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 #define AS_BASE(TYPE) \
146 class ir_##TYPE *as_##TYPE() \
148 assume(this != NULL); \
149 return is_##TYPE() ? (ir_##TYPE *) this : NULL; \
151 const class ir_##TYPE *as_##TYPE() const \
153 assume(this != NULL); \
154 return is_##TYPE() ? (ir_##TYPE *) this : NULL; \
162 #define AS_CHILD(TYPE) \
163 class ir_##TYPE * as_##TYPE() \
165 assume(this != NULL); \
166 return ir_type == ir_type_##TYPE ? (ir_##TYPE *) this : NULL; \
168 const class ir_##TYPE * as_##TYPE() const \
170 assume(this != NULL); \
171 return ir_type == ir_type_##TYPE ? (const ir_##TYPE *) this : NULL; \
175 AS_CHILD(dereference_array
)
176 AS_CHILD(dereference_variable
)
177 AS_CHILD(dereference_record
)
192 * IR equality method: Return true if the referenced instruction would
193 * return the same value as this one.
195 * This intended to be used for CSE and algebraic optimizations, on rvalues
196 * in particular. No support for other instruction types (assignments,
197 * jumps, calls, etc.) is planned.
199 virtual bool equals(const ir_instruction
*ir
,
200 enum ir_node_type ignore
= ir_type_unset
) const;
203 ir_instruction(enum ir_node_type t
)
211 assert(!"Should not get here.");
217 * The base class for all "values"/expression trees.
219 class ir_rvalue
: public ir_instruction
{
221 const struct glsl_type
*type
;
223 virtual ir_rvalue
*clone(void *mem_ctx
, struct hash_table
*) const;
225 virtual void accept(ir_visitor
*v
)
230 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
232 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
234 ir_rvalue
*as_rvalue_to_saturate();
236 virtual bool is_lvalue() const
242 * Get the variable that is ultimately referenced by an r-value
244 virtual ir_variable
*variable_referenced() const
251 * If an r-value is a reference to a whole variable, get that variable
254 * Pointer to a variable that is completely dereferenced by the r-value. If
255 * the r-value is not a dereference or the dereference does not access the
256 * entire variable (i.e., it's just one array element, struct field), \c NULL
259 virtual ir_variable
*whole_variable_referenced()
265 * Determine if an r-value has the value zero
267 * The base implementation of this function always returns \c false. The
268 * \c ir_constant class over-rides this function to return \c true \b only
269 * for vector and scalar types that have all elements set to the value
270 * zero (or \c false for booleans).
272 * \sa ir_constant::has_value, ir_rvalue::is_one, ir_rvalue::is_negative_one
274 virtual bool is_zero() const;
277 * Determine if an r-value has the value one
279 * The base implementation of this function always returns \c false. The
280 * \c ir_constant class over-rides this function to return \c true \b only
281 * for vector and scalar types that have all elements set to the value
282 * one (or \c true for booleans).
284 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_negative_one
286 virtual bool is_one() const;
289 * Determine if an r-value has the value negative one
291 * The base implementation of this function always returns \c false. The
292 * \c ir_constant class over-rides this function to return \c true \b only
293 * for vector and scalar types that have all elements set to the value
294 * negative one. For boolean types, the result is always \c false.
296 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one
298 virtual bool is_negative_one() const;
301 * Determine if an r-value is an unsigned integer constant which can be
304 * \sa ir_constant::is_uint16_constant.
306 virtual bool is_uint16_constant() const { return false; }
309 * Return a generic value of error_type.
311 * Allocation will be performed with 'mem_ctx' as ralloc owner.
313 static ir_rvalue
*error_value(void *mem_ctx
);
316 ir_rvalue(enum ir_node_type t
);
321 * Variable storage classes
323 enum ir_variable_mode
{
324 ir_var_auto
= 0, /**< Function local variables and globals. */
325 ir_var_uniform
, /**< Variable declared as a uniform. */
330 ir_var_function_inout
,
331 ir_var_const_in
, /**< "in" param that must be a constant expression */
332 ir_var_system_value
, /**< Ex: front-face, instance-id, etc. */
333 ir_var_temporary
, /**< Temporary variable generated during compilation. */
334 ir_var_mode_count
/**< Number of variable modes */
338 * Enum keeping track of how a variable was declared. For error checking of
339 * the gl_PerVertex redeclaration rules.
341 enum ir_var_declaration_type
{
343 * Normal declaration (for most variables, this means an explicit
344 * declaration. Exception: temporaries are always implicitly declared, but
345 * they still use ir_var_declared_normally).
347 * Note: an ir_variable that represents a named interface block uses
348 * ir_var_declared_normally.
350 ir_var_declared_normally
= 0,
353 * Variable was explicitly declared (or re-declared) in an unnamed
356 ir_var_declared_in_block
,
359 * Variable is an implicitly declared built-in that has not been explicitly
360 * re-declared by the shader.
362 ir_var_declared_implicitly
,
365 * Variable is implicitly generated by the compiler and should not be
366 * visible via the API.
372 * \brief Layout qualifiers for gl_FragDepth.
374 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
375 * with a layout qualifier.
377 enum ir_depth_layout
{
378 ir_depth_layout_none
, /**< No depth layout is specified. */
380 ir_depth_layout_greater
,
381 ir_depth_layout_less
,
382 ir_depth_layout_unchanged
386 * \brief Convert depth layout qualifier to string.
389 depth_layout_string(ir_depth_layout layout
);
392 * Description of built-in state associated with a uniform
394 * \sa ir_variable::state_slots
396 struct ir_state_slot
{
403 * Get the string value for an interpolation qualifier
405 * \return The string that would be used in a shader to specify \c
406 * mode will be returned.
408 * This function is used to generate error messages of the form "shader
409 * uses %s interpolation qualifier", so in the case where there is no
410 * interpolation qualifier, it returns "no".
412 * This function should only be used on a shader input or output variable.
