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
,
82 ir_type_max
, /**< maximum ir_type enum number, for validation */
83 ir_type_unset
= ir_type_max
88 * Base class of all IR instructions
90 class ir_instruction
: public exec_node
{
92 enum ir_node_type ir_type
;
95 * GCC 4.7+ and clang warn when deleting an ir_instruction unless
96 * there's a virtual destructor present. Because we almost
97 * universally use ralloc for our memory management of
98 * ir_instructions, the destructor doesn't need to do any work.
100 virtual ~ir_instruction()
104 /** ir_print_visitor helper for debugging. */
105 void print(void) const;
106 void fprint(FILE *f
) const;
108 virtual void accept(ir_visitor
*) = 0;
109 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*) = 0;
110 virtual ir_instruction
*clone(void *mem_ctx
,
111 struct hash_table
*ht
) const = 0;
113 bool is_rvalue() const
115 return ir_type
== ir_type_dereference_array
||
116 ir_type
== ir_type_dereference_record
||
117 ir_type
== ir_type_dereference_variable
||
118 ir_type
== ir_type_constant
||
119 ir_type
== ir_type_expression
||
120 ir_type
== ir_type_swizzle
||
121 ir_type
== ir_type_texture
;
124 bool is_dereference() const
126 return ir_type
== ir_type_dereference_array
||
127 ir_type
== ir_type_dereference_record
||
128 ir_type
== ir_type_dereference_variable
;
133 return ir_type
== ir_type_loop_jump
||
134 ir_type
== ir_type_return
||
135 ir_type
== ir_type_discard
;
139 * \name IR instruction downcast functions
141 * These functions either cast the object to a derived class or return
142 * \c NULL if the object's type does not match the specified derived class.
143 * Additional downcast functions will be added as needed.
146 #define AS_BASE(TYPE) \
147 class ir_##TYPE *as_##TYPE() \
149 assume(this != NULL); \
150 return is_##TYPE() ? (ir_##TYPE *) this : NULL; \
152 const class ir_##TYPE *as_##TYPE() const \
154 assume(this != NULL); \
155 return is_##TYPE() ? (ir_##TYPE *) this : NULL; \
163 #define AS_CHILD(TYPE) \
164 class ir_##TYPE * as_##TYPE() \
166 assume(this != NULL); \
167 return ir_type == ir_type_##TYPE ? (ir_##TYPE *) this : NULL; \
169 const class ir_##TYPE * as_##TYPE() const \
171 assume(this != NULL); \
172 return ir_type == ir_type_##TYPE ? (const ir_##TYPE *) this : NULL; \
176 AS_CHILD(dereference_array
)
177 AS_CHILD(dereference_variable
)
178 AS_CHILD(dereference_record
)
193 * IR equality method: Return true if the referenced instruction would
194 * return the same value as this one.
196 * This intended to be used for CSE and algebraic optimizations, on rvalues
197 * in particular. No support for other instruction types (assignments,
198 * jumps, calls, etc.) is planned.
200 virtual bool equals(const ir_instruction
*ir
,
201 enum ir_node_type ignore
= ir_type_unset
) const;
204 ir_instruction(enum ir_node_type t
)
212 assert(!"Should not get here.");
218 * The base class for all "values"/expression trees.
220 class ir_rvalue
: public ir_instruction
{
222 const struct glsl_type
*type
;
224 virtual ir_rvalue
*clone(void *mem_ctx
, struct hash_table
*) const;
226 virtual void accept(ir_visitor
*v
)
231 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
233 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
235 ir_rvalue
*as_rvalue_to_saturate();
237 virtual bool is_lvalue() const
243 * Get the variable that is ultimately referenced by an r-value
245 virtual ir_variable
*variable_referenced() const
252 * If an r-value is a reference to a whole variable, get that variable
255 * Pointer to a variable that is completely dereferenced by the r-value. If
256 * the r-value is not a dereference or the dereference does not access the
257 * entire variable (i.e., it's just one array element, struct field), \c NULL
260 virtual ir_variable
*whole_variable_referenced()
266 * Determine if an r-value has the value zero
268 * The base implementation of this function always returns \c false. The
269 * \c ir_constant class over-rides this function to return \c true \b only
270 * for vector and scalar types that have all elements set to the value
271 * zero (or \c false for booleans).
273 * \sa ir_constant::has_value, ir_rvalue::is_one, ir_rvalue::is_negative_one
275 virtual bool is_zero() const;
278 * Determine if an r-value has the value one
280 * The base implementation of this function always returns \c false. The
281 * \c ir_constant class over-rides this function to return \c true \b only
282 * for vector and scalar types that have all elements set to the value
283 * one (or \c true for booleans).
285 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_negative_one
287 virtual bool is_one() const;
290 * Determine if an r-value has the value negative one
292 * The base implementation of this function always returns \c false. The
293 * \c ir_constant class over-rides this function to return \c true \b only
294 * for vector and scalar types that have all elements set to the value
295 * negative one. For boolean types, the result is always \c false.
297 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one
299 virtual bool is_negative_one() const;
302 * Determine if an r-value is an unsigned integer constant which can be
305 * \sa ir_constant::is_uint16_constant.
307 virtual bool is_uint16_constant() const { return false; }
310 * Return a generic value of error_type.
312 * Allocation will be performed with 'mem_ctx' as ralloc owner.
314 static ir_rvalue
*error_value(void *mem_ctx
);
317 ir_rvalue(enum ir_node_type t
);
322 * Variable storage classes
324 enum ir_variable_mode
{
325 ir_var_auto
= 0, /**< Function local variables and globals. */
326 ir_var_uniform
, /**< Variable declared as a uniform. */
327 ir_var_shader_storage
, /**< Variable declared as an ssbo. */
332 ir_var_function_inout
,
333 ir_var_const_in
, /**< "in" param that must be a constant expression */
334 ir_var_system_value
, /**< Ex: front-face, instance-id, etc. */
335 ir_var_temporary
, /**< Temporary variable generated during compilation. */
336 ir_var_mode_count
/**< Number of variable modes */
340 * Enum keeping track of how a variable was declared. For error checking of
341 * the gl_PerVertex redeclaration rules.
343 enum ir_var_declaration_type
{
345 * Normal declaration (for most variables, this means an explicit
346 * declaration. Exception: temporaries are always implicitly declared, but
347 * they still use ir_var_declared_normally).
349 * Note: an ir_variable that represents a named interface block uses
350 * ir_var_declared_normally.
352 ir_var_declared_normally
= 0,
355 * Variable was explicitly declared (or re-declared) in an unnamed
358 ir_var_declared_in_block
,
361 * Variable is an implicitly declared built-in that has not been explicitly
362 * re-declared by the shader.
364 ir_var_declared_implicitly
,
367 * Variable is implicitly generated by the compiler and should not be
368 * visible via the API.
374 * \brief Layout qualifiers for gl_FragDepth.
376 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
377 * with a layout qualifier.
379 enum ir_depth_layout
{
380 ir_depth_layout_none
, /**< No depth layout is specified. */
382 ir_depth_layout_greater
,
383 ir_depth_layout_less
,
384 ir_depth_layout_unchanged
388 * \brief Convert depth layout qualifier to string.
