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 "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->u
.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->u
.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->u
.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->u
.max_ifc_array_access
[i
] == 0);
517 ralloc_free(this->u
.max_ifc_array_access
);
518 this->u
.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 * Get the max_ifc_array_access pointer
532 * A "set" function is not needed because the array is dynmically allocated
535 inline unsigned *get_max_ifc_array_access()
537 assert(this->data
._num_state_slots
== 0);
538 return this->u
.max_ifc_array_access
;
541 inline unsigned get_num_state_slots() const
543 assert(!this->is_interface_instance()
544 || this->data
._num_state_slots
== 0);
545 return this->data
._num_state_slots
;
548 inline void set_num_state_slots(unsigned n
)
550 assert(!this->is_interface_instance()
552 this->data
._num_state_slots
= n
;
555 inline ir_state_slot
*get_state_slots()
557 return this->is_interface_instance() ? NULL
: this->u
.state_slots
;
560 inline const ir_state_slot
*get_state_slots() const
562 return this->is_interface_instance() ? NULL
: this->u
.state_slots
;
565 inline ir_state_slot
*allocate_state_slots(unsigned n
)
567 assert(!this->is_interface_instance());
569 this->u
.state_slots
= ralloc_array(this, ir_state_slot
, n
);
570 this->data
._num_state_slots
= 0;
572 if (this->u
.state_slots
!= NULL
)
573 this->data
._num_state_slots
= n
;
575 return this->u
.state_slots
;
579 * Enable emitting extension warnings for this variable
581 void enable_extension_warning(const char *extension
);
584 * Get the extension warning string for this variable
586 * If warnings are not enabled, \c NULL is returned.
588 const char *get_extension_warning() const;
591 * Declared type of the variable
593 const struct glsl_type
*type
;
596 * Declared name of the variable
600 struct ir_variable_data
{
603 * Is the variable read-only?
605 * This is set for variables declared as \c const, shader inputs,
608 unsigned read_only
:1;
611 unsigned invariant
:1;
615 * Has this variable been used for reading or writing?
617 * Several GLSL semantic checks require knowledge of whether or not a
618 * variable has been used. For example, it is an error to redeclare a
619 * variable as invariant after it has been used.
621 * This is only maintained in the ast_to_hir.cpp path, not in
622 * Mesa's fixed function or ARB program paths.
627 * Has this variable been statically assigned?
629 * This answers whether the variable was assigned in any path of
630 * the shader during ast_to_hir. This doesn't answer whether it is
631 * still written after dead code removal, nor is it maintained in
632 * non-ast_to_hir.cpp (GLSL parsing) paths.
637 * Enum indicating how the variable was declared. See
638 * ir_var_declaration_type.
640 * This is used to detect certain kinds of illegal variable redeclarations.
642 unsigned how_declared
:2;
645 * Storage class of the variable.
647 * \sa ir_variable_mode
652 * Interpolation mode for shader inputs / outputs
654 * \sa ir_variable_interpolation
656 unsigned interpolation
:2;
659 * \name ARB_fragment_coord_conventions
662 unsigned origin_upper_left
:1;
663 unsigned pixel_center_integer
:1;
667 * Was the location explicitly set in the shader?
669 * If the location is explicitly set in the shader, it \b cannot be changed
670 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
673 unsigned explicit_location
:1;
674 unsigned explicit_index
:1;
677 * Was an initial binding explicitly set in the shader?
679 * If so, constant_value contains an integer ir_constant representing the
680 * initial binding point.
682 unsigned explicit_binding
:1;
685 * Does this variable have an initializer?
687 * This is used by the linker to cross-validiate initializers of global
690 unsigned has_initializer
:1;
693 * Is this variable a generic output or input that has not yet been matched
694 * up to a variable in another stage of the pipeline?
696 * This is used by the linker as scratch storage while assigning locations
697 * to generic inputs and outputs.
699 unsigned is_unmatched_generic_inout
:1;
702 * If non-zero, then this variable may be packed along with other variables
703 * into a single varying slot, so this offset should be applied when
704 * accessing components. For example, an offset of 1 means that the x
705 * component of this variable is actually stored in component y of the
706 * location specified by \c location.
708 unsigned location_frac
:2;
711 * Layout of the matrix. Uses glsl_matrix_layout values.
713 unsigned matrix_layout
:2;
716 * Non-zero if this variable was created by lowering a named interface
717 * block which was not an array.
719 * Note that this variable and \c from_named_ifc_block_array will never
722 unsigned from_named_ifc_block_nonarray
:1;
725 * Non-zero if this variable was created by lowering a named interface
726 * block which was an array.
728 * Note that this variable and \c from_named_ifc_block_nonarray will never
731 unsigned from_named_ifc_block_array
:1;
734 * Non-zero if the variable must be a shader input. This is useful for
735 * constraints on function parameters.
737 unsigned must_be_shader_input
:1;
740 * Output index for dual source blending.
743 * The GLSL spec only allows the values 0 or 1 for the index in \b dual
749 * \brief Layout qualifier for gl_FragDepth.
751 * This is not equal to \c ir_depth_layout_none if and only if this
752 * variable is \c gl_FragDepth and a layout qualifier is specified.
754 ir_depth_layout depth_layout
:3;
757 * ARB_shader_image_load_store qualifiers.
