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 part of a shader storage block.
458 inline bool is_in_shader_storage_block() const
460 return this->data
.mode
== ir_var_shader_storage
&&
461 this->interface_type
!= NULL
;
465 * Determine whether or not a variable is the declaration of an interface
468 * For the first declaration below, there will be an \c ir_variable named
469 * "instance" whose type and whose instance_type will be the same
470 * \cglsl_type. For the second declaration, there will be an \c ir_variable
471 * named "f" whose type is float and whose instance_type is B2.
473 * "instance" is an interface instance variable, but "f" is not.
483 inline bool is_interface_instance() const
485 return this->type
->without_array() == this->interface_type
;
489 * Set this->interface_type on a newly created variable.
491 void init_interface_type(const struct glsl_type
*type
)
493 assert(this->interface_type
== NULL
);
494 this->interface_type
= type
;
495 if (this->is_interface_instance()) {
496 this->u
.max_ifc_array_access
=
497 rzalloc_array(this, unsigned, type
->length
);
502 * Change this->interface_type on a variable that previously had a
503 * different, but compatible, interface_type. This is used during linking
504 * to set the size of arrays in interface blocks.
506 void change_interface_type(const struct glsl_type
*type
)
508 if (this->u
.max_ifc_array_access
!= NULL
) {
509 /* max_ifc_array_access has already been allocated, so make sure the
510 * new interface has the same number of fields as the old one.
512 assert(this->interface_type
->length
== type
->length
);
514 this->interface_type
= type
;
518 * Change this->interface_type on a variable that previously had a
519 * different, and incompatible, interface_type. This is used during
520 * compilation to handle redeclaration of the built-in gl_PerVertex
523 void reinit_interface_type(const struct glsl_type
*type
)
525 if (this->u
.max_ifc_array_access
!= NULL
) {
527 /* Redeclaring gl_PerVertex is only allowed if none of the built-ins
528 * it defines have been accessed yet; so it's safe to throw away the
529 * old max_ifc_array_access pointer, since all of its values are
532 for (unsigned i
= 0; i
< this->interface_type
->length
; i
++)
533 assert(this->u
.max_ifc_array_access
[i
] == 0);
535 ralloc_free(this->u
.max_ifc_array_access
);
536 this->u
.max_ifc_array_access
= NULL
;
538 this->interface_type
= NULL
;
539 init_interface_type(type
);
542 const glsl_type
*get_interface_type() const
544 return this->interface_type
;
548 * Get the max_ifc_array_access pointer
550 * A "set" function is not needed because the array is dynmically allocated
553 inline unsigned *get_max_ifc_array_access()
555 assert(this->data
._num_state_slots
== 0);
556 return this->u
.max_ifc_array_access
;
559 inline unsigned get_num_state_slots() const
561 assert(!this->is_interface_instance()
562 || this->data
._num_state_slots
== 0);
563 return this->data
._num_state_slots
;
566 inline void set_num_state_slots(unsigned n
)
568 assert(!this->is_interface_instance()
570 this->data
._num_state_slots
= n
;
573 inline ir_state_slot
*get_state_slots()
575 return this->is_interface_instance() ? NULL
: this->u
.state_slots
;
578 inline const ir_state_slot
*get_state_slots() const
580 return this->is_interface_instance() ? NULL
: this->u
.state_slots
;
583 inline ir_state_slot
*allocate_state_slots(unsigned n
)
585 assert(!this->is_interface_instance());
587 this->u
.state_slots
= ralloc_array(this, ir_state_slot
, n
);
588 this->data
._num_state_slots
= 0;
590 if (this->u
.state_slots
!= NULL
)
591 this->data
._num_state_slots
= n
;
593 return this->u
.state_slots
;
596 inline bool is_name_ralloced() const
598 return this->name
!= ir_variable::tmp_name
;
602 * Enable emitting extension warnings for this variable
604 void enable_extension_warning(const char *extension
);
607 * Get the extension warning string for this variable
609 * If warnings are not enabled, \c NULL is returned.
611 const char *get_extension_warning() const;
614 * Declared type of the variable
616 const struct glsl_type
*type
;
619 * Declared name of the variable
623 struct ir_variable_data
{
626 * Is the variable read-only?
628 * This is set for variables declared as \c const, shader inputs,
631 unsigned read_only
:1;
635 unsigned invariant
:1;
639 * Has this variable been used for reading or writing?
641 * Several GLSL semantic checks require knowledge of whether or not a
642 * variable has been used. For example, it is an error to redeclare a
643 * variable as invariant after it has been used.
645 * This is only maintained in the ast_to_hir.cpp path, not in
646 * Mesa's fixed function or ARB program paths.
651 * Has this variable been statically assigned?
653 * This answers whether the variable was assigned in any path of
654 * the shader during ast_to_hir. This doesn't answer whether it is
655 * still written after dead code removal, nor is it maintained in
656 * non-ast_to_hir.cpp (GLSL parsing) paths.
661 * Enum indicating how the variable was declared. See
662 * ir_var_declaration_type.
664 * This is used to detect certain kinds of illegal variable redeclarations.
666 unsigned how_declared
:2;
669 * Storage class of the variable.
671 * \sa ir_variable_mode
676 * Interpolation mode for shader inputs / outputs
678 * \sa ir_variable_interpolation
680 unsigned interpolation
:2;
683 * \name ARB_fragment_coord_conventions
686 unsigned origin_upper_left
:1;
687 unsigned pixel_center_integer
:1;
691 * Was the location explicitly set in the shader?
693 * If the location is explicitly set in the shader, it \b cannot be changed
694 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
697 unsigned explicit_location
:1;
698 unsigned explicit_index
:1;
701 * Was an initial binding explicitly set in the shader?
703 * If so, constant_value contains an integer ir_constant representing the
704 * initial binding point.
706 unsigned explicit_binding
:1;
709 * Does this variable have an initializer?
711 * This is used by the linker to cross-validiate initializers of global
714 unsigned has_initializer
:1;
717 * Is this variable a generic output or input that has not yet been matched
718 * up to a variable in another stage of the pipeline?
720 * This is used by the linker as scratch storage while assigning locations
721 * to generic inputs and outputs.
