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. */
331 ir_var_function_inout
,
332 ir_var_const_in
, /**< "in" param that must be a constant expression */
333 ir_var_system_value
, /**< Ex: front-face, instance-id, etc. */
334 ir_var_temporary
, /**< Temporary variable generated during compilation. */
335 ir_var_mode_count
/**< Number of variable modes */
339 * Enum keeping track of how a variable was declared. For error checking of
340 * the gl_PerVertex redeclaration rules.
342 enum ir_var_declaration_type
{
344 * Normal declaration (for most variables, this means an explicit
345 * declaration. Exception: temporaries are always implicitly declared, but
346 * they still use ir_var_declared_normally).
348 * Note: an ir_variable that represents a named interface block uses
349 * ir_var_declared_normally.
351 ir_var_declared_normally
= 0,
354 * Variable was explicitly declared (or re-declared) in an unnamed
357 ir_var_declared_in_block
,
360 * Variable is an implicitly declared built-in that has not been explicitly
361 * re-declared by the shader.
363 ir_var_declared_implicitly
,
366 * Variable is implicitly generated by the compiler and should not be
367 * visible via the API.
373 * \brief Layout qualifiers for gl_FragDepth.
375 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
376 * with a layout qualifier.
378 enum ir_depth_layout
{
379 ir_depth_layout_none
, /**< No depth layout is specified. */
381 ir_depth_layout_greater
,
382 ir_depth_layout_less
,
383 ir_depth_layout_unchanged
387 * \brief Convert depth layout qualifier to string.
390 depth_layout_string(ir_depth_layout layout
);
393 * Description of built-in state associated with a uniform
395 * \sa ir_variable::state_slots
397 struct ir_state_slot
{
404 * Get the string value for an interpolation qualifier
406 * \return The string that would be used in a shader to specify \c
407 * mode will be returned.
409 * This function is used to generate error messages of the form "shader
410 * uses %s interpolation qualifier", so in the case where there is no
411 * interpolation qualifier, it returns "no".
413 * This function should only be used on a shader input or output variable.
415 const char *interpolation_string(unsigned interpolation
);
418 class ir_variable
: public ir_instruction
{
420 ir_variable(const struct glsl_type
*, const char *, ir_variable_mode
);
422 virtual ir_variable
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
424 virtual void accept(ir_visitor
*v
)
429 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
433 * Determine how this variable should be interpolated based on its
434 * interpolation qualifier (if present), whether it is gl_Color or
435 * gl_SecondaryColor, and whether flatshading is enabled in the current GL
438 * The return value will always be either INTERP_QUALIFIER_SMOOTH,
439 * INTERP_QUALIFIER_NOPERSPECTIVE, or INTERP_QUALIFIER_FLAT.
441 glsl_interp_qualifier
determine_interpolation_mode(bool flat_shade
);
444 * Determine whether or not a variable is part of a uniform block.
446 inline bool is_in_uniform_block() const
448 return this->data
.mode
== ir_var_uniform
&& this->interface_type
!= NULL
;
452 * Determine whether or not a variable is the declaration of an interface
455 * For the first declaration below, there will be an \c ir_variable named
456 * "instance" whose type and whose instance_type will be the same
457 * \cglsl_type. For the second declaration, there will be an \c ir_variable
458 * named "f" whose type is float and whose instance_type is B2.
460 * "instance" is an interface instance variable, but "f" is not.
470 inline bool is_interface_instance() const
472 return this->type
->without_array() == this->interface_type
;
476 * Set this->interface_type on a newly created variable.
478 void init_interface_type(const struct glsl_type
*type
)
480 assert(this->interface_type
== NULL
);
481 this->interface_type
= type
;
482 if (this->is_interface_instance()) {
483 this->u
.max_ifc_array_access
=
484 rzalloc_array(this, unsigned, type
->length
);
489 * Change this->interface_type on a variable that previously had a
490 * different, but compatible, interface_type. This is used during linking
491 * to set the size of arrays in interface blocks.
493 void change_interface_type(const struct glsl_type
*type
)
495 if (this->u
.max_ifc_array_access
!= NULL
) {
496 /* max_ifc_array_access has already been allocated, so make sure the
497 * new interface has the same number of fields as the old one.
499 assert(this->interface_type
->length
== type
->length
);
501 this->interface_type
= type
;
505 * Change this->interface_type on a variable that previously had a
506 * different, and incompatible, interface_type. This is used during
507 * compilation to handle redeclaration of the built-in gl_PerVertex
510 void reinit_interface_type(const struct glsl_type
*type
)
512 if (this->u
.max_ifc_array_access
!= NULL
) {
514 /* Redeclaring gl_PerVertex is only allowed if none of the built-ins
515 * it defines have been accessed yet; so it's safe to throw away the
516 * old max_ifc_array_access pointer, since all of its values are
519 for (unsigned i
= 0; i
< this->interface_type
->length
; i
++)
520 assert(this->u
.max_ifc_array_access
[i
] == 0);
522 ralloc_free(this->u
.max_ifc_array_access
);
523 this->u
.max_ifc_array_access
= NULL
;
525 this->interface_type
= NULL
;
526 init_interface_type(type
);
529 const glsl_type
*get_interface_type() const
531 return this->interface_type
;
535 * Get the max_ifc_array_access pointer
537 * A "set" function is not needed because the array is dynmically allocated
540 inline unsigned *get_max_ifc_array_access()
542 assert(this->data
._num_state_slots
== 0);
543 return this->u
.max_ifc_array_access
;
546 inline unsigned get_num_state_slots() const
548 assert(!this->is_interface_instance()
549 || this->data
._num_state_slots
== 0);
550 return this->data
._num_state_slots
;
553 inline void set_num_state_slots(unsigned n
)
555 assert(!this->is_interface_instance()
557 this->data
._num_state_slots
= n
;
560 inline ir_state_slot
*get_state_slots()
562 return this->is_interface_instance() ? NULL
: this->u
.state_slots
;
565 inline const ir_state_slot
*get_state_slots() const
567 return this->is_interface_instance() ? NULL
: this->u
.state_slots
;
570 inline ir_state_slot
*allocate_state_slots(unsigned n
)
572 assert(!this->is_interface_instance());
574 this->u
.state_slots
= ralloc_array(this, ir_state_slot
, n
);
575 this->data
._num_state_slots
= 0;
577 if (this->u
.state_slots
!= NULL
)
578 this->data
._num_state_slots
= n
;
580 return this->u
.state_slots
;
583 inline bool is_name_ralloced() const
585 return this->name
!= ir_variable::tmp_name
;
589 * Enable emitting extension warnings for this variable
591 void enable_extension_warning(const char *extension
);
594 * Get the extension warning string for this variable
596 * If warnings are not enabled, \c NULL is returned.
