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 * Was an initial binding explicitly set in the shader?
688 * If so, constant_value contains an integer ir_constant representing the
689 * initial binding point.
691 unsigned explicit_binding
:1;
694 * Does this variable have an initializer?
696 * This is used by the linker to cross-validiate initializers of global
699 unsigned has_initializer
:1;
702 * Is this variable a generic output or input that has not yet been matched
703 * up to a variable in another stage of the pipeline?
705 * This is used by the linker as scratch storage while assigning locations
706 * to generic inputs and outputs.
708 unsigned is_unmatched_generic_inout
:1;
711 * If non-zero, then this variable may be packed along with other variables
712 * into a single varying slot, so this offset should be applied when
713 * accessing components. For example, an offset of 1 means that the x
714 * component of this variable is actually stored in component y of the
715 * location specified by \c location.
717 unsigned location_frac
:2;
720 * Layout of the matrix. Uses glsl_matrix_layout values.
722 unsigned matrix_layout
:2;
725 * Non-zero if this variable was created by lowering a named interface
726 * block which was not an array.
728 * Note that this variable and \c from_named_ifc_block_array will never
731 unsigned from_named_ifc_block_nonarray
:1;
734 * Non-zero if this variable was created by lowering a named interface
735 * block which was an array.
737 * Note that this variable and \c from_named_ifc_block_nonarray will never
740 unsigned from_named_ifc_block_array
:1;
743 * Non-zero if the variable must be a shader input. This is useful for
744 * constraints on function parameters.
746 unsigned must_be_shader_input
:1;
749 * Output index for dual source blending.
752 * The GLSL spec only allows the values 0 or 1 for the index in \b dual
758 * \brief Layout qualifier for gl_FragDepth.
760 * This is not equal to \c ir_depth_layout_none if and only if this
761 * variable is \c gl_FragDepth and a layout qualifier is specified.
763 ir_depth_layout depth_layout
:3;
766 * ARB_shader_image_load_store qualifiers.
768 unsigned image_read_only
:1; /**< "readonly" qualifier. */
769 unsigned image_write_only
:1; /**< "writeonly" qualifier. */
770 unsigned image_coherent
:1;
771 unsigned image_volatile
:1;
772 unsigned image_restrict
:1;
775 * Emit a warning if this variable is accessed.
778 uint8_t warn_extension_index
;
781 /** Image internal format if specified explicitly, otherwise GL_NONE. */
782 uint16_t image_format
;
786 * Number of state slots used
789 * This could be stored in as few as 7-bits, if necessary. If it is made
790 * smaller, add an assertion to \c ir_variable::allocate_state_slots to
793 uint16_t _num_state_slots
;
797 * Initial binding point for a sampler, atomic, or UBO.
799 * For array types, this represents the binding point for the first element.
804 * Storage location of the base of this variable
806 * The precise meaning of this field depends on the nature of the variable.
808 * - Vertex shader input: one of the values from \c gl_vert_attrib.
809 * - Vertex shader output: one of the values from \c gl_varying_slot.
810 * - Geometry shader input: one of the values from \c gl_varying_slot.
811 * - Geometry shader output: one of the values from \c gl_varying_slot.
812 * - Fragment shader input: one of the values from \c gl_varying_slot.
813 * - Fragment shader output: one of the values from \c gl_frag_result.
814 * - Uniforms: Per-stage uniform slot number for default uniform block.
815 * - Uniforms: Index within the uniform block definition for UBO members.
816 * - Other: This field is not currently used.
818 * If the variable is a uniform, shader input, or shader output, and the
819 * slot has not been assigned, the value will be -1.
824 * Vertex stream output identifier.
829 * Location an atomic counter is stored at.
836 * Highest element accessed with a constant expression array index
838 * Not used for non-array variables.
840 unsigned max_array_access
;
843 * Allow (only) ir_variable direct access private members.
845 friend class ir_variable
;
849 * Value assigned in the initializer of a variable declared "const"
851 ir_constant
*constant_value
;
854 * Constant expression assigned in the initializer of the variable
857 * This field and \c ::constant_value are distinct. Even if the two fields
858 * refer to constants with the same value, they must point to separate
861 ir_constant
*constant_initializer
;
864 static const char *const warn_extension_table
[];
868 * For variables which satisfy the is_interface_instance() predicate,
869 * this points to an array of integers such that if the ith member of
870 * the interface block is an array, max_ifc_array_access[i] is the
871 * maximum array element of that member that has been accessed. If the
872 * ith member of the interface block is not an array,
873 * max_ifc_array_access[i] is unused.
875 * For variables whose type is not an interface block, this pointer is
878 unsigned *max_ifc_array_access
;
881 * Built-in state that backs this uniform
883 * Once set at variable creation, \c state_slots must remain invariant.
885 * If the variable is not a uniform, \c _num_state_slots will be zero
886 * and \c state_slots will be \c NULL.
888 ir_state_slot
*state_slots
;
892 * For variables that are in an interface block or are an instance of an
893 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
895 * \sa ir_variable::location
897 const glsl_type
*interface_type
;
900 * Name used for anonymous compiler temporaries
902 static const char tmp_name
[];
906 * Should the construct keep names for ir_var_temporary variables?