723 unsigned is_unmatched_generic_inout
:1;
726 * If non-zero, then this variable may be packed along with other variables
727 * into a single varying slot, so this offset should be applied when
728 * accessing components. For example, an offset of 1 means that the x
729 * component of this variable is actually stored in component y of the
730 * location specified by \c location.
732 unsigned location_frac
:2;
735 * Layout of the matrix. Uses glsl_matrix_layout values.
737 unsigned matrix_layout
:2;
740 * Non-zero if this variable was created by lowering a named interface
741 * block which was not an array.
743 * Note that this variable and \c from_named_ifc_block_array will never
746 unsigned from_named_ifc_block_nonarray
:1;
749 * Non-zero if this variable was created by lowering a named interface
750 * block which was an array.
752 * Note that this variable and \c from_named_ifc_block_nonarray will never
755 unsigned from_named_ifc_block_array
:1;
758 * Non-zero if the variable must be a shader input. This is useful for
759 * constraints on function parameters.
761 unsigned must_be_shader_input
:1;
764 * Output index for dual source blending.
767 * The GLSL spec only allows the values 0 or 1 for the index in \b dual
773 * \brief Layout qualifier for gl_FragDepth.
775 * This is not equal to \c ir_depth_layout_none if and only if this
776 * variable is \c gl_FragDepth and a layout qualifier is specified.
778 ir_depth_layout depth_layout
:3;
781 * ARB_shader_image_load_store qualifiers.
783 unsigned image_read_only
:1; /**< "readonly" qualifier. */
784 unsigned image_write_only
:1; /**< "writeonly" qualifier. */
785 unsigned image_coherent
:1;
786 unsigned image_volatile
:1;
787 unsigned image_restrict
:1;
790 * ARB_shader_storage_buffer_object
792 unsigned from_ssbo_unsized_array
:1; /**< unsized array buffer variable. */
795 * Emit a warning if this variable is accessed.
798 uint8_t warn_extension_index
;
801 /** Image internal format if specified explicitly, otherwise GL_NONE. */
802 uint16_t image_format
;
806 * Number of state slots used
809 * This could be stored in as few as 7-bits, if necessary. If it is made
810 * smaller, add an assertion to \c ir_variable::allocate_state_slots to
813 uint16_t _num_state_slots
;
817 * Initial binding point for a sampler, atomic, or UBO.
819 * For array types, this represents the binding point for the first element.
824 * Storage location of the base of this variable
826 * The precise meaning of this field depends on the nature of the variable.
828 * - Vertex shader input: one of the values from \c gl_vert_attrib.
829 * - Vertex shader output: one of the values from \c gl_varying_slot.
830 * - Geometry shader input: one of the values from \c gl_varying_slot.
831 * - Geometry shader output: one of the values from \c gl_varying_slot.
832 * - Fragment shader input: one of the values from \c gl_varying_slot.
833 * - Fragment shader output: one of the values from \c gl_frag_result.
834 * - Uniforms: Per-stage uniform slot number for default uniform block.
835 * - Uniforms: Index within the uniform block definition for UBO members.
836 * - Non-UBO Uniforms: explicit location until linking then reused to
837 * store uniform slot number.
838 * - Other: This field is not currently used.
840 * If the variable is a uniform, shader input, or shader output, and the
841 * slot has not been assigned, the value will be -1.
846 * Vertex stream output identifier.
851 * Location an atomic counter is stored at.
858 * Highest element accessed with a constant expression array index
860 * Not used for non-array variables.
862 unsigned max_array_access
;
865 * Allow (only) ir_variable direct access private members.
867 friend class ir_variable
;
871 * Value assigned in the initializer of a variable declared "const"
873 ir_constant
*constant_value
;
876 * Constant expression assigned in the initializer of the variable
879 * This field and \c ::constant_value are distinct. Even if the two fields
880 * refer to constants with the same value, they must point to separate
883 ir_constant
*constant_initializer
;
886 static const char *const warn_extension_table
[];
890 * For variables which satisfy the is_interface_instance() predicate,
891 * this points to an array of integers such that if the ith member of
892 * the interface block is an array, max_ifc_array_access[i] is the
893 * maximum array element of that member that has been accessed. If the
894 * ith member of the interface block is not an array,
895 * max_ifc_array_access[i] is unused.
897 * For variables whose type is not an interface block, this pointer is
900 unsigned *max_ifc_array_access
;
903 * Built-in state that backs this uniform
905 * Once set at variable creation, \c state_slots must remain invariant.
907 * If the variable is not a uniform, \c _num_state_slots will be zero
908 * and \c state_slots will be \c NULL.
910 ir_state_slot
*state_slots
;
914 * For variables that are in an interface block or are an instance of an
915 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
917 * \sa ir_variable::location
919 const glsl_type
*interface_type
;
922 * Name used for anonymous compiler temporaries
924 static const char tmp_name
[];
928 * Should the construct keep names for ir_var_temporary variables?
930 * When this global is false, names passed to the constructor for
931 * \c ir_var_temporary variables will be dropped. Instead, the variable will
932 * be named "compiler_temp". This name will be in static storage.
935 * \b NEVER change the mode of an \c ir_var_temporary.
938 * This variable is \b not thread-safe. It is global, \b not
939 * per-context. It begins life false. A context can, at some point, make
940 * it true. From that point on, it will be true forever. This should be
941 * okay since it will only be set true while debugging.
943 static bool temporaries_allocate_names
;
947 * A function that returns whether a built-in function is available in the
948 * current shading language (based on version, ES or desktop, and extensions).
950 typedef bool (*builtin_available_predicate
)(const _mesa_glsl_parse_state
*);
954 * The representation of a function instance; may be the full definition or
955 * simply a prototype.
957 class ir_function_signature
: public ir_instruction
{
958 /* An ir_function_signature will be part of the list of signatures in
962 ir_function_signature(const glsl_type
*return_type
,
963 builtin_available_predicate builtin_avail
= NULL
);
965 virtual ir_function_signature
*clone(void *mem_ctx
,
966 struct hash_table
*ht
) const;
967 ir_function_signature
*clone_prototype(void *mem_ctx
,
968 struct hash_table
*ht
) const;
970 virtual void accept(ir_visitor
*v
)
975 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
978 * Attempt to evaluate this function as a constant expression,
979 * given a list of the actual parameters and the variable context.