598 const char *get_extension_warning() const;
601 * Declared type of the variable
603 const struct glsl_type
*type
;
606 * Declared name of the variable
610 struct ir_variable_data
{
613 * Is the variable read-only?
615 * This is set for variables declared as \c const, shader inputs,
618 unsigned read_only
:1;
621 unsigned invariant
:1;
625 * Has this variable been used for reading or writing?
627 * Several GLSL semantic checks require knowledge of whether or not a
628 * variable has been used. For example, it is an error to redeclare a
629 * variable as invariant after it has been used.
631 * This is only maintained in the ast_to_hir.cpp path, not in
632 * Mesa's fixed function or ARB program paths.
637 * Has this variable been statically assigned?
639 * This answers whether the variable was assigned in any path of
640 * the shader during ast_to_hir. This doesn't answer whether it is
641 * still written after dead code removal, nor is it maintained in
642 * non-ast_to_hir.cpp (GLSL parsing) paths.
647 * Enum indicating how the variable was declared. See
648 * ir_var_declaration_type.
650 * This is used to detect certain kinds of illegal variable redeclarations.
652 unsigned how_declared
:2;
655 * Storage class of the variable.
657 * \sa ir_variable_mode
662 * Interpolation mode for shader inputs / outputs
664 * \sa ir_variable_interpolation
666 unsigned interpolation
:2;
669 * \name ARB_fragment_coord_conventions
672 unsigned origin_upper_left
:1;
673 unsigned pixel_center_integer
:1;
677 * Was the location explicitly set in the shader?
679 * If the location is explicitly set in the shader, it \b cannot be changed
680 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
683 unsigned explicit_location
:1;
684 unsigned explicit_index
:1;
687 * Do we have a Vulkan (group, index) qualifier for this variable?
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
761 * This is now also used for the Vulkan descriptor set index.
766 * \brief Layout qualifier for gl_FragDepth.
768 * This is not equal to \c ir_depth_layout_none if and only if this
769 * variable is \c gl_FragDepth and a layout qualifier is specified.
771 ir_depth_layout depth_layout
:3;
774 * ARB_shader_image_load_store qualifiers.
776 unsigned image_read_only
:1; /**< "readonly" qualifier. */
777 unsigned image_write_only
:1; /**< "writeonly" qualifier. */
778 unsigned image_coherent
:1;
779 unsigned image_volatile
:1;
780 unsigned image_restrict
:1;
783 * Emit a warning if this variable is accessed.
786 uint8_t warn_extension_index
;
789 /** Image internal format if specified explicitly, otherwise GL_NONE. */
790 uint16_t image_format
;
794 * Number of state slots used
797 * This could be stored in as few as 7-bits, if necessary. If it is made
798 * smaller, add an assertion to \c ir_variable::allocate_state_slots to
801 uint16_t _num_state_slots
;
805 * Initial binding point for a sampler, atomic, or UBO.
807 * For array types, this represents the binding point for the first element.
812 * Vulkan descriptor set for the resource.
817 * Storage location of the base of this variable
819 * The precise meaning of this field depends on the nature of the variable.
821 * - Vertex shader input: one of the values from \c gl_vert_attrib.
822 * - Vertex shader output: one of the values from \c gl_varying_slot.
823 * - Geometry shader input: one of the values from \c gl_varying_slot.
824 * - Geometry shader output: one of the values from \c gl_varying_slot.
825 * - Fragment shader input: one of the values from \c gl_varying_slot.
826 * - Fragment shader output: one of the values from \c gl_frag_result.
827 * - Uniforms: Per-stage uniform slot number for default uniform block.
828 * - Uniforms: Index within the uniform block definition for UBO members.
829 * - Other: This field is not currently used.
831 * If the variable is a uniform, shader input, or shader output, and the
832 * slot has not been assigned, the value will be -1.
837 * Vertex stream output identifier.
842 * Location an atomic counter is stored at.
849 * Highest element accessed with a constant expression array index
851 * Not used for non-array variables.
853 unsigned max_array_access
;
856 * Allow (only) ir_variable direct access private members.
858 friend class ir_variable
;
862 * Value assigned in the initializer of a variable declared "const"
864 ir_constant
*constant_value
;
867 * Constant expression assigned in the initializer of the variable
870 * This field and \c ::constant_value are distinct. Even if the two fields
871 * refer to constants with the same value, they must point to separate
874 ir_constant
*constant_initializer
;
877 static const char *const warn_extension_table
[];
881 * For variables which satisfy the is_interface_instance() predicate,
882 * this points to an array of integers such that if the ith member of
883 * the interface block is an array, max_ifc_array_access[i] is the
884 * maximum array element of that member that has been accessed. If the
885 * ith member of the interface block is not an array,
886 * max_ifc_array_access[i] is unused.
888 * For variables whose type is not an interface block, this pointer is
891 unsigned *max_ifc_array_access
;
894 * Built-in state that backs this uniform
896 * Once set at variable creation, \c state_slots must remain invariant.
898 * If the variable is not a uniform, \c _num_state_slots will be zero
899 * and \c state_slots will be \c NULL.
901 ir_state_slot
*state_slots
;
905 * For variables that are in an interface block or are an instance of an
906 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
908 * \sa ir_variable::location
910 const glsl_type
*interface_type
;
913 * Name used for anonymous compiler temporaries
915 static const char tmp_name
[];
919 * Should the construct keep names for ir_var_temporary variables?