414 const char *interpolation_string(unsigned interpolation
);
417 class ir_variable
: public ir_instruction
{
419 ir_variable(const struct glsl_type
*, const char *, ir_variable_mode
);
421 virtual ir_variable
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
423 virtual void accept(ir_visitor
*v
)
428 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
432 * Determine how this variable should be interpolated based on its
433 * interpolation qualifier (if present), whether it is gl_Color or
434 * gl_SecondaryColor, and whether flatshading is enabled in the current GL
437 * The return value will always be either INTERP_QUALIFIER_SMOOTH,
438 * INTERP_QUALIFIER_NOPERSPECTIVE, or INTERP_QUALIFIER_FLAT.
440 glsl_interp_qualifier
determine_interpolation_mode(bool flat_shade
);
443 * Determine whether or not a variable is part of a uniform block.
445 inline bool is_in_uniform_block() const
447 return this->data
.mode
== ir_var_uniform
&& this->interface_type
!= NULL
;
451 * Determine whether or not a variable is the declaration of an interface
454 * For the first declaration below, there will be an \c ir_variable named
455 * "instance" whose type and whose instance_type will be the same
456 * \cglsl_type. For the second declaration, there will be an \c ir_variable
457 * named "f" whose type is float and whose instance_type is B2.
459 * "instance" is an interface instance variable, but "f" is not.
469 inline bool is_interface_instance() const
471 return this->type
->without_array() == this->interface_type
;
475 * Set this->interface_type on a newly created variable.
477 void init_interface_type(const struct glsl_type
*type
)
479 assert(this->interface_type
== NULL
);
480 this->interface_type
= type
;
481 if (this->is_interface_instance()) {
482 this->u
.max_ifc_array_access
=
483 rzalloc_array(this, unsigned, type
->length
);
488 * Change this->interface_type on a variable that previously had a
489 * different, but compatible, interface_type. This is used during linking
490 * to set the size of arrays in interface blocks.
492 void change_interface_type(const struct glsl_type
*type
)
494 if (this->u
.max_ifc_array_access
!= NULL
) {
495 /* max_ifc_array_access has already been allocated, so make sure the
496 * new interface has the same number of fields as the old one.
498 assert(this->interface_type
->length
== type
->length
);
500 this->interface_type
= type
;
504 * Change this->interface_type on a variable that previously had a
505 * different, and incompatible, interface_type. This is used during
506 * compilation to handle redeclaration of the built-in gl_PerVertex
509 void reinit_interface_type(const struct glsl_type
*type
)
511 if (this->u
.max_ifc_array_access
!= NULL
) {
513 /* Redeclaring gl_PerVertex is only allowed if none of the built-ins
514 * it defines have been accessed yet; so it's safe to throw away the
515 * old max_ifc_array_access pointer, since all of its values are
518 for (unsigned i
= 0; i
< this->interface_type
->length
; i
++)
519 assert(this->u
.max_ifc_array_access
[i
] == 0);
521 ralloc_free(this->u
.max_ifc_array_access
);
522 this->u
.max_ifc_array_access
= NULL
;
524 this->interface_type
= NULL
;
525 init_interface_type(type
);
528 const glsl_type
*get_interface_type() const
530 return this->interface_type
;
534 * Get the max_ifc_array_access pointer
536 * A "set" function is not needed because the array is dynmically allocated
539 inline unsigned *get_max_ifc_array_access()
541 assert(this->data
._num_state_slots
== 0);
542 return this->u
.max_ifc_array_access
;
545 inline unsigned get_num_state_slots() const
547 assert(!this->is_interface_instance()
548 || this->data
._num_state_slots
== 0);
549 return this->data
._num_state_slots
;
552 inline void set_num_state_slots(unsigned n
)
554 assert(!this->is_interface_instance()
556 this->data
._num_state_slots
= n
;
559 inline ir_state_slot
*get_state_slots()
561 return this->is_interface_instance() ? NULL
: this->u
.state_slots
;
564 inline const ir_state_slot
*get_state_slots() const
566 return this->is_interface_instance() ? NULL
: this->u
.state_slots
;
569 inline ir_state_slot
*allocate_state_slots(unsigned n
)
571 assert(!this->is_interface_instance());
573 this->u
.state_slots
= ralloc_array(this, ir_state_slot
, n
);
574 this->data
._num_state_slots
= 0;
576 if (this->u
.state_slots
!= NULL
)
577 this->data
._num_state_slots
= n
;
579 return this->u
.state_slots
;
582 inline bool is_name_ralloced() const
584 return this->name
!= ir_variable::tmp_name
;
588 * Enable emitting extension warnings for this variable
590 void enable_extension_warning(const char *extension
);
593 * Get the extension warning string for this variable
595 * If warnings are not enabled, \c NULL is returned.
597 const char *get_extension_warning() const;
600 * Declared type of the variable
602 const struct glsl_type
*type
;
605 * Declared name of the variable
609 struct ir_variable_data
{
612 * Is the variable read-only?
614 * This is set for variables declared as \c const, shader inputs,
617 unsigned read_only
:1;
620 unsigned invariant
:1;
624 * Has this variable been used for reading or writing?
626 * Several GLSL semantic checks require knowledge of whether or not a
627 * variable has been used. For example, it is an error to redeclare a
628 * variable as invariant after it has been used.
630 * This is only maintained in the ast_to_hir.cpp path, not in
631 * Mesa's fixed function or ARB program paths.
636 * Has this variable been statically assigned?
638 * This answers whether the variable was assigned in any path of
639 * the shader during ast_to_hir. This doesn't answer whether it is
640 * still written after dead code removal, nor is it maintained in
641 * non-ast_to_hir.cpp (GLSL parsing) paths.
646 * Enum indicating how the variable was declared. See
647 * ir_var_declaration_type.
649 * This is used to detect certain kinds of illegal variable redeclarations.
651 unsigned how_declared
:2;
654 * Storage class of the variable.