980 * Returns NULL for non-built-ins.
982 ir_constant
*constant_expression_value(exec_list
*actual_parameters
, struct hash_table
*variable_context
);
985 * Get the name of the function for which this is a signature
987 const char *function_name() const;
990 * Get a handle to the function for which this is a signature
992 * There is no setter function, this function returns a \c const pointer,
993 * and \c ir_function_signature::_function is private for a reason. The
994 * only way to make a connection between a function and function signature
995 * is via \c ir_function::add_signature. This helps ensure that certain
996 * invariants (i.e., a function signature is in the list of signatures for
997 * its \c _function) are met.
999 * \sa ir_function::add_signature
1001 inline const class ir_function
*function() const
1003 return this->_function
;
1007 * Check whether the qualifiers match between this signature's parameters
1008 * and the supplied parameter list. If not, returns the name of the first
1009 * parameter with mismatched qualifiers (for use in error messages).
1011 const char *qualifiers_match(exec_list
*params
);
1014 * Replace the current parameter list with the given one. This is useful
1015 * if the current information came from a prototype, and either has invalid
1016 * or missing parameter names.
1018 void replace_parameters(exec_list
*new_params
);
1021 * Function return type.
1023 * \note This discards the optional precision qualifier.
1025 const struct glsl_type
*return_type
;
1028 * List of ir_variable of function parameters.
1030 * This represents the storage. The paramaters passed in a particular
1031 * call will be in ir_call::actual_paramaters.
1033 struct exec_list parameters
;
1035 /** Whether or not this function has a body (which may be empty). */
1036 unsigned is_defined
:1;
1038 /** Whether or not this function signature is a built-in. */
1039 bool is_builtin() const;
1042 * Whether or not this function is an intrinsic to be implemented
1047 /** Whether or not a built-in is available for this shader. */
1048 bool is_builtin_available(const _mesa_glsl_parse_state
*state
) const;
1050 /** Body of instructions in the function. */
1051 struct exec_list body
;
1055 * A function pointer to a predicate that answers whether a built-in
1056 * function is available in the current shader. NULL if not a built-in.
1058 builtin_available_predicate builtin_avail
;
1060 /** Function of which this signature is one overload. */
1061 class ir_function
*_function
;
1063 /** Function signature of which this one is a prototype clone */
1064 const ir_function_signature
*origin
;
1066 friend class ir_function
;
1069 * Helper function to run a list of instructions for constant
1070 * expression evaluation.
1072 * The hash table represents the values of the visible variables.
1073 * There are no scoping issues because the table is indexed on
1074 * ir_variable pointers, not variable names.
1076 * Returns false if the expression is not constant, true otherwise,
1077 * and the value in *result if result is non-NULL.
1079 bool constant_expression_evaluate_expression_list(const struct exec_list
&body
,
1080 struct hash_table
*variable_context
,
1081 ir_constant
**result
);
1086 * Header for tracking multiple overloaded functions with the same name.
1087 * Contains a list of ir_function_signatures representing each of the
1090 class ir_function
: public ir_instruction
{
1092 ir_function(const char *name
);
1094 virtual ir_function
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1096 virtual void accept(ir_visitor
*v
)
1101 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1103 void add_signature(ir_function_signature
*sig
)
1105 sig
->_function
= this;
1106 this->signatures
.push_tail(sig
);
1110 * Find a signature that matches a set of actual parameters, taking implicit
1111 * conversions into account. Also flags whether the match was exact.
1113 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1114 const exec_list
*actual_param
,
1115 bool allow_builtins
,
1116 bool *match_is_exact
);
1119 * Find a signature that matches a set of actual parameters, taking implicit
1120 * conversions into account.
1122 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1123 const exec_list
*actual_param
,
1124 bool allow_builtins
);
1127 * Find a signature that exactly matches a set of actual parameters without
1128 * any implicit type conversions.
1130 ir_function_signature
*exact_matching_signature(_mesa_glsl_parse_state
*state
,
1131 const exec_list
*actual_ps
);
1134 * Name of the function.
1138 /** Whether or not this function has a signature that isn't a built-in. */
1139 bool has_user_signature();
1142 * List of ir_function_signature for each overloaded function with this name.
1144 struct exec_list signatures
;
1147 * is this function a subroutine type declaration
1148 * e.g. subroutine void type1(float arg1);
1153 * is this function associated to a subroutine type
1154 * e.g. subroutine (type1, type2) function_name { function_body };
1155 * would have num_subroutine_types 2,
1156 * and pointers to the type1 and type2 types.
1158 int num_subroutine_types
;
1159 const struct glsl_type
**subroutine_types
;
1162 inline const char *ir_function_signature::function_name() const
1164 return this->_function
->name
;
1170 * IR instruction representing high-level if-statements
1172 class ir_if
: public ir_instruction
{
1174 ir_if(ir_rvalue
*condition
)
1175 : ir_instruction(ir_type_if
), condition(condition
)
1179 virtual ir_if
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1181 virtual void accept(ir_visitor
*v
)
1186 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1188 ir_rvalue
*condition
;
1189 /** List of ir_instruction for the body of the then branch */
1190 exec_list then_instructions
;
1191 /** List of ir_instruction for the body of the else branch */
1192 exec_list else_instructions
;
1197 * IR instruction representing a high-level loop structure.
1199 class ir_loop
: public ir_instruction
{
1203 virtual ir_loop
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1205 virtual void accept(ir_visitor
*v
)
1210 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1212 /** List of ir_instruction that make up the body of the loop. */
1213 exec_list body_instructions
;
1217 class ir_assignment
: public ir_instruction
{
1219 ir_assignment(ir_rvalue
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
= NULL
);
1222 * Construct an assignment with an explicit write mask
1225 * Since a write mask is supplied, the LHS must already be a bare
1226 * \c ir_dereference. The cannot be any swizzles in the LHS.
