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. */
328 ir_var_shader_shared
, /**< Variable declared as shared. */
333 ir_var_function_inout
,
334 ir_var_const_in
, /**< "in" param that must be a constant expression */
335 ir_var_system_value
, /**< Ex: front-face, instance-id, etc. */
336 ir_var_temporary
, /**< Temporary variable generated during compilation. */
337 ir_var_mode_count
/**< Number of variable modes */
341 * Enum keeping track of how a variable was declared. For error checking of
342 * the gl_PerVertex redeclaration rules.
344 enum ir_var_declaration_type
{
346 * Normal declaration (for most variables, this means an explicit
347 * declaration. Exception: temporaries are always implicitly declared, but
348 * they still use ir_var_declared_normally).
350 * Note: an ir_variable that represents a named interface block uses
351 * ir_var_declared_normally.
353 ir_var_declared_normally
= 0,
356 * Variable was explicitly declared (or re-declared) in an unnamed
359 ir_var_declared_in_block
,
362 * Variable is an implicitly declared built-in that has not been explicitly
363 * re-declared by the shader.
365 ir_var_declared_implicitly
,
368 * Variable is implicitly generated by the compiler and should not be
369 * visible via the API.
375 * \brief Layout qualifiers for gl_FragDepth.
377 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
378 * with a layout qualifier.
380 enum ir_depth_layout
{
381 ir_depth_layout_none
, /**< No depth layout is specified. */
383 ir_depth_layout_greater
,
384 ir_depth_layout_less
,
385 ir_depth_layout_unchanged
389 * \brief Convert depth layout qualifier to string.
392 depth_layout_string(ir_depth_layout layout
);
395 * Description of built-in state associated with a uniform
397 * \sa ir_variable::state_slots
399 struct ir_state_slot
{
406 * Get the string value for an interpolation qualifier
408 * \return The string that would be used in a shader to specify \c
409 * mode will be returned.
411 * This function is used to generate error messages of the form "shader
412 * uses %s interpolation qualifier", so in the case where there is no
413 * interpolation qualifier, it returns "no".
415 * This function should only be used on a shader input or output variable.
417 const char *interpolation_string(unsigned interpolation
);
420 class ir_variable
: public ir_instruction
{
422 ir_variable(const struct glsl_type
*, const char *, ir_variable_mode
);
424 virtual ir_variable
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
426 virtual void accept(ir_visitor
*v
)
431 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
435 * Determine how this variable should be interpolated based on its
436 * interpolation qualifier (if present), whether it is gl_Color or
437 * gl_SecondaryColor, and whether flatshading is enabled in the current GL
440 * The return value will always be either INTERP_QUALIFIER_SMOOTH,
441 * INTERP_QUALIFIER_NOPERSPECTIVE, or INTERP_QUALIFIER_FLAT.
443 glsl_interp_qualifier
determine_interpolation_mode(bool flat_shade
);
446 * Determine whether or not a variable is part of a uniform or
447 * shader storage block.
449 inline bool is_in_buffer_block() const
451 return (this->data
.mode
== ir_var_uniform
||
452 this->data
.mode
== ir_var_shader_storage
) &&
453 this->interface_type
!= NULL
;
457 * Determine whether or not a variable is part of a shader storage block.
459 inline bool is_in_shader_storage_block() const
461 return this->data
.mode
== ir_var_shader_storage
&&
462 this->interface_type
!= NULL
;
466 * Determine whether or not a variable is the declaration of an interface
469 * For the first declaration below, there will be an \c ir_variable named
470 * "instance" whose type and whose instance_type will be the same
471 * \cglsl_type. For the second declaration, there will be an \c ir_variable
472 * named "f" whose type is float and whose instance_type is B2.
474 * "instance" is an interface instance variable, but "f" is not.
484 inline bool is_interface_instance() const
486 return this->type
->without_array() == this->interface_type
;
490 * Set this->interface_type on a newly created variable.
492 void init_interface_type(const struct glsl_type
*type
)
494 assert(this->interface_type
== NULL
);
495 this->interface_type
= type
;
496 if (this->is_interface_instance()) {
497 this->u
.max_ifc_array_access
=
498 rzalloc_array(this, unsigned, type
->length
);
503 * Change this->interface_type on a variable that previously had a
504 * different, but compatible, interface_type. This is used during linking
505 * to set the size of arrays in interface blocks.
507 void change_interface_type(const struct glsl_type
*type
)
509 if (this->u
.max_ifc_array_access
!= NULL
) {
510 /* max_ifc_array_access has already been allocated, so make sure the
511 * new interface has the same number of fields as the old one.
513 assert(this->interface_type
->length
== type
->length
);
515 this->interface_type
= type
;
519 * Change this->interface_type on a variable that previously had a
520 * different, and incompatible, interface_type. This is used during
521 * compilation to handle redeclaration of the built-in gl_PerVertex
524 void reinit_interface_type(const struct glsl_type
*type
)
526 if (this->u
.max_ifc_array_access
!= NULL
) {
528 /* Redeclaring gl_PerVertex is only allowed if none of the built-ins
529 * it defines have been accessed yet; so it's safe to throw away the
530 * old max_ifc_array_access pointer, since all of its values are
533 for (unsigned i
= 0; i
< this->interface_type
->length
; i
++)
534 assert(this->u
.max_ifc_array_access
[i
] == 0);
536 ralloc_free(this->u
.max_ifc_array_access
);
537 this->u
.max_ifc_array_access
= NULL
;
539 this->interface_type
= NULL
;
540 init_interface_type(type
);
543 const glsl_type
*get_interface_type() const
545 return this->interface_type
;
549 * Get the max_ifc_array_access pointer
551 * A "set" function is not needed because the array is dynmically allocated
554 inline unsigned *get_max_ifc_array_access()
556 assert(this->data
._num_state_slots
== 0);
557 return this->u
.max_ifc_array_access
;
560 inline unsigned get_num_state_slots() const
562 assert(!this->is_interface_instance()
563 || this->data
._num_state_slots
== 0);
564 return this->data
._num_state_slots
;
567 inline void set_num_state_slots(unsigned n
)
569 assert(!this->is_interface_instance()
571 this->data
._num_state_slots
= n
;
574 inline ir_state_slot
*get_state_slots()
576 return this->is_interface_instance() ? NULL
: this->u
.state_slots
;
579 inline const ir_state_slot
*get_state_slots() const
581 return this->is_interface_instance() ? NULL
: this->u
.state_slots
;
584 inline ir_state_slot
*allocate_state_slots(unsigned n
)
586 assert(!this->is_interface_instance());
588 this->u
.state_slots
= ralloc_array(this, ir_state_slot
, n
);
589 this->data
._num_state_slots
= 0;
591 if (this->u
.state_slots
!= NULL
)
592 this->data
._num_state_slots
= n
;
594 return this->u
.state_slots
;
597 inline bool is_name_ralloced() const
599 return this->name
!= ir_variable::tmp_name
;
603 * Enable emitting extension warnings for this variable
605 void enable_extension_warning(const char *extension
);
608 * Get the extension warning string for this variable
610 * If warnings are not enabled, \c NULL is returned.