656 * \sa ir_variable_mode
661 * Interpolation mode for shader inputs / outputs
663 * \sa ir_variable_interpolation
665 unsigned interpolation
:2;
668 * \name ARB_fragment_coord_conventions
671 unsigned origin_upper_left
:1;
672 unsigned pixel_center_integer
:1;
676 * Was the location explicitly set in the shader?
678 * If the location is explicitly set in the shader, it \b cannot be changed
679 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
682 unsigned explicit_location
:1;
683 unsigned explicit_index
:1;
686 * Do we have a Vulkan (group, index) qualifier for this variable?
691 * Was an initial binding explicitly set in the shader?
693 * If so, constant_value contains an integer ir_constant representing the
694 * initial binding point.
696 unsigned explicit_binding
:1;
699 * Does this variable have an initializer?
701 * This is used by the linker to cross-validiate initializers of global
704 unsigned has_initializer
:1;
707 * Is this variable a generic output or input that has not yet been matched
708 * up to a variable in another stage of the pipeline?
710 * This is used by the linker as scratch storage while assigning locations
711 * to generic inputs and outputs.
713 unsigned is_unmatched_generic_inout
:1;
716 * If non-zero, then this variable may be packed along with other variables
717 * into a single varying slot, so this offset should be applied when
718 * accessing components. For example, an offset of 1 means that the x
719 * component of this variable is actually stored in component y of the
720 * location specified by \c location.
722 unsigned location_frac
:2;
725 * Layout of the matrix. Uses glsl_matrix_layout values.
727 unsigned matrix_layout
:2;
730 * Non-zero if this variable was created by lowering a named interface
731 * block which was not an array.
733 * Note that this variable and \c from_named_ifc_block_array will never
736 unsigned from_named_ifc_block_nonarray
:1;
739 * Non-zero if this variable was created by lowering a named interface
740 * block which was an array.
742 * Note that this variable and \c from_named_ifc_block_nonarray will never
745 unsigned from_named_ifc_block_array
:1;
748 * Non-zero if the variable must be a shader input. This is useful for
749 * constraints on function parameters.
751 unsigned must_be_shader_input
:1;
754 * Output index for dual source blending.
757 * The GLSL spec only allows the values 0 or 1 for the index in \b dual
760 * This is now also used for the Vulkan descriptor set index.
765 * \brief Layout qualifier for gl_FragDepth.
767 * This is not equal to \c ir_depth_layout_none if and only if this
768 * variable is \c gl_FragDepth and a layout qualifier is specified.
770 ir_depth_layout depth_layout
:3;
773 * ARB_shader_image_load_store qualifiers.
775 unsigned image_read_only
:1; /**< "readonly" qualifier. */
776 unsigned image_write_only
:1; /**< "writeonly" qualifier. */
777 unsigned image_coherent
:1;
778 unsigned image_volatile
:1;
779 unsigned image_restrict
:1;
782 * Emit a warning if this variable is accessed.
785 uint8_t warn_extension_index
;
788 /** Image internal format if specified explicitly, otherwise GL_NONE. */
789 uint16_t image_format
;
793 * Number of state slots used
796 * This could be stored in as few as 7-bits, if necessary. If it is made
797 * smaller, add an assertion to \c ir_variable::allocate_state_slots to
800 uint16_t _num_state_slots
;
804 * Initial binding point for a sampler, atomic, or UBO.
806 * For array types, this represents the binding point for the first element.
811 * Vulkan descriptor set for the resource.
816 * Storage location of the base of this variable
818 * The precise meaning of this field depends on the nature of the variable.
820 * - Vertex shader input: one of the values from \c gl_vert_attrib.
821 * - Vertex shader output: one of the values from \c gl_varying_slot.
822 * - Geometry shader input: one of the values from \c gl_varying_slot.
823 * - Geometry shader output: one of the values from \c gl_varying_slot.
824 * - Fragment shader input: one of the values from \c gl_varying_slot.
825 * - Fragment shader output: one of the values from \c gl_frag_result.
826 * - Uniforms: Per-stage uniform slot number for default uniform block.
827 * - Uniforms: Index within the uniform block definition for UBO members.
828 * - Other: This field is not currently used.
830 * If the variable is a uniform, shader input, or shader output, and the
831 * slot has not been assigned, the value will be -1.
836 * Vertex stream output identifier.
841 * Location an atomic counter is stored at.
848 * Highest element accessed with a constant expression array index
850 * Not used for non-array variables.
852 unsigned max_array_access
;
855 * Allow (only) ir_variable direct access private members.
857 friend class ir_variable
;
861 * Value assigned in the initializer of a variable declared "const"
863 ir_constant
*constant_value
;
866 * Constant expression assigned in the initializer of the variable
869 * This field and \c ::constant_value are distinct. Even if the two fields
870 * refer to constants with the same value, they must point to separate
873 ir_constant
*constant_initializer
;
876 static const char *const warn_extension_table
[];
880 * For variables which satisfy the is_interface_instance() predicate,
881 * this points to an array of integers such that if the ith member of
882 * the interface block is an array, max_ifc_array_access[i] is the
883 * maximum array element of that member that has been accessed. If the
884 * ith member of the interface block is not an array,
885 * max_ifc_array_access[i] is unused.
887 * For variables whose type is not an interface block, this pointer is
890 unsigned *max_ifc_array_access
;
893 * Built-in state that backs this uniform
895 * Once set at variable creation, \c state_slots must remain invariant.
897 * If the variable is not a uniform, \c _num_state_slots will be zero
898 * and \c state_slots will be \c NULL.
900 ir_state_slot
*state_slots
;
904 * For variables that are in an interface block or are an instance of an
905 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
907 * \sa ir_variable::location
909 const glsl_type
*interface_type
;
912 * Name used for anonymous compiler temporaries
914 static const char tmp_name
[];
918 * Should the construct keep names for ir_var_temporary variables?
920 * When this global is false, names passed to the constructor for
921 * \c ir_var_temporary variables will be dropped. Instead, the variable will
922 * be named "compiler_temp". This name will be in static storage.