1228 ir_assignment(ir_dereference
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
,
1229 unsigned write_mask
);
1231 virtual ir_assignment
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1233 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1235 virtual void accept(ir_visitor
*v
)
1240 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1243 * Get a whole variable written by an assignment
1245 * If the LHS of the assignment writes a whole variable, the variable is
1246 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1249 * - Assigning to a scalar
1250 * - Assigning to all components of a vector
1251 * - Whole array (or matrix) assignment
1252 * - Whole structure assignment
1254 ir_variable
*whole_variable_written();
1257 * Set the LHS of an assignment
1259 void set_lhs(ir_rvalue
*lhs
);
1262 * Left-hand side of the assignment.
1264 * This should be treated as read only. If you need to set the LHS of an
1265 * assignment, use \c ir_assignment::set_lhs.
1267 ir_dereference
*lhs
;
1270 * Value being assigned
1275 * Optional condition for the assignment.
1277 ir_rvalue
*condition
;
1281 * Component mask written
1283 * For non-vector types in the LHS, this field will be zero. For vector
1284 * types, a bit will be set for each component that is written. Note that
1285 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1287 * A partially-set write mask means that each enabled channel gets
1288 * the value from a consecutive channel of the rhs. For example,
1289 * to write just .xyw of gl_FrontColor with color:
1291 * (assign (constant bool (1)) (xyw)
1292 * (var_ref gl_FragColor)
1293 * (swiz xyw (var_ref color)))
1295 unsigned write_mask
:4;
1298 /* Update ir_expression::get_num_operands() and operator_strs when
1299 * updating this list.
1301 enum ir_expression_operation
{
1310 ir_unop_exp
, /**< Log base e on gentype */
1311 ir_unop_log
, /**< Natural log on gentype */
1314 ir_unop_f2i
, /**< Float-to-integer conversion. */
1315 ir_unop_f2u
, /**< Float-to-unsigned conversion. */
1316 ir_unop_i2f
, /**< Integer-to-float conversion. */
1317 ir_unop_f2b
, /**< Float-to-boolean conversion */
1318 ir_unop_b2f
, /**< Boolean-to-float conversion */
1319 ir_unop_i2b
, /**< int-to-boolean conversion */
1320 ir_unop_b2i
, /**< Boolean-to-int conversion */
1321 ir_unop_u2f
, /**< Unsigned-to-float conversion. */
1322 ir_unop_i2u
, /**< Integer-to-unsigned conversion. */
1323 ir_unop_u2i
, /**< Unsigned-to-integer conversion. */
1324 ir_unop_d2f
, /**< Double-to-float conversion. */
1325 ir_unop_f2d
, /**< Float-to-double conversion. */
1326 ir_unop_d2i
, /**< Double-to-integer conversion. */
1327 ir_unop_i2d
, /**< Integer-to-double conversion. */
1328 ir_unop_d2u
, /**< Double-to-unsigned conversion. */
1329 ir_unop_u2d
, /**< Unsigned-to-double conversion. */
1330 ir_unop_d2b
, /**< Double-to-boolean conversion. */
1331 ir_unop_bitcast_i2f
, /**< Bit-identical int-to-float "conversion" */
1332 ir_unop_bitcast_f2i
, /**< Bit-identical float-to-int "conversion" */
1333 ir_unop_bitcast_u2f
, /**< Bit-identical uint-to-float "conversion" */
1334 ir_unop_bitcast_f2u
, /**< Bit-identical float-to-uint "conversion" */
1338 * \name Unary floating-point rounding operations.
1349 * \name Trigonometric operations.
1357 * \name Partial derivatives.
1361 ir_unop_dFdx_coarse
,
1364 ir_unop_dFdy_coarse
,
1369 * \name Floating point pack and unpack operations.
1372 ir_unop_pack_snorm_2x16
,
1373 ir_unop_pack_snorm_4x8
,
1374 ir_unop_pack_unorm_2x16
,
1375 ir_unop_pack_unorm_4x8
,
1376 ir_unop_pack_half_2x16
,
1377 ir_unop_unpack_snorm_2x16
,
1378 ir_unop_unpack_snorm_4x8
,
1379 ir_unop_unpack_unorm_2x16
,
1380 ir_unop_unpack_unorm_4x8
,
1381 ir_unop_unpack_half_2x16
,
1385 * \name Lowered floating point unpacking operations.
1387 * \see lower_packing_builtins_visitor::split_unpack_half_2x16
1390 ir_unop_unpack_half_2x16_split_x
,
1391 ir_unop_unpack_half_2x16_split_y
,
1395 * \name Bit operations, part of ARB_gpu_shader5.
1398 ir_unop_bitfield_reverse
,
1407 * \name Double packing, part of ARB_gpu_shader_fp64.
1410 ir_unop_pack_double_2x32
,
1411 ir_unop_unpack_double_2x32
,
1419 ir_unop_subroutine_to_int
,
1421 * Interpolate fs input at centroid
1423 * operand0 is the fs input.
1425 ir_unop_interpolate_at_centroid
,
1428 * Ask the driver for the total size of a buffer block.
1430 * operand0 is the ir_constant buffer block index in the linked shader.
1432 ir_unop_get_buffer_size
,
1435 * Calculate length of an unsized array inside a buffer block.
1436 * This opcode is going to be replaced in a lowering pass inside
1439 * operand0 is the unsized array's ir_value for the calculation
1442 ir_unop_ssbo_unsized_array_length
,
1445 * A sentinel marking the last of the unary operations.
1447 ir_last_unop
= ir_unop_ssbo_unsized_array_length
,
1451 ir_binop_mul
, /**< Floating-point or low 32-bit integer multiply. */
1452 ir_binop_imul_high
, /**< Calculates the high 32-bits of a 64-bit multiply. */
1456 * Returns the carry resulting from the addition of the two arguments.
1463 * Returns the borrow resulting from the subtraction of the second argument
1464 * from the first argument.
1471 * Takes one of two combinations of arguments:
1474 * - mod(vecN, float)
1476 * Does not take integer types.
1481 * \name Binary comparison operators which return a boolean vector.
1482 * The type of both operands must be equal.
1492 * Returns single boolean for whether all components of operands[0]
1493 * equal the components of operands[1].
1497 * Returns single boolean for whether any component of operands[0]
1498 * is not equal to the corresponding component of operands[1].
1500 ir_binop_any_nequal
,
1504 * \name Bit-wise binary operations.
1525 * \name Lowered floating point packing operations.