908 * When this global is false, names passed to the constructor for
909 * \c ir_var_temporary variables will be dropped. Instead, the variable will
910 * be named "compiler_temp". This name will be in static storage.
913 * \b NEVER change the mode of an \c ir_var_temporary.
916 * This variable is \b not thread-safe. It is global, \b not
917 * per-context. It begins life false. A context can, at some point, make
918 * it true. From that point on, it will be true forever. This should be
919 * okay since it will only be set true while debugging.
921 static bool temporaries_allocate_names
;
925 * A function that returns whether a built-in function is available in the
926 * current shading language (based on version, ES or desktop, and extensions).
928 typedef bool (*builtin_available_predicate
)(const _mesa_glsl_parse_state
*);
932 * The representation of a function instance; may be the full definition or
933 * simply a prototype.
935 class ir_function_signature
: public ir_instruction
{
936 /* An ir_function_signature will be part of the list of signatures in
940 ir_function_signature(const glsl_type
*return_type
,
941 builtin_available_predicate builtin_avail
= NULL
);
943 virtual ir_function_signature
*clone(void *mem_ctx
,
944 struct hash_table
*ht
) const;
945 ir_function_signature
*clone_prototype(void *mem_ctx
,
946 struct hash_table
*ht
) const;
948 virtual void accept(ir_visitor
*v
)
953 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
956 * Attempt to evaluate this function as a constant expression,
957 * given a list of the actual parameters and the variable context.
958 * Returns NULL for non-built-ins.
960 ir_constant
*constant_expression_value(exec_list
*actual_parameters
, struct hash_table
*variable_context
);
963 * Get the name of the function for which this is a signature
965 const char *function_name() const;
968 * Get a handle to the function for which this is a signature
970 * There is no setter function, this function returns a \c const pointer,
971 * and \c ir_function_signature::_function is private for a reason. The
972 * only way to make a connection between a function and function signature
973 * is via \c ir_function::add_signature. This helps ensure that certain
974 * invariants (i.e., a function signature is in the list of signatures for
975 * its \c _function) are met.
977 * \sa ir_function::add_signature
979 inline const class ir_function
*function() const
981 return this->_function
;
985 * Check whether the qualifiers match between this signature's parameters
986 * and the supplied parameter list. If not, returns the name of the first
987 * parameter with mismatched qualifiers (for use in error messages).
989 const char *qualifiers_match(exec_list
*params
);
992 * Replace the current parameter list with the given one. This is useful
993 * if the current information came from a prototype, and either has invalid
994 * or missing parameter names.
996 void replace_parameters(exec_list
*new_params
);
999 * Function return type.
1001 * \note This discards the optional precision qualifier.
1003 const struct glsl_type
*return_type
;
1006 * List of ir_variable of function parameters.
1008 * This represents the storage. The paramaters passed in a particular
1009 * call will be in ir_call::actual_paramaters.
1011 struct exec_list parameters
;
1013 /** Whether or not this function has a body (which may be empty). */
1014 unsigned is_defined
:1;
1016 /** Whether or not this function signature is a built-in. */
1017 bool is_builtin() const;
1020 * Whether or not this function is an intrinsic to be implemented
1025 /** Whether or not a built-in is available for this shader. */
1026 bool is_builtin_available(const _mesa_glsl_parse_state
*state
) const;
1028 /** Body of instructions in the function. */
1029 struct exec_list body
;
1033 * A function pointer to a predicate that answers whether a built-in
1034 * function is available in the current shader. NULL if not a built-in.
1036 builtin_available_predicate builtin_avail
;
1038 /** Function of which this signature is one overload. */
1039 class ir_function
*_function
;
1041 /** Function signature of which this one is a prototype clone */
1042 const ir_function_signature
*origin
;
1044 friend class ir_function
;
1047 * Helper function to run a list of instructions for constant
1048 * expression evaluation.
1050 * The hash table represents the values of the visible variables.
1051 * There are no scoping issues because the table is indexed on
1052 * ir_variable pointers, not variable names.
1054 * Returns false if the expression is not constant, true otherwise,
1055 * and the value in *result if result is non-NULL.
1057 bool constant_expression_evaluate_expression_list(const struct exec_list
&body
,
1058 struct hash_table
*variable_context
,
1059 ir_constant
**result
);
1064 * Header for tracking multiple overloaded functions with the same name.
1065 * Contains a list of ir_function_signatures representing each of the
1068 class ir_function
: public ir_instruction
{
1070 ir_function(const char *name
);
1072 virtual ir_function
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1074 virtual void accept(ir_visitor
*v
)
1079 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1081 void add_signature(ir_function_signature
*sig
)
1083 sig
->_function
= this;
1084 this->signatures
.push_tail(sig
);
1088 * Find a signature that matches a set of actual parameters, taking implicit
1089 * conversions into account. Also flags whether the match was exact.
1091 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1092 const exec_list
*actual_param
,
1093 bool allow_builtins
,
1094 bool *match_is_exact
);
1097 * Find a signature that matches a set of actual parameters, taking implicit
1098 * conversions into account.
1100 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1101 const exec_list
*actual_param
,
1102 bool allow_builtins
);
1105 * Find a signature that exactly matches a set of actual parameters without
1106 * any implicit type conversions.