391 depth_layout_string(ir_depth_layout layout
);
394 * Description of built-in state associated with a uniform
396 * \sa ir_variable::state_slots
398 struct ir_state_slot
{
405 * Get the string value for an interpolation qualifier
407 * \return The string that would be used in a shader to specify \c
408 * mode will be returned.
410 * This function is used to generate error messages of the form "shader
411 * uses %s interpolation qualifier", so in the case where there is no
412 * interpolation qualifier, it returns "no".
414 * This function should only be used on a shader input or output variable.
416 const char *interpolation_string(unsigned interpolation
);
419 class ir_variable
: public ir_instruction
{
421 ir_variable(const struct glsl_type
*, const char *, ir_variable_mode
);
423 virtual ir_variable
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
425 virtual void accept(ir_visitor
*v
)
430 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
434 * Determine how this variable should be interpolated based on its
435 * interpolation qualifier (if present), whether it is gl_Color or
436 * gl_SecondaryColor, and whether flatshading is enabled in the current GL
439 * The return value will always be either INTERP_QUALIFIER_SMOOTH,
440 * INTERP_QUALIFIER_NOPERSPECTIVE, or INTERP_QUALIFIER_FLAT.
442 glsl_interp_qualifier
determine_interpolation_mode(bool flat_shade
);
445 * Determine whether or not a variable is part of a uniform or
446 * shader storage block.
448 inline bool is_in_buffer_block() const
450 return (this->data
.mode
== ir_var_uniform
||
451 this->data
.mode
== ir_var_shader_storage
) &&
452 this->interface_type
!= NULL
;
456 * Determine whether or not a variable is the declaration of an interface
459 * For the first declaration below, there will be an \c ir_variable named
460 * "instance" whose type and whose instance_type will be the same
461 * \cglsl_type. For the second declaration, there will be an \c ir_variable
462 * named "f" whose type is float and whose instance_type is B2.
464 * "instance" is an interface instance variable, but "f" is not.
474 inline bool is_interface_instance() const
476 return this->type
->without_array() == this->interface_type
;
480 * Set this->interface_type on a newly created variable.
482 void init_interface_type(const struct glsl_type
*type
)
484 assert(this->interface_type
== NULL
);
485 this->interface_type
= type
;
486 if (this->is_interface_instance()) {
487 this->u
.max_ifc_array_access
=
488 rzalloc_array(this, unsigned, type
->length
);
493 * Change this->interface_type on a variable that previously had a
494 * different, but compatible, interface_type. This is used during linking
495 * to set the size of arrays in interface blocks.
497 void change_interface_type(const struct glsl_type
*type
)
499 if (this->u
.max_ifc_array_access
!= NULL
) {
500 /* max_ifc_array_access has already been allocated, so make sure the
501 * new interface has the same number of fields as the old one.
503 assert(this->interface_type
->length
== type
->length
);
505 this->interface_type
= type
;
509 * Change this->interface_type on a variable that previously had a
510 * different, and incompatible, interface_type. This is used during
511 * compilation to handle redeclaration of the built-in gl_PerVertex
514 void reinit_interface_type(const struct glsl_type
*type
)
516 if (this->u
.max_ifc_array_access
!= NULL
) {
518 /* Redeclaring gl_PerVertex is only allowed if none of the built-ins
519 * it defines have been accessed yet; so it's safe to throw away the
520 * old max_ifc_array_access pointer, since all of its values are
523 for (unsigned i
= 0; i
< this->interface_type
->length
; i
++)
524 assert(this->u
.max_ifc_array_access
[i
] == 0);
526 ralloc_free(this->u
.max_ifc_array_access
);
527 this->u
.max_ifc_array_access
= NULL
;
529 this->interface_type
= NULL
;
530 init_interface_type(type
);
533 const glsl_type
*get_interface_type() const
535 return this->interface_type
;
539 * Get the max_ifc_array_access pointer
541 * A "set" function is not needed because the array is dynmically allocated
544 inline unsigned *get_max_ifc_array_access()
546 assert(this->data
._num_state_slots
== 0);
547 return this->u
.max_ifc_array_access
;
550 inline unsigned get_num_state_slots() const
552 assert(!this->is_interface_instance()
553 || this->data
._num_state_slots
== 0);
554 return this->data
._num_state_slots
;
557 inline void set_num_state_slots(unsigned n
)
559 assert(!this->is_interface_instance()
561 this->data
._num_state_slots
= n
;
564 inline ir_state_slot
*get_state_slots()
566 return this->is_interface_instance() ? NULL
: this->u
.state_slots
;
569 inline const ir_state_slot
*get_state_slots() const
571 return this->is_interface_instance() ? NULL
: this->u
.state_slots
;
574 inline ir_state_slot
*allocate_state_slots(unsigned n
)
576 assert(!this->is_interface_instance());
578 this->u
.state_slots
= ralloc_array(this, ir_state_slot
, n
);
579 this->data
._num_state_slots
= 0;
581 if (this->u
.state_slots
!= NULL
)
582 this->data
._num_state_slots
= n
;
584 return this->u
.state_slots
;
587 inline bool is_name_ralloced() const
589 return this->name
!= ir_variable::tmp_name
;
593 * Enable emitting extension warnings for this variable
595 void enable_extension_warning(const char *extension
);
598 * Get the extension warning string for this variable
600 * If warnings are not enabled, \c NULL is returned.
602 const char *get_extension_warning() const;
605 * Declared type of the variable
607 const struct glsl_type
*type
;
610 * Declared name of the variable
614 struct ir_variable_data
{
617 * Is the variable read-only?
619 * This is set for variables declared as \c const, shader inputs,
622 unsigned read_only
:1;
626 unsigned invariant
:1;
630 * Has this variable been used for reading or writing?
632 * Several GLSL semantic checks require knowledge of whether or not a
633 * variable has been used. For example, it is an error to redeclare a
634 * variable as invariant after it has been used.
636 * This is only maintained in the ast_to_hir.cpp path, not in
637 * Mesa's fixed function or ARB program paths.
642 * Has this variable been statically assigned?
644 * This answers whether the variable was assigned in any path of
645 * the shader during ast_to_hir. This doesn't answer whether it is
646 * still written after dead code removal, nor is it maintained in
647 * non-ast_to_hir.cpp (GLSL parsing) paths.
652 * Enum indicating how the variable was declared. See
653 * ir_var_declaration_type.
655 * This is used to detect certain kinds of illegal variable redeclarations.
657 unsigned how_declared
:2;
660 * Storage class of the variable.
662 * \sa ir_variable_mode
667 * Interpolation mode for shader inputs / outputs
669 * \sa ir_variable_interpolation
671 unsigned interpolation
:2;
674 * \name ARB_fragment_coord_conventions
677 unsigned origin_upper_left
:1;
678 unsigned pixel_center_integer
:1;
682 * Was the location explicitly set in the shader?
684 * If the location is explicitly set in the shader, it \b cannot be changed
685 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
688 unsigned explicit_location
:1;
689 unsigned explicit_index
:1;
692 * Was an initial binding explicitly set in the shader?
