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;
112 bool is_rvalue() const
114 return ir_type
== ir_type_dereference_array
||
115 ir_type
== ir_type_dereference_record
||
116 ir_type
== ir_type_dereference_variable
||
117 ir_type
== ir_type_constant
||
118 ir_type
== ir_type_expression
||
119 ir_type
== ir_type_swizzle
||
120 ir_type
== ir_type_texture
;
123 bool is_dereference() const
125 return ir_type
== ir_type_dereference_array
||
126 ir_type
== ir_type_dereference_record
||
127 ir_type
== ir_type_dereference_variable
;
132 return ir_type
== ir_type_loop_jump
||
133 ir_type
== ir_type_return
||
134 ir_type
== ir_type_discard
;
138 * \name IR instruction downcast functions
140 * These functions either cast the object to a derived class or return
141 * \c NULL if the object's type does not match the specified derived class.
142 * Additional downcast functions will be added as needed.
145 class ir_rvalue
*as_rvalue()
147 assume(this != NULL
);
148 if (ir_type
== ir_type_dereference_array
||
149 ir_type
== ir_type_dereference_record
||
150 ir_type
== ir_type_dereference_variable
||
151 ir_type
== ir_type_constant
||
152 ir_type
== ir_type_expression
||
153 ir_type
== ir_type_swizzle
||
154 ir_type
== ir_type_texture
)
155 return (class ir_rvalue
*) this;
159 class ir_dereference
*as_dereference()
161 assume(this != NULL
);
162 if (ir_type
== ir_type_dereference_array
||
163 ir_type
== ir_type_dereference_record
||
164 ir_type
== ir_type_dereference_variable
)
165 return (class ir_dereference
*) this;
169 class ir_jump
*as_jump()
171 assume(this != NULL
);
172 if (ir_type
== ir_type_loop_jump
||
173 ir_type
== ir_type_return
||
174 ir_type
== ir_type_discard
)
175 return (class ir_jump
*) this;
179 #define AS_CHILD(TYPE) \
180 class ir_##TYPE * as_##TYPE() \
182 assume(this != NULL); \
183 return ir_type == ir_type_##TYPE ? (ir_##TYPE *) this : NULL; \
187 AS_CHILD(dereference_array
)
188 AS_CHILD(dereference_variable
)
189 AS_CHILD(dereference_record
)
204 * IR equality method: Return true if the referenced instruction would
205 * return the same value as this one.
207 * This intended to be used for CSE and algebraic optimizations, on rvalues
208 * in particular. No support for other instruction types (assignments,
209 * jumps, calls, etc.) is planned.
211 virtual bool equals(ir_instruction
*ir
, enum ir_node_type ignore
= ir_type_unset
);
214 ir_instruction(enum ir_node_type t
)
222 assert(!"Should not get here.");
228 * The base class for all "values"/expression trees.
230 class ir_rvalue
: public ir_instruction
{
232 const struct glsl_type
*type
;
234 virtual ir_rvalue
*clone(void *mem_ctx
, struct hash_table
*) const;
236 virtual void accept(ir_visitor
*v
)
241 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
243 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
245 ir_rvalue
*as_rvalue_to_saturate();
247 virtual bool is_lvalue() const
253 * Get the variable that is ultimately referenced by an r-value
255 virtual ir_variable
*variable_referenced() const
262 * If an r-value is a reference to a whole variable, get that variable
265 * Pointer to a variable that is completely dereferenced by the r-value. If
266 * the r-value is not a dereference or the dereference does not access the
267 * entire variable (i.e., it's just one array element, struct field), \c NULL
270 virtual ir_variable
*whole_variable_referenced()
276 * Determine if an r-value has the value zero
278 * The base implementation of this function always returns \c false. The
279 * \c ir_constant class over-rides this function to return \c true \b only
280 * for vector and scalar types that have all elements set to the value
281 * zero (or \c false for booleans).
283 * \sa ir_constant::has_value, ir_rvalue::is_one, ir_rvalue::is_negative_one
285 virtual bool is_zero() const;
288 * Determine if an r-value has the value one
290 * The base implementation of this function always returns \c false. The
291 * \c ir_constant class over-rides this function to return \c true \b only
292 * for vector and scalar types that have all elements set to the value
293 * one (or \c true for booleans).
295 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_negative_one
297 virtual bool is_one() const;
300 * Determine if an r-value has the value negative one
302 * The base implementation of this function always returns \c false. The
303 * \c ir_constant class over-rides this function to return \c true \b only
304 * for vector and scalar types that have all elements set to the value
305 * negative one. For boolean types, the result is always \c false.
307 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one
309 virtual bool is_negative_one() const;
312 * Determine if an r-value is an unsigned integer constant which can be
315 * \sa ir_constant::is_uint16_constant.
317 virtual bool is_uint16_constant() const { return false; }
320 * Return a generic value of error_type.
322 * Allocation will be performed with 'mem_ctx' as ralloc owner.
324 static ir_rvalue
*error_value(void *mem_ctx
);
327 ir_rvalue(enum ir_node_type t
);
332 * Variable storage classes
334 enum ir_variable_mode
{
335 ir_var_auto
= 0, /**< Function local variables and globals. */
336 ir_var_uniform
, /**< Variable declared as a uniform. */
341 ir_var_function_inout
,
342 ir_var_const_in
, /**< "in" param that must be a constant expression */
343 ir_var_system_value
, /**< Ex: front-face, instance-id, etc. */
344 ir_var_temporary
, /**< Temporary variable generated during compilation. */
345 ir_var_mode_count
/**< Number of variable modes */
349 * Enum keeping track of how a variable was declared. For error checking of
350 * the gl_PerVertex redeclaration rules.
352 enum ir_var_declaration_type
{
354 * Normal declaration (for most variables, this means an explicit
355 * declaration. Exception: temporaries are always implicitly declared, but
356 * they still use ir_var_declared_normally).
358 * Note: an ir_variable that represents a named interface block uses
359 * ir_var_declared_normally.
361 ir_var_declared_normally
= 0,
364 * Variable was explicitly declared (or re-declared) in an unnamed
367 ir_var_declared_in_block
,
370 * Variable is an implicitly declared built-in that has not been explicitly
371 * re-declared by the shader.
373 ir_var_declared_implicitly
,
376 * Variable is implicitly generated by the compiler and should not be
377 * visible via the API.
383 * \brief Layout qualifiers for gl_FragDepth.
385 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
386 * with a layout qualifier.
388 enum ir_depth_layout
{
389 ir_depth_layout_none
, /**< No depth layout is specified. */
391 ir_depth_layout_greater
,
392 ir_depth_layout_less
,
393 ir_depth_layout_unchanged
397 * \brief Convert depth layout qualifier to string.
400 depth_layout_string(ir_depth_layout layout
);
403 * Description of built-in state associated with a uniform
405 * \sa ir_variable::state_slots
407 struct ir_state_slot
{
414 * Get the string value for an interpolation qualifier
416 * \return The string that would be used in a shader to specify \c
417 * mode will be returned.