925 * \b NEVER change the mode of an \c ir_var_temporary.
928 * This variable is \b not thread-safe. It is global, \b not
929 * per-context. It begins life false. A context can, at some point, make
930 * it true. From that point on, it will be true forever. This should be
931 * okay since it will only be set true while debugging.
933 static bool temporaries_allocate_names
;
937 * A function that returns whether a built-in function is available in the
938 * current shading language (based on version, ES or desktop, and extensions).
940 typedef bool (*builtin_available_predicate
)(const _mesa_glsl_parse_state
*);
944 * The representation of a function instance; may be the full definition or
945 * simply a prototype.
947 class ir_function_signature
: public ir_instruction
{
948 /* An ir_function_signature will be part of the list of signatures in
952 ir_function_signature(const glsl_type
*return_type
,
953 builtin_available_predicate builtin_avail
= NULL
);
955 virtual ir_function_signature
*clone(void *mem_ctx
,
956 struct hash_table
*ht
) const;
957 ir_function_signature
*clone_prototype(void *mem_ctx
,
958 struct hash_table
*ht
) const;
960 virtual void accept(ir_visitor
*v
)
965 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
968 * Attempt to evaluate this function as a constant expression,
969 * given a list of the actual parameters and the variable context.
970 * Returns NULL for non-built-ins.
972 ir_constant
*constant_expression_value(exec_list
*actual_parameters
, struct hash_table
*variable_context
);
975 * Get the name of the function for which this is a signature
977 const char *function_name() const;
980 * Get a handle to the function for which this is a signature
982 * There is no setter function, this function returns a \c const pointer,
983 * and \c ir_function_signature::_function is private for a reason. The
984 * only way to make a connection between a function and function signature
985 * is via \c ir_function::add_signature. This helps ensure that certain
986 * invariants (i.e., a function signature is in the list of signatures for
987 * its \c _function) are met.
989 * \sa ir_function::add_signature
991 inline const class ir_function
*function() const
993 return this->_function
;
997 * Check whether the qualifiers match between this signature's parameters
998 * and the supplied parameter list. If not, returns the name of the first
999 * parameter with mismatched qualifiers (for use in error messages).
1001 const char *qualifiers_match(exec_list
*params
);
1004 * Replace the current parameter list with the given one. This is useful
1005 * if the current information came from a prototype, and either has invalid
1006 * or missing parameter names.
1008 void replace_parameters(exec_list
*new_params
);
1011 * Function return type.
1013 * \note This discards the optional precision qualifier.
1015 const struct glsl_type
*return_type
;
1018 * List of ir_variable of function parameters.
1020 * This represents the storage. The paramaters passed in a particular
1021 * call will be in ir_call::actual_paramaters.
1023 struct exec_list parameters
;
1025 /** Whether or not this function has a body (which may be empty). */
1026 unsigned is_defined
:1;
1028 /** Whether or not this function signature is a built-in. */
1029 bool is_builtin() const;
1032 * Whether or not this function is an intrinsic to be implemented
1037 /** Whether or not a built-in is available for this shader. */
1038 bool is_builtin_available(const _mesa_glsl_parse_state
*state
) const;
1040 /** Body of instructions in the function. */
1041 struct exec_list body
;
1045 * A function pointer to a predicate that answers whether a built-in
1046 * function is available in the current shader. NULL if not a built-in.
1048 builtin_available_predicate builtin_avail
;
1050 /** Function of which this signature is one overload. */
1051 class ir_function
*_function
;
1053 /** Function signature of which this one is a prototype clone */
1054 const ir_function_signature
*origin
;
1056 friend class ir_function
;
1059 * Helper function to run a list of instructions for constant
1060 * expression evaluation.
1062 * The hash table represents the values of the visible variables.
1063 * There are no scoping issues because the table is indexed on
1064 * ir_variable pointers, not variable names.
1066 * Returns false if the expression is not constant, true otherwise,
1067 * and the value in *result if result is non-NULL.
1069 bool constant_expression_evaluate_expression_list(const struct exec_list
&body
,
1070 struct hash_table
*variable_context
,
1071 ir_constant
**result
);
1076 * Header for tracking multiple overloaded functions with the same name.
1077 * Contains a list of ir_function_signatures representing each of the
1080 class ir_function
: public ir_instruction
{
1082 ir_function(const char *name
);
1084 virtual ir_function
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1086 virtual void accept(ir_visitor
*v
)
1091 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1093 void add_signature(ir_function_signature
*sig
)
1095 sig
->_function
= this;
1096 this->signatures
.push_tail(sig
);
1100 * Find a signature that matches a set of actual parameters, taking implicit
1101 * conversions into account. Also flags whether the match was exact.
1103 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1104 const exec_list
*actual_param
,
1105 bool allow_builtins
,
1106 bool *match_is_exact
);
1109 * Find a signature that matches a set of actual parameters, taking implicit
1110 * conversions into account.
1112 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1113 const exec_list
*actual_param
,
1114 bool allow_builtins
);
1117 * Find a signature that exactly matches a set of actual parameters without
1118 * any implicit type conversions.
1120 ir_function_signature
*exact_matching_signature(_mesa_glsl_parse_state
*state
,
1121 const exec_list
*actual_ps
);
1124 * Name of the function.
1128 /** Whether or not this function has a signature that isn't a built-in. */
1129 bool has_user_signature();
1132 * List of ir_function_signature for each overloaded function with this name.
1134 struct exec_list signatures
;
1137 inline const char *ir_function_signature::function_name() const
1139 return this->_function
->name
;
1145 * IR instruction representing high-level if-statements
1147 class ir_if
: public ir_instruction
{
1149 ir_if(ir_rvalue
*condition
)
1150 : ir_instruction(ir_type_if
), condition(condition
)
1154 virtual ir_if
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1156 virtual void accept(ir_visitor
*v
)
1161 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1163 ir_rvalue
*condition
;
1164 /** List of ir_instruction for the body of the then branch */
1165 exec_list then_instructions
;
1166 /** List of ir_instruction for the body of the else branch */
1167 exec_list else_instructions
;
1172 * IR instruction representing a high-level loop structure.