694 * If so, constant_value contains an integer ir_constant representing the
695 * initial binding point.
697 unsigned explicit_binding
:1;
700 * Does this variable have an initializer?
702 * This is used by the linker to cross-validiate initializers of global
705 unsigned has_initializer
:1;
708 * Is this variable a generic output or input that has not yet been matched
709 * up to a variable in another stage of the pipeline?
711 * This is used by the linker as scratch storage while assigning locations
712 * to generic inputs and outputs.
714 unsigned is_unmatched_generic_inout
:1;
717 * If non-zero, then this variable may be packed along with other variables
718 * into a single varying slot, so this offset should be applied when
719 * accessing components. For example, an offset of 1 means that the x
720 * component of this variable is actually stored in component y of the
721 * location specified by \c location.
723 unsigned location_frac
:2;
726 * Layout of the matrix. Uses glsl_matrix_layout values.
728 unsigned matrix_layout
:2;
731 * Non-zero if this variable was created by lowering a named interface
732 * block which was not an array.
734 * Note that this variable and \c from_named_ifc_block_array will never
737 unsigned from_named_ifc_block_nonarray
:1;
740 * Non-zero if this variable was created by lowering a named interface
741 * block which was an array.
743 * Note that this variable and \c from_named_ifc_block_nonarray will never
746 unsigned from_named_ifc_block_array
:1;
749 * Non-zero if the variable must be a shader input. This is useful for
750 * constraints on function parameters.
752 unsigned must_be_shader_input
:1;
755 * Output index for dual source blending.
758 * The GLSL spec only allows the values 0 or 1 for the index in \b dual
764 * \brief Layout qualifier for gl_FragDepth.
766 * This is not equal to \c ir_depth_layout_none if and only if this
767 * variable is \c gl_FragDepth and a layout qualifier is specified.
769 ir_depth_layout depth_layout
:3;
772 * ARB_shader_image_load_store qualifiers.
774 unsigned image_read_only
:1; /**< "readonly" qualifier. */
775 unsigned image_write_only
:1; /**< "writeonly" qualifier. */
776 unsigned image_coherent
:1;
777 unsigned image_volatile
:1;
778 unsigned image_restrict
:1;
781 * Emit a warning if this variable is accessed.
784 uint8_t warn_extension_index
;
787 /** Image internal format if specified explicitly, otherwise GL_NONE. */
788 uint16_t image_format
;
792 * Number of state slots used
795 * This could be stored in as few as 7-bits, if necessary. If it is made
796 * smaller, add an assertion to \c ir_variable::allocate_state_slots to
799 uint16_t _num_state_slots
;
803 * Initial binding point for a sampler, atomic, or UBO.
805 * For array types, this represents the binding point for the first element.
810 * Storage location of the base of this variable
812 * The precise meaning of this field depends on the nature of the variable.
814 * - Vertex shader input: one of the values from \c gl_vert_attrib.
815 * - Vertex shader output: one of the values from \c gl_varying_slot.
816 * - Geometry shader input: one of the values from \c gl_varying_slot.
817 * - Geometry shader output: one of the values from \c gl_varying_slot.
818 * - Fragment shader input: one of the values from \c gl_varying_slot.
819 * - Fragment shader output: one of the values from \c gl_frag_result.
820 * - Uniforms: Per-stage uniform slot number for default uniform block.
821 * - Uniforms: Index within the uniform block definition for UBO members.
822 * - Other: This field is not currently used.
824 * If the variable is a uniform, shader input, or shader output, and the
825 * slot has not been assigned, the value will be -1.
830 * Vertex stream output identifier.
835 * Location an atomic counter is stored at.
842 * Highest element accessed with a constant expression array index
844 * Not used for non-array variables.
846 unsigned max_array_access
;
849 * Allow (only) ir_variable direct access private members.
851 friend class ir_variable
;
855 * Value assigned in the initializer of a variable declared "const"
857 ir_constant
*constant_value
;
860 * Constant expression assigned in the initializer of the variable
863 * This field and \c ::constant_value are distinct. Even if the two fields
864 * refer to constants with the same value, they must point to separate
867 ir_constant
*constant_initializer
;
870 static const char *const warn_extension_table
[];
874 * For variables which satisfy the is_interface_instance() predicate,
875 * this points to an array of integers such that if the ith member of
876 * the interface block is an array, max_ifc_array_access[i] is the
877 * maximum array element of that member that has been accessed. If the
878 * ith member of the interface block is not an array,
879 * max_ifc_array_access[i] is unused.
881 * For variables whose type is not an interface block, this pointer is
884 unsigned *max_ifc_array_access
;
887 * Built-in state that backs this uniform
889 * Once set at variable creation, \c state_slots must remain invariant.
891 * If the variable is not a uniform, \c _num_state_slots will be zero
892 * and \c state_slots will be \c NULL.
894 ir_state_slot
*state_slots
;
898 * For variables that are in an interface block or are an instance of an
899 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
901 * \sa ir_variable::location
903 const glsl_type
*interface_type
;
906 * Name used for anonymous compiler temporaries
908 static const char tmp_name
[];
912 * Should the construct keep names for ir_var_temporary variables?
914 * When this global is false, names passed to the constructor for
915 * \c ir_var_temporary variables will be dropped. Instead, the variable will
916 * be named "compiler_temp". This name will be in static storage.
919 * \b NEVER change the mode of an \c ir_var_temporary.
922 * This variable is \b not thread-safe. It is global, \b not
923 * per-context. It begins life false. A context can, at some point, make
924 * it true. From that point on, it will be true forever. This should be
925 * okay since it will only be set true while debugging.
927 static bool temporaries_allocate_names
;
931 * A function that returns whether a built-in function is available in the
932 * current shading language (based on version, ES or desktop, and extensions).
934 typedef bool (*builtin_available_predicate
)(const _mesa_glsl_parse_state
*);
938 * The representation of a function instance; may be the full definition or
939 * simply a prototype.
941 class ir_function_signature
: public ir_instruction
{
942 /* An ir_function_signature will be part of the list of signatures in
946 ir_function_signature(const glsl_type
*return_type
,
947 builtin_available_predicate builtin_avail
= NULL
);
949 virtual ir_function_signature
*clone(void *mem_ctx
,
950 struct hash_table
*ht
) const;
951 ir_function_signature
*clone_prototype(void *mem_ctx
,
952 struct hash_table
*ht
) const;
954 virtual void accept(ir_visitor
*v
)
959 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
962 * Attempt to evaluate this function as a constant expression,
963 * given a list of the actual parameters and the variable context.
964 * Returns NULL for non-built-ins.
966 ir_constant
*constant_expression_value(exec_list
*actual_parameters
, struct hash_table
*variable_context
);
969 * Get the name of the function for which this is a signature
971 const char *function_name() const;
974 * Get a handle to the function for which this is a signature
976 * There is no setter function, this function returns a \c const pointer,
977 * and \c ir_function_signature::_function is private for a reason. The
978 * only way to make a connection between a function and function signature
979 * is via \c ir_function::add_signature. This helps ensure that certain
980 * invariants (i.e., a function signature is in the list of signatures for
981 * its \c _function) are met.