419 * This function is used to generate error messages of the form "shader
420 * uses %s interpolation qualifier", so in the case where there is no
421 * interpolation qualifier, it returns "no".
423 * This function should only be used on a shader input or output variable.
425 const char *interpolation_string(unsigned interpolation
);
428 class ir_variable
: public ir_instruction
{
430 ir_variable(const struct glsl_type
*, const char *, ir_variable_mode
);
432 virtual ir_variable
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
434 virtual void accept(ir_visitor
*v
)
439 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
443 * Determine how this variable should be interpolated based on its
444 * interpolation qualifier (if present), whether it is gl_Color or
445 * gl_SecondaryColor, and whether flatshading is enabled in the current GL
448 * The return value will always be either INTERP_QUALIFIER_SMOOTH,
449 * INTERP_QUALIFIER_NOPERSPECTIVE, or INTERP_QUALIFIER_FLAT.
451 glsl_interp_qualifier
determine_interpolation_mode(bool flat_shade
);
454 * Determine whether or not a variable is part of a uniform block.
456 inline bool is_in_uniform_block() const
458 return this->data
.mode
== ir_var_uniform
&& this->interface_type
!= NULL
;
462 * Determine whether or not a variable is the declaration of an interface
465 * For the first declaration below, there will be an \c ir_variable named
466 * "instance" whose type and whose instance_type will be the same
467 * \cglsl_type. For the second declaration, there will be an \c ir_variable
468 * named "f" whose type is float and whose instance_type is B2.
470 * "instance" is an interface instance variable, but "f" is not.
480 inline bool is_interface_instance() const
482 return this->type
->without_array() == this->interface_type
;
486 * Set this->interface_type on a newly created variable.
488 void init_interface_type(const struct glsl_type
*type
)
490 assert(this->interface_type
== NULL
);
491 this->interface_type
= type
;
492 if (this->is_interface_instance()) {
493 this->u
.max_ifc_array_access
=
494 rzalloc_array(this, unsigned, type
->length
);
499 * Change this->interface_type on a variable that previously had a
500 * different, but compatible, interface_type. This is used during linking
501 * to set the size of arrays in interface blocks.
503 void change_interface_type(const struct glsl_type
*type
)
505 if (this->u
.max_ifc_array_access
!= NULL
) {
506 /* max_ifc_array_access has already been allocated, so make sure the
507 * new interface has the same number of fields as the old one.
509 assert(this->interface_type
->length
== type
->length
);
511 this->interface_type
= type
;
515 * Change this->interface_type on a variable that previously had a
516 * different, and incompatible, interface_type. This is used during
517 * compilation to handle redeclaration of the built-in gl_PerVertex
520 void reinit_interface_type(const struct glsl_type
*type
)
522 if (this->u
.max_ifc_array_access
!= NULL
) {
524 /* Redeclaring gl_PerVertex is only allowed if none of the built-ins
525 * it defines have been accessed yet; so it's safe to throw away the
526 * old max_ifc_array_access pointer, since all of its values are
529 for (unsigned i
= 0; i
< this->interface_type
->length
; i
++)
530 assert(this->u
.max_ifc_array_access
[i
] == 0);
532 ralloc_free(this->u
.max_ifc_array_access
);
533 this->u
.max_ifc_array_access
= NULL
;
535 this->interface_type
= NULL
;
536 init_interface_type(type
);
539 const glsl_type
*get_interface_type() const
541 return this->interface_type
;
545 * Get the max_ifc_array_access pointer
547 * A "set" function is not needed because the array is dynmically allocated
550 inline unsigned *get_max_ifc_array_access()
552 assert(this->data
._num_state_slots
== 0);
553 return this->u
.max_ifc_array_access
;
556 inline unsigned get_num_state_slots() const
558 assert(!this->is_interface_instance()
559 || this->data
._num_state_slots
== 0);
560 return this->data
._num_state_slots
;
563 inline void set_num_state_slots(unsigned n
)
565 assert(!this->is_interface_instance()
567 this->data
._num_state_slots
= n
;
570 inline ir_state_slot
*get_state_slots()
572 return this->is_interface_instance() ? NULL
: this->u
.state_slots
;
575 inline const ir_state_slot
*get_state_slots() const
577 return this->is_interface_instance() ? NULL
: this->u
.state_slots
;
580 inline ir_state_slot
*allocate_state_slots(unsigned n
)
582 assert(!this->is_interface_instance());
584 this->u
.state_slots
= ralloc_array(this, ir_state_slot
, n
);
585 this->data
._num_state_slots
= 0;
587 if (this->u
.state_slots
!= NULL
)
588 this->data
._num_state_slots
= n
;
590 return this->u
.state_slots
;
593 inline bool is_name_ralloced() const
595 return this->name
!= ir_variable::tmp_name
;
599 * Enable emitting extension warnings for this variable
601 void enable_extension_warning(const char *extension
);
604 * Get the extension warning string for this variable
606 * If warnings are not enabled, \c NULL is returned.
608 const char *get_extension_warning() const;
611 * Declared type of the variable
613 const struct glsl_type
*type
;
616 * Declared name of the variable
620 struct ir_variable_data
{
623 * Is the variable read-only?
625 * This is set for variables declared as \c const, shader inputs,
628 unsigned read_only
:1;
631 unsigned invariant
:1;
635 * Has this variable been used for reading or writing?
637 * Several GLSL semantic checks require knowledge of whether or not a
638 * variable has been used. For example, it is an error to redeclare a
639 * variable as invariant after it has been used.
641 * This is only maintained in the ast_to_hir.cpp path, not in
642 * Mesa's fixed function or ARB program paths.
647 * Has this variable been statically assigned?
649 * This answers whether the variable was assigned in any path of
650 * the shader during ast_to_hir. This doesn't answer whether it is
651 * still written after dead code removal, nor is it maintained in
652 * non-ast_to_hir.cpp (GLSL parsing) paths.
657 * Enum indicating how the variable was declared. See
658 * ir_var_declaration_type.
660 * This is used to detect certain kinds of illegal variable redeclarations.
662 unsigned how_declared
:2;
665 * Storage class of the variable.
667 * \sa ir_variable_mode
672 * Interpolation mode for shader inputs / outputs
674 * \sa ir_variable_interpolation
676 unsigned interpolation
:2;
679 * \name ARB_fragment_coord_conventions
682 unsigned origin_upper_left
:1;
683 unsigned pixel_center_integer
:1;
687 * Was the location explicitly set in the shader?
689 * If the location is explicitly set in the shader, it \b cannot be changed
690 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
693 unsigned explicit_location
:1;
694 unsigned explicit_index
:1;
697 * Was an initial binding explicitly set in the shader?
699 * If so, constant_value contains an integer ir_constant representing the
700 * initial binding point.