1527 * \see lower_packing_builtins_visitor::split_pack_half_2x16
1530 ir_binop_pack_half_2x16_split
,
1534 * \name First half of a lowered bitfieldInsert() operation.
1536 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1543 * Load a value the size of a given GLSL type from a uniform block.
1545 * operand0 is the ir_constant uniform block index in the linked shader.
1546 * operand1 is a byte offset within the uniform block.
1551 * \name Multiplies a number by two to a power, part of ARB_gpu_shader5.
1558 * Extract a scalar from a vector
1560 * operand0 is the vector
1561 * operand1 is the index of the field to read from operand0
1563 ir_binop_vector_extract
,
1566 * Interpolate fs input at offset
1568 * operand0 is the fs input
1569 * operand1 is the offset from the pixel center
1571 ir_binop_interpolate_at_offset
,
1574 * Interpolate fs input at sample position
1576 * operand0 is the fs input
1577 * operand1 is the sample ID
1579 ir_binop_interpolate_at_sample
,
1582 * A sentinel marking the last of the binary operations.
1584 ir_last_binop
= ir_binop_interpolate_at_sample
,
1587 * \name Fused floating-point multiply-add, part of ARB_gpu_shader5.
1596 * \name Conditional Select
1598 * A vector conditional select instruction (like ?:, but operating per-
1599 * component on vectors).
1601 * \see lower_instructions_visitor::ldexp_to_arith
1608 * \name Second half of a lowered bitfieldInsert() operation.
1610 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1616 ir_triop_bitfield_extract
,
1619 * Generate a value with one field of a vector changed
1621 * operand0 is the vector
1622 * operand1 is the value to write into the vector result
1623 * operand2 is the index in operand0 to be modified
1625 ir_triop_vector_insert
,
1628 * A sentinel marking the last of the ternary operations.
1630 ir_last_triop
= ir_triop_vector_insert
,
1632 ir_quadop_bitfield_insert
,
1637 * A sentinel marking the last of the ternary operations.
1639 ir_last_quadop
= ir_quadop_vector
,
1642 * A sentinel marking the last of all operations.
1644 ir_last_opcode
= ir_quadop_vector
1647 class ir_expression
: public ir_rvalue
{
1649 ir_expression(int op
, const struct glsl_type
*type
,
1650 ir_rvalue
*op0
, ir_rvalue
*op1
= NULL
,
1651 ir_rvalue
*op2
= NULL
, ir_rvalue
*op3
= NULL
);
1654 * Constructor for unary operation expressions
1656 ir_expression(int op
, ir_rvalue
*);
1659 * Constructor for binary operation expressions
1661 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
);
1664 * Constructor for ternary operation expressions
1666 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
, ir_rvalue
*op2
);
1668 virtual bool equals(const ir_instruction
*ir
,
1669 enum ir_node_type ignore
= ir_type_unset
) const;
1671 virtual ir_expression
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1674 * Attempt to constant-fold the expression
1676 * The "variable_context" hash table links ir_variable * to ir_constant *
1677 * that represent the variables' values. \c NULL represents an empty
1680 * If the expression cannot be constant folded, this method will return
1683 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1686 * Determine the number of operands used by an expression
1688 static unsigned int get_num_operands(ir_expression_operation
);
1691 * Determine the number of operands used by an expression
1693 unsigned int get_num_operands() const
1695 return (this->operation
== ir_quadop_vector
)
1696 ? this->type
->vector_elements
: get_num_operands(operation
);
1700 * Return whether the expression operates on vectors horizontally.
1702 bool is_horizontal() const
1704 return operation
== ir_binop_all_equal
||
1705 operation
== ir_binop_any_nequal
||
1706 operation
== ir_unop_any
||
1707 operation
== ir_binop_dot
||
1708 operation
== ir_quadop_vector
;
1712 * Return a string representing this expression's operator.
1714 const char *operator_string();
1717 * Return a string representing this expression's operator.
1719 static const char *operator_string(ir_expression_operation
);
1723 * Do a reverse-lookup to translate the given string into an operator.
1725 static ir_expression_operation
get_operator(const char *);
1727 virtual void accept(ir_visitor
*v
)
1732 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1734 virtual ir_variable
*variable_referenced() const;
1736 ir_expression_operation operation
;
1737 ir_rvalue
*operands
[4];
1742 * HIR instruction representing a high-level function call, containing a list
1743 * of parameters and returning a value in the supplied temporary.
1745 class ir_call
: public ir_instruction
{
1747 ir_call(ir_function_signature
*callee
,
1748 ir_dereference_variable
*return_deref
,
1749 exec_list
*actual_parameters
)
1750 : ir_instruction(ir_type_call
), return_deref(return_deref
), callee(callee
), sub_var(NULL
), array_idx(NULL
)
1752 assert(callee
->return_type
!= NULL
);
1753 actual_parameters
->move_nodes_to(& this->actual_parameters
);
1754 this->use_builtin
= callee
->is_builtin();
1757 ir_call(ir_function_signature
*callee
,
1758 ir_dereference_variable
*return_deref
,
1759 exec_list
*actual_parameters
,
1760 ir_variable
*var
, ir_rvalue
*array_idx
)
1761 : ir_instruction(ir_type_call
), return_deref(return_deref
), callee(callee
), sub_var(var
), array_idx(array_idx
)
1763 assert(callee
->return_type
!= NULL
);
1764 actual_parameters
->move_nodes_to(& this->actual_parameters
);
1765 this->use_builtin
= callee
->is_builtin();
1768 virtual ir_call
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1770 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1772 virtual void accept(ir_visitor
*v
)
1777 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1780 * Get the name of the function being called.
1782 const char *callee_name() const
1784 return callee
->function_name();
1788 * Generates an inline version of the function before @ir,
1789 * storing the return value in return_deref.
1791 void generate_inline(ir_instruction
*ir
);
1794 * Storage for the function's return value.
1795 * This must be NULL if the return type is void.
1797 ir_dereference_variable
*return_deref
;
1800 * The specific function signature being called.
1802 ir_function_signature
*callee
;
1804 /* List of ir_rvalue of paramaters passed in this call. */
1805 exec_list actual_parameters
;
1807 /** Should this call only bind to a built-in function? */
1811 * ARB_shader_subroutine support -
1812 * the subroutine uniform variable and array index
1813 * rvalue to be used in the lowering pass later.