921 * When this global is false, names passed to the constructor for
922 * \c ir_var_temporary variables will be dropped. Instead, the variable will
923 * be named "compiler_temp". This name will be in static storage.
926 * \b NEVER change the mode of an \c ir_var_temporary.
929 * This variable is \b not thread-safe. It is global, \b not
930 * per-context. It begins life false. A context can, at some point, make
931 * it true. From that point on, it will be true forever. This should be
932 * okay since it will only be set true while debugging.
934 static bool temporaries_allocate_names
;
938 * A function that returns whether a built-in function is available in the
939 * current shading language (based on version, ES or desktop, and extensions).
941 typedef bool (*builtin_available_predicate
)(const _mesa_glsl_parse_state
*);
945 * The representation of a function instance; may be the full definition or
946 * simply a prototype.
948 class ir_function_signature
: public ir_instruction
{
949 /* An ir_function_signature will be part of the list of signatures in
953 ir_function_signature(const glsl_type
*return_type
,
954 builtin_available_predicate builtin_avail
= NULL
);
956 virtual ir_function_signature
*clone(void *mem_ctx
,
957 struct hash_table
*ht
) const;
958 ir_function_signature
*clone_prototype(void *mem_ctx
,
959 struct hash_table
*ht
) const;
961 virtual void accept(ir_visitor
*v
)
966 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
969 * Attempt to evaluate this function as a constant expression,
970 * given a list of the actual parameters and the variable context.
971 * Returns NULL for non-built-ins.
973 ir_constant
*constant_expression_value(exec_list
*actual_parameters
, struct hash_table
*variable_context
);
976 * Get the name of the function for which this is a signature
978 const char *function_name() const;
981 * Get a handle to the function for which this is a signature
983 * There is no setter function, this function returns a \c const pointer,
984 * and \c ir_function_signature::_function is private for a reason. The
985 * only way to make a connection between a function and function signature
986 * is via \c ir_function::add_signature. This helps ensure that certain
987 * invariants (i.e., a function signature is in the list of signatures for
988 * its \c _function) are met.
990 * \sa ir_function::add_signature
992 inline const class ir_function
*function() const
994 return this->_function
;
998 * Check whether the qualifiers match between this signature's parameters
999 * and the supplied parameter list. If not, returns the name of the first
1000 * parameter with mismatched qualifiers (for use in error messages).
1002 const char *qualifiers_match(exec_list
*params
);
1005 * Replace the current parameter list with the given one. This is useful
1006 * if the current information came from a prototype, and either has invalid
1007 * or missing parameter names.
1009 void replace_parameters(exec_list
*new_params
);
1012 * Function return type.
1014 * \note This discards the optional precision qualifier.
1016 const struct glsl_type
*return_type
;
1019 * List of ir_variable of function parameters.
1021 * This represents the storage. The paramaters passed in a particular
1022 * call will be in ir_call::actual_paramaters.
1024 struct exec_list parameters
;
1026 /** Whether or not this function has a body (which may be empty). */
1027 unsigned is_defined
:1;
1029 /** Whether or not this function signature is a built-in. */
1030 bool is_builtin() const;
1033 * Whether or not this function is an intrinsic to be implemented
1038 /** Whether or not a built-in is available for this shader. */
1039 bool is_builtin_available(const _mesa_glsl_parse_state
*state
) const;
1041 /** Body of instructions in the function. */
1042 struct exec_list body
;
1046 * A function pointer to a predicate that answers whether a built-in
1047 * function is available in the current shader. NULL if not a built-in.
1049 builtin_available_predicate builtin_avail
;
1051 /** Function of which this signature is one overload. */
1052 class ir_function
*_function
;
1054 /** Function signature of which this one is a prototype clone */
1055 const ir_function_signature
*origin
;
1057 friend class ir_function
;
1060 * Helper function to run a list of instructions for constant
1061 * expression evaluation.
1063 * The hash table represents the values of the visible variables.
1064 * There are no scoping issues because the table is indexed on
1065 * ir_variable pointers, not variable names.
1067 * Returns false if the expression is not constant, true otherwise,
1068 * and the value in *result if result is non-NULL.
1070 bool constant_expression_evaluate_expression_list(const struct exec_list
&body
,
1071 struct hash_table
*variable_context
,
1072 ir_constant
**result
);
1077 * Header for tracking multiple overloaded functions with the same name.
1078 * Contains a list of ir_function_signatures representing each of the
1081 class ir_function
: public ir_instruction
{
1083 ir_function(const char *name
);
1085 virtual ir_function
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1087 virtual void accept(ir_visitor
*v
)
1092 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1094 void add_signature(ir_function_signature
*sig
)
1096 sig
->_function
= this;
1097 this->signatures
.push_tail(sig
);
1101 * Find a signature that matches a set of actual parameters, taking implicit
1102 * conversions into account. Also flags whether the match was exact.
1104 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1105 const exec_list
*actual_param
,
1106 bool allow_builtins
,
1107 bool *match_is_exact
);
1110 * Find a signature that matches a set of actual parameters, taking implicit
1111 * conversions into account.
1113 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1114 const exec_list
*actual_param
,
1115 bool allow_builtins
);
1118 * Find a signature that exactly matches a set of actual parameters without
1119 * any implicit type conversions.
1121 ir_function_signature
*exact_matching_signature(_mesa_glsl_parse_state
*state
,
1122 const exec_list
*actual_ps
);
1125 * Name of the function.
1129 /** Whether or not this function has a signature that isn't a built-in. */
1130 bool has_user_signature();
1133 * List of ir_function_signature for each overloaded function with this name.
1135 struct exec_list signatures
;
1138 inline const char *ir_function_signature::function_name() const
1140 return this->_function
->name
;
1146 * IR instruction representing high-level if-statements
1148 class ir_if
: public ir_instruction
{
1150 ir_if(ir_rvalue
*condition
)
1151 : ir_instruction(ir_type_if
), condition(condition
)
1155 virtual ir_if
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1157 virtual void accept(ir_visitor
*v
)
1162 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1164 ir_rvalue
*condition
;
1165 /** List of ir_instruction for the body of the then branch */
1166 exec_list then_instructions
;
1167 /** List of ir_instruction for the body of the else branch */
1168 exec_list else_instructions
;
1173 * IR instruction representing a high-level loop structure.
