instance_type will be the same
449 * \cglsl_type. For the second declaration, there will be an \c ir_variable
450 * named "f" whose type is float and whose instance_type is B2.
452 * "instance" is an interface instance variable, but "f" is not.
462 inline bool is_interface_instance() const
464 const glsl_type
*const t
= this->type
;
466 return (t
== this->interface_type
)
467 || (t
->is_array() && t
->fields
.array
== this->interface_type
);
471 * Set this->interface_type on a newly created variable.
473 void init_interface_type(const struct glsl_type
*type
)
475 assert(this->interface_type
== NULL
);
476 this->interface_type
= type
;
477 if (this->is_interface_instance()) {
478 this->max_ifc_array_access
=
479 rzalloc_array(this, unsigned, type
->length
);
484 * Change this->interface_type on a variable that previously had a
485 * different, but compatible, interface_type. This is used during linking
486 * to set the size of arrays in interface blocks.
488 void change_interface_type(const struct glsl_type
*type
)
490 if (this->max_ifc_array_access
!= NULL
) {
491 /* max_ifc_array_access has already been allocated, so make sure the
492 * new interface has the same number of fields as the old one.
494 assert(this->interface_type
->length
== type
->length
);
496 this->interface_type
= type
;
500 * Change this->interface_type on a variable that previously had a
501 * different, and incompatible, interface_type. This is used during
502 * compilation to handle redeclaration of the built-in gl_PerVertex
505 void reinit_interface_type(const struct glsl_type
*type
)
507 if (this->max_ifc_array_access
!= NULL
) {
509 /* Redeclaring gl_PerVertex is only allowed if none of the built-ins
510 * it defines have been accessed yet; so it's safe to throw away the
511 * old max_ifc_array_access pointer, since all of its values are
514 for (unsigned i
= 0; i
< this->interface_type
->length
; i
++)
515 assert(this->max_ifc_array_access
[i
] == 0);
517 ralloc_free(this->max_ifc_array_access
);
518 this->max_ifc_array_access
= NULL
;
520 this->interface_type
= NULL
;
521 init_interface_type(type
);
524 const glsl_type
*get_interface_type() const
526 return this->interface_type
;
530 * Declared type of the variable
532 const struct glsl_type
*type
;
535 * Declared name of the variable
540 * For variables which satisfy the is_interface_instance() predicate, this
541 * points to an array of integers such that if the ith member of the
542 * interface block is an array, max_ifc_array_access[i] is the maximum
543 * array element of that member that has been accessed. If the ith member
544 * of the interface block is not an array, max_ifc_array_access[i] is
547 * For variables whose type is not an interface block, this pointer is
550 unsigned *max_ifc_array_access
;
552 struct ir_variable_data
{
555 * Is the variable read-only?
557 * This is set for variables declared as \c const, shader inputs,
560 unsigned read_only
:1;
563 unsigned invariant
:1;
567 * Has this variable been used for reading or writing?
569 * Several GLSL semantic checks require knowledge of whether or not a
570 * variable has been used. For example, it is an error to redeclare a
571 * variable as invariant after it has been used.
573 * This is only maintained in the ast_to_hir.cpp path, not in
574 * Mesa's fixed function or ARB program paths.
579 * Has this variable been statically assigned?
581 * This answers whether the variable was assigned in any path of
582 * the shader during ast_to_hir. This doesn't answer whether it is
583 * still written after dead code removal, nor is it maintained in
584 * non-ast_to_hir.cpp (GLSL parsing) paths.
589 * Enum indicating how the variable was declared. See
590 * ir_var_declaration_type.
592 * This is used to detect certain kinds of illegal variable redeclarations.
594 unsigned how_declared
:2;
597 * Storage class of the variable.
599 * \sa ir_variable_mode
604 * Interpolation mode for shader inputs / outputs
606 * \sa ir_variable_interpolation
608 unsigned interpolation
:2;
611 * \name ARB_fragment_coord_conventions
614 unsigned origin_upper_left
:1;
615 unsigned pixel_center_integer
:1;
619 * Was the location explicitly set in the shader?
621 * If the location is explicitly set in the shader, it \b cannot be changed
622 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
625 unsigned explicit_location
:1;
626 unsigned explicit_index
:1;
629 * Was an initial binding explicitly set in the shader?
631 * If so, constant_value contains an integer ir_constant representing the
632 * initial binding point.
634 unsigned explicit_binding
:1;
637 * Does this variable have an initializer?
639 * This is used by the linker to cross-validiate initializers of global
642 unsigned has_initializer
:1;
645 * Is this variable a generic output or input that has not yet been matched
646 * up to a variable in another stage of the pipeline?
648 * This is used by the linker as scratch storage while assigning locations
649 * to generic inputs and outputs.
651 unsigned is_unmatched_generic_inout
:1;
654 * If non-zero, then this variable may be packed along with other variables
655 * into a single varying slot, so this offset should be applied when
656 * accessing components. For example, an offset of 1 means that the x
657 * component of this variable is actually stored in component y of the
658 * location specified by \c location.
660 unsigned location_frac
:2;
663 * Non-zero if this variable was created by lowering a named interface
664 * block which was not an array.
666 * Note that this variable and \c from_named_ifc_block_array will never
669 unsigned from_named_ifc_block_nonarray
:1;
672 * Non-zero if this variable was created by lowering a named interface
673 * block which was an array.