1174 class ir_loop
: public ir_instruction
{
1178 virtual ir_loop
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1180 virtual void accept(ir_visitor
*v
)
1185 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1187 /** List of ir_instruction that make up the body of the loop. */
1188 exec_list body_instructions
;
1192 class ir_assignment
: public ir_instruction
{
1194 ir_assignment(ir_rvalue
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
= NULL
);
1197 * Construct an assignment with an explicit write mask
1200 * Since a write mask is supplied, the LHS must already be a bare
1201 * \c ir_dereference. The cannot be any swizzles in the LHS.
1203 ir_assignment(ir_dereference
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
,
1204 unsigned write_mask
);
1206 virtual ir_assignment
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1208 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1210 virtual void accept(ir_visitor
*v
)
1215 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1218 * Get a whole variable written by an assignment
1220 * If the LHS of the assignment writes a whole variable, the variable is
1221 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1224 * - Assigning to a scalar
1225 * - Assigning to all components of a vector
1226 * - Whole array (or matrix) assignment
1227 * - Whole structure assignment
1229 ir_variable
*whole_variable_written();
1232 * Set the LHS of an assignment
1234 void set_lhs(ir_rvalue
*lhs
);
1237 * Left-hand side of the assignment.
1239 * This should be treated as read only. If you need to set the LHS of an
1240 * assignment, use \c ir_assignment::set_lhs.
1242 ir_dereference
*lhs
;
1245 * Value being assigned
1250 * Optional condition for the assignment.
1252 ir_rvalue
*condition
;
1256 * Component mask written
1258 * For non-vector types in the LHS, this field will be zero. For vector
1259 * types, a bit will be set for each component that is written. Note that
1260 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1262 * A partially-set write mask means that each enabled channel gets
1263 * the value from a consecutive channel of the rhs. For example,
1264 * to write just .xyw of gl_FrontColor with color:
1266 * (assign (constant bool (1)) (xyw)
1267 * (var_ref gl_FragColor)
1268 * (swiz xyw (var_ref color)))
1270 unsigned write_mask
:4;
1273 /* Update ir_expression::get_num_operands() and operator_strs when
1274 * updating this list.
1276 enum ir_expression_operation
{
1285 ir_unop_exp
, /**< Log base e on gentype */
1286 ir_unop_log
, /**< Natural log on gentype */
1289 ir_unop_f2i
, /**< Float-to-integer conversion. */
1290 ir_unop_f2u
, /**< Float-to-unsigned conversion. */
1291 ir_unop_i2f
, /**< Integer-to-float conversion. */
1292 ir_unop_f2b
, /**< Float-to-boolean conversion */
1293 ir_unop_b2f
, /**< Boolean-to-float conversion */
1294 ir_unop_i2b
, /**< int-to-boolean conversion */
1295 ir_unop_b2i
, /**< Boolean-to-int conversion */
1296 ir_unop_u2f
, /**< Unsigned-to-float conversion. */
1297 ir_unop_i2u
, /**< Integer-to-unsigned conversion. */
1298 ir_unop_u2i
, /**< Unsigned-to-integer conversion. */
1299 ir_unop_d2f
, /**< Double-to-float conversion. */
1300 ir_unop_f2d
, /**< Float-to-double conversion. */
1301 ir_unop_d2i
, /**< Double-to-integer conversion. */
1302 ir_unop_i2d
, /**< Integer-to-double conversion. */
1303 ir_unop_d2u
, /**< Double-to-unsigned conversion. */
1304 ir_unop_u2d
, /**< Unsigned-to-double conversion. */
1305 ir_unop_d2b
, /**< Double-to-boolean conversion. */
1306 ir_unop_bitcast_i2f
, /**< Bit-identical int-to-float "conversion" */
1307 ir_unop_bitcast_f2i
, /**< Bit-identical float-to-int "conversion" */
1308 ir_unop_bitcast_u2f
, /**< Bit-identical uint-to-float "conversion" */
1309 ir_unop_bitcast_f2u
, /**< Bit-identical float-to-uint "conversion" */
1313 * \name Unary floating-point rounding operations.
1324 * \name Trigonometric operations.
1332 * \name Partial derivatives.
1336 ir_unop_dFdx_coarse
,
1339 ir_unop_dFdy_coarse
,
1344 * \name Floating point pack and unpack operations.
1347 ir_unop_pack_snorm_2x16
,
1348 ir_unop_pack_snorm_4x8
,
1349 ir_unop_pack_unorm_2x16
,
1350 ir_unop_pack_unorm_4x8
,
1351 ir_unop_pack_half_2x16
,
1352 ir_unop_unpack_snorm_2x16
,
1353 ir_unop_unpack_snorm_4x8
,
1354 ir_unop_unpack_unorm_2x16
,
1355 ir_unop_unpack_unorm_4x8
,
1356 ir_unop_unpack_half_2x16
,
1360 * \name Lowered floating point unpacking operations.
1362 * \see lower_packing_builtins_visitor::split_unpack_half_2x16
1365 ir_unop_unpack_half_2x16_split_x
,
1366 ir_unop_unpack_half_2x16_split_y
,
1370 * \name Bit operations, part of ARB_gpu_shader5.
1373 ir_unop_bitfield_reverse
,
1382 * \name Double packing, part of ARB_gpu_shader_fp64.
1385 ir_unop_pack_double_2x32
,
1386 ir_unop_unpack_double_2x32
,
1395 * Interpolate fs input at centroid
1397 * operand0 is the fs input.
1399 ir_unop_interpolate_at_centroid
,
1402 * A sentinel marking the last of the unary operations.