983 * \sa ir_function::add_signature
985 inline const class ir_function
*function() const
987 return this->_function
;
991 * Check whether the qualifiers match between this signature's parameters
992 * and the supplied parameter list. If not, returns the name of the first
993 * parameter with mismatched qualifiers (for use in error messages).
995 const char *qualifiers_match(exec_list
*params
);
998 * Replace the current parameter list with the given one. This is useful
999 * if the current information came from a prototype, and either has invalid
1000 * or missing parameter names.
1002 void replace_parameters(exec_list
*new_params
);
1005 * Function return type.
1007 * \note This discards the optional precision qualifier.
1009 const struct glsl_type
*return_type
;
1012 * List of ir_variable of function parameters.
1014 * This represents the storage. The paramaters passed in a particular
1015 * call will be in ir_call::actual_paramaters.
1017 struct exec_list parameters
;
1019 /** Whether or not this function has a body (which may be empty). */
1020 unsigned is_defined
:1;
1022 /** Whether or not this function signature is a built-in. */
1023 bool is_builtin() const;
1026 * Whether or not this function is an intrinsic to be implemented
1031 /** Whether or not a built-in is available for this shader. */
1032 bool is_builtin_available(const _mesa_glsl_parse_state
*state
) const;
1034 /** Body of instructions in the function. */
1035 struct exec_list body
;
1039 * A function pointer to a predicate that answers whether a built-in
1040 * function is available in the current shader. NULL if not a built-in.
1042 builtin_available_predicate builtin_avail
;
1044 /** Function of which this signature is one overload. */
1045 class ir_function
*_function
;
1047 /** Function signature of which this one is a prototype clone */
1048 const ir_function_signature
*origin
;
1050 friend class ir_function
;
1053 * Helper function to run a list of instructions for constant
1054 * expression evaluation.
1056 * The hash table represents the values of the visible variables.
1057 * There are no scoping issues because the table is indexed on
1058 * ir_variable pointers, not variable names.
1060 * Returns false if the expression is not constant, true otherwise,
1061 * and the value in *result if result is non-NULL.
1063 bool constant_expression_evaluate_expression_list(const struct exec_list
&body
,
1064 struct hash_table
*variable_context
,
1065 ir_constant
**result
);
1070 * Header for tracking multiple overloaded functions with the same name.
1071 * Contains a list of ir_function_signatures representing each of the
1074 class ir_function
: public ir_instruction
{
1076 ir_function(const char *name
);
1078 virtual ir_function
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1080 virtual void accept(ir_visitor
*v
)
1085 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1087 void add_signature(ir_function_signature
*sig
)
1089 sig
->_function
= this;
1090 this->signatures
.push_tail(sig
);
1094 * Find a signature that matches a set of actual parameters, taking implicit
1095 * conversions into account. Also flags whether the match was exact.
1097 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1098 const exec_list
*actual_param
,
1099 bool allow_builtins
,
1100 bool *match_is_exact
);
1103 * Find a signature that matches a set of actual parameters, taking implicit
1104 * conversions into account.
1106 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1107 const exec_list
*actual_param
,
1108 bool allow_builtins
);
1111 * Find a signature that exactly matches a set of actual parameters without
1112 * any implicit type conversions.
1114 ir_function_signature
*exact_matching_signature(_mesa_glsl_parse_state
*state
,
1115 const exec_list
*actual_ps
);
1118 * Name of the function.
1122 /** Whether or not this function has a signature that isn't a built-in. */
1123 bool has_user_signature();
1126 * List of ir_function_signature for each overloaded function with this name.
1128 struct exec_list signatures
;
1131 * is this function a subroutine type declaration
1132 * e.g. subroutine void type1(float arg1);
1137 * is this function associated to a subroutine type
1138 * e.g. subroutine (type1, type2) function_name { function_body };
1139 * would have num_subroutine_types 2,
1140 * and pointers to the type1 and type2 types.
1142 int num_subroutine_types
;
1143 const struct glsl_type
**subroutine_types
;
1146 inline const char *ir_function_signature::function_name() const
1148 return this->_function
->name
;
1154 * IR instruction representing high-level if-statements
1156 class ir_if
: public ir_instruction
{
1158 ir_if(ir_rvalue
*condition
)
1159 : ir_instruction(ir_type_if
), condition(condition
)
1163 virtual ir_if
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1165 virtual void accept(ir_visitor
*v
)
1170 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1172 ir_rvalue
*condition
;
1173 /** List of ir_instruction for the body of the then branch */
1174 exec_list then_instructions
;
1175 /** List of ir_instruction for the body of the else branch */
1176 exec_list else_instructions
;
1181 * IR instruction representing a high-level loop structure.
1183 class ir_loop
: public ir_instruction
{
1187 virtual ir_loop
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1189 virtual void accept(ir_visitor
*v
)
1194 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1196 /** List of ir_instruction that make up the body of the loop. */
1197 exec_list body_instructions
;
1201 class ir_assignment
: public ir_instruction
{
1203 ir_assignment(ir_rvalue
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
= NULL
);
1206 * Construct an assignment with an explicit write mask
1209 * Since a write mask is supplied, the LHS must already be a bare
1210 * \c ir_dereference. The cannot be any swizzles in the LHS.
1212 ir_assignment(ir_dereference
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
,
1213 unsigned write_mask
);
1215 virtual ir_assignment
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1217 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1219 virtual void accept(ir_visitor
*v
)
1224 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1227 * Get a whole variable written by an assignment
1229 * If the LHS of the assignment writes a whole variable, the variable is
1230 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1233 * - Assigning to a scalar
1234 * - Assigning to all components of a vector
1235 * - Whole array (or matrix) assignment
1236 * - Whole structure assignment
1238 ir_variable
*whole_variable_written();
1241 * Set the LHS of an assignment
1243 void set_lhs(ir_rvalue
*lhs
);
1246 * Left-hand side of the assignment.
1248 * This should be treated as read only. If you need to set the LHS of an
1249 * assignment, use \c ir_assignment::set_lhs.
1251 ir_dereference
*lhs
;
1254 * Value being assigned
1259 * Optional condition for the assignment.
1261 ir_rvalue
*condition
;
1265 * Component mask written
1267 * For non-vector types in the LHS, this field will be zero. For vector
1268 * types, a bit will be set for each component that is written. Note that
1269 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1271 * A partially-set write mask means that each enabled channel gets
1272 * the value from a consecutive channel of the rhs. For example,
1273 * to write just .xyw of gl_FrontColor with color:
1275 * (assign (constant bool (1)) (xyw)
1276 * (var_ref gl_FragColor)
1277 * (swiz xyw (var_ref color)))
1279 unsigned write_mask
:4;
1282 /* Update ir_expression::get_num_operands() and operator_strs when
1283 * updating this list.