612 const char *get_extension_warning() const;
615 * Declared type of the variable
617 const struct glsl_type
*type
;
620 * Declared name of the variable
624 struct ir_variable_data
{
627 * Is the variable read-only?
629 * This is set for variables declared as \c const, shader inputs,
632 unsigned read_only
:1;
636 unsigned invariant
:1;
640 * Has this variable been used for reading or writing?
642 * Several GLSL semantic checks require knowledge of whether or not a
643 * variable has been used. For example, it is an error to redeclare a
644 * variable as invariant after it has been used.
646 * This is only maintained in the ast_to_hir.cpp path, not in
647 * Mesa's fixed function or ARB program paths.
652 * Has this variable been statically assigned?
654 * This answers whether the variable was assigned in any path of
655 * the shader during ast_to_hir. This doesn't answer whether it is
656 * still written after dead code removal, nor is it maintained in
657 * non-ast_to_hir.cpp (GLSL parsing) paths.
662 * When separate shader programs are enabled, only input/outputs between
663 * the stages of a multi-stage separate program can be safely removed
664 * from the shader interface. Other input/outputs must remains active.
666 unsigned always_active_io
:1;
669 * Enum indicating how the variable was declared. See
670 * ir_var_declaration_type.
672 * This is used to detect certain kinds of illegal variable redeclarations.
674 unsigned how_declared
:2;
677 * Storage class of the variable.
679 * \sa ir_variable_mode
684 * Interpolation mode for shader inputs / outputs
686 * \sa ir_variable_interpolation
688 unsigned interpolation
:2;
691 * \name ARB_fragment_coord_conventions
694 unsigned origin_upper_left
:1;
695 unsigned pixel_center_integer
:1;
699 * Was the location explicitly set in the shader?
701 * If the location is explicitly set in the shader, it \b cannot be changed
702 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
705 unsigned explicit_location
:1;
706 unsigned explicit_index
:1;
709 * Was an initial binding explicitly set in the shader?
711 * If so, constant_value contains an integer ir_constant representing the
712 * initial binding point.
714 unsigned explicit_binding
:1;
717 * Does this variable have an initializer?
719 * This is used by the linker to cross-validiate initializers of global
722 unsigned has_initializer
:1;
725 * Is this variable a generic output or input that has not yet been matched
726 * up to a variable in another stage of the pipeline?
728 * This is used by the linker as scratch storage while assigning locations
729 * to generic inputs and outputs.
731 unsigned is_unmatched_generic_inout
:1;
734 * If non-zero, then this variable may be packed along with other variables
735 * into a single varying slot, so this offset should be applied when
736 * accessing components. For example, an offset of 1 means that the x
737 * component of this variable is actually stored in component y of the
738 * location specified by \c location.
740 unsigned location_frac
:2;
743 * Layout of the matrix. Uses glsl_matrix_layout values.
745 unsigned matrix_layout
:2;
748 * Non-zero if this variable was created by lowering a named interface
749 * block which was not an array.
751 * Note that this variable and \c from_named_ifc_block_array will never
754 unsigned from_named_ifc_block_nonarray
:1;
757 * Non-zero if this variable was created by lowering a named interface
758 * block which was an array.
760 * Note that this variable and \c from_named_ifc_block_nonarray will never
763 unsigned from_named_ifc_block_array
:1;
766 * Non-zero if the variable must be a shader input. This is useful for
767 * constraints on function parameters.
769 unsigned must_be_shader_input
:1;
772 * Output index for dual source blending.
775 * The GLSL spec only allows the values 0 or 1 for the index in \b dual
781 * Precision qualifier.
783 * In desktop GLSL we do not care about precision qualifiers at all, in
784 * fact, the spec says that precision qualifiers are ignored.
786 * To make things easy, we make it so that this field is always
787 * GLSL_PRECISION_NONE on desktop shaders. This way all the variables
788 * have the same precision value and the checks we add in the compiler
789 * for this field will never break a desktop shader compile.
791 unsigned precision
:2;
794 * \brief Layout qualifier for gl_FragDepth.
796 * This is not equal to \c ir_depth_layout_none if and only if this
797 * variable is \c gl_FragDepth and a layout qualifier is specified.
799 ir_depth_layout depth_layout
:3;
802 * ARB_shader_image_load_store qualifiers.
804 unsigned image_read_only
:1; /**< "readonly" qualifier. */
805 unsigned image_write_only
:1; /**< "writeonly" qualifier. */
806 unsigned image_coherent
:1;
807 unsigned image_volatile
:1;
808 unsigned image_restrict
:1;
811 * ARB_shader_storage_buffer_object
813 unsigned from_ssbo_unsized_array
:1; /**< unsized array buffer variable. */
816 * Emit a warning if this variable is accessed.
819 uint8_t warn_extension_index
;
822 /** Image internal format if specified explicitly, otherwise GL_NONE. */
823 uint16_t image_format
;
827 * Number of state slots used
830 * This could be stored in as few as 7-bits, if necessary. If it is made
831 * smaller, add an assertion to \c ir_variable::allocate_state_slots to
834 uint16_t _num_state_slots
;
838 * Initial binding point for a sampler, atomic, or UBO.
840 * For array types, this represents the binding point for the first element.
845 * Storage location of the base of this variable
847 * The precise meaning of this field depends on the nature of the variable.
849 * - Vertex shader input: one of the values from \c gl_vert_attrib.
850 * - Vertex shader output: one of the values from \c gl_varying_slot.
851 * - Geometry shader input: one of the values from \c gl_varying_slot.
852 * - Geometry shader output: one of the values from \c gl_varying_slot.
853 * - Fragment shader input: one of the values from \c gl_varying_slot.
854 * - Fragment shader output: one of the values from \c gl_frag_result.
855 * - Uniforms: Per-stage uniform slot number for default uniform block.
856 * - Uniforms: Index within the uniform block definition for UBO members.
857 * - Non-UBO Uniforms: explicit location until linking then reused to
858 * store uniform slot number.
859 * - Other: This field is not currently used.
861 * If the variable is a uniform, shader input, or shader output, and the
862 * slot has not been assigned, the value will be -1.
867 * Vertex stream output identifier.
872 * Location an atomic counter is stored at.
879 * Highest element accessed with a constant expression array index
881 * Not used for non-array variables.