1404 ir_last_unop
= ir_unop_interpolate_at_centroid
,
1408 ir_binop_mul
, /**< Floating-point or low 32-bit integer multiply. */
1409 ir_binop_imul_high
, /**< Calculates the high 32-bits of a 64-bit multiply. */
1413 * Returns the carry resulting from the addition of the two arguments.
1420 * Returns the borrow resulting from the subtraction of the second argument
1421 * from the first argument.
1428 * Takes one of two combinations of arguments:
1431 * - mod(vecN, float)
1433 * Does not take integer types.
1438 * \name Binary comparison operators which return a boolean vector.
1439 * The type of both operands must be equal.
1449 * Returns single boolean for whether all components of operands[0]
1450 * equal the components of operands[1].
1454 * Returns single boolean for whether any component of operands[0]
1455 * is not equal to the corresponding component of operands[1].
1457 ir_binop_any_nequal
,
1461 * \name Bit-wise binary operations.
1482 * \name Lowered floating point packing operations.
1484 * \see lower_packing_builtins_visitor::split_pack_half_2x16
1487 ir_binop_pack_half_2x16_split
,
1491 * \name First half of a lowered bitfieldInsert() operation.
1493 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1500 * Load a value the size of a given GLSL type from a uniform block.
1502 * operand0 is the ir_constant uniform block index in the linked shader.
1503 * operand1 is a byte offset within the uniform block.
1508 * \name Multiplies a number by two to a power, part of ARB_gpu_shader5.
1515 * Extract a scalar from a vector
1517 * operand0 is the vector
1518 * operand1 is the index of the field to read from operand0
1520 ir_binop_vector_extract
,
1523 * Interpolate fs input at offset
1525 * operand0 is the fs input
1526 * operand1 is the offset from the pixel center
1528 ir_binop_interpolate_at_offset
,
1531 * Interpolate fs input at sample position
1533 * operand0 is the fs input
1534 * operand1 is the sample ID
1536 ir_binop_interpolate_at_sample
,
1539 * A sentinel marking the last of the binary operations.
1541 ir_last_binop
= ir_binop_interpolate_at_sample
,
1544 * \name Fused floating-point multiply-add, part of ARB_gpu_shader5.
1553 * \name Conditional Select
1555 * A vector conditional select instruction (like ?:, but operating per-
1556 * component on vectors).
1558 * \see lower_instructions_visitor::ldexp_to_arith
1565 * \name Second half of a lowered bitfieldInsert() operation.
1567 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1573 ir_triop_bitfield_extract
,
1576 * Generate a value with one field of a vector changed
1578 * operand0 is the vector
1579 * operand1 is the value to write into the vector result
1580 * operand2 is the index in operand0 to be modified
1582 ir_triop_vector_insert
,
1585 * A sentinel marking the last of the ternary operations.
1587 ir_last_triop
= ir_triop_vector_insert
,
1589 ir_quadop_bitfield_insert
,
1594 * A sentinel marking the last of the ternary operations.
1596 ir_last_quadop
= ir_quadop_vector
,
1599 * A sentinel marking the last of all operations.
1601 ir_last_opcode
= ir_quadop_vector
1604 class ir_expression
: public ir_rvalue
{
1606 ir_expression(int op
, const struct glsl_type
*type
,
1607 ir_rvalue
*op0
, ir_rvalue
*op1
= NULL
,
1608 ir_rvalue
*op2
= NULL
, ir_rvalue
*op3
= NULL
);
1611 * Constructor for unary operation expressions
1613 ir_expression(int op
, ir_rvalue
*);
1616 * Constructor for binary operation expressions
1618 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
);
1621 * Constructor for ternary operation expressions
1623 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
, ir_rvalue
*op2
);
1625 virtual bool equals(const ir_instruction
*ir
,
1626 enum ir_node_type ignore
= ir_type_unset
) const;
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(const ir_instruction
*ir
,
1938 enum ir_node_type ignore
= ir_type_unset
) const;
1941 * Return a string representing the ir_texture_opcode.
1943 const char *opcode_string();
1945 /** Set the sampler and type. */
1946 void set_sampler(ir_dereference
*sampler
, const glsl_type
*type
);
1949 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1951 static ir_texture_opcode
get_opcode(const char *);
1953 enum ir_texture_opcode op
;
1955 /** Sampler to use for the texture access. */
1956 ir_dereference
*sampler
;
1958 /** Texture coordinate to sample */
1959 ir_rvalue
*coordinate
;
1962 * Value used for projective divide.
1964 * If there is no projective divide (the common case), this will be
1965 * \c NULL. Optimization passes should check for this to point to a constant
1966 * of 1.0 and replace that with \c NULL.
1968 ir_rvalue
*projector
;
1971 * Coordinate used for comparison on shadow look-ups.
1973 * If there is no shadow comparison, this will be \c NULL. For the
1974 * \c ir_txf opcode, this *must* be \c NULL.
1976 ir_rvalue
*shadow_comparitor
;
1978 /** Texel offset. */
1982 ir_rvalue
*lod
; /**< Floating point LOD */
1983 ir_rvalue
*bias
; /**< Floating point LOD bias */
1984 ir_rvalue
*sample_index
; /**< MSAA sample index */
1985 ir_rvalue
*component
; /**< Gather component selector */
1987 ir_rvalue
*dPdx
; /**< Partial derivative of coordinate wrt X */
1988 ir_rvalue
*dPdy
; /**< Partial derivative of coordinate wrt Y */
1994 struct ir_swizzle_mask
{
2001 * Number of components in the swizzle.
2003 unsigned num_components
:3;
2006 * Does the swizzle contain duplicate components?
2008 * L-value swizzles cannot contain duplicate components.
2010 unsigned has_duplicates
:1;
2014 class ir_swizzle
: public ir_rvalue
{
2016 ir_swizzle(ir_rvalue
*, unsigned x
, unsigned y
, unsigned z
, unsigned w
,
2019 ir_swizzle(ir_rvalue
*val
, const unsigned *components
, unsigned count
);
2021 ir_swizzle(ir_rvalue
*val
, ir_swizzle_mask mask
);
2023 virtual ir_swizzle
*clone(void *mem_ctx
, struct hash_table
*) const;
2025 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2028 * Construct an ir_swizzle from the textual representation. Can fail.