702 unsigned explicit_binding
:1;
705 * Does this variable have an initializer?
707 * This is used by the linker to cross-validiate initializers of global
710 unsigned has_initializer
:1;
713 * Is this variable a generic output or input that has not yet been matched
714 * up to a variable in another stage of the pipeline?
716 * This is used by the linker as scratch storage while assigning locations
717 * to generic inputs and outputs.
719 unsigned is_unmatched_generic_inout
:1;
722 * If non-zero, then this variable may be packed along with other variables
723 * into a single varying slot, so this offset should be applied when
724 * accessing components. For example, an offset of 1 means that the x
725 * component of this variable is actually stored in component y of the
726 * location specified by \c location.
728 unsigned location_frac
:2;
731 * Layout of the matrix. Uses glsl_matrix_layout values.
733 unsigned matrix_layout
:2;
736 * Non-zero if this variable was created by lowering a named interface
737 * block which was not an array.
739 * Note that this variable and \c from_named_ifc_block_array will never
742 unsigned from_named_ifc_block_nonarray
:1;
745 * Non-zero if this variable was created by lowering a named interface
746 * block which was an array.
748 * Note that this variable and \c from_named_ifc_block_nonarray will never
751 unsigned from_named_ifc_block_array
:1;
754 * Non-zero if the variable must be a shader input. This is useful for
755 * constraints on function parameters.
757 unsigned must_be_shader_input
:1;
760 * Output index for dual source blending.
763 * The GLSL spec only allows the values 0 or 1 for the index in \b dual
769 * \brief Layout qualifier for gl_FragDepth.
771 * This is not equal to \c ir_depth_layout_none if and only if this
772 * variable is \c gl_FragDepth and a layout qualifier is specified.
774 ir_depth_layout depth_layout
:3;
777 * ARB_shader_image_load_store qualifiers.
779 unsigned image_read_only
:1; /**< "readonly" qualifier. */
780 unsigned image_write_only
:1; /**< "writeonly" qualifier. */
781 unsigned image_coherent
:1;
782 unsigned image_volatile
:1;
783 unsigned image_restrict
:1;
786 * Emit a warning if this variable is accessed.
789 uint8_t warn_extension_index
;
792 /** Image internal format if specified explicitly, otherwise GL_NONE. */
793 uint16_t image_format
;
797 * Number of state slots used
800 * This could be stored in as few as 7-bits, if necessary. If it is made
801 * smaller, add an assertion to \c ir_variable::allocate_state_slots to
804 uint16_t _num_state_slots
;
808 * Initial binding point for a sampler, atomic, or UBO.
810 * For array types, this represents the binding point for the first element.
815 * Storage location of the base of this variable
817 * The precise meaning of this field depends on the nature of the variable.
819 * - Vertex shader input: one of the values from \c gl_vert_attrib.
820 * - Vertex shader output: one of the values from \c gl_varying_slot.
821 * - Geometry shader input: one of the values from \c gl_varying_slot.
822 * - Geometry shader output: one of the values from \c gl_varying_slot.
823 * - Fragment shader input: one of the values from \c gl_varying_slot.
824 * - Fragment shader output: one of the values from \c gl_frag_result.
825 * - Uniforms: Per-stage uniform slot number for default uniform block.
826 * - Uniforms: Index within the uniform block definition for UBO members.
827 * - Other: This field is not currently used.
829 * If the variable is a uniform, shader input, or shader output, and the
830 * slot has not been assigned, the value will be -1.
835 * Vertex stream output identifier.
840 * Location an atomic counter is stored at.
847 * Highest element accessed with a constant expression array index
849 * Not used for non-array variables.
851 unsigned max_array_access
;
854 * Allow (only) ir_variable direct access private members.
856 friend class ir_variable
;
860 * Value assigned in the initializer of a variable declared "const"
862 ir_constant
*constant_value
;
865 * Constant expression assigned in the initializer of the variable
868 * This field and \c ::constant_value are distinct. Even if the two fields
869 * refer to constants with the same value, they must point to separate
872 ir_constant
*constant_initializer
;
875 static const char *const warn_extension_table
[];
879 * For variables which satisfy the is_interface_instance() predicate,
880 * this points to an array of integers such that if the ith member of
881 * the interface block is an array, max_ifc_array_access[i] is the
882 * maximum array element of that member that has been accessed. If the
883 * ith member of the interface block is not an array,
884 * max_ifc_array_access[i] is unused.
886 * For variables whose type is not an interface block, this pointer is
889 unsigned *max_ifc_array_access
;
892 * Built-in state that backs this uniform
894 * Once set at variable creation, \c state_slots must remain invariant.
896 * If the variable is not a uniform, \c _num_state_slots will be zero
897 * and \c state_slots will be \c NULL.
899 ir_state_slot
*state_slots
;
903 * For variables that are in an interface block or are an instance of an
904 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
906 * \sa ir_variable::location
908 const glsl_type
*interface_type
;
911 * Name used for anonymous compiler temporaries
913 static const char tmp_name
[];
917 * Should the construct keep names for ir_var_temporary variables?
919 * When this global is false, names passed to the constructor for
920 * \c ir_var_temporary variables will be dropped. Instead, the variable will
921 * be named "compiler_temp". This name will be in static storage.
924 * \b NEVER change the mode of an \c ir_var_temporary.
927 * This variable is \b not thread-safe. It is global, \b not
928 * per-context. It begins life false. A context can, at some point, make
929 * it true. From that point on, it will be true forever. This should be
930 * okay since it will only be set true while debugging.
932 static bool temporaries_allocate_names
;
936 * A function that returns whether a built-in function is available in the
937 * current shading language (based on version, ES or desktop, and extensions).
939 typedef bool (*builtin_available_predicate
)(const _mesa_glsl_parse_state
*);
943 * The representation of a function instance; may be the full definition or
944 * simply a prototype.
946 class ir_function_signature
: public ir_instruction
{
947 /* An ir_function_signature will be part of the list of signatures in
951 ir_function_signature(const glsl_type
*return_type
,
952 builtin_available_predicate builtin_avail
= NULL
);
954 virtual ir_function_signature
*clone(void *mem_ctx
,
955 struct hash_table
*ht
) const;
956 ir_function_signature
*clone_prototype(void *mem_ctx
,
957 struct hash_table
*ht
) const;
959 virtual void accept(ir_visitor
*v
)
964 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
967 * Attempt to evaluate this function as a constant expression,
968 * given a list of the actual parameters and the variable context.
969 * Returns NULL for non-built-ins.
971 ir_constant
*constant_expression_value(exec_list
*actual_parameters
, struct hash_table
*variable_context
);
974 * Get the name of the function for which this is a signature
976 const char *function_name() const;
979 * Get a handle to the function for which this is a signature
981 * There is no setter function, this function returns a \c const pointer,
982 * and \c ir_function_signature::_function is private for a reason. The
983 * only way to make a connection between a function and function signature
984 * is via \c ir_function::add_signature. This helps ensure that certain
985 * invariants (i.e., a function signature is in the list of signatures for
986 * its \c _function) are met.