675 * Note that this variable and \c from_named_ifc_block_nonarray will never
678 unsigned from_named_ifc_block_array
:1;
681 * Non-zero if the variable must be a shader input. This is useful for
682 * constraints on function parameters.
684 unsigned must_be_shader_input
:1;
687 * \brief Layout qualifier for gl_FragDepth.
689 * This is not equal to \c ir_depth_layout_none if and only if this
690 * variable is \c gl_FragDepth and a layout qualifier is specified.
692 ir_depth_layout depth_layout
;
695 * Storage location of the base of this variable
697 * The precise meaning of this field depends on the nature of the variable.
699 * - Vertex shader input: one of the values from \c gl_vert_attrib.
700 * - Vertex shader output: one of the values from \c gl_varying_slot.
701 * - Geometry shader input: one of the values from \c gl_varying_slot.
702 * - Geometry shader output: one of the values from \c gl_varying_slot.
703 * - Fragment shader input: one of the values from \c gl_varying_slot.
704 * - Fragment shader output: one of the values from \c gl_frag_result.
705 * - Uniforms: Per-stage uniform slot number for default uniform block.
706 * - Uniforms: Index within the uniform block definition for UBO members.
707 * - Other: This field is not currently used.
709 * If the variable is a uniform, shader input, or shader output, and the
710 * slot has not been assigned, the value will be -1.
715 * Vertex stream output identifier.
720 * output index for dual source blending.
725 * Initial binding point for a sampler or UBO.
727 * For array types, this represents the binding point for the first element.
732 * Location an atomic counter is stored at.
735 unsigned buffer_index
;
740 * ARB_shader_image_load_store qualifiers.
743 bool read_only
; /**< "readonly" qualifier. */
744 bool write_only
; /**< "writeonly" qualifier. */
749 /** Image internal format if specified explicitly, otherwise GL_NONE. */
754 * Highest element accessed with a constant expression array index
756 * Not used for non-array variables.
758 unsigned max_array_access
;
763 * Built-in state that backs this uniform
765 * Once set at variable creation, \c state_slots must remain invariant.
766 * This is because, ideally, this array would be shared by all clones of
767 * this variable in the IR tree. In other words, we'd really like for it
768 * to be a fly-weight.
770 * If the variable is not a uniform, \c num_state_slots will be zero and
771 * \c state_slots will be \c NULL.
774 unsigned num_state_slots
; /**< Number of state slots used */
775 ir_state_slot
*state_slots
; /**< State descriptors. */
779 * Emit a warning if this variable is accessed.
781 const char *warn_extension
;
784 * Value assigned in the initializer of a variable declared "const"
786 ir_constant
*constant_value
;
789 * Constant expression assigned in the initializer of the variable
792 * This field and \c ::constant_value are distinct. Even if the two fields
793 * refer to constants with the same value, they must point to separate
796 ir_constant
*constant_initializer
;
800 * For variables that are in an interface block or are an instance of an
801 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
803 * \sa ir_variable::location
805 const glsl_type
*interface_type
;
809 * A function that returns whether a built-in function is available in the
810 * current shading language (based on version, ES or desktop, and extensions).
812 typedef bool (*builtin_available_predicate
)(const _mesa_glsl_parse_state
*);
816 * The representation of a function instance; may be the full definition or
817 * simply a prototype.
819 class ir_function_signature
: public ir_instruction
{
820 /* An ir_function_signature will be part of the list of signatures in
824 ir_function_signature(const glsl_type
*return_type
,
825 builtin_available_predicate builtin_avail
= NULL
);
827 virtual ir_function_signature
*clone(void *mem_ctx
,
828 struct hash_table
*ht
) const;
829 ir_function_signature
*clone_prototype(void *mem_ctx
,
830 struct hash_table
*ht
) const;
832 virtual void accept(ir_visitor
*v
)
837 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
840 * Attempt to evaluate this function as a constant expression,
841 * given a list of the actual parameters and the variable context.
842 * Returns NULL for non-built-ins.
844 ir_constant
*constant_expression_value(exec_list
*actual_parameters
, struct hash_table
*variable_context
);
847 * Get the name of the function for which this is a signature
849 const char *function_name() const;
852 * Get a handle to the function for which this is a signature
854 * There is no setter function, this function returns a \c const pointer,
855 * and \c ir_function_signature::_function is private for a reason. The
856 * only way to make a connection between a function and function signature
857 * is via \c ir_function::add_signature. This helps ensure that certain
858 * invariants (i.e., a function signature is in the list of signatures for
859 * its \c _function) are met.
861 * \sa ir_function::add_signature
863 inline const class ir_function
*function() const
865 return this->_function
;
869 * Check whether the qualifiers match between this signature's parameters
870 * and the supplied parameter list. If not, returns the name of the first
871 * parameter with mismatched qualifiers (for use in error messages).
873 const char *qualifiers_match(exec_list
*params
);
876 * Replace the current parameter list with the given one. This is useful
877 * if the current information came from a prototype, and either has invalid
878 * or missing parameter names.
880 void replace_parameters(exec_list
*new_params
);
883 * Function return type.
885 * \note This discards the optional precision qualifier.
887 const struct glsl_type
*return_type
;
890 * List of ir_variable of function parameters.