1285 enum ir_expression_operation
{
1294 ir_unop_exp
, /**< Log base e on gentype */
1295 ir_unop_log
, /**< Natural log on gentype */
1298 ir_unop_f2i
, /**< Float-to-integer conversion. */
1299 ir_unop_f2u
, /**< Float-to-unsigned conversion. */
1300 ir_unop_i2f
, /**< Integer-to-float conversion. */
1301 ir_unop_f2b
, /**< Float-to-boolean conversion */
1302 ir_unop_b2f
, /**< Boolean-to-float conversion */
1303 ir_unop_i2b
, /**< int-to-boolean conversion */
1304 ir_unop_b2i
, /**< Boolean-to-int conversion */
1305 ir_unop_u2f
, /**< Unsigned-to-float conversion. */
1306 ir_unop_i2u
, /**< Integer-to-unsigned conversion. */
1307 ir_unop_u2i
, /**< Unsigned-to-integer conversion. */
1308 ir_unop_d2f
, /**< Double-to-float conversion. */
1309 ir_unop_f2d
, /**< Float-to-double conversion. */
1310 ir_unop_d2i
, /**< Double-to-integer conversion. */
1311 ir_unop_i2d
, /**< Integer-to-double conversion. */
1312 ir_unop_d2u
, /**< Double-to-unsigned conversion. */
1313 ir_unop_u2d
, /**< Unsigned-to-double conversion. */
1314 ir_unop_d2b
, /**< Double-to-boolean conversion. */
1315 ir_unop_bitcast_i2f
, /**< Bit-identical int-to-float "conversion" */
1316 ir_unop_bitcast_f2i
, /**< Bit-identical float-to-int "conversion" */
1317 ir_unop_bitcast_u2f
, /**< Bit-identical uint-to-float "conversion" */
1318 ir_unop_bitcast_f2u
, /**< Bit-identical float-to-uint "conversion" */
1322 * \name Unary floating-point rounding operations.
1333 * \name Trigonometric operations.
1341 * \name Partial derivatives.
1345 ir_unop_dFdx_coarse
,
1348 ir_unop_dFdy_coarse
,
1353 * \name Floating point pack and unpack operations.
1356 ir_unop_pack_snorm_2x16
,
1357 ir_unop_pack_snorm_4x8
,
1358 ir_unop_pack_unorm_2x16
,
1359 ir_unop_pack_unorm_4x8
,
1360 ir_unop_pack_half_2x16
,
1361 ir_unop_unpack_snorm_2x16
,
1362 ir_unop_unpack_snorm_4x8
,
1363 ir_unop_unpack_unorm_2x16
,
1364 ir_unop_unpack_unorm_4x8
,
1365 ir_unop_unpack_half_2x16
,
1369 * \name Lowered floating point unpacking operations.
1371 * \see lower_packing_builtins_visitor::split_unpack_half_2x16
1374 ir_unop_unpack_half_2x16_split_x
,
1375 ir_unop_unpack_half_2x16_split_y
,
1379 * \name Bit operations, part of ARB_gpu_shader5.
1382 ir_unop_bitfield_reverse
,
1391 * \name Double packing, part of ARB_gpu_shader_fp64.
1394 ir_unop_pack_double_2x32
,
1395 ir_unop_unpack_double_2x32
,
1403 ir_unop_subroutine_to_int
,
1405 * Interpolate fs input at centroid
1407 * operand0 is the fs input.
1409 ir_unop_interpolate_at_centroid
,
1412 * A sentinel marking the last of the unary operations.
1414 ir_last_unop
= ir_unop_interpolate_at_centroid
,
1418 ir_binop_mul
, /**< Floating-point or low 32-bit integer multiply. */
1419 ir_binop_imul_high
, /**< Calculates the high 32-bits of a 64-bit multiply. */
1423 * Returns the carry resulting from the addition of the two arguments.
1430 * Returns the borrow resulting from the subtraction of the second argument
1431 * from the first argument.
1438 * Takes one of two combinations of arguments:
1441 * - mod(vecN, float)
1443 * Does not take integer types.
1448 * \name Binary comparison operators which return a boolean vector.
1449 * The type of both operands must be equal.
1459 * Returns single boolean for whether all components of operands[0]
1460 * equal the components of operands[1].
1464 * Returns single boolean for whether any component of operands[0]
1465 * is not equal to the corresponding component of operands[1].
1467 ir_binop_any_nequal
,
1471 * \name Bit-wise binary operations.
1492 * \name Lowered floating point packing operations.
1494 * \see lower_packing_builtins_visitor::split_pack_half_2x16
1497 ir_binop_pack_half_2x16_split
,
1501 * \name First half of a lowered bitfieldInsert() operation.
1503 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1510 * Load a value the size of a given GLSL type from a uniform block.
1512 * operand0 is the ir_constant uniform block index in the linked shader.
1513 * operand1 is a byte offset within the uniform block.
1518 * \name Multiplies a number by two to a power, part of ARB_gpu_shader5.
1525 * Extract a scalar from a vector
1527 * operand0 is the vector
1528 * operand1 is the index of the field to read from operand0
1530 ir_binop_vector_extract
,
1533 * Interpolate fs input at offset
1535 * operand0 is the fs input
1536 * operand1 is the offset from the pixel center
1538 ir_binop_interpolate_at_offset
,
1541 * Interpolate fs input at sample position
1543 * operand0 is the fs input
1544 * operand1 is the sample ID
1546 ir_binop_interpolate_at_sample
,
1549 * A sentinel marking the last of the binary operations.
1551 ir_last_binop
= ir_binop_interpolate_at_sample
,
1554 * \name Fused floating-point multiply-add, part of ARB_gpu_shader5.
1563 * \name Conditional Select
1565 * A vector conditional select instruction (like ?:, but operating per-
1566 * component on vectors).
1568 * \see lower_instructions_visitor::ldexp_to_arith
1575 * \name Second half of a lowered bitfieldInsert() operation.
1577 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1583 ir_triop_bitfield_extract
,
1586 * Generate a value with one field of a vector changed
1588 * operand0 is the vector
1589 * operand1 is the value to write into the vector result
1590 * operand2 is the index in operand0 to be modified
1592 ir_triop_vector_insert
,
1595 * A sentinel marking the last of the ternary operations.
1597 ir_last_triop
= ir_triop_vector_insert
,
1599 ir_quadop_bitfield_insert
,
1604 * A sentinel marking the last of the ternary operations.
1606 ir_last_quadop
= ir_quadop_vector
,
1609 * A sentinel marking the last of all operations.
1611 ir_last_opcode
= ir_quadop_vector
1614 class ir_expression
: public ir_rvalue
{
1616 ir_expression(int op
, const struct glsl_type
*type
,
1617 ir_rvalue
*op0
, ir_rvalue
*op1
= NULL
,
1618 ir_rvalue
*op2
= NULL
, ir_rvalue
*op3
= NULL
);
1621 * Constructor for unary operation expressions
1623 ir_expression(int op
, ir_rvalue
*);
1626 * Constructor for binary operation expressions
1628 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
);
1631 * Constructor for ternary operation expressions
1633 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
, ir_rvalue
*op2
);
1635 virtual bool equals(const ir_instruction
*ir
,
1636 enum ir_node_type ignore
= ir_type_unset
) const;
1638 virtual ir_expression
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1641 * Attempt to constant-fold the expression
1643 * The "variable_context" hash table links ir_variable * to ir_constant *
1644 * that represent the variables' values. \c NULL represents an empty
1647 * If the expression cannot be constant folded, this method will return
1650 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1653 * Determine the number of operands used by an expression
1655 static unsigned int get_num_operands(ir_expression_operation
);
1658 * Determine the number of operands used by an expression
1660 unsigned int get_num_operands() const
1662 return (this->operation
== ir_quadop_vector
)
1663 ? this->type
->vector_elements
: get_num_operands(operation
);
1667 * Return whether the expression operates on vectors horizontally.