883 unsigned max_array_access
;
886 * Allow (only) ir_variable direct access private members.
888 friend class ir_variable
;
892 * Value assigned in the initializer of a variable declared "const"
894 ir_constant
*constant_value
;
897 * Constant expression assigned in the initializer of the variable
900 * This field and \c ::constant_value are distinct. Even if the two fields
901 * refer to constants with the same value, they must point to separate
904 ir_constant
*constant_initializer
;
907 static const char *const warn_extension_table
[];
911 * For variables which satisfy the is_interface_instance() predicate,
912 * this points to an array of integers such that if the ith member of
913 * the interface block is an array, max_ifc_array_access[i] is the
914 * maximum array element of that member that has been accessed. If the
915 * ith member of the interface block is not an array,
916 * max_ifc_array_access[i] is unused.
918 * For variables whose type is not an interface block, this pointer is
921 unsigned *max_ifc_array_access
;
924 * Built-in state that backs this uniform
926 * Once set at variable creation, \c state_slots must remain invariant.
928 * If the variable is not a uniform, \c _num_state_slots will be zero
929 * and \c state_slots will be \c NULL.
931 ir_state_slot
*state_slots
;
935 * For variables that are in an interface block or are an instance of an
936 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
938 * \sa ir_variable::location
940 const glsl_type
*interface_type
;
943 * Name used for anonymous compiler temporaries
945 static const char tmp_name
[];
949 * Should the construct keep names for ir_var_temporary variables?
951 * When this global is false, names passed to the constructor for
952 * \c ir_var_temporary variables will be dropped. Instead, the variable will
953 * be named "compiler_temp". This name will be in static storage.
956 * \b NEVER change the mode of an \c ir_var_temporary.
959 * This variable is \b not thread-safe. It is global, \b not
960 * per-context. It begins life false. A context can, at some point, make
961 * it true. From that point on, it will be true forever. This should be
962 * okay since it will only be set true while debugging.
964 static bool temporaries_allocate_names
;
968 * A function that returns whether a built-in function is available in the
969 * current shading language (based on version, ES or desktop, and extensions).
971 typedef bool (*builtin_available_predicate
)(const _mesa_glsl_parse_state
*);
975 * The representation of a function instance; may be the full definition or
976 * simply a prototype.
978 class ir_function_signature
: public ir_instruction
{
979 /* An ir_function_signature will be part of the list of signatures in
983 ir_function_signature(const glsl_type
*return_type
,
984 builtin_available_predicate builtin_avail
= NULL
);
986 virtual ir_function_signature
*clone(void *mem_ctx
,
987 struct hash_table
*ht
) const;
988 ir_function_signature
*clone_prototype(void *mem_ctx
,
989 struct hash_table
*ht
) const;
991 virtual void accept(ir_visitor
*v
)
996 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
999 * Attempt to evaluate this function as a constant expression,
1000 * given a list of the actual parameters and the variable context.
1001 * Returns NULL for non-built-ins.
1003 ir_constant
*constant_expression_value(exec_list
*actual_parameters
, struct hash_table
*variable_context
);
1006 * Get the name of the function for which this is a signature
1008 const char *function_name() const;
1011 * Get a handle to the function for which this is a signature
1013 * There is no setter function, this function returns a \c const pointer,
1014 * and \c ir_function_signature::_function is private for a reason. The
1015 * only way to make a connection between a function and function signature
1016 * is via \c ir_function::add_signature. This helps ensure that certain
1017 * invariants (i.e., a function signature is in the list of signatures for
1018 * its \c _function) are met.
1020 * \sa ir_function::add_signature
1022 inline const class ir_function
*function() const
1024 return this->_function
;
1028 * Check whether the qualifiers match between this signature's parameters
1029 * and the supplied parameter list. If not, returns the name of the first
1030 * parameter with mismatched qualifiers (for use in error messages).
1032 const char *qualifiers_match(exec_list
*params
);
1035 * Replace the current parameter list with the given one. This is useful
1036 * if the current information came from a prototype, and either has invalid
1037 * or missing parameter names.
1039 void replace_parameters(exec_list
*new_params
);
1042 * Function return type.
1044 * \note This discards the optional precision qualifier.
1046 const struct glsl_type
*return_type
;
1049 * List of ir_variable of function parameters.
1051 * This represents the storage. The paramaters passed in a particular
1052 * call will be in ir_call::actual_paramaters.
1054 struct exec_list parameters
;
1056 /** Whether or not this function has a body (which may be empty). */
1057 unsigned is_defined
:1;
1059 /** Whether or not this function signature is a built-in. */
1060 bool is_builtin() const;
1063 * Whether or not this function is an intrinsic to be implemented
1068 /** Whether or not a built-in is available for this shader. */
1069 bool is_builtin_available(const _mesa_glsl_parse_state
*state
) const;
1071 /** Body of instructions in the function. */
1072 struct exec_list body
;
1076 * A function pointer to a predicate that answers whether a built-in
1077 * function is available in the current shader. NULL if not a built-in.
1079 builtin_available_predicate builtin_avail
;
1081 /** Function of which this signature is one overload. */
1082 class ir_function
*_function
;
1084 /** Function signature of which this one is a prototype clone */
1085 const ir_function_signature
*origin
;
1087 friend class ir_function
;
1090 * Helper function to run a list of instructions for constant
1091 * expression evaluation.
1093 * The hash table represents the values of the visible variables.
1094 * There are no scoping issues because the table is indexed on
1095 * ir_variable pointers, not variable names.
1097 * Returns false if the expression is not constant, true otherwise,
1098 * and the value in *result if result is non-NULL.
1100 bool constant_expression_evaluate_expression_list(const struct exec_list
&body
,
1101 struct hash_table
*variable_context
,
1102 ir_constant
**result
);
1107 * Header for tracking multiple overloaded functions with the same name.
1108 * Contains a list of ir_function_signatures representing each of the
1111 class ir_function
: public ir_instruction
{
1113 ir_function(const char *name
);
1115 virtual ir_function
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1117 virtual void accept(ir_visitor
*v
)
1122 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1124 void add_signature(ir_function_signature
*sig
)
1126 sig
->_function
= this;
1127 this->signatures
.push_tail(sig
);
1131 * Find a signature that matches a set of actual parameters, taking implicit
1132 * conversions into account. Also flags whether the match was exact.
1134 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1135 const exec_list
*actual_param
,
1136 bool allow_builtins
,
1137 bool *match_is_exact
);
1140 * Find a signature that matches a set of actual parameters, taking implicit
1141 * conversions into account.
1143 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1144 const exec_list
*actual_param
,
1145 bool allow_builtins
);
1148 * Find a signature that exactly matches a set of actual parameters without
1149 * any implicit type conversions.