759 unsigned image_read_only
:1; /**< "readonly" qualifier. */
760 unsigned image_write_only
:1; /**< "writeonly" qualifier. */
761 unsigned image_coherent
:1;
762 unsigned image_volatile
:1;
763 unsigned image_restrict
:1;
766 * Emit a warning if this variable is accessed.
769 uint8_t warn_extension_index
;
772 /** Image internal format if specified explicitly, otherwise GL_NONE. */
773 uint16_t image_format
;
777 * Number of state slots used
780 * This could be stored in as few as 7-bits, if necessary. If it is made
781 * smaller, add an assertion to \c ir_variable::allocate_state_slots to
784 uint16_t _num_state_slots
;
788 * Initial binding point for a sampler, atomic, or UBO.
790 * For array types, this represents the binding point for the first element.
795 * Storage location of the base of this variable
797 * The precise meaning of this field depends on the nature of the variable.
799 * - Vertex shader input: one of the values from \c gl_vert_attrib.
800 * - Vertex shader output: one of the values from \c gl_varying_slot.
801 * - Geometry shader input: one of the values from \c gl_varying_slot.
802 * - Geometry shader output: one of the values from \c gl_varying_slot.
803 * - Fragment shader input: one of the values from \c gl_varying_slot.
804 * - Fragment shader output: one of the values from \c gl_frag_result.
805 * - Uniforms: Per-stage uniform slot number for default uniform block.
806 * - Uniforms: Index within the uniform block definition for UBO members.
807 * - Other: This field is not currently used.
809 * If the variable is a uniform, shader input, or shader output, and the
810 * slot has not been assigned, the value will be -1.
815 * Vertex stream output identifier.
820 * Location an atomic counter is stored at.
827 * Highest element accessed with a constant expression array index
829 * Not used for non-array variables.
831 unsigned max_array_access
;
834 * Allow (only) ir_variable direct access private members.
836 friend class ir_variable
;
840 * Value assigned in the initializer of a variable declared "const"
842 ir_constant
*constant_value
;
845 * Constant expression assigned in the initializer of the variable
848 * This field and \c ::constant_value are distinct. Even if the two fields
849 * refer to constants with the same value, they must point to separate
852 ir_constant
*constant_initializer
;
855 static const char *const warn_extension_table
[];
859 * For variables which satisfy the is_interface_instance() predicate,
860 * this points to an array of integers such that if the ith member of
861 * the interface block is an array, max_ifc_array_access[i] is the
862 * maximum array element of that member that has been accessed. If the
863 * ith member of the interface block is not an array,
864 * max_ifc_array_access[i] is unused.
866 * For variables whose type is not an interface block, this pointer is
869 unsigned *max_ifc_array_access
;
872 * Built-in state that backs this uniform
874 * Once set at variable creation, \c state_slots must remain invariant.
876 * If the variable is not a uniform, \c _num_state_slots will be zero
877 * and \c state_slots will be \c NULL.
879 ir_state_slot
*state_slots
;
883 * For variables that are in an interface block or are an instance of an
884 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
886 * \sa ir_variable::location
888 const glsl_type
*interface_type
;
892 * A function that returns whether a built-in function is available in the
893 * current shading language (based on version, ES or desktop, and extensions).
895 typedef bool (*builtin_available_predicate
)(const _mesa_glsl_parse_state
*);
899 * The representation of a function instance; may be the full definition or
900 * simply a prototype.
902 class ir_function_signature
: public ir_instruction
{
903 /* An ir_function_signature will be part of the list of signatures in
907 ir_function_signature(const glsl_type
*return_type
,
908 builtin_available_predicate builtin_avail
= NULL
);
910 virtual ir_function_signature
*clone(void *mem_ctx
,
911 struct hash_table
*ht
) const;
912 ir_function_signature
*clone_prototype(void *mem_ctx
,
913 struct hash_table
*ht
) const;
915 virtual void accept(ir_visitor
*v
)
920 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
923 * Attempt to evaluate this function as a constant expression,
924 * given a list of the actual parameters and the variable context.
925 * Returns NULL for non-built-ins.
927 ir_constant
*constant_expression_value(exec_list
*actual_parameters
, struct hash_table
*variable_context
);
930 * Get the name of the function for which this is a signature
932 const char *function_name() const;
935 * Get a handle to the function for which this is a signature
937 * There is no setter function, this function returns a \c const pointer,
938 * and \c ir_function_signature::_function is private for a reason. The
939 * only way to make a connection between a function and function signature
940 * is via \c ir_function::add_signature. This helps ensure that certain
941 * invariants (i.e., a function signature is in the list of signatures for
942 * its \c _function) are met.
944 * \sa ir_function::add_signature
946 inline const class ir_function
*function() const
948 return this->_function
;
952 * Check whether the qualifiers match between this signature's parameters
953 * and the supplied parameter list. If not, returns the name of the first
954 * parameter with mismatched qualifiers (for use in error messages).
956 const char *qualifiers_match(exec_list
*params
);
959 * Replace the current parameter list with the given one. This is useful
960 * if the current information came from a prototype, and either has invalid
961 * or missing parameter names.
963 void replace_parameters(exec_list
*new_params
);
966 * Function return type.
968 * \note This discards the optional precision qualifier.
970 const struct glsl_type
*return_type
;
973 * List of ir_variable of function parameters.
975 * This represents the storage. The paramaters passed in a particular
976 * call will be in ir_call::actual_paramaters.