2030 static ir_swizzle
*create(ir_rvalue
*, const char *, unsigned vector_length
);
2032 virtual void accept(ir_visitor
*v
)
2037 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2039 virtual bool equals(const ir_instruction
*ir
,
2040 enum ir_node_type ignore
= ir_type_unset
) const;
2042 bool is_lvalue() const
2044 return val
->is_lvalue() && !mask
.has_duplicates
;
2048 * Get the variable that is ultimately referenced by an r-value
2050 virtual ir_variable
*variable_referenced() const;
2053 ir_swizzle_mask mask
;
2057 * Initialize the mask component of a swizzle
2059 * This is used by the \c ir_swizzle constructors.
2061 void init_mask(const unsigned *components
, unsigned count
);
2065 class ir_dereference
: public ir_rvalue
{
2067 virtual ir_dereference
*clone(void *mem_ctx
, struct hash_table
*) const = 0;
2069 bool is_lvalue() const;
2072 * Get the variable that is ultimately referenced by an r-value
2074 virtual ir_variable
*variable_referenced() const = 0;
2077 ir_dereference(enum ir_node_type t
)
2084 class ir_dereference_variable
: public ir_dereference
{
2086 ir_dereference_variable(ir_variable
*var
);
2088 virtual ir_dereference_variable
*clone(void *mem_ctx
,
2089 struct hash_table
*) const;
2091 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2093 virtual bool equals(const ir_instruction
*ir
,
2094 enum ir_node_type ignore
= ir_type_unset
) const;
2097 * Get the variable that is ultimately referenced by an r-value
2099 virtual ir_variable
*variable_referenced() const
2104 virtual ir_variable
*whole_variable_referenced()
2106 /* ir_dereference_variable objects always dereference the entire
2107 * variable. However, if this dereference is dereferenced by anything
2108 * else, the complete deferefernce chain is not a whole-variable
2109 * dereference. This method should only be called on the top most
2110 * ir_rvalue in a dereference chain.
2115 virtual void accept(ir_visitor
*v
)
2120 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2123 * Object being dereferenced.
2129 class ir_dereference_array
: public ir_dereference
{
2131 ir_dereference_array(ir_rvalue
*value
, ir_rvalue
*array_index
);
2133 ir_dereference_array(ir_variable
*var
, ir_rvalue
*array_index
);
2135 virtual ir_dereference_array
*clone(void *mem_ctx
,
2136 struct hash_table
*) const;
2138 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2140 virtual bool equals(const ir_instruction
*ir
,
2141 enum ir_node_type ignore
= ir_type_unset
) const;
2144 * Get the variable that is ultimately referenced by an r-value
2146 virtual ir_variable
*variable_referenced() const
2148 return this->array
->variable_referenced();
2151 virtual void accept(ir_visitor
*v
)
2156 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2159 ir_rvalue
*array_index
;
2162 void set_array(ir_rvalue
*value
);
2166 class ir_dereference_record
: public ir_dereference
{
2168 ir_dereference_record(ir_rvalue
*value
, const char *field
);
2170 ir_dereference_record(ir_variable
*var
, const char *field
);
2172 virtual ir_dereference_record
*clone(void *mem_ctx
,
2173 struct hash_table
*) const;
2175 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2178 * Get the variable that is ultimately referenced by an r-value
2180 virtual ir_variable
*variable_referenced() const
2182 return this->record
->variable_referenced();
2185 virtual void accept(ir_visitor
*v
)
2190 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2198 * Data stored in an ir_constant
2200 union ir_constant_data
{
2209 class ir_constant
: public ir_rvalue
{
2211 ir_constant(const struct glsl_type
*type
, const ir_constant_data
*data
);
2212 ir_constant(bool b
, unsigned vector_elements
=1);
2213 ir_constant(unsigned int u
, unsigned vector_elements
=1);
2214 ir_constant(int i
, unsigned vector_elements
=1);
2215 ir_constant(float f
, unsigned vector_elements
=1);
2216 ir_constant(double d
, unsigned vector_elements
=1);
2219 * Construct an ir_constant from a list of ir_constant values
2221 ir_constant(const struct glsl_type
*type
, exec_list
*values
);
2224 * Construct an ir_constant from a scalar component of another ir_constant
2226 * The new \c ir_constant inherits the type of the component from the
2230 * In the case of a matrix constant, the new constant is a scalar, \b not
2233 ir_constant(const ir_constant
*c
, unsigned i
);
2236 * Return a new ir_constant of the specified type containing all zeros.
2238 static ir_constant
*zero(void *mem_ctx
, const glsl_type
*type
);
2240 virtual ir_constant
*clone(void *mem_ctx
, struct hash_table
*) const;
2242 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2244 virtual void accept(ir_visitor
*v
)
2249 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2251 virtual bool equals(const ir_instruction
*ir
,
2252 enum ir_node_type ignore
= ir_type_unset
) const;
2255 * Get a particular component of a constant as a specific type
2257 * This is useful, for example, to get a value from an integer constant
2258 * as a float or bool. This appears frequently when constructors are
2259 * called with all constant parameters.
2262 bool get_bool_component(unsigned i
) const;
2263 float get_float_component(unsigned i
) const;
2264 double get_double_component(unsigned i
) const;
2265 int get_int_component(unsigned i
) const;
2266 unsigned get_uint_component(unsigned i
) const;
2269 ir_constant
*get_array_element(unsigned i
) const;
2271 ir_constant
*get_record_field(const char *name
);
2274 * Copy the values on another constant at a given offset.
2276 * The offset is ignored for array or struct copies, it's only for
2277 * scalars or vectors into vectors or matrices.
2279 * With identical types on both sides and zero offset it's clone()
2280 * without creating a new object.