892 * This represents the storage. The paramaters passed in a particular
893 * call will be in ir_call::actual_paramaters.
895 struct exec_list parameters
;
897 /** Whether or not this function has a body (which may be empty). */
898 unsigned is_defined
:1;
900 /** Whether or not this function signature is a built-in. */
901 bool is_builtin() const;
904 * Whether or not this function is an intrinsic to be implemented
909 /** Whether or not a built-in is available for this shader. */
910 bool is_builtin_available(const _mesa_glsl_parse_state
*state
) const;
912 /** Body of instructions in the function. */
913 struct exec_list body
;
917 * A function pointer to a predicate that answers whether a built-in
918 * function is available in the current shader. NULL if not a built-in.
920 builtin_available_predicate builtin_avail
;
922 /** Function of which this signature is one overload. */
923 class ir_function
*_function
;
925 /** Function signature of which this one is a prototype clone */
926 const ir_function_signature
*origin
;
928 friend class ir_function
;
931 * Helper function to run a list of instructions for constant
932 * expression evaluation.
934 * The hash table represents the values of the visible variables.
935 * There are no scoping issues because the table is indexed on
936 * ir_variable pointers, not variable names.
938 * Returns false if the expression is not constant, true otherwise,
939 * and the value in *result if result is non-NULL.
941 bool constant_expression_evaluate_expression_list(const struct exec_list
&body
,
942 struct hash_table
*variable_context
,
943 ir_constant
**result
);
948 * Header for tracking multiple overloaded functions with the same name.
949 * Contains a list of ir_function_signatures representing each of the
952 class ir_function
: public ir_instruction
{
954 ir_function(const char *name
);
956 virtual ir_function
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
958 virtual void accept(ir_visitor
*v
)
963 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
965 void add_signature(ir_function_signature
*sig
)
967 sig
->_function
= this;
968 this->signatures
.push_tail(sig
);
972 * Find a signature that matches a set of actual parameters, taking implicit
973 * conversions into account. Also flags whether the match was exact.
975 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
976 const exec_list
*actual_param
,
977 bool *match_is_exact
);
980 * Find a signature that matches a set of actual parameters, taking implicit
981 * conversions into account.
983 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
984 const exec_list
*actual_param
);
987 * Find a signature that exactly matches a set of actual parameters without
988 * any implicit type conversions.
990 ir_function_signature
*exact_matching_signature(_mesa_glsl_parse_state
*state
,
991 const exec_list
*actual_ps
);
994 * Name of the function.
998 /** Whether or not this function has a signature that isn't a built-in. */
999 bool has_user_signature();
1002 * List of ir_function_signature for each overloaded function with this name.
1004 struct exec_list signatures
;
1007 inline const char *ir_function_signature::function_name() const
1009 return this->_function
->name
;
1015 * IR instruction representing high-level if-statements
1017 class ir_if
: public ir_instruction
{
1019 ir_if(ir_rvalue
*condition
)
1020 : ir_instruction(ir_type_if
), condition(condition
)
1024 virtual ir_if
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1026 virtual void accept(ir_visitor
*v
)
1031 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1033 ir_rvalue
*condition
;
1034 /** List of ir_instruction for the body of the then branch */
1035 exec_list then_instructions
;
1036 /** List of ir_instruction for the body of the else branch */
1037 exec_list else_instructions
;
1042 * IR instruction representing a high-level loop structure.
1044 class ir_loop
: public ir_instruction
{
1048 virtual ir_loop
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1050 virtual void accept(ir_visitor
*v
)
1055 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1057 /** List of ir_instruction that make up the body of the loop. */
1058 exec_list body_instructions
;
1062 class ir_assignment
: public ir_instruction
{
1064 ir_assignment(ir_rvalue
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
= NULL
);
1067 * Construct an assignment with an explicit write mask
1070 * Since a write mask is supplied, the LHS must already be a bare
1071 * \c ir_dereference. The cannot be any swizzles in the LHS.
1073 ir_assignment(ir_dereference
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
,
1074 unsigned write_mask
);
1076 virtual ir_assignment
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1078 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1080 virtual void accept(ir_visitor
*v
)
1085 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1088 * Get a whole variable written by an assignment
1090 * If the LHS of the assignment writes a whole variable, the variable is
1091 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1094 * - Assigning to a scalar
1095 * - Assigning to all components of a vector
1096 * - Whole array (or matrix) assignment
1097 * - Whole structure assignment
1099 ir_variable
*whole_variable_written();
1102 * Set the LHS of an assignment
1104 void set_lhs(ir_rvalue
*lhs
);
1107 * Left-hand side of the assignment.
1109 * This should be treated as read only. If you need to set the LHS of an
1110 * assignment, use \c ir_assignment::set_lhs.
1112 ir_dereference
*lhs
;
1115 * Value being assigned
1120 * Optional condition for the assignment.
1122 ir_rvalue
*condition
;
1126 * Component mask written
1128 * For non-vector types in the LHS, this field will be zero. For vector
1129 * types, a bit will be set for each component that is written. Note that
1130 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1132 * A partially-set write mask means that each enabled channel gets
1133 * the value from a consecutive channel of the rhs. For example,
1134 * to write just .xyw of gl_FrontColor with color:
1136 * (assign (constant bool (1)) (xyw)
1137 * (var_ref gl_FragColor)
1138 * (swiz xyw (var_ref color)))
1140 unsigned write_mask
:4;
1143 /* Update ir_expression::get_num_operands() and operator_strs when
1144 * updating this list.