978 struct exec_list parameters
;
980 /** Whether or not this function has a body (which may be empty). */
981 unsigned is_defined
:1;
983 /** Whether or not this function signature is a built-in. */
984 bool is_builtin() const;
987 * Whether or not this function is an intrinsic to be implemented
992 /** Whether or not a built-in is available for this shader. */
993 bool is_builtin_available(const _mesa_glsl_parse_state
*state
) const;
995 /** Body of instructions in the function. */
996 struct exec_list body
;
1000 * A function pointer to a predicate that answers whether a built-in
1001 * function is available in the current shader. NULL if not a built-in.
1003 builtin_available_predicate builtin_avail
;
1005 /** Function of which this signature is one overload. */
1006 class ir_function
*_function
;
1008 /** Function signature of which this one is a prototype clone */
1009 const ir_function_signature
*origin
;
1011 friend class ir_function
;
1014 * Helper function to run a list of instructions for constant
1015 * expression evaluation.
1017 * The hash table represents the values of the visible variables.
1018 * There are no scoping issues because the table is indexed on
1019 * ir_variable pointers, not variable names.
1021 * Returns false if the expression is not constant, true otherwise,
1022 * and the value in *result if result is non-NULL.
1024 bool constant_expression_evaluate_expression_list(const struct exec_list
&body
,
1025 struct hash_table
*variable_context
,
1026 ir_constant
**result
);
1031 * Header for tracking multiple overloaded functions with the same name.
1032 * Contains a list of ir_function_signatures representing each of the
1035 class ir_function
: public ir_instruction
{
1037 ir_function(const char *name
);
1039 virtual ir_function
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1041 virtual void accept(ir_visitor
*v
)
1046 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1048 void add_signature(ir_function_signature
*sig
)
1050 sig
->_function
= this;
1051 this->signatures
.push_tail(sig
);
1055 * Find a signature that matches a set of actual parameters, taking implicit
1056 * conversions into account. Also flags whether the match was exact.
1058 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1059 const exec_list
*actual_param
,
1060 bool allow_builtins
,
1061 bool *match_is_exact
);
1064 * Find a signature that matches a set of actual parameters, taking implicit
1065 * conversions into account.
1067 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1068 const exec_list
*actual_param
,
1069 bool allow_builtins
);
1072 * Find a signature that exactly matches a set of actual parameters without
1073 * any implicit type conversions.
1075 ir_function_signature
*exact_matching_signature(_mesa_glsl_parse_state
*state
,
1076 const exec_list
*actual_ps
);
1079 * Name of the function.
1083 /** Whether or not this function has a signature that isn't a built-in. */
1084 bool has_user_signature();
1087 * List of ir_function_signature for each overloaded function with this name.
1089 struct exec_list signatures
;
1092 inline const char *ir_function_signature::function_name() const
1094 return this->_function
->name
;
1100 * IR instruction representing high-level if-statements
1102 class ir_if
: public ir_instruction
{
1104 ir_if(ir_rvalue
*condition
)
1105 : ir_instruction(ir_type_if
), condition(condition
)
1109 virtual ir_if
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1111 virtual void accept(ir_visitor
*v
)
1116 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1118 ir_rvalue
*condition
;
1119 /** List of ir_instruction for the body of the then branch */
1120 exec_list then_instructions
;
1121 /** List of ir_instruction for the body of the else branch */
1122 exec_list else_instructions
;
1127 * IR instruction representing a high-level loop structure.
1129 class ir_loop
: public ir_instruction
{
1133 virtual ir_loop
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1135 virtual void accept(ir_visitor
*v
)
1140 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1142 /** List of ir_instruction that make up the body of the loop. */
1143 exec_list body_instructions
;
1147 class ir_assignment
: public ir_instruction
{
1149 ir_assignment(ir_rvalue
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
= NULL
);
1152 * Construct an assignment with an explicit write mask
1155 * Since a write mask is supplied, the LHS must already be a bare
1156 * \c ir_dereference. The cannot be any swizzles in the LHS.
1158 ir_assignment(ir_dereference
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
,
1159 unsigned write_mask
);
1161 virtual ir_assignment
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1163 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1165 virtual void accept(ir_visitor
*v
)
1170 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1173 * Get a whole variable written by an assignment
1175 * If the LHS of the assignment writes a whole variable, the variable is
1176 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1179 * - Assigning to a scalar
1180 * - Assigning to all components of a vector
1181 * - Whole array (or matrix) assignment
1182 * - Whole structure assignment
1184 ir_variable
*whole_variable_written();
1187 * Set the LHS of an assignment
1189 void set_lhs(ir_rvalue
*lhs
);
1192 * Left-hand side of the assignment.
1194 * This should be treated as read only. If you need to set the LHS of an
1195 * assignment, use \c ir_assignment::set_lhs.
1197 ir_dereference
*lhs
;
1200 * Value being assigned
1205 * Optional condition for the assignment.
1207 ir_rvalue
*condition
;
1211 * Component mask written
1213 * For non-vector types in the LHS, this field will be zero. For vector
1214 * types, a bit will be set for each component that is written. Note that
1215 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1217 * A partially-set write mask means that each enabled channel gets
1218 * the value from a consecutive channel of the rhs. For example,
1219 * to write just .xyw of gl_FrontColor with color:
1221 * (assign (constant bool (1)) (xyw)
1222 * (var_ref gl_FragColor)
1223 * (swiz xyw (var_ref color)))
1225 unsigned write_mask
:4;
1228 /* Update ir_expression::get_num_operands() and operator_strs when
1229 * updating this list.