1108 ir_function_signature
*exact_matching_signature(_mesa_glsl_parse_state
*state
,
1109 const exec_list
*actual_ps
);
1112 * Name of the function.
1116 /** Whether or not this function has a signature that isn't a built-in. */
1117 bool has_user_signature();
1120 * List of ir_function_signature for each overloaded function with this name.
1122 struct exec_list signatures
;
1125 inline const char *ir_function_signature::function_name() const
1127 return this->_function
->name
;
1133 * IR instruction representing high-level if-statements
1135 class ir_if
: public ir_instruction
{
1137 ir_if(ir_rvalue
*condition
)
1138 : ir_instruction(ir_type_if
), condition(condition
)
1142 virtual ir_if
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1144 virtual void accept(ir_visitor
*v
)
1149 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1151 ir_rvalue
*condition
;
1152 /** List of ir_instruction for the body of the then branch */
1153 exec_list then_instructions
;
1154 /** List of ir_instruction for the body of the else branch */
1155 exec_list else_instructions
;
1160 * IR instruction representing a high-level loop structure.
1162 class ir_loop
: public ir_instruction
{
1166 virtual ir_loop
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1168 virtual void accept(ir_visitor
*v
)
1173 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1175 /** List of ir_instruction that make up the body of the loop. */
1176 exec_list body_instructions
;
1180 class ir_assignment
: public ir_instruction
{
1182 ir_assignment(ir_rvalue
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
= NULL
);
1185 * Construct an assignment with an explicit write mask
1188 * Since a write mask is supplied, the LHS must already be a bare
1189 * \c ir_dereference. The cannot be any swizzles in the LHS.
1191 ir_assignment(ir_dereference
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
,
1192 unsigned write_mask
);
1194 virtual ir_assignment
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1196 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1198 virtual void accept(ir_visitor
*v
)
1203 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1206 * Get a whole variable written by an assignment
1208 * If the LHS of the assignment writes a whole variable, the variable is
1209 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1212 * - Assigning to a scalar
1213 * - Assigning to all components of a vector
1214 * - Whole array (or matrix) assignment
1215 * - Whole structure assignment
1217 ir_variable
*whole_variable_written();
1220 * Set the LHS of an assignment
1222 void set_lhs(ir_rvalue
*lhs
);
1225 * Left-hand side of the assignment.
1227 * This should be treated as read only. If you need to set the LHS of an
1228 * assignment, use \c ir_assignment::set_lhs.
1230 ir_dereference
*lhs
;
1233 * Value being assigned
1238 * Optional condition for the assignment.
1240 ir_rvalue
*condition
;
1244 * Component mask written
1246 * For non-vector types in the LHS, this field will be zero. For vector
1247 * types, a bit will be set for each component that is written. Note that
1248 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1250 * A partially-set write mask means that each enabled channel gets
1251 * the value from a consecutive channel of the rhs. For example,
1252 * to write just .xyw of gl_FrontColor with color:
1254 * (assign (constant bool (1)) (xyw)
1255 * (var_ref gl_FragColor)
1256 * (swiz xyw (var_ref color)))
1258 unsigned write_mask
:4;
1261 /* Update ir_expression::get_num_operands() and operator_strs when
1262 * updating this list.
1264 enum ir_expression_operation
{
1273 ir_unop_exp
, /**< Log base e on gentype */
1274 ir_unop_log
, /**< Natural log on gentype */
1277 ir_unop_f2i
, /**< Float-to-integer conversion. */
1278 ir_unop_f2u
, /**< Float-to-unsigned conversion. */
1279 ir_unop_i2f
, /**< Integer-to-float conversion. */
1280 ir_unop_f2b
, /**< Float-to-boolean conversion */
1281 ir_unop_b2f
, /**< Boolean-to-float conversion */
1282 ir_unop_i2b
, /**< int-to-boolean conversion */
1283 ir_unop_b2i
, /**< Boolean-to-int conversion */
1284 ir_unop_u2f
, /**< Unsigned-to-float conversion. */
1285 ir_unop_i2u
, /**< Integer-to-unsigned conversion. */
1286 ir_unop_u2i
, /**< Unsigned-to-integer conversion. */
1287 ir_unop_d2f
, /**< Double-to-float conversion. */
1288 ir_unop_f2d
, /**< Float-to-double conversion. */
1289 ir_unop_d2i
, /**< Double-to-integer conversion. */
1290 ir_unop_i2d
, /**< Integer-to-double conversion. */
1291 ir_unop_d2u
, /**< Double-to-unsigned conversion. */
1292 ir_unop_u2d
, /**< Unsigned-to-double conversion. */
1293 ir_unop_d2b
, /**< Double-to-boolean conversion. */
1294 ir_unop_bitcast_i2f
, /**< Bit-identical int-to-float "conversion" */
1295 ir_unop_bitcast_f2i
, /**< Bit-identical float-to-int "conversion" */
1296 ir_unop_bitcast_u2f
, /**< Bit-identical uint-to-float "conversion" */
1297 ir_unop_bitcast_f2u
, /**< Bit-identical float-to-uint "conversion" */
1301 * \name Unary floating-point rounding operations.