1151 ir_function_signature
*exact_matching_signature(_mesa_glsl_parse_state
*state
,
1152 const exec_list
*actual_ps
);
1155 * Name of the function.
1159 /** Whether or not this function has a signature that isn't a built-in. */
1160 bool has_user_signature();
1163 * List of ir_function_signature for each overloaded function with this name.
1165 struct exec_list signatures
;
1168 * is this function a subroutine type declaration
1169 * e.g. subroutine void type1(float arg1);
1174 * is this function associated to a subroutine type
1175 * e.g. subroutine (type1, type2) function_name { function_body };
1176 * would have num_subroutine_types 2,
1177 * and pointers to the type1 and type2 types.
1179 int num_subroutine_types
;
1180 const struct glsl_type
**subroutine_types
;
1182 int subroutine_index
;
1185 inline const char *ir_function_signature::function_name() const
1187 return this->_function
->name
;
1193 * IR instruction representing high-level if-statements
1195 class ir_if
: public ir_instruction
{
1197 ir_if(ir_rvalue
*condition
)
1198 : ir_instruction(ir_type_if
), condition(condition
)
1202 virtual ir_if
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1204 virtual void accept(ir_visitor
*v
)
1209 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1211 ir_rvalue
*condition
;
1212 /** List of ir_instruction for the body of the then branch */
1213 exec_list then_instructions
;
1214 /** List of ir_instruction for the body of the else branch */
1215 exec_list else_instructions
;
1220 * IR instruction representing a high-level loop structure.
1222 class ir_loop
: public ir_instruction
{
1226 virtual ir_loop
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1228 virtual void accept(ir_visitor
*v
)
1233 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1235 /** List of ir_instruction that make up the body of the loop. */
1236 exec_list body_instructions
;
1240 class ir_assignment
: public ir_instruction
{
1242 ir_assignment(ir_rvalue
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
= NULL
);
1245 * Construct an assignment with an explicit write mask
1248 * Since a write mask is supplied, the LHS must already be a bare
1249 * \c ir_dereference. The cannot be any swizzles in the LHS.
1251 ir_assignment(ir_dereference
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
,
1252 unsigned write_mask
);
1254 virtual ir_assignment
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1256 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1258 virtual void accept(ir_visitor
*v
)
1263 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1266 * Get a whole variable written by an assignment
1268 * If the LHS of the assignment writes a whole variable, the variable is
1269 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1272 * - Assigning to a scalar
1273 * - Assigning to all components of a vector
1274 * - Whole array (or matrix) assignment
1275 * - Whole structure assignment
1277 ir_variable
*whole_variable_written();
1280 * Set the LHS of an assignment
1282 void set_lhs(ir_rvalue
*lhs
);
1285 * Left-hand side of the assignment.
1287 * This should be treated as read only. If you need to set the LHS of an
1288 * assignment, use \c ir_assignment::set_lhs.
1290 ir_dereference
*lhs
;
1293 * Value being assigned
1298 * Optional condition for the assignment.
1300 ir_rvalue
*condition
;
1304 * Component mask written
1306 * For non-vector types in the LHS, this field will be zero. For vector
1307 * types, a bit will be set for each component that is written. Note that
1308 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1310 * A partially-set write mask means that each enabled channel gets
1311 * the value from a consecutive channel of the rhs. For example,
1312 * to write just .xyw of gl_FrontColor with color:
1314 * (assign (constant bool (1)) (xyw)
1315 * (var_ref gl_FragColor)
1316 * (swiz xyw (var_ref color)))
1318 unsigned write_mask
:4;
1321 /* Update ir_expression::get_num_operands() and operator_strs when
1322 * updating this list.
1324 enum ir_expression_operation
{
1333 ir_unop_exp
, /**< Log base e on gentype */
1334 ir_unop_log
, /**< Natural log on gentype */
1337 ir_unop_f2i
, /**< Float-to-integer conversion. */
1338 ir_unop_f2u
, /**< Float-to-unsigned conversion. */
1339 ir_unop_i2f
, /**< Integer-to-float conversion. */
1340 ir_unop_f2b
, /**< Float-to-boolean conversion */
1341 ir_unop_b2f
, /**< Boolean-to-float conversion */
1342 ir_unop_i2b
, /**< int-to-boolean conversion */
1343 ir_unop_b2i
, /**< Boolean-to-int conversion */
1344 ir_unop_u2f
, /**< Unsigned-to-float conversion. */
1345 ir_unop_i2u
, /**< Integer-to-unsigned conversion. */
1346 ir_unop_u2i
, /**< Unsigned-to-integer conversion. */
1347 ir_unop_d2f
, /**< Double-to-float conversion. */
1348 ir_unop_f2d
, /**< Float-to-double conversion. */
1349 ir_unop_d2i
, /**< Double-to-integer conversion. */
1350 ir_unop_i2d
, /**< Integer-to-double conversion. */
1351 ir_unop_d2u
, /**< Double-to-unsigned conversion. */
1352 ir_unop_u2d
, /**< Unsigned-to-double conversion. */
1353 ir_unop_d2b
, /**< Double-to-boolean conversion. */
1354 ir_unop_bitcast_i2f
, /**< Bit-identical int-to-float "conversion" */
1355 ir_unop_bitcast_f2i
, /**< Bit-identical float-to-int "conversion" */
1356 ir_unop_bitcast_u2f
, /**< Bit-identical uint-to-float "conversion" */
1357 ir_unop_bitcast_f2u
, /**< Bit-identical float-to-uint "conversion" */
1361 * \name Unary floating-point rounding operations.
1372 * \name Trigonometric operations.
1380 * \name Partial derivatives.
1384 ir_unop_dFdx_coarse
,
1387 ir_unop_dFdy_coarse
,
1392 * \name Floating point pack and unpack operations.
1395 ir_unop_pack_snorm_2x16
,
1396 ir_unop_pack_snorm_4x8
,
1397 ir_unop_pack_unorm_2x16
,
1398 ir_unop_pack_unorm_4x8
,
1399 ir_unop_pack_half_2x16
,
1400 ir_unop_unpack_snorm_2x16
,
1401 ir_unop_unpack_snorm_4x8
,
1402 ir_unop_unpack_unorm_2x16
,
1403 ir_unop_unpack_unorm_4x8
,
1404 ir_unop_unpack_half_2x16
,
1408 * \name Lowered floating point unpacking operations.
1410 * \see lower_packing_builtins_visitor::split_unpack_half_2x16
1413 ir_unop_unpack_half_2x16_split_x
,
1414 ir_unop_unpack_half_2x16_split_y
,
1418 * \name Bit operations, part of ARB_gpu_shader5.
1421 ir_unop_bitfield_reverse
,
1430 * \name Double packing, part of ARB_gpu_shader_fp64.