1231 enum ir_expression_operation
{
1240 ir_unop_exp
, /**< Log base e on gentype */
1241 ir_unop_log
, /**< Natural log on gentype */
1244 ir_unop_f2i
, /**< Float-to-integer conversion. */
1245 ir_unop_f2u
, /**< Float-to-unsigned conversion. */
1246 ir_unop_i2f
, /**< Integer-to-float conversion. */
1247 ir_unop_f2b
, /**< Float-to-boolean conversion */
1248 ir_unop_b2f
, /**< Boolean-to-float conversion */
1249 ir_unop_i2b
, /**< int-to-boolean conversion */
1250 ir_unop_b2i
, /**< Boolean-to-int conversion */
1251 ir_unop_u2f
, /**< Unsigned-to-float conversion. */
1252 ir_unop_i2u
, /**< Integer-to-unsigned conversion. */
1253 ir_unop_u2i
, /**< Unsigned-to-integer conversion. */
1254 ir_unop_bitcast_i2f
, /**< Bit-identical int-to-float "conversion" */
1255 ir_unop_bitcast_f2i
, /**< Bit-identical float-to-int "conversion" */
1256 ir_unop_bitcast_u2f
, /**< Bit-identical uint-to-float "conversion" */
1257 ir_unop_bitcast_f2u
, /**< Bit-identical float-to-uint "conversion" */
1261 * \name Unary floating-point rounding operations.
1272 * \name Trigonometric operations.
1277 ir_unop_sin_reduced
, /**< Reduced range sin. [-pi, pi] */
1278 ir_unop_cos_reduced
, /**< Reduced range cos. [-pi, pi] */
1282 * \name Partial derivatives.
1286 ir_unop_dFdx_coarse
,
1289 ir_unop_dFdy_coarse
,
1294 * \name Floating point pack and unpack operations.
1297 ir_unop_pack_snorm_2x16
,
1298 ir_unop_pack_snorm_4x8
,
1299 ir_unop_pack_unorm_2x16
,
1300 ir_unop_pack_unorm_4x8
,
1301 ir_unop_pack_half_2x16
,
1302 ir_unop_unpack_snorm_2x16
,
1303 ir_unop_unpack_snorm_4x8
,
1304 ir_unop_unpack_unorm_2x16
,
1305 ir_unop_unpack_unorm_4x8
,
1306 ir_unop_unpack_half_2x16
,
1310 * \name Lowered floating point unpacking operations.
1312 * \see lower_packing_builtins_visitor::split_unpack_half_2x16
1315 ir_unop_unpack_half_2x16_split_x
,
1316 ir_unop_unpack_half_2x16_split_y
,
1320 * \name Bit operations, part of ARB_gpu_shader5.
1323 ir_unop_bitfield_reverse
,
1333 * Interpolate fs input at centroid
1335 * operand0 is the fs input.
1337 ir_unop_interpolate_at_centroid
,
1340 * A sentinel marking the last of the unary operations.
1342 ir_last_unop
= ir_unop_interpolate_at_centroid
,
1346 ir_binop_mul
, /**< Floating-point or low 32-bit integer multiply. */
1347 ir_binop_imul_high
, /**< Calculates the high 32-bits of a 64-bit multiply. */
1351 * Returns the carry resulting from the addition of the two arguments.
1358 * Returns the borrow resulting from the subtraction of the second argument
1359 * from the first argument.
1366 * Takes one of two combinations of arguments:
1369 * - mod(vecN, float)
1371 * Does not take integer types.
1376 * \name Binary comparison operators which return a boolean vector.
1377 * The type of both operands must be equal.
1387 * Returns single boolean for whether all components of operands[0]
1388 * equal the components of operands[1].
1392 * Returns single boolean for whether any component of operands[0]
1393 * is not equal to the corresponding component of operands[1].
1395 ir_binop_any_nequal
,
1399 * \name Bit-wise binary operations.
1420 * \name Lowered floating point packing operations.
1422 * \see lower_packing_builtins_visitor::split_pack_half_2x16
1425 ir_binop_pack_half_2x16_split
,
1429 * \name First half of a lowered bitfieldInsert() operation.
1431 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1438 * Load a value the size of a given GLSL type from a uniform block.
1440 * operand0 is the ir_constant uniform block index in the linked shader.
1441 * operand1 is a byte offset within the uniform block.
1446 * \name Multiplies a number by two to a power, part of ARB_gpu_shader5.
1453 * Extract a scalar from a vector
1455 * operand0 is the vector
1456 * operand1 is the index of the field to read from operand0
1458 ir_binop_vector_extract
,
1461 * Interpolate fs input at offset
1463 * operand0 is the fs input
1464 * operand1 is the offset from the pixel center
1466 ir_binop_interpolate_at_offset
,
1469 * Interpolate fs input at sample position
1471 * operand0 is the fs input
1472 * operand1 is the sample ID
1474 ir_binop_interpolate_at_sample
,
1477 * A sentinel marking the last of the binary operations.
1479 ir_last_binop
= ir_binop_interpolate_at_sample
,
1482 * \name Fused floating-point multiply-add, part of ARB_gpu_shader5.
1491 * \name Conditional Select
1493 * A vector conditional select instruction (like ?:, but operating per-
1494 * component on vectors).
1496 * \see lower_instructions_visitor::ldexp_to_arith
1503 * \name Second half of a lowered bitfieldInsert() operation.