1312 * \name Trigonometric operations.
1320 * \name Partial derivatives.
1324 ir_unop_dFdx_coarse
,
1327 ir_unop_dFdy_coarse
,
1332 * \name Floating point pack and unpack operations.
1335 ir_unop_pack_snorm_2x16
,
1336 ir_unop_pack_snorm_4x8
,
1337 ir_unop_pack_unorm_2x16
,
1338 ir_unop_pack_unorm_4x8
,
1339 ir_unop_pack_half_2x16
,
1340 ir_unop_unpack_snorm_2x16
,
1341 ir_unop_unpack_snorm_4x8
,
1342 ir_unop_unpack_unorm_2x16
,
1343 ir_unop_unpack_unorm_4x8
,
1344 ir_unop_unpack_half_2x16
,
1348 * \name Lowered floating point unpacking operations.
1350 * \see lower_packing_builtins_visitor::split_unpack_half_2x16
1353 ir_unop_unpack_half_2x16_split_x
,
1354 ir_unop_unpack_half_2x16_split_y
,
1358 * \name Bit operations, part of ARB_gpu_shader5.
1361 ir_unop_bitfield_reverse
,
1370 * \name Double packing, part of ARB_gpu_shader_fp64.
1373 ir_unop_pack_double_2x32
,
1374 ir_unop_unpack_double_2x32
,
1383 * Interpolate fs input at centroid
1385 * operand0 is the fs input.
1387 ir_unop_interpolate_at_centroid
,
1390 * A sentinel marking the last of the unary operations.
1392 ir_last_unop
= ir_unop_interpolate_at_centroid
,
1396 ir_binop_mul
, /**< Floating-point or low 32-bit integer multiply. */
1397 ir_binop_imul_high
, /**< Calculates the high 32-bits of a 64-bit multiply. */
1401 * Returns the carry resulting from the addition of the two arguments.
1408 * Returns the borrow resulting from the subtraction of the second argument
1409 * from the first argument.
1416 * Takes one of two combinations of arguments:
1419 * - mod(vecN, float)
1421 * Does not take integer types.
1426 * \name Binary comparison operators which return a boolean vector.
1427 * The type of both operands must be equal.
1437 * Returns single boolean for whether all components of operands[0]
1438 * equal the components of operands[1].
1442 * Returns single boolean for whether any component of operands[0]
1443 * is not equal to the corresponding component of operands[1].
1445 ir_binop_any_nequal
,
1449 * \name Bit-wise binary operations.
1470 * \name Lowered floating point packing operations.
1472 * \see lower_packing_builtins_visitor::split_pack_half_2x16
1475 ir_binop_pack_half_2x16_split
,
1479 * \name First half of a lowered bitfieldInsert() operation.
1481 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1488 * Load a value the size of a given GLSL type from a uniform block.
1490 * operand0 is the ir_constant uniform block index in the linked shader.
1491 * operand1 is a byte offset within the uniform block.
1496 * \name Multiplies a number by two to a power, part of ARB_gpu_shader5.
1503 * Extract a scalar from a vector
1505 * operand0 is the vector
1506 * operand1 is the index of the field to read from operand0
1508 ir_binop_vector_extract
,
1511 * Interpolate fs input at offset
1513 * operand0 is the fs input
1514 * operand1 is the offset from the pixel center
1516 ir_binop_interpolate_at_offset
,
1519 * Interpolate fs input at sample position
1521 * operand0 is the fs input
1522 * operand1 is the sample ID
1524 ir_binop_interpolate_at_sample
,
1527 * A sentinel marking the last of the binary operations.
1529 ir_last_binop
= ir_binop_interpolate_at_sample
,
1532 * \name Fused floating-point multiply-add, part of ARB_gpu_shader5.
1541 * \name Conditional Select
1543 * A vector conditional select instruction (like ?:, but operating per-
1544 * component on vectors).
1546 * \see lower_instructions_visitor::ldexp_to_arith
1553 * \name Second half of a lowered bitfieldInsert() operation.
1555 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1561 ir_triop_bitfield_extract
,
1564 * Generate a value with one field of a vector changed
1566 * operand0 is the vector
1567 * operand1 is the value to write into the vector result
1568 * operand2 is the index in operand0 to be modified
1570 ir_triop_vector_insert
,
1573 * A sentinel marking the last of the ternary operations.
1575 ir_last_triop
= ir_triop_vector_insert
,
1577 ir_quadop_bitfield_insert
,
1582 * A sentinel marking the last of the ternary operations.
1584 ir_last_quadop
= ir_quadop_vector
,
1587 * A sentinel marking the last of all operations.