988 * \sa ir_function::add_signature
990 inline const class ir_function
*function() const
992 return this->_function
;
996 * Check whether the qualifiers match between this signature's parameters
997 * and the supplied parameter list. If not, returns the name of the first
998 * parameter with mismatched qualifiers (for use in error messages).
1000 const char *qualifiers_match(exec_list
*params
);
1003 * Replace the current parameter list with the given one. This is useful
1004 * if the current information came from a prototype, and either has invalid
1005 * or missing parameter names.
1007 void replace_parameters(exec_list
*new_params
);
1010 * Function return type.
1012 * \note This discards the optional precision qualifier.
1014 const struct glsl_type
*return_type
;
1017 * List of ir_variable of function parameters.
1019 * This represents the storage. The paramaters passed in a particular
1020 * call will be in ir_call::actual_paramaters.
1022 struct exec_list parameters
;
1024 /** Whether or not this function has a body (which may be empty). */
1025 unsigned is_defined
:1;
1027 /** Whether or not this function signature is a built-in. */
1028 bool is_builtin() const;
1031 * Whether or not this function is an intrinsic to be implemented
1036 /** Whether or not a built-in is available for this shader. */
1037 bool is_builtin_available(const _mesa_glsl_parse_state
*state
) const;
1039 /** Body of instructions in the function. */
1040 struct exec_list body
;
1044 * A function pointer to a predicate that answers whether a built-in
1045 * function is available in the current shader. NULL if not a built-in.
1047 builtin_available_predicate builtin_avail
;
1049 /** Function of which this signature is one overload. */
1050 class ir_function
*_function
;
1052 /** Function signature of which this one is a prototype clone */
1053 const ir_function_signature
*origin
;
1055 friend class ir_function
;
1058 * Helper function to run a list of instructions for constant
1059 * expression evaluation.
1061 * The hash table represents the values of the visible variables.
1062 * There are no scoping issues because the table is indexed on
1063 * ir_variable pointers, not variable names.
1065 * Returns false if the expression is not constant, true otherwise,
1066 * and the value in *result if result is non-NULL.
1068 bool constant_expression_evaluate_expression_list(const struct exec_list
&body
,
1069 struct hash_table
*variable_context
,
1070 ir_constant
**result
);
1075 * Header for tracking multiple overloaded functions with the same name.
1076 * Contains a list of ir_function_signatures representing each of the
1079 class ir_function
: public ir_instruction
{
1081 ir_function(const char *name
);
1083 virtual ir_function
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1085 virtual void accept(ir_visitor
*v
)
1090 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1092 void add_signature(ir_function_signature
*sig
)
1094 sig
->_function
= this;
1095 this->signatures
.push_tail(sig
);
1099 * Find a signature that matches a set of actual parameters, taking implicit
1100 * conversions into account. Also flags whether the match was exact.
1102 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1103 const exec_list
*actual_param
,
1104 bool allow_builtins
,
1105 bool *match_is_exact
);
1108 * Find a signature that matches a set of actual parameters, taking implicit
1109 * conversions into account.
1111 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1112 const exec_list
*actual_param
,
1113 bool allow_builtins
);
1116 * Find a signature that exactly matches a set of actual parameters without
1117 * any implicit type conversions.
1119 ir_function_signature
*exact_matching_signature(_mesa_glsl_parse_state
*state
,
1120 const exec_list
*actual_ps
);
1123 * Name of the function.
1127 /** Whether or not this function has a signature that isn't a built-in. */
1128 bool has_user_signature();
1131 * List of ir_function_signature for each overloaded function with this name.
1133 struct exec_list signatures
;
1136 inline const char *ir_function_signature::function_name() const
1138 return this->_function
->name
;
1144 * IR instruction representing high-level if-statements
1146 class ir_if
: public ir_instruction
{
1148 ir_if(ir_rvalue
*condition
)
1149 : ir_instruction(ir_type_if
), condition(condition
)
1153 virtual ir_if
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1155 virtual void accept(ir_visitor
*v
)
1160 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1162 ir_rvalue
*condition
;
1163 /** List of ir_instruction for the body of the then branch */
1164 exec_list then_instructions
;
1165 /** List of ir_instruction for the body of the else branch */
1166 exec_list else_instructions
;
1171 * IR instruction representing a high-level loop structure.
1173 class ir_loop
: public ir_instruction
{
1177 virtual ir_loop
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1179 virtual void accept(ir_visitor
*v
)
1184 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1186 /** List of ir_instruction that make up the body of the loop. */
1187 exec_list body_instructions
;
1191 class ir_assignment
: public ir_instruction
{
1193 ir_assignment(ir_rvalue
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
= NULL
);
1196 * Construct an assignment with an explicit write mask
1199 * Since a write mask is supplied, the LHS must already be a bare
1200 * \c ir_dereference. The cannot be any swizzles in the LHS.
1202 ir_assignment(ir_dereference
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
,
1203 unsigned write_mask
);
1205 virtual ir_assignment
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1207 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1209 virtual void accept(ir_visitor
*v
)
1214 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1217 * Get a whole variable written by an assignment
1219 * If the LHS of the assignment writes a whole variable, the variable is
1220 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1223 * - Assigning to a scalar
1224 * - Assigning to all components of a vector
1225 * - Whole array (or matrix) assignment
1226 * - Whole structure assignment
1228 ir_variable
*whole_variable_written();
1231 * Set the LHS of an assignment
1233 void set_lhs(ir_rvalue
*lhs
);
1236 * Left-hand side of the assignment.
1238 * This should be treated as read only. If you need to set the LHS of an
1239 * assignment, use \c ir_assignment::set_lhs.
1241 ir_dereference
*lhs
;
1244 * Value being assigned
1249 * Optional condition for the assignment.
1251 ir_rvalue
*condition
;
1255 * Component mask written
1257 * For non-vector types in the LHS, this field will be zero. For vector
1258 * types, a bit will be set for each component that is written. Note that
1259 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1261 * A partially-set write mask means that each enabled channel gets
1262 * the value from a consecutive channel of the rhs. For example,
1263 * to write just .xyw of gl_FrontColor with color:
1265 * (assign (constant bool (1)) (xyw)
1266 * (var_ref gl_FragColor)
1267 * (swiz xyw (var_ref color)))
1269 unsigned write_mask
:4;
1272 /* Update ir_expression::get_num_operands() and operator_strs when
1273 * updating this list.