1146 enum ir_expression_operation
{
1155 ir_unop_exp
, /**< Log base e on gentype */
1156 ir_unop_log
, /**< Natural log on gentype */
1159 ir_unop_f2i
, /**< Float-to-integer conversion. */
1160 ir_unop_f2u
, /**< Float-to-unsigned conversion. */
1161 ir_unop_i2f
, /**< Integer-to-float conversion. */
1162 ir_unop_f2b
, /**< Float-to-boolean conversion */
1163 ir_unop_b2f
, /**< Boolean-to-float conversion */
1164 ir_unop_i2b
, /**< int-to-boolean conversion */
1165 ir_unop_b2i
, /**< Boolean-to-int conversion */
1166 ir_unop_u2f
, /**< Unsigned-to-float conversion. */
1167 ir_unop_i2u
, /**< Integer-to-unsigned conversion. */
1168 ir_unop_u2i
, /**< Unsigned-to-integer conversion. */
1169 ir_unop_bitcast_i2f
, /**< Bit-identical int-to-float "conversion" */
1170 ir_unop_bitcast_f2i
, /**< Bit-identical float-to-int "conversion" */
1171 ir_unop_bitcast_u2f
, /**< Bit-identical uint-to-float "conversion" */
1172 ir_unop_bitcast_f2u
, /**< Bit-identical float-to-uint "conversion" */
1176 * \name Unary floating-point rounding operations.
1187 * \name Trigonometric operations.
1192 ir_unop_sin_reduced
, /**< Reduced range sin. [-pi, pi] */
1193 ir_unop_cos_reduced
, /**< Reduced range cos. [-pi, pi] */
1197 * \name Partial derivatives.
1205 * \name Floating point pack and unpack operations.
1208 ir_unop_pack_snorm_2x16
,
1209 ir_unop_pack_snorm_4x8
,
1210 ir_unop_pack_unorm_2x16
,
1211 ir_unop_pack_unorm_4x8
,
1212 ir_unop_pack_half_2x16
,
1213 ir_unop_unpack_snorm_2x16
,
1214 ir_unop_unpack_snorm_4x8
,
1215 ir_unop_unpack_unorm_2x16
,
1216 ir_unop_unpack_unorm_4x8
,
1217 ir_unop_unpack_half_2x16
,
1221 * \name Lowered floating point unpacking operations.
1223 * \see lower_packing_builtins_visitor::split_unpack_half_2x16
1226 ir_unop_unpack_half_2x16_split_x
,
1227 ir_unop_unpack_half_2x16_split_y
,
1231 * \name Bit operations, part of ARB_gpu_shader5.
1234 ir_unop_bitfield_reverse
,
1243 * A sentinel marking the last of the unary operations.
1245 ir_last_unop
= ir_unop_noise
,
1249 ir_binop_mul
, /**< Floating-point or low 32-bit integer multiply. */
1250 ir_binop_imul_high
, /**< Calculates the high 32-bits of a 64-bit multiply. */
1254 * Returns the carry resulting from the addition of the two arguments.
1261 * Returns the borrow resulting from the subtraction of the second argument
1262 * from the first argument.
1269 * Takes one of two combinations of arguments:
1272 * - mod(vecN, float)
1274 * Does not take integer types.
1279 * \name Binary comparison operators which return a boolean vector.
1280 * The type of both operands must be equal.
1290 * Returns single boolean for whether all components of operands[0]
1291 * equal the components of operands[1].
1295 * Returns single boolean for whether any component of operands[0]
1296 * is not equal to the corresponding component of operands[1].
1298 ir_binop_any_nequal
,
1302 * \name Bit-wise binary operations.
1323 * \name Lowered floating point packing operations.
1325 * \see lower_packing_builtins_visitor::split_pack_half_2x16
1328 ir_binop_pack_half_2x16_split
,
1332 * \name First half of a lowered bitfieldInsert() operation.
1334 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1341 * Load a value the size of a given GLSL type from a uniform block.
1343 * operand0 is the ir_constant uniform block index in the linked shader.
1344 * operand1 is a byte offset within the uniform block.
1349 * \name Multiplies a number by two to a power, part of ARB_gpu_shader5.
1356 * Extract a scalar from a vector
1358 * operand0 is the vector
1359 * operand1 is the index of the field to read from operand0
1361 ir_binop_vector_extract
,
1364 * A sentinel marking the last of the binary operations.
1366 ir_last_binop
= ir_binop_vector_extract
,
1369 * \name Fused floating-point multiply-add, part of ARB_gpu_shader5.
1378 * \name Conditional Select
1380 * A vector conditional select instruction (like ?:, but operating per-
1381 * component on vectors).
1383 * \see lower_instructions_visitor::ldexp_to_arith
1390 * \name Second half of a lowered bitfieldInsert() operation.
1392 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1398 ir_triop_bitfield_extract
,
1401 * Generate a value with one field of a vector changed
1403 * operand0 is the vector
1404 * operand1 is the value to write into the vector result
1405 * operand2 is the index in operand0 to be modified
1407 ir_triop_vector_insert
,
1410 * A sentinel marking the last of the ternary operations.