2283 void copy_offset(ir_constant
*src
, int offset
);
2286 * Copy the values on another constant at a given offset and
2287 * following an assign-like mask.
2289 * The mask is ignored for scalars.
2291 * Note that this function only handles what assign can handle,
2292 * i.e. at most a vector as source and a column of a matrix as
2296 void copy_masked_offset(ir_constant
*src
, int offset
, unsigned int mask
);
2299 * Determine whether a constant has the same value as another constant
2301 * \sa ir_constant::is_zero, ir_constant::is_one,
2302 * ir_constant::is_negative_one
2304 bool has_value(const ir_constant
*) const;
2307 * Return true if this ir_constant represents the given value.
2309 * For vectors, this checks that each component is the given value.
2311 virtual bool is_value(float f
, int i
) const;
2312 virtual bool is_zero() const;
2313 virtual bool is_one() const;
2314 virtual bool is_negative_one() const;
2317 * Return true for constants that could be stored as 16-bit unsigned values.
2319 * Note that this will return true even for signed integer ir_constants, as
2320 * long as the value is non-negative and fits in 16-bits.
2322 virtual bool is_uint16_constant() const;
2325 * Value of the constant.
2327 * The field used to back the values supplied by the constant is determined
2328 * by the type associated with the \c ir_instruction. Constants may be
2329 * scalars, vectors, or matrices.
2331 union ir_constant_data value
;
2333 /* Array elements */
2334 ir_constant
**array_elements
;
2336 /* Structure fields */
2337 exec_list components
;
2341 * Parameterless constructor only used by the clone method
2347 * IR instruction to emit a vertex in a geometry shader.
2349 class ir_emit_vertex
: public ir_instruction
{
2351 ir_emit_vertex(ir_rvalue
*stream
)
2352 : ir_instruction(ir_type_emit_vertex
),
2358 virtual void accept(ir_visitor
*v
)
2363 virtual ir_emit_vertex
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2365 return new(mem_ctx
) ir_emit_vertex(this->stream
->clone(mem_ctx
, ht
));
2368 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2370 int stream_id() const
2372 return stream
->as_constant()->value
.i
[0];
2379 * IR instruction to complete the current primitive and start a new one in a
2382 class ir_end_primitive
: public ir_instruction
{
2384 ir_end_primitive(ir_rvalue
*stream
)
2385 : ir_instruction(ir_type_end_primitive
),
2391 virtual void accept(ir_visitor
*v
)
2396 virtual ir_end_primitive
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2398 return new(mem_ctx
) ir_end_primitive(this->stream
->clone(mem_ctx
, ht
));
2401 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2403 int stream_id() const
2405 return stream
->as_constant()->value
.i
[0];
2414 * Apply a visitor to each IR node in a list
2417 visit_exec_list(exec_list
*list
, ir_visitor
*visitor
);
2420 * Validate invariants on each IR node in a list
2422 void validate_ir_tree(exec_list
*instructions
);
2424 struct _mesa_glsl_parse_state
;
2425 struct gl_shader_program
;
2428 * Detect whether an unlinked shader contains static recursion
2430 * If the list of instructions is determined to contain static recursion,
2431 * \c _mesa_glsl_error will be called to emit error messages for each function
2432 * that is in the recursion cycle.
2435 detect_recursion_unlinked(struct _mesa_glsl_parse_state
*state
,
2436 exec_list
*instructions
);
2439 * Detect whether a linked shader contains static recursion
2441 * If the list of instructions is determined to contain static recursion,
2442 * \c link_error_printf will be called to emit error messages for each function
2443 * that is in the recursion cycle. In addition,
2444 * \c gl_shader_program::LinkStatus will be set to false.
2447 detect_recursion_linked(struct gl_shader_program
*prog
,
2448 exec_list
*instructions
);
2451 * Make a clone of each IR instruction in a list
2453 * \param in List of IR instructions that are to be cloned
2454 * \param out List to hold the cloned instructions
2457 clone_ir_list(void *mem_ctx
, exec_list
*out
, const exec_list
*in
);
2460 _mesa_glsl_initialize_variables(exec_list
*instructions
,
2461 struct _mesa_glsl_parse_state
*state
);
2464 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state
*state
);
2467 _mesa_glsl_initialize_builtin_functions();
2469 extern ir_function_signature
*
2470 _mesa_glsl_find_builtin_function(_mesa_glsl_parse_state
*state
,
2471 const char *name
, exec_list
*actual_parameters
);
2473 extern ir_function
*
2474 _mesa_glsl_find_builtin_function_by_name(_mesa_glsl_parse_state
*state
,
2478 _mesa_glsl_get_builtin_function_shader(void);
2481 _mesa_glsl_release_functions(void);
2484 _mesa_glsl_release_builtin_functions(void);
2487 reparent_ir(exec_list
*list
, void *mem_ctx
);
2489 struct glsl_symbol_table
;
2492 import_prototypes(const exec_list
*source
, exec_list
*dest
,
2493 struct glsl_symbol_table
*symbols
, void *mem_ctx
);
2496 ir_has_call(ir_instruction
*ir
);
2499 do_set_program_inouts(exec_list
*instructions
, struct gl_program
*prog
,
2500 gl_shader_stage shader_stage
);
2503 prototype_string(const glsl_type
*return_type
, const char *name
,
2504 exec_list
*parameters
);
2507 mode_string(const ir_variable
*var
);
2510 * Built-in / reserved GL variables names start with "gl_"
2513 is_gl_identifier(const char *s
)
2515 return s
&& s
[0] == 'g' && s
[1] == 'l' && s
[2] == '_';
2519 #endif /* __cplusplus */
2521 extern void _mesa_print_ir(FILE *f
, struct exec_list
*instructions
,
2522 struct _mesa_glsl_parse_state
*state
);
2525 fprint_ir(FILE *f
, const void *instruction
);
2532 vertices_per_prim(GLenum prim
);