1505 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1511 ir_triop_bitfield_extract
,
1514 * Generate a value with one field of a vector changed
1516 * operand0 is the vector
1517 * operand1 is the value to write into the vector result
1518 * operand2 is the index in operand0 to be modified
1520 ir_triop_vector_insert
,
1523 * A sentinel marking the last of the ternary operations.
1525 ir_last_triop
= ir_triop_vector_insert
,
1527 ir_quadop_bitfield_insert
,
1532 * A sentinel marking the last of the ternary operations.
1534 ir_last_quadop
= ir_quadop_vector
,
1537 * A sentinel marking the last of all operations.
1539 ir_last_opcode
= ir_quadop_vector
1542 class ir_expression
: public ir_rvalue
{
1544 ir_expression(int op
, const struct glsl_type
*type
,
1545 ir_rvalue
*op0
, ir_rvalue
*op1
= NULL
,
1546 ir_rvalue
*op2
= NULL
, ir_rvalue
*op3
= NULL
);
1549 * Constructor for unary operation expressions
1551 ir_expression(int op
, ir_rvalue
*);
1554 * Constructor for binary operation expressions
1556 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
);
1559 * Constructor for ternary operation expressions
1561 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
, ir_rvalue
*op2
);
1563 virtual bool equals(ir_instruction
*ir
, enum ir_node_type ignore
= ir_type_unset
);
1565 virtual ir_expression
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1568 * Attempt to constant-fold the expression
1570 * The "variable_context" hash table links ir_variable * to ir_constant *
1571 * that represent the variables' values. \c NULL represents an empty
1574 * If the expression cannot be constant folded, this method will return
1577 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1580 * Determine the number of operands used by an expression
1582 static unsigned int get_num_operands(ir_expression_operation
);
1585 * Determine the number of operands used by an expression
1587 unsigned int get_num_operands() const
1589 return (this->operation
== ir_quadop_vector
)
1590 ? this->type
->vector_elements
: get_num_operands(operation
);
1594 * Return whether the expression operates on vectors horizontally.
1596 bool is_horizontal() const
1598 return operation
== ir_binop_all_equal
||
1599 operation
== ir_binop_any_nequal
||
1600 operation
== ir_unop_any
||
1601 operation
== ir_binop_dot
||
1602 operation
== ir_quadop_vector
;
1606 * Return a string representing this expression's operator.
1608 const char *operator_string();
1611 * Return a string representing this expression's operator.
1613 static const char *operator_string(ir_expression_operation
);
1617 * Do a reverse-lookup to translate the given string into an operator.
1619 static ir_expression_operation
get_operator(const char *);
1621 virtual void accept(ir_visitor
*v
)
1626 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1628 ir_expression_operation operation
;
1629 ir_rvalue
*operands
[4];
1634 * HIR instruction representing a high-level function call, containing a list
1635 * of parameters and returning a value in the supplied temporary.
1637 class ir_call
: public ir_instruction
{
1639 ir_call(ir_function_signature
*callee
,
1640 ir_dereference_variable
*return_deref
,
1641 exec_list
*actual_parameters
)
1642 : ir_instruction(ir_type_call
), return_deref(return_deref
), callee(callee
)
1644 assert(callee
->return_type
!= NULL
);
1645 actual_parameters
->move_nodes_to(& this->actual_parameters
);
1646 this->use_builtin
= callee
->is_builtin();
1649 virtual ir_call
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1651 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1653 virtual void accept(ir_visitor
*v
)
1658 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1661 * Get the name of the function being called.
1663 const char *callee_name() const
1665 return callee
->function_name();
1669 * Generates an inline version of the function before @ir,
1670 * storing the return value in return_deref.
1672 void generate_inline(ir_instruction
*ir
);
1675 * Storage for the function's return value.
1676 * This must be NULL if the return type is void.
1678 ir_dereference_variable
*return_deref
;
1681 * The specific function signature being called.
1683 ir_function_signature
*callee
;
1685 /* List of ir_rvalue of paramaters passed in this call. */
1686 exec_list actual_parameters
;
1688 /** Should this call only bind to a built-in function? */
1694 * \name Jump-like IR instructions.
1696 * These include \c break, \c continue, \c return, and \c discard.
1699 class ir_jump
: public ir_instruction
{
1701 ir_jump(enum ir_node_type t
)
1707 class ir_return
: public ir_jump
{
1710 : ir_jump(ir_type_return
), value(NULL
)
1714 ir_return(ir_rvalue
*value
)
1715 : ir_jump(ir_type_return
), value(value
)
1719 virtual ir_return
*clone(void *mem_ctx
, struct hash_table
*) const;
1721 ir_rvalue
*get_value() const
1726 virtual void accept(ir_visitor
*v
)
1731 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1738 * Jump instructions used inside loops
1740 * These include \c break and \c continue. The \c break within a loop is
1741 * different from the \c break within a switch-statement.
1743 * \sa ir_switch_jump
1745 class ir_loop_jump
: public ir_jump
{
1752 ir_loop_jump(jump_mode mode
)
1753 : ir_jump(ir_type_loop_jump
)
1758 virtual ir_loop_jump
*clone(void *mem_ctx
, struct hash_table
*) const;
1760 virtual void accept(ir_visitor
*v
)
1765 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1767 bool is_break() const
1769 return mode
== jump_break
;
1772 bool is_continue() const
1774 return mode
== jump_continue
;
1777 /** Mode selector for the jump instruction. */
1778 enum jump_mode mode
;
1782 * IR instruction representing discard statements.