1815 ir_variable
*sub_var
;
1816 ir_rvalue
*array_idx
;
1821 * \name Jump-like IR instructions.
1823 * These include \c break, \c continue, \c return, and \c discard.
1826 class ir_jump
: public ir_instruction
{
1828 ir_jump(enum ir_node_type t
)
1834 class ir_return
: public ir_jump
{
1837 : ir_jump(ir_type_return
), value(NULL
)
1841 ir_return(ir_rvalue
*value
)
1842 : ir_jump(ir_type_return
), value(value
)
1846 virtual ir_return
*clone(void *mem_ctx
, struct hash_table
*) const;
1848 ir_rvalue
*get_value() const
1853 virtual void accept(ir_visitor
*v
)
1858 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1865 * Jump instructions used inside loops
1867 * These include \c break and \c continue. The \c break within a loop is
1868 * different from the \c break within a switch-statement.
1870 * \sa ir_switch_jump
1872 class ir_loop_jump
: public ir_jump
{
1879 ir_loop_jump(jump_mode mode
)
1880 : ir_jump(ir_type_loop_jump
)
1885 virtual ir_loop_jump
*clone(void *mem_ctx
, struct hash_table
*) const;
1887 virtual void accept(ir_visitor
*v
)
1892 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1894 bool is_break() const
1896 return mode
== jump_break
;
1899 bool is_continue() const
1901 return mode
== jump_continue
;
1904 /** Mode selector for the jump instruction. */
1905 enum jump_mode mode
;
1909 * IR instruction representing discard statements.
1911 class ir_discard
: public ir_jump
{
1914 : ir_jump(ir_type_discard
)
1916 this->condition
= NULL
;
1919 ir_discard(ir_rvalue
*cond
)
1920 : ir_jump(ir_type_discard
)
1922 this->condition
= cond
;
1925 virtual ir_discard
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1927 virtual void accept(ir_visitor
*v
)
1932 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1934 ir_rvalue
*condition
;
1940 * Texture sampling opcodes used in ir_texture
1942 enum ir_texture_opcode
{
1943 ir_tex
, /**< Regular texture look-up */
1944 ir_txb
, /**< Texture look-up with LOD bias */
1945 ir_txl
, /**< Texture look-up with explicit LOD */
1946 ir_txd
, /**< Texture look-up with partial derivatvies */
1947 ir_txf
, /**< Texel fetch with explicit LOD */
1948 ir_txf_ms
, /**< Multisample texture fetch */
1949 ir_txs
, /**< Texture size */
1950 ir_lod
, /**< Texture lod query */
1951 ir_tg4
, /**< Texture gather */
1952 ir_query_levels
, /**< Texture levels query */
1953 ir_texture_samples
, /**< Texture samples query */
1958 * IR instruction to sample a texture
1960 * The specific form of the IR instruction depends on the \c mode value
1961 * selected from \c ir_texture_opcodes. In the printed IR, these will
1964 * Texel offset (0 or an expression)
1965 * | Projection divisor
1966 * | | Shadow comparitor
1969 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1970 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1971 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1972 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1973 * (txf <type> <sampler> <coordinate> 0 <lod>)
1975 * <type> <sampler> <coordinate> <sample_index>)
1976 * (txs <type> <sampler> <lod>)
1977 * (lod <type> <sampler> <coordinate>)
1978 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
1979 * (query_levels <type> <sampler>)
1981 class ir_texture
: public ir_rvalue
{
1983 ir_texture(enum ir_texture_opcode op
)
1984 : ir_rvalue(ir_type_texture
),
1985 op(op
), sampler(NULL
), coordinate(NULL
), projector(NULL
),
1986 shadow_comparitor(NULL
), offset(NULL
)
1988 memset(&lod_info
, 0, sizeof(lod_info
));
1991 virtual ir_texture
*clone(void *mem_ctx
, struct hash_table
*) const;
1993 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1995 virtual void accept(ir_visitor
*v
)
2000 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2002 virtual bool equals(const ir_instruction
*ir
,
2003 enum ir_node_type ignore
= ir_type_unset
) const;
2006 * Return a string representing the ir_texture_opcode.
2008 const char *opcode_string();
2010 /** Set the sampler and type. */
2011 void set_sampler(ir_dereference
*sampler
, const glsl_type
*type
);
2014 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
2016 static ir_texture_opcode
get_opcode(const char *);
2018 enum ir_texture_opcode op
;
2020 /** Sampler to use for the texture access. */
2021 ir_dereference
*sampler
;
2023 /** Texture coordinate to sample */
2024 ir_rvalue
*coordinate
;
2027 * Value used for projective divide.
2029 * If there is no projective divide (the common case), this will be
2030 * \c NULL. Optimization passes should check for this to point to a constant
2031 * of 1.0 and replace that with \c NULL.
2033 ir_rvalue
*projector
;
2036 * Coordinate used for comparison on shadow look-ups.
2038 * If there is no shadow comparison, this will be \c NULL. For the
2039 * \c ir_txf opcode, this *must* be \c NULL.
2041 ir_rvalue
*shadow_comparitor
;
2043 /** Texel offset. */
2047 ir_rvalue
*lod
; /**< Floating point LOD */
2048 ir_rvalue
*bias
; /**< Floating point LOD bias */
2049 ir_rvalue
*sample_index
; /**< MSAA sample index */
2050 ir_rvalue
*component
; /**< Gather component selector */
2052 ir_rvalue
*dPdx
; /**< Partial derivative of coordinate wrt X */
2053 ir_rvalue
*dPdy
; /**< Partial derivative of coordinate wrt Y */
2059 struct ir_swizzle_mask
{
2066 * Number of components in the swizzle.
2068 unsigned num_components
:3;
2071 * Does the swizzle contain duplicate components?
2073 * L-value swizzles cannot contain duplicate components.
2075 unsigned has_duplicates
:1;
2079 class ir_swizzle
: public ir_rvalue
{
2081 ir_swizzle(ir_rvalue
*, unsigned x
, unsigned y
, unsigned z
, unsigned w
,
2084 ir_swizzle(ir_rvalue
*val
, const unsigned *components
, unsigned count
);
2086 ir_swizzle(ir_rvalue
*val
, ir_swizzle_mask mask
);
2088 virtual ir_swizzle
*clone(void *mem_ctx
, struct hash_table
*) const;
2090 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2093 * Construct an ir_swizzle from the textual representation. Can fail.