1433 ir_unop_pack_double_2x32
,
1434 ir_unop_unpack_double_2x32
,
1442 ir_unop_subroutine_to_int
,
1444 * Interpolate fs input at centroid
1446 * operand0 is the fs input.
1448 ir_unop_interpolate_at_centroid
,
1451 * Ask the driver for the total size of a buffer block.
1453 * operand0 is the ir_constant buffer block index in the linked shader.
1455 ir_unop_get_buffer_size
,
1458 * Calculate length of an unsized array inside a buffer block.
1459 * This opcode is going to be replaced in a lowering pass inside
1462 * operand0 is the unsized array's ir_value for the calculation
1465 ir_unop_ssbo_unsized_array_length
,
1468 * A sentinel marking the last of the unary operations.
1470 ir_last_unop
= ir_unop_ssbo_unsized_array_length
,
1474 ir_binop_mul
, /**< Floating-point or low 32-bit integer multiply. */
1475 ir_binop_imul_high
, /**< Calculates the high 32-bits of a 64-bit multiply. */
1479 * Returns the carry resulting from the addition of the two arguments.
1486 * Returns the borrow resulting from the subtraction of the second argument
1487 * from the first argument.
1494 * Takes one of two combinations of arguments:
1497 * - mod(vecN, float)
1499 * Does not take integer types.
1504 * \name Binary comparison operators which return a boolean vector.
1505 * The type of both operands must be equal.
1515 * Returns single boolean for whether all components of operands[0]
1516 * equal the components of operands[1].
1520 * Returns single boolean for whether any component of operands[0]
1521 * is not equal to the corresponding component of operands[1].
1523 ir_binop_any_nequal
,
1527 * \name Bit-wise binary operations.
1548 * \name Lowered floating point packing operations.
1550 * \see lower_packing_builtins_visitor::split_pack_half_2x16
1553 ir_binop_pack_half_2x16_split
,
1557 * \name First half of a lowered bitfieldInsert() operation.
1559 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1566 * Load a value the size of a given GLSL type from a uniform block.
1568 * operand0 is the ir_constant uniform block index in the linked shader.
1569 * operand1 is a byte offset within the uniform block.
1574 * \name Multiplies a number by two to a power, part of ARB_gpu_shader5.
1581 * Extract a scalar from a vector
1583 * operand0 is the vector
1584 * operand1 is the index of the field to read from operand0
1586 ir_binop_vector_extract
,
1589 * Interpolate fs input at offset
1591 * operand0 is the fs input
1592 * operand1 is the offset from the pixel center
1594 ir_binop_interpolate_at_offset
,
1597 * Interpolate fs input at sample position
1599 * operand0 is the fs input
1600 * operand1 is the sample ID
1602 ir_binop_interpolate_at_sample
,
1605 * A sentinel marking the last of the binary operations.
1607 ir_last_binop
= ir_binop_interpolate_at_sample
,
1610 * \name Fused floating-point multiply-add, part of ARB_gpu_shader5.
1619 * \name Conditional Select
1621 * A vector conditional select instruction (like ?:, but operating per-
1622 * component on vectors).
1624 * \see lower_instructions_visitor::ldexp_to_arith
1631 * \name Second half of a lowered bitfieldInsert() operation.
1633 * \see lower_instructions::bitfield_insert_to_bfm_bfi
1639 ir_triop_bitfield_extract
,
1642 * Generate a value with one field of a vector changed
1644 * operand0 is the vector
1645 * operand1 is the value to write into the vector result
1646 * operand2 is the index in operand0 to be modified
1648 ir_triop_vector_insert
,
1651 * A sentinel marking the last of the ternary operations.
1653 ir_last_triop
= ir_triop_vector_insert
,
1655 ir_quadop_bitfield_insert
,
1660 * A sentinel marking the last of the ternary operations.
1662 ir_last_quadop
= ir_quadop_vector
,
1665 * A sentinel marking the last of all operations.
1667 ir_last_opcode
= ir_quadop_vector
1670 class ir_expression
: public ir_rvalue
{
1672 ir_expression(int op
, const struct glsl_type
*type
,
1673 ir_rvalue
*op0
, ir_rvalue
*op1
= NULL
,
1674 ir_rvalue
*op2
= NULL
, ir_rvalue
*op3
= NULL
);
1677 * Constructor for unary operation expressions
1679 ir_expression(int op
, ir_rvalue
*);
1682 * Constructor for binary operation expressions
1684 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
);
1687 * Constructor for ternary operation expressions
1689 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
, ir_rvalue
*op2
);
1691 virtual bool equals(const ir_instruction
*ir
,
1692 enum ir_node_type ignore
= ir_type_unset
) const;
1694 virtual ir_expression
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1697 * Attempt to constant-fold the expression
1699 * The "variable_context" hash table links ir_variable * to ir_constant *
1700 * that represent the variables' values. \c NULL represents an empty
1703 * If the expression cannot be constant folded, this method will return
1706 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1709 * Determine the number of operands used by an expression
1711 static unsigned int get_num_operands(ir_expression_operation
);
1714 * Determine the number of operands used by an expression
1716 unsigned int get_num_operands() const
1718 return (this->operation
== ir_quadop_vector
)
1719 ? this->type
->vector_elements
: get_num_operands(operation
);
1723 * Return whether the expression operates on vectors horizontally.
1725 bool is_horizontal() const
1727 return operation
== ir_binop_all_equal
||
1728 operation
== ir_binop_any_nequal
||
1729 operation
== ir_unop_any
||
1730 operation
== ir_binop_dot
||
1731 operation
== ir_quadop_vector
;
1735 * Return a string representing this expression's operator.
1737 const char *operator_string();
1740 * Return a string representing this expression's operator.
1742 static const char *operator_string(ir_expression_operation
);
1746 * Do a reverse-lookup to translate the given string into an operator.
1748 static ir_expression_operation
get_operator(const char *);
1750 virtual void accept(ir_visitor
*v
)
1755 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1757 virtual ir_variable
*variable_referenced() const;
1759 ir_expression_operation operation
;
1760 ir_rvalue
*operands
[4];
1765 * HIR instruction representing a high-level function call, containing a list
1766 * of parameters and returning a value in the supplied temporary.