1784 class ir_discard
: public ir_jump
{
1787 : ir_jump(ir_type_discard
)
1789 this->condition
= NULL
;
1792 ir_discard(ir_rvalue
*cond
)
1793 : ir_jump(ir_type_discard
)
1795 this->condition
= cond
;
1798 virtual ir_discard
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1800 virtual void accept(ir_visitor
*v
)
1805 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1807 ir_rvalue
*condition
;
1813 * Texture sampling opcodes used in ir_texture
1815 enum ir_texture_opcode
{
1816 ir_tex
, /**< Regular texture look-up */
1817 ir_txb
, /**< Texture look-up with LOD bias */
1818 ir_txl
, /**< Texture look-up with explicit LOD */
1819 ir_txd
, /**< Texture look-up with partial derivatvies */
1820 ir_txf
, /**< Texel fetch with explicit LOD */
1821 ir_txf_ms
, /**< Multisample texture fetch */
1822 ir_txs
, /**< Texture size */
1823 ir_lod
, /**< Texture lod query */
1824 ir_tg4
, /**< Texture gather */
1825 ir_query_levels
/**< Texture levels query */
1830 * IR instruction to sample a texture
1832 * The specific form of the IR instruction depends on the \c mode value
1833 * selected from \c ir_texture_opcodes. In the printed IR, these will
1836 * Texel offset (0 or an expression)
1837 * | Projection divisor
1838 * | | Shadow comparitor
1841 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1842 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1843 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1844 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1845 * (txf <type> <sampler> <coordinate> 0 <lod>)
1847 * <type> <sampler> <coordinate> <sample_index>)
1848 * (txs <type> <sampler> <lod>)
1849 * (lod <type> <sampler> <coordinate>)
1850 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
1851 * (query_levels <type> <sampler>)
1853 class ir_texture
: public ir_rvalue
{
1855 ir_texture(enum ir_texture_opcode op
)
1856 : ir_rvalue(ir_type_texture
),
1857 op(op
), sampler(NULL
), coordinate(NULL
), projector(NULL
),
1858 shadow_comparitor(NULL
), offset(NULL
)
1860 memset(&lod_info
, 0, sizeof(lod_info
));
1863 virtual ir_texture
*clone(void *mem_ctx
, struct hash_table
*) const;
1865 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1867 virtual void accept(ir_visitor
*v
)
1872 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1874 virtual bool equals(ir_instruction
*ir
, enum ir_node_type ignore
= ir_type_unset
);
1877 * Return a string representing the ir_texture_opcode.
1879 const char *opcode_string();
1881 /** Set the sampler and type. */
1882 void set_sampler(ir_dereference
*sampler
, const glsl_type
*type
);
1885 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1887 static ir_texture_opcode
get_opcode(const char *);
1889 enum ir_texture_opcode op
;
1891 /** Sampler to use for the texture access. */
1892 ir_dereference
*sampler
;
1894 /** Texture coordinate to sample */
1895 ir_rvalue
*coordinate
;
1898 * Value used for projective divide.
1900 * If there is no projective divide (the common case), this will be
1901 * \c NULL. Optimization passes should check for this to point to a constant
1902 * of 1.0 and replace that with \c NULL.
1904 ir_rvalue
*projector
;
1907 * Coordinate used for comparison on shadow look-ups.
1909 * If there is no shadow comparison, this will be \c NULL. For the
1910 * \c ir_txf opcode, this *must* be \c NULL.
1912 ir_rvalue
*shadow_comparitor
;
1914 /** Texel offset. */
1918 ir_rvalue
*lod
; /**< Floating point LOD */
1919 ir_rvalue
*bias
; /**< Floating point LOD bias */
1920 ir_rvalue
*sample_index
; /**< MSAA sample index */
1921 ir_rvalue
*component
; /**< Gather component selector */
1923 ir_rvalue
*dPdx
; /**< Partial derivative of coordinate wrt X */
1924 ir_rvalue
*dPdy
; /**< Partial derivative of coordinate wrt Y */
1930 struct ir_swizzle_mask
{
1937 * Number of components in the swizzle.
1939 unsigned num_components
:3;
1942 * Does the swizzle contain duplicate components?
1944 * L-value swizzles cannot contain duplicate components.
1946 unsigned has_duplicates
:1;
1950 class ir_swizzle
: public ir_rvalue
{
1952 ir_swizzle(ir_rvalue
*, unsigned x
, unsigned y
, unsigned z
, unsigned w
,
1955 ir_swizzle(ir_rvalue
*val
, const unsigned *components
, unsigned count
);
1957 ir_swizzle(ir_rvalue
*val
, ir_swizzle_mask mask
);
1959 virtual ir_swizzle
*clone(void *mem_ctx
, struct hash_table
*) const;
1961 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1964 * Construct an ir_swizzle from the textual representation. Can fail.
1966 static ir_swizzle
*create(ir_rvalue
*, const char *, unsigned vector_length
);
1968 virtual void accept(ir_visitor
*v
)
1973 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1975 virtual bool equals(ir_instruction
*ir
, enum ir_node_type ignore
= ir_type_unset
);
1977 bool is_lvalue() const
1979 return val
->is_lvalue() && !mask
.has_duplicates
;
1983 * Get the variable that is ultimately referenced by an r-value
1985 virtual ir_variable
*variable_referenced() const;
1988 ir_swizzle_mask mask
;
1992 * Initialize the mask component of a swizzle
1994 * This is used by the \c ir_swizzle constructors.