1175 class ir_loop
: public ir_instruction
{
1179 virtual ir_loop
*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 /** List of ir_instruction that make up the body of the loop. */
1189 exec_list body_instructions
;
1193 class ir_assignment
: public ir_instruction
{
1195 ir_assignment(ir_rvalue
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
= NULL
);
1198 * Construct an assignment with an explicit write mask
1201 * Since a write mask is supplied, the LHS must already be a bare
1202 * \c ir_dereference. The cannot be any swizzles in the LHS.
1204 ir_assignment(ir_dereference
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
,
1205 unsigned write_mask
);
1207 virtual ir_assignment
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1209 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1211 virtual void accept(ir_visitor
*v
)
1216 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1219 * Get a whole variable written by an assignment
1221 * If the LHS of the assignment writes a whole variable, the variable is
1222 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1225 * - Assigning to a scalar
1226 * - Assigning to all components of a vector
1227 * - Whole array (or matrix) assignment
1228 * - Whole structure assignment
1230 ir_variable
*whole_variable_written();
1233 * Set the LHS of an assignment
1235 void set_lhs(ir_rvalue
*lhs
);
1238 * Left-hand side of the assignment.
1240 * This should be treated as read only. If you need to set the LHS of an
1241 * assignment, use \c ir_assignment::set_lhs.
1243 ir_dereference
*lhs
;
1246 * Value being assigned
1251 * Optional condition for the assignment.
1253 ir_rvalue
*condition
;
1257 * Component mask written
1259 * For non-vector types in the LHS, this field will be zero. For vector
1260 * types, a bit will be set for each component that is written. Note that
1261 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1263 * A partially-set write mask means that each enabled channel gets
1264 * the value from a consecutive channel of the rhs. For example,
1265 * to write just .xyw of gl_FrontColor with color:
1267 * (assign (constant bool (1)) (xyw)
1268 * (var_ref gl_FragColor)
1269 * (swiz xyw (var_ref color)))
1271 unsigned write_mask
:4;
1274 /* Update ir_expression::get_num_operands() and operator_strs when
1275 * updating this list.
1277 enum ir_expression_operation
{
1286 ir_unop_exp
, /**< Log base e on gentype */
1287 ir_unop_log
, /**< Natural log on gentype */
1290 ir_unop_f2i
, /**< Float-to-integer conversion. */
1291 ir_unop_f2u
, /**< Float-to-unsigned conversion. */
1292 ir_unop_i2f
, /**< Integer-to-float conversion. */
1293 ir_unop_f2b
, /**< Float-to-boolean conversion */
1294 ir_unop_b2f
, /**< Boolean-to-float conversion */
1295 ir_unop_i2b
, /**< int-to-boolean conversion */
1296 ir_unop_b2i
, /**< Boolean-to-int conversion */
1297 ir_unop_u2f
, /**< Unsigned-to-float conversion. */
1298 ir_unop_i2u
, /**< Integer-to-unsigned conversion. */
1299 ir_unop_u2i
, /**< Unsigned-to-integer conversion. */
1300 ir_unop_d2f
, /**< Double-to-float conversion. */
1301 ir_unop_f2d
, /**< Float-to-double conversion. */
1302 ir_unop_d2i
, /**< Double-to-integer conversion. */
1303 ir_unop_i2d
, /**< Integer-to-double conversion. */
1304 ir_unop_d2u
, /**< Double-to-unsigned conversion. */
1305 ir_unop_u2d
, /**< Unsigned-to-double conversion. */
1306 ir_unop_d2b
, /**< Double-to-boolean conversion. */
1307 ir_unop_bitcast_i2f
, /**< Bit-identical int-to-float "conversion" */
1308 ir_unop_bitcast_f2i
, /**< Bit-identical float-to-int "conversion" */
1309 ir_unop_bitcast_u2f
, /**< Bit-identical uint-to-float "conversion" */
1310 ir_unop_bitcast_f2u
, /**< Bit-identical float-to-uint "conversion" */
1314 * \name Unary floating-point rounding operations.
1325 * \name Trigonometric operations.
1333 * \name Partial derivatives.
1337 ir_unop_dFdx_coarse
,
1340 ir_unop_dFdy_coarse
,
1345 * \name Floating point pack and unpack operations.
1348 ir_unop_pack_snorm_2x16
,
1349 ir_unop_pack_snorm_4x8
,
1350 ir_unop_pack_unorm_2x16
,
1351 ir_unop_pack_unorm_4x8
,
1352 ir_unop_pack_half_2x16
,
1353 ir_unop_unpack_snorm_2x16
,
1354 ir_unop_unpack_snorm_4x8
,
1355 ir_unop_unpack_unorm_2x16
,
1356 ir_unop_unpack_unorm_4x8
,
1357 ir_unop_unpack_half_2x16
,
1361 * \name Lowered floating point unpacking operations.
1363 * \see lower_packing_builtins_visitor::split_unpack_half_2x16
1366 ir_unop_unpack_half_2x16_split_x
,
1367 ir_unop_unpack_half_2x16_split_y
,
1371 * \name Bit operations, part of ARB_gpu_shader5.
1374 ir_unop_bitfield_reverse
,
1383 * \name Double packing, part of ARB_gpu_shader_fp64.
1386 ir_unop_pack_double_2x32
,
1387 ir_unop_unpack_double_2x32
,
1396 * Interpolate fs input at centroid
1398 * operand0 is the fs input.
1400 ir_unop_interpolate_at_centroid
,
1403 * A sentinel marking the last of the unary operations.
1405 ir_last_unop
= ir_unop_interpolate_at_centroid
,
1409 ir_binop_mul
, /**< Floating-point or low 32-bit integer multiply. */
1410 ir_binop_imul_high
, /**< Calculates the high 32-bits of a 64-bit multiply. */
1414 * Returns the carry resulting from the addition of the two arguments.
1421 * Returns the borrow resulting from the subtraction of the second argument
1422 * from the first argument.