1768 class ir_call
: public ir_instruction
{
1770 ir_call(ir_function_signature
*callee
,
1771 ir_dereference_variable
*return_deref
,
1772 exec_list
*actual_parameters
)
1773 : ir_instruction(ir_type_call
), return_deref(return_deref
), callee(callee
), sub_var(NULL
), array_idx(NULL
)
1775 assert(callee
->return_type
!= NULL
);
1776 actual_parameters
->move_nodes_to(& this->actual_parameters
);
1777 this->use_builtin
= callee
->is_builtin();
1780 ir_call(ir_function_signature
*callee
,
1781 ir_dereference_variable
*return_deref
,
1782 exec_list
*actual_parameters
,
1783 ir_variable
*var
, ir_rvalue
*array_idx
)
1784 : ir_instruction(ir_type_call
), return_deref(return_deref
), callee(callee
), sub_var(var
), array_idx(array_idx
)
1786 assert(callee
->return_type
!= NULL
);
1787 actual_parameters
->move_nodes_to(& this->actual_parameters
);
1788 this->use_builtin
= callee
->is_builtin();
1791 virtual ir_call
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1793 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1795 virtual void accept(ir_visitor
*v
)
1800 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1803 * Get the name of the function being called.
1805 const char *callee_name() const
1807 return callee
->function_name();
1811 * Generates an inline version of the function before @ir,
1812 * storing the return value in return_deref.
1814 void generate_inline(ir_instruction
*ir
);
1817 * Storage for the function's return value.
1818 * This must be NULL if the return type is void.
1820 ir_dereference_variable
*return_deref
;
1823 * The specific function signature being called.
1825 ir_function_signature
*callee
;
1827 /* List of ir_rvalue of paramaters passed in this call. */
1828 exec_list actual_parameters
;
1830 /** Should this call only bind to a built-in function? */
1834 * ARB_shader_subroutine support -
1835 * the subroutine uniform variable and array index
1836 * rvalue to be used in the lowering pass later.
1838 ir_variable
*sub_var
;
1839 ir_rvalue
*array_idx
;
1844 * \name Jump-like IR instructions.
1846 * These include \c break, \c continue, \c return, and \c discard.
1849 class ir_jump
: public ir_instruction
{
1851 ir_jump(enum ir_node_type t
)
1857 class ir_return
: public ir_jump
{
1860 : ir_jump(ir_type_return
), value(NULL
)
1864 ir_return(ir_rvalue
*value
)
1865 : ir_jump(ir_type_return
), value(value
)
1869 virtual ir_return
*clone(void *mem_ctx
, struct hash_table
*) const;
1871 ir_rvalue
*get_value() const
1876 virtual void accept(ir_visitor
*v
)
1881 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1888 * Jump instructions used inside loops
1890 * These include \c break and \c continue. The \c break within a loop is
1891 * different from the \c break within a switch-statement.
1893 * \sa ir_switch_jump
1895 class ir_loop_jump
: public ir_jump
{
1902 ir_loop_jump(jump_mode mode
)
1903 : ir_jump(ir_type_loop_jump
)
1908 virtual ir_loop_jump
*clone(void *mem_ctx
, struct hash_table
*) const;
1910 virtual void accept(ir_visitor
*v
)
1915 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1917 bool is_break() const
1919 return mode
== jump_break
;
1922 bool is_continue() const
1924 return mode
== jump_continue
;
1927 /** Mode selector for the jump instruction. */
1928 enum jump_mode mode
;
1932 * IR instruction representing discard statements.
1934 class ir_discard
: public ir_jump
{
1937 : ir_jump(ir_type_discard
)
1939 this->condition
= NULL
;
1942 ir_discard(ir_rvalue
*cond
)
1943 : ir_jump(ir_type_discard
)
1945 this->condition
= cond
;
1948 virtual ir_discard
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1950 virtual void accept(ir_visitor
*v
)
1955 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1957 ir_rvalue
*condition
;
1963 * Texture sampling opcodes used in ir_texture
1965 enum ir_texture_opcode
{
1966 ir_tex
, /**< Regular texture look-up */
1967 ir_txb
, /**< Texture look-up with LOD bias */
1968 ir_txl
, /**< Texture look-up with explicit LOD */
1969 ir_txd
, /**< Texture look-up with partial derivatvies */
1970 ir_txf
, /**< Texel fetch with explicit LOD */
1971 ir_txf_ms
, /**< Multisample texture fetch */
1972 ir_txs
, /**< Texture size */
1973 ir_lod
, /**< Texture lod query */
1974 ir_tg4
, /**< Texture gather */
1975 ir_query_levels
, /**< Texture levels query */
1976 ir_texture_samples
, /**< Texture samples query */
1977 ir_samples_identical
, /**< Query whether all samples are definitely identical. */
1982 * IR instruction to sample a texture
1984 * The specific form of the IR instruction depends on the \c mode value
1985 * selected from \c ir_texture_opcodes. In the printed IR, these will
1988 * Texel offset (0 or an expression)
1989 * | Projection divisor
1990 * | | Shadow comparitor
1993 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1994 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1995 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1996 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1997 * (txf <type> <sampler> <coordinate> 0 <lod>)
1999 * <type> <sampler> <coordinate> <sample_index>)
2000 * (txs <type> <sampler> <lod>)
2001 * (lod <type> <sampler> <coordinate>)
2002 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
2003 * (query_levels <type> <sampler>)
2004 * (samples_identical <sampler> <coordinate>)
2006 class ir_texture
: public ir_rvalue
{
2008 ir_texture(enum ir_texture_opcode op
)
2009 : ir_rvalue(ir_type_texture
),
2010 op(op
), sampler(NULL
), coordinate(NULL
), projector(NULL
),
2011 shadow_comparitor(NULL
), offset(NULL
)
2013 memset(&lod_info
, 0, sizeof(lod_info
));
2016 virtual ir_texture
*clone(void *mem_ctx
, struct hash_table
*) const;
2018 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2020 virtual void accept(ir_visitor
*v
)
2025 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2027 virtual bool equals(const ir_instruction
*ir
,
2028 enum ir_node_type ignore
= ir_type_unset
) const;
2031 * Return a string representing the ir_texture_opcode.
2033 const char *opcode_string();
2035 /** Set the sampler and type. */
2036 void set_sampler(ir_dereference
*sampler
, const glsl_type
*type
);
2039 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
2041 static ir_texture_opcode
get_opcode(const char *);
2043 enum ir_texture_opcode op
;
2045 /** Sampler to use for the texture access. */
2046 ir_dereference
*sampler
;
2048 /** Texture coordinate to sample */
2049 ir_rvalue
*coordinate
;
2052 * Value used for projective divide.
2054 * If there is no projective divide (the common case), this will be
2055 * \c NULL. Optimization passes should check for this to point to a constant
2056 * of 1.0 and replace that with \c NULL.
2058 ir_rvalue
*projector
;
2061 * Coordinate used for comparison on shadow look-ups.
2063 * If there is no shadow comparison, this will be \c NULL. For the
2064 * \c ir_txf opcode, this *must* be \c NULL.