1996 void init_mask(const unsigned *components
, unsigned count
);
2000 class ir_dereference
: public ir_rvalue
{
2002 virtual ir_dereference
*clone(void *mem_ctx
, struct hash_table
*) const = 0;
2004 bool is_lvalue() const;
2007 * Get the variable that is ultimately referenced by an r-value
2009 virtual ir_variable
*variable_referenced() const = 0;
2012 ir_dereference(enum ir_node_type t
)
2019 class ir_dereference_variable
: public ir_dereference
{
2021 ir_dereference_variable(ir_variable
*var
);
2023 virtual ir_dereference_variable
*clone(void *mem_ctx
,
2024 struct hash_table
*) const;
2026 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2028 virtual bool equals(ir_instruction
*ir
, enum ir_node_type ignore
= ir_type_unset
);
2031 * Get the variable that is ultimately referenced by an r-value
2033 virtual ir_variable
*variable_referenced() const
2038 virtual ir_variable
*whole_variable_referenced()
2040 /* ir_dereference_variable objects always dereference the entire
2041 * variable. However, if this dereference is dereferenced by anything
2042 * else, the complete deferefernce chain is not a whole-variable
2043 * dereference. This method should only be called on the top most
2044 * ir_rvalue in a dereference chain.
2049 virtual void accept(ir_visitor
*v
)
2054 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2057 * Object being dereferenced.
2063 class ir_dereference_array
: public ir_dereference
{
2065 ir_dereference_array(ir_rvalue
*value
, ir_rvalue
*array_index
);
2067 ir_dereference_array(ir_variable
*var
, ir_rvalue
*array_index
);
2069 virtual ir_dereference_array
*clone(void *mem_ctx
,
2070 struct hash_table
*) const;
2072 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2074 virtual bool equals(ir_instruction
*ir
, enum ir_node_type ignore
= ir_type_unset
);
2077 * Get the variable that is ultimately referenced by an r-value
2079 virtual ir_variable
*variable_referenced() const
2081 return this->array
->variable_referenced();
2084 virtual void accept(ir_visitor
*v
)
2089 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2092 ir_rvalue
*array_index
;
2095 void set_array(ir_rvalue
*value
);
2099 class ir_dereference_record
: public ir_dereference
{
2101 ir_dereference_record(ir_rvalue
*value
, const char *field
);
2103 ir_dereference_record(ir_variable
*var
, const char *field
);
2105 virtual ir_dereference_record
*clone(void *mem_ctx
,
2106 struct hash_table
*) const;
2108 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2111 * Get the variable that is ultimately referenced by an r-value
2113 virtual ir_variable
*variable_referenced() const
2115 return this->record
->variable_referenced();
2118 virtual void accept(ir_visitor
*v
)
2123 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2131 * Data stored in an ir_constant
2133 union ir_constant_data
{
2141 class ir_constant
: public ir_rvalue
{
2143 ir_constant(const struct glsl_type
*type
, const ir_constant_data
*data
);
2144 ir_constant(bool b
, unsigned vector_elements
=1);
2145 ir_constant(unsigned int u
, unsigned vector_elements
=1);
2146 ir_constant(int i
, unsigned vector_elements
=1);
2147 ir_constant(float f
, unsigned vector_elements
=1);
2150 * Construct an ir_constant from a list of ir_constant values
2152 ir_constant(const struct glsl_type
*type
, exec_list
*values
);
2155 * Construct an ir_constant from a scalar component of another ir_constant
2157 * The new \c ir_constant inherits the type of the component from the
2161 * In the case of a matrix constant, the new constant is a scalar, \b not
2164 ir_constant(const ir_constant
*c
, unsigned i
);
2167 * Return a new ir_constant of the specified type containing all zeros.
2169 static ir_constant
*zero(void *mem_ctx
, const glsl_type
*type
);
2171 virtual ir_constant
*clone(void *mem_ctx
, struct hash_table
*) const;
2173 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2175 virtual void accept(ir_visitor
*v
)
2180 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2182 virtual bool equals(ir_instruction
*ir
, enum ir_node_type ignore
= ir_type_unset
);
2185 * Get a particular component of a constant as a specific type
2187 * This is useful, for example, to get a value from an integer constant
2188 * as a float or bool. This appears frequently when constructors are
2189 * called with all constant parameters.
2192 bool get_bool_component(unsigned i
) const;
2193 float get_float_component(unsigned i
) const;
2194 int get_int_component(unsigned i
) const;
2195 unsigned get_uint_component(unsigned i
) const;
2198 ir_constant
*get_array_element(unsigned i
) const;
2200 ir_constant
*get_record_field(const char *name
);
2203 * Copy the values on another constant at a given offset.
2205 * The offset is ignored for array or struct copies, it's only for
2206 * scalars or vectors into vectors or matrices.
2208 * With identical types on both sides and zero offset it's clone()
2209 * without creating a new object.
2212 void copy_offset(ir_constant
*src
, int offset
);
2215 * Copy the values on another constant at a given offset and
2216 * following an assign-like mask.
2218 * The mask is ignored for scalars.