1412 ir_last_triop
= ir_triop_vector_insert
,
1414 ir_quadop_bitfield_insert
,
1419 * A sentinel marking the last of the ternary operations.
1421 ir_last_quadop
= ir_quadop_vector
,
1424 * A sentinel marking the last of all operations.
1426 ir_last_opcode
= ir_quadop_vector
1429 class ir_expression
: public ir_rvalue
{
1431 ir_expression(int op
, const struct glsl_type
*type
,
1432 ir_rvalue
*op0
, ir_rvalue
*op1
= NULL
,
1433 ir_rvalue
*op2
= NULL
, ir_rvalue
*op3
= NULL
);
1436 * Constructor for unary operation expressions
1438 ir_expression(int op
, ir_rvalue
*);
1441 * Constructor for binary operation expressions
1443 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
);
1446 * Constructor for ternary operation expressions
1448 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
, ir_rvalue
*op2
);
1450 virtual bool equals(ir_instruction
*ir
, enum ir_node_type ignore
= ir_type_unset
);
1452 virtual ir_expression
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1455 * Attempt to constant-fold the expression
1457 * The "variable_context" hash table links ir_variable * to ir_constant *
1458 * that represent the variables' values. \c NULL represents an empty
1461 * If the expression cannot be constant folded, this method will return
1464 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1467 * Determine the number of operands used by an expression
1469 static unsigned int get_num_operands(ir_expression_operation
);
1472 * Determine the number of operands used by an expression
1474 unsigned int get_num_operands() const
1476 return (this->operation
== ir_quadop_vector
)
1477 ? this->type
->vector_elements
: get_num_operands(operation
);
1481 * Return whether the expression operates on vectors horizontally.
1483 bool is_horizontal() const
1485 return operation
== ir_binop_all_equal
||
1486 operation
== ir_binop_any_nequal
||
1487 operation
== ir_unop_any
||
1488 operation
== ir_binop_dot
||
1489 operation
== ir_quadop_vector
;
1493 * Return a string representing this expression's operator.
1495 const char *operator_string();
1498 * Return a string representing this expression's operator.
1500 static const char *operator_string(ir_expression_operation
);
1504 * Do a reverse-lookup to translate the given string into an operator.
1506 static ir_expression_operation
get_operator(const char *);
1508 virtual void accept(ir_visitor
*v
)
1513 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1515 ir_expression_operation operation
;
1516 ir_rvalue
*operands
[4];
1521 * HIR instruction representing a high-level function call, containing a list
1522 * of parameters and returning a value in the supplied temporary.
1524 class ir_call
: public ir_instruction
{
1526 ir_call(ir_function_signature
*callee
,
1527 ir_dereference_variable
*return_deref
,
1528 exec_list
*actual_parameters
)
1529 : ir_instruction(ir_type_call
), return_deref(return_deref
), callee(callee
)
1531 assert(callee
->return_type
!= NULL
);
1532 actual_parameters
->move_nodes_to(& this->actual_parameters
);
1533 this->use_builtin
= callee
->is_builtin();
1536 virtual ir_call
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1538 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1540 virtual void accept(ir_visitor
*v
)
1545 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1548 * Get the name of the function being called.
1550 const char *callee_name() const
1552 return callee
->function_name();
1556 * Generates an inline version of the function before @ir,
1557 * storing the return value in return_deref.
1559 void generate_inline(ir_instruction
*ir
);
1562 * Storage for the function's return value.
1563 * This must be NULL if the return type is void.
1565 ir_dereference_variable
*return_deref
;
1568 * The specific function signature being called.
1570 ir_function_signature
*callee
;
1572 /* List of ir_rvalue of paramaters passed in this call. */
1573 exec_list actual_parameters
;
1575 /** Should this call only bind to a built-in function? */
1581 * \name Jump-like IR instructions.
1583 * These include \c break, \c continue, \c return, and \c discard.
1586 class ir_jump
: public ir_instruction
{
1588 ir_jump(enum ir_node_type t
)
1594 class ir_return
: public ir_jump
{
1597 : ir_jump(ir_type_return
), value(NULL
)
1601 ir_return(ir_rvalue
*value
)
1602 : ir_jump(ir_type_return
), value(value
)
1606 virtual ir_return
*clone(void *mem_ctx
, struct hash_table
*) const;
1608 ir_rvalue
*get_value() const
1613 virtual void accept(ir_visitor
*v
)
1618 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1625 * Jump instructions used inside loops
1627 * These include \c break and \c continue. The \c break within a loop is
1628 * different from the \c break within a switch-statement.
1630 * \sa ir_switch_jump
1632 class ir_loop_jump
: public ir_jump
{
1639 ir_loop_jump(jump_mode mode
)
1640 : ir_jump(ir_type_loop_jump
)
1645 virtual ir_loop_jump
*clone(void *mem_ctx
, struct hash_table
*) const;
1647 virtual void accept(ir_visitor
*v
)
1652 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1654 bool is_break() const
1656 return mode
== jump_break
;
1659 bool is_continue() const
1661 return mode
== jump_continue
;
1664 /** Mode selector for the jump instruction. */
1665 enum jump_mode mode
;
1669 * IR instruction representing discard statements.