1275 enum ir_expression_operation
{
1284 ir_unop_exp
, /**< Log base e on gentype */
1285 ir_unop_log
, /**< Natural log on gentype */
1288 ir_unop_f2i
, /**< Float-to-integer conversion. */
1289 ir_unop_f2u
, /**< Float-to-unsigned conversion. */
1290 ir_unop_i2f
, /**< Integer-to-float conversion. */
1291 ir_unop_f2b
, /**< Float-to-boolean conversion */
1292 ir_unop_b2f
, /**< Boolean-to-float conversion */
1293 ir_unop_i2b
, /**< int-to-boolean conversion */
1294 ir_unop_b2i
, /**< Boolean-to-int conversion */
1295 ir_unop_u2f
, /**< Unsigned-to-float conversion. */
1296 ir_unop_i2u
, /**< Integer-to-unsigned conversion. */
1297 ir_unop_u2i
, /**< Unsigned-to-integer conversion. */
1298 ir_unop_d2f
, /**< Double-to-float conversion. */
1299 ir_unop_f2d
, /**< Float-to-double conversion. */
1300 ir_unop_d2i
, /**< Double-to-integer conversion. */
1301 ir_unop_i2d
, /**< Integer-to-double conversion. */
1302 ir_unop_d2u
, /**< Double-to-unsigned conversion. */
1303 ir_unop_u2d
, /**< Unsigned-to-double conversion. */
1304 ir_unop_d2b
, /**< Double-to-boolean conversion. */
1305 ir_unop_bitcast_i2f
, /**< Bit-identical int-to-float "conversion" */
1306 ir_unop_bitcast_f2i
, /**< Bit-identical float-to-int "conversion" */
1307 ir_unop_bitcast_u2f
, /**< Bit-identical uint-to-float "conversion" */
1308 ir_unop_bitcast_f2u
, /**< Bit-identical float-to-uint "conversion" */
1312 * \name Unary floating-point rounding operations.
1323 * \name Trigonometric operations.
1328 ir_unop_sin_reduced
, /**< Reduced range sin. [-pi, pi] */
1329 ir_unop_cos_reduced
, /**< Reduced range cos. [-pi, pi] */
1333 * \name Partial derivatives.
1337 ir_unop_dFdx_coarse
,
1340 ir_unop_dFdy_coarse
,
1345 * \name Floating point pack and unpack operations.
1348 ir_unop_pack_snorm_2x16
,
1349 ir_unop_pack_snorm_4x8
,
1350 ir_unop_pack_unorm_2x16
,
1351 ir_unop_pack_unorm_4x8
,
1352 ir_unop_pack_half_2x16
,
1353 ir_unop_unpack_snorm_2x16
,
1354 ir_unop_unpack_snorm_4x8
,
1355 ir_unop_unpack_unorm_2x16
,
1356 ir_unop_unpack_unorm_4x8
,
1357 ir_unop_unpack_half_2x16
,
1361 * \name Lowered floating point unpacking operations.
1363 * \see lower_packing_builtins_visitor::split_unpack_half_2x16
1366 ir_unop_unpack_half_2x16_split_x
,
1367 ir_unop_unpack_half_2x16_split_y
,
1371 * \name Bit operations, part of ARB_gpu_shader5.
1374 ir_unop_bitfield_reverse
,
1383 * \name Double packing, part of ARB_gpu_shader_fp64.
1386 ir_unop_pack_double_2x32
,
1387 ir_unop_unpack_double_2x32
,
1396 * Interpolate fs input at centroid
1398 * operand0 is the fs input.
1400 ir_unop_interpolate_at_centroid
,
1403 * A sentinel marking the last of the unary operations.
1405 ir_last_unop
= ir_unop_interpolate_at_centroid
,
1409 ir_binop_mul
, /**< Floating-point or low 32-bit integer multiply. */
1410 ir_binop_imul_high
, /**< Calculates the high 32-bits of a 64-bit multiply. */
1414 * Returns the carry resulting from the addition of the two arguments.
1421 * Returns the borrow resulting from the subtraction of the second argument
1422 * from the first argument.
1429 * Takes one of two combinations of arguments:
1432 * - mod(vecN, float)
1434 * Does not take integer types.
1439 * \name Binary comparison operators which return a boolean vector.
1440 * The type of both operands must be equal.
1450 * Returns single boolean for whether all components of operands[0]
1451 * equal the components of operands[1].
1455 * Returns single boolean for whether any component of operands[0]
1456 * is not equal to the corresponding component of operands[1].
1458 ir_binop_any_nequal
,
1462 * \name Bit-wise binary operations.
1483 * \name Lowered floating point packing operations.
1485 * \see lower_packing_builtins_visitor::split_pack_half_2x16
1488 ir_binop_pack_half_2x16_split
,
1492 * \name First half of a lowered bitfieldInsert() operation.
1494 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1501 * Load a value the size of a given GLSL type from a uniform block.
1503 * operand0 is the ir_constant uniform block index in the linked shader.
1504 * operand1 is a byte offset within the uniform block.
1509 * \name Multiplies a number by two to a power, part of ARB_gpu_shader5.
1516 * Extract a scalar from a vector
1518 * operand0 is the vector
1519 * operand1 is the index of the field to read from operand0
1521 ir_binop_vector_extract
,
1524 * Interpolate fs input at offset
1526 * operand0 is the fs input
1527 * operand1 is the offset from the pixel center
1529 ir_binop_interpolate_at_offset
,
1532 * Interpolate fs input at sample position
1534 * operand0 is the fs input
1535 * operand1 is the sample ID
1537 ir_binop_interpolate_at_sample
,
1540 * A sentinel marking the last of the binary operations.
1542 ir_last_binop
= ir_binop_interpolate_at_sample
,
1545 * \name Fused floating-point multiply-add, part of ARB_gpu_shader5.
1554 * \name Conditional Select
1556 * A vector conditional select instruction (like ?:, but operating per-
1557 * component on vectors).
1559 * \see lower_instructions_visitor::ldexp_to_arith
1566 * \name Second half of a lowered bitfieldInsert() operation.
1568 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1574 ir_triop_bitfield_extract
,
1577 * Generate a value with one field of a vector changed
1579 * operand0 is the vector
1580 * operand1 is the value to write into the vector result
1581 * operand2 is the index in operand0 to be modified
1583 ir_triop_vector_insert
,
1586 * A sentinel marking the last of the ternary operations.
1588 ir_last_triop
= ir_triop_vector_insert
,
1590 ir_quadop_bitfield_insert
,
1595 * A sentinel marking the last of the ternary operations.
1597 ir_last_quadop
= ir_quadop_vector
,
1600 * A sentinel marking the last of all operations.