2220 * Note that this function only handles what assign can handle,
2221 * i.e. at most a vector as source and a column of a matrix as
2225 void copy_masked_offset(ir_constant
*src
, int offset
, unsigned int mask
);
2228 * Determine whether a constant has the same value as another constant
2230 * \sa ir_constant::is_zero, ir_constant::is_one,
2231 * ir_constant::is_negative_one, ir_constant::is_basis
2233 bool has_value(const ir_constant
*) const;
2236 * Return true if this ir_constant represents the given value.
2238 * For vectors, this checks that each component is the given value.
2240 virtual bool is_value(float f
, int i
) const;
2241 virtual bool is_zero() const;
2242 virtual bool is_one() const;
2243 virtual bool is_negative_one() const;
2244 virtual bool is_basis() const;
2247 * Return true for constants that could be stored as 16-bit unsigned values.
2249 * Note that this will return true even for signed integer ir_constants, as
2250 * long as the value is non-negative and fits in 16-bits.
2252 virtual bool is_uint16_constant() const;
2255 * Value of the constant.
2257 * The field used to back the values supplied by the constant is determined
2258 * by the type associated with the \c ir_instruction. Constants may be
2259 * scalars, vectors, or matrices.
2261 union ir_constant_data value
;
2263 /* Array elements */
2264 ir_constant
**array_elements
;
2266 /* Structure fields */
2267 exec_list components
;
2271 * Parameterless constructor only used by the clone method
2277 * IR instruction to emit a vertex in a geometry shader.
2279 class ir_emit_vertex
: public ir_instruction
{
2281 ir_emit_vertex(ir_rvalue
*stream
)
2282 : ir_instruction(ir_type_emit_vertex
),
2288 virtual void accept(ir_visitor
*v
)
2293 virtual ir_emit_vertex
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2295 return new(mem_ctx
) ir_emit_vertex(this->stream
->clone(mem_ctx
, ht
));
2298 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2300 int stream_id() const
2302 return stream
->as_constant()->value
.i
[0];
2309 * IR instruction to complete the current primitive and start a new one in a
2312 class ir_end_primitive
: public ir_instruction
{
2314 ir_end_primitive(ir_rvalue
*stream
)
2315 : ir_instruction(ir_type_end_primitive
),
2321 virtual void accept(ir_visitor
*v
)
2326 virtual ir_end_primitive
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2328 return new(mem_ctx
) ir_end_primitive(this->stream
->clone(mem_ctx
, ht
));
2331 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2333 int stream_id() const
2335 return stream
->as_constant()->value
.i
[0];
2344 * Apply a visitor to each IR node in a list
2347 visit_exec_list(exec_list
*list
, ir_visitor
*visitor
);
2350 * Validate invariants on each IR node in a list
2352 void validate_ir_tree(exec_list
*instructions
);
2354 struct _mesa_glsl_parse_state
;
2355 struct gl_shader_program
;
2358 * Detect whether an unlinked shader contains static recursion
2360 * If the list of instructions is determined to contain static recursion,
2361 * \c _mesa_glsl_error will be called to emit error messages for each function
2362 * that is in the recursion cycle.
2365 detect_recursion_unlinked(struct _mesa_glsl_parse_state
*state
,
2366 exec_list
*instructions
);
2369 * Detect whether a linked shader contains static recursion
2371 * If the list of instructions is determined to contain static recursion,
2372 * \c link_error_printf will be called to emit error messages for each function
2373 * that is in the recursion cycle. In addition,
2374 * \c gl_shader_program::LinkStatus will be set to false.
2377 detect_recursion_linked(struct gl_shader_program
*prog
,
2378 exec_list
*instructions
);
2381 * Make a clone of each IR instruction in a list
2383 * \param in List of IR instructions that are to be cloned
2384 * \param out List to hold the cloned instructions
2387 clone_ir_list(void *mem_ctx
, exec_list
*out
, const exec_list
*in
);
2390 _mesa_glsl_initialize_variables(exec_list
*instructions
,
2391 struct _mesa_glsl_parse_state
*state
);
2394 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state
*state
);
2397 _mesa_glsl_initialize_builtin_functions();
2399 extern ir_function_signature
*
2400 _mesa_glsl_find_builtin_function(_mesa_glsl_parse_state
*state
,
2401 const char *name
, exec_list
*actual_parameters
);
2404 _mesa_glsl_get_builtin_function_shader(void);
2407 _mesa_glsl_release_functions(void);
2410 _mesa_glsl_release_builtin_functions(void);
2413 reparent_ir(exec_list
*list
, void *mem_ctx
);
2415 struct glsl_symbol_table
;
2418 import_prototypes(const exec_list
*source
, exec_list
*dest
,
2419 struct glsl_symbol_table
*symbols
, void *mem_ctx
);
2422 ir_has_call(ir_instruction
*ir
);
2425 do_set_program_inouts(exec_list
*instructions
, struct gl_program
*prog
,
2426 gl_shader_stage shader_stage
);
2429 prototype_string(const glsl_type
*return_type
, const char *name
,
2430 exec_list
*parameters
);
2433 mode_string(const ir_variable
*var
);
2436 * Built-in / reserved GL variables names start with "gl_"
2439 is_gl_identifier(const char *s
)
2441 return s
&& s
[0] == 'g' && s
[1] == 'l' && s
[2] == '_';
2445 #endif /* __cplusplus */
2447 extern void _mesa_print_ir(FILE *f
, struct exec_list
*instructions
,
2448 struct _mesa_glsl_parse_state
*state
);
2451 fprint_ir(FILE *f
, const void *instruction
);
2458 vertices_per_prim(GLenum prim
);