1669 bool is_horizontal() const
1671 return operation
== ir_binop_all_equal
||
1672 operation
== ir_binop_any_nequal
||
1673 operation
== ir_unop_any
||
1674 operation
== ir_binop_dot
||
1675 operation
== ir_quadop_vector
;
1679 * Return a string representing this expression's operator.
1681 const char *operator_string();
1684 * Return a string representing this expression's operator.
1686 static const char *operator_string(ir_expression_operation
);
1690 * Do a reverse-lookup to translate the given string into an operator.
1692 static ir_expression_operation
get_operator(const char *);
1694 virtual void accept(ir_visitor
*v
)
1699 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1701 ir_expression_operation operation
;
1702 ir_rvalue
*operands
[4];
1707 * HIR instruction representing a high-level function call, containing a list
1708 * of parameters and returning a value in the supplied temporary.
1710 class ir_call
: public ir_instruction
{
1712 ir_call(ir_function_signature
*callee
,
1713 ir_dereference_variable
*return_deref
,
1714 exec_list
*actual_parameters
)
1715 : ir_instruction(ir_type_call
), return_deref(return_deref
), callee(callee
), sub_var(NULL
), array_idx(NULL
)
1717 assert(callee
->return_type
!= NULL
);
1718 actual_parameters
->move_nodes_to(& this->actual_parameters
);
1719 this->use_builtin
= callee
->is_builtin();
1722 ir_call(ir_function_signature
*callee
,
1723 ir_dereference_variable
*return_deref
,
1724 exec_list
*actual_parameters
,
1725 ir_variable
*var
, ir_rvalue
*array_idx
)
1726 : ir_instruction(ir_type_call
), return_deref(return_deref
), callee(callee
), sub_var(var
), array_idx(array_idx
)
1728 assert(callee
->return_type
!= NULL
);
1729 actual_parameters
->move_nodes_to(& this->actual_parameters
);
1730 this->use_builtin
= callee
->is_builtin();
1733 virtual ir_call
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1735 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1737 virtual void accept(ir_visitor
*v
)
1742 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1745 * Get the name of the function being called.
1747 const char *callee_name() const
1749 return callee
->function_name();
1753 * Generates an inline version of the function before @ir,
1754 * storing the return value in return_deref.
1756 void generate_inline(ir_instruction
*ir
);
1759 * Storage for the function's return value.
1760 * This must be NULL if the return type is void.
1762 ir_dereference_variable
*return_deref
;
1765 * The specific function signature being called.
1767 ir_function_signature
*callee
;
1769 /* List of ir_rvalue of paramaters passed in this call. */
1770 exec_list actual_parameters
;
1772 /** Should this call only bind to a built-in function? */
1776 * ARB_shader_subroutine support -
1777 * the subroutine uniform variable and array index
1778 * rvalue to be used in the lowering pass later.
1780 ir_variable
*sub_var
;
1781 ir_rvalue
*array_idx
;
1786 * \name Jump-like IR instructions.
1788 * These include \c break, \c continue, \c return, and \c discard.
1791 class ir_jump
: public ir_instruction
{
1793 ir_jump(enum ir_node_type t
)
1799 class ir_return
: public ir_jump
{
1802 : ir_jump(ir_type_return
), value(NULL
)
1806 ir_return(ir_rvalue
*value
)
1807 : ir_jump(ir_type_return
), value(value
)
1811 virtual ir_return
*clone(void *mem_ctx
, struct hash_table
*) const;
1813 ir_rvalue
*get_value() const
1818 virtual void accept(ir_visitor
*v
)
1823 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1830 * Jump instructions used inside loops
1832 * These include \c break and \c continue. The \c break within a loop is
1833 * different from the \c break within a switch-statement.
1835 * \sa ir_switch_jump
1837 class ir_loop_jump
: public ir_jump
{
1844 ir_loop_jump(jump_mode mode
)
1845 : ir_jump(ir_type_loop_jump
)
1850 virtual ir_loop_jump
*clone(void *mem_ctx
, struct hash_table
*) const;
1852 virtual void accept(ir_visitor
*v
)
1857 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1859 bool is_break() const
1861 return mode
== jump_break
;
1864 bool is_continue() const
1866 return mode
== jump_continue
;
1869 /** Mode selector for the jump instruction. */
1870 enum jump_mode mode
;
1874 * IR instruction representing discard statements.
1876 class ir_discard
: public ir_jump
{
1879 : ir_jump(ir_type_discard
)
1881 this->condition
= NULL
;
1884 ir_discard(ir_rvalue
*cond
)
1885 : ir_jump(ir_type_discard
)
1887 this->condition
= cond
;
1890 virtual ir_discard
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1892 virtual void accept(ir_visitor
*v
)
1897 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1899 ir_rvalue
*condition
;
1905 * Texture sampling opcodes used in ir_texture
1907 enum ir_texture_opcode
{
1908 ir_tex
, /**< Regular texture look-up */
1909 ir_txb
, /**< Texture look-up with LOD bias */
1910 ir_txl
, /**< Texture look-up with explicit LOD */
1911 ir_txd
, /**< Texture look-up with partial derivatvies */
1912 ir_txf
, /**< Texel fetch with explicit LOD */
1913 ir_txf_ms
, /**< Multisample texture fetch */
1914 ir_txs
, /**< Texture size */
1915 ir_lod
, /**< Texture lod query */
1916 ir_tg4
, /**< Texture gather */
1917 ir_query_levels
, /**< Texture levels query */
1918 ir_texture_samples
, /**< Texture samples query */
1923 * IR instruction to sample a texture
1925 * The specific form of the IR instruction depends on the \c mode value
1926 * selected from \c ir_texture_opcodes. In the printed IR, these will
1929 * Texel offset (0 or an expression)
1930 * | Projection divisor
1931 * | | Shadow comparitor
1934 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1935 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1936 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1937 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1938 * (txf <type> <sampler> <coordinate> 0 <lod>)
1940 * <type> <sampler> <coordinate> <sample_index>)
1941 * (txs <type> <sampler> <lod>)
1942 * (lod <type> <sampler> <coordinate>)
1943 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
1944 * (query_levels <type> <sampler>)
1946 class ir_texture
: public ir_rvalue
{
1948 ir_texture(enum ir_texture_opcode op
)
1949 : ir_rvalue(ir_type_texture
),
1950 op(op
), sampler(NULL
), coordinate(NULL
), projector(NULL
),
1951 shadow_comparitor(NULL
), offset(NULL
)
1953 memset(&lod_info
, 0, sizeof(lod_info
));
1956 virtual ir_texture
*clone(void *mem_ctx
, struct hash_table
*) const;
1958 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1960 virtual void accept(ir_visitor
*v
)
1965 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1967 virtual bool equals(const ir_instruction
*ir
,
1968 enum ir_node_type ignore
= ir_type_unset
) const;
1971 * Return a string representing the ir_texture_opcode.