2066 ir_rvalue
*shadow_comparitor
;
2068 /** Texel offset. */
2072 ir_rvalue
*lod
; /**< Floating point LOD */
2073 ir_rvalue
*bias
; /**< Floating point LOD bias */
2074 ir_rvalue
*sample_index
; /**< MSAA sample index */
2075 ir_rvalue
*component
; /**< Gather component selector */
2077 ir_rvalue
*dPdx
; /**< Partial derivative of coordinate wrt X */
2078 ir_rvalue
*dPdy
; /**< Partial derivative of coordinate wrt Y */
2084 struct ir_swizzle_mask
{
2091 * Number of components in the swizzle.
2093 unsigned num_components
:3;
2096 * Does the swizzle contain duplicate components?
2098 * L-value swizzles cannot contain duplicate components.
2100 unsigned has_duplicates
:1;
2104 class ir_swizzle
: public ir_rvalue
{
2106 ir_swizzle(ir_rvalue
*, unsigned x
, unsigned y
, unsigned z
, unsigned w
,
2109 ir_swizzle(ir_rvalue
*val
, const unsigned *components
, unsigned count
);
2111 ir_swizzle(ir_rvalue
*val
, ir_swizzle_mask mask
);
2113 virtual ir_swizzle
*clone(void *mem_ctx
, struct hash_table
*) const;
2115 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2118 * Construct an ir_swizzle from the textual representation. Can fail.
2120 static ir_swizzle
*create(ir_rvalue
*, const char *, unsigned vector_length
);
2122 virtual void accept(ir_visitor
*v
)
2127 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2129 virtual bool equals(const ir_instruction
*ir
,
2130 enum ir_node_type ignore
= ir_type_unset
) const;
2132 bool is_lvalue() const
2134 return val
->is_lvalue() && !mask
.has_duplicates
;
2138 * Get the variable that is ultimately referenced by an r-value
2140 virtual ir_variable
*variable_referenced() const;
2143 ir_swizzle_mask mask
;
2147 * Initialize the mask component of a swizzle
2149 * This is used by the \c ir_swizzle constructors.
2151 void init_mask(const unsigned *components
, unsigned count
);
2155 class ir_dereference
: public ir_rvalue
{
2157 virtual ir_dereference
*clone(void *mem_ctx
, struct hash_table
*) const = 0;
2159 bool is_lvalue() const;
2162 * Get the variable that is ultimately referenced by an r-value
2164 virtual ir_variable
*variable_referenced() const = 0;
2167 ir_dereference(enum ir_node_type t
)
2174 class ir_dereference_variable
: public ir_dereference
{
2176 ir_dereference_variable(ir_variable
*var
);
2178 virtual ir_dereference_variable
*clone(void *mem_ctx
,
2179 struct hash_table
*) const;
2181 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2183 virtual bool equals(const ir_instruction
*ir
,
2184 enum ir_node_type ignore
= ir_type_unset
) const;
2187 * Get the variable that is ultimately referenced by an r-value
2189 virtual ir_variable
*variable_referenced() const
2194 virtual ir_variable
*whole_variable_referenced()
2196 /* ir_dereference_variable objects always dereference the entire
2197 * variable. However, if this dereference is dereferenced by anything
2198 * else, the complete deferefernce chain is not a whole-variable
2199 * dereference. This method should only be called on the top most
2200 * ir_rvalue in a dereference chain.
2205 virtual void accept(ir_visitor
*v
)
2210 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2213 * Object being dereferenced.
2219 class ir_dereference_array
: public ir_dereference
{
2221 ir_dereference_array(ir_rvalue
*value
, ir_rvalue
*array_index
);
2223 ir_dereference_array(ir_variable
*var
, ir_rvalue
*array_index
);
2225 virtual ir_dereference_array
*clone(void *mem_ctx
,
2226 struct hash_table
*) const;
2228 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2230 virtual bool equals(const ir_instruction
*ir
,
2231 enum ir_node_type ignore
= ir_type_unset
) const;
2234 * Get the variable that is ultimately referenced by an r-value
2236 virtual ir_variable
*variable_referenced() const
2238 return this->array
->variable_referenced();
2241 virtual void accept(ir_visitor
*v
)
2246 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2249 ir_rvalue
*array_index
;
2252 void set_array(ir_rvalue
*value
);
2256 class ir_dereference_record
: public ir_dereference
{
2258 ir_dereference_record(ir_rvalue
*value
, const char *field
);
2260 ir_dereference_record(ir_variable
*var
, const char *field
);
2262 virtual ir_dereference_record
*clone(void *mem_ctx
,
2263 struct hash_table
*) const;
2265 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2268 * Get the variable that is ultimately referenced by an r-value
2270 virtual ir_variable
*variable_referenced() const
2272 return this->record
->variable_referenced();
2275 virtual void accept(ir_visitor
*v
)
2280 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2288 * Data stored in an ir_constant
2290 union ir_constant_data
{
2299 class ir_constant
: public ir_rvalue
{
2301 ir_constant(const struct glsl_type
*type
, const ir_constant_data
*data
);
2302 ir_constant(bool b
, unsigned vector_elements
=1);
2303 ir_constant(unsigned int u
, unsigned vector_elements
=1);
2304 ir_constant(int i
, unsigned vector_elements
=1);
2305 ir_constant(float f
, unsigned vector_elements
=1);
2306 ir_constant(double d
, unsigned vector_elements
=1);
2309 * Construct an ir_constant from a list of ir_constant values
2311 ir_constant(const struct glsl_type
*type
, exec_list
*values
);
2314 * Construct an ir_constant from a scalar component of another ir_constant
2316 * The new \c ir_constant inherits the type of the component from the
2320 * In the case of a matrix constant, the new constant is a scalar, \b not
2323 ir_constant(const ir_constant
*c
, unsigned i
);
2326 * Return a new ir_constant of the specified type containing all zeros.
2328 static ir_constant
*zero(void *mem_ctx
, const glsl_type
*type
);
2330 virtual ir_constant
*clone(void *mem_ctx
, struct hash_table
*) const;
2332 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2334 virtual void accept(ir_visitor
*v
)
2339 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2341 virtual bool equals(const ir_instruction
*ir
,
2342 enum ir_node_type ignore
= ir_type_unset
) const;
2345 * Get a particular component of a constant as a specific type
2347 * This is useful, for example, to get a value from an integer constant
2348 * as a float or bool. This appears frequently when constructors are
2349 * called with all constant parameters.
2352 bool get_bool_component(unsigned i
) const;
2353 float get_float_component(unsigned i
) const;
2354 double get_double_component(unsigned i
) const;
2355 int get_int_component(unsigned i
) const;
2356 unsigned get_uint_component(unsigned i
) const;
2359 ir_constant
*get_array_element(unsigned i
) const;
2361 ir_constant
*get_record_field(const char *name
);
2364 * Copy the values on another constant at a given offset.