1429 * Takes one of two combinations of arguments:
1432 * - mod(vecN, float)
1434 * Does not take integer types.
1439 * \name Binary comparison operators which return a boolean vector.
1440 * The type of both operands must be equal.
1450 * Returns single boolean for whether all components of operands[0]
1451 * equal the components of operands[1].
1455 * Returns single boolean for whether any component of operands[0]
1456 * is not equal to the corresponding component of operands[1].
1458 ir_binop_any_nequal
,
1462 * \name Bit-wise binary operations.
1483 * \name Lowered floating point packing operations.
1485 * \see lower_packing_builtins_visitor::split_pack_half_2x16
1488 ir_binop_pack_half_2x16_split
,
1492 * \name First half of a lowered bitfieldInsert() operation.
1494 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1501 * Load a value the size of a given GLSL type from a uniform block.
1503 * operand0 is the ir_constant uniform block index in the linked shader.
1504 * operand1 is a byte offset within the uniform block.
1509 * \name Multiplies a number by two to a power, part of ARB_gpu_shader5.
1516 * Extract a scalar from a vector
1518 * operand0 is the vector
1519 * operand1 is the index of the field to read from operand0
1521 ir_binop_vector_extract
,
1524 * Interpolate fs input at offset
1526 * operand0 is the fs input
1527 * operand1 is the offset from the pixel center
1529 ir_binop_interpolate_at_offset
,
1532 * Interpolate fs input at sample position
1534 * operand0 is the fs input
1535 * operand1 is the sample ID
1537 ir_binop_interpolate_at_sample
,
1540 * A sentinel marking the last of the binary operations.
1542 ir_last_binop
= ir_binop_interpolate_at_sample
,
1545 * \name Fused floating-point multiply-add, part of ARB_gpu_shader5.
1554 * \name Conditional Select
1556 * A vector conditional select instruction (like ?:, but operating per-
1557 * component on vectors).
1559 * \see lower_instructions_visitor::ldexp_to_arith
1566 * \name Second half of a lowered bitfieldInsert() operation.
1568 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1574 ir_triop_bitfield_extract
,
1577 * Generate a value with one field of a vector changed
1579 * operand0 is the vector
1580 * operand1 is the value to write into the vector result
1581 * operand2 is the index in operand0 to be modified
1583 ir_triop_vector_insert
,
1586 * A sentinel marking the last of the ternary operations.
1588 ir_last_triop
= ir_triop_vector_insert
,
1590 ir_quadop_bitfield_insert
,
1595 * A sentinel marking the last of the ternary operations.
1597 ir_last_quadop
= ir_quadop_vector
,
1600 * A sentinel marking the last of all operations.
1602 ir_last_opcode
= ir_quadop_vector
1605 class ir_expression
: public ir_rvalue
{
1607 ir_expression(int op
, const struct glsl_type
*type
,
1608 ir_rvalue
*op0
, ir_rvalue
*op1
= NULL
,
1609 ir_rvalue
*op2
= NULL
, ir_rvalue
*op3
= NULL
);
1612 * Constructor for unary operation expressions
1614 ir_expression(int op
, ir_rvalue
*);
1617 * Constructor for binary operation expressions
1619 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
);
1622 * Constructor for ternary operation expressions
1624 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
, ir_rvalue
*op2
);
1626 virtual bool equals(const ir_instruction
*ir
,
1627 enum ir_node_type ignore
= ir_type_unset
) const;
1629 virtual ir_expression
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1632 * Attempt to constant-fold the expression
1634 * The "variable_context" hash table links ir_variable * to ir_constant *
1635 * that represent the variables' values. \c NULL represents an empty
1638 * If the expression cannot be constant folded, this method will return
1641 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1644 * Determine the number of operands used by an expression
1646 static unsigned int get_num_operands(ir_expression_operation
);
1649 * Determine the number of operands used by an expression
1651 unsigned int get_num_operands() const
1653 return (this->operation
== ir_quadop_vector
)
1654 ? this->type
->vector_elements
: get_num_operands(operation
);
1658 * Return whether the expression operates on vectors horizontally.
1660 bool is_horizontal() const
1662 return operation
== ir_binop_all_equal
||
1663 operation
== ir_binop_any_nequal
||
1664 operation
== ir_unop_any
||
1665 operation
== ir_binop_dot
||
1666 operation
== ir_quadop_vector
;
1670 * Return a string representing this expression's operator.
1672 const char *operator_string();
1675 * Return a string representing this expression's operator.
1677 static const char *operator_string(ir_expression_operation
);
1681 * Do a reverse-lookup to translate the given string into an operator.
1683 static ir_expression_operation
get_operator(const char *);
1685 virtual void accept(ir_visitor
*v
)
1690 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1692 ir_expression_operation operation
;
1693 ir_rvalue
*operands
[4];
1698 * HIR instruction representing a high-level function call, containing a list
1699 * of parameters and returning a value in the supplied temporary.
1701 class ir_call
: public ir_instruction
{
1703 ir_call(ir_function_signature
*callee
,
1704 ir_dereference_variable
*return_deref
,
1705 exec_list
*actual_parameters
)
1706 : ir_instruction(ir_type_call
), return_deref(return_deref
), callee(callee
)
1708 assert(callee
->return_type
!= NULL
);
1709 actual_parameters
->move_nodes_to(& this->actual_parameters
);
1710 this->use_builtin
= callee
->is_builtin();
1713 virtual ir_call
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1715 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1717 virtual void accept(ir_visitor
*v
)
1722 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1725 * Get the name of the function being called.
1727 const char *callee_name() const
1729 return callee
->function_name();
1733 * Generates an inline version of the function before @ir,
1734 * storing the return value in return_deref.
1736 void generate_inline(ir_instruction
*ir
);
1739 * Storage for the function's return value.
1740 * This must be NULL if the return type is void.
1742 ir_dereference_variable
*return_deref
;
1745 * The specific function signature being called.
1747 ir_function_signature
*callee
;
1749 /* List of ir_rvalue of paramaters passed in this call. */
1750 exec_list actual_parameters
;
1752 /** Should this call only bind to a built-in function? */
1758 * \name Jump-like IR instructions.