1589 ir_last_opcode
= ir_quadop_vector
1592 class ir_expression
: public ir_rvalue
{
1594 ir_expression(int op
, const struct glsl_type
*type
,
1595 ir_rvalue
*op0
, ir_rvalue
*op1
= NULL
,
1596 ir_rvalue
*op2
= NULL
, ir_rvalue
*op3
= NULL
);
1599 * Constructor for unary operation expressions
1601 ir_expression(int op
, ir_rvalue
*);
1604 * Constructor for binary operation expressions
1606 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
);
1609 * Constructor for ternary operation expressions
1611 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
, ir_rvalue
*op2
);
1613 virtual bool equals(const ir_instruction
*ir
,
1614 enum ir_node_type ignore
= ir_type_unset
) const;
1616 virtual ir_expression
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1619 * Attempt to constant-fold the expression
1621 * The "variable_context" hash table links ir_variable * to ir_constant *
1622 * that represent the variables' values. \c NULL represents an empty
1625 * If the expression cannot be constant folded, this method will return
1628 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1631 * Determine the number of operands used by an expression
1633 static unsigned int get_num_operands(ir_expression_operation
);
1636 * Determine the number of operands used by an expression
1638 unsigned int get_num_operands() const
1640 return (this->operation
== ir_quadop_vector
)
1641 ? this->type
->vector_elements
: get_num_operands(operation
);
1645 * Return whether the expression operates on vectors horizontally.
1647 bool is_horizontal() const
1649 return operation
== ir_binop_all_equal
||
1650 operation
== ir_binop_any_nequal
||
1651 operation
== ir_unop_any
||
1652 operation
== ir_binop_dot
||
1653 operation
== ir_quadop_vector
;
1657 * Return a string representing this expression's operator.
1659 const char *operator_string();
1662 * Return a string representing this expression's operator.
1664 static const char *operator_string(ir_expression_operation
);
1668 * Do a reverse-lookup to translate the given string into an operator.
1670 static ir_expression_operation
get_operator(const char *);
1672 virtual void accept(ir_visitor
*v
)
1677 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1679 ir_expression_operation operation
;
1680 ir_rvalue
*operands
[4];
1685 * HIR instruction representing a high-level function call, containing a list
1686 * of parameters and returning a value in the supplied temporary.
1688 class ir_call
: public ir_instruction
{
1690 ir_call(ir_function_signature
*callee
,
1691 ir_dereference_variable
*return_deref
,
1692 exec_list
*actual_parameters
)
1693 : ir_instruction(ir_type_call
), return_deref(return_deref
), callee(callee
)
1695 assert(callee
->return_type
!= NULL
);
1696 actual_parameters
->move_nodes_to(& this->actual_parameters
);
1697 this->use_builtin
= callee
->is_builtin();
1700 virtual ir_call
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1702 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1704 virtual void accept(ir_visitor
*v
)
1709 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1712 * Get the name of the function being called.
1714 const char *callee_name() const
1716 return callee
->function_name();
1720 * Generates an inline version of the function before @ir,
1721 * storing the return value in return_deref.
1723 void generate_inline(ir_instruction
*ir
);
1726 * Storage for the function's return value.
1727 * This must be NULL if the return type is void.
1729 ir_dereference_variable
*return_deref
;
1732 * The specific function signature being called.
1734 ir_function_signature
*callee
;
1736 /* List of ir_rvalue of paramaters passed in this call. */
1737 exec_list actual_parameters
;
1739 /** Should this call only bind to a built-in function? */
1745 * \name Jump-like IR instructions.
1747 * These include \c break, \c continue, \c return, and \c discard.
1750 class ir_jump
: public ir_instruction
{
1752 ir_jump(enum ir_node_type t
)
1758 class ir_return
: public ir_jump
{
1761 : ir_jump(ir_type_return
), value(NULL
)
1765 ir_return(ir_rvalue
*value
)
1766 : ir_jump(ir_type_return
), value(value
)
1770 virtual ir_return
*clone(void *mem_ctx
, struct hash_table
*) const;
1772 ir_rvalue
*get_value() const
1777 virtual void accept(ir_visitor
*v
)
1782 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1789 * Jump instructions used inside loops
1791 * These include \c break and \c continue. The \c break within a loop is
1792 * different from the \c break within a switch-statement.
1794 * \sa ir_switch_jump
1796 class ir_loop_jump
: public ir_jump
{
1803 ir_loop_jump(jump_mode mode
)
1804 : ir_jump(ir_type_loop_jump
)
1809 virtual ir_loop_jump
*clone(void *mem_ctx
, struct hash_table
*) const;
1811 virtual void accept(ir_visitor
*v
)
1816 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1818 bool is_break() const
1820 return mode
== jump_break
;
1823 bool is_continue() const
1825 return mode
== jump_continue
;
1828 /** Mode selector for the jump instruction. */
1829 enum jump_mode mode
;
1833 * IR instruction representing discard statements.