2366 * The offset is ignored for array or struct copies, it's only for
2367 * scalars or vectors into vectors or matrices.
2369 * With identical types on both sides and zero offset it's clone()
2370 * without creating a new object.
2373 void copy_offset(ir_constant
*src
, int offset
);
2376 * Copy the values on another constant at a given offset and
2377 * following an assign-like mask.
2379 * The mask is ignored for scalars.
2381 * Note that this function only handles what assign can handle,
2382 * i.e. at most a vector as source and a column of a matrix as
2386 void copy_masked_offset(ir_constant
*src
, int offset
, unsigned int mask
);
2389 * Determine whether a constant has the same value as another constant
2391 * \sa ir_constant::is_zero, ir_constant::is_one,
2392 * ir_constant::is_negative_one
2394 bool has_value(const ir_constant
*) const;
2397 * Return true if this ir_constant represents the given value.
2399 * For vectors, this checks that each component is the given value.
2401 virtual bool is_value(float f
, int i
) const;
2402 virtual bool is_zero() const;
2403 virtual bool is_one() const;
2404 virtual bool is_negative_one() const;
2407 * Return true for constants that could be stored as 16-bit unsigned values.
2409 * Note that this will return true even for signed integer ir_constants, as
2410 * long as the value is non-negative and fits in 16-bits.
2412 virtual bool is_uint16_constant() const;
2415 * Value of the constant.
2417 * The field used to back the values supplied by the constant is determined
2418 * by the type associated with the \c ir_instruction. Constants may be
2419 * scalars, vectors, or matrices.
2421 union ir_constant_data value
;
2423 /* Array elements */
2424 ir_constant
**array_elements
;
2426 /* Structure fields */
2427 exec_list components
;
2431 * Parameterless constructor only used by the clone method
2437 * IR instruction to emit a vertex in a geometry shader.
2439 class ir_emit_vertex
: public ir_instruction
{
2441 ir_emit_vertex(ir_rvalue
*stream
)
2442 : ir_instruction(ir_type_emit_vertex
),
2448 virtual void accept(ir_visitor
*v
)
2453 virtual ir_emit_vertex
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2455 return new(mem_ctx
) ir_emit_vertex(this->stream
->clone(mem_ctx
, ht
));
2458 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2460 int stream_id() const
2462 return stream
->as_constant()->value
.i
[0];
2469 * IR instruction to complete the current primitive and start a new one in a
2472 class ir_end_primitive
: public ir_instruction
{
2474 ir_end_primitive(ir_rvalue
*stream
)
2475 : ir_instruction(ir_type_end_primitive
),
2481 virtual void accept(ir_visitor
*v
)
2486 virtual ir_end_primitive
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2488 return new(mem_ctx
) ir_end_primitive(this->stream
->clone(mem_ctx
, ht
));
2491 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2493 int stream_id() const
2495 return stream
->as_constant()->value
.i
[0];
2502 * IR instruction for tessellation control and compute shader barrier.
2504 class ir_barrier
: public ir_instruction
{
2507 : ir_instruction(ir_type_barrier
)
2511 virtual void accept(ir_visitor
*v
)
2516 virtual ir_barrier
*clone(void *mem_ctx
, struct hash_table
*) const
2518 return new(mem_ctx
) ir_barrier();
2521 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2527 * Apply a visitor to each IR node in a list
2530 visit_exec_list(exec_list
*list
, ir_visitor
*visitor
);
2533 * Validate invariants on each IR node in a list
2535 void validate_ir_tree(exec_list
*instructions
);
2537 struct _mesa_glsl_parse_state
;
2538 struct gl_shader_program
;
2541 * Detect whether an unlinked shader contains static recursion
2543 * If the list of instructions is determined to contain static recursion,
2544 * \c _mesa_glsl_error will be called to emit error messages for each function
2545 * that is in the recursion cycle.
2548 detect_recursion_unlinked(struct _mesa_glsl_parse_state
*state
,
2549 exec_list
*instructions
);
2552 * Detect whether a linked shader contains static recursion
2554 * If the list of instructions is determined to contain static recursion,
2555 * \c link_error_printf will be called to emit error messages for each function
2556 * that is in the recursion cycle. In addition,
2557 * \c gl_shader_program::LinkStatus will be set to false.
2560 detect_recursion_linked(struct gl_shader_program
*prog
,
2561 exec_list
*instructions
);
2564 * Make a clone of each IR instruction in a list
2566 * \param in List of IR instructions that are to be cloned
2567 * \param out List to hold the cloned instructions
2570 clone_ir_list(void *mem_ctx
, exec_list
*out
, const exec_list
*in
);
2573 _mesa_glsl_initialize_variables(exec_list
*instructions
,
2574 struct _mesa_glsl_parse_state
*state
);
2577 _mesa_glsl_initialize_derived_variables(gl_shader
*shader
);
2580 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state
*state
);
2583 _mesa_glsl_initialize_builtin_functions();
2585 extern ir_function_signature
*
2586 _mesa_glsl_find_builtin_function(_mesa_glsl_parse_state
*state
,
2587 const char *name
, exec_list
*actual_parameters
);
2589 extern ir_function
*
2590 _mesa_glsl_find_builtin_function_by_name(const char *name
);
2593 _mesa_glsl_get_builtin_function_shader(void);
2595 extern ir_function_signature
*
2596 _mesa_get_main_function_signature(gl_shader
*sh
);
2599 _mesa_glsl_release_functions(void);
2602 _mesa_glsl_release_builtin_functions(void);
2605 reparent_ir(exec_list
*list
, void *mem_ctx
);
2607 struct glsl_symbol_table
;
2610 import_prototypes(const exec_list
*source
, exec_list
*dest
,
2611 struct glsl_symbol_table
*symbols
, void *mem_ctx
);
2614 ir_has_call(ir_instruction
*ir
);
2617 do_set_program_inouts(exec_list
*instructions
, struct gl_program
*prog
,
2618 gl_shader_stage shader_stage
);
2621 prototype_string(const glsl_type
*return_type
, const char *name
,
2622 exec_list
*parameters
);
2625 mode_string(const ir_variable
*var
);
2628 * Built-in / reserved GL variables names start with "gl_"
2631 is_gl_identifier(const char *s
)
2633 return s
&& s
[0] == 'g' && s
[1] == 'l' && s
[2] == '_';
2637 #endif /* __cplusplus */
2639 extern void _mesa_print_ir(FILE *f
, struct exec_list
*instructions
,
2640 struct _mesa_glsl_parse_state
*state
);
2643 fprint_ir(FILE *f
, const void *instruction
);
2650 vertices_per_prim(GLenum prim
);