1671 class ir_discard
: public ir_jump
{
1674 : ir_jump(ir_type_discard
)
1676 this->condition
= NULL
;
1679 ir_discard(ir_rvalue
*cond
)
1680 : ir_jump(ir_type_discard
)
1682 this->condition
= cond
;
1685 virtual ir_discard
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1687 virtual void accept(ir_visitor
*v
)
1692 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1694 ir_rvalue
*condition
;
1700 * Texture sampling opcodes used in ir_texture
1702 enum ir_texture_opcode
{
1703 ir_tex
, /**< Regular texture look-up */
1704 ir_txb
, /**< Texture look-up with LOD bias */
1705 ir_txl
, /**< Texture look-up with explicit LOD */
1706 ir_txd
, /**< Texture look-up with partial derivatvies */
1707 ir_txf
, /**< Texel fetch with explicit LOD */
1708 ir_txf_ms
, /**< Multisample texture fetch */
1709 ir_txs
, /**< Texture size */
1710 ir_lod
, /**< Texture lod query */
1711 ir_tg4
, /**< Texture gather */
1712 ir_query_levels
/**< Texture levels query */
1717 * IR instruction to sample a texture
1719 * The specific form of the IR instruction depends on the \c mode value
1720 * selected from \c ir_texture_opcodes. In the printed IR, these will
1723 * Texel offset (0 or an expression)
1724 * | Projection divisor
1725 * | | Shadow comparitor
1728 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1729 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1730 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1731 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1732 * (txf <type> <sampler> <coordinate> 0 <lod>)
1734 * <type> <sampler> <coordinate> <sample_index>)
1735 * (txs <type> <sampler> <lod>)
1736 * (lod <type> <sampler> <coordinate>)
1737 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
1738 * (query_levels <type> <sampler>)
1740 class ir_texture
: public ir_rvalue
{
1742 ir_texture(enum ir_texture_opcode op
)
1743 : ir_rvalue(ir_type_texture
),
1744 op(op
), sampler(NULL
), coordinate(NULL
), projector(NULL
),
1745 shadow_comparitor(NULL
), offset(NULL
)
1747 memset(&lod_info
, 0, sizeof(lod_info
));
1750 virtual ir_texture
*clone(void *mem_ctx
, struct hash_table
*) const;
1752 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1754 virtual void accept(ir_visitor
*v
)
1759 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1761 virtual bool equals(ir_instruction
*ir
, enum ir_node_type ignore
= ir_type_unset
);
1764 * Return a string representing the ir_texture_opcode.
1766 const char *opcode_string();
1768 /** Set the sampler and type. */
1769 void set_sampler(ir_dereference
*sampler
, const glsl_type
*type
);
1772 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1774 static ir_texture_opcode
get_opcode(const char *);
1776 enum ir_texture_opcode op
;
1778 /** Sampler to use for the texture access. */
1779 ir_dereference
*sampler
;
1781 /** Texture coordinate to sample */
1782 ir_rvalue
*coordinate
;
1785 * Value used for projective divide.
1787 * If there is no projective divide (the common case), this will be
1788 * \c NULL. Optimization passes should check for this to point to a constant
1789 * of 1.0 and replace that with \c NULL.
1791 ir_rvalue
*projector
;
1794 * Coordinate used for comparison on shadow look-ups.
1796 * If there is no shadow comparison, this will be \c NULL. For the
1797 * \c ir_txf opcode, this *must* be \c NULL.
1799 ir_rvalue
*shadow_comparitor
;
1801 /** Texel offset. */
1805 ir_rvalue
*lod
; /**< Floating point LOD */
1806 ir_rvalue
*bias
; /**< Floating point LOD bias */
1807 ir_rvalue
*sample_index
; /**< MSAA sample index */
1808 ir_rvalue
*component
; /**< Gather component selector */
1810 ir_rvalue
*dPdx
; /**< Partial derivative of coordinate wrt X */
1811 ir_rvalue
*dPdy
; /**< Partial derivative of coordinate wrt Y */
1817 struct ir_swizzle_mask
{
1824 * Number of components in the swizzle.
1826 unsigned num_components
:3;
1829 * Does the swizzle contain duplicate components?
1831 * L-value swizzles cannot contain duplicate components.
1833 unsigned has_duplicates
:1;
1837 class ir_swizzle
: public ir_rvalue
{
1839 ir_swizzle(ir_rvalue
*, unsigned x
, unsigned y
, unsigned z
, unsigned w
,
1842 ir_swizzle(ir_rvalue
*val
, const unsigned *components
, unsigned count
);
1844 ir_swizzle(ir_rvalue
*val
, ir_swizzle_mask mask
);
1846 virtual ir_swizzle
*clone(void *mem_ctx
, struct hash_table
*) const;
1848 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1851 * Construct an ir_swizzle from the textual representation. Can fail.
1853 static ir_swizzle
*create(ir_rvalue
*, const char *, unsigned vector_length
);
1855 virtual void accept(ir_visitor
*v
)
1860 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1862 virtual bool equals(ir_instruction
*ir
, enum ir_node_type ignore
= ir_type_unset
);
1864 bool is_lvalue() const
1866 return val
->is_lvalue() && !mask
.has_duplicates
;
1870 * Get the variable that is ultimately referenced by an r-value
1872 virtual ir_variable
*variable_referenced() const;
1875 ir_swizzle_mask mask
;
1879 * Initialize the mask component of a swizzle
1881 * This is used by the \c ir_swizzle constructors.