2095 static ir_swizzle
*create(ir_rvalue
*, const char *, unsigned vector_length
);
2097 virtual void accept(ir_visitor
*v
)
2102 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2104 virtual bool equals(const ir_instruction
*ir
,
2105 enum ir_node_type ignore
= ir_type_unset
) const;
2107 bool is_lvalue() const
2109 return val
->is_lvalue() && !mask
.has_duplicates
;
2113 * Get the variable that is ultimately referenced by an r-value
2115 virtual ir_variable
*variable_referenced() const;
2118 ir_swizzle_mask mask
;
2122 * Initialize the mask component of a swizzle
2124 * This is used by the \c ir_swizzle constructors.
2126 void init_mask(const unsigned *components
, unsigned count
);
2130 class ir_dereference
: public ir_rvalue
{
2132 virtual ir_dereference
*clone(void *mem_ctx
, struct hash_table
*) const = 0;
2134 bool is_lvalue() const;
2137 * Get the variable that is ultimately referenced by an r-value
2139 virtual ir_variable
*variable_referenced() const = 0;
2142 ir_dereference(enum ir_node_type t
)
2149 class ir_dereference_variable
: public ir_dereference
{
2151 ir_dereference_variable(ir_variable
*var
);
2153 virtual ir_dereference_variable
*clone(void *mem_ctx
,
2154 struct hash_table
*) const;
2156 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2158 virtual bool equals(const ir_instruction
*ir
,
2159 enum ir_node_type ignore
= ir_type_unset
) const;
2162 * Get the variable that is ultimately referenced by an r-value
2164 virtual ir_variable
*variable_referenced() const
2169 virtual ir_variable
*whole_variable_referenced()
2171 /* ir_dereference_variable objects always dereference the entire
2172 * variable. However, if this dereference is dereferenced by anything
2173 * else, the complete deferefernce chain is not a whole-variable
2174 * dereference. This method should only be called on the top most
2175 * ir_rvalue in a dereference chain.
2180 virtual void accept(ir_visitor
*v
)
2185 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2188 * Object being dereferenced.
2194 class ir_dereference_array
: public ir_dereference
{
2196 ir_dereference_array(ir_rvalue
*value
, ir_rvalue
*array_index
);
2198 ir_dereference_array(ir_variable
*var
, ir_rvalue
*array_index
);
2200 virtual ir_dereference_array
*clone(void *mem_ctx
,
2201 struct hash_table
*) const;
2203 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2205 virtual bool equals(const ir_instruction
*ir
,
2206 enum ir_node_type ignore
= ir_type_unset
) const;
2209 * Get the variable that is ultimately referenced by an r-value
2211 virtual ir_variable
*variable_referenced() const
2213 return this->array
->variable_referenced();
2216 virtual void accept(ir_visitor
*v
)
2221 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2224 ir_rvalue
*array_index
;
2227 void set_array(ir_rvalue
*value
);
2231 class ir_dereference_record
: public ir_dereference
{
2233 ir_dereference_record(ir_rvalue
*value
, const char *field
);
2235 ir_dereference_record(ir_variable
*var
, const char *field
);
2237 virtual ir_dereference_record
*clone(void *mem_ctx
,
2238 struct hash_table
*) const;
2240 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2243 * Get the variable that is ultimately referenced by an r-value
2245 virtual ir_variable
*variable_referenced() const
2247 return this->record
->variable_referenced();
2250 virtual void accept(ir_visitor
*v
)
2255 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2263 * Data stored in an ir_constant
2265 union ir_constant_data
{
2274 class ir_constant
: public ir_rvalue
{
2276 ir_constant(const struct glsl_type
*type
, const ir_constant_data
*data
);
2277 ir_constant(bool b
, unsigned vector_elements
=1);
2278 ir_constant(unsigned int u
, unsigned vector_elements
=1);
2279 ir_constant(int i
, unsigned vector_elements
=1);
2280 ir_constant(float f
, unsigned vector_elements
=1);
2281 ir_constant(double d
, unsigned vector_elements
=1);
2284 * Construct an ir_constant from a list of ir_constant values
2286 ir_constant(const struct glsl_type
*type
, exec_list
*values
);
2289 * Construct an ir_constant from a scalar component of another ir_constant
2291 * The new \c ir_constant inherits the type of the component from the
2295 * In the case of a matrix constant, the new constant is a scalar, \b not
2298 ir_constant(const ir_constant
*c
, unsigned i
);
2301 * Return a new ir_constant of the specified type containing all zeros.
2303 static ir_constant
*zero(void *mem_ctx
, const glsl_type
*type
);
2305 virtual ir_constant
*clone(void *mem_ctx
, struct hash_table
*) const;
2307 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2309 virtual void accept(ir_visitor
*v
)
2314 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2316 virtual bool equals(const ir_instruction
*ir
,
2317 enum ir_node_type ignore
= ir_type_unset
) const;
2320 * Get a particular component of a constant as a specific type
2322 * This is useful, for example, to get a value from an integer constant
2323 * as a float or bool. This appears frequently when constructors are
2324 * called with all constant parameters.
2327 bool get_bool_component(unsigned i
) const;
2328 float get_float_component(unsigned i
) const;
2329 double get_double_component(unsigned i
) const;
2330 int get_int_component(unsigned i
) const;
2331 unsigned get_uint_component(unsigned i
) const;
2334 ir_constant
*get_array_element(unsigned i
) const;
2336 ir_constant
*get_record_field(const char *name
);
2339 * Copy the values on another constant at a given offset.
2341 * The offset is ignored for array or struct copies, it's only for
2342 * scalars or vectors into vectors or matrices.
2344 * With identical types on both sides and zero offset it's clone()
2345 * without creating a new object.
2348 void copy_offset(ir_constant
*src
, int offset
);
2351 * Copy the values on another constant at a given offset and
2352 * following an assign-like mask.
2354 * The mask is ignored for scalars.