1973 const char *opcode_string();
1975 /** Set the sampler and type. */
1976 void set_sampler(ir_dereference
*sampler
, const glsl_type
*type
);
1979 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1981 static ir_texture_opcode
get_opcode(const char *);
1983 enum ir_texture_opcode op
;
1985 /** Sampler to use for the texture access. */
1986 ir_dereference
*sampler
;
1988 /** Texture coordinate to sample */
1989 ir_rvalue
*coordinate
;
1992 * Value used for projective divide.
1994 * If there is no projective divide (the common case), this will be
1995 * \c NULL. Optimization passes should check for this to point to a constant
1996 * of 1.0 and replace that with \c NULL.
1998 ir_rvalue
*projector
;
2001 * Coordinate used for comparison on shadow look-ups.
2003 * If there is no shadow comparison, this will be \c NULL. For the
2004 * \c ir_txf opcode, this *must* be \c NULL.
2006 ir_rvalue
*shadow_comparitor
;
2008 /** Texel offset. */
2012 ir_rvalue
*lod
; /**< Floating point LOD */
2013 ir_rvalue
*bias
; /**< Floating point LOD bias */
2014 ir_rvalue
*sample_index
; /**< MSAA sample index */
2015 ir_rvalue
*component
; /**< Gather component selector */
2017 ir_rvalue
*dPdx
; /**< Partial derivative of coordinate wrt X */
2018 ir_rvalue
*dPdy
; /**< Partial derivative of coordinate wrt Y */
2024 struct ir_swizzle_mask
{
2031 * Number of components in the swizzle.
2033 unsigned num_components
:3;
2036 * Does the swizzle contain duplicate components?
2038 * L-value swizzles cannot contain duplicate components.
2040 unsigned has_duplicates
:1;
2044 class ir_swizzle
: public ir_rvalue
{
2046 ir_swizzle(ir_rvalue
*, unsigned x
, unsigned y
, unsigned z
, unsigned w
,
2049 ir_swizzle(ir_rvalue
*val
, const unsigned *components
, unsigned count
);
2051 ir_swizzle(ir_rvalue
*val
, ir_swizzle_mask mask
);
2053 virtual ir_swizzle
*clone(void *mem_ctx
, struct hash_table
*) const;
2055 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2058 * Construct an ir_swizzle from the textual representation. Can fail.
2060 static ir_swizzle
*create(ir_rvalue
*, const char *, unsigned vector_length
);
2062 virtual void accept(ir_visitor
*v
)
2067 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2069 virtual bool equals(const ir_instruction
*ir
,
2070 enum ir_node_type ignore
= ir_type_unset
) const;
2072 bool is_lvalue() const
2074 return val
->is_lvalue() && !mask
.has_duplicates
;
2078 * Get the variable that is ultimately referenced by an r-value
2080 virtual ir_variable
*variable_referenced() const;
2083 ir_swizzle_mask mask
;
2087 * Initialize the mask component of a swizzle
2089 * This is used by the \c ir_swizzle constructors.
2091 void init_mask(const unsigned *components
, unsigned count
);
2095 class ir_dereference
: public ir_rvalue
{
2097 virtual ir_dereference
*clone(void *mem_ctx
, struct hash_table
*) const = 0;
2099 bool is_lvalue() const;
2102 * Get the variable that is ultimately referenced by an r-value
2104 virtual ir_variable
*variable_referenced() const = 0;
2107 ir_dereference(enum ir_node_type t
)
2114 class ir_dereference_variable
: public ir_dereference
{
2116 ir_dereference_variable(ir_variable
*var
);
2118 virtual ir_dereference_variable
*clone(void *mem_ctx
,
2119 struct hash_table
*) const;
2121 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2123 virtual bool equals(const ir_instruction
*ir
,
2124 enum ir_node_type ignore
= ir_type_unset
) const;
2127 * Get the variable that is ultimately referenced by an r-value
2129 virtual ir_variable
*variable_referenced() const
2134 virtual ir_variable
*whole_variable_referenced()
2136 /* ir_dereference_variable objects always dereference the entire
2137 * variable. However, if this dereference is dereferenced by anything
2138 * else, the complete deferefernce chain is not a whole-variable
2139 * dereference. This method should only be called on the top most
2140 * ir_rvalue in a dereference chain.
2145 virtual void accept(ir_visitor
*v
)
2150 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2153 * Object being dereferenced.
2159 class ir_dereference_array
: public ir_dereference
{
2161 ir_dereference_array(ir_rvalue
*value
, ir_rvalue
*array_index
);
2163 ir_dereference_array(ir_variable
*var
, ir_rvalue
*array_index
);
2165 virtual ir_dereference_array
*clone(void *mem_ctx
,
2166 struct hash_table
*) const;
2168 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2170 virtual bool equals(const ir_instruction
*ir
,
2171 enum ir_node_type ignore
= ir_type_unset
) const;
2174 * Get the variable that is ultimately referenced by an r-value
2176 virtual ir_variable
*variable_referenced() const
2178 return this->array
->variable_referenced();
2181 virtual void accept(ir_visitor
*v
)
2186 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2189 ir_rvalue
*array_index
;
2192 void set_array(ir_rvalue
*value
);
2196 class ir_dereference_record
: public ir_dereference
{
2198 ir_dereference_record(ir_rvalue
*value
, const char *field
);
2200 ir_dereference_record(ir_variable
*var
, const char *field
);
2202 virtual ir_dereference_record
*clone(void *mem_ctx
,
2203 struct hash_table
*) const;
2205 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2208 * Get the variable that is ultimately referenced by an r-value
2210 virtual ir_variable
*variable_referenced() const
2212 return this->record
->variable_referenced();
2215 virtual void accept(ir_visitor
*v
)
2220 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2228 * Data stored in an ir_constant
2230 union ir_constant_data
{
2239 class ir_constant
: public ir_rvalue
{
2241 ir_constant(const struct glsl_type
*type
, const ir_constant_data
*data
);
2242 ir_constant(bool b
, unsigned vector_elements
=1);
2243 ir_constant(unsigned int u
, unsigned vector_elements
=1);
2244 ir_constant(int i
, unsigned vector_elements
=1);
2245 ir_constant(float f
, unsigned vector_elements
=1);
2246 ir_constant(double d
, unsigned vector_elements
=1);
2249 * Construct an ir_constant from a list of ir_constant values
2251 ir_constant(const struct glsl_type
*type
, exec_list
*values
);
2254 * Construct an ir_constant from a scalar component of another ir_constant
2256 * The new \c ir_constant inherits the type of the component from the
2260 * In the case of a matrix constant, the new constant is a scalar, \b not
2263 ir_constant(const ir_constant
*c
, unsigned i
);
2266 * Return a new ir_constant of the specified type containing all zeros.
2268 static ir_constant
*zero(void *mem_ctx
, const glsl_type
*type
);
2270 virtual ir_constant
*clone(void *mem_ctx
, struct hash_table
*) const;
2272 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2274 virtual void accept(ir_visitor
*v
)
2279 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2281 virtual bool equals(const ir_instruction
*ir
,
2282 enum ir_node_type ignore
= ir_type_unset
) const;
2285 * Get a particular component of a constant as a specific type
2287 * This is useful, for example, to get a value from an integer constant
2288 * as a float or bool. This appears frequently when constructors are
2289 * called with all constant parameters.