1883 void init_mask(const unsigned *components
, unsigned count
);
1887 class ir_dereference
: public ir_rvalue
{
1889 virtual ir_dereference
*clone(void *mem_ctx
, struct hash_table
*) const = 0;
1891 bool is_lvalue() const;
1894 * Get the variable that is ultimately referenced by an r-value
1896 virtual ir_variable
*variable_referenced() const = 0;
1899 ir_dereference(enum ir_node_type t
)
1906 class ir_dereference_variable
: public ir_dereference
{
1908 ir_dereference_variable(ir_variable
*var
);
1910 virtual ir_dereference_variable
*clone(void *mem_ctx
,
1911 struct hash_table
*) const;
1913 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1915 virtual bool equals(ir_instruction
*ir
, enum ir_node_type ignore
= ir_type_unset
);
1918 * Get the variable that is ultimately referenced by an r-value
1920 virtual ir_variable
*variable_referenced() const
1925 virtual ir_variable
*whole_variable_referenced()
1927 /* ir_dereference_variable objects always dereference the entire
1928 * variable. However, if this dereference is dereferenced by anything
1929 * else, the complete deferefernce chain is not a whole-variable
1930 * dereference. This method should only be called on the top most
1931 * ir_rvalue in a dereference chain.
1936 virtual void accept(ir_visitor
*v
)
1941 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1944 * Object being dereferenced.
1950 class ir_dereference_array
: public ir_dereference
{
1952 ir_dereference_array(ir_rvalue
*value
, ir_rvalue
*array_index
);
1954 ir_dereference_array(ir_variable
*var
, ir_rvalue
*array_index
);
1956 virtual ir_dereference_array
*clone(void *mem_ctx
,
1957 struct hash_table
*) const;
1959 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1961 virtual bool equals(ir_instruction
*ir
, enum ir_node_type ignore
= ir_type_unset
);
1964 * Get the variable that is ultimately referenced by an r-value
1966 virtual ir_variable
*variable_referenced() const
1968 return this->array
->variable_referenced();
1971 virtual void accept(ir_visitor
*v
)
1976 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1979 ir_rvalue
*array_index
;
1982 void set_array(ir_rvalue
*value
);
1986 class ir_dereference_record
: public ir_dereference
{
1988 ir_dereference_record(ir_rvalue
*value
, const char *field
);
1990 ir_dereference_record(ir_variable
*var
, const char *field
);
1992 virtual ir_dereference_record
*clone(void *mem_ctx
,
1993 struct hash_table
*) const;
1995 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1998 * Get the variable that is ultimately referenced by an r-value
2000 virtual ir_variable
*variable_referenced() const
2002 return this->record
->variable_referenced();
2005 virtual void accept(ir_visitor
*v
)
2010 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2018 * Data stored in an ir_constant
2020 union ir_constant_data
{
2028 class ir_constant
: public ir_rvalue
{
2030 ir_constant(const struct glsl_type
*type
, const ir_constant_data
*data
);
2031 ir_constant(bool b
, unsigned vector_elements
=1);
2032 ir_constant(unsigned int u
, unsigned vector_elements
=1);
2033 ir_constant(int i
, unsigned vector_elements
=1);
2034 ir_constant(float f
, unsigned vector_elements
=1);
2037 * Construct an ir_constant from a list of ir_constant values
2039 ir_constant(const struct glsl_type
*type
, exec_list
*values
);
2042 * Construct an ir_constant from a scalar component of another ir_constant
2044 * The new \c ir_constant inherits the type of the component from the
2048 * In the case of a matrix constant, the new constant is a scalar, \b not
2051 ir_constant(const ir_constant
*c
, unsigned i
);
2054 * Return a new ir_constant of the specified type containing all zeros.
2056 static ir_constant
*zero(void *mem_ctx
, const glsl_type
*type
);
2058 virtual ir_constant
*clone(void *mem_ctx
, struct hash_table
*) const;
2060 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2062 virtual void accept(ir_visitor
*v
)
2067 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2069 virtual bool equals(ir_instruction
*ir
, enum ir_node_type ignore
= ir_type_unset
);
2072 * Get a particular component of a constant as a specific type
2074 * This is useful, for example, to get a value from an integer constant
2075 * as a float or bool. This appears frequently when constructors are
2076 * called with all constant parameters.
2079 bool get_bool_component(unsigned i
) const;
2080 float get_float_component(unsigned i
) const;
2081 int get_int_component(unsigned i
) const;
2082 unsigned get_uint_component(unsigned i
) const;
2085 ir_constant
*get_array_element(unsigned i
) const;
2087 ir_constant
*get_record_field(const char *name
);
2090 * Copy the values on another constant at a given offset.
2092 * The offset is ignored for array or struct copies, it's only for
2093 * scalars or vectors into vectors or matrices.
2095 * With identical types on both sides and zero offset it's clone()
2096 * without creating a new object.
2099 void copy_offset(ir_constant
*src
, int offset
);
2102 * Copy the values on another constant at a given offset and
2103 * following an assign-like mask.
2105 * The mask is ignored for scalars.
2107 * Note that this function only handles what assign can handle,
2108 * i.e. at most a vector as source and a column of a matrix as
2112 void copy_masked_offset(ir_constant
*src
, int offset
, unsigned int mask
);
2115 * Determine whether a constant has the same value as another constant
2117 * \sa ir_constant::is_zero, ir_constant::is_one,
2118 * ir_constant::is_negative_one, ir_constant::is_basis
2120 bool has_value(const ir_constant
*) const;
2123 * Return true if this ir_constant represents the given value.