1760 * These include \c break, \c continue, \c return, and \c discard.
1763 class ir_jump
: public ir_instruction
{
1765 ir_jump(enum ir_node_type t
)
1771 class ir_return
: public ir_jump
{
1774 : ir_jump(ir_type_return
), value(NULL
)
1778 ir_return(ir_rvalue
*value
)
1779 : ir_jump(ir_type_return
), value(value
)
1783 virtual ir_return
*clone(void *mem_ctx
, struct hash_table
*) const;
1785 ir_rvalue
*get_value() const
1790 virtual void accept(ir_visitor
*v
)
1795 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1802 * Jump instructions used inside loops
1804 * These include \c break and \c continue. The \c break within a loop is
1805 * different from the \c break within a switch-statement.
1807 * \sa ir_switch_jump
1809 class ir_loop_jump
: public ir_jump
{
1816 ir_loop_jump(jump_mode mode
)
1817 : ir_jump(ir_type_loop_jump
)
1822 virtual ir_loop_jump
*clone(void *mem_ctx
, struct hash_table
*) const;
1824 virtual void accept(ir_visitor
*v
)
1829 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1831 bool is_break() const
1833 return mode
== jump_break
;
1836 bool is_continue() const
1838 return mode
== jump_continue
;
1841 /** Mode selector for the jump instruction. */
1842 enum jump_mode mode
;
1846 * IR instruction representing discard statements.
1848 class ir_discard
: public ir_jump
{
1851 : ir_jump(ir_type_discard
)
1853 this->condition
= NULL
;
1856 ir_discard(ir_rvalue
*cond
)
1857 : ir_jump(ir_type_discard
)
1859 this->condition
= cond
;
1862 virtual ir_discard
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1864 virtual void accept(ir_visitor
*v
)
1869 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1871 ir_rvalue
*condition
;
1877 * Texture sampling opcodes used in ir_texture
1879 enum ir_texture_opcode
{
1880 ir_tex
, /**< Regular texture look-up */
1881 ir_txb
, /**< Texture look-up with LOD bias */
1882 ir_txl
, /**< Texture look-up with explicit LOD */
1883 ir_txd
, /**< Texture look-up with partial derivatvies */
1884 ir_txf
, /**< Texel fetch with explicit LOD */
1885 ir_txf_ms
, /**< Multisample texture fetch */
1886 ir_txs
, /**< Texture size */
1887 ir_lod
, /**< Texture lod query */
1888 ir_tg4
, /**< Texture gather */
1889 ir_query_levels
/**< Texture levels query */
1894 * IR instruction to sample a texture
1896 * The specific form of the IR instruction depends on the \c mode value
1897 * selected from \c ir_texture_opcodes. In the printed IR, these will
1900 * Texel offset (0 or an expression)
1901 * | Projection divisor
1902 * | | Shadow comparitor
1905 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1906 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1907 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1908 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1909 * (txf <type> <sampler> <coordinate> 0 <lod>)
1911 * <type> <sampler> <coordinate> <sample_index>)
1912 * (txs <type> <sampler> <lod>)
1913 * (lod <type> <sampler> <coordinate>)
1914 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
1915 * (query_levels <type> <sampler>)
1917 class ir_texture
: public ir_rvalue
{
1919 ir_texture(enum ir_texture_opcode op
)
1920 : ir_rvalue(ir_type_texture
),
1921 op(op
), sampler(NULL
), coordinate(NULL
), projector(NULL
),
1922 shadow_comparitor(NULL
), offset(NULL
)
1924 memset(&lod_info
, 0, sizeof(lod_info
));
1927 virtual ir_texture
*clone(void *mem_ctx
, struct hash_table
*) const;
1929 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1931 virtual void accept(ir_visitor
*v
)
1936 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1938 virtual bool equals(const ir_instruction
*ir
,
1939 enum ir_node_type ignore
= ir_type_unset
) const;
1942 * Return a string representing the ir_texture_opcode.
1944 const char *opcode_string();
1946 /** Set the sampler and type. */
1947 void set_sampler(ir_dereference
*sampler
, const glsl_type
*type
);
1950 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1952 static ir_texture_opcode
get_opcode(const char *);
1954 enum ir_texture_opcode op
;
1956 /** Sampler to use for the texture access. */
1957 ir_dereference
*sampler
;
1959 /** Texture coordinate to sample */
1960 ir_rvalue
*coordinate
;
1963 * Value used for projective divide.
1965 * If there is no projective divide (the common case), this will be
1966 * \c NULL. Optimization passes should check for this to point to a constant
1967 * of 1.0 and replace that with \c NULL.
1969 ir_rvalue
*projector
;
1972 * Coordinate used for comparison on shadow look-ups.
1974 * If there is no shadow comparison, this will be \c NULL. For the
1975 * \c ir_txf opcode, this *must* be \c NULL.
1977 ir_rvalue
*shadow_comparitor
;
1979 /** Texel offset. */
1983 ir_rvalue
*lod
; /**< Floating point LOD */
1984 ir_rvalue
*bias
; /**< Floating point LOD bias */
1985 ir_rvalue
*sample_index
; /**< MSAA sample index */
1986 ir_rvalue
*component
; /**< Gather component selector */
1988 ir_rvalue
*dPdx
; /**< Partial derivative of coordinate wrt X */
1989 ir_rvalue
*dPdy
; /**< Partial derivative of coordinate wrt Y */
1995 struct ir_swizzle_mask
{
2002 * Number of components in the swizzle.
2004 unsigned num_components
:3;
2007 * Does the swizzle contain duplicate components?
2009 * L-value swizzles cannot contain duplicate components.
2011 unsigned has_duplicates
:1;
2015 class ir_swizzle
: public ir_rvalue
{
2017 ir_swizzle(ir_rvalue
*, unsigned x
, unsigned y
, unsigned z
, unsigned w
,
2020 ir_swizzle(ir_rvalue
*val
, const unsigned *components
, unsigned count
);
2022 ir_swizzle(ir_rvalue
*val
, ir_swizzle_mask mask
);
2024 virtual ir_swizzle
*clone(void *mem_ctx
, struct hash_table
*) const;
2026 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2029 * Construct an ir_swizzle from the textual representation. Can fail.