1602 ir_last_opcode
= ir_quadop_vector
1605 class ir_expression
: public ir_rvalue
{
1607 ir_expression(int op
, const struct glsl_type
*type
,
1608 ir_rvalue
*op0
, ir_rvalue
*op1
= NULL
,
1609 ir_rvalue
*op2
= NULL
, ir_rvalue
*op3
= NULL
);
1612 * Constructor for unary operation expressions
1614 ir_expression(int op
, ir_rvalue
*);
1617 * Constructor for binary operation expressions
1619 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
);
1622 * Constructor for ternary operation expressions
1624 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
, ir_rvalue
*op2
);
1626 virtual bool equals(ir_instruction
*ir
, enum ir_node_type ignore
= ir_type_unset
);
1628 virtual ir_expression
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1631 * Attempt to constant-fold the expression
1633 * The "variable_context" hash table links ir_variable * to ir_constant *
1634 * that represent the variables' values. \c NULL represents an empty
1637 * If the expression cannot be constant folded, this method will return
1640 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1643 * Determine the number of operands used by an expression
1645 static unsigned int get_num_operands(ir_expression_operation
);
1648 * Determine the number of operands used by an expression
1650 unsigned int get_num_operands() const
1652 return (this->operation
== ir_quadop_vector
)
1653 ? this->type
->vector_elements
: get_num_operands(operation
);
1657 * Return whether the expression operates on vectors horizontally.
1659 bool is_horizontal() const
1661 return operation
== ir_binop_all_equal
||
1662 operation
== ir_binop_any_nequal
||
1663 operation
== ir_unop_any
||
1664 operation
== ir_binop_dot
||
1665 operation
== ir_quadop_vector
;
1669 * Return a string representing this expression's operator.
1671 const char *operator_string();
1674 * Return a string representing this expression's operator.
1676 static const char *operator_string(ir_expression_operation
);
1680 * Do a reverse-lookup to translate the given string into an operator.
1682 static ir_expression_operation
get_operator(const char *);
1684 virtual void accept(ir_visitor
*v
)
1689 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1691 ir_expression_operation operation
;
1692 ir_rvalue
*operands
[4];
1697 * HIR instruction representing a high-level function call, containing a list
1698 * of parameters and returning a value in the supplied temporary.
1700 class ir_call
: public ir_instruction
{
1702 ir_call(ir_function_signature
*callee
,
1703 ir_dereference_variable
*return_deref
,
1704 exec_list
*actual_parameters
)
1705 : ir_instruction(ir_type_call
), return_deref(return_deref
), callee(callee
)
1707 assert(callee
->return_type
!= NULL
);
1708 actual_parameters
->move_nodes_to(& this->actual_parameters
);
1709 this->use_builtin
= callee
->is_builtin();
1712 virtual ir_call
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1714 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1716 virtual void accept(ir_visitor
*v
)
1721 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1724 * Get the name of the function being called.
1726 const char *callee_name() const
1728 return callee
->function_name();
1732 * Generates an inline version of the function before @ir,
1733 * storing the return value in return_deref.
1735 void generate_inline(ir_instruction
*ir
);
1738 * Storage for the function's return value.
1739 * This must be NULL if the return type is void.
1741 ir_dereference_variable
*return_deref
;
1744 * The specific function signature being called.
1746 ir_function_signature
*callee
;
1748 /* List of ir_rvalue of paramaters passed in this call. */
1749 exec_list actual_parameters
;
1751 /** Should this call only bind to a built-in function? */
1757 * \name Jump-like IR instructions.
1759 * These include \c break, \c continue, \c return, and \c discard.
1762 class ir_jump
: public ir_instruction
{
1764 ir_jump(enum ir_node_type t
)
1770 class ir_return
: public ir_jump
{
1773 : ir_jump(ir_type_return
), value(NULL
)
1777 ir_return(ir_rvalue
*value
)
1778 : ir_jump(ir_type_return
), value(value
)
1782 virtual ir_return
*clone(void *mem_ctx
, struct hash_table
*) const;
1784 ir_rvalue
*get_value() const
1789 virtual void accept(ir_visitor
*v
)
1794 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1801 * Jump instructions used inside loops
1803 * These include \c break and \c continue. The \c break within a loop is
1804 * different from the \c break within a switch-statement.
1806 * \sa ir_switch_jump
1808 class ir_loop_jump
: public ir_jump
{
1815 ir_loop_jump(jump_mode mode
)
1816 : ir_jump(ir_type_loop_jump
)
1821 virtual ir_loop_jump
*clone(void *mem_ctx
, struct hash_table
*) const;
1823 virtual void accept(ir_visitor
*v
)
1828 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1830 bool is_break() const
1832 return mode
== jump_break
;
1835 bool is_continue() const
1837 return mode
== jump_continue
;
1840 /** Mode selector for the jump instruction. */
1841 enum jump_mode mode
;
1845 * IR instruction representing discard statements.
1847 class ir_discard
: public ir_jump
{
1850 : ir_jump(ir_type_discard
)
1852 this->condition
= NULL
;
1855 ir_discard(ir_rvalue
*cond
)
1856 : ir_jump(ir_type_discard
)
1858 this->condition
= cond
;
1861 virtual ir_discard
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1863 virtual void accept(ir_visitor
*v
)
1868 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1870 ir_rvalue
*condition
;
1876 * Texture sampling opcodes used in ir_texture
1878 enum ir_texture_opcode
{
1879 ir_tex
, /**< Regular texture look-up */
1880 ir_txb
, /**< Texture look-up with LOD bias */
1881 ir_txl
, /**< Texture look-up with explicit LOD */
1882 ir_txd
, /**< Texture look-up with partial derivatvies */
1883 ir_txf
, /**< Texel fetch with explicit LOD */
1884 ir_txf_ms
, /**< Multisample texture fetch */
1885 ir_txs
, /**< Texture size */
1886 ir_lod
, /**< Texture lod query */
1887 ir_tg4
, /**< Texture gather */
1888 ir_query_levels
/**< Texture levels query */
1893 * IR instruction to sample a texture
1895 * The specific form of the IR instruction depends on the \c mode value
1896 * selected from \c ir_texture_opcodes. In the printed IR, these will
1899 * Texel offset (0 or an expression)
1900 * | Projection divisor
1901 * | | Shadow comparitor
1904 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1905 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1906 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1907 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1908 * (txf <type> <sampler> <coordinate> 0 <lod>)
1910 * <type> <sampler> <coordinate> <sample_index>)
1911 * (txs <type> <sampler> <lod>)
1912 * (lod <type> <sampler> <coordinate>)
1913 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
1914 * (query_levels <type> <sampler>)
1916 class ir_texture
: public ir_rvalue
{
1918 ir_texture(enum ir_texture_opcode op
)
1919 : ir_rvalue(ir_type_texture
),
1920 op(op
), sampler(NULL
), coordinate(NULL
), projector(NULL
),
1921 shadow_comparitor(NULL
), offset(NULL
)
1923 memset(&lod_info
, 0, sizeof(lod_info
));
1926 virtual ir_texture
*clone(void *mem_ctx
, struct hash_table
*) const;
1928 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1930 virtual void accept(ir_visitor
*v
)
1935 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1937 virtual bool equals(ir_instruction
*ir
, enum ir_node_type ignore
= ir_type_unset
);
1940 * Return a string representing the ir_texture_opcode.