2125 * For vectors, this checks that each component is the given value.
2127 virtual bool is_value(float f
, int i
) const;
2128 virtual bool is_zero() const;
2129 virtual bool is_one() const;
2130 virtual bool is_negative_one() const;
2131 virtual bool is_basis() const;
2134 * Return true for constants that could be stored as 16-bit unsigned values.
2136 * Note that this will return true even for signed integer ir_constants, as
2137 * long as the value is non-negative and fits in 16-bits.
2139 virtual bool is_uint16_constant() const;
2142 * Value of the constant.
2144 * The field used to back the values supplied by the constant is determined
2145 * by the type associated with the \c ir_instruction. Constants may be
2146 * scalars, vectors, or matrices.
2148 union ir_constant_data value
;
2150 /* Array elements */
2151 ir_constant
**array_elements
;
2153 /* Structure fields */
2154 exec_list components
;
2158 * Parameterless constructor only used by the clone method
2164 * IR instruction to emit a vertex in a geometry shader.
2166 class ir_emit_vertex
: public ir_instruction
{
2168 ir_emit_vertex(ir_rvalue
*stream
)
2169 : ir_instruction(ir_type_emit_vertex
),
2175 virtual void accept(ir_visitor
*v
)
2180 virtual ir_emit_vertex
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2182 return new(mem_ctx
) ir_emit_vertex(this->stream
->clone(mem_ctx
, ht
));
2185 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2187 int stream_id() const
2189 return stream
->as_constant()->value
.i
[0];
2196 * IR instruction to complete the current primitive and start a new one in a
2199 class ir_end_primitive
: public ir_instruction
{
2201 ir_end_primitive(ir_rvalue
*stream
)
2202 : ir_instruction(ir_type_end_primitive
),
2208 virtual void accept(ir_visitor
*v
)
2213 virtual ir_end_primitive
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2215 return new(mem_ctx
) ir_end_primitive(this->stream
->clone(mem_ctx
, ht
));
2218 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2220 int stream_id() const
2222 return stream
->as_constant()->value
.i
[0];
2231 * Apply a visitor to each IR node in a list
2234 visit_exec_list(exec_list
*list
, ir_visitor
*visitor
);
2237 * Validate invariants on each IR node in a list
2239 void validate_ir_tree(exec_list
*instructions
);
2241 struct _mesa_glsl_parse_state
;
2242 struct gl_shader_program
;
2245 * Detect whether an unlinked shader contains static recursion
2247 * If the list of instructions is determined to contain static recursion,
2248 * \c _mesa_glsl_error will be called to emit error messages for each function
2249 * that is in the recursion cycle.
2252 detect_recursion_unlinked(struct _mesa_glsl_parse_state
*state
,
2253 exec_list
*instructions
);
2256 * Detect whether a linked shader contains static recursion
2258 * If the list of instructions is determined to contain static recursion,
2259 * \c link_error_printf will be called to emit error messages for each function
2260 * that is in the recursion cycle. In addition,
2261 * \c gl_shader_program::LinkStatus will be set to false.
2264 detect_recursion_linked(struct gl_shader_program
*prog
,
2265 exec_list
*instructions
);
2268 * Make a clone of each IR instruction in a list
2270 * \param in List of IR instructions that are to be cloned
2271 * \param out List to hold the cloned instructions
2274 clone_ir_list(void *mem_ctx
, exec_list
*out
, const exec_list
*in
);
2277 _mesa_glsl_initialize_variables(exec_list
*instructions
,
2278 struct _mesa_glsl_parse_state
*state
);
2281 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state
*state
);
2284 _mesa_glsl_initialize_builtin_functions();
2286 extern ir_function_signature
*
2287 _mesa_glsl_find_builtin_function(_mesa_glsl_parse_state
*state
,
2288 const char *name
, exec_list
*actual_parameters
);
2291 _mesa_glsl_get_builtin_function_shader(void);
2294 _mesa_glsl_release_functions(void);
2297 _mesa_glsl_release_builtin_functions(void);
2300 reparent_ir(exec_list
*list
, void *mem_ctx
);
2302 struct glsl_symbol_table
;
2305 import_prototypes(const exec_list
*source
, exec_list
*dest
,
2306 struct glsl_symbol_table
*symbols
, void *mem_ctx
);
2309 ir_has_call(ir_instruction
*ir
);
2312 do_set_program_inouts(exec_list
*instructions
, struct gl_program
*prog
,
2313 gl_shader_stage shader_stage
);
2316 prototype_string(const glsl_type
*return_type
, const char *name
,
2317 exec_list
*parameters
);
2320 mode_string(const ir_variable
*var
);
2323 * Built-in / reserved GL variables names start with "gl_"
2326 is_gl_identifier(const char *s
)
2328 return s
&& s
[0] == 'g' && s
[1] == 'l' && s
[2] == '_';
2332 #endif /* __cplusplus */
2334 extern void _mesa_print_ir(FILE *f
, struct exec_list
*instructions
,
2335 struct _mesa_glsl_parse_state
*state
);
2338 fprint_ir(FILE *f
, const void *instruction
);
2345 vertices_per_prim(GLenum prim
);