2031 static ir_swizzle
*create(ir_rvalue
*, const char *, unsigned vector_length
);
2033 virtual void accept(ir_visitor
*v
)
2038 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2040 virtual bool equals(const ir_instruction
*ir
,
2041 enum ir_node_type ignore
= ir_type_unset
) const;
2043 bool is_lvalue() const
2045 return val
->is_lvalue() && !mask
.has_duplicates
;
2049 * Get the variable that is ultimately referenced by an r-value
2051 virtual ir_variable
*variable_referenced() const;
2054 ir_swizzle_mask mask
;
2058 * Initialize the mask component of a swizzle
2060 * This is used by the \c ir_swizzle constructors.
2062 void init_mask(const unsigned *components
, unsigned count
);
2066 class ir_dereference
: public ir_rvalue
{
2068 virtual ir_dereference
*clone(void *mem_ctx
, struct hash_table
*) const = 0;
2070 bool is_lvalue() const;
2073 * Get the variable that is ultimately referenced by an r-value
2075 virtual ir_variable
*variable_referenced() const = 0;
2078 ir_dereference(enum ir_node_type t
)
2085 class ir_dereference_variable
: public ir_dereference
{
2087 ir_dereference_variable(ir_variable
*var
);
2089 virtual ir_dereference_variable
*clone(void *mem_ctx
,
2090 struct hash_table
*) const;
2092 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2094 virtual bool equals(const ir_instruction
*ir
,
2095 enum ir_node_type ignore
= ir_type_unset
) const;
2098 * Get the variable that is ultimately referenced by an r-value
2100 virtual ir_variable
*variable_referenced() const
2105 virtual ir_variable
*whole_variable_referenced()
2107 /* ir_dereference_variable objects always dereference the entire
2108 * variable. However, if this dereference is dereferenced by anything
2109 * else, the complete deferefernce chain is not a whole-variable
2110 * dereference. This method should only be called on the top most
2111 * ir_rvalue in a dereference chain.
2116 virtual void accept(ir_visitor
*v
)
2121 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2124 * Object being dereferenced.
2130 class ir_dereference_array
: public ir_dereference
{
2132 ir_dereference_array(ir_rvalue
*value
, ir_rvalue
*array_index
);
2134 ir_dereference_array(ir_variable
*var
, ir_rvalue
*array_index
);
2136 virtual ir_dereference_array
*clone(void *mem_ctx
,
2137 struct hash_table
*) const;
2139 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2141 virtual bool equals(const ir_instruction
*ir
,
2142 enum ir_node_type ignore
= ir_type_unset
) const;
2145 * Get the variable that is ultimately referenced by an r-value
2147 virtual ir_variable
*variable_referenced() const
2149 return this->array
->variable_referenced();
2152 virtual void accept(ir_visitor
*v
)
2157 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2160 ir_rvalue
*array_index
;
2163 void set_array(ir_rvalue
*value
);
2167 class ir_dereference_record
: public ir_dereference
{
2169 ir_dereference_record(ir_rvalue
*value
, const char *field
);
2171 ir_dereference_record(ir_variable
*var
, const char *field
);
2173 virtual ir_dereference_record
*clone(void *mem_ctx
,
2174 struct hash_table
*) const;
2176 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2179 * Get the variable that is ultimately referenced by an r-value
2181 virtual ir_variable
*variable_referenced() const
2183 return this->record
->variable_referenced();
2186 virtual void accept(ir_visitor
*v
)
2191 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2199 * Data stored in an ir_constant
2201 union ir_constant_data
{
2210 class ir_constant
: public ir_rvalue
{
2212 ir_constant(const struct glsl_type
*type
, const ir_constant_data
*data
);
2213 ir_constant(bool b
, unsigned vector_elements
=1);
2214 ir_constant(unsigned int u
, unsigned vector_elements
=1);
2215 ir_constant(int i
, unsigned vector_elements
=1);
2216 ir_constant(float f
, unsigned vector_elements
=1);
2217 ir_constant(double d
, unsigned vector_elements
=1);
2220 * Construct an ir_constant from a list of ir_constant values
2222 ir_constant(const struct glsl_type
*type
, exec_list
*values
);
2225 * Construct an ir_constant from a scalar component of another ir_constant
2227 * The new \c ir_constant inherits the type of the component from the
2231 * In the case of a matrix constant, the new constant is a scalar, \b not
2234 ir_constant(const ir_constant
*c
, unsigned i
);
2237 * Return a new ir_constant of the specified type containing all zeros.
2239 static ir_constant
*zero(void *mem_ctx
, const glsl_type
*type
);
2241 virtual ir_constant
*clone(void *mem_ctx
, struct hash_table
*) const;
2243 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2245 virtual void accept(ir_visitor
*v
)
2250 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2252 virtual bool equals(const ir_instruction
*ir
,
2253 enum ir_node_type ignore
= ir_type_unset
) const;
2256 * Get a particular component of a constant as a specific type
2258 * This is useful, for example, to get a value from an integer constant
2259 * as a float or bool. This appears frequently when constructors are
2260 * called with all constant parameters.
2263 bool get_bool_component(unsigned i
) const;
2264 float get_float_component(unsigned i
) const;
2265 double get_double_component(unsigned i
) const;
2266 int get_int_component(unsigned i
) const;
2267 unsigned get_uint_component(unsigned i
) const;
2270 ir_constant
*get_array_element(unsigned i
) const;
2272 ir_constant
*get_record_field(const char *name
);
2275 * Copy the values on another constant at a given offset.
2277 * The offset is ignored for array or struct copies, it's only for
2278 * scalars or vectors into vectors or matrices.
2280 * With identical types on both sides and zero offset it's clone()
2281 * without creating a new object.
2284 void copy_offset(ir_constant
*src
, int offset
);
2287 * Copy the values on another constant at a given offset and
2288 * following an assign-like mask.
2290 * The mask is ignored for scalars.