2356 * Note that this function only handles what assign can handle,
2357 * i.e. at most a vector as source and a column of a matrix as
2361 void copy_masked_offset(ir_constant
*src
, int offset
, unsigned int mask
);
2364 * Determine whether a constant has the same value as another constant
2366 * \sa ir_constant::is_zero, ir_constant::is_one,
2367 * ir_constant::is_negative_one
2369 bool has_value(const ir_constant
*) const;
2372 * Return true if this ir_constant represents the given value.
2374 * For vectors, this checks that each component is the given value.
2376 virtual bool is_value(float f
, int i
) const;
2377 virtual bool is_zero() const;
2378 virtual bool is_one() const;
2379 virtual bool is_negative_one() const;
2382 * Return true for constants that could be stored as 16-bit unsigned values.
2384 * Note that this will return true even for signed integer ir_constants, as
2385 * long as the value is non-negative and fits in 16-bits.
2387 virtual bool is_uint16_constant() const;
2390 * Value of the constant.
2392 * The field used to back the values supplied by the constant is determined
2393 * by the type associated with the \c ir_instruction. Constants may be
2394 * scalars, vectors, or matrices.
2396 union ir_constant_data value
;
2398 /* Array elements */
2399 ir_constant
**array_elements
;
2401 /* Structure fields */
2402 exec_list components
;
2406 * Parameterless constructor only used by the clone method
2412 * IR instruction to emit a vertex in a geometry shader.
2414 class ir_emit_vertex
: public ir_instruction
{
2416 ir_emit_vertex(ir_rvalue
*stream
)
2417 : ir_instruction(ir_type_emit_vertex
),
2423 virtual void accept(ir_visitor
*v
)
2428 virtual ir_emit_vertex
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2430 return new(mem_ctx
) ir_emit_vertex(this->stream
->clone(mem_ctx
, ht
));
2433 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2435 int stream_id() const
2437 return stream
->as_constant()->value
.i
[0];
2444 * IR instruction to complete the current primitive and start a new one in a
2447 class ir_end_primitive
: public ir_instruction
{
2449 ir_end_primitive(ir_rvalue
*stream
)
2450 : ir_instruction(ir_type_end_primitive
),
2456 virtual void accept(ir_visitor
*v
)
2461 virtual ir_end_primitive
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2463 return new(mem_ctx
) ir_end_primitive(this->stream
->clone(mem_ctx
, ht
));
2466 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2468 int stream_id() const
2470 return stream
->as_constant()->value
.i
[0];
2477 * IR instruction for tessellation control and compute shader barrier.
2479 class ir_barrier
: public ir_instruction
{
2482 : ir_instruction(ir_type_barrier
)
2486 virtual void accept(ir_visitor
*v
)
2491 virtual ir_barrier
*clone(void *mem_ctx
, struct hash_table
*) const
2493 return new(mem_ctx
) ir_barrier();
2496 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2502 * Apply a visitor to each IR node in a list
2505 visit_exec_list(exec_list
*list
, ir_visitor
*visitor
);
2508 * Validate invariants on each IR node in a list
2510 void validate_ir_tree(exec_list
*instructions
);
2512 struct _mesa_glsl_parse_state
;
2513 struct gl_shader_program
;
2516 * Detect whether an unlinked shader contains static recursion
2518 * If the list of instructions is determined to contain static recursion,
2519 * \c _mesa_glsl_error will be called to emit error messages for each function
2520 * that is in the recursion cycle.
2523 detect_recursion_unlinked(struct _mesa_glsl_parse_state
*state
,
2524 exec_list
*instructions
);
2527 * Detect whether a linked shader contains static recursion
2529 * If the list of instructions is determined to contain static recursion,
2530 * \c link_error_printf will be called to emit error messages for each function
2531 * that is in the recursion cycle. In addition,
2532 * \c gl_shader_program::LinkStatus will be set to false.
2535 detect_recursion_linked(struct gl_shader_program
*prog
,
2536 exec_list
*instructions
);
2539 * Make a clone of each IR instruction in a list
2541 * \param in List of IR instructions that are to be cloned
2542 * \param out List to hold the cloned instructions
2545 clone_ir_list(void *mem_ctx
, exec_list
*out
, const exec_list
*in
);
2548 _mesa_glsl_initialize_variables(exec_list
*instructions
,
2549 struct _mesa_glsl_parse_state
*state
);
2552 _mesa_glsl_initialize_derived_variables(gl_shader
*shader
);
2555 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state
*state
);
2558 _mesa_glsl_initialize_builtin_functions();
2560 extern ir_function_signature
*
2561 _mesa_glsl_find_builtin_function(_mesa_glsl_parse_state
*state
,
2562 const char *name
, exec_list
*actual_parameters
);
2564 extern ir_function
*
2565 _mesa_glsl_find_builtin_function_by_name(const char *name
);
2568 _mesa_glsl_get_builtin_function_shader(void);
2570 extern ir_function_signature
*
2571 _mesa_get_main_function_signature(gl_shader
*sh
);
2574 _mesa_glsl_release_functions(void);
2577 _mesa_glsl_release_builtin_functions(void);
2580 reparent_ir(exec_list
*list
, void *mem_ctx
);
2582 struct glsl_symbol_table
;
2585 import_prototypes(const exec_list
*source
, exec_list
*dest
,
2586 struct glsl_symbol_table
*symbols
, void *mem_ctx
);
2589 ir_has_call(ir_instruction
*ir
);
2592 do_set_program_inouts(exec_list
*instructions
, struct gl_program
*prog
,
2593 gl_shader_stage shader_stage
);
2596 prototype_string(const glsl_type
*return_type
, const char *name
,
2597 exec_list
*parameters
);
2600 mode_string(const ir_variable
*var
);
2603 * Built-in / reserved GL variables names start with "gl_"
2606 is_gl_identifier(const char *s
)
2608 return s
&& s
[0] == 'g' && s
[1] == 'l' && s
[2] == '_';
2612 #endif /* __cplusplus */
2614 extern void _mesa_print_ir(FILE *f
, struct exec_list
*instructions
,
2615 struct _mesa_glsl_parse_state
*state
);
2618 fprint_ir(FILE *f
, const void *instruction
);
2625 vertices_per_prim(GLenum prim
);