1835 class ir_discard
: public ir_jump
{
1838 : ir_jump(ir_type_discard
)
1840 this->condition
= NULL
;
1843 ir_discard(ir_rvalue
*cond
)
1844 : ir_jump(ir_type_discard
)
1846 this->condition
= cond
;
1849 virtual ir_discard
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1851 virtual void accept(ir_visitor
*v
)
1856 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1858 ir_rvalue
*condition
;
1864 * Texture sampling opcodes used in ir_texture
1866 enum ir_texture_opcode
{
1867 ir_tex
, /**< Regular texture look-up */
1868 ir_txb
, /**< Texture look-up with LOD bias */
1869 ir_txl
, /**< Texture look-up with explicit LOD */
1870 ir_txd
, /**< Texture look-up with partial derivatvies */
1871 ir_txf
, /**< Texel fetch with explicit LOD */
1872 ir_txf_ms
, /**< Multisample texture fetch */
1873 ir_txs
, /**< Texture size */
1874 ir_lod
, /**< Texture lod query */
1875 ir_tg4
, /**< Texture gather */
1876 ir_query_levels
/**< Texture levels query */
1881 * IR instruction to sample a texture
1883 * The specific form of the IR instruction depends on the \c mode value
1884 * selected from \c ir_texture_opcodes. In the printed IR, these will
1887 * Texel offset (0 or an expression)
1888 * | Projection divisor
1889 * | | Shadow comparitor
1892 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1893 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1894 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1895 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1896 * (txf <type> <sampler> <coordinate> 0 <lod>)
1898 * <type> <sampler> <coordinate> <sample_index>)
1899 * (txs <type> <sampler> <lod>)
1900 * (lod <type> <sampler> <coordinate>)
1901 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
1902 * (query_levels <type> <sampler>)
1904 class ir_texture
: public ir_rvalue
{
1906 ir_texture(enum ir_texture_opcode op
)
1907 : ir_rvalue(ir_type_texture
),
1908 op(op
), sampler(NULL
), coordinate(NULL
), projector(NULL
),
1909 shadow_comparitor(NULL
), offset(NULL
)
1911 memset(&lod_info
, 0, sizeof(lod_info
));
1914 virtual ir_texture
*clone(void *mem_ctx
, struct hash_table
*) const;
1916 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1918 virtual void accept(ir_visitor
*v
)
1923 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1925 virtual bool equals(const ir_instruction
*ir
,
1926 enum ir_node_type ignore
= ir_type_unset
) const;
1929 * Return a string representing the ir_texture_opcode.
1931 const char *opcode_string();
1933 /** Set the sampler and type. */
1934 void set_sampler(ir_dereference
*sampler
, const glsl_type
*type
);
1937 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1939 static ir_texture_opcode
get_opcode(const char *);
1941 enum ir_texture_opcode op
;
1943 /** Sampler to use for the texture access. */
1944 ir_dereference
*sampler
;
1946 /** Texture coordinate to sample */
1947 ir_rvalue
*coordinate
;
1950 * Value used for projective divide.
1952 * If there is no projective divide (the common case), this will be
1953 * \c NULL. Optimization passes should check for this to point to a constant
1954 * of 1.0 and replace that with \c NULL.
1956 ir_rvalue
*projector
;
1959 * Coordinate used for comparison on shadow look-ups.
1961 * If there is no shadow comparison, this will be \c NULL. For the
1962 * \c ir_txf opcode, this *must* be \c NULL.
1964 ir_rvalue
*shadow_comparitor
;
1966 /** Texel offset. */
1970 ir_rvalue
*lod
; /**< Floating point LOD */
1971 ir_rvalue
*bias
; /**< Floating point LOD bias */
1972 ir_rvalue
*sample_index
; /**< MSAA sample index */
1973 ir_rvalue
*component
; /**< Gather component selector */
1975 ir_rvalue
*dPdx
; /**< Partial derivative of coordinate wrt X */
1976 ir_rvalue
*dPdy
; /**< Partial derivative of coordinate wrt Y */
1982 struct ir_swizzle_mask
{
1989 * Number of components in the swizzle.
1991 unsigned num_components
:3;
1994 * Does the swizzle contain duplicate components?
1996 * L-value swizzles cannot contain duplicate components.
1998 unsigned has_duplicates
:1;
2002 class ir_swizzle
: public ir_rvalue
{
2004 ir_swizzle(ir_rvalue
*, unsigned x
, unsigned y
, unsigned z
, unsigned w
,
2007 ir_swizzle(ir_rvalue
*val
, const unsigned *components
, unsigned count
);
2009 ir_swizzle(ir_rvalue
*val
, ir_swizzle_mask mask
);
2011 virtual ir_swizzle
*clone(void *mem_ctx
, struct hash_table
*) const;
2013 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2016 * Construct an ir_swizzle from the textual representation. Can fail.
2018 static ir_swizzle
*create(ir_rvalue
*, const char *, unsigned vector_length
);
2020 virtual void accept(ir_visitor
*v
)
2025 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2027 virtual bool equals(const ir_instruction
*ir
,
2028 enum ir_node_type ignore
= ir_type_unset
) const;
2030 bool is_lvalue() const
2032 return val
->is_lvalue() && !mask
.has_duplicates
;
2036 * Get the variable that is ultimately referenced by an r-value
2038 virtual ir_variable
*variable_referenced() const;
2041 ir_swizzle_mask mask
;
2045 * Initialize the mask component of a swizzle
2047 * This is used by the \c ir_swizzle constructors.