1942 const char *opcode_string();
1944 /** Set the sampler and type. */
1945 void set_sampler(ir_dereference
*sampler
, const glsl_type
*type
);
1948 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1950 static ir_texture_opcode
get_opcode(const char *);
1952 enum ir_texture_opcode op
;
1954 /** Sampler to use for the texture access. */
1955 ir_dereference
*sampler
;
1957 /** Texture coordinate to sample */
1958 ir_rvalue
*coordinate
;
1961 * Value used for projective divide.
1963 * If there is no projective divide (the common case), this will be
1964 * \c NULL. Optimization passes should check for this to point to a constant
1965 * of 1.0 and replace that with \c NULL.
1967 ir_rvalue
*projector
;
1970 * Coordinate used for comparison on shadow look-ups.
1972 * If there is no shadow comparison, this will be \c NULL. For the
1973 * \c ir_txf opcode, this *must* be \c NULL.
1975 ir_rvalue
*shadow_comparitor
;
1977 /** Texel offset. */
1981 ir_rvalue
*lod
; /**< Floating point LOD */
1982 ir_rvalue
*bias
; /**< Floating point LOD bias */
1983 ir_rvalue
*sample_index
; /**< MSAA sample index */
1984 ir_rvalue
*component
; /**< Gather component selector */
1986 ir_rvalue
*dPdx
; /**< Partial derivative of coordinate wrt X */
1987 ir_rvalue
*dPdy
; /**< Partial derivative of coordinate wrt Y */
1993 struct ir_swizzle_mask
{
2000 * Number of components in the swizzle.
2002 unsigned num_components
:3;
2005 * Does the swizzle contain duplicate components?
2007 * L-value swizzles cannot contain duplicate components.
2009 unsigned has_duplicates
:1;
2013 class ir_swizzle
: public ir_rvalue
{
2015 ir_swizzle(ir_rvalue
*, unsigned x
, unsigned y
, unsigned z
, unsigned w
,
2018 ir_swizzle(ir_rvalue
*val
, const unsigned *components
, unsigned count
);
2020 ir_swizzle(ir_rvalue
*val
, ir_swizzle_mask mask
);
2022 virtual ir_swizzle
*clone(void *mem_ctx
, struct hash_table
*) const;
2024 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2027 * Construct an ir_swizzle from the textual representation. Can fail.
2029 static ir_swizzle
*create(ir_rvalue
*, const char *, unsigned vector_length
);
2031 virtual void accept(ir_visitor
*v
)
2036 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2038 virtual bool equals(ir_instruction
*ir
, enum ir_node_type ignore
= ir_type_unset
);
2040 bool is_lvalue() const
2042 return val
->is_lvalue() && !mask
.has_duplicates
;
2046 * Get the variable that is ultimately referenced by an r-value
2048 virtual ir_variable
*variable_referenced() const;
2051 ir_swizzle_mask mask
;
2055 * Initialize the mask component of a swizzle
2057 * This is used by the \c ir_swizzle constructors.
2059 void init_mask(const unsigned *components
, unsigned count
);
2063 class ir_dereference
: public ir_rvalue
{
2065 virtual ir_dereference
*clone(void *mem_ctx
, struct hash_table
*) const = 0;
2067 bool is_lvalue() const;
2070 * Get the variable that is ultimately referenced by an r-value
2072 virtual ir_variable
*variable_referenced() const = 0;
2075 ir_dereference(enum ir_node_type t
)
2082 class ir_dereference_variable
: public ir_dereference
{
2084 ir_dereference_variable(ir_variable
*var
);
2086 virtual ir_dereference_variable
*clone(void *mem_ctx
,
2087 struct hash_table
*) const;
2089 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2091 virtual bool equals(ir_instruction
*ir
, enum ir_node_type ignore
= ir_type_unset
);
2094 * Get the variable that is ultimately referenced by an r-value
2096 virtual ir_variable
*variable_referenced() const
2101 virtual ir_variable
*whole_variable_referenced()
2103 /* ir_dereference_variable objects always dereference the entire
2104 * variable. However, if this dereference is dereferenced by anything
2105 * else, the complete deferefernce chain is not a whole-variable
2106 * dereference. This method should only be called on the top most
2107 * ir_rvalue in a dereference chain.
2112 virtual void accept(ir_visitor
*v
)
2117 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2120 * Object being dereferenced.
2126 class ir_dereference_array
: public ir_dereference
{
2128 ir_dereference_array(ir_rvalue
*value
, ir_rvalue
*array_index
);
2130 ir_dereference_array(ir_variable
*var
, ir_rvalue
*array_index
);
2132 virtual ir_dereference_array
*clone(void *mem_ctx
,
2133 struct hash_table
*) const;
2135 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2137 virtual bool equals(ir_instruction
*ir
, enum ir_node_type ignore
= ir_type_unset
);
2140 * Get the variable that is ultimately referenced by an r-value
2142 virtual ir_variable
*variable_referenced() const
2144 return this->array
->variable_referenced();
2147 virtual void accept(ir_visitor
*v
)
2152 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2155 ir_rvalue
*array_index
;
2158 void set_array(ir_rvalue
*value
);
2162 class ir_dereference_record
: public ir_dereference
{
2164 ir_dereference_record(ir_rvalue
*value
, const char *field
);
2166 ir_dereference_record(ir_variable
*var
, const char *field
);
2168 virtual ir_dereference_record
*clone(void *mem_ctx
,
2169 struct hash_table
*) const;
2171 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2174 * Get the variable that is ultimately referenced by an r-value
2176 virtual ir_variable
*variable_referenced() const
2178 return this->record
->variable_referenced();
2181 virtual void accept(ir_visitor
*v
)
2186 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2194 * Data stored in an ir_constant
2196 union ir_constant_data
{
2205 class ir_constant
: public ir_rvalue
{
2207 ir_constant(const struct glsl_type
*type
, const ir_constant_data
*data
);
2208 ir_constant(bool b
, unsigned vector_elements
=1);
2209 ir_constant(unsigned int u
, unsigned vector_elements
=1);
2210 ir_constant(int i
, unsigned vector_elements
=1);
2211 ir_constant(float f
, unsigned vector_elements
=1);
2212 ir_constant(double d
, unsigned vector_elements
=1);
2215 * Construct an ir_constant from a list of ir_constant values
2217 ir_constant(const struct glsl_type
*type
, exec_list
*values
);
2220 * Construct an ir_constant from a scalar component of another ir_constant
2222 * The new \c ir_constant inherits the type of the component from the
2226 * In the case of a matrix constant, the new constant is a scalar, \b not
2229 ir_constant(const ir_constant
*c
, unsigned i
);
2232 * Return a new ir_constant of the specified type containing all zeros.