2292 * Note that this function only handles what assign can handle,
2293 * i.e. at most a vector as source and a column of a matrix as
2297 void copy_masked_offset(ir_constant
*src
, int offset
, unsigned int mask
);
2300 * Determine whether a constant has the same value as another constant
2302 * \sa ir_constant::is_zero, ir_constant::is_one,
2303 * ir_constant::is_negative_one
2305 bool has_value(const ir_constant
*) const;
2308 * Return true if this ir_constant represents the given value.
2310 * For vectors, this checks that each component is the given value.
2312 virtual bool is_value(float f
, int i
) const;
2313 virtual bool is_zero() const;
2314 virtual bool is_one() const;
2315 virtual bool is_negative_one() const;
2318 * Return true for constants that could be stored as 16-bit unsigned values.
2320 * Note that this will return true even for signed integer ir_constants, as
2321 * long as the value is non-negative and fits in 16-bits.
2323 virtual bool is_uint16_constant() const;
2326 * Value of the constant.
2328 * The field used to back the values supplied by the constant is determined
2329 * by the type associated with the \c ir_instruction. Constants may be
2330 * scalars, vectors, or matrices.
2332 union ir_constant_data value
;
2334 /* Array elements */
2335 ir_constant
**array_elements
;
2337 /* Structure fields */
2338 exec_list components
;
2342 * Parameterless constructor only used by the clone method
2348 * IR instruction to emit a vertex in a geometry shader.
2350 class ir_emit_vertex
: public ir_instruction
{
2352 ir_emit_vertex(ir_rvalue
*stream
)
2353 : ir_instruction(ir_type_emit_vertex
),
2359 virtual void accept(ir_visitor
*v
)
2364 virtual ir_emit_vertex
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2366 return new(mem_ctx
) ir_emit_vertex(this->stream
->clone(mem_ctx
, ht
));
2369 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2371 int stream_id() const
2373 return stream
->as_constant()->value
.i
[0];
2380 * IR instruction to complete the current primitive and start a new one in a
2383 class ir_end_primitive
: public ir_instruction
{
2385 ir_end_primitive(ir_rvalue
*stream
)
2386 : ir_instruction(ir_type_end_primitive
),
2392 virtual void accept(ir_visitor
*v
)
2397 virtual ir_end_primitive
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2399 return new(mem_ctx
) ir_end_primitive(this->stream
->clone(mem_ctx
, ht
));
2402 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2404 int stream_id() const
2406 return stream
->as_constant()->value
.i
[0];
2413 * IR instruction for tessellation control and compute shader barrier.
2415 class ir_barrier
: public ir_instruction
{
2418 : ir_instruction(ir_type_barrier
)
2422 virtual void accept(ir_visitor
*v
)
2427 virtual ir_barrier
*clone(void *mem_ctx
, struct hash_table
*) const
2429 return new(mem_ctx
) ir_barrier();
2432 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2438 * Apply a visitor to each IR node in a list
2441 visit_exec_list(exec_list
*list
, ir_visitor
*visitor
);
2444 * Validate invariants on each IR node in a list
2446 void validate_ir_tree(exec_list
*instructions
);
2448 struct _mesa_glsl_parse_state
;
2449 struct gl_shader_program
;
2452 * Detect whether an unlinked shader contains static recursion
2454 * If the list of instructions is determined to contain static recursion,
2455 * \c _mesa_glsl_error will be called to emit error messages for each function
2456 * that is in the recursion cycle.
2459 detect_recursion_unlinked(struct _mesa_glsl_parse_state
*state
,
2460 exec_list
*instructions
);
2463 * Detect whether a linked shader contains static recursion
2465 * If the list of instructions is determined to contain static recursion,
2466 * \c link_error_printf will be called to emit error messages for each function
2467 * that is in the recursion cycle. In addition,
2468 * \c gl_shader_program::LinkStatus will be set to false.
2471 detect_recursion_linked(struct gl_shader_program
*prog
,
2472 exec_list
*instructions
);
2475 * Make a clone of each IR instruction in a list
2477 * \param in List of IR instructions that are to be cloned
2478 * \param out List to hold the cloned instructions
2481 clone_ir_list(void *mem_ctx
, exec_list
*out
, const exec_list
*in
);
2484 _mesa_glsl_initialize_variables(exec_list
*instructions
,
2485 struct _mesa_glsl_parse_state
*state
);
2488 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state
*state
);
2491 _mesa_glsl_initialize_builtin_functions();
2493 extern ir_function_signature
*
2494 _mesa_glsl_find_builtin_function(_mesa_glsl_parse_state
*state
,
2495 const char *name
, exec_list
*actual_parameters
);
2497 extern ir_function
*
2498 _mesa_glsl_find_builtin_function_by_name(_mesa_glsl_parse_state
*state
,
2502 _mesa_glsl_get_builtin_function_shader(void);
2505 _mesa_glsl_release_functions(void);
2508 _mesa_glsl_release_builtin_functions(void);
2511 reparent_ir(exec_list
*list
, void *mem_ctx
);
2513 struct glsl_symbol_table
;
2516 import_prototypes(const exec_list
*source
, exec_list
*dest
,
2517 struct glsl_symbol_table
*symbols
, void *mem_ctx
);
2520 ir_has_call(ir_instruction
*ir
);
2523 do_set_program_inouts(exec_list
*instructions
, struct gl_program
*prog
,
2524 gl_shader_stage shader_stage
);
2527 prototype_string(const glsl_type
*return_type
, const char *name
,
2528 exec_list
*parameters
);
2531 mode_string(const ir_variable
*var
);
2534 * Built-in / reserved GL variables names start with "gl_"
2537 is_gl_identifier(const char *s
)
2539 return s
&& s
[0] == 'g' && s
[1] == 'l' && s
[2] == '_';
2543 #endif /* __cplusplus */
2545 extern void _mesa_print_ir(FILE *f
, struct exec_list
*instructions
,
2546 struct _mesa_glsl_parse_state
*state
);
2549 fprint_ir(FILE *f
, const void *instruction
);
2556 vertices_per_prim(GLenum prim
);