2049 void init_mask(const unsigned *components
, unsigned count
);
2053 class ir_dereference
: public ir_rvalue
{
2055 virtual ir_dereference
*clone(void *mem_ctx
, struct hash_table
*) const = 0;
2057 bool is_lvalue() const;
2060 * Get the variable that is ultimately referenced by an r-value
2062 virtual ir_variable
*variable_referenced() const = 0;
2065 ir_dereference(enum ir_node_type t
)
2072 class ir_dereference_variable
: public ir_dereference
{
2074 ir_dereference_variable(ir_variable
*var
);
2076 virtual ir_dereference_variable
*clone(void *mem_ctx
,
2077 struct hash_table
*) const;
2079 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2081 virtual bool equals(const ir_instruction
*ir
,
2082 enum ir_node_type ignore
= ir_type_unset
) const;
2085 * Get the variable that is ultimately referenced by an r-value
2087 virtual ir_variable
*variable_referenced() const
2092 virtual ir_variable
*whole_variable_referenced()
2094 /* ir_dereference_variable objects always dereference the entire
2095 * variable. However, if this dereference is dereferenced by anything
2096 * else, the complete deferefernce chain is not a whole-variable
2097 * dereference. This method should only be called on the top most
2098 * ir_rvalue in a dereference chain.
2103 virtual void accept(ir_visitor
*v
)
2108 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2111 * Object being dereferenced.
2117 class ir_dereference_array
: public ir_dereference
{
2119 ir_dereference_array(ir_rvalue
*value
, ir_rvalue
*array_index
);
2121 ir_dereference_array(ir_variable
*var
, ir_rvalue
*array_index
);
2123 virtual ir_dereference_array
*clone(void *mem_ctx
,
2124 struct hash_table
*) const;
2126 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2128 virtual bool equals(const ir_instruction
*ir
,
2129 enum ir_node_type ignore
= ir_type_unset
) const;
2132 * Get the variable that is ultimately referenced by an r-value
2134 virtual ir_variable
*variable_referenced() const
2136 return this->array
->variable_referenced();
2139 virtual void accept(ir_visitor
*v
)
2144 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2147 ir_rvalue
*array_index
;
2150 void set_array(ir_rvalue
*value
);
2154 class ir_dereference_record
: public ir_dereference
{
2156 ir_dereference_record(ir_rvalue
*value
, const char *field
);
2158 ir_dereference_record(ir_variable
*var
, const char *field
);
2160 virtual ir_dereference_record
*clone(void *mem_ctx
,
2161 struct hash_table
*) const;
2163 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2166 * Get the variable that is ultimately referenced by an r-value
2168 virtual ir_variable
*variable_referenced() const
2170 return this->record
->variable_referenced();
2173 virtual void accept(ir_visitor
*v
)
2178 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2186 * Data stored in an ir_constant
2188 union ir_constant_data
{
2197 class ir_constant
: public ir_rvalue
{
2199 ir_constant(const struct glsl_type
*type
, const ir_constant_data
*data
);
2200 ir_constant(bool b
, unsigned vector_elements
=1);
2201 ir_constant(unsigned int u
, unsigned vector_elements
=1);
2202 ir_constant(int i
, unsigned vector_elements
=1);
2203 ir_constant(float f
, unsigned vector_elements
=1);
2204 ir_constant(double d
, unsigned vector_elements
=1);
2207 * Construct an ir_constant from a list of ir_constant values
2209 ir_constant(const struct glsl_type
*type
, exec_list
*values
);
2212 * Construct an ir_constant from a scalar component of another ir_constant
2214 * The new \c ir_constant inherits the type of the component from the
2218 * In the case of a matrix constant, the new constant is a scalar, \b not
2221 ir_constant(const ir_constant
*c
, unsigned i
);
2224 * Return a new ir_constant of the specified type containing all zeros.
2226 static ir_constant
*zero(void *mem_ctx
, const glsl_type
*type
);
2228 virtual ir_constant
*clone(void *mem_ctx
, struct hash_table
*) const;
2230 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2232 virtual void accept(ir_visitor
*v
)
2237 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2239 virtual bool equals(const ir_instruction
*ir
,
2240 enum ir_node_type ignore
= ir_type_unset
) const;
2243 * Get a particular component of a constant as a specific type
2245 * This is useful, for example, to get a value from an integer constant
2246 * as a float or bool. This appears frequently when constructors are
2247 * called with all constant parameters.
2250 bool get_bool_component(unsigned i
) const;
2251 float get_float_component(unsigned i
) const;
2252 double get_double_component(unsigned i
) const;
2253 int get_int_component(unsigned i
) const;
2254 unsigned get_uint_component(unsigned i
) const;
2257 ir_constant
*get_array_element(unsigned i
) const;
2259 ir_constant
*get_record_field(const char *name
);
2262 * Copy the values on another constant at a given offset.
2264 * The offset is ignored for array or struct copies, it's only for
2265 * scalars or vectors into vectors or matrices.
2267 * With identical types on both sides and zero offset it's clone()
2268 * without creating a new object.
2271 void copy_offset(ir_constant
*src
, int offset
);
2274 * Copy the values on another constant at a given offset and
2275 * following an assign-like mask.
2277 * The mask is ignored for scalars.
2279 * Note that this function only handles what assign can handle,
2280 * i.e. at most a vector as source and a column of a matrix as
2284 void copy_masked_offset(ir_constant
*src
, int offset
, unsigned int mask
);
2287 * Determine whether a constant has the same value as another constant
2289 * \sa ir_constant::is_zero, ir_constant::is_one,
2290 * ir_constant::is_negative_one
2292 bool has_value(const ir_constant
*) const;
2295 * Return true if this ir_constant represents the given value.
2297 * For vectors, this checks that each component is the given value.