2234 static ir_constant
*zero(void *mem_ctx
, const glsl_type
*type
);
2236 virtual ir_constant
*clone(void *mem_ctx
, struct hash_table
*) const;
2238 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2240 virtual void accept(ir_visitor
*v
)
2245 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2247 virtual bool equals(ir_instruction
*ir
, enum ir_node_type ignore
= ir_type_unset
);
2250 * Get a particular component of a constant as a specific type
2252 * This is useful, for example, to get a value from an integer constant
2253 * as a float or bool. This appears frequently when constructors are
2254 * called with all constant parameters.
2257 bool get_bool_component(unsigned i
) const;
2258 float get_float_component(unsigned i
) const;
2259 double get_double_component(unsigned i
) const;
2260 int get_int_component(unsigned i
) const;
2261 unsigned get_uint_component(unsigned i
) const;
2264 ir_constant
*get_array_element(unsigned i
) const;
2266 ir_constant
*get_record_field(const char *name
);
2269 * Copy the values on another constant at a given offset.
2271 * The offset is ignored for array or struct copies, it's only for
2272 * scalars or vectors into vectors or matrices.
2274 * With identical types on both sides and zero offset it's clone()
2275 * without creating a new object.
2278 void copy_offset(ir_constant
*src
, int offset
);
2281 * Copy the values on another constant at a given offset and
2282 * following an assign-like mask.
2284 * The mask is ignored for scalars.
2286 * Note that this function only handles what assign can handle,
2287 * i.e. at most a vector as source and a column of a matrix as
2291 void copy_masked_offset(ir_constant
*src
, int offset
, unsigned int mask
);
2294 * Determine whether a constant has the same value as another constant
2296 * \sa ir_constant::is_zero, ir_constant::is_one,
2297 * ir_constant::is_negative_one
2299 bool has_value(const ir_constant
*) const;
2302 * Return true if this ir_constant represents the given value.
2304 * For vectors, this checks that each component is the given value.
2306 virtual bool is_value(float f
, int i
) const;
2307 virtual bool is_zero() const;
2308 virtual bool is_one() const;
2309 virtual bool is_negative_one() const;
2312 * Return true for constants that could be stored as 16-bit unsigned values.
2314 * Note that this will return true even for signed integer ir_constants, as
2315 * long as the value is non-negative and fits in 16-bits.
2317 virtual bool is_uint16_constant() const;
2320 * Value of the constant.
2322 * The field used to back the values supplied by the constant is determined
2323 * by the type associated with the \c ir_instruction. Constants may be
2324 * scalars, vectors, or matrices.
2326 union ir_constant_data value
;
2328 /* Array elements */
2329 ir_constant
**array_elements
;
2331 /* Structure fields */
2332 exec_list components
;
2336 * Parameterless constructor only used by the clone method
2342 * IR instruction to emit a vertex in a geometry shader.
2344 class ir_emit_vertex
: public ir_instruction
{
2346 ir_emit_vertex(ir_rvalue
*stream
)
2347 : ir_instruction(ir_type_emit_vertex
),
2353 virtual void accept(ir_visitor
*v
)
2358 virtual ir_emit_vertex
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2360 return new(mem_ctx
) ir_emit_vertex(this->stream
->clone(mem_ctx
, ht
));
2363 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2365 int stream_id() const
2367 return stream
->as_constant()->value
.i
[0];
2374 * IR instruction to complete the current primitive and start a new one in a
2377 class ir_end_primitive
: public ir_instruction
{
2379 ir_end_primitive(ir_rvalue
*stream
)
2380 : ir_instruction(ir_type_end_primitive
),
2386 virtual void accept(ir_visitor
*v
)
2391 virtual ir_end_primitive
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2393 return new(mem_ctx
) ir_end_primitive(this->stream
->clone(mem_ctx
, ht
));
2396 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2398 int stream_id() const
2400 return stream
->as_constant()->value
.i
[0];
2409 * Apply a visitor to each IR node in a list
2412 visit_exec_list(exec_list
*list
, ir_visitor
*visitor
);
2415 * Validate invariants on each IR node in a list
2417 void validate_ir_tree(exec_list
*instructions
);
2419 struct _mesa_glsl_parse_state
;
2420 struct gl_shader_program
;
2423 * Detect whether an unlinked shader contains static recursion
2425 * If the list of instructions is determined to contain static recursion,
2426 * \c _mesa_glsl_error will be called to emit error messages for each function
2427 * that is in the recursion cycle.
2430 detect_recursion_unlinked(struct _mesa_glsl_parse_state
*state
,
2431 exec_list
*instructions
);
2434 * Detect whether a linked shader contains static recursion
2436 * If the list of instructions is determined to contain static recursion,
2437 * \c link_error_printf will be called to emit error messages for each function
2438 * that is in the recursion cycle. In addition,
2439 * \c gl_shader_program::LinkStatus will be set to false.
2442 detect_recursion_linked(struct gl_shader_program
*prog
,
2443 exec_list
*instructions
);
2446 * Make a clone of each IR instruction in a list
2448 * \param in List of IR instructions that are to be cloned
2449 * \param out List to hold the cloned instructions
2452 clone_ir_list(void *mem_ctx
, exec_list
*out
, const exec_list
*in
);
2455 _mesa_glsl_initialize_variables(exec_list
*instructions
,
2456 struct _mesa_glsl_parse_state
*state
);
2459 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state
*state
);
2462 _mesa_glsl_initialize_builtin_functions();
2464 extern ir_function_signature
*
2465 _mesa_glsl_find_builtin_function(_mesa_glsl_parse_state
*state
,
2466 const char *name
, exec_list
*actual_parameters
);
2468 extern ir_function
*
2469 _mesa_glsl_find_builtin_function_by_name(_mesa_glsl_parse_state
*state
,
2473 _mesa_glsl_get_builtin_function_shader(void);
2476 _mesa_glsl_release_functions(void);
2479 _mesa_glsl_release_builtin_functions(void);
2482 reparent_ir(exec_list
*list
, void *mem_ctx
);
2484 struct glsl_symbol_table
;
2487 import_prototypes(const exec_list
*source
, exec_list
*dest
,
2488 struct glsl_symbol_table
*symbols
, void *mem_ctx
);
2491 ir_has_call(ir_instruction
*ir
);
2494 do_set_program_inouts(exec_list
*instructions
, struct gl_program
*prog
,
2495 gl_shader_stage shader_stage
);
2498 prototype_string(const glsl_type
*return_type
, const char *name
,
2499 exec_list
*parameters
);
2502 mode_string(const ir_variable
*var
);
2505 * Built-in / reserved GL variables names start with "gl_"
2508 is_gl_identifier(const char *s
)
2510 return s
&& s
[0] == 'g' && s
[1] == 'l' && s
[2] == '_';
2514 #endif /* __cplusplus */
2516 extern void _mesa_print_ir(FILE *f
, struct exec_list
*instructions
,
2517 struct _mesa_glsl_parse_state
*state
);
2520 fprint_ir(FILE *f
, const void *instruction
);
2527 vertices_per_prim(GLenum prim
);