2292 bool get_bool_component(unsigned i
) const;
2293 float get_float_component(unsigned i
) const;
2294 double get_double_component(unsigned i
) const;
2295 int get_int_component(unsigned i
) const;
2296 unsigned get_uint_component(unsigned i
) const;
2299 ir_constant
*get_array_element(unsigned i
) const;
2301 ir_constant
*get_record_field(const char *name
);
2304 * Copy the values on another constant at a given offset.
2306 * The offset is ignored for array or struct copies, it's only for
2307 * scalars or vectors into vectors or matrices.
2309 * With identical types on both sides and zero offset it's clone()
2310 * without creating a new object.
2313 void copy_offset(ir_constant
*src
, int offset
);
2316 * Copy the values on another constant at a given offset and
2317 * following an assign-like mask.
2319 * The mask is ignored for scalars.
2321 * Note that this function only handles what assign can handle,
2322 * i.e. at most a vector as source and a column of a matrix as
2326 void copy_masked_offset(ir_constant
*src
, int offset
, unsigned int mask
);
2329 * Determine whether a constant has the same value as another constant
2331 * \sa ir_constant::is_zero, ir_constant::is_one,
2332 * ir_constant::is_negative_one
2334 bool has_value(const ir_constant
*) const;
2337 * Return true if this ir_constant represents the given value.
2339 * For vectors, this checks that each component is the given value.
2341 virtual bool is_value(float f
, int i
) const;
2342 virtual bool is_zero() const;
2343 virtual bool is_one() const;
2344 virtual bool is_negative_one() const;
2347 * Return true for constants that could be stored as 16-bit unsigned values.
2349 * Note that this will return true even for signed integer ir_constants, as
2350 * long as the value is non-negative and fits in 16-bits.
2352 virtual bool is_uint16_constant() const;
2355 * Value of the constant.
2357 * The field used to back the values supplied by the constant is determined
2358 * by the type associated with the \c ir_instruction. Constants may be
2359 * scalars, vectors, or matrices.
2361 union ir_constant_data value
;
2363 /* Array elements */
2364 ir_constant
**array_elements
;
2366 /* Structure fields */
2367 exec_list components
;
2371 * Parameterless constructor only used by the clone method
2377 * IR instruction to emit a vertex in a geometry shader.
2379 class ir_emit_vertex
: public ir_instruction
{
2381 ir_emit_vertex(ir_rvalue
*stream
)
2382 : ir_instruction(ir_type_emit_vertex
),
2388 virtual void accept(ir_visitor
*v
)
2393 virtual ir_emit_vertex
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2395 return new(mem_ctx
) ir_emit_vertex(this->stream
->clone(mem_ctx
, ht
));
2398 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2400 int stream_id() const
2402 return stream
->as_constant()->value
.i
[0];
2409 * IR instruction to complete the current primitive and start a new one in a
2412 class ir_end_primitive
: public ir_instruction
{
2414 ir_end_primitive(ir_rvalue
*stream
)
2415 : ir_instruction(ir_type_end_primitive
),
2421 virtual void accept(ir_visitor
*v
)
2426 virtual ir_end_primitive
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2428 return new(mem_ctx
) ir_end_primitive(this->stream
->clone(mem_ctx
, ht
));
2431 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2433 int stream_id() const
2435 return stream
->as_constant()->value
.i
[0];
2442 * IR instruction for tessellation control and compute shader barrier.
2444 class ir_barrier
: public ir_instruction
{
2447 : ir_instruction(ir_type_barrier
)
2451 virtual void accept(ir_visitor
*v
)
2456 virtual ir_barrier
*clone(void *mem_ctx
, struct hash_table
*) const
2458 return new(mem_ctx
) ir_barrier();
2461 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2467 * Apply a visitor to each IR node in a list
2470 visit_exec_list(exec_list
*list
, ir_visitor
*visitor
);
2473 * Validate invariants on each IR node in a list
2475 void validate_ir_tree(exec_list
*instructions
);
2477 struct _mesa_glsl_parse_state
;
2478 struct gl_shader_program
;
2481 * Detect whether an unlinked shader contains static recursion
2483 * If the list of instructions is determined to contain static recursion,
2484 * \c _mesa_glsl_error will be called to emit error messages for each function
2485 * that is in the recursion cycle.
2488 detect_recursion_unlinked(struct _mesa_glsl_parse_state
*state
,
2489 exec_list
*instructions
);
2492 * Detect whether a linked shader contains static recursion
2494 * If the list of instructions is determined to contain static recursion,
2495 * \c link_error_printf will be called to emit error messages for each function
2496 * that is in the recursion cycle. In addition,
2497 * \c gl_shader_program::LinkStatus will be set to false.
2500 detect_recursion_linked(struct gl_shader_program
*prog
,
2501 exec_list
*instructions
);
2504 * Make a clone of each IR instruction in a list
2506 * \param in List of IR instructions that are to be cloned
2507 * \param out List to hold the cloned instructions
2510 clone_ir_list(void *mem_ctx
, exec_list
*out
, const exec_list
*in
);
2513 _mesa_glsl_initialize_variables(exec_list
*instructions
,
2514 struct _mesa_glsl_parse_state
*state
);
2517 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state
*state
);
2520 _mesa_glsl_initialize_builtin_functions();
2522 extern ir_function_signature
*
2523 _mesa_glsl_find_builtin_function(_mesa_glsl_parse_state
*state
,
2524 const char *name
, exec_list
*actual_parameters
);
2526 extern ir_function
*
2527 _mesa_glsl_find_builtin_function_by_name(const char *name
);
2530 _mesa_glsl_get_builtin_function_shader(void);
2533 _mesa_glsl_release_functions(void);
2536 _mesa_glsl_release_builtin_functions(void);
2539 reparent_ir(exec_list
*list
, void *mem_ctx
);
2541 struct glsl_symbol_table
;
2544 import_prototypes(const exec_list
*source
, exec_list
*dest
,
2545 struct glsl_symbol_table
*symbols
, void *mem_ctx
);
2548 ir_has_call(ir_instruction
*ir
);
2551 do_set_program_inouts(exec_list
*instructions
, struct gl_program
*prog
,
2552 gl_shader_stage shader_stage
);
2555 prototype_string(const glsl_type
*return_type
, const char *name
,
2556 exec_list
*parameters
);
2559 mode_string(const ir_variable
*var
);
2562 * Built-in / reserved GL variables names start with "gl_"
2565 is_gl_identifier(const char *s
)
2567 return s
&& s
[0] == 'g' && s
[1] == 'l' && s
[2] == '_';
2571 #endif /* __cplusplus */
2573 extern void _mesa_print_ir(FILE *f
, struct exec_list
*instructions
,
2574 struct _mesa_glsl_parse_state
*state
);
2577 fprint_ir(FILE *f
, const void *instruction
);
2584 vertices_per_prim(GLenum prim
);