2299 virtual bool is_value(float f
, int i
) const;
2300 virtual bool is_zero() const;
2301 virtual bool is_one() const;
2302 virtual bool is_negative_one() const;
2305 * Return true for constants that could be stored as 16-bit unsigned values.
2307 * Note that this will return true even for signed integer ir_constants, as
2308 * long as the value is non-negative and fits in 16-bits.
2310 virtual bool is_uint16_constant() const;
2313 * Value of the constant.
2315 * The field used to back the values supplied by the constant is determined
2316 * by the type associated with the \c ir_instruction. Constants may be
2317 * scalars, vectors, or matrices.
2319 union ir_constant_data value
;
2321 /* Array elements */
2322 ir_constant
**array_elements
;
2324 /* Structure fields */
2325 exec_list components
;
2329 * Parameterless constructor only used by the clone method
2335 * IR instruction to emit a vertex in a geometry shader.
2337 class ir_emit_vertex
: public ir_instruction
{
2339 ir_emit_vertex(ir_rvalue
*stream
)
2340 : ir_instruction(ir_type_emit_vertex
),
2346 virtual void accept(ir_visitor
*v
)
2351 virtual ir_emit_vertex
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2353 return new(mem_ctx
) ir_emit_vertex(this->stream
->clone(mem_ctx
, ht
));
2356 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2358 int stream_id() const
2360 return stream
->as_constant()->value
.i
[0];
2367 * IR instruction to complete the current primitive and start a new one in a
2370 class ir_end_primitive
: public ir_instruction
{
2372 ir_end_primitive(ir_rvalue
*stream
)
2373 : ir_instruction(ir_type_end_primitive
),
2379 virtual void accept(ir_visitor
*v
)
2384 virtual ir_end_primitive
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2386 return new(mem_ctx
) ir_end_primitive(this->stream
->clone(mem_ctx
, ht
));
2389 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2391 int stream_id() const
2393 return stream
->as_constant()->value
.i
[0];
2402 * Apply a visitor to each IR node in a list
2405 visit_exec_list(exec_list
*list
, ir_visitor
*visitor
);
2408 * Validate invariants on each IR node in a list
2410 void validate_ir_tree(exec_list
*instructions
);
2412 struct _mesa_glsl_parse_state
;
2413 struct gl_shader_program
;
2416 * Detect whether an unlinked shader contains static recursion
2418 * If the list of instructions is determined to contain static recursion,
2419 * \c _mesa_glsl_error will be called to emit error messages for each function
2420 * that is in the recursion cycle.
2423 detect_recursion_unlinked(struct _mesa_glsl_parse_state
*state
,
2424 exec_list
*instructions
);
2427 * Detect whether a linked shader contains static recursion
2429 * If the list of instructions is determined to contain static recursion,
2430 * \c link_error_printf will be called to emit error messages for each function
2431 * that is in the recursion cycle. In addition,
2432 * \c gl_shader_program::LinkStatus will be set to false.
2435 detect_recursion_linked(struct gl_shader_program
*prog
,
2436 exec_list
*instructions
);
2439 * Make a clone of each IR instruction in a list
2441 * \param in List of IR instructions that are to be cloned
2442 * \param out List to hold the cloned instructions
2445 clone_ir_list(void *mem_ctx
, exec_list
*out
, const exec_list
*in
);
2448 _mesa_glsl_initialize_variables(exec_list
*instructions
,
2449 struct _mesa_glsl_parse_state
*state
);
2452 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state
*state
);
2455 _mesa_glsl_initialize_builtin_functions();
2457 extern ir_function_signature
*
2458 _mesa_glsl_find_builtin_function(_mesa_glsl_parse_state
*state
,
2459 const char *name
, exec_list
*actual_parameters
);
2461 extern ir_function
*
2462 _mesa_glsl_find_builtin_function_by_name(_mesa_glsl_parse_state
*state
,
2466 _mesa_glsl_get_builtin_function_shader(void);
2469 _mesa_glsl_release_functions(void);
2472 _mesa_glsl_release_builtin_functions(void);
2475 reparent_ir(exec_list
*list
, void *mem_ctx
);
2477 struct glsl_symbol_table
;
2480 import_prototypes(const exec_list
*source
, exec_list
*dest
,
2481 struct glsl_symbol_table
*symbols
, void *mem_ctx
);
2484 ir_has_call(ir_instruction
*ir
);
2487 do_set_program_inouts(exec_list
*instructions
, struct gl_program
*prog
,
2488 gl_shader_stage shader_stage
);
2491 prototype_string(const glsl_type
*return_type
, const char *name
,
2492 exec_list
*parameters
);
2495 mode_string(const ir_variable
*var
);
2498 * Built-in / reserved GL variables names start with "gl_"
2501 is_gl_identifier(const char *s
)
2503 return s
&& s
[0] == 'g' && s
[1] == 'l' && s
[2] == '_';
2507 #endif /* __cplusplus */
2509 extern void _mesa_print_ir(FILE *f
, struct exec_list
*instructions
,
2510 struct _mesa_glsl_parse_state
*state
);
2513 fprint_ir(FILE *f
, const void *instruction
);
2520 vertices_per_prim(GLenum prim
);